Lecture Notes in Networks and Systems 89

G. Ranganathan
Joy Chen
Álvaro Rocha   Editors

Inventive
Communication
and Computational
Technologies
Proceedings of ICICCT 2019
Lecture Notes in Networks and Systems

Volume 89

Series Editor
Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences,
Warsaw, Poland

Advisory Editors
Fernando Gomide, Department of Computer Engineering and Automation—DCA,
School of Electrical and Computer Engineering—FEEC, University of Campinas—
UNICAMP, São Paulo, Brazil
Okyay Kaynak, Department of Electrical and Electronic Engineering,
Bogazici University, Istanbul, Turkey
Derong Liu, Department of Electrical and Computer Engineering, University
of Illinois at Chicago, Chicago, USA; Institute of Automation, Chinese Academy
of Sciences, Beijing, China
Witold Pedrycz, Department of Electrical and Computer Engineering,
University of Alberta, Alberta, Canada; Systems Research Institute,
Polish Academy of Sciences, Warsaw, Poland
Marios M. Polycarpou, Department of Electrical and Computer Engineering,
KIOS Research Center for Intelligent Systems and Networks, University of Cyprus,
Nicosia, Cyprus
Imre J. Rudas, Óbuda University, Budapest, Hungary
Jun Wang, Department of Computer Science, City University of Hong Kong,
Kowloon, Hong Kong
The series “Lecture Notes in Networks and Systems” publishes the latest
developments in Networks and Systems—quickly, informally and with high quality.
Original research reported in proceedings and post-proceedings represents the core
of LNNS.
Volumes published in LNNS embrace all aspects and subfields of, as well as new
challenges in, Networks and Systems.
The series contains proceedings and edited volumes in systems and networks,
spanning the areas of Cyber-Physical Systems, Autonomous Systems, Sensor
Networks, Control Systems, Energy Systems, Automotive Systems, Biological
Systems, Vehicular Networking and Connected Vehicles, Aerospace Systems,
Automation, Manufacturing, Smart Grids, Nonlinear Systems, Power Systems,
Robotics, Social Systems, Economic Systems and other. Of particular value to both
the contributors and the readership are the short publication timeframe and the
world-wide distribution and exposure which enable both a wide and rapid
dissemination of research output.
The series covers the theory, applications, and perspectives on the state of the art
and future developments relevant to systems and networks, decision making, control,
complex processes and related areas, as embedded in the fields of interdisciplinary
and applied sciences, engineering, computer science, physics, economics, social, and
life sciences, as well as the paradigms and methodologies behind them.
** Indexing: The books of this series are submitted to ISI Proceedings,
SCOPUS, Google Scholar and Springerlink **

More information about this series at [Link]

G. Ranganathan Joy Chen Álvaro Rocha
• •

Editors

Inventive Communication
and Computational
Technologies
Proceedings of ICICCT 2019

123
Editors
G. Ranganathan Joy Chen
Electronics and Communication Engineering Department of Electrical Engineering
Gnanamani College of Technology Dayeh University
Namakkal, Tamil Nadu, India Chang-Hwa, Taiwan

Álvaro Rocha
University of Coimbra
Rio Tinto, Portugal

ISSN 2367-3370 ISSN 2367-3389 (electronic)

Lecture Notes in Networks and Systems
ISBN 978-981-15-0145-6 ISBN 978-981-15-0146-3 (eBook)
[Link]
© Springer Nature Singapore Pte Ltd. 2020
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with regard
to jurisdictional claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
We dedicated to all the participants of the
conference ICICCT 2019
Foreword

We welcome you to the 2019 International Conference on Inventive Communication

and Computational Technologies (ICICCT 2019) held on April 29–30, 2019, at
Gnanamani College of Technology, Namakkal, Tamil Nadu. ICICCT 2019 provides
a highly competitive forum for reporting the latest developments in the research and
application of communication and computational technologies. We are pleased to
present the proceedings of the conference as its published record.
The conference particularly encouraged the interaction of research students and
developing academics with the more established academic community in an
informal setting to present and to discuss new and current work. Their contributions
helped to make the conference as outstanding as it has been. The papers contributed
the most recent scientific knowledge known in the fields of modern communication
systems which include informatics, data communication and computer networking,
wireless communication, electronics, software engineering, machine learning and
optimization, VLSI design and automation, networking, computing systems, social
networks, Internet of things, cloud and big data.
We hope that this program will further stimulate research in advanced elec-
tronics and communication technologies, artificial intelligence and capsule net-
works, data communication and computer networking and communicating things
networks. We feel honored and privileged to serve the best recent developments in
all the areas of communication technologies to you through this exciting program.
We thank all the authors and participants for their contributions.

Namakkal, India Dr. G. Ranganathan

Conference Chair
ICICCT 2019

vii
Preface

The 2019 International Conference on Inventive Communication and Computational

Technologies (ICICCT 2019) was held on April 29–30, 2019, at Gnanamani College
of Technology, Namakkal, India. ICICCT 2019 aims to cover the recent advance-
ment and trends in the area of communication and computational technologies to
facilitate knowledge sharing and networking interactions on emerging trends and
new challenges.
ICICCT 2019 tends to collect the latest research results and applications on data
communication and computer networking, software engineering, wireless com-
munication, VLSI design and automation, networking, Internet of things, cloud and
big data. It includes a selection of 136 papers from 430 papers submitted to the
conference from universities and industries all over the world. All of the accepted
papers were subjected to strict peer-reviewing by 2–4 expert referees. The papers
have been selected for this volume because of quality and the relevance to the
conference.
We would like to express our sincere appreciation to all the authors for their
contributions to this book. We would like to extend our thanks to all the referees for
their constructive comments on all papers, especially we would like to thank the
Organizing Committee for their hard work. Finally, we would like to thank the
Springer publications for producing this volume.

Namakkal, India Dr. G. Ranganathan

Conference Chair
ICICCT 2019

ix
Acknowledgements

We would like to acknowledge the contribution made to the organization by its

many volunteers. Members contribute their time, energy and knowledge at a local,
regional and international level.
We also thank all the chair persons and conference committee members for their
support.

xi
Contents

Enhanced Security Mechanism in Cloud Based on DNA Excess

3 Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Manjot Kaur and Kiranbir Kaur
TensorFlow-Based Semantic Techniques for Multi-cloud Application
Portability and Interoperability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Tanveer Kaur and Kiranbir Kaur
Enhancing High Availability for NoSQL Database Systems
Using Failover Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Priyanka Gotter and Kiranbir Kaur
Implementation of Automated Bottle Filling System Using PLC . . . . . 33
S. V. Viraktamath, A. S. Umarfarooq, Vinayakgouda Yallappagoudar
and Abhilash P. Hasankar
An Integrated Approach to Network Intrusion Detection
and Prevention . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
B. Bhanu Prakash, Kaki Yeswanth, M. Sai Srinivas, S. Balaji,
Y. Chandra Sekhar and Aswathy K. Nair
A New Design of Equiangular Circular Cum Elliptical Honeycomb
Photonic Crystal Fiber . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Ashish Jain, Ravindra Kumar Sharma, Vimal Agarwal and Sunil Sharma
Survey of Different Countermeasure on Network Layer Attack
in Wireless Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Amruta Chavan, Dipti Jadhav, Nilesh Marathe and Nilima Dongre
Internet of Things: Service-Oriented Architecture Opportunities
and Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
G. R. Sagar and N. Jayapandian

xiii
xiv Contents

Performance Comparison of Multicast Routing Protocols

Based on Route Discovery Process for MANET . . . . . . . . . . . . . . . . . . 79
Kinjal Adhvaryu
New Complex Hybrid Security Algorithm (CHSA) for Network
Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Haider K. Hoomod, Ali Jameel Al-Mousawi and Jolan Rokan Naif
Performance Assessment of Various Encoding Schemes
with Bit Stuffing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
S. Bharath and Gayathri Narayanan
3D(Dimensional)—Wired and Wireless Network-on-Chip (NoC) . . . . . 113
N. Ashokkumar, P. Nagarajan and P. Venkatramana
Performance Analysis of Unified Threat Management (UTM) . . . . . . . 121
Jatin Gharat, Amarsinh Vidhate and Amit Barve
Development of an Efficient Nondestructive Grading Method
for Pomegranate Using Magnetic Resonance Imaging
and Neural Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
Surekha Yakatpure and Krupa Rasane
Implementation of Differential Privacy Using Diffie–Hellman
and AES Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
P. V. Vivek Sridhar Mallya, Aparna Ajith, T. R. Sangeetha, Arya Krishnan
and Gayathri Narayanan
Implementation of Neural Signals in MATLAB (Thought Signals) . . . . 153
Kambhampati Sai Sandilya, Poornima Mohan,
Madiraju Akshay Bharadwaj, B. Maragatha Eswari, K. Namitha,
O. Rajasekhar Reddy and J. Vaishnu Saran
A Framework for Formal Verification of Security Protocols
in C++ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
R. Pradeep, N. R. Sunitha, V. Ravi and Sushma Verma
Locality—Aware Scheduling for Containers in Cloud Computing . . . . 177
G. Charles Babu, A. Sai Hanuman, J. Sasi Kiran and B. Sankara Babu
TorBot: Open Source Intelligence Tool for Dark Web . . . . . . . . . . . . . 187
P. S. Narayanan, R. Ani and Akeem T. L. King
A Novel Approach to View and Modify Data in Cloud Environment
Using Attribute-Based Encryption . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
Swaminathan Subbiah, S. Palaniappan, Sigamani Ashokkumar
and Ananthakrishnan BalaSundaram
Contents xv

Root Cause Detection of Oscillation in Shell and Tube Heat

Exchanger Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
S. Abirami and S. Sivagamasundari
Performance Assessment of Spread Spectrum Communication
Receivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
P. Anand Krisshna, K. Vandith Sreenivas and Gayathri Narayanan
Computationally-Light Metrics to Quantify Link Stability in Mobile
Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
Natarajan Meghanathan
MockRest—A Generic Approach for Automated Mock Framework
for REST APIs Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
Anshu Soni, Virender Ranga and Sandeep Jadhav
Syntactic Interoperability in Real-Time Systems, ROS 2,
and Adaptive AUTOSAR Using Data Distribution Services:
An Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
Navrattan Parmar, Virender Ranga and B. Simhachalam Naidu
IoT-Enabled Water Quality Monitoring System . . . . . . . . . . . . . . . . . . 275
G. Kanagaraj, T. Primya, K. Sashi Rekha, C. Vinothini and P. Anitha
Energy-Efficient Routing-Based Clustering Approaches and Sleep
Scheduling Algorithm for Network Lifetime Maximization in Sensor
Network: A Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
Rajiv R. Bhandari and K. Rajasekhar
Journey of Wireless Communication . . . . . . . . . . . . . . . . . . . . . . . . . . 307
Vignesh Parameswaran and M. Shanmugasundaram
A Survey on K-Means Clustering for Analyzing Variation
in Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
Pratik Patil and A. Karthikeyan
A Review on Mobile Cloud Computing Interoperability Issues
and Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
Tribid Debbarma and K. Chandrasekaran
Filter Bank Modulation in Massive MIMO Scenario . . . . . . . . . . . . . . 335
S. Sruthi and J. Dhoulath Beegum
Autonomous Farming—Visualization of Image Processing
in Agriculture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
Shivam Thakur, Sushant Bawiskar, Sachin Kumar Singh
and M. Shanmugasundaram
Design and Development of End Effector for Domestic Robots . . . . . . 353
V. Indu, Putchala Vinay, Narjala Rohith, Kuppili Puneeth and S. Pramod
xvi Contents

Total Variation and Alternate Direction Method for Deblurring

of Digital Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
S. Rinesh, C. Prajitha, V. Karthick and S. Palaniappan
Design of MIMO Triangular Microstrip Patch Antenna
for IEEE 802.11a Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
M. Arulaalan, K. Mahendran, P. Prabakaran and G. Manikannan
A Hollow Core Bragg Fiber with Multilayered Random Defect
for Refractive Index Sensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
K. Ben Franklin, R. Kumar and C. Nayak
Generation of Multiple Key Based on Monitoring the User
Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
S. Palaniappan, Steward Kirubakaran, V. Parthipan and S. Rinesh
Towards Convolution Neural Networks (CNNs): A Brief Overview
of AI and Deep Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
Preetjot Kaur and Roopali Garg
Populating Indian GST Details into Java Apache Derby Database
Powered by Glassfish Server . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 409
R. Sridevi and S. Srimathi
Intrusion Detection System Using WoSAD Method . . . . . . . . . . . . . . . 423
R. Sridevi and N. Nithya
Blackhole Attack Implementation and Its Performance Evaluation
Using AODV Routing in MANET . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
Anshu Kumari, Madhvi Singhal and Nishi Yadav
Enhancement of Security Using B-RSA Algorithm . . . . . . . . . . . . . . . . 439
Aman Gupta, Saurabh Gupta and Nishi Yadav
Analyzing Different Multiparty Computation Techniques . . . . . . . . . . 451
Tanmay Borade, Dhananjay Dakhane and Tushar Ghorpade
Power and Delay Comparison of 7:3 Compressor Designs
Based on Different Architectures of XOR Gate . . . . . . . . . . . . . . . . . . 461
Rekib Uddin Ahmed and Prabir Saha
Survey of Big Data Warehousing Techniques . . . . . . . . . . . . . . . . . . . . 471
Jaspreet Kaur, Rajashree Shedge and Bharti Joshi
Detecting Denial-of-Service Attacks Using sFlow . . . . . . . . . . . . . . . . . 483
Shivaraj Hublikar, Vijaya Eligar and Arun Kakhandki
Area and Power Efficient Multiplier-Less Architecture for FIR
Differentiator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
Sanjay Eligar and R. M. Banakar
Contents xvii

Optimized Prediction Model to Diagnose Breast Cancer Risk

and Its Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503
Athira Vinod and B. R. Manju
Smart Cane Design for Indoor and Outdoor Navigation:
A Cost-Effective Guide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 517
Vandana Mansur and R. Sujatha
Triclustering of Gene Expression Microarray Data
Using Coarse-Grained Parallel Genetic Algorithm . . . . . . . . . . . . . . . . 529
Shubhankar Mohapatra, Moumita Sarkar, Anjali Mohapatra
and Bhawani Sankar Biswal
Ensemble Feature Selection to Improve Classification Accuracy
in Human Activity Recognition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 541
Nivetha Gopalakrishnan, Venkatalakshmi Krishnan
and Vinodhini Gopalakrishnan
An Exploration on Cloud Computing Security Strategies
and Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 549
Amrita Raj and Rakesh Kumar
A Hybrid Approach-Based Energy Aware Cluster Head Selection
for IOT Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 563
A. Kannammal and S. Suresh
Modified Dive and Rise Technique Incorporating Enhanced
Weighted Centroid Localization Algorithm with Ocean Current
Mobility Model in Underwater Acoustic Sensor Networks . . . . . . . . . . 569
R. Bhairavi and Gnanou Florence Sudha
Tree-Based Approaches for Improving Energy Efficiency and Life
Time of Wireless Sensor Networks (WSN): A Survey and Future
Scope for Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 583
Pranesh and Santhosh L. Deshpande
Automated Cyber Threat Intelligence Generation from Honeypot
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 591
Kumar Sanjeev, B. Janet and R. Eswari
Analysis of MPTCP Packet Scheduling, The Need of Data Hungry
Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 599
Neha Rupesh Thakur and Ashwini S. Kunte
Voice Controlled Home Automation Using Blynk, IFTTT with Live
Feedback . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 619
M. Rajasekhar Reddy, P. Sai Siddartha Reddy, Suthapalli Akhil Sri Harsha
and D. Vishnu Vashista
xviii Contents

AD-LIB: Automated Library System . . . . . . . . . . . . . . . . . . . . . . . . . . 635

Dattatray Sawant and Rohan Patil
Design and Integrate IoT Sensors to RO Water Purifiers for Remote
Monitoring and Allowing Users to Pay Per Usage on the Rented
RO System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 647
H. D. Sundresh and D. Priya
MPTCP a Solution Over Crunch in Bandwidth Requirement
for Multimedia Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 653
Vidya Kubde and Sudhir Sawarkar
Design of AYUSH: A Blockchain-Based Health Record Management
System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 665
A. V. Aswin, K. Y. Basil, Vimal P. Viswan, Basil Reji
and Bineeth Kuriakose
Support Vector Machine-Based Focused Crawler . . . . . . . . . . . . . . . . 673
Vanshita R. Baweja, Rajesh Bhatia and Manish Kumar
Indoor Object Tracking Using Wi-Fi Access Points . . . . . . . . . . . . . . . 687
Vinit Asher, Hardik Thakkar, Suyog Tambe and Prasenjit Bhavathankar
Crime Analysis and Prediction Using Graph Mining . . . . . . . . . . . . . . 699
A. G. Sreejith, Alan Lansy, K. S. Ananth Krishna, V. J. Haran
and M. Rakhee
A Semi-supervised Approach to Detect Malicious Nodes in OBS
Network Dataset Using Gaussian Mixture Model . . . . . . . . . . . . . . . . . 707
Muhammad Kamrul Hossain Patwary and Md. Mokammel Haque
Apache Spark Methods and Techniques in Big Data—A Review . . . . . 721
H. P. Sahana, M. S. Sanjana, N. Mohammed Muddasir
and K. P. Vidyashree
Artificial Intelligence Applications in Defense . . . . . . . . . . . . . . . . . . . . 727
C. Sinchana, K. Sinchana, Shahid Afridi and M. R. Pooja
Hybrid Approach to Enhance Data Security on Cloud . . . . . . . . . . . . . 735
Pratishtha Saxena, Samarjeet Yadav and Neelam Dayal
Security Threats of Embedded Systems in IoT Environment . . . . . . . . 745
Sreeja Rajendran and R Mary Lourde
Decentralized Bagged Stacking Ensemble Mechanism (DBSEM)
for Anomaly Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 755
S. L. Sanjith and E. George Dharma Prakash Raj
The Process and Application of Dual Extruder Three-Dimension
Printer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 767
Kunal Sekhar Pati and A. Karthikeyan
Contents xix

Face Detection and Natural Language Processing System

Using Artificial Intelligence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 773
H. S. Avani, Ayushi Turkar and C. D. Divya
A Review on Detection of Online Abusive Text . . . . . . . . . . . . . . . . . . 781
Bhumika Jain, Chaithra Bekal and S. P. PavanKumar
Security of an IoT Network: A VLSI Point of View . . . . . . . . . . . . . . . 789
Sreeja Rajendran, Azhar Syed and R. Mary Lourde
Anomaly Detection Techniques in Data Mining—A Review . . . . . . . . . 799
K. N. Lakshmi, N. Neema, N. Mohammed Muddasir and M. V. Prashanth
MAIC: A Proficient Agricultural Monitoring and Alerting System
Using IoT in Cloud Platform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 805
A. Srilakshmi, K. Geetha and D. Harini
Open Set Domain Adaptation for Hyperspectral Image Classification
Using Generative Adversarial Network . . . . . . . . . . . . . . . . . . . . . . . . 819
S. Nirmal, V. Sowmya and K. P. Soman
AHP-FVIKOR Based Access Network Selection in Vehicular
Communications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 829
C. Suganthi Evangeline and Vinoth Babu Kumaravelu
Measuring Web Content Credibility Using Predictive Models . . . . . . . 839
R. Manjula and M. S. Vijaya
Survey on Privacy-Preserving and Other Security Issues
in Data Mining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 849
K. Karthika, R. Devi Priya and S. Sathishkumar
A Study on Different Types of Robotics Applications . . . . . . . . . . . . . . 859
B. U. Meghana and M. V. Prashanth
UVRento—Online Product Rental Application Using Cross-Platform
Service . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 865
Rahul Patil, Siddaram Sonnagi, Ashwini Dhummal, Subhas Kadam
and V. Sanju
Autonomous Unmanned Ground Vehicle for Enhancement
of Defence Strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 873
M. Someshwaran, Deepa Jose and P. Paul Jefferson
A Parametric Study of Load Balancing Techniques in Cloud
Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 881
Insha Naz, Sameena Naaz and Ranjit Biswas
Logical Clustering of Similar Vertices in Complex Real-World
Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 891
Md A. Rahman and Natarajan Meghanathan
xx Contents

Capacity Enhancement Using Delay-Sensitive Protocol

in MANETs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 901
V. Sivakumar, J. Kanimozhi, B. Keerthana and R. Muthulakshmi
SCO-RNN: A Behavioral-Based Intrusion Detection Approach
for Cyber Physical Attacks in SCADA Systems . . . . . . . . . . . . . . . . . . 911
N. Neha, S. Priyanga, Suresh Seshan, R. Senthilnathan
and V. S. Shankar Sriram
SSO-IF: An Outlier Detection Approach for Intrusion Detection
in SCADA Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 921
P. S. Chaithanya, S. Priyanga, S. Pravinraj and V. S. Shankar Sriram
EDF-VD Scheduling-Based Mixed Criticality Cyber-Physical
Systems in Smart City Paradigm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 931
G. Naveen Balaji, M. Sethupathi, N. Sivaramakrishnan and S. Theeijitha
Automatic Solid Waste Dumping Alert System . . . . . . . . . . . . . . . . . . 947
K. Bhuvana, S. Deeksha, M. Deepthi and A. D. Radhika
User Authentication for Mobile Phones by Pseudo Signature
Using CNN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 955
Christy James Jose and M. S. Rajasree
Autonomous Mobile Robot Using RGB-D SLAM and Robot
Operating System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 963
Jash Mota, Prutha Edwankar, Yash Shetye, Manisha Sampat
and Dattatray Sawant
Survey on Data Mining and Predictive Analytics Techniques . . . . . . . 971
S. Sathishkumar, R. Devi Priya and K. Karthika
Implementation of In-flight Entertainment System Using Light
Fidelity Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 983
Viraj Savaliya, Kalp Shah, Dharmil Shah, Henish Shah
and Poonam Kadam
A Survey of Pre-processing Techniques Using Wavelets
and Empirical-Mode Decomposition on Biomedical Signals . . . . . . . . . 993
Prasanth M. Warrier, B. R. Manju and Rajkumar P. Sreedharan
Analysis of Process Scheduling Using Neural Network
in Operating System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1003
Harshit Agarwal and Gaurav Jariwala
An Enhanced Trust Based Fuzzy Implicit Cross-Layer Protocol
for Wireless Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1015
Kompalli Anusha and Ambidi Naveena
Contents xxi

Implementation of Public-Key Infrastructure for Smart Parking

System Using MQTT Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1029
Rajilal Manathala Vijayan, R. Ezhilarasie and A. Umamakeswari
Non-bipolar Evaluation and Visualization of Online Text Reviews . . . . 1039
Keerthana Chigateri and Rekha Bhandarkar
Status Monitoring System-Based Defense Mechanism (SMS-BDM)
for Preventing Co-resident DoS Attacks in Cloud Environment . . . . . . 1047
S. Rethishkumar and R. Vijayakumar
Resource Scheduling Algorithms for Cloud Computing
Environment: A Literature Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1059
V. Arulkumar and N. Bhalaji
Cloud-Based Healthcare Portal in Virtual Private Cloud . . . . . . . . . . . 1071
R. Mahaveerakannan, C. Suresh Gnana Dhas and R. Rama Devi
A Hybrid Approach for Video Steganography by Stretching
the Secret Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1081
B. Karthikeyan, M. M. Anishin Raj, D. Yuvaraj
and K. Joseph Abraham Sundar
An Acknowledgment-Based Approach for Implementing Trusted
Routing Protocol in MANET . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1089
K. Dhanya, C. Jeyalakshmi and A. Balakumar
On-line Frequency Reallocation for Call Requests in Linear Wireless
Cellular Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1097
Narayan Patra
A Novel Secure IoT Based Optimizing Sensor Network for Automatic
Medicine Composition Prescribe System . . . . . . . . . . . . . . . . . . . . . . . 1109
M. Bowya and V. Karthikeyan
High-Speed Polar Decoder Architecture for Next Generation 5G
Applications Using Radix-k Processing Engine . . . . . . . . . . . . . . . . . . . 1119
R. Kavipriya and M. Maheswari
PAPR Reduction in F-OFDM Modulation Scheme for 5G Cellular
Networks Using Precoding Technique . . . . . . . . . . . . . . . . . . . . . . . . . 1127
Sasidharan Jiji and M. Ponmani Raja
Privacy Preservation Using Top K Multi-Keyword Synonym Fuzzy
Search Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1139
Manjubala Sekar and Kamalanathan Kandasamy
High-Speed Variable-Length Decoder Using Segmented Code book . . . 1151
Sujata Bhavikatti and R. M. Banakar
xxii Contents

Efficient Cruise Control of Electric Motorcycle Using PID . . . . . . . . . . 1161

Rohan Sawant and Andrew Kidd
Transmission of Watermarked Image in WSN Using ELSM
Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1171
A. Umapriya and P. Nagarajan
International Students’ Gender Impact to Use Mobile Learning
in Tertiary Education . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1179
Sujit Kumar Basak, Marguerite Wotto and Paul Bélanger
Industrial Automation of Process for Transformer Monitoring
System Using IoT Analytics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1191
Vaishali Khairnar, Likhesh Kolhe, Sudhanshu Bhagat, Ronak Sahu,
Ankit Kumar and Sohail Shaikh
A Comparative Study on Various Search Techniques for Gaming
Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1201
B. Sunil, M. R. Naveen Kumar, B. N. Gowrishankar and N. S. Prema
Crowdsourcing Data Analysis for Crowd Systems . . . . . . . . . . . . . . . . 1207
Vignesh Loganathan, Gopinath Subramani and N. Bhaskar
Optimized Database Using Crowdsourcing . . . . . . . . . . . . . . . . . . . . . . 1217
Vignesh Balakrishnan and N. Bhaskar
Anti-Counterfeit on Medicine Detection Using Blockchain
Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1223
R. Anand, Khadheeja Niyas, Sorjeeta Gupta and S. Revathy
Fault Detection in an Indoor Wireless Sensor Network
Using RSSI-Based Machine Learning Technique . . . . . . . . . . . . . . . . . 1233
R. Pradheepa, M. Bhuvaneshwar, S. Ajay Kumar, B. Ajay Raj
and K. S. Anusha
Image Description Generation Using Deep Learning . . . . . . . . . . . . . . 1239
Neha Supe, Deepti Patil, Revathi Mahadevan, Tanvi Pandhre
and Bharti Joshi
Multi-agent-Based Interference Mitigation Technique in Wireless
Communication System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1245
P. Aruna and A. George
Implementing Machine Learning on Edge Devices with Limited
Working Memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1255
A. Harish, Saksham Jhawar, B. S. Anisha and P. Ramakanth Kumar
Contents xxiii

Smart Transportation: An Edge-Cloud Hybrid Computing

Perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1263
Aashish Jaisimha, Salman Khan, B. S. Anisha and P. Ramakanth Kumar
A Low-Cost Ideal Fish Farm Using IoT: In the Context
of Bangladesh Aquaculture System . . . . . . . . . . . . . . . . . . . . . . . . . . . 1273
Md. Kalim Amzad Chy, Abdul Kadar Muhammad Masum,
Mohammad Emdad Hossain, Md. Golam Rabiul Alam,
Shahidul Islam Khan and Mohammed Shamsul Alam
IoT-Based Smart Monitoring System to Ensure Worksite Safety—A
Context of Garment Industry in Bangladesh . . . . . . . . . . . . . . . . . . . . 1285
Abdul Kadar Muhammad Masum, Ahmed Shan-A-Alahi,
Abdullah Al Noman, Mohammad Nazim Uddin, Khairul Islam Azam
and Mohammed Golam Sarwar Rakib
Evading Gratuitous Energy Consumption Due to Activation
of Superfluous Nodes in WSN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1295
Alok Misra and Divakar Singh Yadav
Detection of Phishing Websites Using Machine Learning . . . . . . . . . . . 1307
Ahmed Raad Abbas, Sukhvir Singh and Mandeep Kau
Smart Drip Irrigation Using IOT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1315
Smita Deshmukh, Swati Chavan, Prajakta Zodge, Pooja Dalvi
and Akshaykumar Jadhav
An Efficient Scheduling Algorithm for Sensor-Based IoT Networks . . . 1323
M. Deva Priya, T. Suganya, A. Christy Jeba Malar, E. Dhivyaprabha,
Prajith Kesava Prasad and L. R. Vishnu Vardhan
Cost-Based Meta-Task Scheduling Algorithm for MultiCloud
Computing (CBMTSA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1333
B. J. Hubert Shanthan and L. Arockiam
Implementation of Low-Cost Mobile Robot for Rescue Challenges . . . 1345
Rajesh Kannan Megalingam, Shree Rajesh Raagul Vadivel,
Prasant Kumar Yadav, Katta Nigam, Ravi Teja Geesala
and Ruthvik Chanda
Smog Detection and Pollution Alert System Using Wireless Sensor
Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1357
Manita Rajput, Mrudula Geddam, Priyanka Jondhale and Priyanka Sawant
IoT-Based Women Security System . . . . . . . . . . . . . . . . . . . . . . . . . . . 1369
Megalingam Rajesh Kannan, K. Jyothsna, T. S. Aparna, T. Anjali,
M. Meera and S. D. Amrutha
xxiv Contents

IoT-based Wearable Micro-Strip Patch Antenna with

Electromagnetic Band Gap Structure for Gain Enhancement . . . . . . . 1379
M. Ameena Banu, R. Tamilselvi, M. Rajalakshmi
and M. Pooja Lakshmi
Design and Implementation of a 3 DOF Arm for Disaster
Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1397
Rajesh Kannan Megalingam, Shree Rajesh Raagul Vadivel, Vamsi Gontu,
Deepak Nagalla, Ravi Kiran Pasumarthi and Phanindra Kumar Allada

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1411

Editors and Contributors

About the Editors

Dr. G. Ranganathan has been a Professor at Gnanamani College of Technology

since 2017. Previously, he has worked at Ranganathan Engineering College and
Jayalakshmi Institute of Technology. He has also worked as a Professor at RVS
Technical Campus, Karpagam College of Engineering and P.R. Engineering College.
He completed his Ph.D. and M.E. at the Information and Communication Engineering
Department, Anna University. His areas of interest include biomedical signal pro-
cessing, neural networks, fuzzy logic, networking and VLSI signal processing. He has
published five books and numerous research papers in international journals.

Dr. Joy Chen received his [Link]. in Electronics Engineering from the National
Taiwan Technical University, Taipei, Taiwan, and his [Link]. in Electrical Engineering
from Dayeh University, Changhua, Taiwan, in 1985 and 1995, respectively. He
completed his Ph.D. in Electrical Engineering at the National Defense University,
Tao-Yuan, Taiwan, in 2001 and is currently a Full Professor at the Department of
Communication Engineering, Dayeh University. Prior to joining Dayeh, he worked at
the Control Data Company (Taiwan branch) as a Technical Manager from 1985 to
1996. His research interests include wireless communications, MIMO systems,
OFDM systems, and wireless sensor networks. Dr. Joy I.-Z. Chen has published
numerous SCI/EI journal papers and holds several patents in Taiwan.

Dr. Álvaro Rocha holds a Habilitation (postdoc) in Information Science, Ph.D. in

Information Systems and Technologies, and [Link]. in Information Management. He is
currently a Professor of Information Systems and Software Engineering at the
University of Coimbra, President of AISTI (the Iberian Association for Information
Systems and Technologies), and Chair of IEEE SMC Portugal Section Society
Chapter. He served as Vice-Chair of Experts for the European Commission’s Horizon
2020 Programme, and as an expert at the Italian Ministry of Education, University
and Research. His main research interests are information systems planning and

xxv
xxvi Editors and Contributors

management, maturity models, information system quality, online service quality,

intelligent information systems, software engineering, e-government, e-health, and IT
in education.

Contributors

Ahmed Raad Abbas Department of Information Technology Development (IT),

University Institute of Engineering and Technology (UIET), Punjab University,
Chandigarh, India
S. Abirami Department of Electronics and Instrumentation Engineering,
Annamalai University, Chidambaram, India
Kinjal Adhvaryu Computer Engineering Department, Sankersinh Vaghela Bapu
Institute of Technology, Gandhinagar, Gujarat, India
Shahid Afridi Department of Computer Science and Engineering, Vidyavardhaka
College of Engineering, Mysuru, Karnataka, India
Vimal Agarwal Apex Institute of Engineering & Technology, Jaipur, India
Harshit Agarwal Sarvajanik College of Engineering and Technology, Surat, India
Rekib Uddin Ahmed National Institute of Technology Meghalaya, Shillong,
India
S. Ajay Kumar Department of Electronics and Communication Engineering,
Amrita School of Engineering, Amrita Vishwa Vidyapeetham, Coimbatore, India
B. Ajay Raj Department of Electronics and Communication Engineering, Amrita
School of Engineering, Amrita Vishwa Vidyapeetham, Coimbatore, India
Aparna Ajith Department of Electronics and Communication Engineering,
Amrita Vishwa Vidyapeetham, Amritapuri, Kerala, India
Suthapalli Akhil Sri Harsha School of Electrical and Electronics Engineering,
SASTRA Deemed to Be University, Thanjavur, India
Abdullah Al Noman Department of Computer Science and Engineering,
International Islamic University Chittagong, Chittagong, Bangladesh
Mohammed Shamsul Alam Department of Computer Science and Engineering,
International Islamic University Chittagong, Chittagong, Bangladesh
Phanindra Kumar Allada Department of Electronics and Communication
Engineering, Amrita Vishwa Vidyapeetham, Amritapuri, India
Ali Jameel Al-Mousawi Information Technology Regulation Directory—
Communication and Media Commission of Iraq (CMC), Baghdad, Iraq
Editors and Contributors xxvii

M. Ameena Banu Department of Electronics and Communication Engineering,

Sethu Institute of Technology, Pulloor, Virudhunagar District, Tamilnadu, India
S. D. Amrutha Department of Electronics and Communication Engineering,
Amrita Vishwa Vidyapeetham, Amritapuri, India
R. Anand Department of Information Technology, KCG College of Technology,
Chennai, India
R. Ani Department of Computer Science and Applications, Amrita Vishwa
Vidyapeetham, Kollam, India
B. S. Anisha Information Science and Engineering, R.V. College of Engineering,
Bengaluru, India
P. Anitha Dr. N.G.P. Institute of Technology, Coimbatore, India
T. Anjali Department of Electronics and Communication Engineering, Amrita
Vishwa Vidyapeetham, Amritapuri, India
K. S. Anusha Department of Electronics and Communication Engineering, Amrita
School of Engineering, Amrita Vishwa Vidyapeetham, Coimbatore, India
Kompalli Anusha Branch of WMC, G. Narayanamma Institute of Technology
and Science, JNTUH, Hyderabad, India
T. S. Aparna Department of Electronics and Communication Engineering, Amrita
Vishwa Vidyapeetham, Amritapuri, India
L. Arockiam Department of Computer Science, St. Joseph’s College, Trichy,
India
M. Arulaalan CK College of Engineering & Technology, Cuddalore, India
V. Arulkumar Department of Information Technology, SSN College of
Engineering, Chennai, India
P. Aruna Department of Software Engineering, Periyar Maniammai Institute of
Science and Technology, Thanjavur, India
Vinit Asher Sardar Patel Institute of Technology, Mumbai, India
N. Ashokkumar Department of Electronics and Communication Engineering,
Centre for VLSI and Embedded Systems, Sree Vidyanikethan Engineering College,
Tirupati, India
Sigamani Ashokkumar Department of Computer Science and Engineering,
Saveetha School of Engineering, SIMATS, Chennai, India
A. V. Aswin Muthoot Institute of Technology and Science, Ernakulam, India
H. S. Avani Department of Computer Science and Engineering, Vidyavardhaka
College of Engineering, Mysuru, Karnataka, India
xxviii Editors and Contributors

Khairul Islam Azam Department of Computer Science and Engineering,

International Islamic University Chittagong, Chittagong, Bangladesh
S. Balaji Department of Electronics and Communications Engineering, Amrita
Vishwa Vidyapeetham, Amritapuri, India
Vignesh Balakrishnan Department of Information Technology, KCG College of
Technology, Chennai, India
A. Balakumar K. Ramakrishnan College of Engineering, Trichy, India
Ananthakrishnan BalaSundaram Department of Computer Science and
Engineering, Saveetha School of Engineering, SIMATS, Chennai, India
R. M. Banakar BVB College of Engineering and Technology, Hubli, India
Amit Barve Ramrao Adik Institute of Technology, Navi Mumbai, India
Sujit Kumar Basak Université du Québec à Montréal (UQÀM), Montreal,
Canada
K. Y. Basil Muthoot Institute of Technology and Science, Ernakulam, India
Vanshita R. Baweja Computer Science, PEC University of Technology,
Chandigarh, India
Sushant Bawiskar Department of Embedded Technology, VIT, Vellore, India
Chaithra Bekal Department of Computer Science and Engineering,
Vidyavardhaka College of Engineering, Mysuru, Karnataka, India
Paul Bélanger Université du Québec à Montréal (UQÀM), Montreal, Canada
Sudhanshu Bhagat Information Technology, Terna Engineering College, Navi
Mumbai, India
R. Bhairavi Department of Electronics and Communication Engineering,
Pondicherry Engineering College, Puducherry, India
N. Bhalaji Department of Information Technology, SSN College of Engineering,
Chennai, India
Rajiv R. Bhandari Department of Computer Science Engineering, Koneru
Lakshmaiah Education Foundation Green Fields, Guntur District, Vaddeswaram,
Andhra Pradesh, India
Rekha Bhandarkar Department of ECE, N.M.A.M.I.T, Nitte, India
Madiraju Akshay Bharadwaj Department of Electronics and Communication
Engineering, Amrita School of Engineering, Kollam, Kerala, India
S. Bharath Departtment of Electronics and Communication Engineering, Amrita
Vishwa Vidyapeetham, Amritapuri, India
Editors and Contributors xxix

N. Bhaskar Department of Information Technology, KCG College of

Technology, Chennai, India
Rajesh Bhatia Computer Science, PEC University of Technology, Chandigarh,
India
Prasenjit Bhavathankar Sardar Patel Institute of Technology, Mumbai, India
Sujata Bhavikatti Tontadarya College of Engineering, Gadag, India
K. Bhuvana Department of Computer Science and Engineering, Vidyavardhaka
College of Engineering, Mysuru, Karnataka, India
M. Bhuvaneshwar Department of Electronics and Communication Engineering,
Amrita School of Engineering, Amrita Vishwa Vidyapeetham, Coimbatore, India
Bhawani Sankar Biswal DST-FIST Bioinformatics Laboratory, IIIT
Bhubaneswar, Bhubaneswar, India
Ranjit Biswas Department of Computer Science and Engineering, Jamia
Hamdard, New Delhi, India
Tanmay Borade Ramrao Adik Institute of Technology, Navi Mumbai, India
M. Bowya Department of ECE, Vivekanandha College of Engineering for
Women, Namakkal, India
P. S. Chaithanya School of Computing, Centre for Information Super Highway
(CISH), SASTRA Deemed to be University, Thanjavur, Tamil Nadu, India
Ruthvik Chanda Department of Electronics and Communication Engineering,
Amrita Vishwa Vidyapeetham, Amritapuri, India
K. Chandrasekaran Department of Computer Science and Engineering, National
Institute of Technology Karnataka, Surathkal, Mangalore, India
G. Charles Babu Department of CSE, Malla Reddy Engineering College
(Autonomous), Secunderabad, Telangana, India
Swati Chavan IT Department, Terna Engineering College, Mumbai University,
Navi Mumbai, Maharashtra, India
Amruta Chavan Department of IT Engineering, RAIT, Nerul, Navi Mumbai,
India
Keerthana Chigateri Department of CSE, N.M.A.M.I.T, Nitte, India
A. Christy Jeba Malar Sri Krishna College of Technology, Coimbatore, Tamil
Nadu, India
Dhananjay Dakhane Ramrao Adik Institute of Technology, Navi Mumbai, India
Pooja Dalvi IT Department, Terna Engineering College, Mumbai University, Navi
Mumbai, Maharashtra, India
xxx Editors and Contributors

Neelam Dayal Department of Computer Science & Engineering, Centre for

Advanced Studies, AKTU, Lucknow, India
Tribid Debbarma Department of Computer Science and Engineering, National
Institute of Technology Agartala, Agartala, India
S. Deeksha Department of Computer Science and Engineering, Vidyavardhaka
College of Engineering, Mysuru, Karnataka, India
M. Deepthi Department of Computer Science and Engineering, Vidyavardhaka
College of Engineering, Mysuru, Karnataka, India
Smita Deshmukh IT Department, Terna Engineering College, Mumbai
University, Navi Mumbai, Maharashtra, India
Santhosh L. Deshpande Department of Post Graduate Studies, Visvesvaraya
Technological University (VTU), Belgaum, Karnataka, India
M. Deva Priya Sri Krishna College of Technology, Coimbatore, Tamil Nadu,
India
R. Devi Priya Department of Information Technology, Kongu Engineering
College, Erode, India
K. Dhanya K. Ramakrishnan College of Engineering, Trichy, India
E. Dhivyaprabha Sri Krishna College of Technology, Coimbatore, Tamil Nadu,
India
J. Dhoulath Beegum Department of ECE, TKM College of Engineering, Kollam,
Kerala, India
Ashwini Dhummal Department of Computer Science and Engineering,
NMAMIT, Nitte, Karnataka, India
C. D. Divya Department of Computer Science and Engineering, Vidyavardhaka
College of Engineering, Mysuru, Karnataka, India
Nilima Dongre Department of IT Engineering, RAIT, Nerul, Navi Mumbai, India
Prutha Edwankar Department of Mechatronics Engineering, MPSTME, NMIMS
University, Mumbai, India
Vijaya Eligar KLE Technological University, Hubballi, India
Sanjay Eligar BVB College of Engineering and Technology, Hubli, India
B. Maragatha Eswari Department of Electronics and Communication
Engineering, Amrita School of Engineering, Kollam, Kerala, India
R. Eswari National Institute of Technology, Tiruchirappalli, India
Editors and Contributors xxxi

C. Suganthi Evangeline Karunya Institute of Technology and Sciences,

Coimbatore, India;
Vellore Institute of Technology, Vellore, India
R. Ezhilarasie School of Computing, SASTRA Deemed University, Thanjavur,
India
K. Ben Franklin Department of Electronics and Communication Engineering,
SRMIST, Kattankulathur, Chennai, India
Roopali Garg Associate Professor, UIET, Panjab University, Chandigarh, India
Mrudula Geddam Department of Electronics and Telecommunication
Engineering, Fr. C.R.I.T, Vashi, Navi, Mumbai, India
Ravi Teja Geesala Department of Electronics and Communication Engineering,
Amrita Vishwa Vidyapeetham, Amritapuri, India
K. Geetha SASTRA Deemed to be University, Thanjavur, India
A. George Department of Mathematics, Periyar Maniammai Institute of Science
and Technology, Thanjavur, India
E. George Dharma Prakash Raj Bharathidasan University, Tiruchirappalli,
Tamilnadu, India
Jatin Gharat Ramrao Adik Institute of Technology, Navi Mumbai, India
Tushar Ghorpade Ramrao Adik Institute of Technology, Navi Mumbai, India
Md. Golam Rabiul Alam Department of Computer Science and Engineering,
BRAC University, Dhaka, Bangladesh
Vamsi Gontu Department of Electronics and Communication Engineering,
Amrita Vishwa Vidyapeetham, Amritapuri, India
Nivetha Gopalakrishnan Department of Electronics and Communication
Engineering, University College of Engineering Panruti, Panruti, Tamilnadu, India
Vinodhini Gopalakrishnan Department of Computer Science Engineering,
Annamalai University, Chidambaram, Tamilnadu, India
Priyanka Gotter Computer Engineering and Technology, Guru Nanak Dev
University, Amritsar, Punjab, India
B. N. Gowrishankar Department of Information Science and Engineering,
Vidyavardhaka College of Engineering, Mysuru, Karnataka, India
Aman Gupta Department of C.S.E, Guru Ghasidas Central University, Bilaspur,
India
Saurabh Gupta Department of C.S.E, Guru Ghasidas Central University,
Bilaspur, India
xxxii Editors and Contributors

Sorjeeta Gupta Department of Information Technology, KCG College of

Technology, Chennai, India
V. J. Haran Muthoot Institute of Technology and Science, Kochi, India
D. Harini SASTRA Deemed to be University, Thanjavur, India
A. Harish Information Science and Engineering, R.V. College of Engineering,
Bengaluru, India
Abhilash P. Hasankar Department of E&CE, SDMCET, Dharwad, Karnataka,
India
Haider K. Hoomod Computer Science Department, College of Education—
Mustansiriyah University, Baghdad, Iraq
Mohammad Emdad Hossain Department of Business Administration,
International Islamic University Chittagong, Chittagong, Bangladesh
B. J. Hubert Shanthan Department of Computer Science, St. Joseph’s College,
Trichy, India
Shivaraj Hublikar KLE Technological University, Hubballi, India
V. Indu Department of Electrical and Electronics Engineering, Amrita Vishwa
Vidyapeetham, Amritapuri, India
Akshaykumar Jadhav IT Department, Terna Engineering College, Mumbai
University, Navi Mumbai, Maharashtra, India
Dipti Jadhav Department of IT Engineering, RAIT, Nerul, Navi Mumbai, India
Sandeep Jadhav KPIT Technologies Limited, Pune, Maharashtra, India
Ashish Jain Apex Institute of Engineering & Technology, Jaipur, India
Bhumika Jain Department of Computer Science and Engineering, Vidyavardhaka
College of Engineering, Mysuru, Karnataka, India
Aashish Jaisimha Information Science and Engineering, R.V. College of
Engineering, Bengaluru, India
B. Janet National Institute of Technology, Tiruchirappalli, India
Gaurav Jariwala Sarvajanik College of Engineering and Technology, Surat, India
N. Jayapandian Department of Computer Science and Engineering, CHRIST
(Deemed to Be University), Bangalore, India
C. Jeyalakshmi K. Ramakrishnan College of Engineering, Trichy, India
Saksham Jhawar Information Science and Engineering, R.V. College of
Engineering, Bengaluru, India
Editors and Contributors xxxiii

Priyanka Jondhale Department of Electronics and Telecommunication

Engineering, Fr. C.R.I.T, Vashi, Navi, Mumbai, India
Christy James Jose Government Engineering College, Barton Hill,
Thiruvananthapuram, Kerala, India
Deepa Jose Electronics and Communication, KCG College of Technology,
Chennai, India
K. Joseph Abraham Sundar Computer Vision & Soft Computing Lab, School of
Computing, SASTRA Deemed University, Thanjavur, India
Bharti Joshi Ramrao Adik Institute of Technology, Navi Mumbai, India
K. Jyothsna Department of Electronics and Communication Engineering, Amrita
Vishwa Vidyapeetham, Amritapuri, India
Poonam Kadam Department of Electronics and Telecommunication, Dwarkadas
J. Sanghvi College of Engineering, Mumbai, India
Subhas Kadam Department of Computer Science and Engineering, NMAMIT,
Nitte, Karnataka, India
Arun Kakhandki KLSs VDRIT, Haliyal, India
Md. Kalim Amzad Chy Department of Computer Science and Engineering,
International Islamic University Chittagong, Chittagong, Bangladesh
G. Kanagaraj Kumaraguru College of Technology, Coimbatore, India
Kamalanathan Kandasamy Amrita Center for Cyber Security Systems &
Networks, Amrita Vishwa Vidyapeetham, Amrita University, Kollam, Kerala, India
J. Kanimozhi Department of IT, Adhiyamaan College of Engineering, Hosur,
Tamilnadu, India
A. Kannammal Department of Computer Science and Engineering, Jayam
College of Engineering and Technology, Dharmapuri, TamilNadu, India
V. Karthick Department of Computer Science and Engineering, Saveetha School
of Engineering, Saveetha Institute of Medical and Technical Sciences, Chennai,
India
K. Karthika Department of Information Technology, Adhiyamaan College of
Engineering, Hosur, India
A. Karthikeyan School of Electronics Engg, VIT, Vellore, Tamil Nadu, India;
Department of Embedded System Technologies, VIT, Vellore, India
B. Karthikeyan Computer Vision & Soft Computing Lab, School of Computing,
SASTRA Deemed University, Thanjavur, India
V. Karthikeyan Department of ECE, Vivekanandha College of Engineering for
Women, Namakkal, India
xxxiv Editors and Contributors

Jaspreet Kaur Ramrao Adik Institute of Technology, Navi Mumbai, India

Kiranbir Kaur GNDU (Main Campus), Amritsar, India;
Department of Computer Engineering and Technology, Guru Nanak Dev
University, Amritsar, Punjab, India
Mandeep Kau Department of Information Technology Development (IT),
University Institute of Engineering and Technology (UIET), Punjab University,
Chandigarh, India
Manjot Kaur GNDU (Main Campus), Amritsar, India
Preetjot Kaur PhD Research Scholar, UIET, Panjab University, Chandigarh,
India
Tanveer Kaur Department of Computer Engineering and Technology, Guru
Nanak Dev University, Amritsar, Punjab, India
R. Kavipriya K. Ramakrishnan College of Engineering, Tiruchirappalli, India
B. Keerthana Department of IT, Adhiyamaan College of Engineering, Hosur,
Tamilnadu, India
Vaishali Khairnar Information Technology, Terna Engineering College, Navi
Mumbai, India
Salman Khan Information Science and Engineering, R.V. College of
Engineering, Bengaluru, India
Shahidul Islam Khan Department of Computer Science and Engineering,
International Islamic University Chittagong, Chittagong, Bangladesh
Andrew Kidd Teesside University, Middlesbrough, UK
Akeem T. L. King ISPA Technology, Tampa, FL, USA
Steward Kirubakaran Department of Computer Science and Engineering, KCG
College of Technology, Karapakkam, Chennai, India
Likhesh Kolhe Information Technology, Terna Engineering College, Navi
Mumbai, India
K. S. Ananth Krishna Muthoot Institute of Technology and Science, Kochi, India
Arya Krishnan Department of Electronics and Communication Engineering,
Amrita Vishwa Vidyapeetham, Amritapuri, Kerala, India
Venkatalakshmi Krishnan Department of Electronics and Communication
Engineering, University College of Engineering Tindivanam, Tindivanam,
Tamilnadu, India
P. Anand Krisshna Department of Electronics and Communication Engineering,
Amrita Vishwa Vidyapeetham, Amritapuri, India
Editors and Contributors xxxv

Vidya Kubde Department of Information Technology, Datta Meghe College of

Engineering, Navi Mumbai, India
Ankit Kumar Information Technology, Terna Engineering College, Navi
Mumbai, India
Manish Kumar Computer Science, PEC University of Technology, Chandigarh,
India
R. Kumar Department of Electronics and Communication Engineering, SRMIST,
Kattankulathur, Chennai, India
Rakesh Kumar Department of Computer Science and Engineering (CSE), M.M.
M. University of Technology, Gorakhpur, U.P., India
Vinoth Babu Kumaravelu Vellore Institute of Technology, Vellore, India
Anshu Kumari Department of Computer Science & Engineering, Guru Ghasidas
University, Bilaspur, India
Ashwini S. Kunte Department of Electronics and Telecommunication, Thadomal
Shahani Engineering College, Bandra, Mumbai, India
Bineeth Kuriakose Muthoot Institute of Technology and Science, Ernakulam,
India;
Oslo Metropolitan University, Oslo, Norway
K. N. Lakshmi Department of Information Science and Engineering,
Vidyavardhaka College of Engineering, Mysuru, Karnataka, India
Alan Lansy Muthoot Institute of Technology and Science, Kochi, India
Vignesh Loganathan Department of Information Technology, KCG College of
Technology, Chennai, India
R. Mary Lourde Department of EEE, BITS Pilani Dubai Campus, Dubai, UAE
Revathi Mahadevan Ramrao Adik Institute of Technology, Navi Mumbai,
Maharashtra, India
R. Mahaveerakannan Information Technology St. Peter’s University, Chennai,
India
K. Mahendran CK College of Engineering & Technology, Cuddalore, India
M. Maheswari K. Ramakrishnan College of Engineering, Tiruchirappalli, India
G. Manikannan CK College of Engineering & Technology, Cuddalore, India
B. R. Manju Department of Mathematics, Amrita School of Engineering, Amrita
Vishwa Vidyapeetham, Amritapuri, India
R. Manjula PSGR Krishnammal College for Women, Coimbatore, India
Nilesh Marathe Department of IT Engineering, RAIT, Nerul, Navi Mumbai, India
xxxvi Editors and Contributors

R. Mary Lourde Department of EEE, BITS Pilani, Dubai Campus, Dubai, UAE
Abdul Kadar Muhammad Masum Department of Computer Science and
Engineering, International Islamic University Chittagong, Chittagong, Bangladesh
M. Meera Department of Electronics and Communication Engineering, Amrita
Vishwa Vidyapeetham, Amritapuri, India
Rajesh Kannan Megalingam Department of Electronics and Communication
Engineering, Amrita Vishwa Vidyapeetham, Amritapuri, India
B. U. Meghana Department of Information Science and Engineering,
Vidyavardhaka College of Engineering, Mysuru, Karnataka, India
Natarajan Meghanathan Jackson State University, Jackson, MS, USA;
Computer Science, Jackson State University, Jackson, MS, USA
Alok Misra Dr. A.P.J. Abdul Kalam Technical University, Lucknow, India
N. Mohammed Muddasir Department of Information Science and Engineering,
Vidyavardhaka College of Engineering, Mysuru, Karnataka, India
Poornima Mohan Department of Electronics and Communication Engineering,
Amrita School of Engineering, Kollam, Kerala, India
Anjali Mohapatra DST-FIST Bioinformatics Laboratory, IIIT Bhubaneswar,
Bhubaneswar, India
Shubhankar Mohapatra DST-FIST Bioinformatics Laboratory, IIIT
Bhubaneswar, Bhubaneswar, India
Md. Mokammel Haque Department of Computer Science and Engineering,
Chittagong University of Engineering and Technology, Raozan, Chittagong,
Bangladesh
Jash Mota Department of Mechatronics Engineering, MPSTME, NMIMS
University, Mumbai, India
R. Muthulakshmi Department of IT, Adhiyamaan College of Engineering, Hosur,
Tamilnadu, India
Sameena Naaz Department of Computer Science and Engineering, Jamia
Hamdard, New Delhi, India
Deepak Nagalla Department of Electronics and Communication Engineering,
Amrita Vishwa Vidyapeetham, Amritapuri, India
P. Nagarajan Department of Electronics and Communication Engineering, Centre
for VLSI and Embedded Systems, Sree Vidyanikethan Engineering College,
Tirupati, India;
Department of ECE, Vivekanandha College of Engineering for Women, Namakkal,
India
Editors and Contributors xxxvii

Jolan Rokan Naif Informatics Institute for Postgraduate Studies-Baghdad,

Baghdad, Iraq
Aswathy K. Nair Department of Electronics and Communications Engineering,
Amrita Vishwa Vidyapeetham, Amritapuri, India
K. Namitha Department of Electronics and Communication Engineering, Amrita
School of Engineering, Kollam, Kerala, India
Gayathri Narayanan Department of Electronics and Communication
Engineering, Amrita Vishwa Vidyapeetham, Amritapuri, Kerala, India
P. S. Narayanan Department of Computer Science and Applications, Amrita
Vishwa Vidyapeetham, Kollam, India
G. Naveen Balaji Department of ECE, SNS College of Technology, Coimbatore,
India
M. R. Naveen Kumar Department of Information Science and Engineering,
Vidyavardhaka College of Engineering, Mysuru, Karnataka, India
Ambidi Naveena Department of ETM, G. Narayanamma Institute of Technology
and Science, JNTUH, Hyderabad, India
C. Nayak Department of Electronics and Communication Engineering, SRMIST,
Kattankulathur, Chennai, India
Insha Naz Department of Computer Science and Engineering, Jamia Hamdard,
New Delhi, India
N. Neema Department of Information Science and Engineering, Vidyavardhaka
College of Engineering, Mysuru, Karnataka, India
N. Neha School of Computing, Centre for Information Super Highway (CISH),
SASTRA Deemed to be University, Thanjavur, Tamil Nadu, India
Katta Nigam Department of Electronics and Communication Engineering, Amrita
Vishwa Vidyapeetham, Amritapuri, India
S. Nirmal Center for Computational Engineering and Networking (CEN), Amrita
School of Engineering, Amrita Vishwa Vidyapeetham, Coimbatore, India
N. Nithya Department of Computer Science and Engineering, K. Ramakrishnan
College of Engineering, Trichy, India
Khadheeja Niyas Department of Information Technology, KCG College of
Technology, Chennai, India
S. Palaniappan Department of Computer Science and Engineering, Saveetha
School of Engineering, Saveetha Institute of Medical and Technical Sciences,
Chennai, India;
Department of Computer Science and Engineering, KCG College of Technology,
Karapakkam, Chennai, India
xxxviii Editors and Contributors

Tanvi Pandhre Ramrao Adik Institute of Technology, Navi Mumbai,

Maharashtra, India
Vignesh Parameswaran M. Tech. Embedded Systems, VIT, Vellore, India
Navrattan Parmar National Institute of Technology, Kurukshetra, Haryana, India
V. Parthipan Department of Computer Science and Engineering, KCG College of
Technology, Karapakkam, Chennai, India
Ravi Kiran Pasumarthi Department of Electronics and Communication
Engineering, Amrita Vishwa Vidyapeetham, Amritapuri, India
Kunal Sekhar Pati Department of Embedded Technology, School of Electronics
Engineering, VIT University, Vellore, India
Deepti Patil Ramrao Adik Institute of Technology, Navi Mumbai, Maharashtra,
India
Pratik Patil [Link] Embedded Systems, VIT, Vellore, Tamil Nadu, India
Rahul Patil Department of Computer Science and Engineering, NMAMIT, Nitte,
Karnataka, India
Rohan Patil Automation Department, Jain Irrigation Systems Ltd., Jalgaon, India
Narayan Patra Department of Computer Science and Engineering, Siksha O
Anusandhan Deemed to be University, Bhubaneswar, India
Muhammad Kamrul Hossain Patwary Department of Computer Science and
Engineering, Chittagong University of Engineering and Technology, Raozan,
Chittagong, Bangladesh
P. Paul Jefferson Electronics and Communication, KCG College of Technology,
Chennai, India
S. P. PavanKumar Department of Computer Science and Engineering,
Vidyavardhaka College of Engineering, Mysuru, Karnataka, India
M. Ponmani Raja Electronics and Communication Engineering, Jyothi
Engineering College, Thrissur, India
M. R. Pooja Department of Computer Science and Engineering, Vidyavardhaka
College of Engineering, Mysuru, Karnataka, India
M. Pooja Lakshmi Department of Electronics and Communication Engineering,
Sethu Institute of Technology, Pulloor, Virudhunagar District, Tamilnadu, India
P. Prabakaran CK College of Engineering & Technology, Cuddalore, India
R. Pradeep Department of Computer Science and Engineering, Siddaganga
Institute of Technology, Tumkur, Karnataka, India
Editors and Contributors xxxix

R. Pradheepa Department of Electronics and Communication Engineering,

Amrita School of Engineering, Amrita Vishwa Vidyapeetham, Coimbatore, India
C. Prajitha Department of Biomedical Engineering, Saveetha School of
Engineering, Saveetha Institute of Medical and Technical Sciences, Chennai, India
B. Bhanu Prakash Department of Electronics and Communications Engineering,
Amrita Vishwa Vidyapeetham, Amritapuri, India
S. Pramod Department of Mechanical Engineering, Amrita Vishwa Vidyapeetham,
Amritapuri, India
Pranesh VTU-RRC, Visvesvaraya Technological University (VTU), Belgaum,
Karnataka, India
Prajith Kesava Prasad Sri Krishna College of Technology, Coimbatore, Tamil
Nadu, India
M. V. Prashanth Department of Information Science and Engineering,
Vidyavardhaka College of Engineering, Mysuru, Karnataka, India
S. Pravinraj School of Computing, Centre for Information Super Highway
(CISH), SASTRA Deemed to be University, Thanjavur, Tamil Nadu, India
N. S. Prema Department of Information Science and Engineering, Vidyavardhaka
College of Engineering, Mysuru, Karnataka, India
T. Primya Dr. N.G.P. Institute of Technology, Coimbatore, India
D. Priya Department of Information Science, RV College of Engineering®,
Bangalore, India
S. Priyanga School of Computing, Centre for Information Super Highway
(CISH), SASTRA Deemed to be University, Thanjavur, Tamil Nadu, India
Kuppili Puneeth Department of Electrical and Electronics Engineering, Amrita
Vishwa Vidyapeetham, Amritapuri, India
A. D. Radhika Department of Computer Science and Engineering, Vidyavardhaka
College of Engineering, Mysuru, Karnataka, India
Md A. Rahman Computer Science, Jackson State University, Jackson, MS, USA
M. M. Anishin Raj Department of CSE, Viswajyothi College of Engineering &
Technology, Vazhakulam, Kerala, India
Amrita Raj Department of Computer Science and Engineering (CSE), M.M.M.
University of Technology, Gorakhpur, U.P., India
M. Rajalakshmi Department of Electronics and Communication Engineering,
Sethu Institute of Technology, Pulloor, Virudhunagar District, Tamilnadu, India
xl Editors and Contributors

K. Rajasekhar Department of Computer Science Engineering, Koneru

Lakshmaiah Education Foundation Green Fields, Guntur District, Vaddeswaram,
Andhra Pradesh, India
M. Rajasekhar Reddy School of Computing, SASTRA Deemed to Be
University, Thanjavur, India
M. S. Rajasree APJ Abdul Kalam Technological University, Thiruvananthapuram,
Kerala, India
Sreeja Rajendran Department of EEE, BITS Pilani, Dubai Campus, Dubai, UAE
Megalingam Rajesh Kannan Department of Electronics and Communication
Engineering, Amrita Vishwa Vidyapeetham, Amritapuri, India
Manita Rajput Department of Electronics and Telecommunication Engineering,
Fr. C.R.I.T, Vashi, Navi, Mumbai, India
M. Rakhee Muthoot Institute of Technology and Science, Kochi, India
Mohammed Golam Sarwar Rakib Department of Computer Science and
Engineering, International Islamic University Chittagong, Chittagong, Bangladesh
R. Rama Devi Janson Institute of Technology, Coimbatore, India
P. Ramakanth Kumar Department of Computer Science and Engineering, R.V.
College of Engineering, Bengaluru, India
Virender Ranga National Institute of Technology, Kurukshetra, Haryana, India
Krupa Rasane E&C Department, Jain College of Engineering, Belgavi, India
V. Ravi Department of Computer Science and Engineering, Siddaganga Institute
of Technology, Tumkur, Karnataka, India
O. Rajasekhar Reddy Department of Electronics and Communication
Engineering, Amrita School of Engineering, Kollam, Kerala, India
Basil Reji Muthoot Institute of Technology and Science, Ernakulam, India
S. Rethishkumar School of Computer Sciences, Mahatma Gandhi University,
Kottayam, Kerala, India
S. Revathy Department of Information Technology, KCG College of Technology,
Chennai, India
S. Rinesh Department of Computer Science and Engineering, Saveetha School of
Engineering, Saveetha Institute of Medical and Technical Sciences, Chennai, India;
Department of Computer Science and Engineering, KCG College of Technology,
Karapakkam, Chennai, India
Narjala Rohith Department of Electrical and Electronics Engineering, Amrita
Vishwa Vidyapeetham, Amritapuri, India
Editors and Contributors xli

G. R. Sagar Department of Computer Science and Engineering, CHRIST

(Deemed to Be University), Bangalore, India
Prabir Saha National Institute of Technology Meghalaya, Shillong, India
H. P. Sahana Department of Information Science and Engineering, Vidyavardhaka
College of Engineering, Mysuru, Karnataka, India
Ronak Sahu Information Technology, Terna Engineering College, Navi Mumbai,
India
A. Sai Hanuman Department of CSE, Gokaraju Rangaraju Institute of
Engineering & Technology (Autonomous), Bachupally, Telangana, India
P. Sai Siddartha Reddy School of Computing, SASTRA Deemed to Be
University, Thanjavur, India
Manisha Sampat Department of Mechatronics Engineering, MPSTME, NMIMS
University, Mumbai, India
Kambhampati Sai Sandilya Department of Electronics and Communication
Engineering, Amrita School of Engineering, Kollam, Kerala, India
T. R. Sangeetha Department of Electronics and Communication Engineering,
Amrita Vishwa Vidyapeetham, Amritapuri, Kerala, India
M. S. Sanjana Department of Information Science and Engineering,
Vidyavardhaka College of Engineering, Mysuru, Karnataka, India
Kumar Sanjeev Centre for Development of Advanced Computing (C-DAC),
Mohali, India
S. L. Sanjith Indian Institute of Management Tiruchirappalli, Tiruchirappalli,
Tamilnadu, India
V. Sanju Department of Computer Science and Engineering, NMAMIT, Nitte,
Karnataka, India
B. Sankara Babu Department of CSE, Gokaraju Rangaraju Institute of
Engineering & Technology (Autonomous), Bachupally, Telangana, India
J. Vaishnu Saran Department of Electronics and Communication Engineering,
Amrita School of Engineering, Kollam, Kerala, India
Moumita Sarkar DST-FIST Bioinformatics Laboratory, IIIT Bhubaneswar,
Bhubaneswar, India
K. Sashi Rekha Dr. N.G.P. Institute of Technology, Coimbatore, India
J. Sasi Kiran Department of CSE, Farah Institute of Technology, Chevella,
Telangana, India
Sasidharan Jiji Electronics and Communication Engineering, Jyothi Engineering
College, Thrissur, India
xlii Editors and Contributors

S. Sathishkumar Department of Information Technology, Adhiyamaan College of

Engineering, Hosur, India
Viraj Savaliya Department of Electronics and Telecommunication, Dwarkadas
J. Sanghvi College of Engineering, Mumbai, India
Dattatray Sawant Department of Mechatronics Engineering, MPSTME, NMIMS
University, Mumbai, India
Priyanka Sawant Department of Electronics and Telecommunication
Engineering, Fr. C.R.I.T, Vashi, Navi, Mumbai, India
Rohan Sawant Vidyalankar Institute of Technology, Mumbai, India
Sudhir Sawarkar Department of Computer Engineering, Datta Meghe College of
Engineering, Navi Mumbai, India
Pratishtha Saxena Department of Computer Science & Engineering, Centre for
Advanced Studies, AKTU, Lucknow, India
Manjubala Sekar Amrita Center for Cyber Security Systems & Networks, Amrita
Vishwa Vidyapeetham, Amrita University, Kollam, Kerala, India
Y. Chandra Sekhar Department of Electronics and Communications
Engineering, Amrita Vishwa Vidyapeetham, Amritapuri, India
R. Senthilnathan School of Computing, Centre for Information Super Highway
(CISH), SASTRA Deemed to be University, Thanjavur, Tamil Nadu, India
Suresh Seshan School of Computing, Centre for Information Super Highway
(CISH), SASTRA Deemed to be University, Thanjavur, Tamil Nadu, India
M. Sethupathi Department of ECE, SNS College of Technology, Coimbatore,
India
Dharmil Shah Department of Electronics and Telecommunication, Dwarkadas
J. Sanghvi College of Engineering, Mumbai, India
Henish Shah Department of Electronics and Telecommunication, Dwarkadas
J. Sanghvi College of Engineering, Mumbai, India
Kalp Shah Department of Electronics and Telecommunication, Dwarkadas
J. Sanghvi College of Engineering, Mumbai, India
Sohail Shaikh Information Technology, Terna Engineering College, Navi
Mumbai, India
Ahmed Shan-A-Alahi Department of Computer Science and Engineering,
International Islamic University Chittagong, Chittagong, Bangladesh
V. S. Shankar Sriram School of Computing, Centre for Information Super
Highway (CISH), SASTRA Deemed to be University, Thanjavur, Tamil Nadu, India
Editors and Contributors xliii

M. Shanmugasundaram School of Electronics Engineering, VIT, Vellore, India;

Department of Embedded Technology, VIT, Vellore, India
Ravindra Kumar Sharma Singhania University, Jhunjhunu, Rajasthan, India
Sunil Sharma Rajasthan Technical University, Kota, India
Rajashree Shedge Ramrao Adik Institute of Technology, Navi Mumbai, India
Yash Shetye Department of Mechatronics Engineering, MPSTME, NMIMS
University, Mumbai, India
B. Simhachalam Naidu KPIT Technologies, Bengaluru, India
C. Sinchana Department of Computer Science and Engineering, Vidyavardhaka
College of Engineering, Mysuru, Karnataka, India
K. Sinchana Department of Computer Science and Engineering, Vidyavardhaka
College of Engineering, Mysuru, Karnataka, India
Sukhvir Singh Department of Information Technology Development (IT),
University Institute of Engineering and Technology (UIET), Punjab University,
Chandigarh, India
Sachin Kumar Singh Department of Embedded Technology, VIT, Vellore, India
Madhvi Singhal Department of Computer Science & Engineering, Guru Ghasidas
University, Bilaspur, India
S. Sivagamasundari Department of Electronics and Instrumentation Engineering,
Annamalai University, Chidambaram, India
V. Sivakumar Department of IT, Adhiyamaan College of Engineering, Hosur,
Tamilnadu, India
N. Sivaramakrishnan Department of ECE, SNS College of Technology,
Coimbatore, India
K. P. Soman Center for Computational Engineering and Networking (CEN),
Amrita School of Engineering, Amrita Vishwa Vidyapeetham, Coimbatore, India
M. Someshwaran Electronics and Communication, KCG College of Technology,
Chennai, India
Anshu Soni National Institute of Technology, Kurukshetra, Haryana, India
Siddaram Sonnagi Department of Computer Science and Engineering, NMAMIT,
Nitte, Karnataka, India
V. Sowmya Center for Computational Engineering and Networking (CEN),
Amrita School of Engineering, Amrita Vishwa Vidyapeetham, Coimbatore, India
Rajkumar P. Sreedharan Department of Electronics and Communication,
Amrita Vishwa Vidyapeetham, Amritapuri, India
xliv Editors and Contributors

A. G. Sreejith Muthoot Institute of Technology and Science, Kochi, India

K. Vandith Sreenivas Department of Electronics and Communication
Engineering, Amrita Vishwa Vidyapeetham, Amritapuri, India
R. Sridevi Department of Computer Science and Engineering, K. Ramakrishnan
College of Engineering, Trichy, India
A. Srilakshmi SASTRA Deemed to be University, Thanjavur, India
S. Srimathi Department of Computer Science and Engineering, K Ramakrishnan
College of Engineering Samayapuram, Trichy, India
M. Sai Srinivas Department of Electronics and Communications Engineering,
Amrita Vishwa Vidyapeetham, Amritapuri, India
S. Sruthi Department of ECE, TKM College of Engineering, Kollam, Kerala,
India
Swaminathan Subbiah Department of Computer Science and Engineering,
Saveetha School of Engineering, SIMATS, Chennai, India
Gopinath Subramani Department of Information Technology, KCG College of
Technology, Chennai, India
Gnanou Florence Sudha Department of Electronics and Communication
Engineering, Pondicherry Engineering College, Puducherry, India
T. Suganya Sri Krishna College of Technology, Coimbatore, Tamil Nadu, India
R. Sujatha School of Electronics Engineering, Vellore Institute of Technology,
Vellore, Tamil Nadu, India
H. D. Sundresh RV College of Engineering®, Banglore, India
B. Sunil Department of Information Science and Engineering, Vidyavardhaka
College of Engineering, Mysuru, Karnataka, India
N. R. Sunitha Department of Computer Science and Engineering, Siddaganga
Institute of Technology, Tumkur, Karnataka, India
Neha Supe Ramrao Adik Institute of Technology, Navi Mumbai, Maharashtra,
India
S. Suresh Department of Computer Science and Engineering, P.a College of
Engineering and Technology, Pollachi, Coimbatore, TamilNadu, India
C. Suresh Gnana Dhas Vel Tech High Tech Dr. Rangarajan Dr. Sakunthala
Engineering College, Chennai, India
Azhar Syed Department of EEE, BITS Pilani Dubai Campus, Dubai, UAE
Suyog Tambe Sardar Patel Institute of Technology, Mumbai, India
Editors and Contributors xlv

R. Tamilselvi Department of Electronics and Communication Engineering, Sethu

Institute of Technology, Pulloor, Virudhunagar District, Tamilnadu, India
Hardik Thakkar Sardar Patel Institute of Technology, Mumbai, India
Neha Rupesh Thakur Department of Electronics and Telecommunication,
Thadomal Shahani Engineering College, Bandra, Mumbai, India
Shivam Thakur Department of Embedded Technology, VIT, Vellore, India
S. Theeijitha Department of ECE, SNS College of Technology, Coimbatore, India
Ayushi Turkar Department of Computer Science and Engineering,
Vidyavardhaka College of Engineering, Mysuru, Karnataka, India
Mohammad Nazim Uddin Department of Computer Science and Engineering,
International Islamic University Chittagong, Chittagong, Bangladesh
A. Umamakeswari School of Computing, SASTRA Deemed University,
Thanjavur, India
A. Umapriya Department of ECE, Vivekanandha College of Engineering for
Women, Namakkal, India
A. S. Umarfarooq Department of E&CE, SDMCET, Dharwad, Karnataka, India
Shree Rajesh Raagul Vadivel Department of Electronics and Communication
Engineering, Amrita Vishwa Vidyapeetham, Amritapuri, India
Vandana Mansur School of Electronics Engineering, Vellore Institute of
Technology, Vellore, Tamil Nadu, India
P. Venkatramana Department of Electronics and Communication Engineering,
Centre for VLSI and Embedded Systems, Sree Vidyanikethan Engineering College,
Tirupati, India
Sushma Verma Defence Research and Development Organization, (SAG), New
Delhi, India
Amarsinh Vidhate Ramrao Adik Institute of Technology, Navi Mumbai, India
K. P. Vidyashree Department of Information Science and Engineering,
Vidyavardhaka College of Engineering, Mysuru, Karnataka, India
M. S. Vijaya PSGR Krishnammal College for Women, Coimbatore, India
R. Vijayakumar School of Computer Sciences, Mahatma Gandhi University,
Kottayam, Kerala, India
Rajilal Manathala Vijayan School of Computing, SASTRA Deemed University,
Thanjavur, India
Putchala Vinay Department of Electrical and Electronics Engineering, Amrita
Vishwa Vidyapeetham, Amritapuri, India
xlvi Editors and Contributors

Athira Vinod Amrita Center for Wireless Networks & Applications (AWNA),
Amrita School of Engineering, Amrita Vishwa Vidyapeetham, Amritapuri, India
C. Vinothini Dr. N.G.P. Institute of Technology, Coimbatore, India
S. V. Viraktamath Department of E&CE, SDMCET, Dharwad, Karnataka, India
L. R. Vishnu Vardhan Sri Krishna College of Technology, Coimbatore, Tamil
Nadu, India
D. Vishnu Vashista School of Electrical and Electronics Engineering, SASTRA
Deemed to Be University, Thanjavur, India
Vimal P. Viswan Muthoot Institute of Technology and Science, Ernakulam, India
P. V. Vivek Sridhar Mallya Department of Electronics and Communication
Engineering, Amrita Vishwa Vidyapeetham, Amritapuri, Kerala, India
Prasanth M. Warrier Department of Electronics and Communication, Amrita
Vishwa Vidyapeetham, Amritapuri, India
Marguerite Wotto Université du Québec à Montréal (UQÀM), Montreal, Canada
Divakar Singh Yadav Institute of Engineering and Technology, Lucknow, UP,
India
Nishi Yadav Department of Computer Science & Engineering, Guru Ghasidas
University, Bilaspur, India
Prasant Kumar Yadav Department of Electronics and Communication
Engineering, Amrita Vishwa Vidyapeetham, Amritapuri, India
Samarjeet Yadav Department of Computer Science & Engineering, Centre for
Advanced Studies, AKTU, Lucknow, India
Surekha Yakatpure Electronics and Telecommunication Department, A G Patil
Institute of Technology, Solapur, India
Vinayakgouda Yallappagoudar Department of E&CE, SDMCET, Dharwad,
Karnataka, India
Kaki Yeswanth Department of Electronics and Communications Engineering,
Amrita Vishwa Vidyapeetham, Amritapuri, India
D. Yuvaraj Department of CSE, Chian University, Cihan University, Duhok,
Kurdistan Region, Iraq
Prajakta Zodge IT Department, Terna Engineering College, Mumbai University,
Navi Mumbai, Maharashtra, India
Enhanced Security Mechanism in Cloud
Based on DNA Excess 3 Codes

Manjot Kaur and Kiranbir Kaur

Abstract Exchanging data over the system has generally utilized quick and solid
hot spot for correspondence. Clients from wide devotion utilize this component for
exchanging and retrieving the required data. Portability and operability inside cloud
framework are achieved through disconnected and online mediums, which are per-
sistently alluring, yet the issue of security emerges amid the transmission process.
Security and unwavering quality are the key issues during the exchange process,
which requires serious research consideration. Data security is achieved by utilizing
the public and private key-enabled block-level DNA-based EX-3 code. The analysis
is inferred on the disconnected information as well as on the online information,
for example, Google Docs. Redundancy handling mechanism is utilized to guaran-
tee space for the information storage supplier, which remains a minimum utilized
characteristic as it considers the capacity utilization in DSP. With the proposed sys-
tem, the overall space allocation for heavy documents is decreased and online data
security has been improved by the utilization of byte-level DNA-based EX-3 code.

Keywords DNA · Ex-3 · Data security

1 Introduction

In recent years, due to the rapid development of media transmission and Web, data
security turns out to be increasingly huge in nature. Cryptography is one of the
most ideal paths for ensuring unknown data. Cryptosystems can be separated into
two kinds, secret key cryptosystem and public key cryptosystem. The main sort
(secret key cryptosystem) utilizes a similar encryption key to encipher the plaintext
and decode the figure content. For this reason, this system is additionally called as
symmetric cryptosystem. It has a few disadvantages like excessively numerous keys,

M. Kaur (B) · K. Kaur

GNDU (Main Campus), Amritsar, India
e-mail: manjotk705@[Link]
K. Kaur
e-mail: [Link]@[Link]
© Springer Nature Singapore Pte Ltd. 2020 1
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
2 M. Kaur and K. Kaur

key conveyance issue, verification and non-repudiation issue. The imperative kind
which is the public key cryptosystem is created to take care of the issues of symmetric
cryptosystem, and RSA cryptosystem emerges as the most prevalent approach. As
of late, information security has turned into an essential issue for public, private and
guard associations as a result of the extensive misfortunes of illicit information. To
shield the secret information or data from unapproved access, illegal changes and
propagation, different sorts of cryptographic strategies are utilized. One of these
critical strategies is cryptography, which performs the investigation of writing in
secret frame, and it is isolated into two sorts: symmetric cryptography and asymmetric
cryptography.
Cloud computing is emerging a state-of-the-art storage technology in almost all the
domains, but many experts argue about it [1]. Highly scalable services are provided
by the cloud. Users can utilize the services on pay-per-use basis. Cloud computing
theoretically provides infinite resources, but due to growing number of users, practi-
cally services and resources become limited. The services and resources required to
be distinguished on the basis of scale of utilization along with cost. Further improve-
ment in cloud services could lead to a better framework for concurrent users in order
to access resources more than the capacity of the machine user hold which leads to
more popularity and user community attracted towards cloud services [2].
Cloud interoperability is required during the transmission of data to and from the
cloud servers. The cloud service provides ensures quality of service (QoS) through
security mechanisms. The security mechanisms used may or may not use redundancy
handling mechanism to conserve space. In the proposed system, security mecha-
nism along with redundancy handling mechanism is enforced for ensuring quality
of service. The attributes considered for evaluation are described below.

1.1 Attributes

Cloud computing is used widely from long time and provides opaque framework
where services are visible to the user but internal working is hidden [3]. Key attributes
in cloud computing are described in this section:
• Service-Based: The main objective of cloud is to provide a service-oriented frame-
work by hiding details and showing only necessary features to the user. This
mechanism is also termed as abstraction.
• Scalable and Elastic: Services associated with cloud are not fixed. Services can
be added as and when required depending upon mass usage of services. In other
words, scalable environment is provided by cloud computing [4]. Elasticity in
framework indicates that the resources are provided on different platforms that
remain accessible to multiple users at the same time. In other words, concurrency
is supported through the use of cloud computing framework.
Enhanced Security Mechanism in Cloud Based on DNA Excess 3 Codes 3

• Shared [5]: Pool of resources are provided by the use of advanced computing envi-
ronment. The provided resources are not exclusive in nature. Exclusive resources
cannot be shared, and that required accessing queue should be maintained.
• Metered by Use: Multiple payment modes are supported by the cloud infrastruc-
tures [6]. Services are accessed on pay-per-use basis. Service provider and clients
are bound by the service-level agreement. User needs to pay for accessing the
services mentioned within SLA.
• Uses Internet Technologies: Services are delivered to the user by the use of Inter-
net. Protocol such as Hypertext Transfer Protocol (HTTP), File Transfer Protocol
(FTP), Terminal network (Telnet), etc., are used for this purpose [7].
Symmetric calculations are commonly viewed, and they are appropriate for
preparing extensive stream of information. A portion of the popular and proficient
symmetric calculations incorporate two fish, Serpent, AES, Blowfish and IDEA.
Also, there are non-specific calculations that offer an elective system for encryp-
tion. Hereditary calculations contain three essential administrators; they are mul-
tiplication, hybrid and change. Furthermore, there are distinctive well-known and
productive deviated calculations like RSA, NTRU and elliptic curve cryptography.

2 Related Work

Li et al. [8] describes IBE technique with outsourcing computation and also offloads
the key generation operations to key update the cloud service provider. It also focuses
on critical issues of identity revocation. It accomplishes consistent productivity for
both calculations at PKG and private key size at client; user need not to contact with
PKG amid key update. PKG is permitted to be disconnect and subsequent to send
the disavowal rundown to KU-CSP; no protected channel or client verification is
required amid key update amongst client and KU-CSP.
Seo et al. [9] proposed the main mCL-PKE scheme without blending operations
and gave its formal security. Our mCL-PKE takes care of the key escrow and dis-
avowal issue. Utilizing the mCL-PKE conspires as a key building block; it proposes
an enhanced way to deal safely and share sensitive information out in the public
clouds. This approach supports quick denial and guarantees the classification of the
information put away in an untrusted open cloud while authorizing the entrance
control strategies of the information proprietor. The exploratory outcomes demon-
strate the productivity of fundamental mCL-PKE scheme and enhanced approach for
people in general cloud. Further, for various clients fulfilling similar access control
arrangements, the enhanced approach performs just a solitary encryption of every
datum thing and lessens the general overhead at the information owner.
Wang et al. [10] proposed a variation of CP-ABE to effectively share the various
hierarchical documents in distributed computing. The hierarchical documents are
scrambled with an incorporated access structure, and the ciphertext parts identified
with characteristics could be shared by the records. Thus, both ciphertext storage
4 M. Kaur and K. Kaur

and time cost of encryption are saved. The proposed system has benefits that clients
can decode all approval documents by figuring secret key once. Therefore, the time
cost of decryption is also saved if the client needs to decode various documents.
Additionally, the proposed plot is ended up being secure under DBDH suspicion.
Xu et al. [11] designed a virtual encryption card framework that gives encryption
card usefulness in virtual machines. In this framework, it displayed the vEC-PPM,
which deals with the encryption resource plan. It saved clients’ information utilizing
a trusted equipment of virtualization in view of TPM. It additionally settled a trusted
chain amongst clients and encryption cards in the light of the composed protocols. The
design of the virtual encryption card empowers the security and productivity of the
encryption benefit. A usage examination shows that the effectiveness of framework
is similar to that of the native mode. Later on, it proceeds with examination, trying
to plan a virtual encryption cards bunch to help higher encryption speed and more
reasonable similarity with virtualization.
Alabdulatif et al. [12] proposed a safe billing protocol for smart applications in
distributed computing. It utilized homomorphic encryption through adjusting the
Domingo-Ferrer’s plan, which can perform different number arithmetic operations
to fulfil smart grid billing necessities in a safe way. This plan keeps up the exchange
off amongst security and versatility contrasted and other homomorphic plans that
depend on either secure, yet inelastic in terms of arithmetic operations assortment.
Additionally, it proposed an instrument that guarantees both security and integrity
during correspondence between substances. The execution of the proposed system is
very satisfactory; it is sufficiently productive to use in lightweight applications and
can be helpfully connected to cloud-based applications.
Li et al. [13] proposed a CP-ABE scheme that gives outsourcing key-issuing,
decryption and keyword search work. This scheme is productive since it just needs
to download the fractional decryption ciphertext relating to a particular keyword.
In this scheme, the tedious matching operation can be outsourced to the cloud spe-
cialist organization, while the slight operations should be possible by clients. In this
way, the calculation cost at the two clients and trusted specialist sides is limited.
Besides, the proposed plot supports the capacity of keywords look which can enor-
mously enhance correspondence effectiveness and further ensure the security and
protection of clients. It is difficult to stretch out given KSF-OABE plan to help get
to structure represented by tree in [14]. In this paper, based on contingent intermedi-
ary communicate re-encryption technology, an encrypted information sharing plan
for secure distributed storage is proposed. The plan not just accomplishes commu-
nication information sharing by exploiting communication encryption, yet in addi-
tion accomplishes dynamic sharing that enables adding a client to and expelling a
client from sharing gatherings dynamically without the need to change encryption
open keys. Besides, by utilizing intermediary re-encryption innovation, this scheme
empowers the intermediary (cloud server) to specifically share encoded information
to the objective clients without the intercession of information owner while keeping
information security so significantly enhances the sharing execution. In the mean-
time, the rightness and the security are demonstrated; the execution is broke down,
Enhanced Security Mechanism in Cloud Based on DNA Excess 3 Codes 5

and the test comes about is appeared to confirm the possibility and the productivity
of the proposed plot.
Liu et al. [15] proposed diagram encryption scheme which just makes utilization
of lightweight cryptographic natives, for example, pseudo-arbitrary capacity and
symmetric-key encryption, instead of moderate homomorphic encryptions. Accord-
ingly, the proposed graph encryption scheme is well disposed to a wide arrangement
of graph information-based distributed computing and edge registering applications,
for example, interpersonal organizations, e-maps, criminal investigations and so on.
Contrast with graph anonymization comes nearer from database group, the proposed
system achieves higher security level as the chart itself is encoded and it does not
make any suspicions on the sorts of attacks.
Song et al. [16] discussed the security enhancement mechanisms including sym-
metric, public key, and homomorphic cryptosystems to enable experts to comprehend
encryption plans for information on distributed storage. AES is utilized as a part of
most secure applications for information on distributed storage. Completely homo-
morphic encryption plans are promising for cloud condition however a long way
from being useful due to their execution rate. Homomorphic assessment of AES
has fascinating applications as a reasonable encryption conspires for information on
distributed storage.
Veeraragavan [17] proposed an improved encryption calculation (EEA) for secur-
ing the data in cloud stockpiling. This is a symmetric encryption calculation. It
utilizes same key for encoding and unscrambling the data previously put away into
cloud. Xu et al. [18] proposed a lightweight accessible open key encryption (LSPE)
conspired with semantic security for CWSNs. LSPE decreases countless calculation
escalated operations that are received in past works; along these lines, LSPE has
sought execution near that of some useful accessible symmetric encryption schemes.
Tsai et al. [19] proposed a protected cloud data encryption framework, named the
circulated ecological key (DENK in short), with which all records are encoded by
one encryption key got from numerous coordinating keys which are keys got from
approved clients’ secret key keys and a believed PC’s natural key. El-yahyaoui [20]
proposed to present an effective and unquestionable FHE in the light of another
mathematic structure that is without commotion.
Thomas [21] described the various ways used in cloud computing for data security.
The information is stored on the incorporated area called data centres that have a
substantial size of information storage. In this way, the customers need to put stock in
the supplier on the accessibility and additionally information security. Before moving
information into general society cloud, issues of security gauges and similarity must
be tended to. A trusted screen introduced at the cloud server that can screen or
review the operations of the cloud server. In limiting potential security trust issues
and additionally sticking to administration issues confronting cloud computing, an
essential control measure is to guarantee that a solid cloud computing service-level
agreement (SLA) is set up and kept up when managing outsourced cloud service
suppliers and particular cloud merchants. Cloud computing guarantees to change the
financial matters of the server farm, yet before sensing and managed information
move.
6 M. Kaur and K. Kaur

To resolve the problem with the existing literature, the proposed literature
presents efficient solution. The encryption mechanism with the redundancy handling
mechanism is proposed as described in the next section.

3 Methodology

The methodology of proposed work consists of registration process at first place.

The registration in proposed system will be two-phase process. In the first phase,
registration at data storage provider is made. After successfully registering, user can
load files at data storage provider end. To generate keys, users require performing
registration at key service provider. In order to retrieve the files, users must login to
the DSP and then KSP. The keys generated could be used in order to decrypt the file.
The mechanism also uses redundancy handling mechanism for preserving space for
extra file loading. Also online source of files like Google Docs can be used to retrieve
the files and perform encryption and decryption.
The detailed steps are described as under:

3.1 Registration at DSP

The registration at DSP comprises unique username and password. Username and
password once registered at DSP can be used for accessing file uploading module.

3.2 Registration at KSP

Key service provider (KSP) is used in order to generate the keys for the file which is
uploaded. The proposed system is capable of generating the keys for files generated
from online source.
Enhanced Security Mechanism in Cloud Based on DNA Excess 3 Codes 7

Start

Fetch the file from online or

offline source

Apply Redundancy handing Mechanism

Obtain Cipher text

Ciphertext=c

Generate the key using byte level

deduplication and random key generation
and upload it to the cloud

Generate Plain Text

PlainText=Decode(CipherText)

Print Plain Text

Stop
8 M. Kaur and K. Kaur

3.3 Generating Keys

In order to generate keys, user must login to the KSP. The files uploaded are encrypted,
and corresponding keys are generated. The redundant files are neglected, and rest of
the files are uploaded with the public and private keys generated.

3.4 Encryption and Decryption

For encryption and decryption using byte-level DNA-based EX-3 code, algorithms
are hybridized. The algorithm yields ciphertext after receiving files as plaintext.
Verification of the overall procedure is in terms of time consumed and size of the file
that can be uploaded.

4 Result and Performance Analysis

The result is presented in terms of file size that can be uploaded. Reliability of
encryption and decryption in terms of time consumed is also a performance metric.
The comparison in terms of quality is given below.
Space conservation is achieved using the proposed system. This space conserva-
tion is indicated through Table 1.
The plots correspond to the existing and the proposed literature are given in Fig. 1.
The result deviation obtained using the existing and the proposed system is also
given through the line chart. The line chart is generated by feeding the values obtained
from the simulation as shown in Fig. 2.
The execution time subsequently reduced as file size to be uploaded reduced. The
execution time is estimated by subtracting the end time of simulation from start time
of simulation. The comparison table for the same is given as Table 2.
The plot for the execution time is given using bar and line charts. The chart
generation is done using the facility provided through excel and JFreeCharts (Fig. 3).
The simulation results obtained are also plotted using the line chart. The proposed
system execution time is better as compared to the existing system. The proposed

Table 1 Simulation results

Simulation Block level (KB) Byte level (MB)
with different file sizes
Sim 1 1046 678
Sim 2 1055 500
Sim 3 2010 878
Sim 4 1034 645
Sim 5 1074 788
Enhanced Security Mechanism in Cloud Based on DNA Excess 3 Codes 9

2500

2000

1500
File size

Block
1000 Level(KB)
Byte
Level(MB)
500

0
Sim 1 Sim 2 Sim 3 Sim 4 Sim 5

Number of simulations

Fig. 1 Existing and the proposed file size comparison with bar chart

2500

2000

1500
Block
Level(KB)
1000
Byte
Level(MB)
500

0
Sim 1 Sim 2 Sim 3 Sim 4 Sim 5

Fig. 2 Line chart corresponds to space used by existing and the proposed systems

Table 2 Simulation result of

Simulation Block level (ms) Byte level (ms)
execution time
Sim 1 123 83
Sim 2 103 67
Sim 3 112 78
Sim 4 187 102
Sim 5 118 81
10 M. Kaur and K. Kaur

200
180
160
exection time in ms 140
120
100 Block
80 Level(ms)
60 Byte
40 Level(ms)
20
0
Sim 1 Sim 2 Sim 3 Sim 4 Sim 5
number of simulation

Fig. 3 Execution time of simulation corresponding to various simulations

200
180
160
140
120
100 Block Level(ms)
80 Byte Level(ms)
60
40
20
0
Sim 1 Sim 2 Sim 3 Sim 4 Sim 5

Fig. 4 Execution time corresponding to existing and the proposed systems

system uses the byte-level deduplication, and the existing system uses block-level
deduplication (Fig. 4).

5 Conclusion

From this research paper, we conclude that the cloud computing not only provides the
resources to the users but also develops major security challenge. There are security
requirements for both users and cloud providers, but sometimes it may conflict in
some way. Security of cloud depends upon trusted computing and cryptography. In
Enhanced Security Mechanism in Cloud Based on DNA Excess 3 Codes 11

our review paper, some issues are related to data location, security, storage, availabil-
ity and integrity. Establishing trust in cloud security is the biggest requirement. The
problems and corresponding solutions may require further investigation in terms of
key size and complexity. Complexity of key can be further enhanced by the use of
pseudo-random number generator within the key generation phase.
By incorporating complex key structure, cloud performance and user interaction
can be further enhanced.

References

1. Hwang K, Bai X, Shi Y, Li M, Chen W-G, Wu Y (2016) Cloud performance modeling with
benchmark evaluation of elastic scaling strategies. IEEE Trans Parallel Distrib Syst 27(1):130–
143
2. Noor TH, Sheng QZ, Yao L, Dustdar S, Ngu AHH (2016) CloudArmor: supporting reputation-
based trust management for cloud services. IEEE Trans Parallel Distrib Syst 27(2):367–380
3. Armbrust M et al (2010) A view of cloud computing. Commun ACM 53(4):50
4. Buyya R, Yeo CS, Venugopal S (2008) Market-oriented cloud computing: vision, hype, and
reality for delivering IT services as computing utilities. In: Proceedings of 10th IEEE inter-
national conference on high performance computing and communications, HPCC 2008, pp
5–13
5. Nirmala SJ, Tajunnisha N, Bhanu SMS (2016) Service provisioning of flexible advance
reservation leases in IaaS clouds, vol 3, no 3, pp 154–162
6. Marwan M, Kartit A, Ouahmane H (2016) Secure cloud-based medical image storage using
secret share scheme. In: 2016 5th International Conference on Multimedia Computing and
Systems (ICMCS), IEEE, pp 366–371
7. Dimitrov DV (2016) Medical internet of things and big data in healthcare. Healthc Inform Res
22(3):156–163
8. Li J, Li J, Chen X, Jia C, Lou W (2013) Identity-based encryption with outsourced revocation
in cloud computing. IEEE Trans. Comput. 64(2): 425–437
9. Seo S, Nabeel M, Ding X, Bertino E (2013) An efficient certificateless encryption for secure
data sharing in public clouds. IEEE Trans Knowl Data Eng 26(9): 2107–2119
10. Wang S, Zhou J, Liu JK, Yu J, Chen J, Xie W (2016) An efficient file hierarchy attribute-based
encryption scheme in cloud computing. IEEE Trans Inf Forensics Secur 11(6): 1265–1277.
11. Xu D, Fu C, Li G, Zou D, Zhang H, Liu XY (2017) Virtualization of the encryption card for
trust access in cloud computing. IEEE Access 5, 20652–20667
12. Alabdulatif A, Kumarage H, Khalil I, Atiquzzaman M, Yi X (2017) Privacy-preserving
cloud-based billing with lightweight homomorphic encryption for sensor-enabled smart grid
infrastructure. IET Wirel Sens Syst 7(6):182–190
13. Li J, Lin X, Zhang Y, Han J (2016) KSF-OABE: outsourced attribute-based encryption with
keyword search function for cloud storage. IEEE Trans Serv Comput 10(5): 715–725
14. Jiang L, Guo D (2017) Dynamic encrypted data sharing scheme based on conditional proxy
broadcast re-encryption for cloud storage. IEEE Access 5, 13336–13345.
15. Liu C, Zhu L, Chen J (2017) Graph encryption for top-k nearest keyword search queries on
cloud. IEEE Trans Sustain Comput 2(4): 371–381
16. Song C, Park Y, Gao J, Nanduri SK, Zegers W (2015) Favored encryption techniques for cloud
storage. In: 2015 IEEE First International Conference on Big Data Computing Service and
Applications, IEEE, pp. 267–274
17. Veeraragavan N, Arockiam L, Manikandasaran SS (2017) Enhanced encryption algorithm
(EEA) for protecting users’ credentials in public cloud. In: 2017 International Confer-
ence on Algorithms, Methodology, Models and Applications in Emerging Technologies
(ICAMMAET), IEEE, pp. 1–6
12 M. Kaur and K. Kaur

18. Xu P, He S, Wang W, Susilo W, Jin H (2017) Lightweight searchable public-key encryption

for cloud-assisted wireless sensor networks. IEEE Trans Ind Inf XX(XX):1–12
19. Tsai KL et al (2016) Cloud encryption using distributed environmental keys. In: Proceedings
of 2016 international conference on innovative mobile and internet services in ubiquitous
computing, IMIS 2016, pp 476–481
20. El-Yahyaoui A, El Kettani MD (2017) A verifiable fully homomorphic encryption scheme to
secure big data in cloud computing. In: 2017 International Conference on Wireless Networks
and Mobile Communications (WINCOM), IEEE, pp. 1–5
21. Thomas G, Jose V, Afsar P (2013) Cloud computing security using encryption technique, arXiv
preprint, arXiv:1310.8392
TensorFlow-Based Semantic Techniques
for Multi-cloud Application Portability
and Interoperability

Tanveer Kaur and Kiranbir Kaur

Abstract Cloud computing permits plentiful access to shared pools of resources that
are configurable and provide higher-level services to the user, which can be easily and
hastily granted with minimal management effort. With the advancements in cloud
computing, many cloud service providers have started using the distributed high-
end servers to provide services to its users. However, while designing an efficient
cloud environment, it has been found that the application portability is a key issue.
Portability is usually offered to mitigate supplier lock-in. Nevertheless, shifting from
a single method to an alternative for a minimum of work, seeing that is quite possible
together with box companies, which can also improve the durability along with
scalability. Therefore, in this paper, a novel TensorFlow-based semantic technique
is designed and implemented to significantly port the applications between high-end
servers. Extensive experiments have been carried out to evaluate the effectiveness of
the proposed technique. Extensive experiments reveal that the proposed technique
outperforms the existing techniques.

Keywords Cloud computing · Semantic · Multi-cloud · TensorFlow · Application

portability · Vendor lock-in

1 Introduction

Application portability has the capability of an application to be compactly intro-

duced, sent, got to and oversaw—independent of their conveyance display. This term
communicates an application’s adaptability when this is utilized on numerous dis-
tinctive stages or quickly got to from the Internet, a work area or system. Application
portability coordinates various abilities of an application, and that application can be

T. Kaur (B) · K. Kaur

Department of Computer Engineering and Technology, Guru Nanak Dev University, Amritsar,
Punjab, India
e-mail: tanuhundal05@[Link]
K. Kaur
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 13

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
14 T. Kaur and K. Kaur

got to over the Internet from a Web program. At the point when disconnected, appli-
cation portability portrays an application’s ability that is to be actualized on basic
working framework (OS) conditions [1]. Application portability likewise alludes to
an application that is transmitted and executed by means of versatile gadgets, for
example, a Universal Serial Bus (USB) pen drive.
Portability. It is possible to move or transmit all portable apps between drives
such as USB flash drive and carry anywhere. If you want to have your favourite
programs available with you at all times, in this case this can be more helpful.
Better privacy. Since compact applications don’t leave any utilization follows
or remaining records behind and they don’t should be introduced. This makes them
quite fit for private usage scenarios.
Synchronization with cloud storage services. Portable applications can be syn-
chronized with cloud storage services such as Dropbox. This is really helpful on the
off chance that you utilize a distributed storage benefit on different PCs to match up
your documents.
Custom application settings. The settings are typically spared with the appli-
cation itself, when you indicate any custom settings for a versatile app (e.g. default
download organizer for a compact browser). So you get a steady client encounter,
regardless of the PC utilized for getting to the versatile application(s) [2].
Cloud Computing is often a model that gives consumers limitless computing
electrical power that may be seen from anywhere based on consumer’s convenience.
User could incorporate rely the time like CPU, memory space, safe-keeping and also
bandwidth along with demand for the reasoning provider [3]. User might also make
use of prefabricated expert services like The amazon online marketplace Cognate,
Supple Beanstalk supplied by this reasoning service provider in order to boost smooth
operating regarding a questionnaire along with much less labour pool.

2 Technique Used

2.1 The Semantic-Based Approach

The overall model is a graph-based representation, structured into five conceptual

layers. The five conceptual layers are [4]:

• This parameter level shows this criteria associated with the info sort changed
between providers while input in addition to creation of this operations.
• The operations level shows these syntactic criteria in the operation and also benefits
revealed by the cloud services (described through WSDL).
• The service level presents the particular semantic annotation of the vendor-
dependent impair solutions (exposed via OWL-S) as well as the supporting ontolo-
gies necessary to find the particular cloud service provider reinforced operation,
suggestions plus productivity parameters.
TensorFlow-Based Semantic Techniques for Multi-cloud … 15

• The cloud pattern level represents this semantic explanation involving agnostic
in addition to vendor-dependent cloud patterns becoming aware via an OWL
representation. It contains habits in infrastructural level possibly at program level.
• The application pattern level symbolizes the actual account associated with
behaviour expounding on the applying to be ported.

2.2 TensorFlow

TensorFlow is usually an open-source stockpile with regard to statistical computation.

It was developed by Google in 2015. Offers allocated concurrent unit understanding
based upon general-purpose dataflow graphs.

3 Related Work

Antoniades et al. [5], Enabling Cloud Application Portability, introduced the Cloud
Application Requirement Language (CARL). CARL can be used for describing the
application software and hardware requirements, information that is then encom-
passed into the TOSCA description of the cloud application, beside the applica-
tion blueprint. CAMF’s information service utilizes both these artefacts to provide
IaaS-specific configurations that satisfy the user’s requirements. Ranabahu et al.
[6], Application Portability in cloud computing, November–December, survived an
abstraction-driven way to deal with location the application convenience issues and
spotlight on the application improvement process. It additionally surviving our hypo-
thetical premise and involvement in two down to earth extends in which it connected
the Abstraction driven methodology. Gunka et al. [7], Moving an Application to
the Cloud, provided help understanding this issue by characterizing an evolution-
ary strategy to relocate a current application to the cloud. The principle ventures
of this change are (1) porting the application to an IaaS, (2) including load adjust-
ing the introduction and business rationale layer and (3) incompletely moving the
application to a PaaS. Next to with every one of the means depicted multi-cloud
methodologies will be considered keeping in mind the end goal to have the capacity
to moderate dangers by using repetitive, free frameworks and to guarantee benefit
vicinity for clients at various areas. Moreover, choice help is required to help rec-
ognizing the best-coordinating cloud stages for each progression. Kolba et al. [8],
Towards Application Portability in Platform as a Service, arranged portability issues
of PaaS; it portrays a model of current PaaS contributions and recognizes distinc-
tive conveyability perspectives. Starting from the model, it infers an institutionalized
profile with a typical arrangement of capacities that can be found among PaaS sup-
pliers and relating with each other to check application transportability in view of
16 T. Kaur and K. Kaur

biological system abilities. It approves our discoveries with a comprehensive infor-

mational collection of 68 PaaS contributions together with an electronic application
for versatility coordinating. It additionally distinguishes advance convenience issues
by moving the application to various PaaS merchants, approving biological commu-
nity movability and giving indications to future research headings. Di Martino et al.
[4]. Semantic Technique of Multi-cloud application portability and Interoperability
2016, presented a semantic-based representation of application patterns and cloud
services exhibited, with a case of its utilization in a normal dispersed application,
which shows how the proposed strategy can be effectively worked for the disclosure
and arrangement of cloud services. Kostosk et al. [3], A New Cloud Services Portabil-
ity Platform, proposed another cloud compactness benefit stage and gave an outline
of the present patterns in the zone of cloud benefit portability, particularly breaking
down the TOSCA’s methodology. The new stage proposition depends on establishing
a connector show, which offers the coveted cloud movability. The new arrangement
characterizes the administrations by a XML TOSCA approach. Cretella et al. [9]
discovered functionalities, APIs and assets required for the application improvement
through semantic-based agnostic (vender independent) portrayals of such applica-
tion segments, and re-aiming to fabricate a custom platform as a service (PaaS) that
can acknowledge and exchange establishments of any convenient applications.

4 Gaps in Literature

• Getting a new progress, interconnected atmosphere is confronted with many obsta-

cles, for example, provisioning, protection, acceptance and personality managing,
flexibility, economic vitality proficiency and so forth;
• Vendor lock-in which ends from the possible lack of interoperability along with
portability is said being a dominant obstacle in the route connected with realization
connected with interclouds;
• Communications in Computer and Information Science (CCIS);
• The use of TensorFlow has been pushed aside by means of pre-existing examiner
to provide more cost-effective brings about multi-cloud environment.

5 Proposed Methodology

This section contains the graphical representation of the proposed technique, con-
sisting of various steps which are required to successfully accomplish the suggested
algorithm shown in Fig. 1.
• Step 1. User sends a service request to cloud service provider;
• Step 2. Train the TensorFlow;
TensorFlow-Based Semantic Techniques for Multi-cloud … 17

HES1 HES2 ….. HESn

WHETHER
JOB YES PORT LOAD
PORTABILI OR HESi
TY

TRAIN TENSORFLOW

…….
U1 U2 Un

Fig. 1 Proposed block diagram

• Step 3. It checks that whether a job portability is required; if yes, then port load,
and if no, then service request goes to high-end services (HESn ), and high-end
service provided the services to users.
The proposed technique is designed using ANACODA. TCP uses a neural network
to evaluate the T-worth function. The input for the system is the current, whereas the
output is the relating T-worth for every one of the activity (Table 1).

Table 1 T leaning information

1 Activity (Ac ): Altogether the conceivable moves that the specialist can take
2 State (St ): Present circumstance reverted by environment
3 Reward (Rd ): An instant return send back from the environment to assess the last activity
4 Strategy (π ): The methodology that the specialist utilizes to decide the following activity
dependent on the present state
5 Worth (Ve ): The predictable lasting return with reduction, as opposed to the temporary
reward Rd . Ve π (c) is refer to as the expected lasting return of the present states under
strategy π
6 T-worth or activity-worth (T): T-worth is like worth; then again, actually it takes an
additional stricture, the present activity u
18 T. Kaur and K. Kaur

We train the network dependent on the T-learning inform condition. Review that
the objective T-worth for T-learning is:
 
d + γ max

Q εj+1, u ; σ − (1)
u

The ε is comparable to the state c, whereas the σ represents the strictures in the
neural network, which is not in the space of our dialog. Along these lines, the loss
function for the system is characterized as the squared error among target T-worth
and the Q-worth output from the network.

Algorithm 1: Deep T-learning with experience replay

Originate replay memory M to capacity P
Originate activity-worth function T with random weights σ


Originate target activity-worth function T with weight σ − = σ

For incident = 1, M do
Originate sequences r1 = {m1 } and pre-processed sequence ε1 = ε(r1 )
For t = 1, T do
With probability ε select a random activity ut
Otherwise select ut = argmaxa Q(ε(rt ), u; σ )
Execution action ut in emulator and observe reward dt and image mt+1
Set rt+1 = rt, ut ,mt+1 and pre-process
 εt+1 = ε(rt+1 )
Store transition εt, ut, dt , εt+1 in D  
Sample random minibatch of transitions εj, uj, dj, εj+1 from D

dj if episode terminates at step j + 1
Set nj = 
 
dj + γ maxu T εj+1, u ; σ − otherwise
  2
Perform a gradient descent step on nj − T εj , aj ; σ with respect to the
network parameters σ 

Every G steps reset T = T

End For
End For

• Experience Replay: Since training samples in distinctive RL set-up are excep-

tionally corresponded and less data-efficient, it will tougher convergence for the
network. An approach to resolve of the sample distribution issue is adopting back-
ground replay. Basically, the sample transitions are stowed, which will at that point
be haphazardly chosen from the “transition pool” to refresh the knowledge.
• Separate Target Network: The objective T network has identical structure as the
one that estimates worth. In each C step, as indicated by the above pseudocode,
the target network is reset to another. Along these lines, the fluctuation becomes
less severe, bringing about progressively steady trainings.
TensorFlow-Based Semantic Techniques for Multi-cloud … 19

6 Proposed Methodology

This section defines the graph methodology with the help of parameters like execution
time, number of swaps, load imbalance rate, and overhead time to describe the
effective results in cloud computing environment. Another two strategies are likewise
basic for training TCP.
Figure 2 shows the query execution on VMs, execution at VMs without any failure,
failure at VMs, response to query (−ve) meaning not active and energy consumption
by VMs.
Execution time. The execution time intended for an activity is described as how
much time the whole usually takes to carry out confirmed job; this includes time
utilized executing work time and also program products and services about its behalf.
The particular apparatus and also system familiar with estimate execution time for a
particular job are described as follows:

Fig. 2 Graphical representation of queries on VM’s

20 T. Kaur and K. Kaur

Fig. 3 Execution time

Fig. 4 Waiting time graph

Execution Time = Stop Time−Start Time

The below graphs show the comparison between the existing system and proposed
system with respect to execution time. Figure 3 shows that existing system had
consume more execution time than the proposed system. Figure 4 shows that existing
system had consume more waiting time than the proposed system.

7 Conclusion

Interoperability is the capability regarding more than one devices or apps to switch
details and also to mutually employ the details which were exchanged. Cloud inter-
operability is the ability of a customer’s system to connect to a cloud program or
the flexibility for one cloud that wants to connect to alternative cloud solutions by
swapping details based on some given method to receive estimated results. The inter-
operability obstacles with each one of the clouds program classes (IaaS, PaaS plus
TensorFlow-Based Semantic Techniques for Multi-cloud … 21

SaaS) might be different. That might be more challenging to realize behavioural inter-
operability over a couple of uses as opposed in order to connect a compute preferred
to remote control storage, intended for example. In this paper, to enhance equally
interoperability as well as portability, a new TensorFlow-based interpretability has
been designed and implemented. It trains engines your semantic options that come
with multi-cloud programs to boost the outcome further. Extensive experiments have
been carried out to evaluate the effectiveness of the proposed technique. Extensive
experiments reveal that the proposed technique outperforms the existing techniques.

References

1. Petcu D, Craciun C, Rak M (2011) Towards a cross platform cloud API. In: 1st international
conference on cloud computing and services science, pp. 166–169
2. Arunkumar G, Venkataraman N (2015) A novel approach to address interoperability concern
in cloud computing. Procedia Comput Sci 50:554–559
3. Kostoska M, Gusev M, Ristov S (2014) A new cloud services portability platform. Procedia
Eng 69:1268–1275
4. Di Martino B, Esposito A (2016) Semantic techniques for multi-cloud applications portability
and interoperability. Procedia Comput Sci 97(1): 104–113
5. Antoniades D, Loulloudese N, Foudoulis A, Sophokleous C, Trihinas D, Pallis G, Dika-
iakos M, Kornmayer H (2015) Enabling cloud application portability. In: 2015 IEEE/ACM
8th international conference on utility and cloud computing (UCC), pp. 354–360
6. Ranabahu A, Maximilien EM, Sheth A, Thirunarayan K (2013) Application portability in cloud
computing: an abstraction-driven perspective. IEEE Trans Serv Comput 8(6): 945–957
7. Gunka A, Seycek S, Kühn H (2013) Moving an application to the cloud: an evolutionary
approach. In: Proceedings of the 2013 international workshop on Multi-cloud applications and
federated clouds, ACM, pp. 35–42
8. Kolba S, Wirtz G (2014) Towards application portability in platform as a service. In: 2014
IEEE 8th international symposium on service oriented system engineering, pp. 218–229
9. Cretella G, Di Martino B (2015) A semantic engine for porting applications to the cloud and
among clouds. Softw: Pract Exp 45(12):1619–1637
10. Gondis F, Simons AJH, Paraskakis I, Kourtesis D (2013). Cloud application portability: an
initial view. In: Proceedings of the 6th balkan conference in informatics, ACM, pp. 275–282
Enhancing High Availability for NoSQL
Database Systems Using Failover
Techniques

Priyanka Gotter and Kiranbir Kaur

Abstract NoSQL allows fast processing of real-time large data applications. It

allows database professionals to make use of elastic scalability of database servers.
Most of these databases are created to perform on full throttle despite the presence
of low-cost hardware. Users are able to create the database more quickly due to
the non-relational nature of NoSQL. There is no need of developing the detailed
(fine-grained) database model. As a result, it will save lots of development time. The
primary aim of this paper is to present a novel data integration method as well as
workflow that executes the data integration of various source systems in such a way
that user does not require any programming abilities. Middleware is considered as
main part of the proposed technique which provides high availability of data in case
of failures by utilizing the checkpointing. Experimental results reveal that the pro-
posed technique provides efficient results, especially in case of any kind of failures
in distributed NoSQL servers.

Keywords NoSQL · High availability · Checkpointing · Failures

1 Introduction

The capacity to move duplicate or exchange information effectively from one

database, stockpiling or IT condition to another in a safe and secure way with-
out hindrance to usability is termed as data portability. The information must be in
an organization that is interoperable between a few stages if we have to work with
data portability [1, 2]. This concept comes forward to handle the issue of massive
data. Data is increasing day by day and SQL databases are not able to manage this
huge amount of data. There is a need of non-relational (often known as NoSQL)
database management system since relational databases just work with organized

P. Gotter (B) · K. Kaur

Computer Engineering and Technology, Guru Nanak Dev University, Amritsar, Punjab, India
e-mail: piyuprecious13@[Link]
K. Kaur
e-mail: [Link]@[Link]
© Springer Nature Singapore Pte Ltd. 2020 23
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
24 P. Gotter and K. Kaur

information [8], which can manage both organized and semi-organized information.
To compose the information, i.e., produced from different applications [4], social
database and other customary databases take after unbending structure. However,
NoSQL database gives adaptability in arranging the information that makes it sim-
ple to get to the information. They are not founded on a single model [5] (e.g., social
model of RDBMSs), and every database is relying on the objective usefulness. ACID
properties are imposed in SQL server [7]. Likewise, BASE property is imposed in
NoSQL. The BASE acronym was characterized by Eric Brewer, who is likewise
known for planning the CAP hypothesis. As per CAP hypothesis, in the meantime, a
circulated PC framework cannot ensure the majority of the accompanying three prop-
erties in the meantime, i.e., consistency, availability and partition tolerance. NoSQL
has recently come out and has developed interest and criticism [11]: interest simply
because they manage specifications which can be significant within large-scale uses
and criticism due to the analysis with popular relational achievements [3]. Uniform
data access is a computational idea explaining an evenness of controllability along
with connectivity across numerous target data sources. Data integration consists
of mixing data residing in various sources and offers users with a unified view of
them. This technique gets to be significant in a range of conditions, such as either
industrial (such because if not one but two equivalent businesses should merge his)
or her database and scientific (combining investigation comes from various bioin-
formatics repositories, to get example) domains. A checkpoint is a local state of a
task kept on secure storage. Simply by routinely performing a checkpointing, one
can possibly help you save a reputation of an activity with regular intervals. In case
any node fails, one may resume computation from the earlier checkpoints, thereby
avoiding restating execution from the beginning.

2 Related Work

This paper offers a novel integration methodology to be able to question information

through different relational databases along with NoSQL repository system [12]. The
recommended technique is dependent on meta-model strategy and it also protects the
structural, semantic plus syntactic heterogeneities of origin system. To demonstrate
the particular usefulness of recommended method, an Internet program is definitely
designed, named hybrid DB. It possesses a database layer which provides whole
database services over origin system. The parameters defined are entire running
time, data retrieval time and native query time. Leavitt [6] learnt that many companies
gather huge amount of client, scientific, sales and other details with regard to long-
term evaluation. Traditionally, these types of companies have saved organized data
in relational databases for long-term accessibility and evaluation. However, growing
quantity of designers plus users has begun looking at various types of non-relational,
right now frequently termed NoSQL databases. Different NoSQL databases carry
various approaches. What we share can be that they are non-relational. Its major
benefit is that contrary to relational databases, they hold unstructured information
Enhancing High Availability for NoSQL Database Systems … 25

like files, mails and media efficiently. By using the advantages of NoSQL to migrate
data, Nikam et al. introduce a framework that is helpful to map the MySQL query
to MongoDB query format [9]. Soon after, it fetches the data faster by presenting a
decision maker that chooses the database from which data can be fetched at a rel-
atively faster rate, thus uplifting the system performance. The previous years have
experienced an extreme rise in the quantity as well as the heterogeneity regarding
NoSQL data stores [10]. As a result, researches along with comparison of such infor-
mation stores have gotten difficult. Recently, several info-accessibility middleware
programs with regard to NoSQL have emerged that include use of different NoSQL
info stores out of standard APIs. Rafique et al. presented two complementary studies.
First, Rafique et al. measure the efficiency overhead described by these systems for
the CRUD operations. Second, costs of migration with and without these kinds of
programs are usually compared.

3 Limitations of Earlier Work

Subsequent section describes the limitations of the existing work.

3.1 High Availability

When it is continuing to grow quickly, the NoSQL local community is actually

innovative additionally along with falls short of the maturity on the MySQL end user
base. Definitely, NoSQL will be quickly increasing, but it is experiencing the high
availability issue.

3.2 Lack of Reporting Tools

The unavailability of reporting methods for evaluation and performance testing is

the major issue with NoSQL databases. On the other hand, by using MySQL, you
can acquire many reporting tools to aid you to validate your individual application’s
validity.

3.3 Lack of Standardization

To ensure that NoSQL is growing, it requires an average query language including

SQL. This is a significant problem pointed out by experts at Microsoft, who seem to
report that NoSQL’s absence of standardization could cause an issue for the duration
26 P. Gotter and K. Kaur

of migration. Apart from this, standardization will be very significant to the repository
sector to be able to bring together themselves while in the future.

4 Proposed Methodology

This section contains the graphical representation of proposed technique, consisting

of various steps which are required to successfully accomplish the suggested algo-
rithm. Each time a checkpoint function happens, many unclean data file web pages
for any repository are usually prepared for NoSQL web server (all web pages may
have improved with memory since they were examined from NoSQL web server as
well as the previous checkpoint), irrespective of the point of transaction that pro-
duced the particular transformation. Before the page is usually prepared for NoSQL
web server, every log data including the new log data conveys a big difference for
that web page that is usually prepared to NoSQL web server (yes, log records may
be cached with RAM too).

Use of checkpointing:

• Preserve condition of task from steady intervals;

• Allow few CSP to supply higher access to your requests of clients in case of failure;
• Decrease the loss of calculations out while in the indication of failures.

To show the checkpointing structure, 2 metrics are used:

• Checkpoint overhead (due to checkpoint implementation, execution period of task

is increased);
• Checkpoint latency (duration of their time needed in order to save lots of your
checkpoints).

All the jobs are submitted by the clients to request manager. The given job is
splitted into threads by request manager and also assigns one site (service manager)
to threads and updating of global checkpoint took place. In First in First out (FIFO)
fashion, the threads are selected by every site and casually loaded service node is
assigned to it. That thread is executed by service nodes, or if the execution of any
other thread is going on, then this thread may be added in its waiting queue. N1 in
order to N12 is the service node which will supply services for the clients.

Algorithm:

• Step1: Clients send their tasks to distributed NoSQL service provider (DNSP)
present at local site;
• Step2: DNSP split the task in threads. Then min loaded site is distributed to the
jobs;
• Step3: After distribution of sites, updating of global checkpoint took place;
• Step4: Checkpoint will probably function periodically;
Enhancing High Availability for NoSQL Database Systems … 27

Fig. 1 Existing technique

without failure

• Step5: By analyzing checkpoint DNSP, it checks whether any site has failed or no
failure occurs. New save point is generated if no failure took place and checkpoint
updating took place;
• Step6: The work is transferred from failed node to available site in case of failure
and updating of checkpoint will take place.

Figure 1 shows the existing technique without failure. User submits the NoSQL
query to the query server. Query server performed its job and it decomposes the
query so that it can be run at their respective sites and generates the result. Later on,
the output is fetched from each and every site so that final result is merged, and it
will return back to the user. On the other hand, Fig. 2 describes what if the site 2 is
crashed due to any reason. It will not generate its output. Other sites will wait for
the output so that final result can be merged. This leads to increase the data retrieval
time, and also, final result is not merged (Fig. 3).
This methodology introduces the checkpointing scenario.
The steps included are:
• Step1: Initially, the failures are monitored that are occurring due to any condition;
• Step2: Save the checkpoints that are used to recover the data later;
• Step3: If failure does not occur, it will return to the user back;
• Step4: If failure occurs, then rollback to save point is the phenomenon by which
the availability of data got increased;
• Step5: If queries are terminated, the responses are automatically sent to the user,
and if it is not, then repeat steps from starting.
28 P. Gotter and K. Kaur

Fig. 2 Existing technique

with failure

Fig. 3 Flowchart of
proposed algorithm

Algorithm 1
 
Algorithm initPPreDeCon Dbuf , d , ∈, μ, λ, e
/ ∗ every place in D is noted as unclassified and Dj is a
collection of point Proce
Processor n, 0 ≤ n < p do
for each unclassified 0 ∈ Dbuf do
pref
If C̄ORE (h)then/ ∗ enlarge a new cluster ∗ /
den
build new microClus q − micro
Enhancing High Availability for NoSQL Database Systems … 29

make new cluster ID

put entirey ∈ N̄∈W̄O (h)in queueU ;
W hile U = ∅ do
p = initial point inU ; 
calculate Z = y ∈ DIR REACH pref (q, y) ;
den
for every y ∈ Z do
if y is classified , in that case
put y in U ;
allot present cluster ID to y
put y in microClus
eliminate p fromU ;
otherwise
mark h as noise;
finish.
finish
Add q − micro in q − microClus record
finish.
//combine step
Processor n, 0 ≤ n < P do
For every classified m ∈ Dj do
if |clusID| > 1 in that case
//amount of clustered that point gets
put Min of them as clusID;
p
eliminate m point from microclus in recordlj
finish

Note: Q_Si = query size, QQ = query queue, QS = query status, AV_QQ =

available query queue, AV_RS = available resources status, qi = independent queries,
R_CAP = resources capacity, vm = virtual machine, rvm = reconfigurable virtual
machine, pr = available physical resources, mips = million instructions per seconds.

Algorithm 2
1. Initialize input parameters:
Ni → Number of queries, Aj → arrival time of Ni , rs → resources
2. Sorting queries:  
AV _ QQ[] = SORT Ni , Aj
j = 0;
AV _ RS = “ready”
for every resources Rs in resource group do
while(Rs is in running state)
skip and choose the next one
30 P. Gotter and K. Kaur

3. Select resources

IfRs j selected  
Rs selected [] = Rs j
j + +;
k = 0; //for every selected resources
4. Check query dependency:
if (dependent queries in Rs selected [])
assign AV _QQ[k] to Rs selected []
5. Check available resources:
if (R_CAP > Q_size)
submit queries to selected resources
elseif (R_ CAP < Q_ size) &&(AV _ RS = ‘terminated ’)
make new vm to match the Q_Si
6. Check vm status:
If (vm failure = true)in that case
make new vm from rvm
check
if (New vm’s capacity(mips) > Q_Si)
assign queries
otherwise
6.1 make vm using one PR to satisfy the Q_size
If (Q_size < avail_(PR))
Create vm
Update vm pool
6.2 Create vm from combining PR to satisfy the Q_size
//for(i = 0; i < +n_PR; i + +)
//for(j = 1; j < n_PR;
 j + +)
elseif (avail_ PRij > Q_size)
create new vmi + vmj
7. Check status of dependent queries:
if qs = executed process Qi
else
repeat the process from step 5
8. Queuing independent queries based on priority:
Select Qi and replicate the similar process from step 5
9. Check status of each and every query:
If (qs = executed & &QQ = empty)
then stop the process
Enhancing High Availability for NoSQL Database Systems … 31

Fig. 4 Entire running time

analysis

Fig. 5 Data retrieval time

analysis

5 Results

This section defines the graph methodology by using the following parameters:
• Entire running time defines the time taken by the queries to give appropriate results
according to the values;
• Data retrieval time defines the time at which the data is retrieved from the database
management system and there are many options to store the retrieved data like files.
The comparison between the existing system and proposed system with respect
to native query time, data retrieval time and entire running time is shown in Figs. 4
and 5.

6 Conclusion

A NoSQL checkpoint is a kind of test operation that verifies data retrieved from the
database by comparing that data with the baseline copy stored in your project. That
32 P. Gotter and K. Kaur

may be needed, for example, when you test an application that modifies a database
and want to verify that the appropriate tables are updated correctly. Baseline data
used by the checkpoint for verification is stored at middleware end. Each element
contains connection settings for a database and a baseline copy of the data to be used
for verification. Experimental results reveal that the proposed technique provides
efficient results, especially in case of any kind of failures in distributed NoSQL
servers.

References

1. Alomari E et al (2015) CDPort: a portability framework for NoSQL datastores. Arab J Sci Eng
40(9):2531–2553
2. Androcec D (2018) Data portability among providers o platform as a service. Res Pap Fac
Mater Sci Technol Slovak Univ Technol 21(Special-Issue):7–11
3. Arora R, Aggarwal RR (2013) An algorithm for transformation of data from MySQL to NoSQL
(MongoDB). Int J Adv Stud Comput Sci Eng 2(1):6–12
4. Atzeni P et al (2014) Uniform access to non-relational database systems: The SOS plat-
form. Lecture Notes in Computer Science (including Subseries Lecture Notes in Artificial
Intelligence, Lecture Notes in Bioinformatics), vol 7328, pp 160–174
5. Chaure SD et al (2015) Web based ETL approach to transform relational database to graph
database, 7
6. Leavitt N (2010) Will NoSQL databases live up to their promise? Computer (Long Beach Calif)
43(2):12–14
7. Lotfy AE et al (2016) A middle layer solution to support ACID properties for NoSQL databases.
J King Saud Univ—Comput Inf Sci 28(1):133–145
8. Mior MJ et al (2017) NoSE: schema design for NoSQL applications. IEEE Trans Knowl Data
Eng 29(10):2275–2289
9. Nikam P et al (2016) Migrate and map: a framework to access data from Mysql, Mongodb or
Hbase using Mysql Queries. IOSR J Comput Eng 18(3):13–17
10. Rafique A et al (2018) On the performance impact of data access middleware for NoSQL data
stores a study of the trade-off between performance and migration cost. IEEE Trans Cloud
Comput 6(3):843–856
11. Rith J et al (2014) Speaking in tongues: SQL access to NoSQL systems. In: Proceedings of
29th Annual ACM Symposium on Applied Computing—SAC ’14, pp 855–857
12. Vathy-Fogarassy Á, Hugyák T (2017) Uniform data access platform for SQL and NoSQL
database systems. Inf Syst 69:93–105
Implementation of Automated Bottle
Filling System Using PLC

S. V. Viraktamath, A. S. Umarfarooq, Vinayakgouda Yallappagoudar

and Abhilash P. Hasankar

Abstract Automatic bottle filling using programmable logic controller (PLC) con-
stitutes a user-specified volume selection, in which the user can input the desired
amount of liquid or water to be inserted in the bottles. It is generally used where
many bottles of same volume are to be filled by passing bottles over the conveyor
belt. PLC is a main functional block in the automation which tries to minimize the
complexity and increases safety and cost reduction. Using of PLC in filling the bot-
tles allows us to select the required amount of liquid by the ladder language. Filling
is done by using motor, sensors, conveyor belt, PLC, solenoid valve, and so on. The
whole system is more flexible and time saving. The process of filling is carried out for
packaging of liquid and beverages. This is an interdisciplinary branch of engineer-
ing which includes mechanical, computer, and electronics parts. The process also
enhances the knowledge of fabrication, programming, design, planning, and presen-
tation skills. The PLC is gaining popularity because it is easy for troubleshooting
which makes programming easier and reduces downtime.

Keywords Automation · PLC · Sensor · Solenoid valve · Glass motor

S. V. Viraktamath · A. S. Umarfarooq (B) · V. Yallappagoudar · A. P. Hasankar

Department of E&CE, SDMCET, Dharwad, Karnataka, India
e-mail: umarfs0707@[Link]
S. V. Viraktamath
e-mail: svvmath@[Link]
V. Yallappagoudar
e-mail: vinayakgouda1997@[Link]
A. P. Hasankar
e-mail: abhilashhasankar2@[Link]

© Springer Nature Singapore Pte Ltd. 2020 33

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
34 S. V. Viraktamath et al.

1 Introduction

Automation is the use of control system for operating equipment’s and informa-
tion technology with minimal or reduced human intervention in the production of
goods and services [1, 2]. Apart from manufacturing, the automation has found a
notable impact in industries. Automation has been achieved by various means includ-
ing mechanical, hydraulic, pneumatic, electrical, electronic devices, and computers,
usually in combination. Major part of the world economy is played by automation.
Automation finds its application in the field where a liquid must be filled continu-
ously for fixed interval or based on volume of the bottle such as soft drinks and other
beverage industries [3, 4]. The work presented is an application of automation, and
the processes are controlled using the PLC. The PLC that was used in this work is
Rexroth-L20. This PLC operates on 24 V DC and variable number of inputs and
outputs.

2 Literature Survey

Control engineering has been the best field of engineering [1–4]. In the past, humans
were employed for controlling the system [5, 6]. Always, there is an eagerness in
human beings to develop better technique in any working process to attain or to
provide less mental stress and more comfort to the operators. The PLC is selected
based on scan time, communication protocol, memory size, and software loaded into
it. Nowadays, electricity is being used as control for automation, but in older days
relays were used as control automation. Relays are very expensive, since they are
mechanical devices it has limited usage and lifetime. The improvements of low-cost
computers have changed trends in the automation fields. The programmable logic
controller was started in the 1970s and has become the most common choice for the
large-scale manufacturing controls in all automation industries [7, 8].

3 Component Description

3.1 Inputs

Infrared (IR) sensor is used as an input. An IR sensor is a device that senses changes
in its surrounding. An IR sensor can sense the heat, and it can detect the motion. In
our project, IR sensor is used to detect the bottle position [7]. The IR sensor is shown
in Fig. 1.
Implementation of Automated Bottle Filling System Using PLC 35

Fig. 1 IR sensor

3.2 Outputs

The various output devices used are motor and solenoid valve. The motor is used to
drive the conveyor in forward direction. One end of solenoid is connected to the tank
and other for filling process [9, 10]. A solenoid valve operates electromechanically.
Electric current controls functioning of the valve. In fluidics, solenoid valves are used
most commonly to control elements. Their functions are to close, open, distribute,
or mix liquids. Solenoids offer good reliability, easy switching, compatibility with
other devices, and small size. The solenoid valve is shown in Fig. 2.

Fig. 2 Solenoid valve

36 S. V. Viraktamath et al.

3.3 Regulator

The outputs that are obtained from PLC cannot be directly fed to the output devices.
Hence, regulators are used in order to supply with the amount of voltage that is
required by the output devices. In this work, 12-V regulator has been used to step
down the output from 24 to 12 for the motor that was used to drive the conveyor
belt. The regulator is covered by a heat sink in order to control the amount of heat
dissipated while transferring the regulator 7812 IC as shown in Fig. 3.

4 Methodology

Bottles are placed in required position on a conveyor belt. When the power is
switched, the conveyor belt starts to move along with the bottles on it. Once the
bottles appear in front of IR sensor which is placed to detect the bottle, conveyor
belt stops making the bottle positioned still below the solenoid valve. After 7-s time
interval, solenoid valve opens. This results in flowing of liquid from reservoir to
bottle through solenoid valve. Up to 20 s, liquid will be continuously flowing into
the bottle making it full. After 7 s, the conveyor starts to move and carries away the
filled bottles unless and until next bottle is detected. This process continues till the
power is on [11] as shown in Fig. 4.

5 Description

The filling operation takes place when the bottles are kept in sequential manner on
a conveyor belt. The bottles are filled one at a time irrespective of number of bottles
on the conveyor belt. The process is also provided with a volume defined by the user
by adjusting the time interval provided to fill the bottle. The block diagram is shown
in Fig. 5.

Fig. 3 7812 regulator

Implementation of Automated Bottle Filling System Using PLC 37

Fig. 4 Flowchart of bottle filling system

Fig. 5 Block diagram of

PLC interfacing

5.1 Bottle Detection Using IR Sensor

Bottles are kept in position over the conveyor belt at the input side. IR sensors are
used to detect the presence of bottles. The sensor will detect their position based on
bottle motion. The IR sensor sensing the bottle is shown in Fig. 6.
38 S. V. Viraktamath et al.

Fig. 6 IR sensor sensing the

bottle and filling

5.2 Bottle Filling Operation

Once the bottles are in front of IR sensor, they are detected. Once the bottle reaches
the desired position, the conveyor belt stops, and filling operation begins after 7 s of
delay. Seven seconds of delay is provided to settle the bottle over the conveyor belt.

5.3 Specified Volume of Liquid Introduced

The filling operation is associated with the volume defined by the user. The desired
volume of liquid is fed into the bottle depending on the volume that is already stored
in the program. In this work, the preset value of timer is defined which ensures the
solenoid opening for that time instant. This process continues until there are bottles
on the conveyor belt. The overall setup is power dependent; hence, once the power
is turned off, the system stops. When the bottles are on the conveyor and the power
is on, the process of filling simply continues.

6 Simulation Results

a. When power is switched on, conveyor belt starts moving as shown in Fig. 7.
b. When IR sensor is on, the bottle is sensed as shown in Fig. 8.
c. Solenoid valve opens and closes after filling the bottle as shown in Fig. 9.
Implementation of Automated Bottle Filling System Using PLC 39

Fig. 7 Conveyor belt ON

Fig. 8 IR sensor is ON

7 Advantages

• Compact and Portable system [12].

• Economical system.
40 S. V. Viraktamath et al.

Fig. 9 Solenoid valve opens and closes after filling the bottle

8 Limitations

• It uses electronic components and circuits; any amount of water leakage will be
dangerous.
• Using this system, only one bottle can be filled at a time.

9 Future Scope

As a future scope, the project can also be employed with various sizes, shapes, and
weights of the bottles and densities of the liquid. A level sensor can be used to detect
the level of liquid filled. Instead of filling one bottle at a time, it can be made to fill
the multiple bottles at a time reducing the time, but design complexity increases. The
process can also be extended to stuffing of definite quantity of capsules in bottles.
The process can be made more efficient using flow sensor which detects the water
flow. The process can be implemented on other controllers and the cost factors. The
system can be made more feasible with advancement in technology. The automation
system finds wide range of applications.
Implementation of Automated Bottle Filling System Using PLC 41

10 Conclusion

This paper describes a method to develop an automated bottle filling system based
on user-defined volume. The process is controlled by PLC. The entire system is
more flexible. Implementation of automated system increases the productivity and
economy. The language used to design this model is easy and understandable by
common people. This system can be used in every industry. With this work, it is
possible to design other applications of automated system.

References

1. Somavanshi AP, Austkar SB, More SA (2013) Automatic bottle filling using microcontroller
volume correction. Int J Eng Res Technol (IJERT) 2(3)
2. Saeed AUA, Al Mamun M, Zadidul Karim AHM (2012) Industrial application of PLCs in
Bangladesh. Int J Sci Eng Res (IJSER) 3(6). ISSN: 2229-5518
3. Kalaiselvi T, Praveena R (2012) PLC based bottle filling and capping using user defined volume
selection. ISSN: 2250-2459
4. Mashilkar B, Khaire P, Dalvi G (2015) Automated bottle filling system. IRJET 02(07). ISSN:
2395-0072
5. Chakraborty K, Roy I, De P (2015) Controlling process of a bottling plant using PLC and
SCADA. Indonesian J Electr Eng Inf (IJEEI) 3(1):39–44. ISSN: 2089-3272
6. Boyer SA (2009) SCADA—supervisory control and data acquisition, 4th edn. International
Society of Automation, USA
7. Ahuja H, Singh A, Tandon S, Shrivastav S, Patil S (2014) Automatic filling management system
for Industries. IJETAE 4(Special Issue 1)
8. Dakre A, Sayed JG, Thorat EA, Chaudhary AMA (2015) Implementation of bottle filling and
capping using PLC and SCADA. IRJET 02(9). ISSN: 2395-0072
9. Mallaradhya HM, Prakash KR (2013) Automatic liquid filling to bottles of different heights
using programmable logic controller. ISSN: 23202092
10. Savita, Lokeshwar (2012) Implementation and performance analysis of bottle filling plant using
ladder language. ISSN: 2319-7064
11. Patel J (2015) Based automatic bottle filling. ISSN: 2091-2730
12. Katre S, Hora J (2016) Microcontroller based automated solution filling module. Int Res J Eng
Technol (IRJET) 03(05)
An Integrated Approach to Network
Intrusion Detection and Prevention

B. Bhanu Prakash, Kaki Yeswanth, M. Sai Srinivas, S. Balaji,

Y. Chandra Sekhar and Aswathy K. Nair

Abstract At present, with the expansion of size of the internet, security plays a
crucial role in computer networks. Also with the advancement of Internet of things,
earlier technology like firewall, authentication and encryption are not effective in
ensuring the complete security. This has lead to the development of Intrusion Detec-
tion Systems (IDS) which monitors the events in computer networks to recognize
the threats that violates computer security. With the help of various machine learn-
ing algorithms we have carried out the implementation of IDS. Machine learning
technique increases the accuracy of anomaly detection in real-time scenario. This
work focuses on K-Nearest Neighbor (KNN) classifier and Support Vector Machine
(SVM), which classify the program behavior as intrusive or not. To prevent DoS
(Denial-of-Service) attacks, a new method is implemented in this paper. The KNN
classified data which provides malicious IP address are blocked in routers through
Standard Access-list.

Keywords Computer security · Intrusion detection · KNN

1 Introduction

Computer security vulnerabilities prevail as long as we have flawed security mea-

sures, insecure software programs, weak authentication mechanisms, and network
protocols. Almost all types of information are communicated and stored in public

B. B. Prakash (B) · K. Yeswanth · M. S. Srinivas · S. Balaji · Y. C. Sekhar · A. K. Nair

Department of Electronics and Communications Engineering, Amrita Vishwa Vidyapeetham,
Amritapuri, India
e-mail: bhanu_18@[Link]
K. Yeswanth
e-mail: kyeswanth1@[Link]
S. Balaji
e-mail: balu9198@[Link]
A. K. Nair
e-mail: aswathyscorp@[Link]
© Springer Nature Singapore Pte Ltd. 2020 43
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
44 B. B. Prakash et al.

switched communication network. This may lead to illegal interception, wiretap-

ping and tampering thereby, damage the entire system that results in irreparable loss
[1, 2]. Security products and Intrusion Detection System (IDS) are the existing meth-
ods to tackle the threats in computer networks. The latter is more intelligent as it
gathers all the information and study the behavior and finally based on that it could
characterize a program’s normal behavior [3]. IDS detects network intrusions and
protects the computer network from unauthorized access. It helps in distinguishing
the bad connections called Intrusions and good connections [4]. The distinguish-
ing is done by using different machine learning tools such as KNN-Classification,
K-means clustering, and probabilistic measures [5]. The term machine learning is
a branch of Computer Science that is closely related to data mining. It is a study
of algorithms and statistical models to improve the performance of a specific task
[6]. These algorithms are used in applications of Intrusion Detection in Computer
Networks, spam emails and some other areas where it is impracticable to develop an
algorithm for performing the tasks [7–9]. There are many ways in which the machine
learns, but the three important ways are:
A. Supervised Learning
It is the branch of machine learning that maps an input variable (x) to an output
variable (Y ) based on the labeled training data. The algorithm is intelligent enough
to map the function, f (x) when we have a new input data, x and we could predict the
output, Y. As the name implies the supervised learning algorithm acts as a supervisor
which teaches the machine using well-labeled data and when a new set of data is
given as input it could predict the correct output from labeled data [10, 7].
B. Unsupervised Learning
It is the branch of machine learning that tests the data which has not been labeled
or categorized as in the case of supervised learning. Here the role of algorithm is to
classify or group the unsorted data itself based on patterns, similarities or differences
[11]. The algorithm has to draw the inference by itself from the data set and because
of this such type of algorithm is used in exploratory data analysis.
C. Reinforcement Learning
It is the branch of machine learning that works on the principle of feedback of the
data given. Reinforcement agent has to learn from the environment and it is bound
to learn from its own experience [12]. There are various issues in intrusion detection
that is still needed to be addressed in detail. This paper has adopted two methods. One
is to detect the intrusion based on KNN algorithm. To increase the authenticity and
reliability of the system a second phase of IP filtering technique is also added. The
paper is organized as follows. Section 2 describes the methods adopted for intrusion
detection. Section 3 gives the implementation method of the system and Sect. 4
discusses results and analysis and the paper is concluded with Sect. 5.
An Integrated Approach to Network Intrusion … 45

2 Methods Adopted

Intrusion detection is performed using KNN algorithm which is the simplest one and
has less complexity compared to other algorithms. Using KNN classifier, it is able
to classify the program behavior in the network as normal or intrusive. The KNN
detection system is integrated with IP filtering technique that increases the reliability
of the system. In IP filtering method, the system is designed to block the incoming
connections which are malicious.
The data set for intrusion detection is obtained from KDD cup99 database which
contains a large sample of attacked data set. For KNN based intrusion detection,
the data set is split into training and testing data set using suitable split ratio. Based
on the distance between the attributes, the class votes are given with respect to the
majority. The Intrusive classes are classified out based on the attributes. After KNN
classifier, an IP filtering is performed in the router. We have performed a simulation
test network for IP filtering using Cisco Packet Tracer software. In Cisco Packet
Tracer, IP address of malicious nodes are listed out and the router is bounded to filter
out the malicious connection with the help of Internet Service Provider (ISP). These
methods are given in Fig. 2.

2.1 KNN (K-Nearest Neighbors)

KNN is one of the simplest algorithms used for classification and it is also most used
learning algorithm. KNN classifiers need two parameters: K value and threshold
value. K value denotes the number of adjacent neighbors and threshold value is
used for the judgment of abnormal neighbors. KNN uses the KDD CUP database
which contains data points that are separated into two separate classes to predict
the classification of a new data point. The model structure is formed from the data
by providing data set along with training data. The KNN was initially executed for
random samples in training set and having observed the accuracy, the same was
performed on test data. The similarity between the know class and unknown class
(unclassified) are identified using Euclidean-distance. Based on nearest neighboring
(Euclidean distance) it allocates new unclassified data point to one of the cluster.

2.2 Cisco Packet Tracer-Blocking Malicious IP Address

Metropolitan Area Network and Malicious IP addressing is implemented in Cisco

Packet Tracer. It is a Simulation tool that allows users to create their own network
topologies. This software also allows user to simulate the configuration of router and
switches using user interface and command-line interface.
46 B. B. Prakash et al.

Fig. 1 Flowchart of KNN

3 Implementation

3.1 KNN-Classification

The flow chart for KNN is shown in Fig. 1. We have implemented the K-Nearest
Neighbor Algorithm with the combination of defining several methods
(i) Importing Data: We have defined a method in Python language to import the
data in csv. format. This is for the ease of implementation so that we could
extract the data set smoothly.
(ii) Classifying the data into Training Set and Testing Set: We have used the split
ratio as 0.66:0.33; to split the data, and made use of an inbuilt-function random()
which returns the next floating-point number in the range [0.0, 1.0). We have
classified the data set into Training Set if the value of [Link] is greater
than the split value.
(iii) Calculating the Euclidean-Distance: Calculated the Euclidean-Distance by
Implementing the Distance formula.
An Integrated Approach to Network Intrusion … 47

(iv) Extracting the Neighbors based on the K value: Calculated distances are sorted
and the K-nearest neighbors are appended in a list. The K parameter plays an
important role in this method.
(v) Extracting the Response from the K-Nearest Neighbors: Based on the majority
votes from the neighbors, we return the most up-voted class as the predicted
class.
(vi) Calculating the Accuracy: Based on the ratio of the correct value and predicted
value the accuracy is measured.

Fig. 2 Overview of
intrusion detection and
prevention system
48 B. B. Prakash et al.

3.2 IP Filtering

In this technique, the router will filter out the malicious IP address that is listed in
the router local database before the route is announced. This is performed using
Access Control List (ACL) that works like filters and deny the packet entering into
the network. The ISP (Internet Service Provider) learns the blacklisted IP address
that is not delegated by IANA (Internet Assigned Numbers Authority) and based on
this router takes the decision for input filtering the malicious IP address. Establishing
MAN in Cisco Packet Tracer using multistep procedure.
(1) Assigning IP address and Configuration: For proper communication in network,
every Personal Computer (PC) is assigned with IP address, Subnet Mask, Default
gateway. PC of same network ID are connected through Switch. Switches helps
to forward the data between PCs and it also forwards and receives the data
packets of Router. In this step, each PC, router, and gateway are assigned IP
addresses.
(2) Router Configuration: Router helps to communicate between Remote Networks
(PC of different Network). PC of one network tries to access web server located
at another network, it should communicate through Routers. Every router forms
its own routing table so that it provides path for data packet to reach destination.
Each router has knowledge of only networks that are connected to it. So, in this
step PCs belong to different networks are connected through routers. Default
gateway at PC helps to reach the router it has connected. Also, configuration of
router hops is performed in this step.
(3) Connection Establishment: Step1 and step 2 make successful connection estab-
lishment between PCs of different networks. In this step 3, we are going to
check whether receiving data packets of one PC to another are in same network
or in remote network through PING command. In cisco packet tracer every PC
is provided with terminal. In that terminal ping IP address helps us to know
whether proper network is established between them.
(4) Malicious IP address detection: Malicious IP Address blocking using Standard
Access-list is performed here. In this step PC with malicious IP address are
blocked in the router of sever located. Access Control List (ACL) works like
filters and enables to control packets that is permitted in or denied out of a
network. Every Router in Cisco packet tracer is provided with CLI (Command
Line Interface). Commands for denying the IP address in router through standard
access-list are en helps to enable the router, configure terminal helps us to
configure terminal, sh access-lists gives list of access-list created, access-list
no deny IP address wildcard subnet blocks the specified IP address the router,
access-list no permit any permits all the IP address in network except the blocked
IP address.
An Integrated Approach to Network Intrusion … 49

Fig. 3 Metropolitan area network in Cisco Packet tracer

4 Results and Analysis

The KNN classifier was run for different values of K, and for each K, the algorithm
studies the network features and anomaly detection. After running the program mul-
tiple times, the best K value is chosen that gives less false rate. Simulation result of
number of neighbors, K in KNN versus accuracy is plotted in Fig. 4.
The schematic diagram and configuration of a simulation test network using packet
tracer is given in Fig. 3 Cisco Packet.
Based on several characteristics and observations, most reliable value of K (Near-
est Neighbors) is chosen. Figure 4 shows the accuracy obtained from KNN which
is 89.09%. The screenshot obtained from Python code is given as accuracy = 89.09
(Fig. 5).
Tracer is simulated in a Metropolitan Area Network (MAN). All communicating
nodes in Metropolitan Area Network tries to access web server through Routers and
Switches which are created in our simulation. So we identify the PCs with malicious
IP address and block those PCs in Router. Figure 6 shows that connections from
malicious node addresses to one of the PC is denied at the router.
50 B. B. Prakash et al.

Fig. 4 Accuracy versus K-neighbors

Fig. 5 Accuracy report of KNN algorithm

Fig. 6 Blocked IP address denied by router

5 Conclusion

In this project, we have used KNN algorithm for identifying the program behavior
and label as intrusion or not. To increase the reliability of the system, an IP filtering
technique is also integrated with the system. Using IP filtering the incoming connec-
tions which are malicious are blocked. For experimental findings, we have worked
An Integrated Approach to Network Intrusion … 51

on Denial of Service (DoS) attack. Our results give a good accuracy for intrusion
detection using KNN algorithm and decreased false rate. The work may be extended
to time-stamping method for accuracy and better detection of DoS attack.

References

1. Prem Sankar AU, Poornachandran P, Ashok A, Manu RK, Hrudya P (2017) B-secure: a dynamic
reputation system for identifying anomalous BGP paths. Adv Intell Syst Comput 515:767–775
2. Sankaran S, Sridhar R (2015) Modeling and analysis of routing for IoT networks. In: Interna-
tional conference on computing and network communications (CoCoNet). IEEE, Trivandrum,
India
3. Hindy H, Brosset D, Bayne E, Seeam A, Tachtatzis C, Atkinson R, Bellekens X A Taxonomy
and survey pf intrusion detection system, design techniques, network threats and datasets
4. Vijayarani1 S, Sylviaa M Assistant Professor and M. Phil Research Scholar from the
Department of Computer Science Intrusion detection system a study. Bharathiar University,
Coimbatore
5. Paliwal S, Gupta R (2012) Denial-of-service, probing remote to user (R2L) attack detection
using genetic algorithm. Int J Comput Appl
6. Panda M, Patra MR Network intrusion detection using naive bayes. Department of E TC
Engineering, G.I.E.T, Gunupur, India and from Department of Computer Science, Berhampur
University, Berhampur, India
7. Zhang M, Xu B, Gong J (2015) An anamoly detection model based on one-class SVM to detect
network intrusions. In: 11th conference on mobile ad-hoc and sensor networks
8. Xiaofeng Z, Xiaohong H (2017) Research on intrusion detection based on improved combi-
nation of k-means and multi-level SVM. In: K. Elissa (ed) 17th international conference on
communication technology. Title of paper if known, unpublished
9. Ghanem K, Aparacio Navarro FJ, Chambers JA (2017) Support vector machine for network
intrusion and cyber-attack detection. IEEE 2017
10. Seo J, Lee C, Shon T, Cho K-H, Moon J (2005) A new DDoS detection model using multiple
SVMs and TRA, IFIP
11. Hutchins E, Cloppert M, Amin RM (2011) Intelligence-driven computer network defence
informed by analysis of adversary campaigns and intrusion kill chains, USA. In: 6th
International conference on warfare and security
12. Yan F, Jain-Wen Y, Lin C (2015) Computer network security and technology research. IEEE
2015
A New Design of Equiangular Circular
Cum Elliptical Honeycomb Photonic
Crystal Fiber

Ashish Jain, Ravindra Kumar Sharma, Vimal Agarwal and Sunil Sharma

Abstract The design of Silica Glass photonic crystal fiber is proposed for lowering
the chromatic dispersion. For this design, Finite Difference Time Domain (FDTD)
methodology along with the transparent boundary condition (TBC) is applied. This
method produces zero dispersion at 0.5 to 1.5 µm diameter of circular and elliptical air
holes. These types of PCFs have high potential as like the dispersion compensating
fiber (DCF) in optical window. The refractive index is calculated with Sellmeier
equation in this method which is equal to the conventional silica glass, i.e., 1.457.
The proposed design is also used to show the non-linear effects of the material used.

Keywords Photonic crystal fiber (PCF) · Chromatic dispersion · Transparent

boundary condition (TBC) · FDTD (Finite Difference Time Domain)

1 Introduction

Photonic crystal fibers are available as an independent and new technology which acts
as a pioneered in the filled of communication. It was pinned by Phillip Russell. These
are available in the form of a finite element, two-dimensional or three-dimensional
photonic crystals that have a defect in the center core region. Basically the photonic
crystals are classified into solid core and hollow-core fibers. Number of parameter
can be varied to design a photonic crystal fiber. It includes the lattice symmetry
structure, the hole to hole distance, i.e., pitch and hole size which can be varied
by varying the diameter of hole. The solid core types of fibers shows a high-index
profile of solid core with a lower effective index surrounding medium known as
cladding. On the other hand the hollow core fibers have a lower index profile than

A. Jain · V. Agarwal
Apex Institute of Engineering & Technology, Jaipur, India
R. K. Sharma (B)
Singhania University, Jhunjhunu, Rajasthan, India
e-mail: ravindra.8810@[Link]
S. Sharma
Rajasthan Technical University, Kota, India
© Springer Nature Singapore Pte Ltd. 2020 53
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
54 A. Jain et al.

the surrounding medium of cladding. These Photonic crystals can be widely used as
periodic optical nanostructures. They show high bandwidth range and lower attenu-
ation characteristics. The optical transmission wavelengths available are 850, 1310,
and 1550 nm.

2 Design Considerations for Fiber Optics

It is very important to know about design parameters before designing any type of
fiber optic system. Some key factors must be taken into consideration for this design.
The prime consideration of this design is to transmit lossless signals over the wide
range of channels so that we can have a better reception of signals.
So we must be aware about:
• Modulation and multiplexing techniques that must be used.
• Enough signal strength or power level must be known of the receiver signal, i.e.,
power budget.
• Rising time along with bandwidth must be known.
• What type of fiber/amplifier is best suited for the application?
• Cost-Effectiveness.
• Design structure of the system.
• Dispersion profile of the system.
• Confinement Loss of the system.
• Refractive Index profile of material used for the system design.
Among all these parameter here we have consider dispersion, confinement loss,
refractive index, and design structure for the proposed work.

3 Proposed Structure

The proposed structure is designed by considering various parameters like air hole
diameter is varying from 0.5 to 1.5 µm range. The air hole spacing, i.e., pitch is
maintaining at 2 µm for nine-layer structure (Fig. 1).
This proposed structure is used to minimize dispersion along with maintaining
refractive index equal to the silica material, i.e., 1.457 (Figs. 2, 3).
Above Figs. 4, 5, 6 and 7 shows the 3D viewer of the proposed structure but
with varying parameters. All the above-mentioned figures are showing changes with
respect to change in dimensions and parameters selections. They also show variation
in the amplitude and phase accordingly.
A New Design of Equiangular Circular … 55

Fig. 1 Proposed structure of silica glass

Fig. 2 2D structure of proposed design

56 A. Jain et al.

Fig. 3 3D viewer of proposed design

Fig. 4 3D viewer of proposed design

A New Design of Equiangular Circular … 57

Fig. 5 3D viewer of proposed design

Fig. 6 3D viewer of proposed design

58 A. Jain et al.

Fig. 7 3D viewer of proposed design

4 Results and Discussions

For the proposed structure of silica glass fiber, the dispersion is so obtained and a
graph is plotted for the design which shows that it is zero for the selected diameter
range. The variation of dispersion is shown below in Fig. 8.
The refractive index profile of the proposed structure is obtained and plotted in a
graph which shows that it is quite similar to the silica glass, i.e., 1.457. It is shown
above in the Fig. 9.
The transmission curve of the proposed structure is shown below in Fig. 10.

5 Conclusion

Finally, it is concluded that the fiber parameters selected are yield to optimize and ana-
lyze the best-suited results among the data available. The discrepancy may observe
somewhere which is only due to higher wavelength region. Since we know that the
refractive index profile of the material and the effective cladding index are wave-
length dependent. It is also observed that silica glass PCF provides much better
A New Design of Equiangular Circular … 59

Fig. 8 Dispersion obtained

of proposed design

Fig. 9 Refractive index

Fig. 10 Transmission curve

of proposed design
60 A. Jain et al.

dispersion results as compared to others of the same structure and parameters. Here
we get nearly zero dispersion in between 0.6 and 1.6 µm wavelength region. We also
have a refractive index of silica as 1.457 at 0.2–1.2 µm wavelength region. Around
90% data transmission capability is assumed with the proposed structure.
Survey of Different Countermeasure
on Network Layer Attack in Wireless
Network

Amruta Chavan, Dipti Jadhav, Nilesh Marathe and Nilima Dongre

Abstract In recent times, wireless network builds a bridge between the physical
and virtual worlds. It poses security threats at different layers. The most vital secu-
rity issue is observed on the network layer on different wireless networked systems.
Defense mechanisms are used to shield the network layer from these repetitive mali-
cious attacks. Malicious hubs are detected and differentiated along the network by
using these mechanisms. These networks truly require a unified security answer
for ensuring both course and information transmission tasks on the network layer.
Without a suitable security solution, malicious nodes in the network can disturb
rapidly across the network. This just irritates the system activity from exact delivery
of packets and malevolent hubs that can disturb transmissions or drop all packets
going through them. The main objective is to outline the different attacks and the
countermeasures used to ensure the network security against pernicious attacks.

Keywords Security · Wireless network · Network layerattack · Countermeasure

1 Introduction

Security has been described as a safe environment to remain free from dangers posed
by adversaries purposefully or unknowingly. With the recent technological advances,
the systems never remain safe from assailants and from any unprotected framework

A. Chavan (B) · D. Jadhav · N. Marathe · N. Dongre

Department of IT Engineering, RAIT, Nerul, Navi Mumbai, India
e-mail: amruta.chavan1994@[Link]
D. Jadhav
e-mail: [Link]@[Link]
N. Marathe
e-mail: [Link]@[Link]
N. Dongre
e-mail: nilimraj@[Link]

© Springer Nature Singapore Pte Ltd. 2020 61

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
62 A. Chavan et al.

which can undoubtedly be disregarded from unapproved sources to take the data for
malignant purposes.
Network security alludes to the usage of network management for monitoring and
specialized measures to guarantee that information protection, integrity, and acces-
sibility, which can be ensured in a network domain. The security of the PC network
covers two viewpoints: physical safety and logical security. However, the network
suppliers are concerned with network data security, as well as how to manage sudden
catastrophic events, for example, a military strike against system equipment harm,
and how to reestablish network communication and keep up network communication
coherence in unusual circumstances.
The main advantage of a wireless network over a wired network is that the users can
move freely within the network with their wireless devices and access the internet
wherever they need. Users are allowed to share files between devices without a
physical connection. The installation time and cost of wireless networks are low
when compared to wired networks. In addition to these advantages, the wireless
networks are prone to several vulnerabilities discussed in subsequent sessions.

1.1 The Need for Security

The security is needed for the following given reasons:

• To protect vital information while still enabling access to trading secrets, medical
records, etc.
• To provide resource authentication and access control.
• To guarantee resource availability.

2 Analysis of Different Attack on Network Layer

2.1 Attacks

1. Blackhole Attack. In this assault, a pernicious node drops all data packets
through this as matter and energy go away in a black hole from our universe. In the
event, the assault node is connected between nodes of two parts of that arrange,
isolates the network, it separates the network effectively into two disengaged seg-
ments. All of the data from the fooled neighbor is sent to the deceitful node instead
of the destined base station. The sensor network’s specialized communication
pattern and multi-hop nature make it susceptible to this attack.
2. Wormhole Attack. This type of assault involves receiving information packets
on some point and invokes them to a different fake node. There is a channel or
tunnel between the two nodes called a wormhole attack [1]. The tunnel forms
Survey of Different Countermeasure … 63

between one or more malicious nodes during this attack. The distant nodes appear
as close neighbors so that this node’s energy resources are quickly exhausted. This
attack is effective when combined with selective forwarding and Sybil attack,
where detection is very difficult.
3. Sybil Attack. The attacker fools neighboring nodes with several identities [2].
By hacking the identities of the neighboring nodes, the attackers have access to
the corresponding node information. A defective hub or an opponent might have
several identities into show up and work as various separate nodes. The opponent
can then overhear communication or act maliciously after becoming part of the
network. The enemy can control the network significantly by displaying multiple
identities [3].
4. Grayhole Attack. The assailant draws in the information packets by publiciz-
ing himself the smallest way to the goal and after that captures and drops the
information packet. This assault is otherwise called a misbehavior attack, which
prompts to messages being dropped. The hubs drop the packets selectively in
this attack [4]. In the main phase, the hub advertises itself as having a substantial
route to the destination, while in the second phase, the hubs drop intercepted
packets with a specific likelihood.
5. Denial of Service Attack. An opponent tries to disturb communication in a
network by flooding the network with a large number of packages, for example.
When a service attack is denied, the attackers try to temporarily make the network
resources or a node or machine unavailable to their actual users. They send false
requests to the target so that it is unavailable to serve its intended users. The target
may be temporarily down or destroyed in such attacks. The routing process can be
interrupted on the network layer by changing the packet control routing, dropping
selectively, overflowing the table or poisoning [5].
6. Selective Forwarding Attack. In this attack, only certain packets are selec-
tively dropped by a malicious node. Usually, messages received are faithfully
forwarded by nodes in sensor networks. The message or information received
and sent should be the same for reliable and secure communication. When some
compromised node refuses to forward and this result in loss of important data
[2].
7. Information Disclosure Attack. During the communication process, every con-
fidential exchange of information must be protected. Critical information stored
in nodes must also be protected against unauthorized access. Control data are
sometimes more important for safety than traffic data. A node may release secret
or critical information into unapproved network hubs. Such data may contain
data on the topology of the system, the geographical area of hubs or ideal routes
to authorized hubs in the system.
64 A. Chavan et al.

Table 1 Comparative study of attacks in network layer

Types of attacks Attack threat Affects on Purpose Attack goal
Sybil Availability, Routing, voting, Unfairness; Control traffic
authenticity, fair resource disrupt the attack
integrity allocation, authentication
distributed
storage
Warm hole Confidentiality, Reveres Unfairness; Control traffic
authenticity engineering, disrupt attack
cipher breaking communication
to be
authenticated
Selective Availability, Multi-hoping Unfairness Data traffic
forwarding integrity protocols attack
Dos attack Availability, Essential Disrupt service Control traffic
integrity, services to be attack
confidentiality, authenticate
authenticity
Gray hole Availability, Normal Unfairness Data traffic
authenticity, communication, attack
integrity network node,
network
partition
Information Confidentiality, Normal Disrupt Data traffic
disclosure authenticity communication communication attack
to be
authenticate

2.2 Comparative Study of Attacks in Network Layer

As network layer is susceptible against dissimilar routing assaults, they can gain
access to routing paths and redirect the traffic, propagating fake routing information
into the system. The following Table 1 compares routing attacks on the network layer
based on a threat model, purpose, and security services.

3 Countermeasure Against Attack of Network Layer

3.1 Protection Agaınst Blackhole Attacks

A Detection, Prevention and Reactive AODV (DPRAODV) are considered a black

hole attack countermeasure. Topology graph-based anomaly detection (TOGBAD)
is planning a new centralized approach for identifying hubs attempting to make a
black hole. Adhoc routing protocol (SAR), which can be utilized to protect beside
Survey of Different Countermeasure … 65

black hole attacks, is also proposed. The black hole is a kind of steering attack in
which a vindictive node has the shortest route to all hubs in the earth by sending a
phony course response Thus, the noxious hub can deny the first hub traffic.
DPRAODV (Detection, Prevention, and Reactive AODV): In the Detection,
Prevention and Reactive AODV protocol, the threshold value is dynamically updated
[2]. It is intended to counteract black hole security dangers by telling different hubs
in the occurrence network. It also hinders the rehashed response from the malevolent
hub by diminishing system activity and in this manner, it disconnects the malignant
hub from the network. The results of the simulation in NS2 show that this convention
anticipates black hole attacks as well enhances the general execution of (typical)
within the sight of black hole attacks.
TOGBAD: It a new concentrated methodology utilizing topology diagrams to
detect black hole nodes. Utilize entrenched strategies to pick up learning of network
topology and utilize this information to perform believability minds on the routing
data spread through the network hubs. In this methodology must consider a hub that
generates false data about routing as pernicious. So, if the plausibility check fails,
this method must trigger an alarm. In addition, a hopeful first recreation result is
present. With this new methodology, the attempt to create a black hole can already
be detected before the actual impact before the genuine effect takes place.
SAR (Security-Aware Ad Hoc Routing Protocol): It can be utilized to secure
beside black hole attacks. The ad hoc routing protocol for security-aware depends
on request protocols, for example, AODV or DSR. In the greater part of the routing
protocol, they primarily expect to discover the most routes from source to goal,
which is the length of the route only. A security metric is extra to the RREQ bundle
in SAR and another route detection process is utilized. Intermediate hubs get an
RREQ bundle at a certain level of security or hope. In the event that the security
metric or hope level is fulfilled in intermediate nodes, the hub processes the RREQ
bundle and propagates it to its neighbors by means of prohibited floods. If not, the
RREQ bundle is dropped. If you can find a way with the required security properties,
the target will create an RREP bundle with the particular security metric. In the event
that the destination node does not discover a course with the required level of security
or confidence, it sends a warning to the sender and allows the sender to the sender
and enables the sender to modify the dimension of security to discover a route.

3.2 Protection Against Wormhole Attacks

To prevent the wormhole attack, WAP (Wormhole Attack Prevention) without uti-
lizing particular equipment is proposed. A TrueLink mechanism, a planning-based
defense for the wormhole assault is proposed. A packet with a leash protocol is
considered a treatment to the wormhole. The SECTOR component is proposed for
recognizing wormholes without synchronization of the timer. Directional antennas
are likewise designed to avert attacks with the wormhole.
66 A. Chavan et al.

Wormhole Attack Prevention (WAP): To prevent the wormhole attack, WAP

without utilizing particular equipment is planned. Not exclusively does the WAP
detect the phony route, but it also takes defensive measures next to the reappearance
of wormhole nodes amid the disclosure phase. The consequences of the simulation
demonstrate that wormholes can be detected and remotely detected during the route
detection stage.
Packet Leashes: Packet leashes for detecting wormhole attacks are proposed [6].
Mainly leashes mean a packet which limits a packet’s maximum transmission space.
A transitory packet leash sets a limit to a packet’s lifetime, which adds a limit to
its movement remove. A sender incorporates the time and area of transmission in
the note. The recipient checks if the packet has ventured to every part of the separa-
tion between the sender and itself inside the period among receipt and transmission.
Transitory packet leashes require tight tickers and exact area information. In geo-
graphic leashes, ensure that the receiver of the packets is some tickers and exact
area information. The area data and inexactly synchronized timekeepers confirm the
neighboring relationship in geographical leashes.
SECTOR Mechanism: This mechanism is primarily founded on separation bor-
dering procedures single-way hash chains and the hash tree of Merkle. It can be used
with no clock synchronization or area information to avoid wormhole attacks on the
Wireless network. It can also be utilized to anchor routing protocols in the wireless
network through recent encounters and to detect checks through topology following.

3.3 Protection Against Denial of Service Attacks

A DoSP-MAC convention is proposed to enhance network equality. A new firewall-

based format is offered as a countermeasure. In order to prevent a DoS attack, a DoS
moderation procedure utilizing a digital signature is proposed. Efficient on-the-fly
search technology to track DoS attackers was also proposed.
DoSP-MAC Protocol: The execution of the MAC (Media Access Control) pro-
tocol in wireless networks majorly affects generally network performance. In MAC
protocols dependent hubs’ access to the common channel isn’t synchronized and they
battle for the channel when packets are ready to be sent in their buffers. The quick
forward mechanism and the quick trade system are used to reduce self-contention;
however, these mechanisms exacerbate the issue of equality. It will, in this man-
ner, prompt refusal of DoS attacks on the grounds that the last champ is constantly
supported among neighborhood battling nodes, a continually transmitting node can
generally catch the channel and cause other nodes to perpetually disappear. In this
context, a DoSP-MAC protocol is proposed to enhance the reasonableness of the
network in order to improve its execution. This can enhance the medium usage
dramatically [7].
Novel System: A novel system dependent on a firewall. This firewall can discrim-
inate against the assault packets from the bundles sent by authentic clients dependent
on the packet’s stamping quality and along these lines sift through the greater part of
Survey of Different Countermeasure … 67

the assault packets. Our system is very efficient and has a very low cost of deployment
compared to other package marking solutions. In implementing this scheme, only
about 10% of Internet routers would need to cooperates in the checking procedure
and servers would need to make encrypted markings for secure transmission [6]. The
plan enables the firewall to be identified and keeps the first packet assaults on the
DDoS [8, 9].
Mitigation Technique: The digital signature mitigation method for confirming
legitimate packets and crash packets that do not pass the verification. Since nodes are
narrowing minded, they can’t verify the overhead so they can abstain from paying.
A terrible packet escaping verification throughout the entire network way will carry
a punishment for all its forwarders. A network diversion can be developed in which
hubs along a system way are urged to act all in all to modify terrible packets with
the end goal to improve their very own advantages. Analytical outcomes show that
Nash balance can be achieved for players in the proposed game, in which important
advantages can be given to with the goal that a significant number of the terrible
packets are verified.

4 Comparison of Different Countermeasure Techniques

Table 2 describes the attack type, merits, demerits and detection mechanism used
in each technique. Here each technique is proposed in order to detect and prevent
attacks at the network layer.

5 Conclusion

Although there are many strategies to counter security attacks, technological

upgrades are still required. The attacks act against secure communication data avail-
ability target. They drain the energy of the nodes and reduce the life of the network.
In this report, we endeavored to examine attacks and presented countermeasures of
security attacks on the network layer in the wireless system. Existing network layer
attacks like Sybil attack, Wormhole attack, Dos Attack, Selective forwarding attack,
Grayhole attack, and Information Disclosure attack are analyzed. These attacks affect
voting, routing, network resources, multi-hope protocols and so on. A comparative
study of the network layered attacks is based on the security classes, the location of
security, the nature of the attacker, the target of the attacker. In future, based on the
analysis of these vulnerabilities, an efficient intrusion detection mechanism can be
proposed to overcome most of these attacks.
68 A. Chavan et al.

Table 2 Comparison of different countermeasure techniques

Name of Attack type Detection Merits Demerits Remark
algorithm technique
DPRAODV Blackhole Checks if the Blackhole Identifies Fixed the
destination detection normal node threshold
sequence as a malicious value
number of node and enter
RREP is higher to blacklist
than the ALARM
threshold value broadcast
or not make network
overhead
TOGBAD Blackhole Compares the Blackhole The method is Decrease
number of detection ineffective in delay time
neighboring reacting
nodes it has protocol
with a number
of neighboring
nodes in
accordance
with topology
graph
SAR Blackhole Nodes checks Blackhole Cannot Decrease
if security detection guarantee detection
metrics or shortest route time
requirements discovery
are satisfied or
not
TELSA Wormhole Checks if the Wormhole Strict Detect in
recipient is in detection requirements required
certain distance in timing time only
or not
SECTOR Wormhole Bounding Wormhole Need special Maintain
maximum detection hardware to the accuracy
distance ensure the
between two accuracy of
neighboring time
nodes by series
of fast one-bit
exchange
DoSP-MAC DoS Nodes access DoS exacerbate the Improve
protocol to the shared detection issue of equality of
channel is not equality network
synchronized, services
and they
contend for the
channel
whenever there
are packets in
their buffers
ready to be sent
(continued)
Survey of Different Countermeasure … 69

Table 2 (continued)
Name of Attack type Detection Merits Demerits Remark
algorithm technique
Novel system DoS Compared to DoS Enables to be Improve
other detection identified and detection
packet-marking keeps the first framework
based solutions packet attack
on the DDoS
Mitigation DoS The digital DoS Confirm only Maintain
technique signature detection legitimate packet
mitigation packets delivery
method to ratio
confirm
legitimate
packets and
crash packets
that don’t pass
the verification

References

1. Singh RK, Nand P (2016) Literature review of routing attacks in MANET. In: 2016 interna-
tional conference on computing, communication and automation (ICCCA). Noida, pp 525–530.
[Link]
2. Meddeb R, Triki B, Jemili F, Korbaa O (2017) A survey of attacks in mobile ad hoc networks.
In: 2017 international conference on engineering & MIS (ICEMIS). Monastir, pp 1–7
3. Yao et al Y (2018) Multi-channel based sybil attack detection in vehicular ad hoc networks
using RSSI. In: IEEE Trans Mob Comput. [Link]
4. Kumar S, Goyal M, Goyal D, Poonia RC (2017) Routing protocols and security issues in
MANET. In: 2017 international conference on Infocom technologies and unmanned systems
(trends and future directions) (ICTUS). Dubai, pp 818–824
5. Wang Q, Dunlap T, Cho Y, Qu G (2017) DoS attacks and countermeasures on network devices.
In: 2017 26th wireless and optical communication conference (WOCC). Newark, NJ, pp 1–6
6. Korde S, Sarode MV (2017) Review on network layer attacks detection and prevention tech-
niques in mobile ad hoc networks. In: 2017 international conference on inventive systems and
control (ICISC). Coimbatore, pp 1–5
7. Chang S, Hu Y (2017) SecureMAC: securing wireless medium access control against insider
denial-of-service attacks. IEEE Trans Mob Comput 16(12):3527–3540
8. Simpson S, Shirazi SN, Marnerides A, Jouet S, Pezaros D, Hutchison D (2018) An inter-domain
collaboration scheme to remedy DDoS attacks in computer networks. IEEE Trans Netw Serv
Manage 15(3):879–893
9. Wang A, Chang W, Chen S, Mohaisen A (2018) Delving into internet DDoS attacks by Botnets:
characterization and analysis. In: IEEE/ACM Trans Netw. [Link]
2874896
10. SunilKumar KN, Shivashankar (2017) A review on security and privacy issues in wireless
sensor networks. In: 2017 2nd IEEE international conference on recent trends in Electronics,
Information & Communication Technology (RTEICT), Bangalore, pp 1979–1984. [Link]
org/10.1109/rteict.2017.8256945
11. Zhang L, Nie G, Ding G, Wu Q, Zhang Z, Han Z (2018) Byzantine attacker identification in
collaborative spectrum sensing: a robust defense framework. In: IEEE Trans Mob Comput.
[Link]
70 A. Chavan et al.

12. Surya SR, Magrica GA (2017) A survey on wireless networks attacks. In: 2017 2nd international
conference on computing and communications technologies (ICCCT), Chennai, pp 240–247
13. Soliman JN, Mageed TA, El-Hennawy HM (2017) Countermeasures for layered security attacks
on cognitive radio networks based on modified digital signature scheme. In: 2017 eighth inter-
national conference on intelligent computing and information systems (ICICIS). Cairo, pp
2–8
14. Kumar P, Tripathi M, Nehra A, Conti M, Lal C (2018) Safety: early detection and mitigation
of TCP SYN flood utilizing entropy in SDN. In: IEEE Trans Netw Serv Manage. [Link]
org/10.1109/tnsm.2018.2861741
15. Deeksha AK, Bansal M (2017) A review on VANET security attacks and their countermeasure.
In: 2017 4th international conference on signal processing, computing and control (ISPCC).
Solan, pp 580–585
16. Santhi G, Sowmiya R (2017) A survey on various attacks and countermeasures in wireless
sensor networks. Int J Comput Appl 159(7):0975–8887
17. Sinha P, Jha VK, Rai AK, Bhushan B (2017) Security vulnerabilities, attacks and countermea-
sures in wireless sensor networks at various layers of OSI reference model: a survey. In: 2017
international conference on signal processing and communication (ICSPC). Coimbatore, pp
288–293. [Link]
18. Tomić I, McCann JA (2017) A survey of potential security issues in existing wireless sensor
network protocols. IEEE Internet Things J 4(6):1910–1923. [Link]
2749883
19. Chen K, Zhang Y, Liu P (2018) Leveraging information asymmetry to transform android apps
into self-defending code against repackaging attacks. IEEE Trans Mob Comput 17(8):1879–
1893
20. Moudni H, Er-rouidi M, Mouncif H, Hadadi BE (2016) Secure routing protocols for mobile ad
hoc networks. In: 2016 international conference on information technology for organizations
development (IT4OD). March 2016, pp 1–7
21. Moubarak J, Filiol E, Chamoun M (2018) On blockchain security and relevant attacks. In: 2018
IEEE middle east and north africa communications conference (MENACOMM). Jounieh, pp
1–6. [Link]
Internet of Things: Service-Oriented
Architecture Opportunities
and Challenges

G. R. Sagar and N. Jayapandian

Abstract “Internet of Things” is now a subject that is increasingly growing on both

the job and modern devices. It is a concept that maybe not just get the potential
to influence how we live but in addition how we work. Intelligent systems in IoT
machines in many cases are used by various events; consequently, simultaneous
information collection and processing are often anticipated. Such a characteristic
that is exclusive of systems has imposed brand new challenges towards the designs
of efficient data collection processes. This article is to be discussing various layers
in Internet of things. Those layers are sensing layer, network layer, service layer
and application layer. Various data processing techniques are integrated along with
data filtering and data conversion. Protocol transformation is also feeling the major
challenges faced by enterprises wanting to shift to the style in brand new technology.

Keywords Internet of things · Service-oriented architecture · Security · Protocol ·

IoT layers

1 Introduction

The shift is in regards to both understanding and thinking about IoT. The IoT is
holds: IoT is really a network of connected devices which includes unique identi-
fiers in respect of an IP along with inbuilt technologies which help them to collect,
feeling information and communicate about the surroundings by which they prevail.
But within the broader sense and between the shifts pointed out, the IoT definition
does matter much if not seen with a bigger viewpoint [1]. Now it is perhaps not the
proper time for definitions, and it is time for development, value, business and IoT
in action. Over the last few years, there has been a great development in the area of

G. R. Sagar · N. Jayapandian (B)

Department of Computer Science and Engineering, CHRIST (Deemed to Be University), Kengeri
Campus, Bangalore, India
e-mail: jayapandian.n@[Link]
G. R. Sagar
e-mail: sagar.r@[Link]

© Springer Nature Singapore Pte Ltd. 2020 71

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
72 G. R. Sagar and N. Jayapandian

science and technology which has major impact on our daily lives either directly or
indirectly. Internet of things is one such technology which has major contribution to
this development. This IoT is mainly implemented in healthcare industry [2]. The
main idea of IoT is to connect the objects to Internet which will create a path for an
efficient interaction between the physical objects and the cyber-systems which are
also termed as cyber-physical-systems (CPS) [3]. The term IoT gained its popularity
by the value that it promises to create a scalable market and predict the future trend.
This will indeed create smart objects which can understand the user requirements and
act accordingly without explicit instructions. The term IoT has several definitions as
each author tries to concentrate on any one of its fundamental characteristics. Due to
this, it is often depicted as Web of Things allowing exchange of data. Although this
is the fundamental characteristic of IoT, the application of it is so diverse because
the solutions developed are totally dependent on the end-user needs which vary from
user to user. IoT can be broadly classified as consumer IoT (CIoT) and industrial IoT
(IIoT) where CIoT is more user-centric, that is, the interaction is between the user
and the objects which improve the awareness of the surroundings and saving both
money and time [4]. But IIoT is all about digital and smart manufacturing that is
creating a platform for operational technology and information technology to inter-
act [5]. As a consequence, digital manufacturing focuses on product life cycle and
creates ability to counter the dynamic changes in the requirement. Along with this,
the machine-oriented architecture of IIoT finds its application in different market
sectors. This in turn creates a path for understanding the manufacturing process and
enables sustainable and efficient production. In the past, ad hoc solutions were used
by the industries as it provided the necessary flexibility and scalability for IoT com-
munications. Recently, new standards such as WirelessHART and ISA100.11a were
introduced, and cellular technologies such as 4G and 5G are capable of connect-
ing devices over long range [6]. However, the limitations in terms of scalability and
long-range communications are to be solved as even the cellular technologies require
large infrastructure support and licence band. The large heterogeneous devices are
connecting and it provides following features, that is low cost, low power, reliable
and security.

2 State of Art

IoT has been built on various layers such as edge layer, gateway access layer, Internet
layer, middleware layer and application layer which are stacked upon each other.
This architecture is designed to meet the requirements of various domains such as
industries, governments, healthcare and so on. The architecture covers various aspects
such as scalability, modularity and interoperability among different devices using
different technologies. Physical layer—this is the bottommost layer in the stack [5].
This layer is similar to the network physical layer to convert the human sensor signals
into machine language. This physical layer is a main protocol of this IoT device,
Internet of Things: Service-Oriented … 73

because most of the IoT device is automated signal detection. This type of auto-
detection helps to convert the machine language. It is made up of sensor networks,
RFID and embedded systems which are deployed on the field to capture the data.
This sensor is the heart of the IoT, and without this sensor, there is no meaning in IoT
device. This IoT technology is the combination of computer and electronic devices.
This combination is implemented in mechanical, electrical, transport, logistics and
buildings. This sensor is also fixed in smart city, and the concept of smart city is to
detect the building activity and send it to the server. This signal is stored by cloud
computing technology. The combined IoT and cloud computing is not having local
server; it’s always depend on third party service provider. The purpose of third party
is to maintain the quality and lower the cost. Gateway access layer—next on the
stack is gateway layer. This gateway is an interface between one particular device
and server. This gateway is using some protocol system, and this protocol helps to
provide the better security. This security system is using some encryption algorithm.
The reason for using this algorithm is providing the information security while data
stored in cloud server. The reason for using this algorithm is third-party cloud server
usage [2] (Fig. 1).
This layer consists of various routers which are used for routing, publishing and
performing cross-platform communication when needed. The first level of data han-
dling takes place in this layer. Middleware layer—over the gateway layer comes
the middleware layer. This middleware is the concept of interconnecting the device
mechanism. This device middleware is used to establish and monitor the different
types of IoT device. There are many companies producing the IoT product in same
application. In that kind of situation, this middleware layer is helped to connect dif-
ferent devices with different company products. The most critical functions such as
information management, design management, data filtering, semantic analysis and

Fig. 1 Internet of things

layer
74 G. R. Sagar and N. Jayapandian

access control are taken care by this layer. As it is located in-between the hardware
layer and application layer, it also enables the bidirectional flow of data by acting
as an interface. Application layer—on the top comes the application layer which
is responsible for creating a platform for different users to access the data. These
are all some important layers of IoT device to enhance the network and device con-
nection. The individual device ID sensors are generating the encryption key for the
purpose of device authentication [7]. The healthcare industry is also using this IOT
device to monitor patients’ health information [8]. These layers are used in some
fingerprint-based security system [9].

3 Service-Oriented Architecture in IoT

Service Oriented Architecture (SOA) concept is used in different parts of IoT. This
is working on various domain and interoperability between complex heterogeneous
devices. This method is used to reuse software and impose any specific technology
during software implementation [10]. Sensing layer is smart sensors, and RFID tags
are connected to devices to gather the data automatically and provide information
exchange among heterogeneous devices. As number of sensors increases, the com-
plexity of connecting devices reduces. The IoT device is having the unique ID that
technology named as Universal Unique Identifier (UUID). This UUID is used to
track and monitor the IoT device. The goal and working principle of network layer
is to establish the connection in heterogeneous system in inter- and intra-network.
As IoT deals with heterogeneous environments, it is also important to involve QoS
management and control according to clients’ requirements. Security and privacy
issues are addressed in this layer along with other key features such as energy effi-
ciency, data and signal processing and QoS. This security problem is dealt with
symmetric and asymmetric encryption [11]. This QoS is major parameter in hybrid
encryption [12]. The middleware layer is a functional and important layer in service
layer structure. This particular layer is providing higher cost-effective platform in
both software and hardware [13]. It also takes care of other service-oriented issues
such as identification of devices that offer the required services, developing APIs
that can manage the interoperability among different services, data management and
information exchange and storage [14]. Interface layer IoT deals with heterogeneous
devices and many problems related to interoperability and information exchange
arises. This layer consists of user interface and application APIs [15]. This archi-
tecture also creates platform where every object’s functionality can be offered as a
standard service, and each service can be uniquely identified by a virtual element.
The IoT service-oriented protocol middleware problem is to solve various solutions.
The probabilistic protocol is one of the best solutions for this problem. Then face
heterogeneity is also dealt with this problem [16] (Fig. 2).
An emerging trend is to develop production systems integrated with sophisticated
electronics, interconnecting them and using them in the traditional business IT sys-
tems. As a result, IIoT creates a platform for flexible, low cost and resource-saving
Internet of Things: Service-Oriented … 75

Fig. 2 Service-oriented
architecture layer

productions. With the help of smart services, IoT can also improve the efficiency of
real-time data processing and can reduce the gap between the heterogeneous com-
ponents in current digital economy. Industries deployed with IoT systems follow
greener business models, are more profitable and optimise the resource usage. The
acceptance of this green technology totally depends on the protection of the private
data and information, the trust mechanism, communication security and security of
services and applications. This is also a major challenge in IoT because inter- and
intra-communications between every “thing” is via Internet, and the main source of
internet for IoT is through the conventional mobile networks and sensor networks.
Most of the services which are provided are for consumer IoT, and the same can-
not be used for the IIoT because the number of attack vectors increases rapidly. As
industrial applications require a reliable and secure data transfer methods, a num-
ber of research studies have been done to overcome these major challenges. Based
on the properties of spatial correlation, low-ranked structure and temporal stability,
a data monitoring algorithm for packet loss and re-establishing of sensor network
76 G. R. Sagar and N. Jayapandian

was proposed by Kong et al. Lu et al. designed a digital watermarking technique to

establish secure communication. The problem of continuous data collections where
the lower bounds for a single set of data and continuous data collections is derived
and also shortened the data collection process using multiple transceivers. Unlike in
traditional WSNs where the devices are maintained by one user, in IoT applications
the devices can be communicating with several devices deployed and users due to
which data aggregations and processing have to be executed concurrently on same
set of nodes. The delay-aware network structure can reduce the delays of collecting
data simultaneously.

4 Challenges and Opportunities

The only reason for accepting IoT by the industries is to increase their efficiency
and enhance their productions. Even though IoT promises to improve efficiency
and productions, there are major challenges that industries have to look in before
they leap in. These challenges are listed below. Real-time processing is one of the
main features of IoT which attracted the industries. This real time processing unit
is facing different challenges. That challenges are maintain reliability which help
to satisfy the end-to-end connection. Various packets scheduling algorithms such
as time-slotted packet scheduling are developed to achieve the desired QoS. As the
growth of IIoT was dramatically increased in terms of scale and complexity, the
level of difficulty in ensuring real-time performance has also increased. Inter- and
Intra-device communications devices being connected are increasing rapidly inter-
connecting all devices and tracking them becomes a major challenge. To overcome
these, smart devices which can detect, classify and migrate are required. The lack
of interoperability among the devices increases the cost factor and complexities of
the system deployment. Due to this achieving, seamless communication will be even
more complicated. Energy efficiency as IoT devices are deployed on site should be
constantly on for long years and the major power source is batteries, and it is an impor-
tant factor which cannot be ignored. Therefore, green networking plays an important
role in IIoT to develop a low power consuming and efficient devices at low cost.
Although there are numerous technologies developed for low power consumption
for WSN, these cannot be directly applied for IoT. Due to this, we need to establish
a high power energy generating units which can also produce high density energy
and use with the present low power nano-circuits and allows designing self-powered
smart sensors. Security and privacy are two major factors which define how pow-
erful the IoT system. Due to the large number of sensors and mobility, overcoming
these challenges is complex. Even though there exist a large number of encryption
algorithms, they have to be redesigned to make it faster and consume less power.
Privacy is the next factor which is causing hurdles for IoT. Heterogeneous devices
used in fields add more complexity to achieve privacy. Anonymity for data and net-
work has to be provided, and powerful authentication systems have to be developed.
The major significant of IoT device are connectivity, compatibility and longevity.
Internet of Things: Service-Oriented … 77

The communication devices and platforms are centred on the fundamental idea and
are being changed such that clients do not go through serious communication issues.
Currently, if a community utilising the client-server, users can communicate through
different channels. Issue that is primary would be the fact that it is well suited for
a number that is large. This technology should increase the quality of service up
to the maximum level. A problem that is similarly independent of the connectivity
would be the scalability. The final outcome of this method is not maintaining security
parameter in designing and development phase. Machines which are manufactured
keeping the security of the product in requirement documents get services and prod-
ucts, and this can be generic. They don’t come because of the storage that’s needed is
capability, computation energy, and system that is operating. This single application
device technology is used in many latest devices that is named as WiFi. Also same
technology is used in ZigBee. The heterogeneous devices are used to collect the
information during this process device compatibility is one of the major issue. The
applications required for connection purposes will not support the hardware used.
As there is huge amount of research and development happening all over the globe,
the technology being used today might not be reliable in the coming up years. There
is a huge scope for researchers in IoT because the technology has its applications in
various domains. Few of the areas are scalable communications, privacy and security,
identification technology, authentication, interoperability of heterogeneous devices;
powerful energy-efficient devices, etc. Developing a context-aware IoT middleware
is another area where a lot of research has to be done. Combining IoT, AI and cloud
can develop smart objects which are self-configured, self-healing, self-protection
and more capabilities.

5 Conclusion

IoT is made up by bringing multiple technologies such as information technology,

network and communications and heterogeneous devices all connected to Internet,
thus creating a new era of smart things. This study should take from nationwide fed-
eral government report. This report helps to improve the future IoT development. The
main development of IoT technology is depending on technological revolution. This
technology should provide energy efficient in heterogeneous devices. The primary
advantage of this technology is decreasing computational cost and secondary advan-
tage is increasing device efficiency. The growth associated with the IoT revealed
many brand new difficulties like the not enough fundamental principle encouraging,
ambiguous design and immature criteria. To satisfy these difficulties, we provide a
structure that is four-layer systems. This technology is maintaining standard protocol;
this protocol helps to establish the quality connections between one device to other
device. The continuing future of IoT will undoubtedly be anticipated to be unified,
smooth and pervading. Large-scale solution implementation has to be framed in just
a pair of requirements. Therefore, the improvements of IoT being a system that is
smart to be continuing with interoperability, power durability, privacy and safety. IoT
78 G. R. Sagar and N. Jayapandian

has grown to be a style that is unavoidable of data business, which bound to create
brand new changes in the way we live.

References

1. Whitmore A, Agarwal A, Da Xu L (2015) The internet of things—a survey of topics and trends.
Inf Syst Front 17(2):261–274
2. Bitra VS, Jayapandian N, Balachandran K (2018) Internet of things security and privacy issues
in healthcare industry. In: International conference on intelligent data communication technolo-
gies and internet of things (ICICI) 2018. Lecture notes on data engineering and communications
technologies 26. Springer
3. Lee EA (2008) Cyber physical systems: design challenges. In: Proceedings of symposium on
object oriented real-time distributed computing (ISORC). IEEE, pp 363–369
4. Garge GK, Balakrishna C, Datta SK (2018) Consumer health care: current trends in consumer
health monitoring. IEEE Consum Electron Mag 7(1):38–46
5. Palattella MR, Accettura N, Grieco LA, Boggia G, Dohler M, Engel T (2013) On optimal
scheduling in duty-cycled industrial IoT applications using IEEE802. 15.4 e TSCH. IEEE
Sens J 13(10):3655–3666
6. Petersen S, Carlsen S (2011) WirelessHART versus ISA100. 11a: the format war hits the factory
floor. IEEE Industr Electron Mag 5(4):23–34
7. Sridhar S, Smys S (2017) Intelligent security framework for IoT devices cryptography based
end-to-end security architecture. In: Proceedings of inventive systems and control (ICISC).
IEEE, pp 1–5
8. Bitra VS, Jayapandian N, Balachandran K (2018) Internet of things security and privacy issues
in healthcare industry. In: Proceedings of intelligent data communication technologies and
internet of things. Springer, pp 967–973
9. Jayapandian N, Rahman AMZ, Koushikaa M, Radhikadevi S (2016) A novel approach to
enhance multi level security system using encryption with fingerprint in cloud. In: Proceedings
of futuristic trends in research and innovation for social welfare (startup conclave). IEEE, pp
1–4
10. Teixeira T, Hachem S, Issarny V, Georgantas N (2011) Service oriented middleware for the inter-
net of things: a perspective European conference on a service-based internet. In: Proceedings
of European conference on a service-based internet. Springer, pp 220–229
11. Jayapandian N, Rahman AMZ, Radhikadevi S, Koushikaa M (2016) Enhanced cloud security
framework to confirm data security on asymmetric and symmetric key encryption. In: Proceed-
ings of futuristic trends in research and innovation for social welfare (startup conclave). IEEE,
pp 1–4
12. Jayapandian N, Rahman AMZ (2017) Secure and efficient online data storage and sharing
over cloud environment using probabilistic with homomorphic encryption. Cluster Comput
20(2):1561–1573
13. Duan R, Chen X, Xing T (2011) A QoS architecture for IOT. In: Proceedings of internet of
things (iThings/CPSCom), cyber, physical and social computing. IEEE, pp 717–720
14. Grønbæk I (2008) Architecture for the internet of things (IoT): API and interconnect. In:
Proceedings of sensor technologies and applications SENSORCOMM’08. IEEE, pp 802–807
15. Krco S, Pokric B, Carrez F (2014) Designing IoT architecture (s): a European perspective. In:
Proceedings of internet of things (WF-IoT). IEEE, pp 79–84
16. Issarny V, Bouloukakis G, Georgantas N, Billet B (2016) Revisiting service-oriented archi-
tecture for the IoT: a middleware perspective. In: Proceedings of service-oriented computing.
Springer, pp 3–17
Performance Comparison of Multicast
Routing Protocols Based on Route
Discovery Process for MANET

Kinjal Adhvaryu

Abstract Performance of routing protocols used for multicasting is highly depend-

ing on its route searching approach, which is the first step of routing process. Expand-
ing ring search (ERS) is one of the widely used concepts for route searching process
which is also used to reduce the broadcast overhead in MANET. To understand the
effect of route discovery process on the performance of routing protocols, we have
considered two different multicast routing protocols, namely energy-efficient ERS
(E2 ERS) and Multicast Ad hoc On-demand Distance Vector Protocol (MAODV)
which are widely used for multicasting in MANET. MAODV uses ERS approach
whereas E2 ERS has modified the ERS approach by modifying the transmissions
of RREQ during route discovery process. Here, we have used different parameters
like end-to-end delay, goodput and packet delivery fraction for varying node speed
and varied TTL_Increment values for varied number of receivers to compare the
performance of MAODV and E2 ERS protocols.

Keywords Goodput · Control overhead · Broadcasting · Ad hoc network

1 Introduction

Multicasting is one of the efficient ways to set up the group communication when
one sender node wants to send same set of data/messages to more than one receiver.
Performance of routing protocols used for multicasting is highly depending on its
route-finding approach which is the first step of routing process. The goal of the
ERS during route-finding process is to find mobile nodes that have the required route
information to the destination mobile node in their route table or the destination
mobile node by flooding of the route request packets (RREQs) in a controlled manner
[1–4]. To restrict the flooding of RREQs, TTL-based ERS is used by various routing
protocols in MANET. ‘The ERS based on TTL restricts its searching range by giving

K. Adhvaryu (B)
Computer Engineering Department, Sankersinh Vaghela Bapu Institute of Technology,
Gandhinagar, Gujarat, India
e-mail: kinjalvk@[Link]

© Springer Nature Singapore Pte Ltd. 2020 79

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
80 K. Adhvaryu

Fig. 1 Route discovery in

MAODV

RREQs with a predefined TTL value. The TTL value indicates the radium of a
searching space. Each time if it fails to find any destination node or any node that has
route information to the destination node, the source node rebroadcasts the RREQ
with an increased TTL number to allow the RREQ to encoder a larger area. Even
with the controlled manner of flooding, expanding ring search still suffers from high
overhead because each time route discovery process starts from the source node.
Research says that when a larger area of the network needs to be searched, the cost
drastically increases’ [1, 4].
Figure 1 shows the working of ERS which is used by MAODV protocol. Here,
we assume that node S wants to find the destination node. Then it will start its
searching process by broadcasting the RREQ to all nodes which are one hope away,
if TTL_START = 1. Else if the target node is not found, then TTL will be incremented
based on TTL_Increment value. Based on the updated TTL value, again node S will
broadcast the RREQ. Node S will repeat the same process until TTL crosses the
TTL_Threshold or in between destination node is found [4–8]. From Fig. 1, we
can see that more number of RREQ packets is required during the route discovery
process.

2 Energy-Efficient ERS Expanding Ring Search (E2 ERS)

Here, an alternative energy-efficient ERS scheme is proposed which reduces the

overhead during route discovery, and in this way, it reduces the processing load
on mobile nodes and provides support to reactive types of MANET protocols. The
Performance Comparison of Multicast Routing … 81

Fig. 2 Route discovery in

E2 ERS

basic route-finding structure of energy-efficient ERS (E2 ERS) is similar to that of

the TTL-based ERS. Main differences from the TTL-based ERS are that the E2 ERS
does not transmit the RREQ every time during its routing search procedure from
the sender/source node to all other mobile nodes. Instead of that E2 ERS unicast the
RREQ, if destination/target node is not found then it broadcast the RREQ and find
the destination/target node. Still destination/target node is not found then unicast the
RREQ and find the destination/target node. In this way, alternatively unicasting and
broadcasting of RREQ will be done instead of every time broadcasting of the RREQ
to search the destination/target node during path searching process.
In Fig. 2, optimized ERS based on modifying the forwarding RREQ packets is
shown. Instead of broadcasting the RREQ packets, E2 ERS is unicasting the RREQ,
then broadcasting the RREQ, then unicasting the RREQ, then broadcasting the RREQ
and so on until interested node is found. Black arrow shows the unicasting the RREQ,
and blue arrows show the broadcasting the RREQ. During unicasting, RREQ will
be forwarded to furthest node from the neighbor list based on current TTL value.
To justify the proposed concept, assume ad hoc network with network diameter 5
and consider total of 5 mobile nodes in radio range of every node, where network
diameter is the longest path between any two mobile nodes. To find out the route to
the farthest node located 5 hops away from the source node, broadcasting RREQ will
start with TTL equal to 1 and increasing by TTL_Increment with each re-broadcast
of RREQ up to the predefined network diameter or TTL_Threshold.
Now for designed network, total 4 + 8 + 20 + 32 + 68 = 132 transmission of
RREQ will be required by E2 ERS instead of 4 + 16 + 52 + 160 + 484 = 716
transmissions of RREQ done by TTL-based ERS (as shown in Table 1) which shows
the effective reduction in control overhead during route discovery process. Due to
82 K. Adhvaryu

Table 1 Required transmissions of RREQ

Diameter: 5, 5 nodes in radio range of every node
TTL Traditional ERS Modified ERS
1 4 4 Unicasting and 4 4
2 4 + (4 * 3) 16 broadcasting 4+4 8
alternatively
3 4 + 12 + (4 * 3 * 3) 52 4 + 4 + (4 * 3) 20
4 4 + 12 + 36 + 160 4 + 4 + 12 + 12 32
(4 * 3 * 3 * 3)
5 4 + 12 + 36 + 108 484 4 + 4+ 12 + 12 + 68
+ (4 * 3 * 3 * 3 * 3) (12 * 3)
Total PKTS 4 + 16 + 52 + 160 + 484 4 + 8 + 20 + 32 + 68 = 132
= 716

reduction in control overhead, processing load on mobile node will be reduced, and in
this way, lifespan of mobile node will be prolonged. And this difference in reduction
will go on increasing for bigger values of network size.

3 Performance Comparison

The performance of E2 ERS-based protocol is compared with the known multicast

routing protocol MAODV which uses the TTL-based ERS. For the simulation of the
E2 ERS-based protocol, NS-2.26 simulator has been used. Data traffic was generated
using constant bit rate (CBR) UDP traffic with 32 network size with one sender and
2, 5 and 10 varying mobile nodes acting as receivers in the multicast group. All
wireless mobile nodes are randomly distributed in a square of 500 m × 500 m. The
nodes use the IEEE 802.11 radio and MAC model provided by the CMU extensions.
Each simulation executes for 200 s [5, 6, 9, 10]. The number of receiver mobile nodes
is varied from 2 to 10 nodes to see the effect of the number of receiver nodes on
the performance on the system performance. Also other parameter TTL_Increment
value is varied to see the performance of proposed concept. TTL_Increment value
is varied with 1, 2, 3, 4 and 5 and each time, proposed concept is compared with
original approach for various performance metrics. To analyze the effect of number
of receiver nodes, another node speed parameter is also considered with 4, 12 and
20 m/s varying values. The metrics used for performance evaluation are: (i) Packet
delivery fraction—‘the ratio obtained by dividing the number of data packets cor-
rectly received by the destination by the number of data packets originated by the
source’ [5]. (ii) Goodput—‘which is also known as throughput [8, 9, 11–13] of the
network.’ (iii) Average end-to-end delay—‘this includes all possible delays caused
by queuing delay at the interface, buffering during route discovery, retransmission
delays at the MAC, propagation and transfer times’ [2, 3, 9, 14, 15].
Performance Comparison of Multicast Routing … 83

Figures 3, 4 and 5 show the comparison between E2 ERS and MAODV protocols
for 2, 5 and 10 receiver nodes for varying node speed values for all three metric
values. From results, it is clear that E2 ERS gives better results compared to MAODV
for more number of receiver nodes with increasing node speeds.
Figures 6, 7 and 8 show the comparison between E2 ERS and MAODV protocols
for 2, 5 and 10 receiver nodes for varying TTL_Increment values for all three metric
values. From results, it is clear that E2 ERS gives better results compared to MAODV
for more number of receiver nodes with increasing TTL_Increment values.

Fig. 3 PDF improvements PDF Improvement (%) in E2ERS V/S MAODV

80
10R 5R 2R
PDF Improvement (%)

60

40

20

0
1 2 3 4 5
TTL _INCREMENT

Fig. 4 Goodput Goodput Improvement (%) in E2ERS V/S MAODV

improvements 80
Goodput Improvement (%)

10R 5R 2R
60

40

20

0
1 2 3 4 5
TTL_INCREMENT

Fig. 5 E2E delay decrement E2E Delay Decrement (%) in E2ERS V/S MAODV
E2E Delay Decrement (%)

80
10R 5R 2R
60

40

20

0
1 2 3 4 5
TTL_INCREMENT
84 K. Adhvaryu

Fig. 6 PDF improvements PDF Improvement (%) in E2ERS V/S MAODV

80

PDF Improvement (%)

10R 5R 2R
60

40

20

0
4 12 20
NODE SPEED

Fig. 7 Goodput Goodput Improvement (%) in E2ERS V/S MAODV

Goodput Improvement (%)

improvements 80
10R 5R 2R
60

40

20

0
4 12 20
NODE SPEED

Fig. 8 E2E delay decrement E2E Delay Decrement (%) in E2ERS V/S MAODV
80
10R 5R 2R
E2 Delay Decrement (%)

60

40

20

0
4 12 20
NODE SPEED

4 Conclusion

From the presented results, it is clear that selection of route discovery process affects
the performance of routing protocols. Apart from it, number of receiver nodes and
selection of node speed also affect the outcome of multicast routing protocols. It
is also clear that E2 ERS, by modifying the expanding ring search technique which
is used by MAODV, which is state of the art tree-based multicast routing protocol,
Performance Comparison of Multicast Routing … 85

reduces the transmission of RREQ packets. Due to reduction in required transmis-

sions of RREQ during route discovery process, processing load on mobile nodes
decreases and so that the lifetime of mobile nodes and of network is also prolonged.
It also leads less chance of network clogging and less overheads as well and makes
routing process more energy efficient.

References

1. Junhai L, Danxia Y, Liu X, Mingyu F (2009) A survey of multicast routing protocols for mobile
ad-hoc networks. IEEE Commun Surv Tutorials 11(1)
2. Moh S, Yu C, Lee B, Youn HY (2002) Energy efficient and robust protocol for mobile ad hoc
networks. In: Proceedings of the 2002 Pacific Rim international symposium on dependable
computing (PRDC’02), IEEE. 0-7695-1852-4/02
3. Hwang I-S, Pang W-H (2007) Energy efficient clustering technique for multicast routing
protocol in wireless ad hoc networks. IJCSNS Int J Comput Sci Netw Secur 7(8)
4. Pu I (2007) Energy efficient expanding ring search. In: First Asia international conference on
modelling & simulation (AMS 07), 03/2007
5. Bakht H (2011) Survey of routing protocols for mobile ad-hoc network. Int J Inf Commun
Technol Res 1(6)
6. Wang N-C, Chen Y-S A power-aware dual-tree-based multicast routing protocol for mobile ad
hoc Networks. IET Commun 6(7):724–732
7. Adhvaryu KU, Kamboj P (2017) Performance comparison between multicast routing protocols
in MANET. In: IEEE international conference on electrical, computer and electronics—2017
(UPCON 2017). GLA University, Mathura, 26–28 Oct 2017
8. Sreedhar GS, Damodaram A (2011) Tree based multicast routing protocols for mobile ad hoc
networks and current state of the art. Int J Sci Adv Technol
9. Nagaratna M, Kamakshi Prasad V, Raghavendra Rao C (2011) Performance evaluation of
tree—based multicast routing protocols in MANETs. IJCST 2(3)
10. Vassileva N, Barcelo-Arroyo F (2008) A survey of routing protocols for maximizing the lifetime
of ad hoc wireless networks. Int J Softw Eng Appl 2(3); Young M (2002) The technical writer’s
handbook. Mill Valley, CA
11. Wei W, Zakhor A (2007) Multiple tree video multicast over wireless ad hoc networks. IEEE
Trans Circ Syst Video Technol 17(1)
12. Moustafa H, Labiod H (2003) A performance comparison of multicast routing protocols in
ad hoc networks. In: Proceedings of IEEE personal, indoor and mobile radio conference, pp
497–501
13. Adhvaryu KU, Kamboj P (2017) Effect of node speed on performance of multicast routing
protocols in MANET. In: IEEE international conference on telecommunication, power analysis
and computing techniques—2017 (ICTPACT—2017). Bharath Institute of Higher Education
& Research, Chennai, 6–8 Ap 2017
14. Qabajeh MM, Abdalla A, Khalifa O, Qabajeh LK (2011) A tree-based QoS multicast routing
protocol for MANETs. In: 2011 4th international conference on mechatronics (ICOM)
15. Adhvaryu KU, Kamboj P (2013) Survey of various energy efficient multicast routing pro-
tocols for MANET. In: Fifth international conference on advances in recent technologies in
communication and computing, ARTCom2013. Bangalore published by IET, Elsevier
16. Adhvaryu KU, Kamboj P (2016) Efficient multicast routing in mobile ad-hoc network. J Mob
Comput Commun Mob Netw 3(3). ISSN: 2349-901X (online)
New Complex Hybrid Security Algorithm
(CHSA) for Network Applications

Haider K. Hoomod, Ali Jameel Al-Mousawi and Jolan Rokan Naif

Abstract The raising of Cyberwarfare level require to have the maximum protec-
tion levels in the networks and its applications. Therefore, these systems must be
protected from unauthorized usage of their data using the most powerful security
algorithms. Many and growing challenges facing cybersecurity require cryptogra-
phers and data security engineers to develop stronger and more powerful algorithms
to resist this type of attack. There are many characteristics that measure the strength of
the electronic security system, including the speed of the algorithm and the strength
of the system against the cryptanalysis. This paper discusses the design, building,
and implementation of a new hybrid algorithm that combines a number of security
algorithms in order to obtain the best results in terms of resistance to multiple cyber-
attacks. This paper also contains an analytical study of the algorithms used and their
effectiveness in resistance to attacks represented by statistical results and graphs.

Keywords Hybrid algorithm · Cybersecurity · Network security · System security

1 Introduction

Many electronic systems around the world have been exposed to cyber breaches by
hackers or by some malicious programs that have enabled hackers to access these
systems and use their data unauthorized. Most electronic systems rely on the principle
of encryption algorithms to protect their data and convert clear data into encrypted

H. K. Hoomod (B)
Computer Science Department, College of Education—Mustansiriyah University, Baghdad, Iraq
e-mail: [Link]@[Link]; drhjnew@[Link]
A. J. Al-Mousawi
Information Technology Regulation Directory—Communication and Media Commission of Iraq
(CMC), Baghdad, Iraq
e-mail: aburgiefali@[Link]
J. R. Naif
Informatics Institute for Postgraduate Studies-Baghdad, Baghdad, Iraq
e-mail: newjolan@[Link]

© Springer Nature Singapore Pte Ltd. 2020 87

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
88 H. K. Hoomod et al.

data that can not be decoded into using special arguments called cryptographic keys.
These systems vary according to the nature of the security algorithms used. Some of
them use cryptographic algorithms of great complexity and high strength. They are
not fast, and those who use fast security algorithms are relatively less safe compared
to their complex predecessors [1].
In order to design a new algorithm, consider the analytical studies of existing or
previously used algorithms and compare them to find the strengths and weaknesses
of each algorithm. As it is known, each algorithm has a number of features and
some disadvantages. The hybrid algorithm is a combination of the properties of
the existing algorithms to take advantage of the characteristics of these algorithms
in improving the results of the new algorithm [2]. For example, when combining
two cryptographic algorithms, one of these algorithms has certain disadvantages,
and their integration with another algorithm may be a reason to address these
disadvantages. It is also possible that new defects will emerge as a result of the
process of integration of the algorithms, so the process of hybridization of algorithms
in the case of increased disadvantages is unsuccessful.
The characteristics of the hybrid algorithm depend mainly on the type of selected
Algorithms. In the hybridization process [3]. The more powerful algorithms are
selected, the better the results are obtained. There are systems based on speed in the
work where the speed is a sensitive issue cannot be compromised in terms of work
so not all algorithms can work efficiently in this type of systems, an example these
systems are wireless systems and most of the Web sites. On the other side, there is
another type of system that requires maximum security regardless of time. It requires
security algorithms with a very high level of complexity and a high level of security
in order to provide the security requirements of information and data such as military
systems and research facilities systems [4].
There are some basic characteristics that must be met in cryptographic algorithms.
One of the most important features is its strength in the face of code analysis and ran-
domization in generating cryptographic keys, increasing randomization and confu-
sion within the algorithm’s body and other properties. Algorithms that are adopted on
systems of a sensitive nature must have these characteristics at their highest levels [5].

2 Related Work

In this section of this research, it is important to look at the previous studies in

this area, i.e., the most prominent security algorithms in which there is a kind of
hybridization in order to know what characteristics have emerged and what are the
disadvantages that emerged as a result of the hybridization process in the security
algorithms.
Jignesh R. Patel, Rajesh S. Bansode and Vikas Kaul have invented a hybrid
encryption system based on a block cipher [6]. This system was built based on
two main block cipher algorithm which is advanced encryption standard (AES) and
data encryption standard (DES) by merging the two techniques together attempting
New Complex Hybrid Security Algorithm … 89

to have both advantages of both systems. In this experiment, a hybrid block cipher
algorithm deals with 64-bit input message were each input message that exceeded
64 bit will be divided into equal size 64-bit blocks. The key size of the algorithm is
128-bit size. The results showed that the hybrid algorithm takes more time to process
the data then each of its components because the processing time increased compared
with the single algorithm. The resultant from the hybridization process shows that
the complexity level increased and the security level increased compared with its
component.
Kirtiraj Bhatele, Amit Singhal, and Mayank Pathak in their paper design an archi-
tecture for a hybrid encryption system using multiple algorithms [7]. The algorithms
used are AES, RSA, Elliptic Curve (EC), and additionally, it uses an MD5 algo-
rithm for hashing the component of the encryption algorithm. This architecture also
takes into consideration the use of duel RSA algorithm and its effect compared with
normal RSA. The proposed hybrid system generates a hash value using MD5 and
places it within the encryption algorithms. Meanwhile, the plaintext enters the AES,
RSA and Elliptic curve cryptographic function in order to be encrypted by using the
encryption key. The plaintext is retrieved at the end of the algorithm and by using
of MD5 authentication. When the hash value approved, the Ciphertext decrypted to
retrieve the original plaintext. The resulting ciphertext of this system is resistance to
the square attacks. The algorithm use high bandwidth and low power consumption
as a measurable advantage of this hyper system. A disadvantage of this system it
is noted that the expected execution time is as twice as the normal execution time
of its component. This make this algorithm suitable for a high-security system but
in the same time it is unsuitable for applications and systems that require real-time
execution or fast execution of security algorithm.
Prabhat K. Panda and Sudipta Chattopadhyay create a hybrid encryption algorithm
based on RSA and public-key cryptographic systems [8]. The merging in this process
made by using four prime numbers in order to generate both of public and private keys
so as the complexity in key generation increased. The hybrid RSA (HRSA) results
produced from the experiments made by the researchers show that HRSA is more
secure than normal RSA and also the time taken for the encryption and decryption
process is close to the original execution time of normal RSA. The hybridization
came from the idea of inserting new arguments to the original RSA by using the
properties of prime numbers to generate secure keys. Another focal point that is
noticed in this algorithm is that the randomness in key generation increased and as a
result of that affect the attack resistance capability increased because whenever the
randomness period increased the more powerful the algorithm will be resistance to
all attacks that depend on the predictability.

2.1 Proposed Work (CHSA Design)

Each algorithm has a specific structural security representation that determines the
mode of data flow within it. The more algorithm structure parts correlated together
90 H. K. Hoomod et al.

the more algorithm is faster in execution. The algorithms whose structure is intercon-
nected have a fast execution time (encryption–decryption) because of this correlation.
On the unrelated or low-correlation algorithms implementation time will be more
than the previous one, but the complexity and strength of the algorithm will be very
large because the process of code analysis will be very difficult resulting from the
structural parts of this algorithm are not understood intelligently. There is a kind of
interconnection between the structural parts of the algorithm that is incomprehen-
sible to the normal algorithm tracer or is very difficult for the code analyst, which
makes this algorithm has a strong structure and thus will be difficult to break. The
general structure of CHSA consists of six basic classes summarized as follows:
1. Block cipher: the truncating of the message made by using block cipher. The
message truncated into the equal-sized block in order to be processed by the other
algorithm steps [9]. All message types processed by the algorithm withier if it is
a multimedia or a string file. The string file truncated into blocks and processed
to the other steps as a binary string while the multimedia file are encoded and
converted into equal size binary blocks. Meanwhile, a conversion method used to
convert the converted binary message into a hexadecimal form. For mathematical
and programming reasons, all messages need to be in hexadecimal form to be
compatible with other algorithm arguments.
2. Stream cipher: after processing the truncated message into hexadecimal blocks
the next step is to process the block in stream cipher form. Each character in the
block processed one-by-one using a stream cipher algorithm. The stream cipher
increased the speed in execution time and therefore it decreased the overall time
needed to encrypt the message and convert it into ciphertext form. Stream cipher
algorithm needs to have an initial seed to encrypt the message using this seed as
a key for the encryption process. This step considers the first encryption in the
CHSA using a stream cipher method. The stream cipher key generated using the
CHSA key generation method. CHSA provided with a method for the random
key generation whatever it is requested. The stream cipher algorithm used in this
step is RC4 [10]. The algorithm sends a request to the key generation method
within CHSA in order to receive a random key according to the algorithm type
(more details in Sect. 3).
3. S-Box confusion: one of the basic steps in all cryptographic algorithms is the
confusion [11]. The confusion added by using the s-box matrix. Normal algo-
rithms use pre-build s-boxes in order to distribute the change over the current data
chunk. S-boxes in CHSA is randomly generated by the key generation method.
The generated pattern will be organized into a matrix. The main objective of
the s-boxes in CHSA is to increase the confusion of the message structure and
process the message to the next step. The increase in confusion makes it harder
to analyze the data chunk in case of sniffing during the data transmission or
cryptanalysis attack.
4. Add encryption key: this is the stage where the message is truly encrypted.
Each encryption algorithm takes its strength from the encryption key. The more
the encryption key is strong the more strength will gain against cryptanalysis
New Complex Hybrid Security Algorithm … 91

and more resistance to all known attack types. At this stage, CHSA requesting
from the generation center the encryption key to encrypt the current block. The
complexity of the hybrid algorithm is that CHSA requests a key for each block
that is under process. This would add a lot of complexity to the system meanwhile
it required to generate multiple encryption keys (the same number of blocks).
This behavior within CHSA is similar to the behavior of one-time pad algorithm
[12] which considered unbreakable algorithm since the generated encryption key
is as long as the message and used only once in encryption and decryption and
then destroyed. The generated hexadecimal key added to the current block which
himself is a hexadecimal quantity and therefore an X-OR operation performed in
this stage. Generated ciphertext blocks transferred into another step to perform
the chaining process within other blocks.
5. Block organization: the resultant blocks from the previous step will be organized
into a chain of the block. Each message block enters the algorithm separately from
the other blocks. CHSA works in the parallel mode were each block processed
and encrypted independently from other blocks. Each block of the data needs to
correlate with other blocks in order to produce the final block cipher. The known
methods for the block cipher called mode of operations. These modes used to join
the blocks after they are encrypted in a manure that can be decrypted later using
the same method in reverse. However, the chaining process of the block cipher
considered a focal point in the algorithm secrecy [13]. CHSA joining the blocks
in a random pattern way generated by the generation center of CHSA. When
the block generated and attended to be chained with other blocks the algorithm
generate a request to the generation center and receive a pattern of the block
order to join the blocks together and when decrypt the original message it will
use the same pattern generated previously to retrieve the original blocks.
6. Probability distribution: in mathematics, the probability is the mean of an event
that may or may not occur according to a specific condition [14]. Adding proba-
bility distribution to the algorithms increases the confusion level along with the
randomness of the hall encryption systems [15]. CHSA added this feature to the
ciphertext by using CSHA probability distribution sector in the algorithm body.
This processing method generating a random pattern for the distribution of the
ciphertext algorithm. The input block will be permutated according to the gen-
erated pattern from the distribution center. Probability plays a central role in the
changing event of the block cipher parts. After the chaining process completed,
the ciphertext block enters the probability distribution stage which examines the
ciphertext block according to the pattern. If the probability occurs then the block
is merged with the distribution pattern generated from the probability distribution
center. Probability distribution center have a communication link with the block
organization stage because of shared objectives, the distribution center request
from the block organizer to provide the distribution center with the block bound-
aries in order to identify the block border and act according to this border [14, 16].
Whenever the block boundaries received a pattern will be generated automati-
cally and randomly. Ciphertext changed according to this pattern. The form of
change occurs by the probability distribution pattern is multi-form change. What
92 H. K. Hoomod et al.

this means is that it contain multiple actions to change the ciphertext like shift-
ing, simple stream cipher encryption, substitution encryption, or transposition
encryption. The generated pattern need a key in case of substitution, transposi-
tion, or even stream cipher so this step needs to be provided with keys from the
generation center within CHSA.
All details of CHSA algorithm structure and general design appear clearly in
Fig. 1. Connections, steps, and data flow is represented in details.

2.2 CHSA Generation Center (GC)

The GC is the main nerve of the new hybrid algorithm as it generates all kinds of
keys and patterns used by the algorithm in the performance of various functions.
GC receives the generation requests from the joints of the system. Each of these
joints has a special request to the GC to generate the cryptographic keys. It generates
the patterns of the process of reallocating the bits in the encoded text during the
conversion process from its current form to a more complicated one. The process of
reallocating bits in the current message form requires a pattern generated by the GC
for each step that requires reallocation. Mainly, there are three main steps in CSHA
that requires a pattern for the distribution process. In another hand, there are three
main phases that require keys for encryption.
The hybrid chaotic system used in this proposed system is composed of the Logis-
tic system (Eq. 1) and Lorenz system (Eqs. 2, 3, and 4) as shown in Fig. 2. The Logistic
chaos outputs are used in two manners (as KS1 and KS5). First, they are used for
encryption processing. Second, they are used to initialize the Lorenz chaotic system.
The last significant number of each result from the Logistic chaotic system is added
to the initial periods of the Lorenz initial conditions

f (kn+1 ) = r · kn (1 − kn ) (1)

xn+1 = a · (y − xn ) (2)

yn+1 = c · x − x · z − y (3)

z n+1 = x · y − b · z (4)

where a, b, c, r is the chaos parameters. While x 0 , y0 , z0 , k 0, and v0 is the initial

conditions for chaos map.
The GC consists mainly of two main sectors. The first sector is the Chaos Random
Generation Center (CRGC) which is responsible for providing random keys from
hybrid chaotic system, the patterns depending on random key and pattern generation
algorithm. Blum-Blum algorithm summarized in the following equation:
 Plaintext Message
PDF, JPEG, PNG, WAV, MP3, MP4, etc.

Probability DistribuƟon
Re-distribute the ciphertext according to the probability paƩern. If the paƩern contains a changing in some
blocks within the ciphertext then it is executed and stored in the storing center. PaƩern generated from the
generaƟon center
TruncaƟng Step
TruncaƟng message into equal sized block Probability PaƩern ((b2,enc(RSA, key)), ( b5,enc(RC4,key) …)

Ciphertext Message
Block #1 Block #2 .... Block #n

GeneraƟon Center (GC)

Responds to all requests from the algorithm steps and respond to them by
providing all keys s-boxes, and paƩerns Block Organizer
Stream Processing Step OrganizaƟon of the blocks is performed at
Using RC4 to process and encrypt each block on stream cipher method Chaos Random GeneraƟon Center (CRGC) this step. The paƩern of chaining between
blocks generated by the generation center
Generate random keys , paƩerns, and S-Boxes randomly
New Complex Hybrid Security Algorithm …

Block #1 "∑ STREAM_KEY (k1, Block_1)

Storing Center (SC) Block Chaining PaƩern
(b1,b4) , (b5,b2) , ……..
Stores all the informaƟon generated from each step in order to retrieve these
The key is generated from
informaƟon when decrypt the ciphertext
generaƟon center
RC4 Algorithm

Cipher_Block #n

1A976F3E418B……….
Random Hexadecimal key as long as the block size and a key per block is generated Cipher_Block #3
Confusion Processing Step
Emerging confusion to the block using S-Boxes generated randomly from generaƟon
Cipher_Block #2
center
D_Block #1 "∑ Cipher_Block #1
Block #1 DistribuƟon D_Block #1 Cipher_Block #1

Adding EncrypƟon Key

[S-box #1] Adding the encrypƟon key generated by the generaƟon center. A key is gen erated
An S-Box created for each block generated by the for each block so the final key will be as long as message
generaƟon center

Fig. 1 General CSHA structure with the details of each step

93
94 H. K. Hoomod et al.

Parameters and initial value for Logistic chaos system

2D Logistic Chaotic System KS5

Generate Initial Values for Lorenz

system
KS1

Parameter 3D Lorenz Chaotic System

KS2 KS3 KS4

Fig. 2 The hybrid chaotic system used in the proposed system

z n+1 = z n2 mod M (5)

where M = p ∗ q and p, q are two large prime numbers. The seed z n2 should be
an integer value that its value is co-prime with M. The main purpose behind using
Blum-Blum hybrid chaotic system algorithm is its properties of generating a random
number that approximates the properties of the true random number generation from
hybrid chaotic system [13, 17].
The second sector in the GC is the Storing Center (SC) which is responsible
of storing form holding the key streams and patterns generated by the CRGC. The
storing strategy takes the form of table with identifiers in order to identify the type
of storing operation. Each record in the database of the system contains key streams
or a pattern composed with it the type of the algorithm that request it and the step of
the request. Figure 3 shows a detailed structural view of the GC and its components.

Generation Center (GC)

Responds to all requests from the algorithm steps and respond to them by providing all
keys s-boxes, and patterns

Chaos Random Generation Center (CRGC)

Generate random key , patterns, and S-Boxes randomly

Id Type Block Keystream or Pattern

Storing Center (SC) 3 Key B1 A193E1F9912DB14236
Stores all the information generated from each step in order to retrieve these 4 Pattern B3 ((B1,[1,A,3,D,E,9,6,F]))
information when decrypt the ciphertext
5 Key B5 5438762A2ED3C5691A
6 Pattern B11 ((B5),[4,A,6,3,1,E,C]
7 Key B4 B234876E9FE2321129A

Fig. 3 GC internal design and storing strategy inside SC unit in the GC saves the data on a table
with type of the storing element, block that is related to the stored element, and the generated stream
or pattern
New Complex Hybrid Security Algorithm … 95

2.3 CHSA Syntax

The data flow within CHSA goes from truncated blocks and ending with ciphertext
blocks. The encryption process takes six basic steps. All steps within CHSA are
represented in the following algorithm.
a. CHSA Encryption
Below the proposed CHSA encryption algorithm.
Algorithm 1: CHSA Encryption Syntax

The encryption process takes an action under for loop as long as the block size in
order to process all blocks within the plaintext message.
96 H. K. Hoomod et al.

b. CHSA Decryption
The decryption process as in all encryption standards and algorithms performed by
the inverse of the encryption process was all process taken place in reverse order
to retrieve the original plaintext. CHSA decrypts the ciphertext through the use of
stored keys and patterns inside GC. GC provides the CHSA with all necessary terms
for the decryption process.
Algorithm 2: CHSA Decryption Syntax

Input:
Output:

The decryption process within CHSA depends on the reverse processing and
storing center within GC. All key streams and patterns stored in SC. The process of
specifying the address of the pattern or the key is performed using Address record
stored in the GC. SC contain a special table called Addr essT able that containing
the full address of each block and each pattern. The Addr ess of the block stored to
be retrieved by the desirable phase.
The complexity of the encryption/decryption process came from the mathematical
processing and pseudo-random pattern and key generation. The chaining process,
confusion process, distribution process, and organizing process make it very hard
and complex for the cryptanalyst to attempt retrieving all the system arguments
because each block has a key that differs from the other keys stored in the SC.
New Complex Hybrid Security Algorithm … 97

3 Implementation and Results Analysis

The process of implementation of the new algorithm requires a programming lan-

guage with a large capacity to deal with data of all kinds and also must have the ability
to deal with networks in order to secure the files and transmission over the network
to the recipient. A new system was implemented to implement the new hybrid algo-
rithm within the Visual Studio environment. The algorithm succeeded in encrypting
the data in a new way. It worked on splitting the clear text and then coding it using the
RC4 algorithm and then integrating the key and some additional processes to get the
encrypted text. Table 1 shows the results for each of the algorithm’s operations for,
as well as the total duration of several types of experiments represented in Table 2.
In Table 2, it is possible to note that the time taken to perform encryption operations
ranges between 0–2 s and 12 s as the maximum time in the encryption process and
the maximum encoding process can be observed in video files which are the largest
file types in size in network applications. Table 2 represents the time taken by each
encryption process within the hybrid algorithm for the creation of encoded text
since each process has a different key from the other because of the key generation
center, which gives each algorithm a different key depending on the semi-random
generation algorithms. Table 3 shows the results of organization information and
S-box distribution operation performed by the CHSA.
Because the last stage of the algorithm is the probability distribution variable by
semi-random pattern, we find that it is found in Tables 2 and 3. This phase contains
several sub-phases, such as RSA and process stages to increase the randomness
and strength of the encoded text Permutation. Overall, the processes in Table 3
took relatively less time than the encodings in Table 2 because the mathematical
structure of these processes was less than the coding processes. Figure 4 illustrates
the results gained from the encryption algorithms in CHSA with file encryption and
the operations of CHSA on files.
These results were in terms of files within the operating systems within the com-
puters either in the network systems, the implementation of the algorithm is through
the use of PHP programming language as a common language in the applications
of the Internet and networks and various applications of the internal sharing of files.
The CHSA algorithm was built by the PHP programming language and also per-
formed file encryption experiments during the transmission and receiving process.
The results showed that the time spent in the encryption process exceeded what it
was in the operating system due to the time it took to load the file and the time it took
to convert the file to the formula And finally the time spent in the process of sending
to the receiving party and all these additional times increased the overall time spent
in the encryption process using the CHSA algorithm. Table 4 shows the time spent
in encrypting data using the CHSA algorithm for network applications as well as
Fig. 5 that illustrate the difference between the OS operation and network operation
in CHSA.
 98

Table 1 Results came from string encryption when using CHSA

Data type Block cipher Stream cipher Adding encryption key Probability distribution
String 65537d410d00c52ed7d34dd7e03 93dd6077dd2424e64effffd3 ba65567b3ac7b100767b582fb 5ad0abb839b660c3b5f423eb3
plaintext 563345997cf61d872be2c9451e2 1993826635ce2f98cc5c7a67 6af11d244a14184ba8c57772a3 95cbed3aa239d3f4ef7ffb1b87a
c03abb69156af664ad7619d24d 34f28370ab502ce9ae539ac9936ac2 7ee128acad2881461cc911001 f977c21a89f253291938ab9121
d1cbf58e4956d9516e 581d9253488d0b893d5a1b707
e9e59998cebd5e42f37bc87077
42f217ddff693da73ade9e2132
4f75580cf2350079e4547eabd3
3543de7b37
H. K. Hoomod et al.
New Complex Hybrid Security Algorithm … 99

Table 2 Results came from file encryption when using CHSA

Data type File size Block cipher Stream Adding Probability
(s) cipher (s) encryption distribution
key (s) (s)
String 180 character 0.2 0.2 0.3 0.4
plaintext
.TEXT 4 KB 0.5 0.8 0.6 1
.DOC 12 KB 0.9 0.7 0.9 1
.PNG 40 KB 3 3.5 4 2
.PDF 231 KB 7 8 7 8
.MP4 3240 KB 10 11.3 12 10.2

Table 3 Results came from file encryption when using CHSA (organization processes)
Data type File size S-box (s) Block organizer (s) Probability distribution
(s)
String plaintext 180 character 0.3 0.11 0.4
.TEXT 4 KB 0.51 0.55 1
.DOC 12 KB 0.7 0.76 3
.PNG 40 KB 1 2.15 1.5
.PDF 231 KB 4 3.9 4
.MP4 3240 KB 5 4.16 1.2

(a) (b) Orginization Process Results

6
5
4
3
2
1
0

doc png pdf mp4

Fig. 4 CHSA results from both encryption and security functions

4 Conclusion

The hybrid algorithm is the process of integrating more than one algorithm at a time
in order to take advantage of the existing features in each algorithm. The purpose
of increasing the complexity of the CHSA algorithm is to increase data security.
The greater the complexity, the greater the security of the data. The structure of the
algorithm consists of 6 main parts in each of these parts is a mathematical algorithm
100 H. K. Hoomod et al.

Table 4 Results came from file encryption when using CHSA (network application using PHP)
File File size Upload Conversion Block Stream S-box Adding Probability
type time time (s) cipher cipher (s) encryption distribution
(s) (s) (s) key (s) (s)
String 220 0.4 0.1 0.2 0.2 0.1 0.4 0.2
data character
.TEXT 3 KB 1 0.3 0.5 0.4 0.3 0.2 0.9
.DOC 15 KB 1.6 0.5 1.3 1 0.5 0.4 1.4
.PDF 1 MB 4 3 5 5.2 3 1.2 2
.PNG 1.4 MB 4.3 3.1 4 5.2 3.2 1.5 4
.MP4 4 MB 9 6.3 7 6 5.1 2.3 8.1

specialized in data security. In the first part, the algorithm divides the clear text into
blocks in order to be passed to the other stages. This is known as a block cipher
since the data is cut into blocks of equal size. The data then flows to the stream
cipher process, which in turn encodes the blocks that were cut at the beginning of the
algorithm. The third stage is to add more randomness to the system using the S-box.
The fourth stage is to add the key-encryption key to clear text parts. The fifth stage is
the process of redistributing the blocks, which are scattered in a random order. The
sixth and final stage is the process of authentication or coding at random according
to the pattern sent by the generation center.
The CHSA algorithm contains a special center for generating keys called the gen-
eration center, which is responsible for generating all types of keys for all algorithms
within the CHSA. Each of the six algorithms has special keys that differ from the
other algorithms. Therefore, the task of the generation center is to provide these
algorithms with the necessary keys to perform their functions. As for the method
of storage of these keys, there is a storage facility located within the center of the
generation used to store the keys that are generated through the center of genera-
tion and use by other algorithms. The storage unit at the generating station shall be
responsible for storing and retrieving these keys at the time of need in the decryption
process for the recipient in the case of network systems and for the user if he uses
the same operating system.
The algorithm was tested by encrypting a number of files within the operating
system. These files were text string, text file, images, video, and e-books in pdf
format. The algorithm showed its ability to encrypt data in such a complex way
that it is difficult to analyze the code for the algorithm. The results showed that the
minimum time spent in the encryption process in the CHSA algorithm is 0.5 s, which
is the total time spent in the coding process of the text string, which is the simplest
input type for the encryption algorithm, while the largest time taken by the algorithm
is about 32 s when encrypting video files in mp4 format which are considered The
largest input type of the algorithm in terms of size and found that the algorithm
in total takes more time than the rest of the algorithms is relative because of the
multiple stages of encryption contained in the algorithm and because of these multiple
stages, the encryption takes more time than in traditional algorithms. However, the
 CHSA OS Opera Ɵon Results CHSA Network OperaƟon Results
14 25

12 12 20
11.3
10.2 15
10
10
doc 10
8

AXIS TITLE
8 8
New Complex Hybrid Security Algorithm …

png
5
7 7 pdf
6
mp4 0
4 1.8 0.4 0.3 0.2 0.9
4 3 3.5
doc 3.4 1 0.5 0.4 1.4
2 png 11.4 5.2 3.2 1.5 4
2
0.7 0.9
0.9 1 pdf 12 5.2 3 1.2 2
0 mp4 22 6 5.1 2.3 8.1

Fig. 5 CHSA results comparison between OS application and network application

101
102 H. K. Hoomod et al.

most important feature of the CHSA algorithm is the strength in encryption and
complexity of the encoded text, since the process of parsing the code is almost
impossible given the number of random processes contained in the algorithm and its
encryption algorithms and reliability.
After obtaining results and statistics for the CHSA algorithm, it is necessary to
mention the effect of this algorithm in the face of attacks targeting both the operating
system and network applications. Here lies the real strength of security algorithms in
responding to these attacks. The CHSA algorithm proved effective against Trojans
and viruses that open the gaps between the victim’s operating system and the hacker
to analyze the code as the data will be useless for the hacker. As for the network’s
many attacks, the most powerful types of attacks that network systems may face and
the application of the algorithm have been chosen to determine their effectiveness
in preventing these attacks. It has proven that it has been able to prevent 3 of the 4
attacks considered the most dangerous in the network world.
As a future development of the CHSA algorithm, two main directions must be
developed to develop its work. The first is to increase the complexity of the algorithm
by adding new and complex stages in order to increase the strength in the face of as
many attacks on both the operating system and the network applications. Work on
increasing the speed of the algorithm to make it more suitable for applications that
require real-time processing process, i.e., video applications over networks, which
require high-speed encryption algorithms so increasing the speed of implementation
of this algorithm is an important trend very in the development process.

References

1. Ismail M, Gerard C (2012) Evaluation of different cryptographic algorithms on wireless sensor

network nodes. IEEE Int Conf Wirel Commun Undergr Confin Areas. [Link]
ICWCUCA.2012.6402500
2. Timothy DP, Santra AK (2017) A hybrid cryptography algorithm for cloud computing security.
Int Conf Microelectr Dev Circ Syst (ICMDCS), IEEE. [Link]
8211728
3. Li X, Li X, Wei L (2013) The application of hybrid encryption algorithm in software security.
In: 3rd international conference on consumer electronics, communications and networks, IEEE.
[Link]
4. Yu L, Wang Z, Wang W (2012) The application of hybrid encryption algorithm in software
security. In: Fourth international conference on computational intelligence and communication
networks, IEEE. [Link]
5. Mandal AK, Parakash C, Tiwari A (2012) Performance evaluation of cryptographic algorithms:
DES and AES. In: IEEE students’ conference on electrical, electronics and computer science,
IEEE. [Link]
6. Patel JR, Bansode RS, Kaul V (2012) Hybrid security algorithms for data transmission using
AES-DES. In Int J Appl Inf Syst (IJAIS) 2(2). ISSN: 2249-0868 Foundation of Computer
Science FCS, New York, USA
7. Bhatele KR, Sinhal A, Pathak M (2012) A novel approach to the design of a new hybrid security
protocol architecture. In: IEEE international conference on advanced communication control
and computing technologies, IEEE. [Link]
New Complex Hybrid Security Algorithm … 103

8. Panda PK, Chattopadhyay S (2017) A hybrid security algorithm for RSA cryptosystem. In:
4th international conference on advanced computing and communication systems (ICACCS),
IEEE. [Link]
9. Mitchell CJ (2016) On the security of 2-key triple DES, IEEE transactions on information
theory 62(11). [Link]
10. Weerasinghe TDB (2013) An effective RC4 stream cipher. In: IEEE 8th International con-
ference on industrial and information systems, IEEE. [Link]
6731957
11. Ahmad A, Farooq M, Amin M (2016) SBoxScope: a meta s-box strength evaluation framework
for heterogeneous confusion boxes. In: 49th Hawaii international conference on system sciences
(HICSS), IEEE. [Link]
12. Wang X, Wang S, Zhang Y, Luo C (2018) A one-time pad color image cryptosystem based
on SHA-3 and multiple chaotic systems. Optics Lasers Eng 103:1. [Link]
optlaseng.2017.11.009
13. Vybornova YD (2017) Password-based key derivation function as one of Blum-Blum-Shub
pseudo-random generator applications. Proc Eng 201. [Link]
09.669
14. AL-Mousawi AJ, AL-Hassani HK (2017) A survey in wireless sensor network for explosives
detection. Comput Electr Eng. [Link]
15. Narasimha Mallikarjunan K, Muthupriya K, Mercy Shalinie S (2016) A survey of distributed
denial of service attack. In: 10th international conference on intelligent systems and control
(ISCO), IEEE. [Link]
16. Smyth D, McSweeney S, O’Shea D, Cionca V (2017) Detecting link fabrication attacks in
software-defined networks. In: 26th international conference on computer communication and
networks (ICCCN), IEEE. [Link]
17. Shaikh AA (2016) Attacks on cloud computing and its countermeasures. International confer-
ence on signal processing, communication, power and embedded system (SCOPES), IEEE.
[Link]
Performance Assessment of Various
Encoding Schemes with Bit Stuffing

S. Bharath and Gayathri Narayanan

Abstract This paper discusses the implementation of a novel technique of encoding

data bits using the concept of bit stuffing in addition to the conventional methods
of source coding. This technique can be applied to any of the existing methods
of source encoding under controlled conditions. In particular, the method is very
efficient when the encoded bits have more number of ones or zeros than a predefined
threshold, at any point of time and in any part of the stream. Usually, bit stuffing
is a common method used for data compression in data communication layers to
reduce the bandwidth. In this paper, we have attempted to incorporate bit stuffing in
various encoding schemes and have compared the improvement in performance with
and without bit stuffing. The software used for simulation is MATLAB. The primary
motivation of this work is to determine the maximum amount of bandwidth savings
that can be achieved due to bit stuffing for a random series of alphabets.

Keywords Arithmetic encoding · Bit stuffing · Huffman coding · Redundancy

1 Introduction

Source encoding is the process of encoding the data bits and preparing it for trans-
mitting across the channel length in an efficient manner so as to decrease the overall
bandwidth and increase the data rate. Source encoding is done to eliminate redun-
dancy. Redundancy translates to the wastage of bandwidth in the channel. Source
encoding is accomplished by reducing redundancy and compressing the data before
it is sent to the channel while channel encoding adds redundancy to ensure reliable
reception of data stream. Channel encoding is done to overcome the effect of noise in
the channel. A source encoding scheme is applied to a system based on some specific
pattern that the source follows, which could be either the probability of each symbol

S. Bharath (B) · G. Narayanan

Departtment of Electronics and Communication Engineering, Amrita Vishwa Vidyapeetham,
Amritapuri, India
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 105

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
106 S. Bharath and G. Narayanan

Fig. 1 Model of a simple

digital communication
channel

or uniqueness in the occurrence of the symbols. Any source encoding scheme should
generally satisfy the following properties:
• Uniqueness
• Prefix free
• Instantaneous
The coding scheme should provide unique code words to each symbol used. A
prefix code is a code that follows ‘prefix property’ for example, a code with code
words [4 0 15] may be said to be a prefix code while another code with code words [4
0 41] does not follow prefix property. The code word of any symbol should not be a
prefix of a code word for another symbol. A prefix code is usually uniquely decodable
because given a complete sequence the receiver can identify each word without
requiring a marker between each word. The final code should be instantaneously
decodable which means that the decoded data should not be ambiguous. In arithmetic
encoding scheme, we transmit a tag value based on the probability distribution in each
symbol instead of making code words for each symbol. The concept of ‘Information’
in Information Theory deals with the amount of uncertainty in an event. Average
entropy is the average rate at which information is produced by a particular stochastic
source. Redundancy is then referred to be the difference of entropy H(X) of an
ensemble X and its maximum possible value, i.e. if there are N ensembles for equal
probability distribution, then (Fig. 1).

R = H (X ) − log2 (1/N )

2 Related Works

There are certain encoding schemes where probability distribution is not used for
coding the data bits like the Lempel Ziv method [1]. This paper focuses on Huff-
man and Arithmetic encoding schemes by incorporating bit stuffing. Considering the
whole set of existing encoding schemes Huffman and Arithmetic encoding are con-
sidered to be the most effectual based on the compression efficiency. Bit stuffing is a
Performance Assessment of Various Encoding Schemes … 107

novel idea as presented in [2] where we insert a non-informative bit in between the
data stream so as to break a common pattern followed in the data stream. Bit stuffing
is useful only if the data stream follows a specific pattern. There are some existing
works on the use of bit stuffing in steganography [3] to improve the encryption of
data. The above-mentioned paper is about ‘graphstega’ using Huffman encoding to
hide a potential secret. Bit stuffing is also found useful for crosstalk avoidance in
high-speed switching [4].
Huffman code is a particular optimum variable-length prefix code usually used in
the lossless data compression. It is based on the probabilities of each symbol used in
the stream. In one of the reference papers [5], the authors suggest a modified Huffman
coding to improve its performance and redundancy based on the minimum variance
Huffman coding. A universal design of arithmetic coding algorithm using the soft-
ware and hardware design is presented in [2] where 256 ASCII codes of distinct
symbols, as a particular example, are accommodated in the alphabet. Accordingly,
both the coding equations are modified by indicating the code values as the lower
end-point value of the coding range and the width of this range. Thus the flow of
sending output codes, solving the so-called underflow problem, and updating the
coding range can be integrated and simply controlled by the value of the coding
range.
Various variants of arithmetic coding schemes are now available for different
purposes like image and audio compression. In [6], a method of context-based adap-
tive arithmetic coding is suggested for compression of image. The concept is based
on parallel leading zero detection and bit stuffing handling where the symbols are
encoded and decoded in one cycle. In this paper, we deal with text as the data
entity and encoding it according to its probability distribution. We present three
main functional blocks namely, ‘arithmetic encoding and decoding’, ‘bit stuffing’,
‘bit de stuffing’ to compare the efficiency of transmitted data after source encoding
with arithmetic and Huffman method.

3 Proposed Work

3.1 Arithmetic Encoding and Decoding

The state of the art in source encoding is arithmetic encoding scheme as it represents
the data more compactly. It has an optimal performance without the need for blocking
of input data. It encourages a clear separation between the model for representing data
and the encoding of information with respect to that model. In arithmetic encoding,
the message output data is represented by an interval between 0 and 1. The range
of this interval reduces as the size of the message data is bigger. As the size of this
interval reduces more no of bits are required to represent the data. In this paper,
simulation is done with MATLAB and we found that it’s difficult to figure out the
tag number if direct arithmetic coding is applied to an audio signal. Many modified
108 S. Bharath and G. Narayanan

arithmetic logics are presented before like [7] which could compress an image,
audio or video. This paper is to account for a pattern observed in arithmetic encoded
sequences where it’s common to have observed more number of consecutive similar
bits in the code. As the message size increases this pattern is observed to increase.
Consecutive symbols of the message reduce the size of the interval according to the
symbol probabilities generated by the model. The more probable symbols reduce the
range lesser than the least probable symbols adding fewer bits to represent the data.

3.2 Bit Stuffing

As in bit stuffing concepts normally used, it is a method of data compression that

takes care of a certain known pattern used repeatedly in a sequence of data. If a
pattern repeats in a binary sequence, then it could eventually be replaced by a bit
‘1’ stuffed in between the sequence. This requires that a code must check for a
certain pattern regularly and stuff either a ‘0’ for absence or ‘1’ for presence of
that pattern. The encoder and decoder should know the pattern followed and the
method used for bit stuffing so as to decode the message accurately. There are many
existing papers on modification to bit stuffing algorithm as in [8] which are applied
in different contexts. There are some existing works [9] about deriving lower bounds
on the capacity of certain two-dimensional constraints by considering bounds on the
entropy of measures induced by bit stuffing encoders.
In this module used for simulation, the program searches for five consecutive
one’s or zero’s if two of them are already found successively. The module writes the
first two consecutively repeated bits for transmitting and adds an extra bit that says
if the third, fourth and fifth bits are being repeated the same value. This module adds
or stuffs ‘1’ if they are repeated and ‘0’ for the other condition. If the stuffed bit is ‘1’
then the further third fourth and fifth repeated bits are ignored. On the other hand, if
the stuffed bit is ‘0’ no bit is ignored. Thus, if two bits repeating pattern are more in
our sequence with no five-bit repeating pattern, it leads to a considerable increase in
bandwidth.

3.3 Bit De-Stuffing

This is the process by which the bit stuffed sequence at encoder is recovered at
the destination decoder. Here whatever modifications made for bit stuffing which
accounts for the extra stuffed bits indicating the presence or absence of the pattern is
eliminated. In this module used for simulation, every third bit that comes after two
repeated bits is ignored. Also that if this third bit is ‘1’ three repeated 1’s are stuffed
instead of this ignored third bit.
Performance Assessment of Various Encoding Schemes … 109

4 Result Analysis

MATLAB is one of the most widely used programming platforms in which a wide
range of operations can be performed on the input signals and their functions can be
plotted. MATLAB considers any signal as a matrix of numbers and performs opera-
tions on them. This paper focuses on encoding text messages where the probabilities
of individual symbols are taken from Pavel Mika’s database which cites Robert
Leward’s cryptological mathematics. For arithmetic encoding, the probabilities of
individual symbols are calculated according to their occurrences and frequencies in
the sentence. Before transmitting anything the range of the message is from 0 to 1, i.e.
0 ≤ x < 1. An extract from Stephan Hawking’s ‘A Brief History Of Time’ is taken as
the message for the simulation purpose. The basic procedure for arithmetic encoding
is demonstrated below with a simple example as shown in Table 1. Suppose that
‘HOUSE’ is to be coded under arithmetic encoding scheme. We have the individual
probability distribution of ‘H’, ‘O’, ‘U’, ‘S’, ‘E’ as 0.2, 0.2, 0.2, 0.2, 0.2 taking prob-
abilities based on their occurrence frequencies of each alphabet. As each alphabet
occur only once in the message string, the probability of each alphabet is assumed
to be the same. Initially, both the encoder and decoder know the entire range to be
[0, 1]. As each alphabet is coded the range to be considered decreases. The range for
each letter is assigned by taking the cumulative probability for the ascending order
of alphabets. Thus the range for these alphabets will lie as in Table 1.
When the encoder see ‘H’ it limits the range to [0.2, 0.4] and sets the new range.
When ‘O’ is sensed the range further narrows to the third one-fifth of the above
range and it goes on until the end of the message. Finally, the encoder is left with a
range whose difference is called the tag which is transmitted in arithmetic encoding
scheme.
This binary tag is passed through bit stuffing to realize that the bandwidth is
reduced considerably. Similarly, a Huffman coded sequence is passed through the
bit stuffing module and was found that the bandwidth decrease was less compared
to arithmetic code which gives insight to the pattern followed in the sequence as
mentioned above.
It’s found that bit stuffing has less effect on Huffman coding or rather it may
increase the bandwidth of encoded message for short messages. Considering a text
message of length 5 in MATLAB, length of coded bits is 20 bits without bit stuffing
but the length increases to 27 bits on bit stuffing indicating the presence of 7 stuffed

Table 1 Sample alphabet set

Alphabet Probability Range
and their probabilities
E 0.2 [0, 0.2]
H 0.2 [0.2, 0.4]
O 0.2 [0.4, 0.6]
S 0.2 [0.6, 0.8]
U 0.2 [0.8, 1]
110 S. Bharath and G. Narayanan

0’s and no bit savings. For a length 10 text message, the bit stuffed code indicates the
presence of 13 stuffed 0’s and no bit savings. But when the length of the text message
increased to 15, Huffman coded sequence saved 24 bits indicating 12 stuffed 1’s and
added an extra 6 bits indicating 6 stuffed 0’s through bit stuffing. Thus 18 net saved
bits are found with bit stuffing. This says that as the size of the message increases
bit stuffing has a considerable effect on Huffman encoding.
In case of arithmetic coding, bit stuffing has a considerable effect on any length
of message. For a length 5 text message in MATLAB, the coded sequence is 24 bits
long without bit stuffing but it reduced to 16 bits after bit stuffing indicating the
presence of more consecutive similar bits. Similarly, for a length 10 text message,
the code length was 28 bits without bit stuffing which reduced to 22 bits after bit
stuffing. For a length 15 text message the code length reduced from 64 bits before bit
stuffing to 43 bits after bit stuffing. Length of the code varies depending on the text
message and the frequency of each letters in them but this trade-off is taken care off
by suppling similar messages in each case. Thus, arithmetic encoding scheme stands
out as better data compression with bit stuffing.
For random text messages of size 10 and 20 the graphs in Figs. 2 and 3 shows
the effect of the length of Huffman code without bit stuffing, length of an arithmetic
code without bit stuffing, traditional encoding with N bits for 2n symbols without
bit stuffing, normal arithmetic code without bit stuffing, normal arithmetic code with
bit stuffing respectively for a, b, c, d and e shown in X-axis. Figure 2 is about 25
different, random, size 10, text messages generated by ‘rand’ command in MATLAB
environment. Figure 3 is about 7 different, random, size 20, text messages. Comparing
the two figures, one can observe that as the text size of the message increased the bit
stuffing gets more effective. More sequences are given here to generalize the concept.
We also observe that we have a common convergence point which is because of the
fact that MATLAB software random function generates equal no of different letters

Fig. 2 Code lengths of 25

different, random, size 10,
text messages
Performance Assessment of Various Encoding Schemes … 111

Fig. 3 Lengths of 7
different, random, size 20,
text messages

Fig. 4 Length of varies code

for the size of text messages
N = 5, 10, 15, 20 and 25

most of the time. Figure 4 clearly shows that as size of text message increases bit
stuffing becomes more efficient in arithmetic codes.

5 Conclusion

This work presents a performance analysis of various encoding schemes like Huffman
and Arithmetic encoding with bit stuffing. The chosen schemes are the most efficient
in terms of their source encoding efficiency. This work attempts to improve the
112 S. Bharath and G. Narayanan

efficiency of Huffman and Arithmetic coding schemes by incorporating the concept

of bit stuffing, under chosen conditions. It shows the importance of bit stuffing in
data compression as the length of the text message increases. The simulation studies
were carried out using MATLAB software. It was observed that arithmetic encoding
with bit stuffing stands out in performance as a better data compression or source
encoding scheme.

References

1. Savari SA (1997) Bell Labs Lucent Tehnol Murray Hill USA: redundancy of the Lempel Ziv
codes. In: Proceedings of IEEE international symposium for information theory
2. Jiang J (1995) Novel design of arithmetic coding for data compression. IEEE Proc Comput
Digital Technol 142(6):419
3. Akhter F (2016) Secured word-by-word graph steganography using Huffman encoding. In: 2016
international conference on computer communication and informatics (ICCCI)
4. Chang C-S, Cheng J, Huang T-K, Huang X-C, Lee D-S, Chen C-Y (2015) Bit-stuffing algorithms
for crosstalk avoidance in high-speed switching. IEEE Trans Comput 64(12):3404
5. Sandeep GS, Sunil Kumar BS, Deepak DJ (2015) An efficient lossless compression using double
Huffman minimum variance encoding technique. In: International conference on applied and
theoretical computing and communication technology (iCATccT)
6. Ong KK, Chang W-H, Tseng Y-C, Lee Y-S, Lee C-Y (2002) A high throughput context-based
adaptive arithmetic codec for JPEG2000. IEEE 2002
7. Marpe D, Schwarz H, Wiegand T (2003) Context-based adaptive binary arithmetic coding in
the H.264/AVC video compression standard. IEEE Trans Circ Syst Video Technol 13:620
8. Aviran S, Siegel PH, Wolf JK (2005) An improvement to the bit stuffing algorithm. IEEE Trans
Inf Theory 51(8):2885
9. Halevy S, Chen J, Roth RM, Siegel PH, Wolf JK (2004) Improved bit-stuffing bounds on
two-dimensional constraints. In: Proceedings IEEE international symposium on information
theory
3D(Dimensional)—Wired and Wireless
Network-on-Chip (NoC)

N. Ashokkumar, P. Nagarajan and P. Venkatramana

Abstract Network on Chip is a special unique case of parallel computing systems

defined by the tight constraints such as availability of resources, area, cost of the NoC
architecture and power consumption. NoC is designed with three main components:
switches, Network Interfaces (NIs) and links. NoC is used for several application
domains, such as multi-media processing, consumer electronics, biological applica-
tions, etc. NoC is the technology proposed to solve the shortcoming of buses. This
technique is used to design communication subsystem among IP cores (Intellectual
property core) in a SoC design. In this chapter, we have discussed about 3D inte-
grated circuits, 3D wired and wireless NoC, Emerging Technologies, and Literature
Survey.

Keywords 3D NoC (3Dimensional Network-on-Chip) · TSV (Through silicon

via) · Wired and wireless NoC · Topology

1 Introduction 3D-ICS

The issues connected with the high wiring network necessities of large scale joining
circuit configuration is investigated in [1] alongside how 3D ICs improve avail-
ability while decreasing the quantity of long interconnects. So also, the creators
of [2, 3] research how 3D ICs can be utilized to battle the developing proportion
of interconnect to entryway delay as highlight sizes diminish. A general diagram
of 3D innovations and the inspirations driving outlining 3D coordinated circuits is
introduced in [4]. The vertical interconnect innovations contain micro bumps, wire

N. Ashokkumar (B) · P. Nagarajan · P. Venkatramana

Department of Electronics and Communication Engineering, Centre for VLSI and Embedded
Systems, Sree Vidyanikethan Engineering College, Tirupati, India
e-mail: ashoknoc@[Link]
P. Nagarajan
e-mail: [Link]@[Link]
P. Venkatramana
e-mail: pvramana2406@[Link]
© Springer Nature Singapore Pte Ltd. 2020 113
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
114 N. Ashokkumar et al.

holding remote interconnects utilizing capacitive/inductive coupling, and through-

silicon vias (TSVs), of which TSVs suggest high-thickness vertical interconnects
and are by a wide margin the most encouraging method [5]. The advantages of uti-
lizing 3D NoC rather than 2D NoC are investigated by Feero and Pande [6]. These
work approaches on the performance and area effects of the network structures rather
than the power and performance tradeoffs of various technologies. 3D ICs gives an
equivalent set by long inbuilt through stacking active silicon layer. In this manner,
one can expect huge increments in execution and lessening in force utilization and
territory with conceivable integration of CMOS circuits with different innovations
[5, 7, 8]. 3D ICs suggest various favorable positions contrasted with 2DICs. These
incorporate:
• Shorter global interconnects
• Superior performance
• Smaller area (footprint)
• Low power consumption
• Scope of mixed-technology ICs
• Higher packing density.
In any case, 3D ICs have huge worries as warm thought. While the general force
dispersal in 3D ICs might be lower because of shorter and less worldwide inter-
connects, the force thickness is much higher because of vertically stacked silicon
layers. Thusly, capable warm organization is the best approach to guarantee the
execution redesigns offered by 3D ICs. For more unmistakable cognizance on this
subject, Emerging Interconnect Technologies for 3D Networks-on-Chip per users
are referred to the distinctive warm methodologies, for instance, physical arrange-
ment streamlining, usage of warm vias, and microfluidic cooling of the vertical stack
reported in the writing [9–11].
The 3D reconciliation innovation we utilized depends on Tezzaron [7] that uti-
lizations TSV for fringe IOs and microbumps for bury bite the dust associations.
The two-level 3D stacking technique depends on wafer-to-wafer holding, up close
and personal technique with by means of first approach as outlined in Fig. 1. The
between kick the bucket micro bumps give high interconnection thickness up to
40,000 per mm2 without interfering to FEOL (front-end-offline) device or routing
layers. It is additionally conceivable to execute four levels by stacking through con-
secutive utilizing TSV of the two face to face stacking so as to have higher outline
many-sided quality yet it won’t be secured in this work.
In advanced technology, we analyze the 3D NoC architectures we select 45 nm
standard library from ST Microelectronic [8]. We utilize comparative 3D structure
for between level associations utilizing microbumps as a part of Tezzaron innovation
however we replace the 130 nm technology of Global Foundries with 45 nm ST
Microelectronic standard library. The 45 nm innovation utilized as a part of this study
has seven metal layers where metal seven is utilized for holding and the steering is
constrained until metal six.
3D(Dimensional)—Wired and Wireless Network-on-Chip (NoC) 115

CMOS
Global Foundries 130nm,LVT
process,1.5v core,3.3V I/o

6 metal layers, metal 6

for bonding

FACE-TO-FACE COPPER BUMP 3.4 μm diameter,5 um

pitch bumps

Via-first method, Tungsten

TSV

1.2 μm diameter,2.5um pitch,6

μm depth

CMOS

Fig. 1 Cross section of Tezzaron 3D IC technology with corresponding parameters

2 3D-Wired and Wireless NoC

2.1 Wired NoC

As one of the more well known vertical network technologies, through silicon via
(TSVs) and some of their manufacturing methods are described in [5] alongside TSV
electrical characteristics extraction and demonstrating. TSVs add extra many-sided
quality to the fabricating process for 3D ICs yet they lead to offer power, performance
of execution, and chip range qualities.

2.2 Wireless NoC

In [7], a low power and high information rate inductive coupling transceiver is pro-
posed. Inductive coupling is a vertical technology innovation that does not require
adjustments to the assembling procedure, however, the force, execution, and chip
range overheads are regularly restrictive to the appropriation of the innovation. The
design and implementation of a capacitive coupling handset is broke down in [8]
where the force, execution, and region overheads are examined and also limitations
that capacitive coupling joins put on how the layers of the 3D ICs are collected.
Capacitive coupling additionally does not oblige changes to the assembling proce-
dure but rather confines vertical scaling to two layers put confronted to confront
116 N. Ashokkumar et al.

rather than numerous layers put face to back. It too displays poor performance, exe-
cution, and chip zone overheads in respect to inductive coupling what’s more, wired
procedures.

3 Emerging Technologies

Some experimental advances show potential for being powerful at decreasing vital-
ity utilization and increase execution yet are not secured in this work. One of the
all the more encouraging innovations is photonic interconnects. Photonic intercon-
nects exchange information by sending signals over optical waveguides. In [5], TSVs
and a reconfigurable photonic system are used to decrease vitality utilization while
looking after execution. Photonic interconnects have the advantage of their data
transfer capacity being autonomous of the correspondence separation. Lamentably,
there are additional fabricating steps that are required to assemble circuits that incor-
porate photonic interconnects. These additional strides add to the unpredictability
and general expense of these frameworks. Another innovation for associating cen-
ters in a framework uses remote interconnects. Radio recurrence handsets can be
incorporated with the chip and used to transmit information crosswise over bigger
separations with less power and less inertness than customary wires. Little world
systems and millimeter-wave remote systems on chip are investigated in [9, 10]. In
[11], remote interconnects that use CDMA to permit different remote handsets to
work in the meantime are reenacted to break down their execution and vitality quali-
ties. Remote interconnects can likewise be used for exchanging information between
layers of 3D ICs as in [12].
Previously, the possibility of remote correspondence in planar 2D NoCs has been
proposed by a couple of researchers. A remote direct in 2D NoCs wears down
high data transfer capacity, single-skip, long-go correspondence, as against multi-
hop correspondence in steady wired NoCs, realizing lower stillness, lower power use,
besides, less difficult controlling arrangements. The possibility of remote correspon-
dence in 2D NoCs was at first highlighted by Floyd et al., where a clock dispersal
framework was executed using remote interconnects. In addition, the remote 2D
NoCs using an ultra-wideband coincidence plan have been suggested. Consistently,
a remote redirect in a planar NoC has an accepting wire, framework building, and
handset circuits as its key definitive parts. The idea here is to divide the system into
various subnets with wired intra-subnet correspondence and remote joins conferring
between the subnets. Each subnet involves a base station for setting up the remote
association. Customary wired NoCs (WiNoCs) have an individual subnets, distinc-
tive structures and heterogeneous configuration also those outcomes in essentially
made strides inertness and throughput.
The thought of 3D remote NoC using inductive coupling was first illustrated and
the planned systems predict a 4 × 4 × 3 3D network system. We have used a coordi-
nation based handset circuit for inductive coupling. The intralayer associations are
3D(Dimensional)—Wired and Wireless Network-on-Chip (NoC) 117

32 bits wide, and the guiding system is dimensional wormhole controlling. Particu-
lar switches have eight data handsets and a solitary clock handset. The handset has
separate twists for data and clock transmission with a resultant vertical association of
transfer speed of 8 bits. The total region of the six-port switch in 90 nm development
is 0.13 mm2 , which consolidates a region spending arrangement of 0.03 mm2 for the
nine twists.

4 Survey Results

Many issues in 2D NoC designing and it’s have been concentrated on over the
previous years by covering different perspectives, for example, plan stream, usage
assessment, and configuration space investigation. Nonetheless, examine in 3D NoC
is still new and numerous issues stay unexplored particularly in genuine configuration
and usage. Outline space investigation of 3D NoC topologies through cycle-exact
reproduction have been performed demonstrating the advantages of 3D outline as far
as throughput, idleness and vitality dispersal for cross-section based, what’s more,
tree-based NoC engineering [1]. In [2], zero load latency also, control utilization
expository models of different 3D NoC, topologies have been assessed demonstrat-
ing the benefits of consolidating 3D IC with 3D NoC engineering. We base upon this
writing to examine further the outcomes by doing investigation from physical config-
uration execution results. Another work [3] proposed a novel 3D switch engineering
by disintegrating the switch into various measurements to give better execution over
other 3D NoC structures. We contrast from the past reported fills in as we concen-
trate on apportioning 3D NoC models and assess their execution through design
level netlist for more exact investigation of wire length, timing, territory and force
utilization.
In [13], the study of various 3D situation strategies on the execution of three 3D
designs demonstrated that genuine 3D position strategy produces the most elevated
execution change over other strategies at old innovation (130 nm) denote the signifi-
cance of 3D-mindful apparatuses to get most extreme advantages of 3D combination.
Be that as it may, no past work has been given nitty-gritty execution assessment on
different physical configuration measurements (wirelength, timing, effect of wire
length) of 3D NoC engineering specifically with 3D Mesh-based NoC engineering.
In the course of the most recent quite a while, there has been a developing enthusiasm
for 3D ICs as a way to reduce the interconnect bottleneck as of now confronting 2D
ICs. A key test with 3D ICs is their high warm thickness due to various centers being
stacked together, that can unfavorably affect chip execution and unwavering quality.
In this way a few scientists have proposed warm mindful floor planning systems for
3D ICs [7, 8]. A couple of specialists have investigated interconnect designs for 3D
ICs for example, 3D work and stacked cross-section NoC topologies [5] and a half
breed transport NoC topology [9]. Some late work has taken a gander at breaking
down centers (processors [12], NoC switches [10], and on-chip reserve [11]) into
the third measurement which permits decreasing wire dormancy at the intra-center
118 N. Ashokkumar et al.

level, rather than the between center level. Circuit level models for TSVs were intro-
duced in [14]. A couple of later works have investigated the framework level effect
of utilizing half and half electro-photonic interconnect models and proposed half
and half Cu-photonic crossbars (Corona, Firefly), Clos systems, fat-trees and torii.
This paper portrays the advantages and disadvantages of various creation techniques
including up close and personal holding and face-to-back holding. With a SOI face-
to-back procedure, the through pitch is minimized, at 0.4 m with a partition of 2 m
between layers of SOI gadgets [15]. Jacob et al. [16] propose utilizing 3D ICs to
enhance the execution of chip by shaping a processor memory stack. They demon-
strate that the mix of processor worries by applying genuine movement designs in a
cycle-precise recreation and by measuring execution through set up measurements
for 3D NoC structures. In [17], 3D ICs were proposed to enhance execution of chip
multiprocessors. Drawing upon 3D IC research, they picked a hybridization of trans-
ports and systems to give the interconnect fabric amongst CPUs and L2 reserves.
The execution of this combination of NoC and transport models was assessed uti-
lizing standard CPU benchmarks. Be that as it may, this examination relates just to
chip multiprocessors and does not consider the utilization of 3D system structures
for application-particular SoCs. Three-dimensional NoCs are broke down as far as
temperature in [5]. Pavlidis and Friedman [9] contrasted 2D MESH structures and
their 3D partners by breaking down the zero-load inertness and force utilization of
every system. This is an assessment that demonstrates a portion of the benefits of
3D NoCs, yet it not one or the other applies any genuine activity design nor does
it gauge other significant execution measurements. We intend to address these wor-
ries by applying genuine movement designs in a cycle-precise reproduction and by
measuring execution through built-up measurements for 3D NoC structures.

References

1. Bayan AF, Wan T, Ramadass S (2010) Delay analysis and system capacity control for mobile
WiMax relay networks. J Comput Sci 6(10):1137
2. Bazzi L, Richardson R, Urbanke R (2001) Exact thresholds and optimal codes for the binary
symmetric channel and Gallager’s decoding algorithm. IEEE Trans Inform Theory 47
3. Benini L, Bertozzi D (2005) Network-on-chip architectures and design methods. IEE Comput
Digit Tech 152(2):272
4. Benini L, Micheli G (2002) Networks on chips: a new SoC paradigm. Computer 35(1):70–78
5. Brooks L, Fife K (2004) Hardware efficient lossless image compression engine. Proc IEEE Int
Conf Acou Speech Sign Process 5:17–21
6. Badrouchi S, Zitouni A, Torki K, Tourki R (2005) Asynchronous NoC router design. J Comput
Sci 1(3):429–436
7. Bollobas B (1978) Extremal graph theory. Academic Press, New York
8. Bollobas B (1985) Random graphs. Academic Press, Londan
9. Burshtein D, Miller G (2001) Expander graph arguments for message-passing algorithms. IEEE
Trans Inform Theory 47:782–790
10. Byers J, Luby M, Mitzenmacher M, Rege R (1998) A digital fountain approach to reliable
distribution of bulk data. In: Proceedings of ACM SIGCOMM’98
3D(Dimensional)—Wired and Wireless Network-on-Chip (NoC) 119

11. Cantor D, Gerla M (1974) Optimal routing in a packet-switched computer network. IEEE Trans
Comput 23(10):1062–1069
12. Chandra A (2002) Test data compression and decompression based on internal scan chains and
Golomb coding. IEEE Trans Comput Aided Design Integr Circ Syst 21(6):715–722
13. Berrou C, Glavieux A, Thitimajshima P (1993) Near Shannon limit error correcting coding and
decoding: turbo-codes. In: Proceedings of the IEEE international conference on communication
technical program, May 23–26, IEEE Xplore Press, Geneva, pp 1064–1070. [Link]
1109/icc.397441
14. Charles RKJ, Vanchinathan T, Kharthik K (2013) Design of serdes transceiver with fixed and
high throughput implementation on FPGA. Life Sci J 10:2849–2857
15. Denecker K, Ville DVD, Habils F, Meeus M, Brunfaut M, Lemahieu I (2002) Design of
an improved lossless halftone image compression codec. Signal Process: Image Commun
17(3):277–292
16. Di C, Proietti D, Telatar E, Richardson T, Urbanke R (2002) Finite-length analysis of low-
density parity-check codes on the binary erasure channel. IEEE Trans Inform Theory 48:1579
17. Divsalar D, Jin H, McEliece R (1998) Coding theorems for ‘Turbo-like’ codes. In: Proceedings
of the Allerton conference, p 210
Performance Analysis of Unified Threat
Management (UTM)

Jatin Gharat, Amarsinh Vidhate and Amit Barve

Abstract With the substantial increase in the internet usage and the growing threat
of hackers to infect as many devices as possible, security has become important to
prevent data breaches and industrial sabotage. Unified Threat Management are the
Next-Generation network security appliances that include multiple security features
along with the performance required for future networks. But packet processing
usually consumes 70% of the CPU time and during heavy load it degrades the UTM
performance by dropping important packets. To overcome such limitations, many
techniques and algorithms have been proposed by the researchers. In this paper,
survey and numerical analysis of each technique is done based on the overall packet
processing time. Based on the numerical analysis, we suggest the best technique to
reduce the overall packet processing time in UTM and hence reduce the load under
heavy traffic conditions.

Keywords UTM · Splay tree · Multicore · Graphics processing unit

1 Introduction

With the recent increase in cyber-attacks and security breaches, more security features
are required inside the network. Increasing the number of security devices makes it
difficult to manage each device separately and consumes other resources. A unified
threat management (UTM) is a device used to incorporate number of network and
security features like firewall, packet classification, bandwidth management, etc.
in a single device [1]. Generally, UTM includes functions like firewall, intrusion

J. Gharat (B) · A. Vidhate · A. Barve

Ramrao Adik Institute of Technology, Nerul, Navi Mumbai, India
e-mail: jatin.gharat87@[Link]
A. Vidhate
e-mail: [Link]@[Link]
A. Barve
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 121

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
122 J. Gharat et al.

detection, and prevention (IDS), anti-spam, anti-spyware, anti-virus, content filtering,

virtual private network, and other security functions.
We will look into the major application of UTM, i.e., packet processing. Some
packets are targeted to pass through the entire packet filter rule list to finally get
rejected. This can effectively decrease the UTM performance. Also, under severe
load conditions, for example, unwanted traffic, DoS attack, etc. efficiency of UTM
decreases considerably. To improve the efficiency, we need to search for good possible
solutions which can reduce the load on UTM under such extreme conditions. Hence,
we need to study and implement improved packet filtering techniques to provide
up-to-date security.
Subsequent sections are organized in the following order: In Sect. 2 we perform
literature survey giving an overview on the work done in this field. In Sect. 3, we
study some techniques used to improve UTM performance. In Sect. 4, we perform
analysis on the techniques used by authors. Finally, in Sect. 5 we conclude this study.

2 Literature Survey

The research areas in UTM are divided into packet filtering, firewall enhancements
and other techniques like using multicore functionality of CPU and GPU. While
these techniques have resulted in great successes, a lot of problems are yet to be
resolved.
Trabelsi et al. [1] have proposed a hybrid approach using Splay Filters and DPI
Rules Reordering (SFRO) algorithm to dynamically optimize the order of splay
tree filters such that network packets are either early accepted/rejected to reduce
the redundancy introduced by multiple filters in UTM. Gummandi et al. [2] used
OpenMP parallelization methods to reduce the overload on UTM by scheduling log-
ical CPUs for packet processing, firewall, spam filtering, and other security features.
Recent firewall devices are considered complex, time-consuming and error-prone
thus resulting in slow filtering action. Sahoo et al. [3] have presented a unique
approach using CUDA programming for GPU which processes packet in parallel
thread blocks and handles firewall security efficiently.
Trabelsi et al. [4] proposed Dynamic Rule and Rule-Fields Ordering with Decision
(DR-RFOD) algorithm to improve the firewall performance and packet classification
by optimizing the firewall filtering rules and rule-fields using network traffic statistics.
Another algorithm proposed by Trabelsi et al. [5] uses Statistical Splaying Filters
with Binary Search on Prefix Length (SSF-BSPL) to optimize the matching time
for wanted/unwanted traffic packets. Neji et al. [6] proposed Self-Adjusting Binary
Search on Prefix Length (SA-BSPL) which can easily find matching for protocol field,
range matching for port numbers, and prefix matching for IP address by dynamic
packet filtering. Adel El-Atawy et al. [7] have proposed Relaxed Policy Expression
(RPE) technique which uses Binary Decision Diagrams (BDD) data structure to
reduce packet matching cost by dynamically changing the pre-filter phase that can
reside on top of any filtering mechanism.
Performance Analysis of Unified Threat Management (UTM) 123

Harvath et al. [8] proposed pNEGI (parallel NEGI) to parallelize the process of
NEGI, a software SoR simulator, to analyze the layer-7 information using DPI which
is generally difficult. Yunchun Li et al. [9] states that with the increasing demand for
more bandwidth by applications, packet processing capabilities decreases in serial
processing systems. Hence the authors have proposed a parallel processing model
and compared with serial processing system to analyze the performance. Chonka
et al. [10] states that implementing security, in general, affects the performance of
the application. Hence the authors propose a multicore defense framework to deal
with the performance issues.
Shuai Mu et al. [11] states that there always has been a trade-off between through-
put and programmability in modern IP router designs. To overcome these limi-
tations, the authors have proposed a technique using GPU, prefix matching and
string-matching algorithms to achieve a much higher performance. Che-Lun Hung
et al. [12] have provided a solution to the computationally intensive process of pat-
tern matching [19] on CPU by using the computational power of graphics processing
unit to accelerate pattern-matching operations. Kang Kang et al. [13] states that high-
performance routers use proprietary hardware to classify packets but are excessively
costly, consume more power and are less scalable. Hence the authors proposed GPU
based linear search framework which uses metaprogramming technique to enhance
packet processing efficiency. Present network intrusion detection systems (NIDS)
must handle increasing network traffic and more complex packet processing task.
Giorgos Vasiliadis, et al. [14] proposed a prototype system Gnort, which utilizes
computational capabilities of graphics card to perform packet processing operations.

2.1 Survey Analysis

From the literature analysis, we found that we can categorize the papers according
to the techniques used. Hence, we divided the papers into three different categories.
The detailed survey analysis for the three categories is provided below:
• Category I: The first category is based on only the algorithmic techniques used
to enhance the overall packet processing performance. The papers in this category
[1, 4–7] mostly use tree filters as a mechanism to early accept or reject a packet
by checking against a set of rules. The use of network traffic to adapt the changing
traffic load helps to achieve better efficiency.
• Category II: The second category is based on the use of multicore functionality of
CPU based system [2, 8–10]. Multiple functions in UTM like packet processing,
URL filtering, spam filtering, etc. can be processed simultaneously thus increasing
the performance of UTM.
• Category III: The third category is based on the usage of Graphics Processing Unit
(GPU) system as a special packet processing device [3, 11–14]. The development of
CUDA programming architecture has made it easy to perform the general task more
efficiently. Because the number of threads present in GPUs surpasses any other
124 J. Gharat et al.

device, the packet processing operation gains considerable amount of performance

improvement from the GPU system.

3 Techniques to Improve UTM Performance

UTM appliances maintain a database of rules for packet classification. Each packet
classification rule consists of a prefix (or range of values) for each possible header
field, which matches a subset of packets. Thus, when a packet arrives, UTM finds a
rule that matches the packet headers; however, if more than one match is found; the
first matching rule is applied.
The main disadvantage of such an approach is the poor scalability since the time
to perform a classification increases substantially with the number of rules in the
rule set. This issue can be resolved by either adjusting the rule set or by using some
parallel processing techniques.
Here we study three techniques: Splay trees with network statistics, Parallel
Processing using OpenMP, and Parallel Processing using GPU.

3.1 Splay Trees with Network Statistics

Splay trees are self-adjusting binary data structure which makes use of splaying
property such that the frequently accessed node (element) is splayed to the root of
the tree. Several authors [1, 4–6] made use of network traffic statistics to effectively
monitor the performance in real-time and predict the patterns of rule and rule-fields
that best suit the next traffic window. Filtering rules and rule-fields orders are updated
if the stability test reaches a certain threshold.
To decide which rule applies to a given packet, five filters, i.e., protocol, source
IP address, destination IP address, source port, and destination port are used. Nodes
that are frequently accessed are given position close to the root of the splay-tree.
The longest matching prefix is updated whenever a match is found. When a leaf
is encountered, the search process stops the best length node is splayed to the root
(splaying property). This results in early acceptance of repeated packets with limited
memory access. Similarly, a packet will be rejected if node with minimum length
in splay tree does not match the corresponding packet value. Decision whether the
current splay filters order should be preserved or not for the next traffic window is
taken on the basis of traffic statistics.
Performance Analysis of Unified Threat Management (UTM) 125

3.2 Parallel Processing Using OpenMP

Parallel processing is a type of computation in which many calculations or the exe-

cution of processes are carried out simultaneously. Large problems can often be
divided into smaller ones, which can then be solved at the same time. A multicore
processor is a processor that includes multiple processing units (called “cores”) on
the same chip. A multicore processor can issue multiple instructions per clock cycle
from multiple instruction streams.
Multiple independent processes in UTM can be parallelly processed using
OpenMP [18]. Assigning different CPU cores to perform different packet processing
operations like URL filtering, spam filtering, etc. parallelly can reduce the overall
time consumption in heavy traffic conditions.

3.3 Parallel Processing Using GPU

A graphics processing unit (GPU) is a specialized electronic circuit designed to

rapidly manipulate and alter memory accelerating the computations in a frame
buffer intended for output. General-purpose computing on graphics processing units
(GPGPU) is the use of a GPU, which typically handles computation only for com-
puter graphics, to perform computation in applications traditionally handled by the
central processing unit (CPU). A single GPU-CPU framework provides advantages
that multiple CPUs on their own do not offer due to the specialization in each chip.
CUDA is a parallel computing platform and application programming interface (API)
model created by Nvidia. The CUDA platform is a software layer that gives direct
access to the GPU’s virtual instruction set and parallel computational elements, for
the execution of compute kernels. The general process flow of data between CPU
and GPU is as follows:
1. Copy input data from CPU memory to GPU memory.
2. Load GPU program and execute, caching data on-chip for performance.
3. Copy results from GPU memory to CPU memory.

4 Comparative Analysis

In this section, we perform comparative analysis on different techniques proposed

by authors. Table 1 shows the notations used for analysis.
126 J. Gharat et al.

Table 1 Notations
Notation Description
P Total number of packets
S Size of a single packet segment/queue
T1 Number of threads in a GPU
T2 Number of threads in a multicore system
m Cost of memory latency
T in Time taken to modify/queue incoming packet
T out Time taken to update result/rule set
t lock Cost of thread locking
t unlock Cost of thread unlocking
T total Total packet processing time in ms
t update Cost of the rule/result table update
T process Time taken to process packet using an
algorithm/technique
C switch1 Cost of context switching in GPU
C switch2 Cost of context switching in the CPU

4.1 Analysis

Let T total be the total time taken for packet processing. T total can be represented as:

Ttotal (ms) = Tin + Tprocess + Tout (1)

where T in is the time taken to queue/modify a packet or transfer it to threads. T process

is the time taken to process packet using an algorithm, and T out time required update
the result/rule set.
Category I: Packet Processing Using Splay Trees and Other Algorithms
In Category I, we consider the techniques which make use of different algorithms
for packet processing. The methods discussed in this category do not use the multi-
threading. Hence, cost of context switching is not involved.
In the paper, proposed by Trabelsi [1], considering Eq. (1) and rewriting it, we
get:
 
Ttotal = (P/S) + (P × 2m) + P × tupdate /S (2)

where Tin = P/S, here P/S represents number of incoming packet segments,
T process = P × 2 m, here m represents memory access provided for each packet
which is at most two memory accesses, and T out = (P × t update )/S, where t update is
time required to update the splay tree filter only if previous traffic is not same. Hence,
we consider the update for different packet segments S.
Performance Analysis of Unified Threat Management (UTM) 127

In a similar paper proposed by Trabelsi [4], considering Eq. (1) and rewriting it,
we get:
 
Ttotal = (P/S) + (P × m) + P × 2 × tupdate /S (3)

where Tin = P/S, here P/S represents number of incoming packet segments,
T process = P × m, here m represents memory access provided for each packet, and
T out = (P × 2 × t update )/S, where t update is time required to update the splay tree filter
only if previous traffic is not same and to calculate the optimum window size. Hence,
we consider the update for different packet segments S.
Again, in paper [5], considering Eq. (1) and rewriting it, we get:
 
Ttotal = (P/S) + (P × 2m) + P × tupdate /S (4)

where Tin = P/S, T process = P × 2 m, here m represents memory access provided for
each packet, T out = (P × t update )/S, where t update is time required to update the splay
tree filter order.
Evaluating paper [6], considering Eq. (1) and rewriting it, we get:
 
Ttotal = (P/S) + (P × 4m) + P × tupdate (5)

where Tin = P/S, T process = P × 4 m, since memory access is frequent for the
greater number of rules, and T out = (P × t update ), here, splay filter gets updated more
frequently.
Evaluating paper [7], considering Eq. (1) and rewriting it, we get:

Ttotal = (P/S) + (P × 15m) + (P × ( tupdate + 2m)) (6)

where Tin = P/S, T process = P × 15 m, here memory is accessed more frequently

and packet needs to be converted to boolean form, and T out = (P × (t update + 2 m)),
updating the filter, calculating the network statistics requires memory usage.
Category II: Parallel Processing Using Multicore Processors
Multicore Processors can parallelly process packets by assigning them to multiple
threads. We evaluate the techniques used in this section. Here context switching with
CPU usage is involved. The thread locking and unlocking delays are ignored in some
techniques (included wherever necessary) since it is very small compared to other
delays.
In the paper [2], considering Eq. (1) we can write:
 
Ttotal = (P × Cswitch 2 ) + (P/T2 ) + P × tupdate /T2 (7)

where T in = (P × C switch2 ), the context switching associated with each packet,

T process = P/T 2 , T 2 represents number of threads created in CPU, and T out = P ×
t update /T 2 , t update represent the time to update the accepted/rejected packet list.
128 J. Gharat et al.

For paper [8], evaluating Eq. (1) we get,

  
Ttotal = (P × 2 × Cswitch 2 × tlock /S) + (P/T2 ) + P × tupdate + tunlock /S (8)

where T in = (P × 2 × C switch2 × t lock /S), here the main thread creates child threads
and queues are locked before multithreading, T process = P/T 2 , T out = (P × (t update +
t unlock /S)), here t unlock represents cost of thread unlocking.
Similarly, for paper [9], the threads are queued according to the stream they belong
to. The queues need to be locked to protect them. Hence rewriting Eq. (1) we get,
 
Ttotal = (P × Cswitch 2 × tlock /S) + (P/T2 ) + P × tupdate /T2 + t + tunleck /SS ))
(9)

where T in = (P × C switch2 × t lock /S), T process = P/T 2 , T out = (P × (t update /T 2 + t +

t unlock /S)), here let t be the delay caused by asynchronous computation between
threads. Let t = m and then we can rewrite Eq. (9) as:
  
Ttotal = (P × Cswitch 2 × tlock /S) + (P/T2 ) + P × tupdate /T2 + m + tunlock /S
(10)

Using the technique presented in paper [10] and re-evaluating Eq. (1) we get,
  
Ttotal = (P × Cswitch ) + (P/T2 ) + P × tupdate + t + 2m (11)

where T in = (P × C switch2 ), T process = P/T 2, T out = (P × (t update + t + 2 m)) here let t

represent the intercommunication delay and m is the memory latency involved. Let
t = m, then Eq. (11) becomes

Ttotal = (P × Cswitch 2 ) + (P/T2 ) + (P × (tupdate + 3m)) (12)

Category III: Parallel Processing Using GPU

In this category, we consider the techniques which make use of highly efficient
Graphic Processing Unit (GPU) for packet processing. The context switching,
C switch1 , in the following papers do not involve CPU time since the packets are
transferred using Direct Memory Access (DMA).
In the paper [3], considering Eq. (1) we can write:
 
Ttotal = (P × Cswitch 1 ) + (P/T1 ) + P × tupdate /T1 (13)

where, T in = P × C switch1 , here only the context switching time for packet queue is
involved, T process = P/T 1 , packets are distributed among threads for processing, and
T out = P × t update /T 1 is the time taken to update the list of accepted/rejected packets.
This process happens in GPU itself.
For paper [11], evaluating Eq. (1) we get,
Performance Analysis of Unified Threat Management (UTM) 129
  
Ttotal = (P × Cswitch ) + (P/Tl ) + P × 4 × tupdate + m (14)

where, T in = P × C switch1 , the context switching time for packet queue is involved,
T process = P/T 1 , T out = (P x (4 × t update + m)), here transition table has to be accessed
frequently. Hence the algorithm is memory bound.
Similarly, for paper [12], rewriting Eq. (1) we get,

Ttotal = (P × Cswitch ) + (P/Tl ) + (P × tupdate ) (15)

where T in = (P × C switch1 ), T process = P/T 1 , T out = (P × t update ), here the result of list
of packets are updated in the host.
Using the technique presented in paper [13] and re-evaluating Eq. (1) we get,

Ttotal = (P × Cswitch ) + (P/T1 ) + (P × (tupdate + 2m)) (16)

where T in = P × C switch1 , T process = P/T 1 , T out = (P × (t update + 2 m)), here metapro-

gramming involves compilation time of the C Program in CPU. Linear Search also
require some time to find the exact match.
Evaluating paper [14] using Eq. (1) we get,
  
Ttotal = (P × Cswitchl ) + (P/Tl ) + P × 4 × tupdate + m (17)

where T in = P × C switch1 , T process = P/T 1 , T out = (P × (4 × t update + m)), here four addi-
tional factors are required to be checked to correctly identify packets. This process
is done in the CPU.

4.2 Numerical Analysis Results

In this study, we used a dataset consisting of 32,506 packets containing 97.6% TCP,
2.3% UDP, and 0.1% ICMPv6 of the total traffic. The average packet size is 730
bytes with 528.5 packets per second. Let us assume that the traffic window contains
S segments with each segment containing 8 packets. For the context switching time,
we can say that C switch1 < C switch2 . This is because while transferring packets from
host to GPU we use DMA technique which reduces the switching time. This is not the
case in multicore CPU systems where CPU processing cost is involved. From paper
[17], we can approximate the values of C switch1 ≈ 4.2 µs and C switch2 ≈ 5 µs. The
memory latency involved is approximated to be m ≈ 100 µs [15]. The table update
cost is approximated [15] to be around t update ≈ 50 µs. The thread locking/unlocking
delay is approximately taken around t lock = t unlock ≈ 100 ns [16]. We consider number
of threads in multicore system, T 2 = 16 while that in Graphics Processing Unit, taking
32 warps and each warp can execute 32 threads concurrently, T 1 = 1024.
The results are shown in Tables 2, 3 and 4. We compare the total packet processing
time (in ms) for each method.
130 J. Gharat et al.

Table 2 Results for Category I

S. No. Title Result (in ms)
1 Hybrid mechanism towards network packet early acceptance and 4064.10
rejection for unified threat management [1]
2 Dynamic traffic awareness statistical model for firewall performance 4064.31
enhancement [4]
3 Multilevel early packet filtering technique based on traffic statistics 4064.10
and splay trees for firewall performance improvement [5]
4 Dynamic scheme for packet classification using splay trees [6] 4066.18
5 Adaptive early packet filtering for depending firewall against DoS 4070.40
attack [7]

Table 3 Results for Category II

S. No. Title Result (in ms)
1 Effective utilization of multicore processor for unified threat 2031.89
management functions [2]
2 Parallel packet processing on multicore and many-core processors [8] 2033.25
3 A parallel packet processing method on multicore systems [9] 2033.58
4 Protecting information systems from DDoS attack using multicore 2034.39
methodology trees [10]

Table 4 Results for Category III

S. No. Title Result (in ms)
1 Firewall engine based on graphics processing unit [3] 31.88
2 IP routing processing with graphic processors [11] 38.71
3 An efficient parallel-network packet pattern-matching approach 33.51
using GPUs [12]
4. Scalable packet classification via GPU metaprogramming [13] 34.16
5 Gnort: high-performance network intrusion detection using graphics 38.71
processors [14]

From the above analysis, we can conclude the following:

• In Category I, the method presented in paper [1] is more efficient than others (see
Fig. 1). The memory access time is limited in this technique to at most two accesses
which helps in faster packet processing.
• In Category II, the method proposed in paper [2] is more efficient (see Fig. 2). This
is due to the fact that multicore functionality is implemented not only for packet
processing but also for packet reception/transmission. Also, the memory is highly
managed to further improve the system.
Performance Analysis of Unified Threat Management (UTM) 131

Fig. 1 Category I

Fig. 2 Category II

• In Category III, the method proposed in paper [3] is most efficient compared
to other techniques (see Fig. 3). The GPU handles most of the network traffic
efficiently and using the zero-copy technique the transferring time is greatly
reduced.
In our further analysis, we compare the best paper in all three categories. From
the overall analysis (see Fig. 4) we can conclude that the technique presented by
Sahoo et al. [3] is most efficient in terms of total packet processing time. Thus, by

Fig. 3 Category III

132 J. Gharat et al.

Fig. 4 Overall comparison

incorporating the processing power of Graphics Card, we can substantially increase

the packet processing speed and improve the system performance to the next level.

5 Conclusion

A UTM device provides high-end security in a single device. We have seen that an
increase in network traffic can create problems and it consumes almost 70% of the
resources. From our analysis, we can conclude the following.:
• The performance can be increased by using multicore functionality in the device
itself. From the analysis, it can be noticed that the processing capabilities of GPU
system outperform any other device capability. Hence, we can say that incorpo-
rating GPUs in our UTM system, we can drastically improve the performance and
overcome any load situations arising as a result of heavy traffic.
• In our future work, we plan to use the multicore functionality of CPU for other
UTM functions like anti-virus, anti-spam detection, etc. once the packet processing
results are received from GPU. Also, we plan to use a self-adjusting tree data
structure for rule-fields ordering in GPU and compare the results for dynamically
changing traffic environment especially for DoS attacks.

Acknowledgements We would like to extend our sincere thanks to Dr. Ramesh Vasapanavara,
Principal, Ramrao Adik Institute of Technology for his constant encouragement and generous
support throughout the completion of this work.

References

1. Trabelsi Z, Zeidan S, Masud MM (2017) Hybrid mechanism towards network packet early
acceptance and rejection for unified threat management. IET Inf Secur 11(2):104–113
2. Gummandi S, Shanmugasundaram R (2012) Effective utilization of multicore processor for
unified threat management functions. J Comput Sci 8(1):68–75
Performance Analysis of Unified Threat Management (UTM) 133

3. Sahoo AK, Das A, Tiwary M (2014) Firewall engine based on graphics processing unit. ISBN
No. 978-1-4799-3914-5, IEEE ICACCCT
4. Trabelsi Z, Zhang L, Zeidan S, Ghoudi K (2013) Dynamic traffic awareness statistical model
for firewall performance enhancement. Comput Secur 39:160–172
5. Trabelsi Z, Zeidan S (2012) Multilevel early packet filtering technique based on traffic statis-
tics and splay trees for firewall performance improvement. IEEE international conference on
communications (ICC)
6. Neji NB, Bouhoula A (2009) Dynamic scheme for packet classification using splay trees. J
Inform Assur Secur 4:133–141
7. El-Atawy A, Al-Shaer E, Tran T, Boutaba R (2009) Adaptive early packet filtering for depending
firewall against DoS attacks. IEEE INFOCOM
8. Harvath A, Nishi H (2015) Parallel packet processing on multi-core and many-core processors.
International conference on parallel and distributed processing techniques and applications
9. Li Y, Qiao X (2011) A parallel packet processing method on multi-core systems. In: IEEE 10th
international symposium on distributed computing and applications to business, engineering
and science
10. Chonka A, Zhou W, Knapp K, Yang X (2008) Protecting information systems from DDoS
attack using multicore methodology. In: IEEE 8th international conference on computer and
information technology workshops
11. Mu S, Zhang X, Zhang N, Lu J, Deng YS, Zhang S (2010) IP routing processing with graphic
processors. IEEE 2010 Des Autom Test Eur Conf Exhib
12. Hung Che-Lun, Lin Chun-Yuan, Wang Hsiao-Hsi (2014) An efficient parallel-network packet
pattern-matching approach using GPUs. Elsevier J Syst Archit 60:431–439
13. Kang K, Deng YS (2011) Scalable packet classification via GPU metaprogramming. IEEE
2011 Des Autom Test Eur
14. Vasiliadis G, Antonatos S, Polychronakis M, Markatos EP, Ioannidis S (2008) Gnort: high
performance network intrusion detection using graphics processors. In: International workshop
on recent advances in intrusion detection
15. Stackoverflow. [Link]
various-caches-and-main-memory
16. .NET Reference Guide. The cost of locks. [Link]
[Link]
17. Li C, Ding C, Shen K (2007) Quantifying the cost of context switch. In: ExpCS’07 proceedings
of the 2007 workshop on experimental computer science
18. Wikipedia. OpenMp. [Link]
19. Rathod PM, Marathe N, Vidhate AV (2014) A survey on finite automata based pattern matching
techniques for network intrusion detection system (NIDS). In: 2014 international conference
on advances in electronics, computers and communi cations (ICAECC)
Development of an Efficient
Nondestructive Grading Method
for Pomegranate Using Magnetic
Resonance Imaging and Neural Network

Surekha Yakatpure and Krupa Rasane

Abstract The pomegranate fruit has gained popularity due to its nutritional values
and pharmacological properties. India ranks high among growers of pomegranates
across the world and hence there is tremendous potential for its export. Here in
our work the Bhagwa, a prime Indian pomegranate cultivar was studied. Total sol-
uble solids (TSS) were measured experimentally. Internal images of the fruit were
obtained nondestructively using Magnetic Resonance Imaging (MRI). The textural
features from image were given as input to the nonlinear autoregressive neural net-
work. The results showed that T1-weighted MR images were sensitive to physical
and chemical changes. The R-value for measured TSS and model-predicted TSS for
training data was 0.99 and testing data was 0.92. This study shows that MRI has
higher potential for nondestructive method of grading pomegranate fruit based on
the chemical values which are the basis for determining the maturity of the fruit.

Keywords Pomegranate · Nondestructive · MRI · Texture · Neural network

1 Introduction

India ranks high among growers of pomegranates across the world as per the statistics
of APEDA [1] and therefore India can use this natural produce to increase the export
of this fruit [2]. To increase exports, it is necessary to improve the quality. If we make
improvement in the quality of fruit then only we can cater the ever-increasing need
and stand tall in the market. For the obvious reason there is demand of developing
innovative non-destructive techniques to characterize internal quality attributes to
cater the ever-increasing market throughout the year by further extending the shelf-
life. To gain consumer appreciation and increase local and export marketing supply of

S. Yakatpure (B)
Electronics and Telecommunication Department, A G Patil Institute of Technology, Solapur, India
e-mail: [Link]@[Link]
K. Rasane
E&C Department, Jain College of Engineering, Belgavi, India
e-mail: kru_ran@[Link]

© Springer Nature Singapore Pte Ltd. 2020 135

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
136 S. Yakatpure and K. Rasane

graded and sorted fruits becomes a norm. Outer quality attributes considered are size,
shape, color, tenderness, hardness can be obtained by just observations and touch
senses. But these methods which are thus conventionally graded can lead to a risk of
internal faults as pale arils, dead tissues and browning, which go undetected, as they
are not externally seen. This marks a grave problem to the food processing units based
on fruits and also endangers the export market causing rejections and reputation of
the cultivar used for export to the destination country. Much work has been done to
estimate the maturity index and quality of fruit, analytically. In many works, data
analysis has been done over different physiochemical values as total soluble solids
(TSS), pH, weight, titratable acidity (TA), etc. were obtained from the instruments
in the laboratory which are laborious, cause loss of fruit and above all much slower
and hence less efficient for speedy quality determination. Nondestructive detection
of internal quality factors is a great challenge. Researchers have been working to
develop new methods which can determine the internal quality attributes without
breaking the product under observation. The various techniques based on operating
principles of optics, acoustics, and ultrasonics do model few physical values related
to quality. Moreover, where some nondestructive techniques as X-ray and CT used
for interpreting internal qualities of fruit, turn hazardous for the biological cells due to
electromagnetic radiations. On the other side, safer techniques of optic and acoustic
result in lesser penetration in the fruit due to varying thickness of skin and pulp.
These methods are not able to measure internal physicochemical values required for
grading based on internal qualities [3]. Magnetic resonance imaging (MRI) has been
introduced around middle of 1980s for studying internal physiochemical values of
various vegetables and fruits. MRI had certain limitations used for inline grading,
firstly due to the small sample size, secondly the rate of processing. As today’s
medical field needed improved power magnets, technology has devised the higher end
requirements for them and hence these facilities can be used for other application as
for nondestructive method for literally diagnosing the fruit internally and thoroughly
for quality grading.

1.1 Magnetic Image Processing Techniques Used for Fruits

The operating principle of MRI consists of secured interaction among radio waves
and hydrogen from the water molecules in the fruit under the influence of strong
magnetic field made available in the assembly. Unlike X-ray and CT imaging, it
bears least ill effects concerned with the magnetic field. MRI, as is sensitive to
hydrogen nuclei, present ample in any biological cell, every detail can be observed
inside the soft tissues of the fruit. Highest benefits of MRI are its potential to vary
the contrast of the images far better than X-ray and CT. With the advancement of
MRI hardware and efficient algorithms for the acquired data and improved methods
of analyzing data, the applications based on MRI have spread the tentacles in the
domain of food science and technology including the fruit industry [11, 12]. Various
work done are summarized in Table 1.
Development of an Efficient Nondestructive Grading Method … 137

Table 1 Physiochemical values determination and grading for fruits and food using MRI as
nondestructive method
S. No. Purpose Type of fruit/food Author Reference
1 Mealiness assessment Apple and peach P. Barriero [4]
and internal
breakdown
2 Water-core Apples Wang S. Y., [5, 6]
Herremans,
3 Internal damage Tomato Milczarek [7]
4 Internal quality Orange Wasiu A. Balogun, [8]
5 TSS Pomegranate Khusroo [10, 11]
6 Physiochemical values Loin Daniel Caballero [12]

Mealiness in apple has been detected using MRI technique [4]. In the apples,
those features representing mealiness were correctly obtained as markers of affected
fruit. Here skewness in histogram was checked and it was observed that the effected
ones were having more skewness than those which are not affected by mealiness.
Water core in apple fruit was related with increased water content and was observed
by MRI technique. It was seen in the image that water core-affected tissues generally
got obvious due to increased intensity for the corresponding region in the image
[5]. The study was done over many varieties and a classifier was modeled using
gray-scale histograms of MRI images to classify the water core apples from the
unaffected ones [6]. Neural network method was used to classify MRI image of
orange fruit into defect or good classes based on pixel intensity level. Here evaluation
of models performance was tested based on R-value and mean squared error [8].
Using this method study was carried out to make the prediction of those features
of the pomegranate which define its internal quality. In the study, it was concluded
that MRI method can estimate many quality features which can be used for grading
[9]. Also, this technique has provided greater knowledge of the internal variations
taking place throughout the stages of growth in pomegranate and has helped detect
maturity in pomegranate fruits [10]. Image attributes were quantified using eleven
co-occurrence matrix texture features and five run-length matrix texture features
extracted from the MRI images of pomegranates. In another work of same author, the
features were used as input to a MLP Neural networks and predicted the maturity with
correlation coefficient, R = 0.93 and R = 0.90 for training and test data, respectively
[11]. Recently, researchers have devised the model to predict the physicochemical
values of loins and related them with 3D features of features obtained from the
MRI imaging. [12]. The aim for this present study is to optimize the work done for
correlation coefficients [11] and develop an efficient method of grading pomegranate
using MRI as a nondestructive method based on the maturity index which is defined
by the physiochemical values. In the work here a model is developed wherein texture
attributes from MRI of pomegranate, combined with nonlinear autoregressive neural
138 S. Yakatpure and K. Rasane

networks technique to improve the correlation coefficient for training and testing
data increasing the grading potential of the model.

2 Materials and Methods

Pomegranates of “Bhagwa”, cultivar which is a prime export variety in India, had

been collected from local market at normal temperature and sent to Shri Markandaya
Solapur, Sahakari Rugnalya and Research Center, for scanning purpose. We have
used a 1.5T MRI scanner (Siemens) with 2D spin-echo sequence for our study. Table 2
shows the comparative table of the parameters for the sample of our project and the
reference work [11]. Figure 1 shows magnetic resonance image of the pomegranate.

Table 2 Comparative parameters used in study

Author MRI TR TE FOV Slice Inter Total Matrix
scanner (ms) (ms) (cm) thickness slice slices
(mm) gap
(cm)
Alireza 1.5T 800 18 29.6 3 1.56 12 336 by
Khoshroo scanner 512
Our 1.5T 800 18 29.6 3 1.56 18 256 by
sample scanner 256

Fig. 1 Magnetic resonance

image of the pomegranate
Development of an Efficient Nondestructive Grading Method … 139

After MRI imaging as shown as an example in Fig. 1, TSS of fruits are obtained
from laboratory of National Research Center of Pomegranate, Solapur. In this study,
to estimate total soluble solids of pomegranates, texture features which are obtained
from texture analysis of magnetic resonance images after preprocessing are combined
with regressive neural network, along with the laboratory values.
Here we have considered a dynamic neural network for our study, where this
model having memory help predict the new coming value of TSS, based on the past
input values. The model can be defined in a simple manner mathematically as

y(to ) = f {y(to − 1), y(to − 2), y(to − 3), u(to − 1), u (to − 2), u (to − 3) . . . u (to − n)}

where the next outcoming value is obtained by feedback of the past output and the
corresponding input. The new output is based on these new inputs to the neural
network. Again the two-layered feed-forward network which is used for correctly
approximating the function defined over the inputs has an advantage that it is best
suited for the training purpose. Also the added grace is that inputs to this second layer
are very accurate. Here the neurons required were 20, and epochs for the improved
correlation were 12.

2.1 Methodological Steps in Detail

Step 1 Pre-processed all images using imadjust which assign I image of particu-
lar intensity values to that of different intensity values present in J image,
wherein this newly formed image shows improved contrast.
Step 2 Our thresholding function

f (i, j) ≥ T r
fˆ(i, j) =
0 otherwise

Here, f (i, j) being original and fˆ(i, j) the filtered image and Tr is
thresholding value, here we applied Tr = 100
Step 3 Calculated GLCM 4 features using GLCM function
Step 4 Created feature vector and target vector based on TSS values
Step 5 Experimental setup: Database is divided into 50, 15, and 35% for training,
validation, and testing. Created neural network model and then tested for
the generated feature space with target values of TSS from the lab values,
which finally predicted TSS values. Later calculated correlation coefficient
using both values and obtained the results. The comparative steps involved
in the experiments and with the experiment performed in the earlier work
[10, 11] referred with due courtesy are shown in Table 3.
 140

Table 3 Comparison of experimental steps for the referred work and present work
Author Pre-processing Parameter Feature extraction Classification Post-processing Outcomes
AlirezaKhoshroo [10, NA TSS GLCM 10 feature Neural network Correlation Training = 0.93
11] using MaZda 2.11 and LR = 0.1 coefficient Between Testing = 0.90
RLM (run length Momentum = 0.7–0.7 Experimental TSS
matrix) 5 feature Neurons = 16 and Predicted TSS
Epoch = 5000
Transfer function =
hyperbolic-tangent
Ours Noise removal (using TSS GLCM 4 features Neural network Correlation Training = 0.99
stdfil, modified (using Matlab 17) LR = 0.1 coefficient between Testing = 0.92
thresholding filter), Momentum = 0.7–0.7 experimental TSS and
image enhancement Neurons = 20 predicted TSS
(histogram stretching) Epoch = 12
S. Yakatpure and K. Rasane
Development of an Efficient Nondestructive Grading Method … 141

Fig. 2 R values obtained

while training of the data

Fig. 3 R values obtained

while testing of the data

3 Results

For the experiment, from 324 images of pomegranate, 162 images for training, 113
images for testing and 49 images for cross-validation were used. Textural features
extracted from the images were given as input to the network. The R-value for
measured TSS and model-predicted TSS for training data was 0.99 and testing data
was 0.92, and are shown in Figs. 2 and 3 in the order. Experimentation has been done
with Matlab 17.

4 Conclusions

Nondestructive imaging of pomegranate based on magnetic resonance imaging pro-

vides a greater potential to determine maturity and internal breakdown, which are
controlled by the internal physiochemical values. For the internal attributes, four co-
occurrence matrix texture features were extracted from the images. These features
were provided as input to the NARX neural network which predicted higher correla-
tion coefficient, R = 0.99 and R = 0.92 for training and test data, respectively, as com-
pared to the MLP architecture. With the improved correlation results obtained, the
proposed future work for grading the pomegranate based on physiochemical values
142 S. Yakatpure and K. Rasane

satisfying those of export values can be carried out. With the advent of deep learning
neural network techniques, fruit grading for different cultivars of pomegranate can
be implemented to predict the maturity and internal breakdown of the fruit extending
the availability of the fruit in the market. The improved results states that MRI as
nondestructive method of grading is evolving with a promising technique for on-line
sorting of pomegranate fruit and with ever-increasing improvement in MRI facilities
this goal will be achievable in near future.

Acknowledgements I admire the technical support of National Research Center on Pomegranate

(NRCP), Solapur. I express my gratitude to Dr. Jyotsna Sharma, Director In charge, NRCP and Dr.
K. Dhinesh Babu, Principal Scientist, for their support.

References

1. The Agricultural and Processed Food Products Export Development Authority (APEDA)
2. Ganeshkumar Rede, Santosh Yumnam (2016) Performance of pomegranate export from India.
Econ Aff 61:575–580. [Link]
3. Kim SM, Chen P, McCarthy MJ, Zion B (1999) Fruit internal quality evaluation using online
nuclear magnetic resonance sensors. J Agric Eng Res 74:293–301
4. Barriero P, Ruiz-Cabello J, Fernandez-Valle ME, Ortiz C, Ruiz M (2000) Mealiness assessment
in apples and peaches using MRI techniques Article. Magn Reson Imag 18(9):1175–1181
5. Wang SY, Wang PC, Faust M (1988) Non-destructive detection of water-core in apple with
nuclear magnetic resonance imaging. Scientia Hortic 35:224–234
6. Herremans E, Melado-Herreros A, Defraeye T, Verlinden B, Hertog M, Verboven P, Nicolai BM
(2014) Comparison of X-ray CT and MRI of water core disorder of different apple cultivars.
Postharvest Biol Tech 87:42–50
7. Milczarek R (2009) Multivariate image analysis: an optimization tool for characterizing
damage-related attributes in magnetic resonance images of processing tomatoes. Ph.D.
dissertation, University of California, Davis, CA., USA
8. Balogun WA, Salami MJE, McCarthy MJ, Mustafah YM, Aibinu AM (2013) Intelligent tech-
nique for grading tropical fruit using magnetic resonance imaging. Int J Sci Eng Res 4(7):216
(ISSN 2229-5518)
9. Zhang L, McCarthy MJ (2013) Assessment of pomegranate postharvest quality using nuclear
magnetic resonance. Postharvest Biol Technol 77:59–66
10. Khoshroo A, Keyhani A, Zoroofi RA, Rafiee S, Zamani Z, Alsharif MR (2009) Classification
of pomegranate fruit using texture analysis of MR images. Agric Eng Int CIGR E J. 11:1182
11. Khoshroo A, Keyhani A, Zoroofi RA, Yaghoobi G (2011) Non-destructive Inspection of
pomegranate maturity using magnetic resonance imaging and neural networks. CIGR Section
VI international symposium on towards a sustainable food chain food process, bio processing
and food quality management
12. Caballero Daniel, Antequera Teresa, Durán Marisa, Caro Andres (2017) Applying 3D texture
algorithms on MRI to evaluate quality traits of loin. Art J Food Eng 222:258–266
Implementation of Differential Privacy
Using Diffie–Hellman and AES
Algorithm

P. V. Vivek Sridhar Mallya, Aparna Ajith, T. R. Sangeetha, Arya Krishnan

and Gayathri Narayanan

Abstract Differential privacy is a method adopted to check for any privacy breach
that occurs during communication for the exchange of confidential information. Here,
in this work, differential privacy is being implemented in the context of vehicular
ad hoc networks (VANETs). In this paper, we implement the concept of differential
privacy using the Diffie–Hellman key exchange algorithm and the advanced encryp-
tion standards (AES) algorithms that are very powerful in terms of their performance.
Algorithms like Laplace and Gaussian algorithms, which are currently the most com-
monly implemented algorithms, have been used for verification. The algorithms were
analyzed by considering a situation where an initial location and final location have
been defined and these have been encrypted using the mentioned algorithms and the
privacy has been preserved.

Keywords Differential privacy · VANETs · Diffie–Hellman algorithm · AES

1 Introduction

Differential privacy helps us obtain accurate data from a collected set of data and also
to handle the privacy issues connected to it while dealing with data. One common
application of this concept is with regard to individual database. Differential privacy
aims in identifying the difference within the database of each and every individual.
Even if the individual is an active or passive entry in the database, there should
not be any difference to the individual. An adversary should not be able to access
anything new from the database containing the information of the individual. For
this purpose, different algorithms have been adopted such as Diffie–Hellman key
exchange algorithm and advanced encryption algorithm.
At first, we consider a few set of auxiliary information of a particular individual.
The auxiliary information of the individual or group like name, occupation, etc.,

P. V. Vivek Sridhar Mallya · A. Ajith · T. R. Sangeetha (B) · A. Krishnan · G. Narayanan

Department of Electronics and Communication Engineering, Amrita Vishwa Vidyapeetham,
Amritapuri, Kerala, India
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 143

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
144 P. V. Vivek Sridhar Mallya et al.

constitutes the statistical database. The statistical database contains the private infor-
mation of a particular group which has been collected by a particular entity. This
information might be medical reports of an individual or routine information which
are associated with the identity of a person. A privacy breach occurs when an adver-
sary hacks this particular data. The adversary hacks this information by accessing a
statistical database [1].
Privacy is of supreme importance, especially in today’s transparent society. It
demands the protection and safe handling of data associated with individuals. The
concept of differential privacy aims at maximizing the correctness of the queries
from the defined databases while measuring how the privacy is affected on individ-
uals whose information is stored in the database. Search engines, hospital records,
corporate firms, and other such institutions possess large amounts of data much of
which are very sensitive. Such data are no longer just important to their domain
agencies but are becoming widely common to all those who use that data. In mak-
ing the data available for public, they also need to face the consequences in terms
of the legal, financial, and moral pressure since they are also bound to protect the
identities of the individuals. Hence, a trade-off between privacy and utility is of great
importance.

2 Differential Privacy

2.1 Need for Differential Privacy

Consider a situation in which we want to take the medical records of a person “K.” If
a privacy breach occurs and all identifiable information such as name, identity proof
number, mobile number, etc., had to be removed from the original data set, then the
available records would contain only information such as date of birth, gender, and
address details. These fields are more than sufficient to provide a unique combination
that they are often required to identify an individual from the database. The technique
used here is to pick out the information that is linked with “unidentified” records by
tagging it with some additionally available data. Generally, when people are asked
for their personal information, they are hesitant to provide it, for fear of the data
being misused in some way or the other. Nevertheless, if the data are used in a
differentially private manner, then the individual can be assured that the data will be
virtually hidden and that it will not be available to any third party for misuse. This
would prompt the users to provide their data since they know that it is secured in the
right sense.
Implementation of Differential Privacy Using Diffie–Hellman … 145

2.2 Epsilon-Differential Privacy

Dwork, McSherry, Nissim, and Smith in 2006, in their article, introduced the concept
of epsilon-differential privacy. The main agenda behind the definition of epsilon-
differential privacy in 2006 is that it is not possible to sacrifice the privacy of an
individual by a statistical release of their data from the database. So, the goal here is
to provide each and every individual the same privacy that would result from having
their data removed; that is, the statistical functions running on the database should
not depend too much on their data of any of the individual. A mechanism provides
differential privacy if for all databases x, x  which are adjacent and for all z ∈ Z, we
have
P(K = z/ X = x)
≤ e
P(K = z/ X = x  )

The result is the -differential privacy [2]. Here,  (privacy parameter) is a positive
real number.  can be viewed as a relative measure of privacy. Its value can be
chosen upon analyzing the goal of hiding any person’s presence or absence in a given
database [3].

2.3 Application Scenario—VANETs

VANETs or vehicular ad hoc networks are exchange networks where data packets
are exchanged between vehicles which are commonly referred to as common nodes
traveling on defined constrained path. In vehicular networks, there is a chance of
privacy risks where vehicles could be tracked by the information transmitted by the
vehicle-to-vehicle (v2v), vehicle-to-infrastructure (V2I), or vehicle-to-x (V2X) com-
munications implemented with the dedicated short-range communications (DSRC)
standards operating at 5.9 GHz. VANET requires fully decentralized network control
since no central entity could or should organize the network [4].
In VANETs, the nodes, which are vehicles, are installed with on-board units
(OBUs) to them to communicate with other vehicles in the vicinity and also to sta-
tionary structures as the application demands. VANETs therefore form a special case
of sensor networks. Here, in wireless sensor networks (WSNs) the sensor nodes are
installed near the roadside to monitor the road condition and to send the information
about dangerous conditions to vehicles regardless of the connectivity of the VANET.
Each vehicle has the functionality of being a packet transmitter, receiver, or router
which enables the vehicles to communicate with other vehicles or access points
which form the stationary roadside units (RSUs) [4].
There are three main types of vehicular communication system. They are:
1. Vehicle-to-Vehicle (V2V)—Using vehicle-to-vehicle, a vehicle can detect the
position of and movement of other vehicles which are half a kilometer away.
146 P. V. Vivek Sridhar Mallya et al.

This topology allows mobile-to-mobile interface among vehicles. Pure ad hoc

communication is actually vehicle-to-vehicle communication. Each vehicle in
V2V communication has a lot of gadgets attached to global positioning system
(GPS) networking devices, sensors, and digital map which contain the route map
and relevant information and computing devices. These vehicles communicate
with neighboring vehicles by sensing its own traffic messages and sending beacon
messages to the other vehicles periodically. The data communication techniques
implemented in V2V communication are unicast and multicast packet forwarding
techniques between the source and destination vehicles. Here, unicast means
sending or receiving packets to and from its own direct neighbors. Unlikely,
multicast forwarding helps to exchange packets with remote vehicles also by
making use of intermediate vehicles as relays.
2. Vehicle-to-Infrastructure (V2I)—V2I is the intelligent technology for the next
generation. This topology allows interface between vehicles and roadside units.
Here, the infrastructure plays an important role in coordinating the communica-
tion by collecting local or global information on the traffic conditions in the road
and then suggesting some of the behaviors on a group of vehicles. For example,
one technique which employs V2I is the ramp metering. It requires limited sen-
sors and actuators for measuring the traffic density. V2I communication which is
enabled by a system of hardware, software, and firmware, by nature is wireless.
The major implication is that infrastructure components such as lane markings,
signboards, and traffic signals or any other infrastructure can wirelessly provide
information to the vehicle and vice versa.
3. Vehicle-to-X (V2X)—It is referred to as vehicle-to-everything communication.
In V2X communication, information from various sensors on the OBUs and other
sources travel via a large bandwidth, low latency, and high-reliability links, giving
the experience of fully autonomous driving. All the other types of networks fall
under this topology. By following this type of network configuration, vehicles
can easily communicate with other vehicles, to infrastructural units like traffic
signals or smart parking lots, to pedestrians holding their cell phones, and also
to central hubs or data centers using cellular networks.

3 System Model and Implementation

Trajectories are generated by vehicles in large numbers, which are significantly dif-
ferent from pedestrian locations. Assuming T = {location 1, location 2, location 3,
…, location n}, where T is a set consisting of vehicular locations appearing in a
sequential manner. Every point denotes a location in trajectory, which is an ordered
pair comprising of longitude and latitude. In order to conform to the standards that
are required to be met in order to ensure the protection of privacy, some interference
items are added to generate the published trajectory TR = {T 1, T 2, T 3, …, T n}. By
referring to the map, the third-party hackers can clip the interference items because
Implementation of Differential Privacy Using Diffie–Hellman … 147

of the limitation of roads in VANETs [5]. The most precarious of the situation would
be that if the hackers are able to obtain even a small set of values out of the total
location space, they will be able to interpolate the points in between and thereby
generate a trajectory which will bear a very close resemblance to the original trajec-
tory. Keeping this in perspective, this project aims at creating a cost-effective indoor
localization and tracking system that does not require any infrastructure assistance.
Just with the help of smartphones, the user can be localized.
A differential privacy protection protocol is implemented in the Java language
based on the Diffie–Hellman key exchange algorithm and the (AES) Advanced
Encryption Standard 256 algorithm. Firstly, a symmetric key is generated by the
D–H key exchange algorithm, and then, the AES 256 algorithm is used to encrypt
the transmitted content, thereby protecting the privacy of the user information. The
idea of the differential privacy protection prototype can be understood with the help
of a simple observation: when the data set D contains the individual Alice, it is set to
perform random query operations on D (such as counting, summing, mean, median,
interquartile range, or other related queries). The result obtained is f (D). On carrying
out the operation, if the result yielded on querying Alice’s information from D is still
f (D), it can be considered that Alice’s information is not included in data set D. In the
middle of it creates additional risks. Differential privacy protection is to ensure that
any individual in the data set or in the data set has little impact on the final published
query results. The goal of differential privacy is to maximize query accuracy and
minimize the risk of privacy breaches [6].
The Diffie–Hellman key exchange algorithm, denoted as D–H algorithm, is a very
powerful protocol in public-key encryption. It allows the parties to institute a key
over an unsecured channel with no previous intimation from the other party. The
advantage being that this key can be used as a tool, a symmetric key, to encrypt the
communication content in ensuing communications [7].

3.1 D–H Key Exchange Algorithm

Consider a scenario where we have Alice, Bob, and Eve as shown in Fig. 1. Alice
and Bob are exchanging some sort of data among each other, and Eve being an
intermediator who is constantly watching Alice and Bob. But, here Eve, she does not
disturb the subject or content of the communication. Now, consider a public prime
base say g = 3 which is known to all the three. Next, consider a public data known
to Alice, Bob, and Eve which are p = 12. Now, consider Alice’s key (private key)
which is known to her alone which is a = 6. Next, consider Bob’s private key known
only to Bob as b = 15.
Let A denotes the public key of Alice known to Alice, Bob, and Eve where
 
A = G a mod p = 9

Let B denotes the public key of Bob known to Alice, Bob, and Eve where
148 P. V. Vivek Sridhar Mallya et al.

Fig. 1 Conceptual representation of the D–H algorithm

 
B = G b mod p = 7

Next, Alice and Bob exchange each of their public keys. Let s denotes the shared
secret key common to both Alice and Bob, but which is not known to Eve.
For Alice,
 
s = B a mod p = 1

For Bob,
 
s = Ab mod p = 1

So, even if a hacker attacks, he will not be able to provide the actual data because he
does not know the key to decode it. In real scenario, consider Alice and Bob represent
two vehicular ad hoc networks, while Eve represents the transmitting medium or
mediator. Thus, the above algorithm is real-life implementation of VANETs [8].
Implementation of Differential Privacy Using Diffie–Hellman … 149

3.2 Advanced Encryption Standard Algorithm

The advanced encryption standard is an encryption method in which the characters

are replaced by some other kind of operation such as addition, subtraction, exclusive-
or, or any such similar operators. By this, it is possible to produce infinite number
of combinations. Subbytes, shiftrows, mixcolumns, and addroundkey are some of
the techniques which can be used for making rotations. The AES algorithm consists
of AES-128, AES-192, and AES-256 block ciphers. The three variants of AES are
based on different key sizes (128, 192, and 256 bits) [9].

4 Results

This section shows the results that have been taken in terms of the location values,
encrypted and protected. The values considered are assumed to be data taken from
the on-board unit of a car which conveys the exact location of the vehicle. As can be
seen, the initial location conveyed is the location of the house, assuming that the car
is parked at the house and is starting from there (Figs. 2 and 3).
In this work, we are presenting the comparison between the above algorithms for
the following choice of values as shown below:
k = 28
q = 117686837408564885453421841793708809347228146689137677955112976
83943548586519985787950858001297977310173110996763173093591148833
987835653326488053279071057
p = 3295231447439816792695811570223846661722388107295854982743163351
5041936042255960206262402403634336468484710790936884662055216735165
9398293141665491813989597
g = 2936079689421990213596959618831699609892128325127375389414219745
98430263629524708683392530750033382195532008794395609507801074545519
29347513371119440534090946249252931698618289251122473408140692956145
46100216572732567527559463356954948075734550545288593249436517685280
135130178110107249280453035685425826946616

Setup:

k01 = 728873398441318548710301394139479150676176393886425534408519502
48251796148100167365253903634953054476322942772847795760914393651709
75316064521074730852743

N = 50

Setup use time is 95 ms

150 P. V. Vivek Sridhar Mallya et al.

Fig. 2 Output data as computed by the algorithm

ke1 = 0

Phase 2: Encrypt use time is 7 ms

ke2 = 0

Encrypt xN = 26520

Decrypt xN = 26520

Decrypt use time is 9 ms

k is the number of bits in the data, p and q are the encryption key known to two
vehicles, and g is known to vehicles and the adversary. After using Diffie–Hellman
algorithm and advanced encryption method, keys are produced which is not known
to adversary; that is k01, ke1, and ke2 are the key used for encryption. The encryption
time and decryption time have also been calculated [10].
Implementation of Differential Privacy Using Diffie–Hellman … 151

Fig. 3 Output data as computed by the algorithm

5 Conclusion

Differential privacy can be effectively used to ensure security in vehicular ad hoc

networks. The contribution of the work is that this approach has been implemented in
the context of security in vehicular ad hoc networks. The implementation of the algo-
rithms for finding the trajectory of motion of vehicle when start and end longitude
and latitude is available along with a matrix defined containing the locations nearby
is obtained. Also, the comparison between Laplace algorithm and Median algorithm
was done which are commonly used to implement differential privacy in VANETs.
Along with this, we also obtained the result for the implementation of two currently
used algorithms Diffie–Hellman key exchange algorithm and advanced encryption
standards algorithm. In today’s society driven by big data, where these data hold enor-
mous significance, it is necessary to ensure that privacy is not compromised. This
work is a small attempt in that direction. The project aims at creating a cost-effective
indoor localization and tracking system that does not require any infrastructure assis-
tance. Thus, just with help of smartphones, the user can be localized. Although it is
a promising area of research and one of extreme importance, this concept has not
been explored to its full capability due to limitations in understanding it completely
and also due to its non-trivial method of implementation.
152 P. V. Vivek Sridhar Mallya et al.

References

1. Roth A (2010) Algorithms for preserving differential privacy, CMU-CS-10-135 (references)

2. Kressner D, Implementing differential privacy. Institute of Information System, School of
Business and Economics Humboldt University of Berlin, Term paper for the IT Security
Seminar
3. Ding Z, Wang Y, Wang G, Zhang D, Kifer D (2019) Detecting violations of differential privacy.
arXiv:1805.10277v4 [[Link]]
4. Baldini G, Giuliani R (2017) Analysis of the privacy threat in vehicular ad-hoc networks due
to radio frequency fingerprinting
5. Nelson B, Olovsson T (2017) Introducing differential privacy to the automotive domain: oppor-
tunities and challenges. In: 2017 IEEE 86th vehicular technology conference. [Link]
10.1109/vtcfall.2017.8288389
6. Pradeep LN, Bhattacharjya A (2013) Random key and key dependent S-box generation for
AES cipher to overcome known attacks. In: Security in computing and communications. Berlin,
Heidelberg
7. Hay M, Machanavajjhala A, Mikalu G, Chen Y, Zheng D, Principled evaluation of differentially
private algorithms using D.P Bench
8. Diffie-Hellman key exchange. [Link]
key_exchange
9. The Advanced Encryption Algorithm. [Link]
e32fdd267aaa4240a4464723bc74d0a5
10. Neha, Kaur M (2016) Enhanced security using hybrid encryption algorithm. Int J Innovative
Res Comput Commun Eng 4(7)
11. Rewagad P, Pawar Y (2013) Use of digital signature with Diffie Hellman key exchange and
AES encryption algorithm to enhance data security in cloud computing. In: Proceedings of the
international conference on communication systems and network technologies
12. Ganesh AR, Manikandan PN, Sethu SP, Sundararajan R, Pargunarajan K (2011) An improved
AES-ECC hybrid encryption scheme for secure communication in cooperative diversity
based wireless sensor networks. In: International conference on recent trends in information
technology, ICRTIT 2011, Chennai, pp 1209–1214
Implementation of Neural Signals
in MATLAB (Thought Signals)

Kambhampati Sai Sandilya, Poornima Mohan,

Madiraju Akshay Bharadwaj, B. Maragatha Eswari, K. Namitha,
O. Rajasekhar Reddy and J. Vaishnu Saran

Abstract The most intelligent creatures on the earth are homosapiens since their
brain is gifted with the ability of thinking and expressing emotions through different
means. The human brain is a universe consisting of a cluster of neurons connected
to each other. There are about 100 billion neurons in the human brain. Estimated
that a neuron connection transmits at least one signal per second, and some theories
proved that some of the specialized connections transmit up-to 10,000 signals per
second [1]. Briefly, we can say that when we experience something through sense
organs which are having direct contact with our brain, release some or other chemical
called hormones. For instance, if we feel something ‘very good to our heart!’ or like
start loving something by sensing anything, our brain produces a hormone called
‘oxytocin’. As there is an extraction of this hormone happens in a ‘Patterned’ manner
from an import gland of our human body called as the pituitary gland, the pattern
creates an electrical signal/impulse which transmitted to our brain in a fraction of
seconds, making us experience the emotion called love on that particular object or

K. S. Sandilya (B) · P. Mohan · M. A. Bharadwaj · B. M. Eswari · K. Namitha · O. R. Reddy ·

J. V. Saran
Department of Electronics and Communication Engineering, Amrita School of Engineering,
Kollam, Kerala, India
e-mail: knvsspss@[Link]
P. Mohan
e-mail: poornimamohan@[Link]
M. A. Bharadwaj
e-mail: akshaybharadwaj98@[Link]
B. M. Eswari
e-mail: maragathaeswari@[Link]
K. Namitha
e-mail: namitha96k@[Link]
O. R. Reddy
e-mail: orsr97@[Link]
J. V. Saran
e-mail: jvaishnusaran@[Link]

© Springer Nature Singapore Pte Ltd. 2020 153

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
154 K. S. Sandilya et al.

person. The electrical impulse created here and transmitted to the brain is nothing
but a thought/emotion signal. Our main aim is to extract a converted written text or
an image of the electrical impulse created, this electrical impulse is extracted out
from a human brain through EEG signals.

Keywords EEG signals · Patterned signals · Electrical impulses · Neurons

1 Introduction

1.1 What Is a Thought Signal?

Thoughts are the effect of simulation that happens when we feel touch, smell, taste,
sound or sight through the five sense organs namely skin, nose, tongue, ears and
eyes. These stimuli when we sense something will have direct neural contact with
the brain generating a special kind of fluid from the glands present in our body known
as the hormones. The amount of production, the speed, and timing of the production
of these hormones produce stimuli or an electrical impulse in our brain, which are
generally a form of signals in electronics terminology and are known as thoughts
in general case. These can also be ideas, believes on self or on others. Assume, the
human body as a machine made up of a microcontroller called the brain, which is
made of some sensors or microprocessors called sense organs which can send and
receive the impulses or signals from the brain. These signals are passed through a
medium for which these sensors are connected to the brain, which is called nerves.
Every sensor is made up of a very large scale integrated circuits (VLSI circuits)
called transistors [2]. Similarly, our nerves are made of nerve cells or neurons [3, 4].
Let us consider an example of a tennis player playing a tennis match. As the player
watches the approaching ball, two types of senses activate immediately, his/her sight
on the ball, and the sound of the air through which the ball is coming [5, 6]. Then
the impulses are sent to the brain and the brain sends some signals to the motors of
the machine that our hands and legs to move, to hit the ball. Everything and every
thought process happen the same [7, 8]. But, it is not known how to visually look at
these signals, which are in our mind. Let us have an assumed theory, to visually find
our thoughts. With this, we have another application also, we can see our dreams as
an image signal [9, 10] (Fig. 1).
The above figure illustrates how a thought signal is generated in a tennis player
when he is standing in the tennis court and the ball is running towards him.
Implementation of Neural Signals in MATLAB (Thought Signals) 155

Fig. 1 Stimulation of thought signals in a tennis player

2 Methodology

2.1 Brain and Computer Interface (BCI)

At present in the field of Biomedical Engineering, Brain and Computer Interface

(BCI) has a huge demand [11]. BCI technique is that creates a great link between
the computer and the human brain. Nowadays BCI technique is mostly used in
Artificial Intelligence (AI), that is in controlling robots, through the human brain,
mobility commanding system, the best example is Stephen Hawking’s chair and
communication fields. Here the signals are captured as an Electroencephalography
(EEG) signals through a device called EMOTIV EPOC EEG amplifier [1] (Fig. 2).
This amplifier consists of electrodes that are fit to the head of the subject and
extracts the signal. This raw signal is transferred into the computer through a
Bluetooth interface and saved as a .edf file [11].

2.2 Methods to Implement

Selecting the Subjects. The EEG information to be collected from some people aged
20–30 years approximately [11].
156 K. S. Sandilya et al.

Fig. 2 EEG measurement

positions on the head

Training the Subjects. The subjects are given with different kind of inputs. For
instance, Subject A is given something to look at, and similarly, all other subjects
are given something to sense with so that their brains start producing the electrical
impulse [11] (Fig. 3).
Collecting and Storing the Signal. The generated signal is collected from the
EMOTIV EPOC EEG amplifier and stored into the computer in the form of .edf or
.mat files through Bluetooth and taken into MATLAB for further evaluation. The
EEG measurements on the brain are explained in the following Table 1.
Signal and Image Processing. EEG signals are generated in the form of Analog
signals. Analog Signals are continuous signals which have time-varying quantities.

Fig. 3 Block diagram illustrating the signal transmission into MATLAB

Implementation of Neural Signals in MATLAB (Thought Signals) 157

Table 1 The EEG functional

Lobe Function EMOTIV EEG
measurements on the brain
(positions/channels)
[11]
Frontal Emotions, cognition AF3, AF4, F3, F4, F7,
F8, FC5, FC6
Parietal Movement P7, P8
Occipital Sight, vision O1, O2
Temporal Hearing T7, T8

EEG signals vary from 5 to 500 Hz of the sampling frequency. As the .edf file is
converted into the .mat form, it can be directly used into the MATLAB. To plot the
EEG signal in MATLAB, initially, after loading the .mat file into the MATLAB,
subtract the value provided ‘Gain’ and divide the whole value with the provided
‘Base’.

val = load(‘Subject00_2_ed f [Link]’) (1)

eeg = (val − Gain)/Base; (2)

Now for plotting the cure, we need to calculate the time of the curve with the
provided sampling frequency and obtained signal.

time = (0 : length(eeg) − 1)/Fs (3)

Fs = Sampling Fr equency

When we try to plot the signal with these specifications given, a continuous analog
signal with about (1 × 5000) double length is plotted as shown in (Fig. 4)
Now, the signal is generated as an Analog signal. This Analog signal needs to
be converted to Digital signal using Analog to Digital conversion (ADC) technique.
In ADC technique, calculate the quantization value with the given needs range and
number of bits of the required plot.
 
q = r/ 2n − 1 ; (4)

q = quantization, r = range of the signal, n = number of bits.

Now, we divide the signal with the obtained quantization value and will get a
quantized signal, and use decimal to binary conversion of the quantized analog values,
which will convert the analog values of the quantized EEG signal into binary values.
Plot the binary values with respect to the original raw EEG plot, with a decrease
in the number of values in the original signal to make the process easy. But, the
drawback in the method is we need to take the absolute values of the original signal
158 K. S. Sandilya et al.

Fig. 4 Plot of raw EEG signal

to plot the binary plot. The below plot represents the binary converted EEG signal
[12] with respect to the original raw signal.
Figure 5 represents the plot of digitalized EEG signal with respect to the absolute
raw signal.
Figure 6 represents the illustration of the only Digitalised EEG signal. This is the
plot of all the converted digital values of quantized Analog values.

Fig. 5 Plot of a digitalized EEG signal with respect to raw EEG signal
Implementation of Neural Signals in MATLAB (Thought Signals) 159

Fig. 6 Plot of a digitalized EEG signal

3 Results and Future Scope

3.1 Results

After digitalizing the signal, modulate the digital signal either in Amplitude (AM) or
Frequency (FM) format if needed. Digital values are obtained from the digitalized
signal. These digitalized values are arranged as a matrix, such that when the matrix
is plotted in the form an image using ‘imshow’, an in-built command in MATLAB,
converts the produced matrix into an image by considering each value in the matrix
as a pixel. Here, a rough sketched image of the digitalized EEG signal is generated,
where ‘0’ value of the signal is taken as black colour and ‘1’ value as white and
displays a pixelated image.
Figure 7 Interpreters an object that is shown to ‘Subject00_2’, one of the subject
in the experiment.

3.2 Future Scope

This BCI (Brain and Computer Interface) can be improved further and can be used
in getting a visual image of the dreams of people. Most of its applications can be
used for those people who are in the state called ‘COMA’, but can hear and couldn’t
respond to the outer world. Their thoughts can be reconstructed as an image and
can rectify what the problem is. Involving the machine learning concept to this may
increase the accuracy and speed of the resultant.
160 K. S. Sandilya et al.

Fig. 7 Plot of a digitalized

EEG signal in the form of an
image signal

4 Conclusion

The above-mentioned research work clearly explains, how a EEG thought signal
collected from a human brain can be interpreted as an image. The output must be
furthermore modelized in a more equipped and developed apparatus, like increasing
the pixels of the image and using other image processing techniques such that the
subjects’ thoughts are purely displayed as an image in either black and white or RGB
frame.

References

1. Thomas T, James M, Shaji RR, Pillai BC (2016) Interpretation of human stages from EEG
signals using LabVIEW. Int J Adv Res Comput Commun Eng 5(3)
2. Soman KP, Manikandan S, A novel method for detecting R-peaks in electrocardiogram (ECG)
signal. J Biomed Sig Process Control
3. Gurumurthy S, Mahit VS, Ghosh R (2013) Analysis and simulation of brain signal data by
EEG signal processing technique using MATLAB. Int J Eng Technol 5:2771–2776
4. Sutradhar S, Sayadat N, Rahman A, Munira S, Fazlul Haque AKM, Sakib S (2017) IIR
based digital filter design and performance analysis:1–6. [Link]
8343596
5. Joy J, Peter S et al (2013) Denoising using soft thresholding. Int J Adv Res Electr Electron
Instrum Eng 2(3):1027–1031
6. Walters-Williams J, Li Y (2012) BMICA independent component analysis based on B-Spline
mutual information estimation for EEG signals. Can J Biomed Eng Technol 3(4):63–79
7. Kroupi E, Yazdani A et al (2011) Ocular artifact removal from EEG: a comparison of subspace
projection and adaptive filtering methods. In: 19th European signal processing conference.
Barcelona, Spain
Implementation of Neural Signals in MATLAB (Thought Signals) 161

8. Akhtar MT et al (2009) Focal artifact removal from ongoing EEG-A hybrid approach based on
spatially constrained ICA and wavelet denoising. In: 31st annual international signal conference
of the IEEE EMBS Minneapolis, Minnesota, USA, 2–6 Sept 2009
9. Prasad VVKDV et al (2013) A new wavelet packet based method for denoising of biological
signals. Int J Res Comput Commun Technol 2(10):1056–1062. Simranpreet Kaur1 IJECS
Volume 3 Issue 8 August, 2014 Page No.7965-7973 Page 7973
10. Abdullah H et al (2013) Double density wavelet for EEG signal denoising. In: 2nd international
conference on machine learning and computer science, Kuala Lumpur, Malaysia, 25–26 Aug
2013
11. Sulaiman N, Hau CC, Hadi AA, Mustafa M, Jadin S (2014) Interpretation of human thought
using EEG signals and LabVIEW. In: E2014 IEEE international conference on control system,
computing and engineering, Penang, Malaysia, 28–30 Nov 2014
12. Kazi MM, Rode YS, Dabhade S, Dawla NAl, Mane AV, Manza RR, Kale KV, Image retrieval
and interested image detection for thought processing using EEG signals. UACEE Int J Comput
Sci Appl 2(2). [ISSN 2250 - 3765]
A Framework for Formal Verification
of Security Protocols in C++

R. Pradeep, N. R. Sunitha, V. Ravi and Sushma Verma

Abstract Every communication system is a safety-critical system, in which the

communicating entities share the confidential data over the untrusted public network
by using a set of cryptographic security protocols (CSPs). Many security protocols
proved secure were cracked within a short span of time, and the best example is
Needham–Schroeder authentication protocol. The quality assurance about the cor-
rectness of security protocols is one of the key challenges. In software testing, it is
not possible to prove the correctness of security protocols, because testing has got
major drawbacks and the tester cannot predict what knowledge an attacker may gain
about the communication system by interacting with several runs of the protocol,
and also testing shows the presence of bugs but can never show the absence of bugs.
Formal verification has proved to be a reliable solution as the correctness of the CSP
can be proved mathematically. In the proposed work, a new framework is proposed,
which includes a library of functions to specify a security protocol in C++ by follow-
ing a set of rules (syntax and semantics), a interpreter to interpret the C++ code to
security protocol description language (SPDL), and finally a model checker Scyther
backend verification engine. The proposed framework is successful in identifying
the attacks on IKE version-1. Also the Skeme and Oakley versions were verified for
their correctness.

Supported by Defence Research and Development Organisation (DRDO), SAG, New Delhi.

R. Pradeep (B) · N. R. Sunitha · V. Ravi

Department of Computer Science and Engineering,
Siddaganga Institute of Technology, Tumkur 572103, Karnataka, India
e-mail: pradeepr@[Link]
N. R. Sunitha
e-mail: nrsunitha@[Link]
V. Ravi
e-mail: ravi@[Link]
S. Verma
Defence Research and Development Organization, (SAG), New Delhi, India
e-mail: sushmaverma@[Link]

© Springer Nature Singapore Pte Ltd. 2020 163

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
164 R. Pradeep et al.

Keywords Formal verification · Model checking · Internet Key Exchange (IKE) ·

Scyther · Skeme · Oakley · CSP

1 Introduction

CSPs help to achieve secure communication between the communicating agents.

They are at the core level of most of the communication systems, such as secure
Internet communications, bank transactions, or any electronic way of fund trans-
fers. Such applications should not be delivered without verifying the crypto-analytic
weaknesses of the underlying algorithms.
Security protocols are difficult to design, even under the assumption of perfect
cryptography. Many CSPs claimed as safe and used for many years were “cracked”
within a short span of time, one such protocol is Needham–Schroeder authentication
protocol [1] proposed in 1978, and this protocol was used for decades to provide
authentication for agents using a hybrid method which involves both public-key and
symmetric-key cryptosystems. The Needham–Schroeder protocol was cracked by
Galvin-Lowe [2] in 1995 using FDR [3], and a new updated version called Needham–
Schroeder-Lowe protocol was released. This shows that there is a need for formal
verification of security protocols.
In the development phase of security protocols, if the design errors or functionality
errors are introduced, then it is very difficult to detect and debug with testing. To detect
and correct the design errors and to achieve high reliability of CSP, formal verification
is one of the key solutions and it gives a mathematical proof for the correctness for the
security protocols. There are three main approaches for formal verification of security
protocols [4]: 1. BAN-logic, 2. Casper/FDR (model checking), and 3. Stranspace.
BAN-logic is a light-weight method under the assumption of honest agents and a
passive intruder. The Casper/FDR approach translates a high-level description of a
security protocol along with its security requirements and a particular instantiation
into CSP that can be machine-verified using the FDR model checker [5], and the
intruder is in control of the network and is allowed to participate as one of the agents.
The same holds for the strand space approach which is basically a Dolev-Yao model
[6] which represents a strong adversary and his capabilities. For of all these three
approaches, the state space explosion problem is a challenge, in which for a system
with “N” number of concurrent processes the number of states increases in the global
state graph may grow exponentially with N. [7] Explains a solution to avoid state
explosion problem with the help of temporal logics.
In order to use the CSPs in network devices like router, switch, and modems, the
protocols have to be implemented using a programming language like C, C++, or any
other similar high-level programming language. But there is no mechanism available
to formally verify a security protocol which is implemented using a programming
language C++. In the proposed work, a new framework is proposed to perform formal
verification of security protocols in which a library of security primitives is speci-
fied in C++ by following a set of rules (syntax and semantics). A new interpreter
A Framework for Formal Verification of Security Protocols in C++ 165

is proposed, which converts a CSP specified in C++ language to security protocol

description language (SPDL), and a library of APIs is proposed to support crypto-
graphic primitives for CSP specification. In the proposed framework, the Scyther
model checker [8] is used as back end verification engine, Scyther has support for
multiprotocol black-box analysis mechanism, it is efficient in handling state space
explosion problem, and Scyther uses Dolev-Yao channel to model the adversary and
guarantees termination of verification process.
To communicate securely over a public network, computers make use of vir-
tual private network (VPN), and VPNs use IKE security protocol to exchange keys
between the agents, to authenticate mutually without revealing the identities of the
agents. This protocol needs to be formally verified because the agents will be sharing
private data over a VPN which is created over a public network. The main challenge
in formal verification of security protocols is, building abstract models from the for-
mal specifications, identifying the security properties to be satisfied by a security
protocol. In the proposed framework, the secrecy and authentication security prop-
erties can be verified for the security protocols, for verification of IKE protocol, the
formal specifications are drawn from the RFC 2409 [9], the framework is successful
in verifying the IKE version-1, Skeme [10], and Oakley. The framework is successful
in identifying the attacks for IKE version-1. The Oakley and Skeme version security
properties are proved safe.

2 Proposed Framework to Formally Verify Security

Primitives of CSP Coded in C++

Formal verification is a challenging task and requires a lot of effort and knowledge,
but the automated tools help to perform verification with ease, these tools are called
model checkers, and the verification technique is called formal verification using
model checking. A model checker is required to evaluate the security properties by
searching through the state space to check whether even a single state violates the
security properties specified.
The main challenge in formal verification of security protocols is generating the
abstract CSP model and identifying the security properties to be verified. Building an
abstract CSP model depends totally on the model checker being used, it changes from
one model checker to other, and there is no mechanism to verify a CSP implemented in
a high-level language. To solve this problem, in the proposed work, a new framework
for specifying a security protocol in C++ language is developed and it is interpreted
to the backend model checker Scyther using an interpreter program. Figure 1 shows
the architecture of the proposed framework, in which the CSPs will be represented
in an abstract level and the implementation details are hidden in order to build the
CSP model easily. Using this framework, the CSP model can be specified in C++
language using a set of cryptographic library functions. The library contains all
the cryptographic primitives such as generation of keys, nonces values, encryption,
166 R. Pradeep et al.

Fig. 1 Framework architecture

decryption, hash functions, key exchange, send, and recv. The library is extendable
for user requirements and the definitions for the library functions can be changed
based on the user requirements.
To build a CSP model using the proposed framework, a protocol structure has
to be followed as shown in Listing 1.1. The nesting of classes is used to bind the
communicating agent behaviors in a single class. The outer class represents the
protocol name along with the communicating role names, and the inner classes
represent the communicating agents named as roles. The agents communicate with
each other by sending and receiving messages. This takes place by calling the send
and recv functions.
The role definition is the set of library function calls, which are bind in a single-
member function. The role objects communicate by calling the single-member func-
tion, and thereby, all the communication takes place. Listing 1.2 gives an overview
about the CSP specification using the proposed framework.

2.1 Scyther Model Checker as Verification Engine

Scyther is a formal analysis and verification tool, and it uses black-box analysis
technique for verification of CSPs under the perfect cryptography assumption, which
states that adversary learns nothing from the ciphertext unless he has the decryption
key. The tool is basically used to identify the design errors of the security protocols.
The protocols either proved correct or attacks will be detected.
Scyther uses Dolev-Yao threat model [11] as channel to model the adversary,
on which the agents communicate, and the adversary has the upper hand and he
can eavesdrop every message, perform reply attacks, can have knowledge about the
cryptographic primitives, can remove sent messages and examine their contents,
insert his own messages, or reroute or simply retransmit messages.
A Framework for Formal Verification of Security Protocols in C++ 167

The key feature of Scyther is multiprotocol analysis, in which during the

communication between the roles, the communicating agents use multiple CSPs
in parallel, and then the attacker can gain the knowledge about the cryptographic
primitives from one protocol run and can successfully perform an attack on the
other. Scyther helps to perform formal analysis and verification of multi-protocols
CSPs as well as a single CSP in isolation.

2.2 Dolev-Yao Channel

Dolev-Yao adversary model [11] is used to model the adversary, on which the agents
(communicating parties either a sender or receiver) communicate, and the adversary
has the upper hand and he can eavesdrop every message, perform reply attacks, can
have knowledge about the cryptographic primitives, can remove sent messages and
examine their contents, insert his own messages, or reroute or simply retransmit
messages.
The Dolev-Yao adversary model will be running in parallel with the security pro-
tocol model. This model views the message space as a term algebra. Term derivation
rules specify how agents can obtain new terms from old ones. Fix countable sets A,
N, K, and V, denoting the set of agents, nonces, keys, and variables, respectively.
The set of basic terms is B = A ∪ N ∪ K. For each A, B ∈ A, assume that sk(A),
pk(A) and shk(A,B) are keys. Further, each k ∈ K has an inverse defined as follows:
inv(pk(A)) = sk(A), inv(sk(A)) = pk(A), and inv(k) = k for the other keys.
The important features of the Dolev-Yao model are: 1. secrecy properties, 2.
stateless parties, 3. concurrent execution, and 4. public-key cryptography and infras-
tructure. [11] Explains all these features in detail.
The intruder has access to all public information and can block and replay any
terms sent by honest agents. The intruder is essentially the network itself—any terms
sent by honest agents are received by the intruder and then forwarded onto the
intended recipient (or not), and any terms received by honest agents are assumed to
have come from the intruder. He can also send terms under masquerade, in the name
of any agent (all agents’ names are public). The intruder can send out any terms he
can derive from the set of terms he has access to. An agent A might send out a term
t for the intended recipient B, and the intruder might send a term t to B, pretending
to be A, thereby effecting a forgery in A’s name. A key component in checking for
what terms the intruder can get access to is solving the derivability problem. The DY
channel helps to model the channel which gives a strong adversary for verification
of CSP.
168 R. Pradeep et al.

2.3 The C++ to SPDL Interpreter

Formal verification cannot be done directly for a CSP implemented in a high-level

language like C++, because the model checker cannot process C++ code. In order to
overcome this language mismatch and to use a model checker for verification of CSP
in C++ language, an interpreter is required which can convert C++ code to security
protocol description language (SPDL) code. In the proposed framework, to specify a
protocol in C++ a set of rules (syntax and semantics) must be followed. The protocol
structure will be as follows.
Listing 17.1 Protocol structure
hashfunction g( ) ;
class protocol_protocolname_rolename1_rolename2 / / Outer class
{
class rolename1 / / Inner class1
{
void function1 ()
{

}
};

class rolename2 / / Inner class2

{
void function2 ()
{

}
};
};
int main()
{
protocol_protocolname_rolename1−rolename2 : : rolename1 obj1 ;
protocol_protocolname_rolename1−rolename2 : : rolename2 obj2 ;
obj1 . function1 ( ) ;
obj2 . function2 ( ) ;
return 0;

The communication between the agents takes place by sending and receiving
messages, and this is specified in the protocol description by using send and receive
functions. The syntax for send and recv functions is as follows:
send(sequence − number, Sender − r ole − name, Receiver − r ole − name,
data);
r ecv(sequence − number, Sender − r ole − name, Receiver − r ole − name,
data);
sequence-number—is a message sequence number, Sender-role-name- class name
of Sender role, Receiver-role-name- class name of Responder role.
An example of message exchange is illustrated in Listing 1.2
A Framework for Formal Verification of Security Protocols in C++ 169

Listing 17.2 Message exchange

hashfunction g( ) ;
class protocol_demo_role_Initiator_role_Responder
{
class role_Initiator
{
fresh x;
void fun1()
{
send(1 , Initiator , Responder, encrypt (g(x) ,k( I ,R) ) ) ;
claim( Initiator , Secret , x ) ;
}
};
class role_Responder
{
var x;
void fun2()
{
recv(1 , Initiator , Responder, decrypt (g(x) ,k( I ,R) ) ) ;
claim( Responder, Secret , x ) ;
}
};

};
The data can be sent with encryption or without encryption during the commu-
nication by using send function. If the data has to be sent securely, then encryption
can be used, and this is done using an encryption function in the data field of send
function, as shown in Listing 1.2. For the encryption function, the last parameter
must be a key, either a public key, secret key, or a session key, and its syntax is as
follows.
encrypt(parameters... , key);
decrypt(parameters... , key);
The encryption function encrypts all the parameters passed to it by using the key
which is the last parameter in the encryption function, and it returns a ciphertext. In the
proposed framework, the RSA encryption/decryption algorithm is used for public-
key encryption and 128-bit AES algorithm is used for symmetric-key encryption.
Keys—pk(I) represents the long-term public key of role I, sk(I) represents the
long-term secret key, and k(I,R) represents the long-term session key of role I and
role R. If a public key or secret key is used to encrypt the data, then a public-key
encryption function RSA is called. If a symmetric key is used to encrypt the data,
then AES encryption library function is called. To use a cryptographic hash function,
a hash function has to be declared as global functions as shown in Listing 1.2, in
order to use it by the roles. In the proposed framework, MD5 hash algorithm is used
to generate unique hash values. Finally, the security properties to be formally verified
are represented using the claim functions as shown in Listing 1.2. The syntax for
claim function is as follows:
claim(Role-name,Security-Property,data);
170 R. Pradeep et al.

Role-name: The name of the role which is claiming the property, Security-
Property: Property to be satisfied by the role for the data (the different security
properties are discussed in the below paragraph), data: The data on which the prop-
erty to be checked.
Security properties for CSPs are very important, and to verify a CSP, one should
know the exact security properties to be satisfied by a CSP. The security properties
are represented in claim functions, and its mathematical meaning is as follows.
Claim—Let γ be a claim role event of a protocol P. For a security property, it is
required that some predicate Q on traces(P) × ClaimRunEv holds for each instance
of γ in all traces of P. A property is true for the protocol if and only if: ∀ t ∈ traces(P)
∀(inst, γ ) ∈ t : Q(t, (inst, γ )) where we use the notation e ∈ t as an abbreviation for
∃ i: ti = e. There will be many security properties for a particular CSP to be proven
correct, and the proposed framework focuses on secrecy and authentication.
Secrecy property states that the sent or received message must be secretly deliv-
ered to the honest agent without revealing the message to the intruder.
Authentication is identification of the right communication partner across the
public network. There are three forms of authentication, namely 1. aliveness, 2.
synchronization, and 3. message agreement.
Aliveness is checking of the existence of communication partner.
Synchronization verifies the sequence of messages exchanged between the
agents, and thereby, the behaviors of the communicating agents are checked. In the
presence of an active adversary, other behavior (not defined by the protocol descrip-
tion) might occur. For example, if there is no agent performing a certain role, this
contradicts the protocol description. We consider any behavior that is not specified, to
be unauthenticated. This leads to a natural definition of strong authentication, which
is called synchronization.
Message Agreement requires that the contents of the received messages cor-
respond to the sent messages, as specified by the protocol. As a result, after the
execution of the protocol the contents of the variables will be exactly as specified by
the protocol. After the execution of the protocol, the parties agree on the values of
variables. Modex [12] is a tool to extract verification models from implemented C
code to the SPIN model checker. The proposed interpreter works similar to Modex
by interpreting each line of CSP C++ code, with help of regular expressions and
pattern matching the C++ code is converted to SPDL code, Finally, the generated
model is executed by the Scyther (Figs. 2 and 3).

Fig. 2 Interpreter
A Framework for Formal Verification of Security Protocols in C++ 171

Fig. 3 Interpreter architecture

3 IKE-Internet Key Exchange Protocol

For formal verification of IKE protocol, the formal specifications are drawn from
the RFC-2409 [9], and the security properties to be verified are also obtained from
the same RFC. IKE is the protocol used to set up a security association (SA) in the
IPsec protocol suite. IKE builds upon ISAKMP, Oakley, and Skeme. It uses X.509
certificates for authentication and uses either pre-shared or distributed keys using
DNS and a Diffie–Hellman key exchange to set up a shared session secret from
which cryptographic keys are derived. IKE is used in most of the communication
devices to provide authentication and key exchange between the agents, particularly
in establishing a VPN connection to a remote host by hiding the identities of the end
parties. It uses two basic methods to establish an authenticated key exchange—Main
Mode and Aggressive Mode. [9] Gives the details about these modes. The ISAKMP
SA is the shared policy and keys used by the negotiating peers in IKE protocol to
protect the communication. Oakley and Skeme each define a method to establish
an authenticated key exchange. Oakley defines modes, and Skeme defines “phases.”
IKE uses public-key encryption for authentication and exchanges key between the
entities with the help of Nonces. SA is a negotiation payload with one or more
security proposals provided by the sender. An initiator can provide multiple proposals
for security negotiation, but the responder must reply with only one SA. IKE uses
hash functions to uniquely identify the messages, and prf() is the pseudo-random
function—often a hash function—used to generate a deterministic output that appears
pseudo-random. prfs are used both for key derivations and for authentication. Listing
1.3 shows the IKE communications. HDR is an IKE header whose exchange type
is the mode, HDR* indicates payload encryption, SA is an SA negotiation payload
with one or more proposals, and KE is the key exchange payload which contains
the public information exchanged in a Diffie–Hellman exchange. The IKE Oakley
172 R. Pradeep et al.

version uses three modes—1. Main Mode, 2. Aggressive Mode, and 3. Quick Mode.
[9] Explains these three modes in detail along with packets exchanged.
Listing 17.3 IKE message exchanges
Initiator Responder
−−−−−−−−−− −−−−−−−−−−−
HDR, SA −−> HDR, SA
<−−

HDR, KE, Ni −−> HDR, KE, Nr

<−−

HDR∗, IDii , HASH_I −−> HDR∗, IDir , HASH_R

<−−

4 Verification

A model checker is required to generate all possible behaviors of protocol under

verification, it is called state space or search space, and it is stored in a tree structure.
The verification of each security property is done by the model checker by searching
through the state space for each security property represented as claim events. For
search operation, the model checker uses a DFS or BFS algorithm. The claim events
represent the safety property of verification, which states something bad is never
going to happen. For a security property, if there is a trace pattern in the state space
which violates the claim, then the corresponding attacks can be obtained with the
help of model checker by traversing from the initial state to the state where the claim
property was failed, the path or trace gives the execution or behavior of the system
which violates particular property, and the trace includes the detail how the claim
got failed and the possible attack can be derived from the attack graph. If there is
no trace pattern which violates the security property claim, then it is proved safe. A
trace is a partially ordered set of symbolic events, and a pattern represents a set of
traces. The set of pattern events or Event is defined by the following BNF grammar:
AdvEvent ::= decr ({| RunTerm |} RunTerm ) |
encr ({| RunTerm |} RunTerm ) |
app (Func(RunTerm∗ )) | i n i t | know (RunTerm) ,
SendRecv ::= send | recv ,
RolePos ::= Role^#RID | Agent,
CommEvent ::= SendRecv Label (RolePos, RolePos,RunTerm)^RID ,
ClaimEvent ::= claim Label (RolePos, Claim [ , RunTerm]) ,
AgEvent ::= CommEvent | ClaimEvent,
PatternEvent ::= AdvEvent | AgEvent.
A Framework for Formal Verification of Security Protocols in C++ 173

5 IKE Verification Results

The IKE version-1 in main mode is vulnerable to the brute force attack on keys
exchanged between the agents. The agents share a set of (falsely) authenticated sym-
metric keys with the attacker, the attacker gets the symmetric keys, and the authentica-
tion security property gets violated. The proposed framework successfully identified
this attack on IKE version-1, and it is shown in Fig. 4. IKE Oakley and Skeme ver-
sion uses quick mode for security association negotiation and exchange of nonces
to provide protection against replay attack. The nonces are used to generate fresh
key material and prevent replay attacks from generating bogus security associations,
with the help of nonces and Diffie–Hellman key exchange, and the protocol is proved
safe in Skeme and Oakley. The results for Skeme and Oakley versions were shown
in Figs. 5 and 6, respectively.

6 Conclusion and Future Work

The formal verification of cryptographic security protocols has the potential to

improve the reliability of the security protocols and can give mathematical proofs
for CSP’s correctness. Building the model for the security protocols and identifying
the security properties is one of the key challenges in formal verification of CSP.
By using the proposed framework, the CSP models and its security properties can
be easily specified and verified for CSPs specified in C++. The IKE security proto-
col is formally verified using the proposed framework, and the results are obtained

Fig. 4 IKE version-1 verification result

174 R. Pradeep et al.

Fig. 5 IKE-Skeme version verification result

Fig. 6 IKE Oakley version verification result

as shown in Figs. 4, 5, and 6. The framework shows how a model checker can be
used to verify CSP code specified using a high-level programming language C++.
Finally using proposed framework, we can achieve high reliability of the CSPs. The
framework can be further extended to additional security property requirements, and
the same methodology can be incorporated to specify CSPs in other high-level pro-
gramming languages like Java and Python, and the new interpreters are required to
interpret CSPs specified in high-level languages like Java or Python with different
model checkers.
A Framework for Formal Verification of Security Protocols in C++ 175

References

1. Needham RM, Schroeder MD (1978) Using encryption for authentication in large networks of
computers. Commun ACM 21(12):993–999. [Link] http://
[Link]/10.1145/359657.359659
2. Lowe G (1996) Breaking and fixing the Needham-Schroeder public-key protocol using FDR.
In: Margaria T, Steffen B (eds) Tools and algorithms for the construction and analysis of
systems. Springer, Berlin, pp 147–166
3. Gibson-Robinson T, Armstrong P, Boulgakov A, Roscoe AW (2014) FDR3—a modern refine-
ment checker for CSP. In: Ábrahám E, Havelund K (eds) Tools and algorithms for the con-
struction and analysis of systems. Springer, Berlin, pp 187–201
4. Cremers C, Mauw S, de Vink E (2003) Formal methods for security protocols: three examples
of the black-box approach. NVTI Newsl 7:21–32. Newsletter of the Dutch Association for
Theoretical Computing Scientists
5. Roscoe AW (1994) Model-checking CSP. In: A classical mind. Prentice Hall International
(UK) Ltd., Hertfordshire, UK, pp 353–378. [Link]
6. Herzog J (2005) A computational interpretation of Dolev-Yao adversaries. Theor Comput Sci
340(1):57–81
7. Clarke EM, Grumberg O (1987) Avoiding the state explosion problem in temporal logic model
checking. In: Proceedings of the sixth annual ACM symposium on principles of distributed
computing. ACM, pp 294–303
8. Cremers C, Mauw S (2012) Operational semantics and verification of security protocols.
Springer
9. Carrel D, Harkins D (1998) The Internet Key Exchange (IKE). RFC 2409. [Link]
17487/RFC2409, [Link]
10. Krawczyk H (1996) Skeme: a versatile secure key exchange mechanism for internet. In: Pro-
ceedings of the 1996 symposium on network and distributed system security (SNDSS ’96).
SNDSS ’96, IEEE Computer Society, Washington, DC, USA, p 114.[Link]
cfm?id=525423.830460
11. Herzog J (2005) A computational interpretation of Dolev-Yao adversaries. Theor Comput
Sci 340(1):57–81. [Link] [Link]
science/article/pii/S0304397505001179. Theoretical Foundations of Security Analysis and
Design II
12. Holzmann G (2003) The spin model checker: primer and reference manual, 1st edn. Addison-
Wesley Professional
Locality—Aware Scheduling
for Containers in Cloud Computing

G. Charles Babu, A. Sai Hanuman, J. Sasi Kiran and B. Sankara Babu

Abstract The cutting edge scheduler of containerized cloud administrations con-

siders load balance as the main rule, numerous other imperative properties, including
application execution, are ignored. In the period of Big Data, applications advance
to be progressively more information escalated and subsequently performed inade-
quately when conveyed on containerized cloud administrations. With that in mind,
this paper means to enhance the present cloud administration by considering appli-
cation execution for the cutting edge compartments. The more explicitly, in this
work we fabricate and break down another model that regards both burden equaliza-
tion and application execution. Dissimilar to earlier examinations, our model edited
compositions the predicament between burden equalization and application execu-
tion into brought together steaming issue and after that utilizes a factual technique
to effectively settle it. The most difficult part is that some sub-issues are amazingly
unpredictable (for instance, NP-hard) and heuristic calculations must be formulated.
To wrap things up, we actualize a framework model of the proposed planning proce-
dure for containerized cloud administrations. Exploratory outcomes demonstrate that
our framework can fundamentally support application execution white safeguarding
generally high burden balance.

Keywords Cloud computing · Load balance · Containerized cloud

administration · Safeguarding · NP-hard

G. Charles Babu (B)

Department of CSE, Malla Reddy Engineering College (Autonomous), Secunderabad, Telangana,
India
e-mail: Charlesbabu26@[Link]
A. Sai Hanuman · B. Sankara Babu
Department of CSE, Gokaraju Rangaraju Institute of Engineering & Technology (Autonomous),
Bachupally, Telangana, India
J. Sasi Kiran
Department of CSE, Farah Institute of Technology, Chevella, Telangana, India

© Springer Nature Singapore Pte Ltd. 2020 177

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
178 G. Charles Babu et al.

1 Introduction

The previous couple of years have seen a developing number of versatile and sensor
applications that depend on Cloud support. The job of the Cloud is to enable these
asset constrained gadgets to offload and execute a portion of their register serious
undertakings in the Cloud for vitality sparing and additionally quicker preparing.
Notwithstanding, such offloading to the Cloud may result in high system overhead
which isn’t reasonable for some portable/sensor applications that require low idle-
ness. In this paper, we propose a territory mindful burden-sharing method that permits
edge assets to share their remaining task at hand so as to keep up the low inertness
necessity of Mobile-Cloud applications. In particular, we examine how to figure out
which edge hubs ought to be utilized to impart the remaining task at hand to and
the amount of the outstanding burden ought to be shared to every hub. Our trials
demonstrate that our region mindful burden-sharing procedure can keep up low nor-
mal start to finish idleness of portable applications with low inertness variety while
accomplishing great use of assets within the sight of a dynamic remaining task at
hand.

Objectives
• To examination prescient investigation of territory mindful storage tier information
obstructs over Hadoop;
• To plan area mindful burden part taking in versatile distributed computing;
• To structure toward territory mindful planning for containerized cloud adminis-
trations;
• To examination territory mindful booking for holders in distributed computing.

2 Literature Review

2.1 Structuring Your Paper

LaValle et al. [1] the term “Enormous Data investigation’s alludes to an extensive
scale answer for overseeing mammoth datasets in a parallel domain. Hadoop is a
biological community that forms vast datasets in appropriated processing situation.
The biological community is additionally classified into four sub-ventures, for exam-
ple, HDFS, MapReduce, YARN, and Hadoop Commons. The Hadoop Distributed
File System (HDFS) is a spine of biological system, which helps putting away and
handling substantial datasets. As of late, HDFS is moved up to heterogeneous capac-
ity level condition that adapts to information square handling over various capacity
gadgets for example Plate, SSD, and RAM. The square position strategy dispatches
information squares to the gadgets without figuring I/O exchange parameters and
territory points of view. In addition, HDFS chooses arbitrary Datanodes that could
Locality—Aware Scheduling for Containers in Cloud Computing 179

be situated into the following rack having longer way than neighborhood rack. This
expands the information square handling dormancy and results in a colossal post-
ponement for reproduction the board in heterogeneous capacity level. To determine
this issue, we propose a prescient examination that fabricate a territory mindful
capacity level hub outline and anticipate the most adjacent accessible capacity level
for square occupation preparing. The test assessment delineates that the proposed
methodology diminishes information square exchange time overhead, copy exchange
time overhead and diminishes hub ways to an ideal availability over the bunch.
Cloudera [2] in this paper, we propose a system for security safeguarding redis-
tributed medication revelation in the cloud, which we allude to as POD. In particular,
POD is intended to enable the cloud to safely utilize different medication recipe sup-
pliers” medication equations to prepare Support Vector Machine (SVM) given by
the logical model supplier. In our methodology, we configuration secure calculation
conventions to enable the cloud server to perform generally utilized number and divi-
sion calculations. To safely prepare the SVM, we plan a safe SVM parameter choice
convention to choose two SVM parameters and develop a safe consecutive negligible
enhancement convention to secretly invigorate both chose SVM parameters. The pre-
pared SVM classifier can be utilized to decide if a medication substance compound
is dynamic or not in a security safeguarding way. In conclusion, we demonstrate that
the proposed POD accomplishes the objective of SVM preparing and concoction
compound order without protection spillage to unapproved parties, just as exhibiting
its utility and productivity utilizing three certifiable medication datasets.
Kala Karun and Chitharanjan [3] cloud-helped Internet of Things (IoT) gives a
promising answer for information blasting issues for the capacity imperatives of
individual items. Nonetheless, with the influence of cloud, IoT faces new security
challenges for information commonality between two gatherings, which is presented
without precedent for this paper and not as of now tended to by customary method-
ologies. We research a protected cloud-helped IoT information overseeing technique
to keep information classification when gathering, putting away, and getting to IoT
information with the help of a cloud with the thought of clients increase. The pro-
posed framework novelly applies an intermediary re-encryption plot, which was
proposed in \cite{XJW15}. Henceforth, a protected IoT under our proposed strategy
could oppose most assaults from the two insiders and untouchables of IoT to break
information classification, and in the interim with consistent correspondence cost for
re-encryption against gradual size of IoT. We further demonstrate the technique is
handy by numerical outcomes.
Abbas et al. [4] Information spillage is the coincidental revelation of delicate data
through relationship of records from a few databases/accumulations of a cloud infor-
mation distribution center. Vindictive insiders represent a genuine danger to cloud
information security and this legitimizes the emphasis on data spillage because of
maverick representatives or to pariahs utilizing the qualifications of authentic work-
ers. The exchange in this paper is confined to NoSQL databases with an adaptable
construction. Information encryption can lessen data spillage, however, it is illogical
to scramble huge databases and additionally all fields of database records. Encryption
restricts the tasks that can be carried on the information in a database. It is along these
180 G. Charles Babu et al.

lines, basic to recognize touchy reports in an information stockroom and focus on

endeavors to secure them. The limit of a spillage divert presented in this work eval-
uates the instinctively evident intends to trigger cautions when an insider aggressor
utilizes unreasonable PC assets to correspond data in various databases. The Sensi-
tivity Analysis dependent on Data Sampling (SADS) presented in this paper adjusts
the exchange offs between higher effectiveness in recognizing the dangers presented
by data spillage and the exactness of the outcomes gotten by testing vast accumula-
tions of reports. The paper gives an account of trials surveying the viability of SADS
and the utilization of specific disinformation to confine data spillage. Cloud admin-
istrations distinguishing touchy records and decreasing the danger of data spillage
are additionally talked about.
Tsuruoka [5] with the appearance of distributed computing, an ever-increasing
number of individuals will in general re-appropriate their information to the cloud.
As a basic information use, secure watchword look over scrambled cloud information
has pulled in light of a legitimate concern for some specialists as of late. Be that as
it may, the majority of existing looks into depend on a perfect suspicion that the
cloud server is “interested however genuine”, where the list items are not checked.
In this paper, we think about an all the more difficult model, where the cloud server
would most likely carry on unscrupulously. In view of this model, we investigate
the issue of result confirmation for the protected positioned catchphrase seek. Not
the same as past information confirmation plans, we propose a novel obstacle based
plan. With our cautiously formulated check information, the cloud server can’t know
which information proprietors, or what number of information proprietors trade stay
information which will be utilized for confirming the cloud server’s rowdiness. With
our methodically structured confirmation development, the cloud server can’t know
which information proprietors’ information are installed in the check information
support, or what number of information proprietors’ confirmation information are
really utilized for confirmation. All the cloud server knows is that when he carries on
untrustworthily, he would be found with a high likelihood, and rebuffed truly once
found. Besides, we propose to enhance the estimation of parameters utilized in the
development of the mystery confirmation information cradle.

3 Methodology

The predictive analysis consists of two phases, i.e., (i) Storage-tier summary container
and (ii) Predict the most nearby Datanode.
The storage-tier summary container collects all the Datanode and storage media
information, i.e., computing capacity and storage-tier devices with volume statis-
tics. Moreover, the media predictor performs training sessions over the dataset and
predicts the most nearby Datanode with available storage-tier media as seen from
Fig. 1.
Locality—Aware Scheduling for Containers in Cloud Computing 181

Fig. 1 Storage tier

predictive analysis over
HDFS cluster

Fig. 2 Storage-tier
summary architecture

Storage-tier Summary Container

The summary container consists of Datanode information, i.e., CPU, storage media,
accessibility time, 1 MB data block receiving time and volume sizes of each storage
(Fig. 2).

4 Results

Environment

The biological community comprises of Intel Xeon processor with 8 CPUs, 32 GB

memory, and capacity gadgets, for example, 1 TB Hard plate drive and 128 GB
Samsung SSD. Notwithstanding that, we use Intel center i5 with 4 Core, 16 GB
memory and capacity gadgets for example 1 TB Hard circle drive and 128 GB
Samsung SSD. We introduce 5 virtual machines having virtual box 5.0.16 as observed
from Table.
182 G. Charles Babu et al.

Fig. 3 Storage-tier summary

container message collector

The exploratory dataset comprises of:

(i) 250 arbitrary SSD word check information squares of 64 MB (40 GB size),
(ii) 250 irregular DISK word tally information squares (40 GB size) and
(iii) 250 arbitrary RAM word check information squares (40 GB size).

Storage-tier Summary Collector

The collector fetches event traces of computing capacity, storage-tier I/O, Acces-
sibility Path timestamp, Datablock receiver timestamp and Volume Space statistics
over container. The message length varies between 0.5 ≤ size ≥ 5 KB and consumes
a resource between 0.2 ≤ Bandwidth ≥ 500 KB/s. The summary container stores
2.7 GB of log information over 120 GB data blocks as observed from Fig. 3.
After generating container messages, we perform prediction simulations over
three “250” random data blocks. In the first hour of simulation, we observe that
predictor detects pattern of “109” SSD data blocks, “78” DISK data blocks and “63”
RAM data blocks. In the second hour of simulation, we use “500” random data blocks
and analyze that predictor observes pattern of “211” SSD data blocks, “192” DISK
data blocks and “97” RAM data blocks. In the third hour of simulation, we evaluate
“750” random data blocks and evaluate that predictor observes pattern of “322” SSD
data block, “219” DISK data blocks and “109” RAM data blocks as observed from
Fig. 4.
The media predictor depicts locality-aware nearby Datanode statistics with avail-
able functional media. The predictor lists processing timestamp, accessibility times-
tamp, and completion timestamp of data blocks over respective storage media. As
a result, we calculate locality-aware path and observes that proposed processing,
node, and storage-tier approaches are 39.1, 54.7, and 22.9% efficient than default
data block processing. This reduces storage-tier latency, node latency, and overall
processing latency as observed from Fig. 5.
Locality—Aware Scheduling for Containers in Cloud Computing 183

Fig. 4 Storage-tier media

block job prediction

Fig. 5 Latency optimization

over HDFS cluster

5 Conclusion

In this paper, we think about the issue of burden-sharing to deal with runtime ele-
ments in a Mobile-Edge Computing (MEC) condition. Our inspiration depends on
the dynamic property of the outstanding burden in MEC alongside the low idle-
ness prerequisite for a considerable lot of the present portable/IoT applications. The
Edge Cloud stage that has been proposed to give computational loading backing to
versatile applications faces extra difficulties in taking care of outstanding burden
elements since the hubs in Edge Clouds are ordinarily associated by WAN with high
system inertness and constrained transmission capacity. We propose a region mindful
burden-sharing method that permits edge hubs to share their outstanding task at hand
184 G. Charles Babu et al.

to different hubs to meet the low dormancy prerequisite of the versatile applications
on account of remaining task at hand increments. Our heap sharing system enables
hubs to
(1) Intelligently decides if to share their outstanding burden to different hubs,
(2) Selectively picks which hubs the outstanding burden ought to be imparted to,
and
(3) Determines the amount of the remaining burden ought to be shared. Our trial
results dependent on a genuine Twitter’s follow demonstrate that our territory
mindful burden-sharing strategy can keep the general dormancy of versatile
applications near the applications’ ideal objectives just as better use assets even
on account of dynamic outstanding task at hand.
Headings should be capitalized (i.e., nouns, verbs, and all other words except
articles, prepositions, and conjunctions should be set with an initial capital) and
should, with the exception of the title, be aligned to the left. Only the first two
levels of section headings should be numbered. Kindly refrain from using “0” when
numbering your section headings.

References

1. LaValle S et al (2011) Big data, analytics and the path from insights to value. MIT Sloan Manag
Rev 52.2:21
2. Cloudera (2016) The modern platform for data management and analytics, Cloudera [Online].
Available [Link] Accessed 13 Mar 2017
3. Kala Karun A, Chitharanjan K (2013) A review on Hadoop—HDFS infrastructure extensions.
In: 2013 IEEE conference on information and Communication Technologies
4. Abbas A, Wu Z, Siddiqui IF, Lee SUJ (2016) An approach for optimized feature selection in
software product lines using union-find and genetic algorithms. Indian J Sci Technol 9(17)
5. Tsuruoka Y (2016) Cloud computing—current status and future directions. J Inf Process
24(2):183–194
6. Welcome to Apache™ Hadoop®! (2014) [Online]. Available [Link]
Accessed 13 Mar 2017
7. M. Technologies, “Featured customers” (2016) [Online]. Available [Link]
Accessed 13 Mar 2017
8. Apache Hadoop 2.7.2—Apache Hadoop YARN (2016) [Online]. Available [Link]
[Link]/docs/r2.7.2/hadoopyarn/hadoop-yarn-site/[Link]. Accessed 13 Mar 2017
9. Apache Hadoop 2.7.2—MapReduce Tutorial (2016) [Online]. Available [Link]
org/docs/stable/hadoopmapreduce-client/hadoop-mapreduce-clientcore/MapReduceTutorial.
html. Accessed 13 Mar 2017
10. Apache Hadoop 2.7.2—HDFS users guide (2016) [Online]. Available [Link]
org/docs/stable/hadoopprojectdist/hadoophdfs/[Link]. Accessed 13 Mar 2017
11. Abbas A, Siddiqui IF, Lee SUJ (2016) Multi-objective optimization of feature model in software
product line: perspectives and challenges. Indian J Sci Technol 9(45)
Locality—Aware Scheduling for Containers in Cloud Computing 185

12. Abbas A, Siddiqui IF, Lee SUJ (2017) Contextual variability management of IoT application
with xml-based feature modelling. J Theor Appl Inf Technol 95(6)
13. Rodríguez-Quintana C, Díaz AF, Ortega J, Palacios RH, Ortiz A (2016) A new scalable
approach for distributed metadata in HPC. In Algorithms and architectures for parallel
processing. Springer Nature, pp 106–117
14. White T (2012) Hadoop: the definitive guide. O’Reilly Media, Inc
TorBot: Open Source Intelligence Tool
for Dark Web

P. S. Narayanan, R. Ani and Akeem T. L. King

Abstract The dark web has turned into a dominant source of illegal activities. With
several volunteered networks, it is becoming more difficult to track down these ser-
vices. Open source intelligence (OSINT) is a technique used to gather intelligence on
targets by harvesting publicly available data. Performing OSINT on the Tor network
makes it a challenge for both researchers and developers because of the complexity
and anonymity of the network. This paper presents a tool which shows OSINT in the
dark web. With the use of this tool, researchers and Law Enforcement Agencies can
automate their task of crawling and identifying different services in the Tor network.
This tool has several features which can help extract different intelligence.

Keywords Dark web · Osint · Security · Tor

1 Introduction

We know that the Internet is an ocean of data that is scattered across the Internet.
Open source intelligence (OSINT) is a technique used to analyze this scattered data
hastily by identifying meaningful relationships among data points to get meaningful
information [1].
Data becomes valuable when it disseminates information, and unfortunately, all
data is not self-descriptive and requires context to reveal information. The Internet
has been a major source of this type of data in large quantities, and we can generate
valuable information if we can provide context to it. OSINT refers to the analysis

P. S. Narayanan (B) · R. Ani

Department of Computer Science and Applications, Amrita Vishwa Vidyapeetham, Amritapuri,
Kollam, India
e-mail: thepsnarayanan@[Link]
R. Ani
e-mail: anir@[Link]
A. T. L. King
ISPA Technology, Tampa, FL, USA
e-mail: akeemtlking@[Link]

© Springer Nature Singapore Pte Ltd. 2020 187

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
188 P. S. Narayanan et al.

of unclassified information that is spread across the Internet [2]. OSINT plays an
important role in criminal investigations. There are many OSINT tools available to
gather intelligence, for example,
• Maltego, an inbuilt tool in Kali Linux, helps to perform a significant reconnaissance
against targets.
• Recon-Ng, which is also included in the Kali Linux distribution, has various inbuilt
modules.
• Shodan, a search engine for hackers, gives a huge footprint of IoT devices which
are connected to the Internet.
• Reaper [3], a credential and threat intelligence automation tool.
A smaller portion of the Internet exists which is surreptitious from the normal web
traffic. This part can be accessed only using special software like Tor, I2P [4], etc. This
hidden part is known as the dark web. It comprises numerous unindexed web pages.
This is so because the traditional web crawlers and spiders are not able to access this
part of the Internet [5, 6]. The dark web has been entertaining criminals for performing
many illicit activities and deploying illegal services. Identifying such actions is not a
simple procedure, especially doing so with the dark web. Information gathering is one
way to identify the relations among dark web contents. Law Enforcement Agencies
and researchers rely on the manual investigation, which is time-consuming and thus
inefficient [7, 8].
Similarly, there exists a larger portion of the Internet known as the deep web.
Deep web may be defined as the part of the network in which the contents are not
indexed by standard search engines. It is a reference to any web page that cannot be
accessed using a conventional search engine, such data remains obscured from the
users.
The Tor network ensures a better privacy and security for Internet users by creating
a series of virtual tunnels for communication rather than making a direct connection.
Tor is a tool which secures the users from surveillance otherwise identified as traffic
analysis by distributing the requests and responses over several places on the Internet.
Tor allows the data to be incorrigible and does not lose its integrity. Tor is mainly
used by individuals, journalists, whistleblowers, etc. [9, 10].
TorBot is an open source intelligence tool developed in Python which primar-
ily focuses for the dark web content. It simplifies the process of identification and
analysis of onion services and gathers intelligence about dark web service. Manual
intelligence collection and classification in the dark web is not efficient. To speed
up the process and to help researchers, we developed TorBot. Extensive use of dark
web for communication of terrorism-related information makes it a challenge for
Law Enforcement Agencies. TorBot should be able to fetch data and later using the
data on different machine learning algorithms, and it should be able to identify such
illegal activities that are happening in this encrypted network. Therefore, this tool
will be able to ease the task of finding such activities by an intelligence group or
researchers, thus making this the main objective of TorBot.
TorBot: Open Source Intelligence Tool for Dark Web 189

2 Literature Review

2.1 Challenges

Brian Nafziger has presented a study in which he discussed the challenges when
collecting open source intelligence data from the Darknet. This includes using a spe-
cialized set, refining, and evaluation tools. He proposed an automation tool set that
helps us to scan across the Darknet connecting, gathering, refining, and analyzing
the data. The tool is created on behalf of anonymity system, the collection of intel-
ligence using a web crawler system, the refining using a big data system, and the
evaluation by implementing an NLP technique and a relational linking system [11].

2.2 Accessing Dark Pages

Another research by Ahmed T. Zulkarnine et al. proposed an enhanced Dark Crawler.

It was able to access the Tor network while accessing the surface Internet. It searches
services within the Tor network based on a set of pre-defined keywords, then stores
this unprocessed data in a database. Some challenges they have faced are a high
volume of data, low-quality content, and unintentional DOS attack when crawling.
They have also discussed and implemented solutions for these problems. For this,
they have used a dataset containing over 10,000 distinct Tor domains and 50,000 Tor
web pages. After that, they have constructed a directed graph using Tor web pages
it got where each web page is treated as a vertex and the hyper-links as edges [8].

2.3 Analysis of Tor Hidden Services

In a study conducted by Iskander Sanchez-Rola et al., the design and privacy analysis
of Tor hidden service is measured using a dedicated analysis platform which they
applied it to crawl and analyze millions of Tor URLs. According to their study,
Tor services are organized in an infrequent but highly connected graph. They have
also discussed about the connection that exists between Tor services and the surface
web. Their design of dark web crawler was implemented on top of the headless
browser PhantomJS. For privacy, they have implemented a number of advanced
hiding techniques. Using their tool, they randomly extracted a sample of the initial
seed domains and evaluated the page loading time [12].
190 P. S. Narayanan et al.

2.4 Open Source Intelligence

The research conducted by Blake Butler et al. proposed a tool named REAPER for
automated threat intelligence and OSINT. Their primary aim was to find the distri-
bution and source of where a credential dump first appeared while also maintaining
an in-depth study into the intelligence data that can be achieved by examining the
criminal activities associated with it. They measured the effectiveness of the tool
using 30 unique credential dumps which were found from different sources across
the surface web and dark web. Then, the focus was given primarily to the identifica-
tion of an earlier release, identification of ‘Dark’ domains associated with the dump,
identification of additional dumps, and finally, the number of credentials identified
per-domain [3].

3 Design and Methodology

The TorBot includes several modules, and the core functionality is its crawler and
the visualizer module. This section provides a closer look into our tool design and
features.

3.1 Crawler

TorBot’s crawler is designed such that it crawls the dark pages fast and recursively
through a breadth-first exploration. It further identifies other links, e-mails along with
metadata, and text. It utilizes the multi-threading feature for improved performance.
It has a sub-module called randomizer which randomizes the header information and
IP address after every ‘n’ requests. This helps in maintaining anonymity. Moreover,
this helps prevent IP blocking from certain Web sites which are protected by web
application firewalls (Fig. 1).
The crawler by default checks for ‘onion’ domains only. But, this can be overridden
by specifying the additional domains using TorBot’s ‘-e’ flag. Before adding a link
to the queue, it checks whether the site is live or not. This is simply by sending a
request to the site address and analyzing the response.
TorBot: Open Source Intelligence Tool for Dark Web 191

URLs = input(url)
while (URLs is not empty) do
dequeue url
request page
parse for Links
for (link in Links) do
if link is live && link is not visited then
add link to URLs
store page content

Fig. 1 Pseudo code for crawling the page

3.2 Intelligence Extractor

A typical OSINT tool comprises several modules which makes it a perfect tool. One
such module is the intelligence collection module. Just as it sounds, it collects Intel
from the parsed web page. This Intel can be later used for classification or to find
some key information. The TorBot’s Intel module collects as much information as
possible. It collects scripts, robots, files, e-mails, fuzz URLs and checks for basic
web vulnerabilities [13–15]. The output is stored for future reference.
• [Link]: This function checks for [Link] file. This file is created by web-
masters to instruct web robots how to crawl their Web site. It is included in the
robots exclusion protocol (REP). The basic structure of [Link] is:
User-agent: [user-agent name]
Disallow: [URL string not to be crawled]

This function checks for URLs in the [Link] file, and then, the new URLs (if
present) are added to the queue.
• E-mails: The Intel module searches for e-mails in the page using regex. This e-mail
can be used as Intel information. E-mails often help to map the link with surface
web. Therefore, e-mail plays a vital role in open source intelligence.
• Files: Files present in web pages are fetched using this feature. This includes
images, PDFs, etc.
• Bitcoin Hashes: Bitcoin is unregulated peer-to-peer virtual currency [16]. Bitcoin
hashes are more widely present in the dark web than in the surface web.
192 P. S. Narayanan et al.

Fig. 2 Architecture diagram of TorBot

3.3 Visualizer

The data harvested by TorBot is presented visually using this module. It has the
ability to examine the relationship of links using a tree with the given link as the root
node. Currently, it is capable of checking the length of the tree if a link exists within
the tree. The visualizer module creates an interactive tree graph and displays it to the
user (Fig. 2).
The generated tree can be saved as PNG, SVG, or PDF.

4 Results and Discussion

The result presented here is the final output of extracting information from a given
random Tor service. After successful execution, we can see that it extracted an e-mail
address associated with the page along with URLs and a Bitcoin address as shown
in Fig. 3. The e-mail address extracted is an address associated with a surface web.
Therefore, this hidden service can be mapped with the corresponding surface web.
The visualizer was able to produce a tree graph of the links after further crawling the
dark page as shown in Fig. 4. The crawler was able to find over 200 links which were
directly or indirectly related to the given Tor service. After analyzing, the Bitcoin
hash extracted from the service, and we found a Silkroad transaction, which was one
of the best-known drug markets in the dark web.
TorBot: Open Source Intelligence Tool for Dark Web 193

Fig. 3 TorBot intelligence report

5 Conclusion and Future Work

In this paper, we introduced a tool that can be used for crawling and extracting deep
web contents. Furthermore, it provides a visual module which represents the data in
tree form. GUI for TorBot is currently in development, and users can expect a simple
beginner-friendly GUI for easy interaction.
Currently, TorBot does not have a machine learning model to classify the services
and images. A good ML model could help in classification of services as well as
images [17–21]. Other specific directions for future work are to integrate social
media with TorBot which could help generate more information and to create a data
collection feature for creation of datasets for classification and analysis.
194 P. S. Narayanan et al.

Fig. 4 Visualizer tree graph

References

1. Glassman M, Kang MJ (2012) Intelligence in the internet age: the emergence and evolution of
open source intelligence (OSINT). Comput Human Behav 28(2):673–682
2. Bradbury D (2011) In plain view: open source intelligence. Comput Fraud Secur 2011(4):5–9
3. Butler B, Wardman B, Pratt N (2016) Reaper: an automated, scalable solution for mass
credential harvesting and OSINT. APWG Symp Electron Crime Res 1–10
4. Zantout B, Haraty RA (2014) I2P Data communication system I2P data communication system.
April 2002
5. Qin J, Zhou Y, Lai G, Reid E, Sageman M, Chen H (2005) The dark web portal project:
collecting and analyzing the presence of terrorist groups on the web. In: Proceedings of the
2005 IEEE international conference on intelligence and security informatics, pp 623–624
6. Moore D, Rid T (2016) Cryptopolitik and the Darknet. Survival 6(3):38
7. Weimann G (2016) Going dark: terrorism on the dark web. Stud Confl Terror 39(3):195–206
8. Zulkarnine AT, Frank R, Monk B, Mitchell J, Davies G (2016) Surfacing collaborated networks
in dark web to find illicit and criminal content. In: IEEE conference on intelligence and security
informatics (ISI), pp 109–114
9. Minárik T, Osula A-M (2016) Tor does not stink: use and abuse of the Tor anonymity network
from the perspective of law. Comput Law Secur Rev 32(1):111–127
10. Loesing K, Murdoch SJ, Dingledine R (2010) A case study on measuring statistical data in
the {T}or anonymity network. In: Proceedings of the workshop on ethics in computer security
research (WECSR)
11. Nafziger B (2017) Data mining in the dark: Darknet intelligence automation
12. Sanchez-Rola I, Balzarotti D, Santos I (2017) The onions have eyes: a comprehensive structure
and privacy analysis of tor hidden services. In: Proceedings of the 26th international conference
on world wide web, pp 1251–1260
13. Mouli VR, Jevitha KP (2016) Web services attacks and security-a systematic literature review.
Proced Comput Sci 1(93):870–877
TorBot: Open Source Intelligence Tool for Dark Web 195

14. Cova M, Felmetsger V, Vigna G (2007) Vulnerability analysis of web-based applications. In:
Test and analysis of web services, Springer, Berlin, Heidelberg, pp 363–394
15. Holland BR (2012) Enabling open source intelligence (OSINT) in private social networks
16. Nakamoto S (2009) Bitcoin: a peer-to-peer electronic cash system. Cryptogr. Mail. List https://
[Link]
17. Wesam M, Nabki A, Fidalgo E, Alegre E, De Paz I (2017) Classifying illegal activities on Tor
network based on web textual contents, vol 1, pp 35–43
18. Sathyadevan S, Gangadharan S (2014) Crime analysis and prediction using data mining. In:
2014 first international conference on networks and soft computing (ICNSC), August 19, IEEE,
pp 406–412
19. Chau M, Chen H (2008) A machine learning approach to web page filtering using content and
structure analysis. Decis supp syst 44(2):482–494
20. Ani R, Jose J, Wilson M, Deepa OS (2018) Modified rotation forest ensemble classifier for med-
ical diagnosis in decision support systems. In: Progress in advanced computing and intelligent
engineering, Springer, Singapore, pp 137–146
21. Ani R, Augustine A, Akhil NC, Deepa OS (2016) Random forest ensemble classifier to predict
the coronary heart disease using risk factors. In: Proceedings of the international conference
on soft computing systems, Springer, New Delhi, pp 701–710
A Novel Approach to View and Modify
Data in Cloud Environment Using
Attribute-Based Encryption

Swaminathan Subbiah, S. Palaniappan, Sigamani Ashokkumar

and Ananthakrishnan BalaSundaram

Abstract The Big data and cloud integration is a challenging Task. To enhance the
data security issues, ABE can be deployed. In proposed model, a improved concept
has been implemented and the integration of cloud and Big data is achieved. Security
is the major threat for cloud computing applications. Every user has to feed user
name, password, and primary key for Data access into the cloud data center. Data
owner generates a new key to the users for accessing the data. Policy updating is also
implemented in the proposed system, that is the accountability for the data access
has also been implemented. In case of the change of policy, the altered data stored
in the cloud is not affected. In addition to that, admin generates policy key based on
the user’s profile. If any user tries to misbehave, an immediate alert is sent to the data
owner. Data owner can change the policy key and access policy in the run time. Our
system should be able to update its policy automatically.

Keywords Big data · Cloud computing · ABE

1 Introduction

Big data is described using 3 V’s, they are volume, velocity, variety and this was
defined in the year of 2011. In the year of 2013, in addition to that, another V’s are
introduced, namely veracity, variability, visualization, and value. Nowadays, the Big
data is getting humongous in size. It also provides a way to process a large quantity

S. Subbiah (B) · S. Palaniappan · S. Ashokkumar · A. BalaSundaram

Department of Computer Science and Engineering, Saveetha School of Engineering, SIMATS,
Chennai, India
e-mail: subbussp2007@[Link]
S. Palaniappan
e-mail: [Link]@[Link]
S. Ashokkumar
e-mail: sashokkumarkavi@[Link]
A. BalaSundaram
e-mail: balawins2day@[Link]
© Springer Nature Singapore Pte Ltd. 2020 197
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
198 S. Subbiah et al.

of data that are stored in the cloud server and also uses analytics. Hadoop is one of
most important frameworks to process the big data. It is used to save a file and also
process the file. To save the file, it uses hadoop distributed file system and to process
or to do analytics, it uses map reduce algorithm. Hadoop helps to handle modification
process. In this concept, Big data helps to modify the content dynamically without
waiting for process to complete. It also has grains of what are all modifications done.
It also used for processing the variety of gathered information in high velocity. The
speed of processing this kind of data is faster than comparing with other previous
database management systems. Big data also recovers and discovers the select data
that user wants to do. It helps to improve the decision making and error detection
dynamically. Because of this high velocity data processing in big data is introduced
because previous database management tools are not suitable for process that kind
of large data. For this kind of processing, big data uses cloud to store and process
these large data sets. This can be accessed using efficient way by user. So that user
can process this data at anytime and anywhere. After storing this processed data. In
cloud, security is major threat . Sometimes, data owners not trusted the cloud parties.
Attribute-based encryption (ABE) is one of the most important techniques used in
the cloud. It helps to provide end-to-end data security when the data stored in the
cloud. It allows user to have access polices, using that user or data owner can encrypt
and decrypt the data according to their access policies. When multiple organizations
and enterprises stores data into the cloud, then changing the policy becomes a major
issue as data access policies may be changed dynamically and frequently by users
or data owners. If the data is transferred back to the local site from the cloud, again
the encrypted data contains new access policy after updating that policy it should be
moved back to the cloud server.

2 Issues and Limitations

In the existing system contains policy update issue, but we can neglect this using
attribute-based encryption method [1]. Because, once data owner farm out the data
into cloud server, they won’t store in client system or their own system. To modify
the policy of cipher text in the data storage unit, it helps to get the data and reencrypt
under the new outcome policy, and then send it back to the cloud server. But it contains
some difficulties like high communication overhead and heavy computation on data
owners.
In this system also takes more time for updating, using ABE we can easily update
the policy. That is when we go for any integration in cloud and big data it is somehow
a challenging task. Because during that time, we must shut the cloud server access
it also takes more time.
For example: If we enter into any college portal, it provides different data for
different users, if you are a student most of time, you do not have to update in your
portal. In case of Professor or Head of department, sometimes, they need to update
some data (like attendance, exam marks, exam timetable, etc….) during that time
A Novel Approach to View and Modify Data in Cloud Environment … 199

they should have their own user ID, password and secret key using this they can
login it also provides security. In case of any content updating by data owner in
cloud server they cannot allow to modify or view the data. To neglect that we created
view and modify policy, During the normal time, if user login through their account,
it shows two buttons one for view only and another one for view and modify, If any
of updating take place by data owner then data which are all present in the cloud
server before the update is transferred to secondary server.
So in that time users only allowed viewing that content they cannot allow them
to do update.

3 Modified Approach

It is already known that, admin generates policy key based on the User’s Profile.
Some would have view policy and others would have modify policy. If any user tries
to misbehave, automatically their secret key will be changed by the data owner. Data
owner can change the policy key and access policy dynamically. So the misbehaved
user does not have time to do move further. This policy is provided according to their
role in the organization and their role decides to limit the authentication, i.e. if the
person is working as a Professor or Head of the department in a desired institution,
their limit only resides in their department, they will not get access to view or modify
beyond their limits. If he or she tries to access their policy key, it will be immediately
changed. If the person is a student, he gets view policy and he will not be allowed
to mody the content. To find their role in the organization, we are using big data
which used to analyze the streaming of data. In this paper, we have introduced policy
method but usually we have two methods, one is view policy and another one is
modify policy. Using this modify policy, we can modify the data in cloud; the view
policy is used for read only access. In case of content change or implementation, the
modify policy is unbuttoned. So, we cannot allow doing any changes at a time. We
can only allow viewing and once the implementation is done, it allows users to work
normally.

4 Data Flow Explanation

Data owner uploaded the data, policy settings, and policy key for user based on their
user profile. This user profile explains the role and responsibility of each user. The
secret policy is also generated by the data owner, it helps to find out the misbehaved
user. Once it is done, the data is stored in the cloud server. When the user login
through the search engine, the cloud displays login form and it contains the identity
field; after submitting the form with the relevant information, the browser checks and
encrypt the data and send to the relevant cloud server. If the entered are authenticated,
then it allows the user to access the data according to their access policy. After that,
200 S. Subbiah et al.

it asks user to type their policy key, then this policy is forwarded to the data owner.
Then, data owner sends access key to the user mobile or any other devices. Then, it
asks the user to enter the access key then it allows the user to modify the contents or
data that resides within their access limit. Data manipulation is done.

5 Architecture Diagram

User
Data owner
Misbehavior
Cloud
Big data
(1) User
A person who utilizes resources (like computer, software product, websites, etc…)
is called user. User often has an account, it contains identity like username, login ID,
password, etc. It helps to know authorized user and accessing the data. User often
called as end user, they also defined as operators. This user identity also helps to
create unique ID for each user. They login their account by submitting the identity,
sometimes password, fields are defined as case sensitive, it does not allow special
characters. In this paper, user is people, who viewing and modifying the data in the
cloud server.
(2) Data owner
A person who can authorize or deny the access to certain data, he is also responsible
for accuracy, security, integrity, and authentication. It is act of having legal rights
and overall control over a single or set of data elements. He has the ability to create
a new data field for user, it helps to edit, modify, and share over the networks. He
A Novel Approach to View and Modify Data in Cloud Environment … 201

can also be allowed to restrict access for particular elements and also distributed the
policy. Sometimes, they uses legal actions to claim the copyrights, because if there
any illegitimately breached by unknown users. In this paper, data owner stores data
in cloud server.
(3) Misbehavior
Here, the details of the misbehaved users will be stored for the reference of the data
owner. Then only the system will identify the users when they request for the key
generation, the policy will be taken whether they can be allowed to allocate the policy
to modify/view the data. Due to this, the data stored in the cloud server will be kept
in the safe environment as well as the unwanted modifications of the data can be
avoided.
(4) Cloud server
Cloud is a centralized storage. It is also known as on-demand computing and ubiqui-
tous computing. The information, data, and resources are shared with other computer
and other devices (e.g. network, servers, storage, applications, and services) if they
want. Cloud is used for storage purpose and sharing the resources. It also helps the
large companies to avoid upfront infrastructure costs and focus on projects. Nowa-
days, cloud computing has become a highly demanded service, because it provides
high-speed computation, less cost and easy to maintain characteristics. Security is
the major threat in the cloud computing. It provides three categories of service such
as Platform as a Service (PaaS), Infrastructure as a Service (IaaS), and Software as
a Service (SaaS). It helps the user to work on any software product at any platform
without any cost. In this, the cloud server is used for data storage which is created
by the data owner and provides multiuser access through the network.
(5) Big data
Big data is the term used to define large amount of both structured (traditional data)
and unstructured data (like images, text, audio, video, etc…). It reduces the challenges
like analysis, searching, sharing, storage, querying, etc…). Big data overcomes the
challenges that were faced in the traditional method. Because in the traditional data
processing method, we use database management system (DBMS) and (relational
database management system (RDBMS), which processes only the business data. The
business data is arranged in the form of table (intersection of rows and columns). Each
row holds record, each column is called as tuple. In that, storing the large amount
of unstructured data is a major problem because of that manipulation of data. To
overcome that, we have introduced big data and its capacity to store terabytes. Hadoop
is the open-source framework used to develop distributed storage and distributed data
processing. In this, hadoop is used to perform analytics in user profile to know their
access policy limitations and modify the new contents.
202 S. Subbiah et al.

6 Modules

A. User registration
B. Big data deployment
C. Dynamic policy generation misbehavior detection
D. Cloud deployment
E. Policy key generation

A. User Registration

Once the user creates an account, they are allowed to login into their account to
access the applications. Based on the user’s request, the server will respond to the
user. All the user details will be stored in the database of the server. Every time user
tries to login, this checks the authentication of the username and password.
Special characters and ASCII value are not allowed in the password; therefore,
password becomes case sensitive in nature. This password is used for security purpose
and also used to check whether the authorized user is accessing the data.
B. Big data deployment
Big data is described using 3 V’s they are volume, velocity, variety, this was given in
the year of 2011. In the year of 2013, in addition to that, another V’s are introduced
veracity, variability, visualization, and value. In day by day, the quantity of big data
emerges in the form of increasing orders. It also provides a way to process a large
quantity of data that are stored in the cloud server and also uses analytics. Hadoop
is one of most important frameworks to process a big data. It is used to save a
file and also process the file, inorder to save the file, it uses hadoop distributed file
system and to process or to do analytics, it uses map reduce algorithm. Hadoop helps
to handle the modification process. In this concept, Big data helps to modify the
content dynamically without waiting for the process to complete.
C. Dynamic policy generation misbehavior detection
In this module, we create a dynamic policy generation, i.e. data owner will give the
file access permission to some data user to view and some user to edit, if they edit
without the permission or misbehave, then the policy can be changed dynamically.
After that, if the user enters previous key, he or she will not be allowed to access
data.
D. Cloud deployment
Cloud is a centralized mobile data access center. We can use any device to access
the cloud data. It also provides platform as a Service (PaaS), Software as a Service
(Saas) and Infrastructure as a Service (Iaas ) platforms. Multi and different devices
connected using cloud; this provides unique and different information for a particular
entity. Cloud also provides data storage. Hence, the data can be accessed at anywhere
and at anytime.
A Novel Approach to View and Modify Data in Cloud Environment … 203

E. Policy key generation

In cloud, policy key generation is the most important paradigm and it is an efficient
way to secure the data from the external and internal attacks. It is also used to tackle
that kind of attacks. The cloud server will set the policy key for each and every user
based on their designation. So that, legitimate users can view the data stored in the
cloud only up to their privilege level. They are not allowed to view the data beyond
their privileges. This policy key is also used for developing an unique identity. The
format of the policy key differs according to their departments. The first two digits are
used to identify their department and the next one digit is used for their designation.
It improves the data security for both data owner and users.

7 ABE Implementation

ABE is defined has attribute-based encryption, it is one of the public key encryptions
in which the secret key depends upon the attributes or characteristics (like person
name, age, account number, etc…). It can be also used for log encryption. Instead of
encrypting each part of a log with the keys of all recipients, it is possible to encrypt the
log only with attributes that match recipient’s attributes. For the decryption process,
the same key can be used. This primitive can also be used for broadcast encryption
in order to decrease the number of keys used.

8 Conclusion and Future Work

In this paper, cloud and big data integration challenges are explained. Analysis has
been done on the policy updating problem by using the big data control systems and
derived some challenging tools that are needed to solve this problem. We have also
created an useful model to update the date in the server, which can satisfy all the
needs in the big data process. We have also explained a detailed attribute-based access
control for big data in the cloud server, and designed policy updating algorithms for
different types of access policies. Furthermore, we can use this method to check the
data integrity. Data security in the cloud environment has been much improved by
the proposed model and also monitors and records the misbehaved users. Hence, the
system can improve its decision making when it goes for the new key assignment
to the users by this the system and also avoids the unauthorized user access into the
system.
204 S. Subbiah et al.

References

1. Goyal V, Pandey O, Sahai A (2006) Attribute-based encryption for fine-grained access control
of encrypted data. In: CCS’06. ACM, pp 89–98
2. Bethencourt J, Sahai A, Waters B (2007) Ciphertext-policy attribute based encryption. In:
S&P’07. IEEE, pp 321–334
3. Waters B (2011) Ciphertext-policy attribute-based encryption: an expressive, efficient, and
provably secure realization. In: PKC’11. Springer, pp 53–70
4. Lewko AB, Okamoto T, Sahai A (2010) Fully secure functional encryption: attribute-based
encryption and (hierarchical) inner product encryption. In: EUROCRYPT’10. Springer, pp
62–91
5. Lewko AB, Waters B (2011) Decentralizing attribute-based encryption. In: EUROCRYPT’11.
Springer, pp 568–588
6. Yu S, Wang C, Ren K et al (2010) Achieving secure, scalable, and fine-grained data access
control in cloud computing. In: INFOCOM’10. IEEE, pp 534–542
7. Yang K, Jia X, Ren K (2013) Attribute-based fine-grained access control with efficient
revocation in cloud storage systems. In: AsiaCCS’13. ACM, pp 523–528
8. Yang K, Jia X, Ren K, Zhang B, Xie R (2013) DAC-MACS: Effective data access control for
multiauthority cloud storage systems. IEEE Trans Info Forensics Secur 8(11):1790–1801
9. Yang K, Jia X (2014) Expressive, efficient, and revocable data access control for multi-authority
cloud storage. IEEE Trans Parallel Distrib Syst 25(7):1735–1744
10. Sahai A, Seyalioglu H, Waters B (2012) Dynamic credentials and ciphertext delegation for
attribute-based encryption. In: CRYPTO’12. Springer, pp 199–217
11. Subbiah S, Selvamuthukumaran S, Ramkumar T (2015) An approach for enhancing secure
cloud storage using vertical partitioning algorithm. Middle-East J Sci Res 23(2):223–230.
ISSN:1990-9233
12. Subbiah S, Selvamuthukumaran S (2015) Distributed data security for data prevention in cloud
computing using one time password for user authentication. J Environ Sci Comput Sci Eng
Technol 4(4):752–758. E-ISSN: 2278-179X
13. Roshan, MR, Kumar, MS, Magesh S, Palaniappan S (2017) Forwarding secured data in cloud
interface systems. Int J Pure Appl Math 116(23):179–183
Root Cause Detection of Oscillation
in Shell and Tube Heat Exchanger
Process

S. Abirami and S. Sivagamasundari

Abstract The key emphasis on Control Loop Performance Monitoring (CLPM)

includes the detection of oscillations in control systems. Oscillations are the results
of plant performance degradation and are a very common problem that occurs in the
control loops of the process. This paper discusses the technique for detecting oscil-
lations in process variables. The occurrence of oscillation in control loops, results in
deviation from the setpoint, hence reducing the productivity and thus the profitability.
Oscillations in control loops may be due to several causes such as aggressive tuning of
the controller, external disturbances and sometimes may be due to stiction in control
valve. Low maintenance of valves frequently produces large oscillations in a process
which in turn affects the throughput. In time domain, detection of oscillations is tough
when the signal includes disturbances. In frequency domain, Bispectrum analysis is
a great tool for the detection and analysis of oscillations. To detect the oscillatory
behaviors, the Modified Bispectrum tool was applied to a highly nonlinear Shell and
Tube Heat Exchanger (STHX) process.

Keywords CLPM · Oscillations · Plant performance degradation · Control loops ·

Control valves · Modified bispectrum · STHX

1 Introduction

Root-cause detection of oscillations in process control loop is still a problem at

the current juncture; with which one detects the malfunctions of a loop in many
conditions [1]. Poor control loop performance is commonly the result of unnoticed
deterioration arising in control valves and external disturbances. The field of data-
driven approaches for Control loop Performance Monitoring (CPM) is significantly

S. Abirami (B) · S. Sivagamasundari

Department of Electronics and Instrumentation Engineering, Annamalai University,
Chidambaram 608002, India
e-mail: abiramiselvaraju@[Link]
S. Sivagamasundari
e-mail: sivagamasundari67@[Link]

© Springer Nature Singapore Pte Ltd. 2020 205

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
206 S. Abirami and S. Sivagamasundari

improved by having comparable and standardized sets of data which is used for
testing. They [2] anticipated that companies involved in production gradually and
steadily use CPM tools to tackle loop performance problems. Stiction is the frequently
found control valve problem in process industries. Several attempts have been made
to understand, model and then detect stiction in control valves. The data-driven
method of valve stiction [3, 4] is used with regular plant model to get the required
oscillating data. HOSA with closed-loop data detects the root cause for poor control
loop performance utilizing its tools namely cumulants or moments, bispectrum and
bicoherence to develop the non-gaussianity and the nonlinearity indices so as to
detect and quantify the source of nonlinearity [5]. The techniques of HOSA have
been extensively used in various fields. One such major contribution includes the
bio-medical area [6].
The works done in this paper has been originated from the ideas gained by sur-
veying the literature, specifically from the various chapters of [7–10]. This paper
emphases on the detection of oscillations present in the control loops of STHX pro-
cess using modified bispectrum analysis.

2 Mathematical Modeling

2.1 Energy Balance Equations

The Energy Balance Equations derived from “Rate of energy stored in the control
volume is equal to the rate of gain of energy from neighboring control volume”, for
shell-side and tube-side [11], respectively, are given below.

ρs Cs Vs dTco h s As
Shell Side: ∗ = ṁ s Cs (Tci − Tco ) + (Tho − Tco ) (1)
N dt N
ρt Ct Vt dTho h t At
Tube Side: ∗ = ṁ t Ct (Thi − Tho ) + (Tco − Tho ) (2)
N dt N

2.2 Process Parameters and PID Controller Parameters

The controller parameters are calculated using the process parameters by Zeigler
Nichols tuning technique [11]. The process and PID controller parameters for various
operating regions are computed and specified in Table 1.
Root Cause Detection of Oscillation in Shell and Tube Heat … 207

Table 1 Process and PID controller parameters

Flow rate Operating Gain Time Time delay Kc Ki Kd
(lps) region (°C) (°C/lps) constant (s) (s)
0.02–0.04 44.96–45.5 −33.3 0.827 0.137 −0.218 −0.796 −0.0149
0.04–0.06 44.65–44.96 −15.5 0.773 0.134 −0.447 −1.668 −0.0299
0.08–0.10 44.42–44.52 −5.5 0.606 0.122 −1.084 −4.443 −0.066
0.12–0.10 44.41–44.34 −3.5 0.335 0.161 −0.7134 −2.216 −0.057
0.08–0.06 44.65–44.5 −7.5 0.776 0.159 −0.781 −2.456 −0.0621
0.06–0.04 44.96–44.5 −14 0.816 0.192 −0.3643 −0.949 −0.03497

3 Valve Stiction Model

3.1 Structure of Pneumatic Control Valve

Figure 1 shows the general structure of a pneumatic control valve. The valve is
opened by air force and closed by elastic pressure. The position of the plug regulates
the balance between elastic pressure and air force thus regulating the flow rate. The
valve stem connected to the plug is moved in contrary to static force caused by gland
packing, a device which is sealed to prevent process fluid leakage.

Fig. 1 Structure of
pneumatic control valve
208 S. Abirami and S. Sivagamasundari

Fig. 2 Closed loop system with valve stiction model

3.2 Closed Loop System with Valve Stiction Model

Figure 2 shows the block diagram of closed-loop system with the control valve
stiction model. The controller chosen here is PID controller and the process taken
is Shell and Tube Heat Exchanger. The single parameter model [12] is utilized for
valve nonlinearity as valve stiction model.

3.3 One Parameter Model

The actual valve position (xt ) differs from the control signal (ut ) if the valve undergoes
stiction, hence resulting in a poor control loop performance. For detection problem,
a simple valve model uses model-based approach [12]. The actual valve position (xt )
is considered to be piecewise constant, as a function of time.

xt−1 , if|u t − xt−1 | ≤ d
xt = (3)
ut , otherwise

4 Oscillation Detection

4.1 Bispectrum Analysis

Scrutinizing the nonlinear signals in Higher Order Statistics encloses the relations
between phase components. The bispectrum B (f 1 , f 2 ) of a non-Gaussian signal, x (t),
is a 2-D Fourier transforms of the third-order cumulants gives the info not presented
by the spectral domain is defined as

C(m, n) = E [x(k) x(k + m) x(k + n)] (4)

where E is the Expectation function. The bispectrum formula related to (4) is:

B( f 1 , f 2 )  E [X ( f 1 )X ( f 2 )X ∗ ( f 1 + f 2 )] (5)
Root Cause Detection of Oscillation in Shell and Tube Heat … 209

where X (f ) is the Fourier transform of x (t) and * represents its complex conjugate.
Bispectrum contains the facts about the relation of phase between the frequency
components at f 1 , f 2, and f 1 + f 2 [11].

4.2 Modified Bispectrum Analysis

In Eq. (5) some modifications are made to bispectrum formula and hence termed
as Modified Bispectrum analysis, which is used to detect the root cause for process
control loop oscillations of STHX process.

B M ( f 1 , f 2 )  E [X ( f 2 + f 1 ) X ( f 2 − f 1 )X ( f 2 )X ∗ ( f 2 )X ∗ ( f 2 )] (6)

The total phase of modified bispectrum is,

∅ M ( f 1 , f 2 ) = ∅( f 2 + f 1 ) + ∅( f 2 − f 1 ) − ∅( f 2 ) − ∅( f 2 ) (7)

As the two components f 1 and f 2 are in coupling, their phases are related as

∅( f 2 + f 1 ) = ∅( f 2 ) + ∅( f 1 )
∅( f 2 − f 1 ) = ∅( f 2 ) − ∅( f 1 ) (8)

By substituting Eq. (7) in (6) for modulated signal bispectrum, the total phase
is zero and modulated signal bispectrum amplitude is the product of the four mag-
nitudes, which contributes to the maximum of the complex product. Therefore, a
bispectral peak appears at (f 1 , f 2 ) [13]. For checking the probable presence of a non-
linear element in control loop, the process variable (PV) data is used by the modified
bispectrum.

5 Results and Discussions

Detection of oscillations in process control loop is essential as it reduces the plant

profitability. Root causes for oscillations are not only due to the occurrence of nonlin-
earities (Dead band, Hysteresis, Stiction, etc.,) but also due to external disturbances.
The stiction in control valve causes the process output to oscillate around the set
point. The manipulated variable considered in the STHX process is cold water flow
rate and the controlled variable is hot water outlet temperature. White noise is intro-
duced in the closed-loop to collect datasets due to external disturbance. For getting
nonlinearity induced oscillatory data, a stiction model is introduced in control loop
with known values of stiction parameters. From these responses simulated sets of data
were collected and used as input vector for modified bispectrum analysis. Modified
210 S. Abirami and S. Sivagamasundari

bispectrum plots of external oscillatory disturbance for various operating regions in

STHX (PV) are displayed from Figs. 3, 4, 5, 6, 7 and 8. Modified bispectrum plots of
stiction (both weak and strong stiction) for various operating regions in STHX (PV)
are presented from Figs. 9, 10, 11, 12, 13 and 14. The estimated modified bispectrum
values for external oscillatory disturbance and for stiction are presented in Tables 2
and 3.
The bispectral amplitude with graphical plots has been used to diagnose the root
causes for the bad performance of control loops. The threshold value is specified
to be 0.1. If the bispectrum value exceeds the threshold value, nonlinearity can be
confirmed. The threshold limit for nonlinearity detection is chosen based on the
experience of using this tool in process performance diagnosis and for this case
clearly detects the stiction nonlinearity present in the process output. Comparing
the plots, highest bispectrum value is observed for the PV data which is affected by
the existence of stiction in control valve, i.e., the higher value indicates significant

Fig. 3 Positive step change—Region 1 (44.96–45.5)

Fig. 4 Positive step change—Region 2 (44.65–44.5)

Fig. 5 Positive step change—Region 3 (44.41–44.34)

Root Cause Detection of Oscillation in Shell and Tube Heat … 211

Fig. 6 Negative step change—Region 1 (44.96–44.5)

Fig. 7 Negative step change—Region 2 (44.65–44.5)

Fig. 8 Negative step change—Region 3 (44.41–44.34)

Fig. 9 Weak and strong stiction for [positive step change—Region 1 (44.96–45.5)]

Fig. 10 Weak and strong stiction [positive step change—Region 2 (44.65–44.5)]

212 S. Abirami and S. Sivagamasundari

Fig. 11 Weak and strong stiction [positive step change—Region 3 (44.41–44.34)]

Fig. 12 Weak and strong stiction [negative step change—Region 1 (44.96–44.5)]

Fig. 13 Weak and strong stiction [negative step change—Region 2 (44.65–44.5)]

Fig. 14 Weak and strong stiction [negative step change—Region 3 (44.41–44.34)]

Table 2 Estimated modified bispectrum values for external disturbance

Disturbance Positive step change Negative step change
Region 1 Region 2 Region 3 Region 1 Region 2 Region 3
(44.96–45.5) (44.65– (44.42– (44.96–44.5) (44.65–44.5) (44.41–
44.96) 44.52) 44.34)
Modified 0.064 0.068 0.08 0.098 0.069 0.07
bispectrum
Root Cause Detection of Oscillation in Shell and Tube Heat … 213

Table 3 Estimated modified

Regions Modified bispectrum values
bispectrum values for stiction
Weak Strong
stiction stiction
Positive step Region 1 0.37 0.99
change (44.96–45.5)
Region 2 0.4 1
(44.65–
44.96)
Region 3 0.35 0.88
(44.42–
44.52)
Negative step Region 1 0.15 0.68
change (44.96–44.5)
Region 2 0.37 0.65
(44.65–44.5)
Region 3 0.12 0.57
(44.41–
44.34)

nonlinearity. The bispectrum values less than 0.1 is due to external disturbance. In
this case, the bispectrum is found to have multiple peaks with lesser amplitude.

5.1 Modified Bispectrum of External Disturbance Induced

Oscillations for Various Operating Regions in STHX (PV)

See Figs. 3, 4, 5, 6, 7 and 8.

5.2 Modified Bispectrum of Stiction (Weak Stiction

and Strong Stiction) Induced Oscillations for Various
Operating Regions in STHX (PV)

See Figs. 9, 10, 11, 12, 13 and 14.

6 Conclusion

Higher Order Statistical technique such as bispectrum is used in this paper to detect
the oscillations present in the process variable for STHX process. The bispectral
values with graphical plots have been used to diagnose the root causes for the bad
214 S. Abirami and S. Sivagamasundari

performance of control loops. The magnitude of bispectrum indicates the presence

or absence of process nonlinearity. The results clearly shows that modified bispectral
analysis gives promising results for oscillation detection. The future scope of this
work involves compensating techniques to overcome stiction nonlinearity present in
control valves.

References

1. Bauer M, Horch A, Xie L, Jelali M, Thornhill N (2016) The current state of control loop
performance monitoring—A survey of application in industry. J Process Control 38:1–10
2. Bauer M, Auret L, le Roux D, Aharonson V (2018, May 9–11) An industrial PID data repository
for Control Loop Performance Monitoring (CPM). In: Preprints of the 3rd IFAC conference
on advances in proportional-integral-derivative control, Ghent, Belgium, pp 823–828
3. Choudhury MAAS, Thornhill NF, Shah SL (2005) Modelling valve stiction. Control Eng Pract
13:641–658
4. Shoukat Choudhury MAA, Jain M, Shah SL (2008) Stiction—definition, modelling, detection
and quantification. J Process Control 18:232–243
5. Shoukat Choudhury MAA, Shah SL, Thornhill NF (2004) Diagnosis of poor control-loop
performance using higher-order statistics. Automatica 40:1719–1728
6. Goshvarpour A, Goshvarpour A (2013) Comparison of higher order spectra in heart rate signals
during two techniques of meditation: Chi and Kundalini meditation. Cogn Neurodyn 7:39–46
7. Shoukat Choudhury MAA, Shah SL, Thornhill NF (2008) Diagnosis of process nonlinearities
and valve stiction-data driven approaches. Advances in industrial control. Springer, Berlin,
Heidelberg
8. Matsuo T, Sasaoka H, Yamashita Y (2003) Detection and diagnosis of oscillations in process
plants. In: Palade V, Howlett RJ, Jain L (eds) Knowledge-based intelligent information and
engineering systems, KES 2003, vol 2773. Lecture notes in computer science. Springer, Berlin,
Heidelberg, pp 1258–1264
9. Karra S, Karim M (2010) Oscillation root-cause detection and quantification under multiple
faults. In: Jelali M, Huang B (eds) Detection and diagnosis of stiction in control loops. Advances
in industrial control. Springer, London, pp 267–293
10. Karra S, Jelali M, Karim M, Horch A (2010) Detection of oscillating control loops. In: Jelali
M, Huang B (eds) Detection and diagnosis of stiction in control loops. Advances in industrial
control. Springer, London, pp 61–100
11. Abirami S, Sivagamasundari S (2018) Comparison of controller tuning techniques for a tem-
perature control of STHX Process. Int J Comput Sci Eng 6(8):316–320
12. Stenman A, Gustafsson F, Forsman K (2003) A segmentation-based method for detection of
stiction in control valves. Int J Adapt Control Sig Process 17:625–634
13. Shaeboub A, Gu F, Lane M, Haba U, Wu Z, Ball AD (2017) Modulation signal bispectrum
analysis of electric signals for the detection and diagnosis of compound faults in induction
motors with sensorless drives. Syst Sci Control Eng 5(1):252–267. [Link]
21642583.2017.1331769
Performance Assessment of Spread
Spectrum Communication Receivers

P. Anand Krisshna, K. Vandith Sreenivas and Gayathri Narayanan

Abstract This paper mainly focuses on implementing a new approach rather than
using the conventional method of generating an intermediate frequency in a mixer.
The ordinary mixer is compared with a switching mixer through several techniques
in order to strengthen the point that switching mixers have a better performance.
Theoretically, we know that switching mixers are more efficient but here we also
give it a practical justification through this paper. Software such as MATLAB and
Proteus has been used for the same. The main objective of this research work is
to determine the maximum amount of noise that can be removed to obtain a good
reconstruction of the input signal.

Keywords AGWN · Monte-Carlo simulation · Modulation schemes · BER · SNR,

DSSS, FHSS

1 Introduction

A mixer is a basic electronic circuit that mixes a radio frequency signal with a
local oscillator signal and produces the sum and difference frequency as the output.
The output port is also called as the Intermediate Frequency port. The mixer is an
important component of the receiver in any given communication system. The Fig. 1
gives a mathematical representation of a mixer circuit.

f IF = f LO ± f RF

They limit the dynamic range of a system through the following specifications
mentioned below
• Noise figure and LO noise
• Power compression

P. A. Krisshna · K. V. Sreenivas (B) · G. Narayanan

Department of Electronics and Communication Engineering, Amrita Vishwa Vidyapeetham,
Amritapuri, India
e-mail: vandithsreenivas@[Link]
© Springer Nature Singapore Pte Ltd. 2020 215
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
216 P. A. Krisshna et al.

Fig. 1 Model of a simple

mixer

• Conversion Efficiency
• Port to port isolation
• Single tone inter-modulation distortion
• Multi-tone inter-modulation distortion.
There are basically two ways in which the process of mixing is done, one is
the non-linear way and the other is by switching depending upon the local oscillator
signal. The non-linear frequency mixer creates new frequencies from the two applied
signals, whereas the switching mixer aims to perform linear operations on the signal
by hard switching that will be driven by the local oscillator.

2 Related Work

Commercially switching mixers are preferred to non-linear ones and the advantage
that it gives is a lower noise figure along with a larger conversion gain, by applying
the same effort, because the switching diodes act like an open switch as well as a
closed switch and in both the case only a minimal noise will be added [1]. Switching
mixers are the mixers that are most used in the field and theoretically it has around
3.9 dB efficiency. They are roughly split up into two, the passive and the active.
The passive ones can further be divided to diode based and FET based. The noise
generated in mixers is caused due to the diodes, transistors, and agitation of electrons
in conductor that cause resistive losses which further becomes thermal noise. Mixers
have a large range of applications in communication systems. The Super-heterodyne
receiver as well as the direct conversion receiver architectures use mixers at the
input to down-convert and demodulate the digital information [2]. Therefore mix-
ers are widely used in analog or RF front end of receivers. They can be used for
demodulation and also as analog multipliers. The application of spread spectrum
techniques is not restricted to mixers. Here, the signal is prone to jamming, which
affects transfer of information, especially in critical cases such as the transmission
of encrypted data. As reference [3] suggests, a Frequency Hopping Spread Spectrum
(FHSS) system is often deployed to protect wireless communication from jamming
or to hinder undesired reception of the signal. [4] suggests a scheme of combining
spread spectrum technology with the MSK-LFM waveforms. The spread spectrum
Performance Assessment of Spread Spectrum Communication Receivers 217

technology is applied to the radar signal, which can remarkably improve the perfor-
mance of the signal ambiguity function. Using adaptive frequency hopping, a system
implemented to enhance immunity toward frequency interference by avoiding usage
of congested frequency channels in hopping sequence [5]. CMOS mixers can also be
used to implement these techniques and they perform well in noisy environments [6].
In Cognitive Radio (CR) networks, less utilized spectra can be detected using spread
spectrum techniques due to their ability to cover wider spectra [7]. Conventionally,
Additive White Gaussian Noise (AWGN) is the noise type that will be encountered
during transmission, so analysis of AWGN helps in simulation of the technique in
MATLAB [8]. BSIM models are used for transistor design (MOSFETS), designed by
UC Berkeley. These transistors hare hence used in receiver circuits [9]. SDR (Soft-
ware Defined Radio) reduces hardware and maintenance costs due to the majority of
its functionality being software. In particular RTL_SDR (Realtek Software Defined
Radio) reduces the cost further thus making it a good option for transmission of
signals over a channel [10].
As mentioned above, CMOS Radio Frequency (RF) circuits are used in transmit-
ters and receivers [11]. A bandpass sampling receiver of a certain design will help in
efficient transmission of signals over multi-channel networks; these radio receivers
can be built using switching mixers for lesser BER [12].

3 Proposed Work

The principal objective of this work is to compare the spread spectrum techniques by
using both ordinary and switching mixers and to verify that the bit error rate (BER)
of the switching mixer is lesser than that of the ordinary mixer, thus implying that
switching mixers are more efficient. The verification is done using MATLAB, where
a random value is chosen for further operations.
For effective communication, the transmitted signal has to be received as it is on
the other side. This is ideal, since over a channel (in most cases air), the signal may
experience the effect of noise, which will hamper its efficiency. The interference of
noise may be unintentional, such as another signal being transmitted over the same
channel, or intentional, such as a signal jammer or an enemy trying to intercept
signals. One type of technique that can avoid this kind of interference is the ‘Spread
Spectrum ‘technique. This technique creates a special code for the transmitted signal
that increases its bandwidth substantially and makes it look like a noise signal in
itself. This code is known only to the transmitter and receiver. The most commonly
used techniques are DSSS (Direct Sequence Spread Spectrum) and FHSS (Frequency
Hopping Spread Spectrum).
218 P. A. Krisshna et al.

Figure A. FHSS

Bandwidth and power are the major shortcomings faced in the study of digital
communication systems. These important parameters need to be improved in order
to achieve effective performance. But, there is a trade-off in this efficiency in order
to provide an important objective in communications i.e., Security. There will be
no meaning for a system, where messages can be detected by unwanted listeners.
The major advantage of Spread Spectrum (SS) is its ability to counter interference
whether unintentional (i.e. another user trying simultaneously to transmit over the
channel) or intentional (i.e. a jammer or interceptor).

Figure B. DSSS

4 Result Analysis

A. Signal Generation Using Different Modulation Schemes

Performance Assessment of Spread Spectrum Communication Receivers 219

MATLAB is one of the most commonly used programming languages in which

different operations are performed on the input data or signals and their functions
can be plotted. In order to analyze the difference between a normal frequency mixer
and a switching mixer, we have used three kinds of modulation schemes (BPSK,
QPSK, and QAM). In BPSK modulation a carrier wave and a binary sequence is
multiplied and it is further given as the input. According to the value of the binary
input the phase of the output changes from 0° to 180°. In QPSK modulation there is a
two-bit input in which the even and the odd bit are separately multiplied by the same
carrier. The output signal will have a 90-degree phase shift. In QAM the number
of input bits can be two or more. The input signal for each of these schemes was
generated respectively. The signal that reaches the mixer usually contains noise as it
passes through the receiver channel. Since the input signal is randomly generated and
the continuous probability distribution of thermal noise is Gaussian, thus the noise
in MATLAB is represented by Additive White Gaussian Noise (AWGN). Noise is
added to the transmitted signal and the power spectral density of the noise is constant
and has a Gaussian distribution of power spectral density. Inside the mixer the input
signal gets multiplied with the LO signal and turns out to give an IF signal as the
output. In a normal mixer the LO signal is usually a sine wave whereas in a switching
mixer it is a square pulse. On comparing the outputs from the two, it is evident that the
switching mixer output contains lesser noise and gives a much better reconstruction
of the input signal as compared to the normal frequency mixer.
Figures 2, 3, 4, 5, 6, 7, 8, and 9 represents the time and frequency domain effects
of various kinds of modulation schemes. As is the principal objective of the paper,
the responses of the modulation exhibited by both the ordinary and switching mixers
are compared.

Fig. 2 Time domain analysis of ordinary mixer a QAM input signal, b input signal with noise,
c LO signal, d mixer output without noise, e mixer output with noise
220 P. A. Krisshna et al.

Fig. 3 Time domain analysis of switching mixer a QAM input signal, b input signal with noise,
c LO signal, d mixer output without noise, e mixer output with noise

Fig. 4 Frequency spectrum of ordinary mixer a BPSK input signal, b input signal with noise, c LO
signal, d mixer output without noise, e mixer output with noise

B. Comparison of Responses
Figures 2 and 3 show the time domain analyses of a QAM modulated signal for
ordinary and switching mixers respectively. The output with noise is the last plot in
each figure, this is the signal that will be acted upon using spread spectrum techniques
to give a noise like appearance. From here on, Figs. 4, 5, 6, 7, 8, and 9 show frequency
spectra. Figures 4 and 5 show BPSK modulated outputs with noise for both mixers.
Figures 6 and 7 show QPSK modulated outputs for ordinary and switching mixers
respectively. It is observed that the switching mixer gives a better amplitude than that
Performance Assessment of Spread Spectrum Communication Receivers 221

Fig. 5 Frequency spectrum of switching mixer a BPSK input signal, b input signal with noise,
c LO signal, d mixer output without noise, e mixer output with noise

Fig. 6 Frequency spectrum of ordinary mixer a QPSK input signal, b input signal with noise, c LO
signal, d mixer output without noise, e mixer output with noise

of the ordinary mixer. Figures 8 and 9 show frequency responses of QAM modulated
signals for both the ordinary and switching mixers, respectively.
The responses show that using a switching mixer will produce better plots
attributed to the switching by the local oscillator (LO).
C. Performance Analysis
In order to compare the performance of the different modulation schemes, the SNR
vs. BER curve is plotted as shown in Figs. 10 and 11. BER gives the number of
222 P. A. Krisshna et al.

Fig. 7 Frequency spectrum of switching mixer a QPSK input signal, b input signal with noise,
c LO signal, d mixer output without noise, e mixer output with noise

Fig. 8 Frequency spectrum of ordinary mixer a QAM input signal, b input signal with noise, c LO
signal, d mixer output without noise, e mixer output with noise

bits that are in error or have variations due to noise as it passes through the AWGN
channel. SNR is a ratio of the signal uncorrupted by noise to the amount of noise
present. An increase in the SNR value indicates that there is more of the signal and
less amount of noise and a decrease in BER. BER varies for the different modulation
schemes. The BER of communication receivers operating in AWGN Monte Carlo
Simulations are used. This is commonly used when there is interference from a large
number of random variables and at the same time model the probability of different
outcomes in a process that cannot be easily determined.
Performance Assessment of Spread Spectrum Communication Receivers 223

Fig. 9 Frequency spectrum of switching mixer a QAM input signal, b input signal with noise,
c LO signal, d mixer output without noise, e mixer output with noise

Fig. 10 Mixer and switching

mixer output for QPSK

D. Application of Spread Spectrum Schemes

Frequency Hopping Spread Spectrum (FHSS) is a method of transmitting radio

signals by rapidly switching a carrier among many frequency channels, using a
pseudo-random sequence known to both transmitter and receiver (as shown in Figure
A). The FHSS output is shown in Fig. 12. The spread signal resembles a noise
signal, and thus ensures uninterrupted transmission. Inclusion of Additive White
Gaussian Noise (AWGN) in the technique creates a noise-like pattern that can trick the
intercepting party into thinking that the transmitted signal is just some random noise
224 P. A. Krisshna et al.

Fig. 11 Mixer and switching mixer output for 16 QAM

Fig. 12 FHSS spread signal

signal. Direct Sequence Spread Spectrum (DSSS) is a spread spectrum technique

whereby the original data signal is multiplied with a pseudo-random noise spreading
code (as shown in Figure B). This spreading code has a higher chip rate (this the bit
rate of the code), which results in a wide band time-continuous scrambled signal.
Figure Fig. 13 shows the reconstructed signal.
Performance Assessment of Spread Spectrum Communication Receivers 225

Fig. 13 DHSS spread and reconstructed signal

5 Conclusion

This work aims at simulating the performance of the conventional and switching
mixer receivers in communication systems which employ spread spectrum modula-
tion techniques. The simulations were carried out in MATLAB and the objective was
verified. It was observed that the BER is higher in the case of the conventional mixer
as compared to switching mixer receivers and hence we see that the performance of
switching mixer is much more efficient. Thus usage of switching mixers can help
achieve better results.

References

1. Marki F, Marki C (2001) T3 mixer primer: a mixer for the 21st century. MarkiMicrowaven,
Morgan Hills, CA, [Link]
2. Amita M, Anjali G, Kurup DG (2015) Combined amplitude and phase noise effects in
QAM direct conversion receivers. In: International conference on microwave, optical and
communication engineering (ICMOCE), IIT-Bhuvaneshwar
3. Ebrahimzadeh A, Falahati A (2013) Frequency hopping spread spectrum security improvement
with encrypted spreading codes in a partial band noise jamming environment. J Inf Secur
4(1):1–6
226 P. A. Krisshna et al.

4. Dou Z, Zhong XK, Zhang WX (2017) Radar-communication integration based on MSK-LFM

spread spectrum signal. Int J Commun Netw Syst Sci 10:108–117. [Link]
ijcns.2017.108B012
5. Motlagh N (2010) Performance improvement of wireless communications using frequency
hopping spread spectrum. Int J Commun Netw Syst Sci 3(10):805–810
6. Terrovitis M, Meyer R (1999, June) Noise in current-commutating CMOS mixers. IEEE J
Solid-State Circ 34(6):772–783
7. Mahamuni S, Mishra V (2014) Performance evaluation of spectrum detection in cognitive radio
network. Int J Commun Netw Syst Sci 7:485–496
8. [Link]
basics-tutorial
9. [Link] of BSIM3 model
10. Sruthi MB, Abirami M, Manikkoth A, Gandhiraj R, Soman KP (2013) Low cost digi-
tal transceiver design for software defined radio using RTL-SDR. In: Proceedings—2013
IEEE international multi conference on automation, computing, control, communication and
compressed sensing, iMac4 s 2013. Kerala, pp 852–855
11. Lee TH (2004) The design of CMOS radio-frequency integrated circuits, 2nd edn. Cambridge
University Press
12. Avantika S, Devika SK, Gomathy V, Manjukrishna SKSA, Kurup DG (2015) Design and
experimental characterization of a bandpass sampling receiver. In: International conference on
communication systems, ICCS-2015, American Institute of Physics (AIP), Pilani, India
Computationally-Light Metrics
to Quantify Link Stability in Mobile
Sensor Networks

Natarajan Meghanathan

Abstract We propose three innovative location and mobility-independent

computationally-light metrics to quantify the stability of links in mobile sensor net-
works (MSNs). The proposed metrics (Normalized Neighbor Degree: NND, One
Hop Two Hop Neighbors: OTH, and Fraction of Shared and Unshared Neighbors:
FSU) are computed on the egocentric network of an edge and the hypothesis is that
larger the extent of shared neighborhood between the end vertices of an edge, larger
the stability (lifetime) of the link in the MSN. The computation times of all the three
metrics are about 15–40 times lower than the computation times of the bipartivity
index (BPI) and algebraic connectivity (ALGC) metrics that were adapted from Net-
work Science in an earlier research to quantify link stability in MSNs. The lifetimes
of the DG trees obtained with the proposed computationally-light link stability met-
rics are appreciably larger or comparable to that of the ALGC and BPI-based DG
trees.

Keywords Link stability · Computationally-light metrics · Computationally-heavy

metrics · Mobile sensor networks · Data gathering

1 Introduction

A mobile sensor network (MSN) is a distributed network of dynamically changing

topology and the communication structures of the sensor nodes (like network-wide
data gathering trees) need to be frequently reconfigured. In an earlier paper [1] as well
as this paper, we address the problem of quantifying the stability of links in a MSN
using the neighborhood information of the nodes. In [1], we adapted two Network
Science metrics (such as Bipartivity Index: BPI [2] and Algebraic Connectivity:
ALGC [3]) and computed them on the egocentric network of an edge [1] to quantify
the stability of the edge (also referred to as ‘link’) in a MSN. The egocentric network
of an edge includes the end vertices and their neighbors as vertices and the edges

N. Meghanathan (B)
Jackson State University, Jackson, MS 39217, USA
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 227

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
228 N. Meghanathan

connecting the end vertices with their neighbors as the edges. The lifetimes of the BPI
(=1 − BPI) based data gathering (DG) trees were observed [1] to be significantly
larger than those of the predicted LET (link expiration time [4, 5], adapted from
mobile ad hoc networks [6])-based DG trees.
Prior to [1], the LET approach [4, 5] was the only well-known approach to deter-
mine stable DG trees in MSNs. However, the LET metric requires the mobility and
location information of the nodes. Other relevant approaches (such as self-healing
[7]) proposed in the literature of wireless networks also cannot be adapted for MSNs
as these would also require the location information of the nodes. On the other hand,
one does not require the mobility and location information of the nodes to compute the
BPI and ALGC metrics. However, a major weakness of these two metrics is that they
are computationally-heavy and could impose a significant burden on the resource-
constrained sensor nodes. Hence, the motivation is to develop computationally-light
metrics (that are also location- and mobility-independent, like BPI and ALGC) to
quantify the stability of links in MSNs. Our hypothesis is that the link whose end
vertices are closer is likely to be more stable (i.e., exist for a longer time) than the link
between two end vertices that are farther away from each other. The above hypothesis
was observed to be true [1] for both BPI and ALGC, and we anticipate it to hold
true for the computationally-light link stability assessment metrics proposed in this
paper.
The rest of the paper is organized as follows: Sect. 2 introduces the three
computationally-light metrics to quantify the stability of a link in a MSN. Section 3
presents simulation results for computation times incurred with the five met-
rics (including BPI and ALGC), explores the rank-based correlation between the
computationally-light and computationally-heavy metrics and compares the DG
tree lifetimes incurred using the link stability scores quantified using these met-
rics. Section 4 summarizes the contributions of this work and concludes the paper.
Throughout the paper, we interchangeably use the terms: ‘link’ and ‘edge’, ‘node’
and ‘vertex’, ‘network’ and ‘graph’. They mean the same.

2 Metrics to Quantify Link Stability

Our hypothesis is that larger the extent of shared neighborhood between the end
vertices of an edge (u, v), the larger the stability of the link (edge). We use the
notion of “egocentric network of an edge” proposed in our earlier work [1] for all
the analysis. From notation point of view, for an edge (u, v): let N(u) represent the
set of neighbors of a vertex u and let EgoN (u, v) represent the set of vertices (vertices
u, v and their neighbors, with each vertex, represented exactly once) that are part of
the egocentric network of the edge.
Computationally-Light Metrics to Quantify Link Stability … 229

2.1 Normalized Neighbor Degree (NND)

For an edge (u, v), its NND (see formulation 1 below) is the ratio of the square
root of the sum of the squares of the degrees of the neighbor nodes and the number
of neighbor nodes in its egocentric edge network. The neighbor nodes of the end
vertices u and v in the egocentric network of edge (u, v) could have a degree of either
1 or 2. If a neighbor node has degree 2, it implies the node is a neighbor of both
u and v and is part of the shared neighborhood. If a neighbor node has degree 1,
then the node is not part of the shared neighborhood. Hence, the larger the number
of neighbor nodes with degree 2, the larger the extent to which the neighborhood is
shared.
 
ki2
i∈Ego N (u,v)−{u,v}
NND(u, v) =   (1)
Ego (u, v) − {u, v}
N

2.2 One Hop Two Hop (OTH) Neighborhood

For an edge (u, v), its OTH is computed on the basis of the number of one hop and
two hop neighbors of the end vertices u and v in its egocentric edge network. From
the point of the end vertices u and v, it is desirable that all the vertices in EgoN (u,
v) be their one hop neighbors. However, if a vertex in EgoN (u, v) is not present
in the neighborhood of either u or v, then it becomes a two hop neighbor of the
corresponding end vertex. The presence of two hop neighbors in an egocentric edge
network decreases the extent with which the neighborhood is shared; on the other
hand, the presence of one hop neighbors increases the extent of shared neighborhood.
We propose an auxiliary metric called weighted neighborhood hop count (WNH)
as a weighted sum of the one hop and two hop neighbors of the end vertices u
and v (see formulation 2). As seen in the WNH formulation for an end vertex, we
complement the end vertex for having one hop neighbors and penalize for having two
hop neighbors (by a factor of 2). In other words, the presence of two hop neighbors
is modeled to weaken the stability of the neighborhood and negate the advantage that
comes with the presence of one hop (shared) neighbors.
We opine the sigmoid function (typically used in the formulation of the cost func-
tion for logistic regression [8] or artificial neural networks [9]) to be an appropriate
function to introduce the needed non-linearity (a stronger decay or a stronger ascent
even for egocentric edge networks with moderate differences in the number of one
hop and two hop neighbors) that is not incorporated in the WNH linear formulation.
The OTH for an edge (u, v) is thus computed using a sigmoid formulation (see for-
mulation 3) involving the WNH scores of u and v. Also, a sigmoid formulation will
transform the input (WNH values) to an output that is confined in the range of (0,
…, 1).
230 N. Meghanathan
 
WNH(u) = |N (u)| − 2 ∗ Ego N (u, v) − N (u)
 
WNH(v) = |N (v)| − 2 ∗ Ego (u, v) − N (v)
N (2)

1
OTH(u, v) = (3)
1+ e−(WNH(u)+WNH(v))

2.3 Fraction of Shared and Unshared (FSU) Neighbors

For an edge (u, v): let f ushared and f uunshared denote respectively the fraction of shared
neighbors and fraction of unshared neighbors among the neighbors of vertex u;
likewise, let f vshared and f vunshared , respectively denote the fraction of shared and
unshared neighbors of vertex v. The fraction of shared neighbors for an end vertex u
(or v) in an edge (u, v) is the ratio of the number of neighbors that are shared by both
the two end vertices u and v and the number of neighbors of the end vertex u (or v),
excluding the other end vertex v (or u). Likewise, the fraction of unshared neighbors
for an end vertex u (or v) in an edge (u, v) is the ratio of the number of neighbors
of u (or v) that are not shared with the other end vertex v (or u) and the number of
neighbors of the end vertex u (or v), excluding the other end vertex v (or u). We define
an auxiliary metric called “shared unshared ratio (SUR)” as shown in formulation
(4) that is in turn used to compute the FSU metric for edge (u, v): see formulation
(5). The SUR ratio values could range from −1 to 1; the FSU formulation basically
transforms the SUR values to a scale of 0 to 1.
Like in the case of WNH metric (see above), we can notice in the formulation for
SUR(u, v) that the presence of unshared neighbors for a vertex could be considered to
negate the advantages that come with the presence of shared neighbors. However, the
difference lies in the extent of penalty given for the presence of unshared neighbors.
The SUR/FSU metric uses a linear penalty alone, whereas the WNH/OTH metric
incorporates both linear and non-linear penalties.
   
f ushared + f vshared − f uunshared + f vunshared
SUR(u, v) =  shared    (4)
fu + f vshared + f uunshared + f vunshared
 
1 + SUR(u, v)
FSU(u, v) = (5)
2
Computationally-Light Metrics to Quantify Link Stability … 231

3 Simulations

We conducted exhaustive simulations to evaluate the following aspects with respect to

the proposed computationally-light metrics (vis-a-vis BPI and ALGC) to quantify
link stability as a measure of the extent with which the neighborhood is shared:
(1) Computation time (2) Correlation (3) Stability of DG trees. The node mobility
model used is the Random Waypoint model [10]; the maximum velocity (vmax ) of the
nodes is varied with values of 1 m/s (low mobility), 5 m/s (low-moderate mobility),
10 m/s (moderate-high mobility) and 30 m/s (high mobility). The number of nodes
is varied with values of 50 (low density) and 100 (high density). We generated
100 mobility profile files for each of the above combinations of node mobility and
density scenarios, with the simulation time being 1000 s. We averaged the results for
the following for each of the above scenarios and each of the five metrics: (1) the
Computation time (on a per-edge basis); (2) the Spearman’s correlation coefficient
(at randomly sampled time instants in the mobility profile files) between any two
of the five metrics; (3) the lifetime of the DG trees. The computer on which the
simulations are ran is a Dell workstation with Intel i-2620 M CPU @ 2.70 GHz and
8 GB RAM.

3.1 Computation Time

Table 1 presents the average computation times (in microseconds) to determine the
link stability score for an edge with respect to each of the five metrics. For this
purpose, we sampled a total of 1000 randomly chosen edges in the network at ran-
domly chosen time instants in each of the 100 mobility profile files and averaged the
computation times. The OTH metric incurs the lowest computation time for both low-
density and high-density networks. The computation times for the FSU metric has
been observed to be slightly larger than the OTH metric. The NND metric incurs the
largest computation time among the three computationally-light metrics proposed in
this research. The computation times for the NND metric are about 2–3 times larger
than the computation times of the OTH and NND metrics. The computation time for
the ALGC metric is the largest among all the five location and mobility-independent

Table 1 Computation times (in microseconds) for the link stability metrics
Condition BPI ALGC NND OTH FSU
50 nodes, vmax = 10 m/s 38.177 42.643 6.514 2.450 2.771
50 nodes, vmax = 30 m/s 36.255 39.889 6.181 2.419 2.596
100 nodes, vmax = 10 m/s 83.839 93.825 4.902 1.967 2.062
100 nodes, vmax = 30 m/s 79.574 89.133 4.960 1.881 2.230
232 N. Meghanathan

metrics. Between BPI and ALGC, for a given node velocity: the difference in the
computation times increases as the node density increases.
The computation times for BPI and ALGC are about 15 and 40 times larger than
that of OTH and FSU in low-density networks and high-density networks, respec-
tively. For a given node velocity, the computation times for BPI and ALGC almost
double (i.e., increase by a factor of 2 or more) as the number of nodes is increased
from 50 to 100. On the other hand, for a given node velocity, the computation times for
the three computationally-light metrics (NND, OTH, and FSU) decreases (by about
20–25%) as the number of nodes is increased from 50 to 100. As a result, we can
say that the time savings obtained with the computationally-light metrics (vis-a-vis
the computationally-heavy metrics) increase with increase in node density.

3.2 Rank-Based Correlation

We determined the Spearman’s rank-based correlation coefficient values [11] for the
edges between the computationally-heavy versus computationally-light link stabil-
ity metrics for every second during the simulation time of 1000 s in each of the 100
mobility profile files and averaged these correlation coefficient values for each pair
of the link stability metrics (as reported in Table 2). For each of BPI and ALGC, we
identify the computationally-light metric for which the largest value for the correla-
tion coefficient is obtained (the entries in Table 2 are shaded). We observe BPI to be
very strongly correlated with OTH for all the four scenarios (in each case, the corre-
lation coefficient is above 0.90). We thus establish OTH to be a computationally-light
alternative to rank the edges in lieu of the computationally-heavy BPI . In the case
of ALGC, it is strongly correlated with NND in low-density scenarios and with FSU
in high-density scenarios. The ALGC-FSU correlation in high-density scenarios is
more stronger than the ALGC-NND correlation in low-density scenarios. For a given
node mobility, with increase in node density, the correlation coefficients associated
with ALGC increase and the correlation coefficients associated with BPI decrease.

Table 2 Spearman’s rank-based correlation coefficient between computationally-heavy versus

computationally-light link stability metrics
Scenarios NND OTH FSU
50 nodes, vmax = 10 m/s BPI 0.3283 0.9566 0.9260
ALGC 0.8029 0.7183 0.7766
50 nodes, vmax = 30 m/s BPI 0.2985 0.9514 0.9287
ALGC 0.7946 0.7389 0.7878
100 nodes, vmax = 10 m/s BPI 0.2583 0.9252 0.8444
ALGC 0.8168 0.8185 0.9051
100 nodes, vmax = 30 m/s BPI 0.2414 0.9291 0.8487
ALGC 0.8045 0.8158 0.9054
Computationally-Light Metrics to Quantify Link Stability … 233

3.3 Data Gathering Tree Lifetime

In Fig. 1, we report the normalized values of the DG tree lifetimes (normalized using
the square root of the sum of the squares of the raw lifetime values) incurred with
the five metrics. The lifetimes of the DG trees determined based on each of the three
computationally-light metrics are greater than that of the ALGC-based DG trees.
Between BPI and ALGC, the BPI -based DG trees incurred the largest lifetime for
both networks of low density and high density. The largest tree lifetimes incurred
with the {NND, OTH, FSU} metrics are very much comparable with the largest
tree lifetimes incurred with the {ALGC, BPI } metrics. The maximum difference in
the tree lifetimes could be observed in the low mobility and high-density scenario
(vmax = 1 m/s and 100 nodes). For a given mobility level of the nodes, the DG tree
lifetimes for all the five metrics increases with increase in network density, due to
the increase in the extent of shared neighborhood (and thereby an increase in the
stability of the links). Among the three computationally-light metrics, the OTH-
based DG trees incurred the largest lifetime in low-density networks that is almost
the same (or even sometimes larger) as the lifetime of the BPI -based DG trees. The
FSU-based DG trees incurred the largest lifetime in high-density networks that is also
almost the same (or even sometimes larger) as that of the BPI -based DG trees. Thus,
the OTH and FSU metrics could, respectively, be computationally-light alternatives
for BPI in low and high-density networks.
Note that a common theme among the BPI , OTH and FSU metrics is that they
follow the strategy of penalizing the link stability score (LSS) of an edge if its
egocentric network comprises of unshared neighbors and rewarding if the egocentric
network comprises of shared neighbors. The relatively larger normalized lifetime
values for the BPI , OTH, and FSU-based DG trees (vis-a-vis the ALGC and NND-
based DG trees) could be attributed to such a combined strategy of rewarding as
well as penalizing the LSS scores of the edges depending on the extent of shared
and unshared neighborhoods, respectively. On the other hand, the ALGC and NND
metrics follow the strategy of just rewarding the LSS score of an edge if it comprises

Fig. 1 Normalized lifetime of the data gathering trees with respect to link stability metrics
234 N. Meghanathan

of egocentric networks with shared neighborhood and not penalizing if the egocentric
networks comprise of unshared neighborhood.

4 Conclusions and Future Work

We have developed three computationally-light metrics (NND, OTH, and FSU),

two of which are shown to be potential alternatives (OTH and FSU) to the
computationally-heavy metrics BPI and ALGC. The computation times of BPI and
ALGC are 15–40 times larger than that of OTH and FSU (the difference increases
with increase in node density). On the other hand, we observe strong-very strong pos-
itive correlation (for ranking the edges) between OTH and BPI as well as between
FSU and ALGC. The lifetimes of the data gathering (DG) trees determined using
the OTH and FSU metrics (as the link stability scores) are observed to be as large as
those determined using the BPI metric in low-density and high-density networks,
respectively. In an earlier research [1], the BPI metric was observed to determine the
most stable DG trees for mobile sensor networks (when compared to the use of the
predicted link expiration time, LET [4, 5]). Thus, we are very confident to have iden-
tified two computationally-light metrics based on the extent of shared neighborhood
to quantify link stability in mobile sensor networks. In future, we plan to predict the
actual BPI values for the links using their OTH and FSU values itself so that we could
determine BPI -based DG trees by avoiding a heavy computation overhead. We also
plan to incorporate computationally-light secure encryption techniques (like [12])
to exchange the neighborhood information of the nodes for building the egocentric
edge networks.

References

1. Meghanathan N (2018) Complex network analysis-based graph theoretic metrics to determine

stable data gathering trees for mobile sensor networks. Comput J 61(2):199–222
2. Estrada E, Rodriguez-Velazquez JA (2005) Spectral measures of bipartivity in complex
networks. Phys Rev E 72(046105):1–6
3. Fiedler M (1973) Algebraic connectivityh of graphs. Czechoslov Math J 23(98):298–305
4. Meghanathan N (2012) Link expiration time and minimum distance spanning trees based
distributed data gathering algorithms for wireless mobile sensor networks. Int J Commun
Netw Inf Secur 4(3):196–206
5. Su W, Gerla M (1999) IPv6 flow handoff in Ad hoc wireless networks using mobility prediction.
In: IEEE Global Telecommunications Conference, pp. 271–275. IEEE Press, New York
6. Abolhasan M, Wysocki T, Dutkiewicz E (2004) A review of routing protocols for mobile ad
hoc networks. Ad Hoc Netw 2(1):1–22
7. Smys S, Raj JS (2015) A self-organized structure for mobility management in wireless
networks. J Comput Electr Eng 49C:153–163
8. Osborne J (2014) Best practices in logistic regression, 1st edn. Sage Publications, Thousand
Oaks
9. Agarwal CC (2018) Neural networks and deep learning strategies, 1st edn. Springer, Berlin
Computationally-Light Metrics to Quantify Link Stability … 235

10. Bettstetter C, Hartenstein H, Perez-Costa X (2004) Stochastic properties of the random-way

point mobility model. Wirel Netw 10(5):555–567
11. Strang G (2016) Introduction to linear algebra, 5th edn. Wellesley-Cambridge Press, Wellesley
12. Praveena A, Smys S (2016) Efficient cryptographic approach for data security in wireless
sensor networks using MES V-U. In: 10th international conference on intelligent systems and
control, pp 1–6. IEEE Press, New York
MockRest—A Generic Approach
for Automated Mock Framework
for REST APIs Generation

Anshu Soni, Virender Ranga and Sandeep Jadhav

Abstract Mock is an object that replicates the behavior of a real object in a

disciplined way and improves unit testing. Unit testing is a testing where each indi-
vidual or component is tested. The purpose of unit testing is to validate each unit
of designed software and allow to verify the generated code is working properly,
regardless of its dependencies. A system under test has some external dependencies
like APIs and creating a mock object based on that kind of dependencies would be
efficient rather than generate a test case on the actual instance of the dependencies.
A real working system such as banking, autonomous vehicles, online-supply chain
businesses, and E-commerce platforms are heavily dependent on a server and facing
difficulty while testing with a real server. Mock server helps in testing by simulating
the behavior of a real server. Mocks could be used for testing and developing the
front-end even when the back-end is not available. The aim of our research work is to
propose a generic approach in which we propose a mock framework named Mock-
Rest for REST API in Java. The main reason to propose such kind of framework is to
get a consistent response while real API is down at the moment by creating a mock
of REST API as it allows the developer to stay constructive while the API is being
implemented. Application Programming Interface (API) allows interaction between
software programs, exchanges their information while REST is an architectural style,
and applies to the design of API. A Web API that follows the standards of REST
architectural style is a REST API. Based on the description of Web services by its
interface, Mock simulates its behavior.

Keywords MockRest · Mock · REST API · Unit testing · Framework ·

WireMock · SOAP UI · Jersey · JAX-RS · Tomcat · JAXB

A. Soni (B) · V. Ranga

National Institute of Technology, Kurukshetra, Haryana, India
e-mail: anshusoni1995@[Link]
V. Ranga
e-mail: [Link]@[Link]
S. Jadhav
KPIT Technologies Limited, Pune, Maharashtra, India
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 237

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
238 A. Soni et al.

1 Introduction

Unit testing targets to test a single unit of a program individually. It is difficult to test
in isolation as code can interact with external dependencies such as a server, database,
and API. For example, test a unit that interacts with external dependencies, which
requires “real” file to be existed or to be created in order to ensure unit testing. A Java
Web application consists of two constituents—front-end and back-end server. Both
are dependent on each other and run simultaneously. The front-end developer depends
on the back-end for a server, third-party APIs, database, and other external services
required. For effective and efficient testing, a major challenge is to handle the various
dependencies such as external data, third-party services, and external libraries. In
unit testing, tester tests the software component without incorporation of external
dependencies. In a real, system such as banking, autonomous vehicles, online-supply
chain businesses, and E-commerce platforms exchanges their data through a server
and a database where customer services are handled by a different server, and admin
handled their services through another server. It would be burdensome for tester as
well as for a developer to do testing. During the development phase, in order to
ensure test with a real server, it would be difficult to evaluate test cases by a tester or
a developer. So, to provide effective testing, there is a requirement of a mock server.
It will remove the dependency on a real server and allows testing by the developer
itself. Figure 1 gives a visual representation of a mock server in place of a real server.
Client or any Web application sends HTTP request to the real server and server return
HTTP response get back to the client over the Internet. Now, in order to effectively
utilize our time and improve testing mechanism, there is a requirement of a mock
server that behaves as a real server and mimics its dependencies. In order to simulate
external dependency and model the environment, developers can create a mock.

Fig. 1 Mock server

MockRest—A Generic Approach for Automated Mock … 239

Fig. 2 Mock server

Mock objects can be used to reconstitute the real software dependencies by mim-
icking the admissible feature of software dependency. Reduction of time takes place
while testing software development by the creation of mock objects. Some of the
mocking frameworks available to unit test java application are Jmock, Mockito, and
EasyMock.
REST stands for “Representational State Transfer,” an architectural style for devel-
oping Web services. API is a software intermediary that allows two applications to
talk to each other. RESTful Web services in Java applications have been developed
by JAX-RS with jersey.
When front-end and back-end developers work parallelly, then front-end devel-
oper needs mock responses even API is not available or not implemented yet. By
mocking REST API, simply means creating a mock server which returns pre-defined
responses and other parameters such as status codes and headers on matching request
by the client.
This research paper proposes a framework of our generic approach that generates a
mock of REST API. Wiremock and SoapUI are the existing tools available generating
REST API mock, but they have some limitations as they are not creating mock
responses as many REST APIs simultaneously, and generate stub responses. There
is a lack of study that explores REST API mock creation; this paper gives an idea
of framework creation for mock REST API. Figure 2 shows the behavior of mock
server.
As discussed above a generic idea of mock server and the basic steps used to
implement a mock server is shown below:
1. Pass the REST API to the framework that is created with the code.
2. Expectations can be set in the mock server according to our requirements.
3. Mock server returns responses according to the API defined and pre-defined
responses.
4. Finally, code under test verifies the responses by the mock server without including
dependency of real server.
240 A. Soni et al.

2 Related Work

Mock object is a technique that assists unit testing by simulating the real object.
Thomas et al. [1] is the well-known literature about mock objects. This paper demon-
strates the usage of mock objects. These mock objects can be used when real objects
have non-deterministic behavior, difficult to set up, has a user interface. When a real
object does not exist yet, then there is a need of mock objects.
Kim et al. [2] identify a mock object model for test-driven development(TDD).
They identify the features and limitations of the existing mock object model and
propose a new model to overcome limitations of previuos proposed work.
Spadini et al. [3] observe why developers apply mock objects after conducted
a survey over 105 developers from software testing communities. They have done
interviewed of developers to understand why some dependencies were mocked and
some are not as well creation of MockExtractor,a tool that extracts a list of mocked
and non-mocked dependencies in test suites. We have observed the following facts
after doing the literature review thoroughly:
– Database, Web services, and external dependency are mocked dependencies.
Domain objects are only ∼36% mocked.
– In integration testing, mock is not used.
– Interfaces are mocked so that there is no need to depend on specific
implementations.
– Test support and Java libraries are not mocked.
While creating mock, developers faced most of the challenges that are maintaining
mock behavior of original class along with its compatibility. Mocking with legacy
system (where architecture is not well developed) is the major challenge.
Mostafa et al. [4] have given a study on four most popular mocking frameworks,
i.e., EasyMock, Mockito, JMock, JMockit by considering 5000 Github open-source
software projects. They have also found out that how developers have used mocking
frameworks in open-source software projects. They have shown that 23% of software
projects used mocking frameworks.
Singhal et al. [5] propose an approach that quickly generates a mock in front-
end that behaves exactly like an actual back-end. It makes front-end and back-end
developer independent and utilizes their time effectively.
Ashikhmin et al. [6] describe a architecture and its implementation which can
generate mock services based on RAML specification and deploy with Docker con-
tainer. Define Gateway, Request validator, Path Resolver, and Response Generator
are the four components in mock services as well give an overview of microservices.
Adamczyk [7] distinguishes two architectural styles REST and SOAP and explains
briefly about the RESTful Web services and demonstrates four principles of REST
coined by Roy Fielding based on a survey of existing Web services.
Simple rules have been defined to design Web services as discussed in [8]. The
authors also discuss why REST API should be designed and configured and define
the tips to address the client’s needs.
MockRest—A Generic Approach for Automated Mock … 241

Kao et al. [9] introduce the performance testing framework for REST-based Web
applications and provide software tester an integrated process from test case design,
generation of test scripts to the execution of tests. This framework decreases the
effort to understand the design and implementation of application.
Haupt et al. [10] provide a framework for structural analysis of REST APIs which
consists of metrics and graphical representation of APIs. They also convert the
description of REST API, present in different languages into a canonical model
that describes the structure of REST API explicitly.
Ed-douibi et al. [11] focus on REST API automated test case generation by using
OpenAPI for automatic REST API development. They have proposed an approach
to generate test cases automatically and test API specifications according to the
requirements and provide a tool for testing 91 OpenAPI specifications. They conclude
that 41% test APIs are failed. Their approach depends on a model-based procedure
to automate test case generation.
Arcuri [12] considers the testing at the development phase by the developers
having full code access. They have proposed an approach that uses an evolutionary
algorithm for automatic test case generation based on white-box testing approach.
They have done their experiments of proposed approach on EvoMaster, open-source
tool.
Munonye and Martinek [13] analyzed the performance of REST Web services
based on Java and .Net implementations and concluded their performance. On various
test scenarios, they found that Java-based Web services perform better for GET
method around 80.36%, and .Net performs better for PUT method and performance
is 11.6% lower. They came across the conclusion that Java-based Web services should
be used on business-to-business purposes while .Net-based Web services preferable
on forms updating applications having user involvement regular update and deletion
takes place.
Sinha et al. [14] scrutinize a new approach for application integration on cloud
with REST Web services.
Giessler et al. [15] identify best practices for developing REST services and
clarify the usage of Web services on different applications. This paper illustrates
eight different categories of Web services and 23 best practices.
Qiu et al. [16] simplify the usage of API in Java programming language. They
discuss how API engages in software projects. They have done the empirical analysis
on 5000 open-source Java projects, understand the core API and third-party API in
Java, knowing the concept of API libraries. They conclude that core API is not widely
used, most of the APIs that are deprecated are still used.
242 A. Soni et al.

3 Background Study

3.1 Unit Testing

A chunk of code written by a developer that handles a very tiny, specific area of code
being tested comprises unit testing [17]. Unit testing basically done over different
methods or classes. It is done to prove that chunk of code does exactly what developers
think. It should be done over all modules independently before two modules combine
together. Will make developer’s life easier by reducing the amount of time spent on
debugging. Use assertion to check whether a chunk of code behaves exactly as a
developer wants. An assertion is a method that verifies whether testing statement
returns true or not. For the Java programming language, JUnit framework is used.
There are frameworks available for Java are NUnit, TestNG, and many more. Many
assert methods available in JUnit library like
assertEquals(String[message],expected,actual),
assertTrue(string[message],boolean condition) and much more.
The objective of the testing is to exercise one method at a time, but when method
depends on the external database, third-party APIs, then there is a need of mock
objects to reduce time handling these external services. Figure 3 shows a flow chart
of unit testing and our focus on testing with mock.

3.2 Mock Objects

Mock objects are referred to as a simulated object that simulates the nature of real
objects in a disciplined way. To simulating the required environment, mock objects
[17] are required. It is essential for test case generation. EasyMock, Mockito, Pow-
erMock, and JMock are the Java mocking frameworks available. The purpose of a
mock object is to enable unit testing by testing the code if it has interaction with
external dependencies by simulating them.
Interfaces are the basic concept in object-oriented systems. Interfaces are lan-
guage constructs that contain only a set of methods signature, expected behavior
of the object, does not contain any implementation details [18]. Their implementa-
tion is defined by its class. The system adopts interfaces have three characteristics:
Flexibility, Extensibility, Pluggability.
For unit testing, mock objects can be used in place of real objects by simulating the
interfaces required. Mocks are easiest to use while interface-based design systems.
Test suites and domain code can be improved by mock objects used in unit testing.
Term Endo-Testing [19] have been announced, in which code has been tested from
inside where mock objects are passed to the target domain code. Mock objects are
the good technique to make testing easy for developers without interfering with
external dependency, third-party server when they are not available and not yet been
implemented. There is some pattern to be followed for unit testing with mock objects:
MockRest—A Generic Approach for Automated Mock … 243

Fig. 3 Unit testing flow

diagram

– Build an instance of mock object.

– Define states of mock objects.
– Prescribed expectations, responses in mock objects.
– Set mock object as parameter under domain code.
– Verify mock objects under testing.

3.3 REST API

An architectural style for developing Web services is REST(Representational State

Transfer) was first conferred by Roy Fielding in 2000. It defines a set of constraints
for developing Web services. Fielding [20] defines seven constraints described below
in their dissertation report:
– Starting with NULL style: It is an empty set of constraints. It is the starting point
to describe REST.
– Client—Server: separation of client and server architectural style to evolve inde-
pendently of each other, thus improves the portability under user interface and
scalability by facilitating server components.
244 A. Soni et al.

– Stateless: Communication between client and server must be stateless. Any request
of the client will not be saved. No session nor history be made of a client request.
Management of resources is not required. This constraint brings about the prop-
erties of scalability, visibility, and reliability.
– Cache: To improve network efficiency, cache constraint is added between client
and server communication. This constraint requires that response is declared as
cacheable or non-cacheable explicitly or implicitly. Cacheable response requires
the client to reuse the response. Caching improves the performance over client-side
and scalability be improved over server-side.
– Uniform Interface: REST architectural style different from other network style
is through uniform interface constraint. It decouples and simplifies the architec-
ture, which enables to evolve independently. Four interface constraints have been
defined, i.e., identification of resources, manipulation of resources through repre-
sentations, self-descriptive messages, and hypermedia as the engine of application
state (HATEOAS).
– Layered System: To improve Internet-scale requirements, the layered constraint
has been added. In this, architecture is composed of hierarchical layers so that
each component was not able to view beyond the immediate layer toward they
were connected. A layered constraint is applied to enclose legacy services. The
disadvantage over this system is that it increases the overhead and delay to the
data processing.
– Code-on-demand: Functionality be added under REST to downloading and exe-
cuting code. It allows the client to reduce features that would be needed to be
pre-implemented. This constraint improves system extensibility.
API (Application Programming Interface) acts as an interface that allows communi-
cation between two software programs. API that uses HTTP protocol for interaction
between different programs are Web services. Basically, the interaction of Web ser-
vices is the request and response type between client and server. An API follows
REST rules or standard for creating Web services, which are REST API. REST is an
architectural style, not a protocol that develops over HTTP. For request and response,
REST [11, 12] uses HTTP (HyperText Transfer Protocol) to define the desired action.
To allow communication, there are resource methods or request types available:
– GET: Retrieve resource information.
– POST: Create new resource.
– PUT: Update the existing resource.
– DELETE: Delete the existing resource.
Figure 4 represents REST API workflow following HTTP methods and returns
response to clients in JSON/XML format. HTTP methods described above can be
safe and idempotent [15] as well.
Table 1 exhibits the properties of HTTP methods which are idempotent and safe.
Idempotent methods are those HTTP methods that can be called as many times
without any effect on stored data, whereas safe methods are those HTTP methods
which do not update the resources, these methods are read-only.
MockRest—A Generic Approach for Automated Mock … 245

Fig. 4 REST flowchart

Table 1 Properties of HTTP HTTP methods Idempotent Safe

Methods
GET Yes Yes
POST No No
PUT Yes No
DELETE Yes No

The representation of state can be in JSON or XML format. Usually, today most
preferable response format will be in JSON format. There is some structure [8] that
is to be followed while creating a REST API.
– Base URI: To identify resources Uniform Resource Identifier(URI) be used.
Example: [Link]
URI format:
scheme“/”authority“/”path[“?”query][“#”fragment]
– Resource modeling: Representation of data is called resource. It can be a singleton
or can be a collection, and it may contain sub-resources also. Resource modeling
can be separated by a forward slash in the URI path.
Example: [Link]
– Query Design(Parameter): Query string is separated by “?” symbol. It contributes
to the identification of resources uniquely. It contains parameters that identify
hierarchically along with path segment.
Example: [Link]
Figure 5 shows rest API structure consists of BaseURI, resources, Parameters, and
Query String.
246 A. Soni et al.

Fig. 5 REST API structure with example

4 Existing Tools

4.1 WireMock

WireMock [21] is an HTTP mock server, a simulator for APIs. It can be used to mock
API for testing. Today’s, mocking REST services are necessary for our development.
It can give the canned responses (known as Stubbing) to an individual request match-
ing. It is considered as a mock server or service virtualization tool. It facilitates to stay
productive when API not yet developed or down at a time. By any JVM application,
it can be used as a library or can be used as a standalone and configured via JAVA
API, JSON files, or JSON over HTTP.
To include wiremock into test cases can be configured with JUnit rule. With default
port (8080), basic usage of wiremock into test cases by @Rule annotation described
below.
@Rule
public WireMockRule wireMockRule = new WireMockRule();
With programmatically, it can create, start, and stop as specified with command.
It can do stubbing with stubFor method which takes HTTP methods and URL for
request matching and will return a response in JSON mapping format. It is a bit
faster as it avoids sending a command over HTTP. It will concur with a third-party
server and record tested API and response as a cache. Basically, it behaves like a
cache while API testing. For one API server, one wiremock instance is configured. It
will not create a client object if anyone can want to mock multiple services it should
be declared as a rule(@Rule) per service. It will deploy as a WAR into a servlet
container.
Limitations

– Too slow and complicated as don’t test for the same database again and again.
– Daily update Wiremock if it has date function calls.
– All test cases can’t be mapped in one set.
– Mocking is done with only one instance of wiremock. If a client wants to connect
to more than one server, multiple wiremock instances run on different ports and
pointing to one proxy server.
MockRest—A Generic Approach for Automated Mock … 247

– If user resets the API, then it cannot turn back to original API could be done by
restarting the server again.

4.2 SOAP UI

A tool is used for testing Web services. It is a complete Automation framework

for API (SOAP or REST) testing. It contains many features like inspection Web
services, development, Mocking API, and simulation. It will create a static mock
implementation in seconds and also create a Load and Functional tests of mock
Web service before their actual implementation. With SOAP UI, multiple responses
depending on request by scripting and also allows different status codes according to
the request and response defined. With scripting, modify the mock response headers,
contents at the time dispatching to a client. Able to add more than one responses
according to the request initiated dynamically.
Limitations
– Assertion to JSON response not easy.
– Having a delay in response.
– Don’t handle different REST API at a time and generate mock server accordingly.

5 Motivation

As discussed before, we can use mock objects to test code in isolation without
integrating with external dependencies. To verify the interaction of class under test
(CUT), the mock objects are used. It returns responses to method calls that are pre-
defined under the mock framework. While testing, developer has to decide accord-
ingly whether to test a unit in isolation by simulating all of its dependencies or test
along with its dependencies together. By simulating its dependencies, developers
gain focus over testing a single unit not on its dependencies as far as testing with-
out simulations slows down the execution of a test, testers require full control over
external dependencies, and it is costly to prepare for testing. For getting a quick and
consistent response, there is a need to mock REST API. Back-end service that is
slow to respond should be mocked. Not to get dependent on back-end developers,
front-end developers require dummy data or response.
As shown in (Fig. 6), the process of developing REST API or any other software
gone through these steps without mock.
1. Define interface of REST API, their Base URI.
2. Implement REST API resource methods their associated parameters, query string
along with Base URI.
3. Development server testing has been done by the developers. It is an essence layer
in software development where all set up like hardware, software, and other nec-
essary component required to debugging and test the code directly by developers.
248 A. Soni et al.

Fig. 6 Development process without Mock

4. Stagging server testing is a pre-deployment process. In this server, simulation is

set up as a real environment for testing, all production configuration has been set
up. It is the endwise step before final production of software.
5. Production server where Web site, REST API get deploy and hosted, also referred
to as a live server. Its environment is actually identical to staging server. It is the
terminal for all the accomplished task which has been done.
In this complete development process cycle, development server testing takes
almost 3 h, and staging server takes around 4 h for testing the REST API. Without
creating a mock object takes lots of time and efforts for development. By creating
a mock object, we save our time and utilize it efficiently. This research paper gives
an idea of how we can create a mock of REST API. There are already existing tools
available like wiremock and SoapUI for mock REST API, but they have limitations
while creating mock. They don’t create different instances of mock of different REST
APIs. They worked efficiently for only one instance at a time. To overcome these
limitations and save time during development, there is a need of mock.

6 Proposed Approach

6.1 Challenges

In this research paper, we have shown a general approach to create a mock framework
for REST API. Getting independence from back-end services and improving the
execution of a class under test is done by creating a mock server. For creating mock
objects, most common Java frameworks available are JMock and EasyMock. They
allow us to create mock objects and define the behavior of it, exactly what you expect
when we call methods on the mock object. This research work proposes an approach
of how a mock framework of REST API is generated so that there is no need to
interfere with the real server code according to our need. We create a framework of
mock server and use it according to our requirements. Our mock server takes input
data through a XML file which is not hard-coded in the framework design.
MockRest—A Generic Approach for Automated Mock … 249

6.2 MockRest Framework

In our proposed approach, we define a framework of a mock server that mock the
third-party component in which expectations are already defined according to our
requirements against API. We have tested our system against API and got a response
from the mock server accordingly. We propose a framework that makes it easier to
generate mocks for REST API services. Our approach supports as many API calls
as required. Our framework allows taking data through an XML file not to be hard-
coded in the framework design. Return responses, throw exceptions, events for API
call if required and allows Test Driver to verify REST API calls, their behavior with
the mock server not with the real server.
An approach described in Fig. 7 gives MockRest Framework operations and meth-
ods and their behavior while generating mock instances. As said above, framework
takes their data through an external document containing XML/JSON data and gen-
erating instances of mocked REST.
A mock server contains all those mocked instances and return responses with the
status code, exceptions, and events as defined according to the requirements when
test case runs. To assert with test cases, mocked instances have to pass as an argument
of the constructor of the code under test. This is how Mocked REST API helps in
testing during the development phase. Figure 8 shows the document containing XML
data that would be passed in a MockRest framework to read data through an XML
file by JAXB. Figure 10 shows the sequence diagram of RestMock (Fig. 9).

6.3 Create REST API with Java (JAX-RS) Using Jersey

Jersey is an open-source framework for developing REST services in java that sup-
ports JAX-RS APIs. JAX-RS has been designed to make it easier for development
of RESTful Web services in java.

Fig. 7 Approach of MockRest framework

250 A. Soni et al.

Fig. 8 [Link]

Fig. 9 Jersey dependencies in maven

Technolgies used:
– Jersey version—2.25
– Apache Maven—3.8.0
– Tomcat 7.0
– Eclipse Java IDE Mars 2.0.
In order to add jersey in Maven Project creating REST API, dependency should be
added as shown in Fig. 9.
MockRest—A Generic Approach for Automated Mock … 251

Fig. 10 Sequence diagram of RestMock

Fig. 11 [Link]

Under the [Link], these changes to be reflected while creating REST service with
URL Pattern “rest” as shown in Fig. 11. Table 2 describes some of the annotations
by JAX-RS.

6.4 Read XMl File Through JAXB

JAXB is a Java standard that defines API to convert Java object to and from XML. It
stands for Java Architecture for XML Binding. It provides structure to write(marshal)
Java objects into XML and unmarshal(read) XML into Java object. There are no
252 A. Soni et al.

Table 2 JAX-RS annotations

Annotations Description
@Path Specify the URI path
@GET Annotate HTTP GET request method
@POST Annotate HTTP POST request method
@DELETE Annotate HTTP DELETE request method
@PathParam Extract URI path parameters
@QueryParam Extract URI path query parameters
@Produces Specify media type to be produced
@Consumes Specify media type to be produced by REST

Table 3 JAXB annotations

Annotations Description
@XmlRootElement(namespace = Define the global(root) element of xml tree
“namespace”)
@XmlElement Annotate field that need to be in output
@XmlType Mention order of xml elements
@XmlAttribute Maps java Bean property to xml attribute

external libraries for JAXB as long as jersey builds on the project. There are some
of the annotations required to support XML binding described in Table 3. To convert
XML data into Java objects, Unmarshal() method to be used. In JAXB, there is an
abstract class named JAXBContext that provides client’s entry point to the JAXB API,
gets initialized with the class used, creates Unmarshaller, and returns an unmarshaller
object that can be used to convert the XML data. Unmarshaller interface provides
client XML data and converts into Java object by the function named unmarshal().

7 Results Observed

Two REST APIs created dependent on each other.

[Link]
and
[Link]
Both REST API for messenger one for retrieving messages and other for retrieving
comments are given by messageid. As shown above, comments are dependent on
message-id. By giving message-id in the REST API, comments are retrieved by Get
Http method. File [Link] shown in Fig. 8 has been passed into the framework, and
mock server generates expected result after creating mock of REST API as shown
in Fig. 12. As observed, mock server returns comment and its created date with
MockRest—A Generic Approach for Automated Mock … 253

Fig. 12 Result after creating Mock

status code 200, a status message with type Success and header of application/xml
and mocked response as a message with message-id “Messageid:1 with comments
returned Successfully.” in response tag.

8 Conclusion

This research work proposed a generic approach to create a mock framework for
REST API. The MockRest framework generated a mocked object of REST APIs.
Mock server generated a pre-defined response, events, and exceptions to improve
testing by the developers without getting involved with a real server. Mocked APIs
could be used by testers to simulate external dependencies and make efficient testing
with mocked APIs.
254 A. Soni et al.

The MockRest framework has also taken the dependent list of APIs and generated
mocked objects on existing mock server which are not been able to do as they have a
one-to-one relationship. Future benefits of RestMock are that it could manage third-
party services, speed up the testing and development, and speed up the generation of
test cases.

References

1. Thomas D, Hunt A (2002) MOCK objects. IEEE Softw (Published in Journal) 19(3):22–24
2. Kim T, Park C, Wu C (2006) Mock object models for test driven development. In: Fourth interna-
tional conference on software engineering research, management and applications (SERA06).
IEEE
3. Spadini D, Aniche M, Bruntink M, Bacchelli A (2017) To mock or not to mock? An empirical
study on mocking practices. In: IEEE/ACM 14th international conference on mining software
repositories (MSR), pp 402–412
4. Mostafa S, Wang X (2014) An empirical study on the usage of mocking frameworks in software
testing. In: 2014 IEEE 14th International conference on quality software. [Link]
1109/QSIC.2014.19
5. Singhal N, Jain H (2016) Automock: automated mock backend generation for javascript based
applications, vol. 16, Issue 7 Version 1.0. Online ISSN: 0975-4172 & Print ISSN: 0975-4350
6. Ashikhmin N, Radchenko G, Tchernykh A (2017) RAML-based mock service generator for
microservice applications testing. [Link]
7. Adamczyk P, Smith PH, Johnson RE, Hafiz M (2011) REST and web services: in theory and in
practice. Springer Science+Business Media, New York. [Link]
8303-9_2
8. Masse M (2011) Designing consistent RESTful web services interface. OReilly Media. ISBN:
978-1-449-31050-9
9. Kao CH, Lin CC, Chen JN (2013) Performance testing framework for REST-based web appli-
cations. In: 2013 IEEE 13th international conference on quality software. [Link]
1109/QSIC.2013.32
10. Haupt F, Leymann F, Scherer A, Vukojevic-Haupt K (2017) A framework for the structural
analysis of rest APIs. In: IEEE international conference on software architecture. [Link]
org/10.1109/ICSA.2017.40
11. Ed-douibi H, Izquierdo JLC, Cabot J (2018) Automatic generation of test cases for REST
APIs: a specication-based approach. In: IEEE 22nd international enterprise distributed object
computing conference. [Link]
12. Arcuri A (2017) RESTful API automated test case generation. In: IEEE international conference
on software quality, reliability and security
13. Munonye K, Martinek P (2018) Performance analysis of the microsoft .net- and java-based
implementation of REST web services. In: IEEE 16th international symposium on intelligent
systems and informatics, Subotica, Serbia, 13–15 September 2018
14. Sinha R, Khatkar M, Gupta SC (2014) Design & development of a REST based web service
platform for applications integration on cloud. IJISET—Int J Innov Sci Eng Technol 1(7):385-
389
15. Giessler P, Gebhart M, Sarancin D, Steinegger R, Abeck S (2015) Best practices for the design
of RESTFul web services. In: Tenth international conference on software engineering advances,
Barcelona
16. Qiu D, Li B, Leung H (2016) Understanding the API usage in Java. In: Information and software
technology, Elsevier, Amsterdam, pp 81–100
MockRest—A Generic Approach for Automated Mock … 255

17. Hunt A, Thomas D (2003) Pragamatic unit testing with JUnit, First printing, September 2003.
ISBN 0-9745140-1-2
18. Nandigam J, Gudivada VN, Hamou-Lhadj A, Tao Y (2009) Interface-based object-oriented
design with mock objects. In: 2009 IEEE sixth international conference on information tech-
nology: new generations, pp 713–718
19. Mackinnon T, Freeman S, Craig P (2001) Endo-testing: unit testing with mock objects
20. Fielding RT (2000) Architectural styles and the design of network-based software architectures.
Dissertation
21. WireMock (2018). [Link] Accessed 8 September 2018
22. Marri MR, Xie T, Tillmann N, de Halleux J, Schulte W (2009) An empirical study of testing file-
system-dependent software with mock objects. In: ICSE workshop on automation of software
test, AST 2009, pp 149–153
23. SoapUI (2018). [Link] Accessed 12 October 2018
24. Jersey (2018). [Link] Accessed 5 Oct 2018
Syntactic Interoperability in Real-Time
Systems, ROS 2, and Adaptive
AUTOSAR Using Data Distribution
Services: An Approach

Navrattan Parmar, Virender Ranga and B. Simhachalam Naidu

Abstract DDS is a real-time protocol for fast communication. It implements Data-

Centric Publish–Subscribe (DCPS) implementation and an optional higher-layer
Data Local Reconstruction Layer (DLRL). DCPS ensures the reliability of message
delivery to proper recipient and uses in syntactic interoperability in different plat-
forms and languages. ROS is a widely used platform to develop robots, drones, and
other cyber–physical systems (CPSs). ROS 2 is built on top of middleware DDS and
provided abstraction in communication. Adaptive AUTOSAR (Automotive Open
System Architecture) also adopted the DDS standards as one of the communication
bindings. This research paper proposes connection establishment and interoperabil-
ity between ROS 2 and Adaptive AUTOSAR software using DDS as a middleware.
Interoperability is the major challenge with the increasing number of IoT devices,
being solved by DDS. The outcome of this research is useful for autonomous cars,
and the proposed concept can be extended for fog computing and other interoperabil-
ity problems. DDS will bring a revolution in the near future in automotive industry,
smart grid, smart homes, and other smart applications.

Keywords Syntactic interoperability · DDS · Adaptive AUTOSAR · ROS 2.0 ·

Fog computing · Cyber–physical system (CPS) · Electric control unit (ECU) ·
Automotive software stack · Adaptive application (AA) · Adaptive Platform (AP) ·
Inter-process communication (IPC)

N. Parmar (B) · V. Ranga

National Institute of Technology, Kurukshetra, Haryana, India
e-mail: parmarnavrattan@[Link]
V. Ranga
e-mail: [Link]@[Link]
B. Simhachalam Naidu
KPIT Technologies, Bengaluru, India
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 257

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
258 N. Parmar et al.

1 Introduction

DDS is a decentralized messaging architecture where we need not worry about inter-
face interaction from the development point of view. It is persistent, scalable, and
fault tolerant. It can control throughput, data latency, data availability, and data deliv-
ery. Now, with the support of DDS in many softwares, it is possible to communicate
between them.

1.1 ROS 2.0 and Adaptive AUTOSAR

Robot Operating System (ROS) is the widely used platform nowadays to develop
robots, drones, and other cyber–physical systems (CPSs) as explained previously. It is
a very popular middleware that can provide abstraction at communication level. ROS
is maintained by Open Source Robotics Foundation (OSRF) and Willow Garage. It
uses publish–subscribe transport method [1, 2], multiple libraries (e.g., rcl, OpenCV,
and the Point Cloud Library (PCL) [3]), and tools to help low-level abstraction and
enhance the productivity.
Adaptive AUTOSAR is the automotive software stack that provides standard
interfaces. It is currently being used in automotive industry for developing state-of-
the-art ECUs running on multi-core processors. Its interfaces allow the equipment
manufactures to implement autonomous driving, over-the-air updates, IoT features,
media streaming, and other advanced services.
The main aim of this research work is to propose an approach to solve interoper-
ability problem using DDS. In this research paper, a systematic approach is proposed
for syntactic interoperability between ROS 2.0 and Adaptive AUTOSAR. Our work
demonstrates the publisher–subscriber method and communication between them.
Figure 1 shows a communication model through DDS.

2 Related Work

In the new era of communications, for a large number of IoT devices, people from
different domains, systems, and knowledge are required to communicate with each
other [4] and heterogeneity is the major problem. It can be solved by DDS. Next
section shows the literature review done by us.

2.1 ROS Literature Review

ROS is the middleware that was initially used to build small robots, but it is now being
used to make drones’ applications as well as in large projects. Our proposed research
Syntactic Interoperability in Real-Time Systems, ROS 2 . . . 259

Fig. 1 Communication via DDS bus

work aims to integrate ROS [5–7] with Adaptive AUTOSAR for autonomous drive
capabilities in automobile industry.
Rhoades et al. [8] describe ROS, CAN, and QRE to enable control vehicle via
ROS and enable the independent development and communication. The authors
use RX63n board for evaluation and control. RX63n is used to send message to
ROS master. ATV and QRE are used to receive the messages from ROS master and
controlled the vehicles.
Mauyama et al. [9] give in-depth overview of ROS 2. With the adaptation of
DDS in ROS 2, there is an evolution in the ROS. Now, it can be used for real-time
communication. This paper analyzes the DDS capabilities and evaluates various
challenges, overhead, and constraints of DDS and ROS. ROS adopts DDS only
for inter-process communication. The experimental part shows the capability and
efficiency of DDS in ROS 2, and it shows that data is not lost even if message size is
4 MB. The authors evaluate the performance of ROS 2 and ROS 1 in various cases
of remote and local communication and also evaluate the parameter of QoS in ROS
2, number of threads, memory consumption, throughput, and latencies. Finally, this
paper concludes the advantages of using DDS. The authors show that DDS makes
ROS real time and capable of running on multiple platforms.
Liu et al. [6] observe DDS in ROS 2. They conduct the experiments for the
messages of high priority and come up with the result that it would be great if the
system could buffer six or more messages. The authors propose a non-preemptive
system and analyze its properties using auto-meta. It is noted that the high-priority
nodes can preferentially transfer data with system deadlock-free and secure also.
260 N. Parmar et al.

Mishra et al. [5] propose a robot–human interaction with an experimental setup.

They use TurtleBot platform for the simulation and rtabmap_ros for mapping and
localization. They propose a Adaptive Monte Carlo Localization for navigation and
visualized using rviz. They show that robot can run successfully by navigating itself
to target even if robot is placed at any different locations.
Dauphin et al. [10] brief about the implementation, evaluation, and design of a
system for mini-bots. They use ROS 2 on top of RIOT and aim to reduce gap between
robotics and IoT. Their research work investigates the future of open-source robotic
platform with IoT to command mini-bots.
Reliability and Achievements of ROS 2 This section focuses on finding most
appropriate background of ROS 2 applications. In the implementation:
– We evaluate latency and throughput performance of ROS 2.
– We conduct the experiment on DDS and VPN, and compare them.
– We implement Real Time Publish–Subscribe (RTPS) communication standards,
under middleware ROS 2.
Returning to general OpenSSL library with FIPS module may refine target module’s
security and its security functions. It is emanated from research that VPN may be a
better choice for simple system architecture.

2.2 Data Distribution Services

Koksal and Tekinerdogan [11] propose a well-organized review paper. They have
done a large review of 468 papers. They introduce the study on DDS, because DDS
has a huge scope. They categorize DDS into 11 topics for a systematic study. Their
classification is shown below:
– Problems in DDS.
– Conduct, quantification, and progress in DDS.
– Execution of DDS protocol.
– DDS incorporation in WAN.
– DDS use in wireless network and mobile computing.
– Ability to communicate among different DDS vendors.
– Data uniformity in DDS.
– Security and robustness of DDS.
– Scaling-up of nodes in DDS.
– Certainty and reliability of DDS protocol.
– Integration of DDS with real-time systems (Table 1).
Yim et al. [12] propose a paper in the category of DDS in real-time system. They
deal with the interoperability and heterogeneity problems in data/services/events.
Following is the description of topic given by the authors on DDS as illustrated in
Fig. 2.
Syntactic Interoperability in Real-Time Systems, ROS 2 . . . 261

Table 1 Integration of real-time systems with DDS

Research papers Years Important points
S.A. Hadiwardoyo et al. 2018 Information passing between the train wagon and station
Adding more number of subscribers does not effect
Y. Park et al. 2015 High-level architecture for vehicles-to-everything communication using
OpenSlice DDS
Experiments using SUMO v0.22.0 for traffic simulation and OMNeT++ v4.6
for network simulation
It achieves interoperability and workability of HLA features
Research affix control functions of network QoS, network conduct, and
scalability from DDS features
H. Perez et al. 2016 A point of view to use DDS and hypervisor technologies in multi-core systems
is suggested
DDS design is consistent with ARINC-analogous segregated systems
They endorse through a dossier from the energy domain
P. Bellavista et al. 2013 It is the comparison of different DDS implementations (PrismTech and RTI)
We can infer that OpenSplice has superior execution for small data
RTI scale-up adeptly due to a better accomplishment message split-up and
memory management tasks
Z. Heng et al. 2017 Application of DDS, i.e., live virtual and constructive training used in US Army
It is the real-time integration concept presented and reliability test for DDS
DDS uses Interface Description Language (IDL) which makes it platform
independent
W. Zeyu et al. 2017 Designs the test system for DDS
It is the real time integration testing presented and reliability test for DDS
M. Takrouni et al. 2017 Designed a Simulink DDS blockset
It solves the problem of manually setting up the publisher–subscriber in DDS
They also explained the XML Script to understand functioning of blockset
A. Alaerjan et al. 2017 Practical conduct of DDS using Object Constraint Language (OCL)
It defines briefly distinct layers in the DDS, i.e., DLRL, DCPS, and RTPS and
conduct of DCPS
DCPS behaviors are explained for generating entities and publish–subscribe
data
T. Chen et al. 2018 They used ARINC 653 (Avionics Application Standard Software Interface) and
VxWorks boards
DDS was integrated into VxWorks
ARINC 653 and VxWorks were integrated via AFDX bus
Setup had a delay due to VxWorks and AFDX drivers which is noticeable
DDS integration inside ARINC 653 is a difficult task according to researchers
H. Yuefeng 2018 DCPS data transmission
Experiment was conducted in VxWorks5.5 environment
Transmission of data was steady and reliable
Conclusion: DDS can be used for real-time systems
A. Alaerjan et al. 2018 DDS in smart grid
Transmission of data was steady and reliable
Conclusion: DDS can be used for real-time systems
262 N. Parmar et al.

Fig. 2 Topic description file structure

2.3 Adaptive AUTOSAR

Adaptive AUTOSAR is a automotive software that is being used for automation of

driving, over-the-air updates, and car-to-everything architectures (Car-2-X). Here is
the literature review of Adaptive AUTOSAR. Furst et al. [13] describe Adaptive
AUTOSAR for automotive and its characteristics as well. Adaptive AUTOSAR is a
very well-suited platform for adapting and accepting the new challenges.
– It is very well suited for different software platforms.
– The mode of communication, i.e., service-oriented and signal-based communica-
tion, was inefficient for huge volumes of dynamic data.
Han et al. [1] describe TCP/IP analysis in Adaptive AUTOSAR. The average
number of ECUs in vehicles are increasing rapidly. In order to fulfill its needs,
Adaptive Platform is developed to enhance the performance of some of features of
TCP/IP. Address Resolution Protocol (ARP) cache entries should be updated based
only on ARP packets. When service is not provided by the peer at the requested
protocol or port, it should respond with RST (reset) rather than responding with port
unreachable.

3 Background Study

3.1 Robot Operating System (ROS)

In contrast to ROS 1, many proposed changes are adopted in ROS 2 as it inculcated

DDS in its architecture. ROS 2 does not require a master for publish–subscribe
communication, thus making it more secure.
Syntactic Interoperability in Real-Time Systems, ROS 2 . . . 263

Fig. 3 Architecture Of ROS

ROS Architecture. Figure 3 illustrates ROS architecture.

• Nodes: It is individualistic module of userland code.
• Topics: Identification of message/data between publisher–subscriber.
• Message: Message in ROS and ROS 2 is very similar to C structs. It sends complete
table for communication. So, we do not need different interfaces.
• Publisher: Sender of the data/client.
• Subscriber: Receiver of data/server.
• Package: ROS 2 software stack is composed of packages, where each package
provides important functionality.
ROS 2 is middleware for development of small/large robots. User code is built on
top of ROS client libraries. ROS provides rclcpp for C++ and rclpy for Python as the
basic client libraries for development. Support for other languages is being provided
by the community. Client libraries are responsible for the execution of a thread,
IPC, services, parameters, console, and logging. RCL and RMW implementation
are bound by a library. RMW is the ROS middleware implementation. RMW is
responsible for different DDS implementations, QoS, and discovery.
264 N. Parmar et al.

Fig. 4 ROS 2 file structure

ROS 2 File Structure. File structure of C++ package of ROS is shown in Fig. 4:
– build: Temporary files are generated during the build. It can be used to check the
built files.
– [Link]: This folder contains package dependencies and how to build them,
how to generate binaries and libraries, and where to store them.
– install: It contains files to be included in its subfolders, i.e., lib and share.
– include: Stores all the necessary files/directories for building and linked in cmake.
– log: Contains information about build.
– [Link]: It tells us about the type of build and dependencies.
– src: It contains the files developed by the ROS 2 developer.
File structure of Python packages in ROS 2 is very similar to C++ package, but only
difference is [Link] does not exist in Python package; instead, [Link] file
is used. [Link] performs similar function.

3.2 Data Distribution Services

DDS stack has three layers: Data Local Reconstruction Layer (DLRL), Data-Centric
Publish–Subscribe (DCPS), and Real-Time Publish–Subscribe (RTPS). DCPS is
compulsory for interaction of DDS components. DLRL is a discretionary layer.
RTPS supports interoperability between different DDS layers. DDS architecture is
depicted in Fig. 5. DCPS contains modules, and each of the modules has multiple
classes for its operations.
• Domain Module: It manages various DDS modules through domain ID differen-
tiation. It has the following classes.
Syntactic Interoperability in Real-Time Systems, ROS 2 . . . 265

Fig. 5 Architecture of DDS

– DomainParticipantListener (interface)
– DomainParticipant
– DomainParticipantFactory
• Publisher Module: It aids publication.
– Publisher
– Publication type specific classes
– PublisherListener (interface)
– DataWriterListener (interface)
• Subscriber Module: This module contains all the support for subscription.
– Subscriber
– Subscription type specific classes
– Subscription type specific classes
– DataSample
– SampleInfo (struct)
– SubscriberListener (interface)
– DataReaderListener (interface)
– ReadCondition
– QueryCondition
• Topic Description Module: It specifies topic objects and specifies quality of service
for them has the following classes.
– TopicDescription (abstract)
– Topic
– ContentFilteredTopic
– MultiTopic
– TopicListener (interface)
• Infrastructure Module: This module contains the abstract classes and interfaces,
which are used by the different modules. It assists interoperability between the
266 N. Parmar et al.

application and the Data Distribution Service. Interaction between DDS and appli-
cation may be event- or state-based. It contains the following classes.
– QosPolicy (abstract, struct)
– Listener (interface)
– Status (abstract, struct)
– WaitSet
– Condition
– GuardCondition
– StatusCondition
– Entity (abstract)
– DomainEntity (abstract).

3.3 Adaptive AUTOSAR

The software platform, AUTOSAR Adaptive Platform (AP), provides a software

stack for development of automotive and other ECUs. AP offers high computing
functions and communication and offers flexible software design based on software-
oriented architecture (SOA), e.g., to support software update over the air, distributed
architecture, hybrid and central fusion. Figure 6 shows the modules of AP which are
listed below.

Fig. 6 Physical architecture of Adaptive AUTOSAR

Syntactic Interoperability in Real-Time Systems, ROS 2 . . . 267

• AP uses Operating System Interface (OSI) rather than building its own OS. OSI
is POSIX compliable interface. Adaptive application (AA) shall use PSE51 as an
OSI. Scheduling is also defined under POSIX standard, and OS provides multi-
threading and multi-process support. AP provides “freedom of interferences”
among functional clusters and adaptive application. Multiple processes can run
together, and multiple instance of a single process can run in different address
spaces.
• Execution Manager (EM): It is responsible for start and shutdown of application.
It views all the functional clusters and an application as well. EM is responsible
for application life cycle.
• Communication Module (CM): It is responsible for IPC and application interac-
tion. There is no direct communication between adaptive applications.
• RESTful Communication: ara::com and ara::rest are used for communication
between AAs. It ensures interoperability with non-AUTOSAR clients, as an exam-
ple ara::rest service can transfer data with a adaptable client and vice versa.
• Persistency: It provides mechanism to store data and makes available Key-Value
Storage, File-Proxy Storage on the device, and the combination of two. Persistent
data is always private to each application.
• Time Synchronization: TS performs synchronization between adaptive applica-
tions, ECUs, and other entities.
• Update and Configuration Management : It handles all the update request. Its
function is same as to dpkg or YUM in LINUX.
• Diagnostics: It perceives ISO 14229-5 (UDSonIP). ISO 14229-5 is based on ISO
14229-1 (UDS) and ISO 13400-2 (DoIP).

4 Motivation

• Hussain et al. [4] propose a detailed study of various issues in IoT like scalabil-
ity, interoperability, heterogeneity, quality of service, and security. The proposed
research work focuses on heterogeneity.
• In automotive industry, “interoperability between ROS 2.0 and Adaptive
AUTOSAR will bring two heterogeneous softwares to communicate among them-
selves to increase automation, efficiency, and ease of development.”
• This research provides a lot of innovative ideas with DDS.
• It will be boon to the students and researchers as it will give them a direction
and quick start in world of autonomous vehicles. There is only a limited literature
available for the Adaptive AUTOSAR and ROS 2.
• This work will not only help in automotive but also make the us aware scope of
DDS.
• Adaptive AUTOSAR and ROS are having high cohesion and low coupling. AAs
are developed independently on different machines and OS. So, there is a scope
of syntactic interoperability if DDS implementation is carried in IPC of ROS and
adaptive.
268 N. Parmar et al.

Fig. 7 Case 1

Fig. 8 Case 2

5 Approach

After the ROS 2 is integrated with DDS and Adaptive AUTOSAR is also adopted
the DDS standard. Many papers in the literature propose the idea and feasibility of
syntactic interoperability between ROS and Adaptive AUTOSAR. Figures 7 and 8
give an idea of communication between ROS 2 and Adaptive AUTOSAR by con-
sidering two different cases with first on the same machine and second on different
machines, respectively.

5.1 Communication Management in Adaptive AUTOSAR:

Network Socket

AP can be depicted as follows in Fig. 9.

Layer named ara::com is responsible for AA communication. Service-oriented
communication, i.e., in addition to the operating system such as intra-/inter-machine
communication, is also carried out by com layer.
Illustration of application registry and discovery services in adaptive is shown in
Fig. 10.
Language binding is carried out by a source code generator that is provided by
the service interface definition (ARXML) file. It defines how service is translated
into available identifiers by implementing target language characteristics. ARXML
is the interface description and fed as input to tools for code generation. It is strongly
Syntactic Interoperability in Real-Time Systems, ROS 2 . . . 269

Fig. 9 Socket approach of Adaptive AUTOSAR

Fig. 10 Service registry and discovery of an application

typed, i.e., all arguments are pre-defined. Network binding in ara::com can be using
SOME/IP (default), IPC using DDS, or other transport layers.
Network binding in ara::com can be using SOME/IP (default), REST, DDS or
other Transport Layers.

5.2 ROS 2 Publish–Subscriber

The following steps may be followed for workflow in ROS 2

– Create the ROS node: Initialize the ROS client library you want to use and create
node using create_node() method.
– Publish it: After the contents and parameter of ROS topic is decided, we publish
it.
– rmw and rmw_implementation: ROS message is converted to DDS message and
published.
– Adaptive Platform understands the message and decodes it.
270 N. Parmar et al.

Fig. 11 Plug and socket approach for connection

5.3 The Connection

Figure 11 shows the view connection. We explain the connection using plug and
socket approach, where either of ROS or adaptive can act as plug/socket.

6 Results

We have conducted many experiments with the talker and listener scenario in our
system and observed the results on the same machine and different machines. The fol-
lowing results have been achieved. Figure 12 depicts the talker and listener instances
on a same machine.
We compare RTPS and UDP based on different parameters as shown in Fig. 13.
– Reliability: It is measured as overall consistency of a protocol. It is shown that
UDP is an unreliable protocol, whereas DDS offers abstraction on TCP/IP or UDP
and is reliable protocol.
– QoS Setting: It is observed that DDS offers Process Interface Management (PIM),
configuration, and interoperability. It can use both TCP and UDP and can have a
large set of QoS setting as well.
– Latency: It is observed that UDP has less latency over an unreliable network, but
if we want to offer reliability in the protocol itself, the latency of DDS is the same.
– Interoperability: The pub–sub concept itself was designed to solve the problems
of interoperability. By having different configurations, it can communicate.
– Complexity: DDS is easier to implement because of abstraction it offers.
Figure 14 shows talker and listener instances on different machines.
Syntactic Interoperability in Real-Time Systems, ROS 2 . . . 271

Fig. 12 Running instances of ROS on single machine

Fig. 13 Comparison of DDS (RTPS) and UDP

7 Conclusion and Future Work

– It is possible to connect two different software stacks using DDS.

– IPC implementation in both ROS 2 and Adaptive AUTOSAR needs to be imple-
mented for syntactic interoperability between ROS 2 and Adaptive AUTOSAR. It
is clear from the study of ARINC 650 [14, 15] that there is a need to have build
support DDS inside your software.
– DDS can be implemented on top of unreliable protocol like UDP, but it offers
reliability setup over it. It can use unicast or multicast, depending on the network.
272 N. Parmar et al.

Fig. 14 Running instances of ROS on different machines using Ethernet cable

– Interoperability is solved using DDS by using different mechanisms, QoS services,

shared space, etc. DDS is easier to implement due to abstraction. It also offers
implementation in many different programming languages.
– Our research gives full conceptual view on DDS, AP, and ROS 2. Even more
softwares can be integrated according to need of the future.
– DDS in Fog Computing: If we want to shift our computing from cloud to edge,
interoperability could be a major challenge addressed by DDS. Fog computing
is also another critical part of autonomous vehicles in remote areas, large farms
using autonomous vehicles in a slow connection.
– As the number of IoT devices are increasing day by day at a very fast rate, DDS
can solve such interoperability and heterogeneity problem.
– We feel that DDS will bring a revolution in IoT in the near future.

References

1. Han S et al (2016) On AUTOSAR TCP/IP performance in in-vehicle network environ-

ments, automotive networking and applications. IEEE. [Link]
1500167CM
2. Zeyu W, Jinsong Y, Wubin S (2017) Distributed test system based on publish/subscribe mid-
dleware. In: IEEE 13th international conference on electronic measurement & instruments
3. DDS Security, Object Management Group (OMG) Std. ptc/17-09-20, Rev. 1.1, September
2018. [Link]
4. Hussain MI. Internet of Things: challenges and research opportunities. Published in interna-
tional conference on dependable systems and networks workshops, vol 35, pp 123–126
5. Mishra R et al (2018) ROS based service robot platform. In: 2018 4th international conference
on control, automation and robotics. [Link]
6. Liu Y, Guan Y, Wang R, Zhang J, Li X (2018) Formal analysis and verification of DDS in ROS2.
In: 16th ACM/IEEE international conference on formal methods and models for system design
(MEMOCODE). IEEE. [Link]
Syntactic Interoperability in Real-Time Systems, ROS 2 . . . 273

7. Bellavista P, Corradi A, Foschini L, Pernafini A (2013) Data distribution service (DDS): a

performance comparison of open splice and RTI implementations. In: IEEE symposium on
computers and communications (ISCC). IEEE. [Link]
8. Rhoades BB et al. Design and development of a ROS enabled CAN based all-terrain vehicle
platform
9. Maruyama Y, Kato S, Azumi T (2016) Exploring performance of ROS2. ACM. [Link]
10.1145/2968478.2968502. ISBN 978-1-4503-4485-2/16/10
10. Dauphin L, Baccell E, Adjih C (2018) RIOT-ROS2: low-cost robots in IoT controlled via
information-centric networking. IFIP
11. Koksal O, Tekinerdogan B (2017) Obstacles in data distribution service middleware: a sys-
tematic review. Futur Gener Comput Syst ScienceDirect 68(2017):191–210. [Link]
10.1016/[Link].2016.09.020
12. Yim H-J, Seo D, Jung H, Back M-K, Kim I, Lee K-C (2017) Description and classification
for facilitating interoperability of heterogeneous data/events/services in the Internet of Things.
Science Direct. [Link]
13. Furst S et al (2016) AUTOSAR for connected and autonomous vehicles. In: 46th annual
IEEE/IFIP international conference on dependable systems and networks workshops. https://
[Link]/10.1109/DSN-W.2016.24
14. Perez H (2017) Handling heterogeneous partitioned systems through ARINC-653 and DDS.
Comput Stand Interf ScienceDirect. [Link]
15. Chen T, Hu X, Zhang G, Xiao J (2018) Implementation of data distribution service interface
based on ARINC653 system. In: 13th IEEE conference on industrial electronics and applica-
tions (ICIEA). [Link]
16. Kim J, Smereka JM, Cheung C, Nepal S, Grobler M (2018), Security and performance con-
siderations in ROS 2: a balancing act. Cornell University. arXiv:1809.09566
17. eProsima Fast RTPS Performance. eProsima, June 2018. [Link]
resources-all/performance/40-eprosima-fast-rtps-performance
18. Kim J et al (2018) Security and performance considerations in ROS 2: a balancing act
19. Vignesh UP (2017) ROS based stereo vision system for autonomous vehicle. In: IEEE interna-
tional conference on power, control, signals and instrumentation engineering (ICPCSI-2017).
[Link]
20. Auliya RS, Sheu RK, Liang D, Wang WJ (2018) IIoT testbed: a DDS-based emulation tool
for industrial IoT applications. Int Conf Sys Sci Eng (ICSSE). [Link]
2018.8520091
21. Alaerjan A, Kim DK, Kafaf DA (2017) Modeling functional behaviors of DDS. IEEE
22. Yuefeng H (2018) Study on data transmission of DCPS Publish–Subscribe model. In: 2018 2nd
IEEE advanced information management,communicates, electronic and automation control
conference (IMCEC 2018)
23. Shokrollahi S, Shams F (2017) Rich Device-Services (RDS): a service-oriented approach to
the Internet of Things (IoT). Wireless Pers Commun (2017) 97:3183–3201. [Link]
1007/s11277-017-4669-2
24. Heng Z, Jianguo H, Xiaoyuan Z, Hanqiang D (2017) Research on LVC real-time integra-
tion based on DDS. In: 2017 4th international conference on information science and control
engineering. [Link]
25. Alaerjan A, Kim DK, Ming H, Malik K (2018) Using DDS based on unified data model to
improve interoperability of smart grids. In: 2018 the 6th IEEE international conference on
smart energy grid engineering
26. Priyadarshi D, Behura A (2018) Analysis of different IoT protocols for heterogeneous devices
and cloud platform. In: International conference on communication and signal processing
27. Park Y, Min D (2015) Distributed traffic simulation using DDS-communication based HLA
for V2X. In: 2015 seventh international conference on ubiquitous and future networks. https://
[Link]/10.1109/ICUFN.2015.7182584
28. Takrouni M, Gdhaifi M, Hasnaoui A, Mejri I, Hasnaoui S (2017) Design and implementation of
a simulink DDS blockset and its integration to an active frame steering blockset conformed to
274 N. Parmar et al.

SAE ElectricVehicle. In: 2017 IEEE/ACS 14th international conference on computer systems
and applications
29. Hadiwardoyo SA, Gao L (2018) Integrating a middleware DDS application for safety purposes
in an underground railway environment. In: 2018 3rd international conference on computer
and communication systems
30. Lera FJR, Llamas CF, Guerrero AM, Olivera VM (2017) Cybersecurity of Robotics and
Autonomous Systems: Privacy and Safety. In: Dekoulis G (ed) IntechOpen Robotics—legal,
ethical and socioeconomic impacts, Chapter 5
31. Source code using ROS2 evaluations. [Link]
ardent-20180307
IoT-Enabled Water Quality Monitoring
System

G. Kanagaraj, T. Primya, K. Sashi Rekha, C. Vinothini and P. Anitha

Abstract In case of water quality monitoring, smart solutions are gaining more
importance with communication technology. By using a specific equipment, the
quality of water is tested on each attributes like minerals, temperature and so on.
This process consumes more time to complete for the given sample. This paper
contains two main activities. In application side, a detailed survey on recent work has
been carried out to perform smart water quality monitoring in terms of application,
communication technology used, and types of sensors employed. The next is by
using controller with inbuilt Internet connectivity module to monitor parameters such
as temperature and turbidity using low cost and less complex smart water quality
monitoring system. The system contains an appropriate webpage for enhancing the
user convenience on the deviation of water quality parameters.

Keywords Arduino · Ethernet shield · ThinkSpeak · Turbidity sensor

1 Introduction

Due to the enormous sources of pollutants, ensuring the safety of water is considered
as a major challenge. Water over exploitation of natural resources is the main cause
for it. Water pollution to a large extend is due to the rapid industrialization and
agricultural growth. Sometimes, the problem is aggravated due to the non-uniform
distribution of rainfall. To determine the quality of water, individual practices play a
major role [1].
The point and non-point sources of pollution tend to adversely affect the water
quality, which includes discharge from sewage, industrial discharge, agricultural
fields’ run-off and run-off from urban. The next level sources of water contamination
include droughts and floods and user’s lack of awareness and education. To maintain

G. Kanagaraj
Kumaraguru College of Technology, Coimbatore, India
T. Primya (B) · K. Sashi Rekha · C. Vinothini · P. Anitha
Dr. N.G.P. Institute of Technology, Coimbatore, India
e-mail: primyacse@[Link]
© Springer Nature Singapore Pte Ltd. 2020 275
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
276 G. Kanagaraj et al.

the quality of water resources, user involvement needed in the areas such as hygiene,
storage of critical elements, disposal of waste materials and environmental sanitation.
Diseases can spread through poor quality water, which may lead to death and also
socio-economic progress hampers [2].
In order to provide data, contents such as define real-time conditions and trend
happenings are taken into account at specified locations and ordered intervals. The
aim of water quality monitoring through online contains critical water quality mea-
surement parameters such as physical and chemical properties, microbial, to identify
parameters deviations and supply early warning hazards identification. The monitor-
ing system provides actual time investigation of data composed and suggests suitable
corrective measures [1].

2 Automated Water Quality Monitoring System

2.1 Using Raspberry PI

The use of raspberry pi along with Arduino kit helps in the storage of data in a Secure
digital card (SD card) and then inspected when the memory is full or at a periodic
time. The data sent from the sensors are collected from the Arduino and the raspberry
pi stores the data along with the time in a specified format for retrieval. This type of
storing the data and analysing it in a periodic interval are possible if the water quality
is measured for a small range or it is measured near to the analysis area.
Drawbacks
The raspberry pi can be used as a storage for the sensor data but there is no constant
maintenance of the values when the tanks are placed at a distance. And if there are
more than one storage places, SD card should be provided to all the places and the
collection of SD cards after memory is full or it should be done at periodic intervals.
This is a difficult task [1].

2.2 Digital LCD Display

The digital LCD form of display can be utilized for viewing the values that are read
from the sensors. This LCD display helps in the real-time monitoring of data that are
produced by the sensors. Any change in sensor value can be seen and the rectifying
steps can be taken within minutes. The LCD display is a module connected to the
Arduino. Sensors provide data to the Arduino and Arduino helps in viewing the data
in the LCD display. LCD display can be mounted on the central maintenance or the
place where there is a constant maintenance of the tanks.
Drawbacks
IoT-Enabled Water Quality Monitoring System 277

For a small project with one/two storage spaces, it is easy to manipulate the data
from the LCD display. But it cannot be maintained for a large scale like colleges as
there are more number of storage tanks. Thus, the usage of LCD display appears as
a major drawback [3].

3 IoT-Enabled Water Quality Monitoring System

Water is utilized for numerous activities such as utilization, gardening, and travel,
which may concern mainly on the water quality. The water quality monitoring
is essential, which includes numerous substance parameters like alkaline, redox
potential, conductivity, dissolved oxygen, ammonium and chloride ion amount.
The existing system will observe the water bodies, and it is agreed that the methods
used in laboratory are too unhurried to develop a running response and it does not
provide a plane of public health fortification in real time. Owing to the infinite increase
in output of the global industry, and the over-utilization of terrain and marine property,
the excellence of water available to public has been deteriorated deeply.
The overall decrease of water quality has internationally contributed with the
high use of fertilizers in farms and other chemicals in sectors like withdrawal and
production. Water is an important necessitate for human endurance and there must
be mechanisms position in place to powerfully analyse the quality of water that has
been offered for drinking in urban and municipality. The accessibility of superior
eminence water is overriding to prevent water-borne diseases as fit as recovering the
worth of life. The surface water monitoring improvement in network is an essential
element in the evaluation and fortification of water quality. A prototype is developed
for easy to mount expertise by which the diverse surface water value indicators can
be considered [4].
With the help of Ethernet shield, the details will be uploaded constantly in real
time from the Arduino board. The management and upload of this data to cloud and
the daily report will be published in the website specially designed for this project.

3.1 Architectural Design

Figure 1 shows the architectural design of IOT-enabled water quality monitoring

system which consists of the subsequent key components (Fig. 2).

3.2 Implementation

The implementation of quality monitoring is explained by the flow chart given in

Fig. 3.
278 G. Kanagaraj et al.

Fig. 1 Architecture design

Fig. 2 Architectural view

IoT-Enabled Water Quality Monitoring System 279

Fig. 3 Implementation
design

4 Components Involved

4.1 Data Retrieval

The data retrieval is a module to collect all the measurement data at a periodic interval
from the storage and sent it directly to the Arduino. This module can only collect
the data. It cannot do further processing from the data collected. The further process
with the data can be taken care of the Arduino board. The data retrieval is done using
different types. Here, we use the sensors to collect the data from the storage tanks.
The sensors are connected to the Arduino and programmed in a way that the entire
280 G. Kanagaraj et al.

sensor sense the environment at the time of filling of tanks and sent the sensed data
to the Arduino. The model diagram for the data retrieval is shown Figs. 4, 5 and 6.

Fig. 4 Arduino connection to temperature sensor

Fig. 5 Data retrieval from temperature sensor

IoT-Enabled Water Quality Monitoring System 281

Fig. 6 Data retrieval from other sensors

This is the connection of the Arduino with the temperature sensor that can send
the temperature of the water at the particular time. The measurements are sent to the
Arduino and it is then uploaded to the cloud.

4.2 Ethernet Shield Connectivity

To connect the Ethernet shield with Arduino hardware and PC, the Ethernet shield is
placed firmly on the Arduino hardware. An Ethernet shield stacked on the Arduino
hardware is shown. The Ethernet shield can be connected to a network router, or to
computer, using an RJ45 cable (Figs. 7 and 8).
An Arduino board is linked with Internet using Arduino Ethernet Shield 2. The
WiZnet W5500 provides a complex stack consist of mutual TCP and UDP. It supports
up to eight concurrent socket relations. The Ethernet library is second-hand to inscribe
sketches that join to the Internet by means of the shield. The Ethernet Shield 2
connects to an Arduino board with long wire-wrap headers throughout the shield. It
keeps the pin arrangement intact and allows a different shield to be stacked on the
peak of it.

4.3 ThinkSpeak–Cloud Connectivity

The cloud architecture used to store the data here is ThinkSpeak. The collected
information from the sensors is then send to the Arduino where the records are sent
to cloud services with the aid of Ethernet shield.
282 G. Kanagaraj et al.

Fig. 7 Ethernet shield connectivity

Fig. 8 Arduino with Ethernet shield

The statistics is maintained in the cloud and the periodic information is sent to
the management team in that particular area who maintains the water quality system.
Any deviation in measurements is intimated to the team in order to take quick actions
(Figs. 9 and 10).
ThinkSpeak is an IoT analytics podium service that allows aggregating, visualiz-
ing and analysing survive statistics streams in the cloud. It gives immediate visual-
izations of facts posted by the devices to ThingSpeak. To perform MATLAB code in
ThingSpeak, carry out online examination and dispensation of the data as it comes
inside. ThingSpeak is worn for prototyping and evidence of idea IoT systems that
need analytics.

ThingSpeak Key Features

IoT-Enabled Water Quality Monitoring System 283

Fig. 9 Cloud inculcated with IOT

Fig. 10 ThinkSpeak with IOT

The key capabilities of ThingSpeak include the capability to:

• Simply configure devices to convey data to ThingSpeak by means of trendy IoT
protocols.
• Envisage the sensor data in real time.
• Collective data on-demand beginning third-party sources.
• Use the power of MATLAB to make sense of your IoT data.
• Scamper IoT analytics repeatedly based on schedules or actions.
• Example and construct IoT systems without location up servers or initial Web
software.

4.4 Website Creation

The website is created to provide easy maintenance and to get the real-time values
updated as a graph so that the management can easily view the current situation
happening in the place where this project is set. The site is done offline and site
284 G. Kanagaraj et al.

Fig. 11 Water level sensor

is hosted in the heroku hosting platform. Heroku, a container-based Platform as a

Service (PaaS) in cloud.
The developers use Heroku to organize, administer and extent modern applica-
tions. Its platform is graceful, supple and effortless to use, offering developers the sim-
plest path to getting their apps to market. It is completely managed give developers the
liberty to centre on their middle creation lacking the disruption of maintaining servers,
hardware, or communications. The Heroku understanding gives forces, equipment,
workflows and linguist sustain, all considered to improve developer efficiency.
In this heroku hosting the heroku hosting customer for windows have to be down-
loaded from the executive site. Then once setting up, the command line can be worn
for easy hosting after login in the command prompt.

5 Hardware Requirements

5.1 Water Level Sensor

Without any affecting parts, the liquid plane sensor calculates the fluid level in tanks,
reservoirs and in its surroundings. The sensing probe element comprises of an extraor-
dinary wire cable which is accomplished for truthfully sensing the exterior level of
almost any fluid, together with water, salt water and oils. Sensor element is electri-
cally insulated and secluded from the water into which it is inserted, and will not
deteriorate in the overload of time. Contrasting other sensors, the quantity range is
changeable from a few cm to a few metres. The appraisal is reported back as an ana-
logue voltage ranging from 0 to 3 V where 0 V represent the non-immersed sensor,
and 3 V represents the highest water level (Fig. 11 and 12).

5.2 Temperature Sensor

It includes a humidity sensing component, a NTC temperature sensor and an IC on

the reverse side of the sensor. The humidity sensing component is used to measure
humidity, which has two electrodes with dampness holding substrate among them.
IoT-Enabled Water Quality Monitoring System 285

Fig. 12 Temperature sensor

Fig. 13 Temperature graph

The conductivity of the substrate varies or the resistance among these electrodes
varies as the humidity changes. The variation in resistance is calculated and processed
by the IC which makes it prepared to be converted by a microcontroller.
To assess temperature, NTC temperature sensor or a thermistor is used. A thermis-
tor is a changeable resistor that varies its resistance with variation of the temperature.
This type of sensors is finished by sintering of semi conductive equipment such as
ceramics or polymers in order to supply overweight changes in the resistance with
just tiny changes in warmth. The term “NTC” means “negative temperature coef-
ficient”, which income that the confrontation decreases with augment of the high
temperature (Fig. 13).

5.3 Turbidity Sensor

In a fluid, turbidity is the quantitative enumerate of pending particles. It may be soil

in water or chocolate flake in favourite milk shake. While chocolate is somewhat
so wanted in our drinks, soil particles are completely undesired. There are several
manufacturing and family circle solutions that make use of water in some or other
way keeping aside the moveable purposes. A car use water to unsoil the windshield,
a power plant desires it to cool the reactors, washing machines and dish washers
286 G. Kanagaraj et al.

Fig. 14 Turbidity sensor

Fig. 15 Arduino UNO

depends on water like fish. How does this equipment get to be familiar with concern-
ing the turbidity? The nature’s evolutionary contribution of senses to discover soil
in the water, but what about washing machines? No eyes to see, no tongue to taste,
no skin to feel but just a plastic body with some buttons and motor inside. How does
it so elegant to work as per soil deferment? Turbidity sensor, which the length of
with a micro controller unit, takes care of turbidity capacity. It is craft with synthetic
and some metal-alloy traces, turbidity sensor uses light to put across in sequence in
relation to turbidity in water (Figs. 14 and 15).

5.4 Arduino Uno

Arduino Uno is a microcontroller board based on the ATmega328P. It includes 14 dig-

ital input/output pins, 6 analogue inputs, a 16 MHz quartz crystal, a USB connection,
a power jack, an ICSP header and a reset button. The ATmega328 on the Arduino Uno
comes pre-programmed with a boot loader that allows uploading innovative code to
it devoid of the use of an external hardware programmer. It communicates during the
original STK500 procedure. The board can be power-driven via the USB connection
or with an outside power supply. The power source is particular by design. It can
activate on an external contribute from 6 to 20 volts. If it is complete with less than
IoT-Enabled Water Quality Monitoring System 287

7 V, the 5 V pin may used to supply less than 5 volts and the board may become
unbalanced. The use of more than 12 V, the voltage regulator may overheat and
injure the board. The suggested range is 7 to 12 V. Arduino/Genuino Uno contains
quantity of facilities for communicating with a computer, another Arduino/Genuino
board or other microcontrollers. The ATmega328, provides UART TTL (5 V) serial
communication, is accessible on digital pins 0 (RX) and 1 (TX).

5.5 Ethernet Shield

Arduino board is used to connect the Arduino Ethernet Shield 2 with the Internet.
The WiZnet W5500 provides a network stack capable of together TCP and UDP.
It chains up to eight concurrent socket connections. The Ethernet library is used to
inscribe sketches that attach to the Internet by means of the shield (Fig. 16).
The Ethernet Shield 2 connects to an Arduino board with long wire-wrap headers
from end to end the shield. It keeps the pin layout unbroken and allows a different
shield to be stacked on apex of it. The current modification of the panel exposes the
1.0 pinout on rev 3 of the Arduino UNO Board. It has a paradigm RJ-45 connection,
with an incorporated line transformer and Power greater than Ethernet enabled. It
gives an onboard micro-SD card slot, it can be used to store files for helping over
the network and it is well-suited with the Arduino Uno and Mega using the Ethernet
library.

Fig. 16 Ethernet shield

288 G. Kanagaraj et al.

6 Software Requirements

6.1 Arduino IDE

For a compiler, a program for Arduino may be printed in any programming language
that produces binary machine code for the objective processor. It gives a development
surroundings for their microcontrollers, AVR Studio and the newer Atmel Studio.

6.2 HTML

To create Web pages and Web applications, Hypertext Markup Language (HTML)
is the standard markup language The Web browsers, obtain HTML documents from
a Web server or from local storage space and provide the permit into multimedia
Web pages. Hypertext Markup Language defines the constitution of a Web page
semantically and initially included cues for the exterior of the document. HTML may
embed programs written in a scripting language such as JavaScript which affects the
performance and substance of Web pages. Insertion of CSS defines the appear and
design of content.

6.3 CSS

Cascading Style Sheets, referred to as CSS, is a trouble-free design language intended

to make things easier the process of making Web pages reasonable. CSS controls the
colour of the text, the style of fonts, the spacing among paragraphs and it explains
how columns are sized and laid out.

6.4 Bootstrap

For faster and easier Web development Bootstrap is an authoritative front–end frame-
work. It too comprises HTML- and CSS-based design templates for familiar user
interface components like Dropdowns, Alerts, Modals Typography, Forms, Buttons,
Tables, Accordion, Carousel Navigations, Tabs, and many other as well as optional
JavaScript extensions.
IoT-Enabled Water Quality Monitoring System 289

7 Summary

The result of the proposed project lies in developing a water monitoring IOT set-up
with a website linked to the set-up to provide a real-time management of the quality
of water. The scope of the project lies in integrating the hardware sensors using
Internet of things and the software part of the website. It also helps all the people
in management team to know the current status through the website. The use of the
website hosted is very helpful in monitoring the quality of water and ease of use.

8 Snapshots

See Figs. 17, 18 and 19.

Fig. 17 Data retrieval

290 G. Kanagaraj et al.

Fig. 18 Website creation

Fig. 19 ThingSpeak water monitoring

IoT-Enabled Water Quality Monitoring System 291

References

1. Purohit A, Gokhale U (2014) Real time water quality measurement system based on GSM.
IOSR (IOSR-JECE) 9(3), Ver. V
2. Lom M, Pribyl O, Svitek M (2016) Industry 4.0 as a part of smart cities. 978-1-4673-9948-7/16
© IEEE
3. Kedia N (2015) Water quality monitoring for rural areas—a sensor cloud based economical
project’ by in (NGCT-2015) 4–5. 978-1-4673-6809-4/15/$31.00 ©2015 IEEE
4. Daigavane VV, Gaikwad MA (2017) Advances in wireless and mobile communications, vol
10, no 5, pp 1107–1116. ISSN 0973-6972
5. Bhatt J, Patoliya J (2013) Real time water quality monitoring system. 978-1-47990792-
2/13/$31.00 © IEEE
6. Kartakis S, Yu W, Akhavan R, McCann JA (2015) Adaptive edge analytics for distributed
networked control of water systems. In: IEEE first international conference on internet-of-things
design and implementation (IoTDI)
7. Daigavane VV (2017) IoT based water quality monitoring system. Int J Ind Electron Electr
Eng 4. ISSN: 2347-6982
8. Sun Z, Li CH, Bisdikian C, Branch JW, Yang B (2013) QOI-aware energy management in IoT
sensory environments, pp 1–9. [Link]
9. Karthik Kumar R, Chandra Mohan M, Vengateshapandiyan S, Mathan Kumar M, Eswaran R
(2014) Solar based advanced water quality monitoring system using wireless sensor network.
Int J Sci Eng Technol Res (IJSETR) 3(3):385–389
10. Cheng P, Wang X-L (2010) The design and implementation of remote-sensing water quality
monitoring system based on SPOT-5. In: Second IITA international conference on geoscience
and remote sensing, pp 6–10
11. Nagarajan R, Dhanasekaran R (2013) Implementation of wireless data transmission in moni-
toring and control. In: International conference on communication and signal processing, India,
April, pp 83–87
12. Primya T (2018) Hybrid algorithm based user authentication and data security using image
comparison and OTP. Asia Pac J Res I(LXXXVII)
Energy-Efficient Routing-Based
Clustering Approaches and Sleep
Scheduling Algorithm for Network
Lifetime Maximization in Sensor
Network: A Survey

Rajiv R. Bhandari and K. Rajasekhar

Abstract Along with number of problems associated with WSNs, one of the major
issue we chose to study is their clustering topology and energy utilization techniques.
Both these parameters are very much responsible in determining the life of nodes,
quality of service, delay in data transmission, etc. So, it is very important to examine
the cluster-based routing protocols and energy optimization protocols. The cluster
routing is based on the selection of cluster head nodes for data transmission, and the
parameters which are used for the making of an optimized network mainly depend
on number of nodes, position of base station, and the network size. So, a well-settled
cluster is tried to be designed with the study of various algorithms early proposed. And
for energy optimization of network, we studied “sleep/wake-up” algorithm, which
makes the nodes sleep during its ideal mode and wake it up when data transmission is
to be done. With this sleep/wake-up algorithm important point to be kept an eye is the
delay during this shifting between sleep and wake up or making sure that the selected
path should not be engaged as well as it must be shortest path from base station. So,
based on these two protocols, various papers are studied, and the comparative result
of this protocol in various scenarios is carried out.

Keywords Routing · Clustering · Wireless sensor network · Sleep/wake up

1 Introduction

In WSN, network is formed by sensor nodes that are aligned arbitrarily or in immobile
manner according to the network application [1]. The typical working of WSN is
that the sensor nodes sense data and transmit it to neighbor node or to sink node
within network or outside the network which totally depends upon the application.
Since this sensor nodes are running on the battery, and hence, due to this limitation
of energy, it must be optimized in order to give a better durability of these sensors. So

R. R. Bhandari (B) · K. Rajasekhar

Department of Computer Science Engineering, Koneru Lakshmaiah Education Foundation Green
Fields, Guntur District, Vaddeswaram, Andhra Pradesh 522502, India
e-mail: rajivrbhandari@[Link]

© Springer Nature Singapore Pte Ltd. 2020 293

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
294 R. R. Bhandari and K. Rajasekhar

sensor nodes are designed with consideration parameters like tiny node size, minimal
costing, etc. These are considered in order to improve the efficiency of sensors and
also make the network fast and steady. So, to overcome such network issues we
come with the comparative study of many algorithms so as to make a fully optimized
network which is smart as well as responsive when needed. For this, we are mainly
looking forward to two types of protocols one is cluster-based routing protocol and
another one is sleep/wake-up protocol.
In routing based cluster protocol most of the time depends on network structure
and can be roughly divided into two categories: hierarchical and uniform routing. In
a uniform topology, all nodes accomplish same jobs and have same functionalities of
every node within network. In this topology, data communication is achieved in hop-
by-hop, flooding, and gossiping manner. In a hierarchical topology, different nodes
performed different tasks and are well organized in cluster according to specific
requirements or metrics. Generally, there is cluster head along with other member
ordinary nodes, and all are arranged in hierarchical levels. The communication and
data processing are carried out by a monitoring node called as cluster head node
(CH), and all other nodes are called as member nodes and are used for sensing the
information. Among many routing protocols which are used in WSNs, low-energy
adaptive clustering hierarchy (LEACH) is the classic one. Here, we are making
comparative study of various such routing protocols. And we have also made a study
based on sleep/wake-up protocols too, which aims to minimize idle processing time
[2]. In order to study sleep/wake-up protocols, adjustments of sleep and awake time
for sensors during particular period are important to study. Basically, during ideal
state of node, the energy supplied is wasted as during this ideal time it won’t transmit
any data. Hence, the motto of sleep/wake-up scheduling is to minimize or make the
adjustment in wake time of sensor. These methodologies for wake-up methods are
classified into three types: (1) on-demand, (2) synchronous, and (3) asynchronous. So,
the comparative study of various algorithms associated with the energy optimization
and time scheduling is also done here.

1.1 Clustering and Routing in WSN

The basic structure of standard cluster consists of number of sensors (for data cap-
turing and analysis) and transceiver (to transmit data through gateway node), power
source (mainly batteries), sink node (to sense data and communicate with base sta-
tion), and many other parts. Figure 1 shows a simple cluster-based network. In
intra-cluster, the communication is done between a node within a cluster, but in
inter-cluster, communication is carried out between one cluster head with other and
with the sink. This is to be done so as to save significant energy required for all
the nodes. With the use of these techniques, we can make communication between
clusters so that all the nodes need not to be active all the time and also the range
of particular base station is increased as nodes from one cluster can communicate
with other node within different cluster. This technique is very useful in making an
Energy-Efficient Routing-Based Clustering Approaches … 295

Fig. 1 Cluster and routing

system

efficient WSN routing. Moreover, this technique allows not only reduction in routing
tables calculations for both CH selections as well as for member nodes selection but
also making the bandwidth reusability possible.
Here, in this figure, we can see the inter-cluster and intra-cluster communication
is done. It makes a system energy efficient as we can see we do not need all this node
to be awakened, and the range of particular cluster is extended.
Calculation for the cluster head selection depends on its score, as node with higher
score has priority to be elected as CH and other nodes join this CH in its proximity.
The calculation of score is done by the following Eq. (1).


number of neighbours
1
score(v) ← (1)
i=1
dist2 (v, i)

In Eq. (1), the score of node v is calculated. Basically, the selection depends
up to two metrics-node degree (defines number of neighbors in cluster range) and
node centrality (centrality of a node among its neighbor). Cluster head formation is
dependent on the value of this node v calculated from the neighboring node, and if
this distance is small, ultimately, the value of score(v) is increased and this value also
depends on the number of nodes (proportionally increases). So, we can say node v
is more likely to be CH and creates the intense network. For applications, creating
more closely compacted clusters is very supportive to group nodes in close locality.
The basic steps involved in cluster routing are shown in Fig. 2.
Figure 2 shows the complete procedure of the clustering process algorithm in
which the selection of CH is calculated by using the following equation:
 
E residual
CHprob = MAX(Cprob ∗ , pmin ) (2)
E max

where E residual is current energy of node and E max is maximum energy. It is to

note that the value of CHprob should fall below pmin . After this, the while loop is
iterated (from Fig. 2). Up to step 8, the node decision took place means whether
the node will become a CH or will join any other CH. Next iteration is performed
296 R. R. Bhandari and K. Rajasekhar

Fig. 2 Clustering process

after collecting the coded and doubled value of CHprob from pervious iteration level.
When the CHprevious reaches value 1, the respective node will terminate the loop.
So, termination of node with higher energy level will be done prior to other nodes
with lower energy. After successful execution of 17–20 steps, member nodes having
lower energy level will join the deterministic CH. But, it is to be noted that during
this complete cluster formation, every deterministic CH or candidate CH can send
CH_msg only once. Hence, like this, we can form the more efficient routing protocol
for WSN.

1.2 Sleep/Wake-up Algorithm

Another protocol is sleep/wake-up protocol, which is mainly used to make system

energy efficient. In this protocol, the nodes are made sleep while in its ideal mode,
means it consumes very less amount of energy. In order to make system energy
efficient, during this ideal mode, the node won’t transfer any data but wakes up
whenever needed with very low response time. The sleep/wake-up algorithm mainly
depends upon the duty cycle. Figure 3 shows the duty cycle working of node.
In this figure, we can see how the sleep and wake-up switching is done. The packet
transmits and receive is done whenever required after listen and in rest of the time, the
node is in ideal state or sleeping state. Energy in MAC-based sleep/wake-up protocol
is determined by its sleep time. In time slot TR , for some period, time interval is given
Energy-Efficient Routing-Based Clustering Approaches … 297

Fig. 3 Duty cycle of node

as (1 − waketime) × TR . A new node will be in sleeping phase for entire time length,
and during this interval, there will be no transmission of data taking place. The time
interval for this protocol is given as:

E(0)TR
t= ×d
T
where T stands for frame duration and d is for duty cycle.
In this system, energy saved by each node can be calculated as:
 2×n h 
n g −1
E(0) × TR × d × Pidle × ng
E save =
T
where Pidle represents power consumption of node during its ideal state. And the
power of “n” number of nodes can be easily calculated by multiplying the above
energy-saving equation with this number of nodes, i.e., “n” [3].
Based on these two protocols, we have studied number of papers which have done
research on same topics.

2 Literature Survey

2.1 Worked Based on Clustering Protocols

This paper presents a study of WSN clustering protocols in order to make more reli-
able and optimized network. Various routing indoor and outdoor network protocols
are presented in below section.
Amjad et al. [4], in that paper, presented a quality of service-based routing method
for heterogeneously clustered WSNs which is done by using nodes with different level
of energies and also having different initial energy. So, minimal energy is the point to
be calculated in order to achieve the energy optimization of network. The simulated
298 R. R. Bhandari and K. Rajasekhar

results show improvement in terms of network life, consistency, throughput, and

delay minimization.
Kang et al. [5], in that paper, presented a distributed delay-efficient scheme named
DEDAS-D to address the MLAS problem in duty-cycled WSNs. DEDAS-D cov-
ers two new algorithms for data accumulation, tree construction, and scheduling.
This method while addressing MLAS problem gives better simulation results like
reduction in data aggregation, and delay of DEDAS-D is reduced by 50.95%.
Zhou et al. [6], in that paper, proposed a new clustering protocol in which the
relay nodes are used to share the load of the cluster heads, and for energy optimiza-
tion, they proposed an improved PSO algorithm which improves the efficiency of
cluster structure for longer distance coverage. The improvement in cluster energy
management is successfully shown by simulation results.
Wang et al. [7], in that paper, presented a load-balancing and routing protocol
which is a multilayered and multihop routing protocol. Although TLCBMH have
more message and protocol complexity than RDCA, but still, they both have similar
clustering power. In this TLCBMH-based process, the base station access is done via
multihop frequency and it makes the system robust. The result comparison of both
these techniques shows that the TLCBMH provides better service as it does not lose
any clustering ability.
Deepa et al. [1], in that study, presented a routing algorithm analysis with or
without data aggregations called as hybrid hierarchal cluster-based secure (HHCS)
routing algorithm. The method provides good level of nodal level security as it con-
sists of both group key and pair wise key. Also, it reduces traffic of the network using
data aggregation method. The comparative result with LEACH algorithm showed
that this HHCS algorithm is more secure as well as more energy efficient.
Sreevidya et al. [8], in this paper, suggested a new energy monitoring and opti-
mization technique for cluster-based routing protocol. In this algorithm, mainly, the
energy in data transfer is optimized and end-to-end delay is also optimized, which
increase overall lifespan of network. The comparative results with that of AODV
protocol show that the life of nodes is enhanced after implementation of this new
protocol.
Gnanambigai et al. [9], in this paper, suggested a new hybrid algorithm in order to
reduce the energy of system and improve the durability of system and they called this
system as QBLEACH. This algorithm limits the usage of number of active nodes
for transmission of data which is done in order to reduce the energy required to
transmit the data. The simulation results precisely describe the energy required to
transfer data, and accordingly, the distribution is done. Depending upon this energy,
the distribution path and the number of nodes selection are done which ultimately
improve the life of sensor as well as the better energy optimization is observed.
Behera et al. [10], in this paper, defined a best-suited cluster for IoT applications.
This algorithm works in the selection of shortest path for data transmission and also
selects minimum distance between the active cluster heads. This method gives very
efficient results by optimizing lot of energy while transferring data to base station
or vice versa. They also proved that the results obtained from this protocol are quite
better than that of HEED, PEGASIS, HEER, and TEEN.
Energy-Efficient Routing-Based Clustering Approaches … 299

Xiao et al. [11], in this paper, introduced the 6LoWPAN hierarchical routing
protocol which is based on a switched hierarchical structure. In this protocol, the
transport path is merged with the IP address which stores very small amount of
routing information. In this protocol, the data can be transferred from any node
within cluster as well as outside the cluster to another node of different cluster even
cluster in like a node. Another benefit of this protocol is that even if any node gets
fail, then the network failure will not affect as data will be passed through another
node. The compared result with the traditional RPL protocol shows that the HCPR
has faster efficiency of transmission and has less storage of routing information.
Shaha et al. [12], in this paper, defined the new routing protocol for optimizing
the routing path for transmission of data from cluster head to respective nodes and
finally to sink. And if the node is in direct contact with sink, then the information
will be directly sent to sink itself. They used shortest path algorithm to define the
most optimized path and faster delivery of packets. They also grouped the nearby
nodes in such a way that all these nodes are combined to form a greater network.
Acharjee et al. [13], in this paper, suggested a modified hierarchical cluster-based
routing protocol which works in reducing the load on the mesh point portal and group
heads. In this algorithm, they introduced helping nods for cluster head, means in case,
if the cluster head fails, then this assistant cluster head will take the responsibility
of the defined CH node. This algorithm is mainly useful for large area network and
hence is more independent and reliable and also gives more throughput than older
network. The result of this algorithm has much better throughput for larger network,
but for less number of nodes, packet delivery ratio is less than that of HWMP, but
for larger number of nodes, it provides lower PDR than BATMAN algorithm.
Misra et al. [14], in that paper, surveyed the clustering-based algorithms based
on the state of art. Comparative study based on energy competence, cluster stabil-
ity/reliability, and load balancing is carried out with respect to this proposed algorithm
and other algorithms. This algorithm serves and educates in terms of clustering tech-
niques which is a concern for both military and civil applications. This algorithm is
utilized for various applications like security surveillance, environment monitoring,
and healthcare system.
Lenka et al. [15], in this paper, suggested a cluster-based rendezvous routing pro-
tocol. With respect to this algorithm, the clusters are structured within the rendezvous
area in order to achieve scalability and to maintain network load. To find the sink
position, not only cluster heads are part of communication, but also nodes are not
involved in sink selection. But, in other protocols like LBDD, ring routing, railroad
and rendezvous routing protocol, all the nodes inside the virtual infrastructure are
involved in transmitting the information from source to sink and vice versa. But, in
this proposed strategy, only cluster heads establish the efficient route to send the data
to the sink.
Pant et al. [16], in that paper, proposed a routing algorithm which is based on
multihop routing technique and uses an EEBCDA method. In this work, sensor nodes
along the network are divided into unequal grid as on EEBCDA and this division is
done in such a way that grid which is away from the sink has a larger size and has
more number of sensor nodes within this network. The formation of grids and the
300 R. R. Bhandari and K. Rajasekhar

structure of this grid are well elaborated in this framework, also, the path for the hop
is defined, and all these works improve the stability, coverage, and overall lifetime
and efficiency of network.
Based on the above study, the following algorithms are compared with each other
as given in Table 1.

2.2 Worked Based on Sleep/Wake-up Protocols

The study of sleep/wake-up protocol is very important as it deals with the energy
optimization of network node. It is important to study because many times the node
may be located at remote geography, so many times it is not possible to keep a watch
on supply, and network may get interrupted due to power cutoff. So, to avoid such
issues, sleep/wake-up protocol is an healthy approach to extend the life of power
source by making node sleep while it is in ideal state and wake it up when data is to
be transmitted. Here is some part of the study done based on this protocol.
Ye et al. [2], in that paper, suggested a self-adaptive sleep/wake-up scheduling
algorithm. In prior works, the duty cycle technique was used, which fails to deliver
the expected result in terms of both packet delivery wait and also energy optimization.
But, the suggested system is fast enough and energy efficient too as it is not based on
duty cycling. The proposed technique is corroboration for learning-based technique,
in which each node decides its individual role means weather it will sleep, listen, or
transmit data for particular time slot and in distributed manner during each time slot
in a decentralized way. Hence, this paper discourses a good knowledge regarding
sleep/wake-up algorithm, and the results lead to energy efficiency of system as well.
Panahi et al. [17], in that paper, worked on the power consumption of base stations
(BSs) by the use of sleep/wake-up algorithm. For base stations (BSs) of a hetero-
geneous network (HetNet), energy-proficient sleep/wake-up protocol is suggested
using a fuzzy Q-learning (FQL). This system works by monitoring the BSs, means
the less used BSs will get switched off in accordance with the local traffic profile,
required area coverage, and cell EE, and all this leads to energy efficiency of system
as less number of nodes are active. Their result shows much improvement in energy
consumption.
Kovásznai et al. [18], in that paper, proposed a satisfiability modulo theories
(SMT) validation for WSNs, which is used in and is based on sleep/wake-up schedul-
ing algorithm for solving various network issues. This method can only be used with
optimization modulo theories (OMT) which is another WSN sleep/wake-up schedul-
ing. They addressed the energy optimization issues by using graph standards and
examined their associations with the degree of challenge for OMT solvers.
Panahi et al. [19], in their paper, proposed a methodology for energy conservation
in WSN. They used device-to-device (D2D)-based communication in order to enlarge
the network coverage area so that the area of which the BS is not in working condition
also gets network service. An energy optimization is also done with the help of this
methodology.
Table 1 Comparison of routing protocols
S. No. Name Author Algorithms/techniques used Result
1 For wireless sensor networks, Muh. Amjad Energy-efficient routing protocol Simulation results gives improvement in
QoS-aware, and heterogeneously many aspects like network life,
clustered routing Protocol constancy, efficiency, and minimal
end-to-end stay
2 A distributed delay-efficient data B. Kang Aggregation tree construction algorithm, The proposed system solves MLAs
aggregation scheduling for duty-cycled fast distributed aggregation scheduling problem and gives better result as
WSNs algorithm compared with other algorithms
3 Clustering hierarchy protocol in wireless Y. Zhou Energy-efficient and life-extending The simulation result of this protocol
sensor networks using an improved PSO routing algorithm gives better energy efficient results
algorithm
4 A multilayer Layer Clustering-based K. Wang Establishment of cluster protocol Simulation shows TLCBMH improves
multihoping routing protocol the WSNs lifetime
5 HHCS: hybrid hierarchical cluster-based C. Deepa Hybrid hierarchical cluster-based secure HHCS algorithm gives more energy
secure routing protocol for wireless routing (HHCS) protocol and LEACH efficient and secured results than LEACH
Energy-Efficient Routing-Based Clustering Approaches …

sensor networks protocol algorithm

6 Enhanced energy-optimized Sreevidya B CBRP Results show overall improvement in
cluster-based on-demand routing terms of energy efficiency, better
protocol for wireless sensor networks throughput, and reduction in end-to-end
delay
7 An improved hierarchical cluster-based T. Acharjee Hierarchical routing protocol For small number of nodes, this protocol
routing approach for wireless mesh gives lower delay as compared with
network BATMAN protocol
8 Cluster-based rendezvous routing R. K. Lenka Rendezvous routing protocol With 1 hopping communication, this
protocol for WSN method consumes less energy and also
reduces delay
301
302 R. R. Bhandari and K. Rajasekhar

Lin et al. [20], in this paper, elaborated the use of body area networks (BANs)
which enable the data exchange and monitoring between wearable/implanted
devices. Due to lack of channel fluctuation consideration in many previous tech-
niques, the improper information is inefficient in a BAN. Along with this, channel-
aware polling-based MAC protocol CPMAC is also used which heals the network
failure or loses of packets issues. These techniques trigger sensors to spread or receive
data whenever the channel is sufficiently strong to promise fast, steady, and more
unfailing transmission. This is really very important problem-solving techniques
having very efficient results.
Latif et al. [3], in their paper, proposed a communication protocol for small area
network like home. This method targets to solve issues like energy wastage during the
idle heeding, crashing situation, and overhearing processes. The suggested algorithm
performs various tasks like reducing the energy of each node, eases communication
for stable traffic, and provides adaptive regulated load for fluctuating traffic load.
The results based on this MAC protocol give better results as compared with other
older techniques.
Alhalaf et al. [21], in that paper, presented an energy cross-layer design for WSN
which they named “green task-based sensing” (gTBS) scheme. In the suggested gTBS
techniques, they blend the sleep and wake-up techniques with power adjustment
which active or evoke the active nodes. The use of gradient-oriented unicast technique
helps to get over the synchronization issue of data transmission, diminishes network
traffic issues, and reduces the whole general energy utilization of the network. The
result output shows the improvement in network energy.
Vaiyshnavi et al. [22], in their proposed paper, specified a scheduling method
called (LECSA) load and energy consumption-based scheduling algorithm. In this
algorithm, the cluster heads have the power to decide the path for data transmission,
means based on the certain calculations, the CH will derive the nearest active node
which transfers the data. The dynamic selection of nearest node is calculated by using
an energy balanced tree construction, and this schedule is dependent on load and
residual energy at that particular moment. The improvement by use of this algorithm
is seen in the result obtained.
Kuo et al. [23], in their paper, elaborated the working of WSN using asynchronous
sleep-wake scheduling algorithm. This method is an effective tool to optimize the
overall energy consumption as this method defines the sleep schedule for sensor
nodes. The proposed system is designed with combinations of two algorithms which
are asynchronous sleep-wake schedules and opportunistic routing called ASSORT.
The simulation results with the use of ASSORT lead to lifetime extension.
Zheng et al. [24], in their paper, proposed an organized method for designing
asynchronous wake up of sensor nodes for an ad hoc network. This work revolves
around power management with consideration of two factors: on slot-bases and on-
demand bases, combining these factors gives better overall power enhancement. This
algorithm addresses various wake-up problems and forms an efficient network.
Kumar et al. [25], in their paper, tried to solve the problem of lesser life span of
sensors in a WSN by defining a more optimal algorithm. This algorithm not only
Energy-Efficient Routing-Based Clustering Approaches … 303

results in lifetime improvement of sensor but also deals with maintaining barrier
coverage and also provides options for fault-tolerant connectivity.
Zhang et al. [26], in this paper, introduced a general intermittent energy-aware
(IEA) and (EH-WSN) energy-harvesting wireless sensor network platform. The sys-
tem works as it adopts a double-stage capacitor structure to make sure node’s syn-
chronization without energy harvesting, and an integrator is used in order to achieve
ultra-low power measurement. Number of experiments is performed to verify perfor-
mance of IEA in terms of life and reliability of network. The results of IEA platform
utilized show ultra-low power consumption and high accuracy (Table 2).

3 Conclusion

This paper offers protocols which make relative study and give appropriate solutions
for various WSN issues. But, among various network problems, the main concentra-
tion of the study is to find appropriate cluster routing protocols and the sleep/wake-up
protocol for more optimized network. The main motto of this study is to make them
more reliable and energy-efficient network, and after this study, the utilization of
more trusted protocol is done in the future implementation. The comparison of these
two protocols with the other parameters used in optimization of WSNs is shown in
graphical manner as follows:

From the above study, we come to know that we can design an optimized and more
reliable WSN by working on these two parameters (cluster routing and sleep/wake-
up). Other parameters like network lifetime, security, and data gathering also play
important part in making healthy WSN, but these two play a very durable and more
efficient role in terms of network lifetime betterment and energy management, as
we can see with the improvement in these two protocols, the overall improvement in
WSN is seen.
304 R. R. Bhandari and K. Rajasekhar

Table 2 Comparison of the various algorithms is done here

S. Name Author Algorithm/techniques used Results
No.
1 A self-adaptive D. Ye 1. Sleep/wake-up protocol The proposed
sleep/wake-up 2. During each time, the system doesn’t
approach for slot packet transmission work on the duty
wireless sensor depends upon the node cycling algorithm,
networks instead it uses an d
sleep/wake-up
algorithm for more
optimized network
2 Green F. H. Panahi Fuzzy Q-learning (FQL) The results are
heterogeneous and device-to-device more optimized for
networks via an (D2D) communications small area network
intelligent
sleep/wake-up
mechanism and
D2D
Communications
3 Investigations of G. Optimization modulo With the use of this
graph properties Kovásznai theories (OMT) and OMT solver, more
in terms of satisfiability modulo graphical optimizes
wireless sensor theories (SMT) criteria of WSN are
network gathered
optimization
4 Green F. H. Panahi Fuzzy Q-learning (FQL) Result of their
heterogeneous proposed method
networks via an leads them to
intelligent power achieve an
control strategy operative
and D2D energy-optimizing
communications sleep/wake-up
system
5 Channel-aware C. H. Lin Channel-aware The result shows
polling-based polling-based MAC that improvement
MAC protocol for protocol CPMAC in polling periods
body area improves energy
networks: design efficiency
and analysis adaptively
6 Improved M. P. (LECSA) load and The results
scheduling Vaiyshnavi energy-consumption-based obtained are more
algorithm using scheduling algorithm efficient as
dynamic tree compared with
Construction for other
wireless sensor sleep/wake-up
networks protocols
Energy-Efficient Routing-Based Clustering Approaches … 305

References

1. Deepa C et al (2014) HHCS: hybrid hierarchical cluster based secure routing protocol for
wireless sensor networks, In: ICICES
2. Ye D et al (2017) A self-adaptive sleep/wake-up scheduling approach for wireless sensor
networks. IEEE Trans Cybern
3. Latif S et al (2013) A greener MAC layer protocol for smart home wireless sensor networks.
In: IEEE online conference on green communications, pp 169–174
4. Amjad M et al (2017) QoS-aware and heterogeneously clustered routing protocol for wireless
sensor networks. IEEE Access 5:10250–10262
5. Kang B et al (2017) A Distributed delay-efficient data aggregation scheduling for duty-cycled
WSNs. IEEE Sens J 17:3422–3437
6. Zhou Y et al (2017) Clustering hierarchy protocol in wireless sensor networks using an
improved PSO algorithm. IEEE Access 5:2241–2253
7. Wang K et al (2014) A Two-layered clustering-based MultiHop routing protocol. In: Fifth
international conference on intelligent systems, modelling and simulation, pp 573–578
8. Sreevidya B et al (2017) Enhanced energy optimized cluster based on demand routing protocol
for wireless sensor networks. IEEE, pp 2016–2019
9. Gnanambigai J et al (2014) A clustering based hybrid routing protocol for enhancing network
lifetime of wireless sensor network. In: 2nd international conference on devices, circuits and
systems (ICDCS), pp 1–4
10. Behera TM et al (2017) Work-in-progress: DEEC-VD: a hybrid energy utilization cluster-based
routing protocol for WSN for application in IoT. In: International conference on information
technology, pp 97–100
11. Xiao X et al (2017) 6LoWPAN hierarchical cluster routing protocol. In: International
conference on cyber-enabled distributed computing and knowledge discovery, pp 80–83
12. Shaha SAN et al (2017) Cluster based routing protocol using rendezvous agent. In: International
conference on energy, communication, data Analytics and soft computing (ICECDS), pp 1537–
1542
13. Acharjee T et al (2016) An improved hierarchical cluster based routing approach for wire-
less mesh network. In: International conference on computer communication and informatics
(ICCCI)
14. Misra S et al (2016) A literature survey on various clustering approaches in wireless sensor
network. In: IEEE 2nd international conference on communication, control and intelligent
systems (CCIS), pp 18–22
15. Lenka RK et al (2017) Cluster-based rendezvous routing protocol for wireless sensor network.
In: International conference on computing, communication and automation (ICCCA), pp 748–
752
16. Pant P et al (2015) A Multihop routing protocol for wireless sensor network based on grid
clustering. IEEE, pp 137–140
17. Panahi FH et al (2018) Green heterogeneous networks via an intelligent sleep/wake-up
mechanism and D2D communications. IEEE Trans Green Commun Netw
18. Kovásznai G et al (2018) Investigations of graph properties in terms of wireless sensor network
optimization. Institute of Mathematics and Informatics Eszterházy Károly University, Eger,
Hungary
19. Panahi FH et al (2017) Green heterogeneous networks via an intelligent power control strategy
and D2D communications. IEEE
20. Lin C-H et al (2016) Channel-aware polling-based MAC protocol for body area networks:
design and analysis. IEEE Sens J
21. Alhalaf A et al (2016) “gTBS: a green task-based sensing for energy efficient wireless sensor
networks. In: IEEE conference on computer communications workshops
22. Vaiyshnavi MP et al (2015) Improved scheduling algorithm using dynamic tree construction
for wireless sensor networks. In: IEEE ICCSP conference, pp 0846–0849
306 R. R. Bhandari and K. Rajasekhar

23. Kuo M-S et al (2012) Joint design of asynchronous sleep-wake scheduling and opportunistic
routing in wireless sensor networks. IEEE
24. Zheng R et al (2003) Asynchronous wakeup for ad hoc networks
25. Kumar S et al (2007) Optimal sleep-wake up algorithms for barriers of wireless sensors. The
Ohio State University
26. Zhang Y et al (2018) An efficient EH-WSN energy management mechanism, vol 23, no 4, pp
407–418
Journey of Wireless Communication

Vignesh Parameswaran and M. Shanmugasundaram

Abstract Development of mobile communication is rapid, with different methods

and techniques being introduced in wireless communications. The next few pages
will deal with the detailed study of wireless cellular technologies—first, second, third,
and fourth generations eventually leading to the fifth generation as well. This will
visualize the evolution from analog system transmissions to digital transmissions
which brought the usage of audio, graphics, video, etc. The evolution also gave
rise to Internet on the cellular mobile phones which were once realizable only in
computers through broadband connections. With improved technologies, we saw the
development of fourth-generation cell phones which harnessed the use of LTE. The
world is moving very fast, and the corporates involved in mobile communications
are in a strong tussle to achieve the fifth generation of mobile networks which will
shift the world’s entire way of functioning from autonomous vehicles to IoT and
many other things which could be seen only as part of sci-fi movies.

1 Introduction

The exchange of data which can be in any form between points that are not connected
physically is wireless communication. It helps us avoid the recurring cost incurred
in setting up physical means of communication such as cables. It is the fastest devel-
oping field in the world today. This development has been triggered by the need of
employing the major use case of wireless communication—voice transmission, to
be accompanied by video, packet, and data transmission on the air. There was a time
when there used to be research conducted to increase the capacity of wired lines.
This has reached a point wherein the world is researching on increasing the capacity

V. Parameswaran (B)
M. Tech. Embedded Systems, VIT, Vellore, India
e-mail: viguinuniverse@[Link]
M. Shanmugasundaram
School of Electronics Engineering, VIT, Vellore, India
e-mail: phdsundaram@[Link]

© Springer Nature Singapore Pte Ltd. 2020 307

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
308 V. Parameswaran and M. Shanmugasundaram

of the wireless modes [1] of communication to accommodate the fifth generation

of wireless communication so-called 5G. As there are limitations in increasing the
bandwidth and maintaining power requirements, research is widely carried out on
the types of signal transmission and the signal processing methods used in receiver.
Due to rapid increase in the number of mobile subscribers, many issues such as
congestion, low speed, and low bandwidth are faced. But above all this there are
many advantages of wireless communication as well which include saving cost of
installation of cables, saving time of installing the same, and at the time, instance
creating mobility of devices connected to a network. But wireless communication is
not only restricted to the cellular communication part. It also includes radio satellites
which are used to communicate across the world as well as the networks which work
inter-continentally. Even though we have come this far in the evolvement, there is
still scope for perfection. The journey of coming to the fifth generation of networks
is very long, and many researchers and companies have had to give in to the needs of
the present world. Those who could not adapt were left back. For instance, Nippon
Telegraph and Telephone (NTT) were the first to start a commercial cellular network
which was automated in 1979 creating huge popularity which went to become the
first nationwide 1G network in Japan. NTT is still at the forefront in the world in
revenue terms. Nordic Mobile Telephone (NMT) was an early 1G network mostly
used in Nordic countries. As the NMT specifications were free and open, many com-
panies tried their recipes and companies like Nokia, Ericsson, and Motorola [2] came
to the fore in manufacturing communication equipment. Some of these companies,
however, could not survive the retail market beyond the 2G networks because they
could not adapt to the changing needs of the people in cellular phones. And then
came many other mobile operating systems which were all outclassed by Android
and IOS. Chip manufacturing companies like Qualcomm, MediaTek, Broadcom, etc.
have had their share of successes till date. As 2G evolved into 3G, people started
expecting more and eventually came 4G which changed the outline of the cellular
market. The corporates involved in the development of 4G dished out all companies
surviving on 2G and 3G, and it has come to a stage where service providers like
AT&T, Verizon, Jio, etc. have either completely shut down the GSM network-related
infrastructure or are in the process of phasing it out soon. As companies started
cashing on the success of 4G, a survey claims that more than 54% of the companies
involved in technology and as service providers have started developing technologies
for 5G with 16% already partially deployed.

2 Evolution of Mobile Cellular Networks

Wired communication using landlines and setting up public switched telephone net-
work (PSTN) is considered the most reliable source of communication wherein we
see very good speech quality and high-speed broadband services. Cable television
providers still use wired systems from their control room because of the quality and
reliability in spite of being costly. The evolution of the mobile cellular networks has
Journey of Wireless Communication 309

been phenomenal, and lots of scene changing episode have made this field evergreen.
It was presumed that satellite phones first used for communication between boats will
become a sensation, but this vision was proved wrong by the upcoming discoveries
in the wireless ocean.

2.1 First Generation

This generation of wireless mobile communications supported communication which

was only analogue in nature. They were used predominantly for voice. They improved
on earlier systems by providing automatic switching, handover of calls to different
cells thus using the cellular concept. Japan’s NTT paved way to different adapta-
tions by different operators in various countries. NMT was the first system to sup-
port automated handover and international roaming. This was used in most parts
of Europe. AMPS from USA and total access communication system (TACS) from
Europe (ETACS) and Japan (NTACS) were more successful. These systems had only
a major difference in channel bandwidth with AMPS deploying a 30 kHz channel
whereas NTACS deployed 12.5 kHz and ETACS used 25 kHz [3]. A quick summary
can be seen in Fig. 1.
AMPS used frequency division multiple access (FDMA). Subscribers were
assigned a pair of voice channels (forward and reverse) for the duration of their
call. AMPS had a coverage of 2100 square miles. It had only ten base stations. Each
of them had height of antenna tower around 150–550 ft. They mostly deployed a
frequency reuse pattern of 7 cells with 3 sectors for each cell. The Federal Commu-
nications Commission (FCC) allocated the spectrum to two operators per market in
USA. Each operator supported a total of 416 AMPS channels and assigned 20 MHz
of spectrum, in each market. Of the 416 channels, control information was sent in 21
channels and the remaining 395 channels for voice traffic. Control information was

Fig. 1 Major first generation cellular systems

310 V. Parameswaran and M. Shanmugasundaram

sent using frequency-shift keying (FSK) and frequency modulation (FM) at higher
frequencies typically 150 MHz and above was used for the transmission of analog
voice. Such typical systems had a lifetime of a decade from 1980 to 1990. The rate
at which data was sent was around 2.4 Kbps. When such 1G phones started get-
ting popular, the number of people who started using it rose to 20 million by 1990
which meant an astonishing growth rate of around 40%. It has bad voice links, less
capacity, more noise interference, not so reliable handoff, and susceptibility to eaves
dropping by hackers. Also the phones were big in size and had a poor battery life.
Even after 2G came into picture, such systems were used as a fallback network by
service providers and also for providing roaming between different operators with
incompatible 2G systems.

2.2 Second Generation

This generation evolved mainly due to hardware platforms depicting increased pro-
cessing abilities. It also aimed toward efficient voice communication only but adapted
digital modulation in contrast to 1G systems. This shift also increased the system
capacity through use of digital speech codes which were efficient spectrally, thus
deploying time or code division multiple access to multiplex many users on the same
frequency channel. It could also realize frequency reuse due to better digital mod-
ulation, coding, and equalization techniques. Better speech codecs and signal-level
processing also improved voice quality. Security was also worked upon in 2G to pre-
vent eavesdropping through simple encryption techniques which differed for every
channel access and known only to mobile station and the infrastructure.
Global System for Mobile Communications (GSM), IS-136 TDMA, and IS-95
CDMA systems were the prominent cellular systems in this generation among many
others. IS-95 was used mostly in some parts of Asia and North America. IS-136,
touted as a digital improvement over AMPS using 30 kHz channels, was a TDMA-
based system. However, GSM was the most used technique and widely adapted which
went on to become a standard. GSM is also based on TDMA, and one 200 kHz channel
is time shared between 8 users in different time slots [4]. It used Gaussian minimum-
shift keying (GMSK) for modulation because of its ability to provide constant enve-
lope and providing efficient use of the spectrum along with good power characteristics
[5]. GSM did not stop here and went on to provide numerous services. Short mes-
saging service (SMS) was one of them, and this does not describe at all. Besides
voice and SMS, circuit-switched properties started to get imbibed. Around middle of
the 1990s, European Telecommunications Standards Institute (ETSI) that had taken
charge of the GSM standard introduced GSM Packet Radio Systems (GPRS) as a
stepping stone toward increasing the data rates. The architecture for GSM and GPRS
was almost the same in the signaling links and also usage of the same bands as well
as slots in time. GPRS could garner average data rate of 30 kbps. GSM also got a
further push during the early part of 1997 in the data domain with the innovation
of Enhanced Data Rate for GSM Evolution (EDGE). EDGE could increase the data
Journey of Wireless Communication 311

rate to around 59 kbps per slot by addition of the 8 PSK scheme of modulation. This
was almost 3 times as that of GPRS. Practically EDGE could provide average user
rates around 100 Kbps.
CDMA was claimed by Qualcomm in 1989 as a more efficient and higher quality
technology. The unique property which it provided was the usage of same frequency
at same time by many users. It used a 12.5 MHz bandwidth to transmit a voice signal at
around 9.2 Kbps. Other advantages include every cell using same frequency channel
in turn simplifying the planning for frequency and increasing the capacity. It was
also instrumental in the innovation of a mobile station able to connect to a new base
station before the disconnection from the previous one which was referred to as soft
handoff [6]. It also was able to provide one dedicated channel for data at 9.6 kbps
which went on to become around 14.4 kbps later on. CDMA was instrumental in
the fast transition to 3G through newer versions like CDMA2000-1X and EV-DO
thus challenging GSM which was undertaking a subtle evolution through GPRS and
EDGE to 3G.

2.3 Third Generation

The second generation improved voice capacity and quality and began data support
for Internet. But the circuit-switched way was inefficient for data providing low data
rate. Hence, the need for third-generation systems to provide increased data rate
and advanced services and applications was felt thus triggering the International
Telecommunications Union (ITU) to invite proposals for such systems. It laid out
criteria such as 2 Mbps in fixed, 384 Kbps in pedestrian, and 144 Kbps in wide area
vehicular environments. There were parallel evolutions going on in the GSM and
CDMA track which were carried out by consortium bodies for standards called Third-
Generation Partnership Project (3GPP) and 3GPP2 (spearheaded by Qualcomm),
respectively [7].
CDMA2000-1X was 3GPP2’s first bet (here 1X implied that it uses same band-
width as IS-95). As it provided specifications less than the standards set, it was
referred to as 2.5G. CDMA-1X increased capacity by doubling the number of for-
ward channels as compared to previous 64. In contrast, 3GPP was created as a con-
sortium of 6 regional standards in telecommunication. The 3G system based on the
evolution of GSM was Universal Mobile Telephone Service (UMTS), indigenously
developed by ETSI. The characteristics of UMTS were a core network (CN) which
provided routing, user management in addition to switching. It also comprised of
UMTS Terrestrial Radio Access Network (UTRAN) and the User Equipment (UE).
UMTS was consistent with the architecture used for GSM/GPRS. It is described in
Fig. 2. The air interface used was Wide-band CDMA (W-CDMA) inspired by IS-95.
The system supported 100 voice calls at the same time and maximum data rates from
384 to 2048 kbps due to bandwidth of 5 MHz.
Further additions to 3G family included EV-DO which was the first system to
deliver speeds comparable to broadband developed by Qualcomm (the 3GPP2 group)
312 V. Parameswaran and M. Shanmugasundaram

Fig. 2 GSM network architecture

in 2002, 3 years ahead of HSDPA deployed by GSM. It provided up to 2.4 Mbps down-
stream and up to 153 kbps upstream. The modulation and coding could be dynami-
cally adapted according to conditions by EV-DO. High-Speed Data Access (HSPA), a
combination of HSDPA and HSUPA (downlink and uplink), was additional enhance-
ment to the 3GPP standard. It introduced new advanced techniques such as adapting
modulation and coding according to channel conditions, fast dynamic scheduling of
packets, and hybrid automatic repeat request (H-ARQ), an improved retransmission
technique, to soft-combine multiple erroneous retransmissions to recover from errors
more quickly. Various other additions occurred to the family which are considered
as part of 4G or 3G by different explanations which will be seen further as part of
4G here.

2.4 Fourth Generation

This generation has lots of arguments of technologies not meeting the standards set
by ITU for 4G. HSPA+, WiMAX, and LTE are some of the most prominent ones.
HSPA+ uses higher order modulation such as 64QAM and 16QAM, in addition to
multiple-input multiple-output (MIMO) wherein two transmit and receive antennas
each are used to enhance diversity, beamforming, and spatial multiplexing. It uses
dual-carrier downlink operation for doubling data rates and could enhance capacity
in addition to data rates. It also gave attention to the draining batteries of the devices
using discontinuous transmission and reception. WiMAX has many features adapted
by LTE, the most prominent one being orthogonal frequency division multiplexing
Journey of Wireless Communication 313

(OFDM) [8]. OFDM can operate even when not in line of sight and can resist mul-
tipath. Its peak data rate is 74 Mbps. Support for advanced antenna techniques and
dynamic allocation of resource to a user are other characteristics of OFDM.
LTE, referred to as Long-Term Evolution, has many features inherited from
HSPA+ and WiMAX. When 3G was at its peak, the world was moving rapidly to high
bandwidth applications such as music downloads, video streaming, and IPTV. It was
the time when smart devices started proliferating. We started using laptops, netbook
computers, large screen devices, devices for gaming, camcorders and projectors with
built-in wireless interfaces, portable media players, cameras, and other machine-to-
machine communication devices. Also service providers and device manufacturers
started cashing in on this and the need was felt for bringing the performance on wire-
less devices on par with broadband in addition to lowering the cost per megabyte of
data for the users to make the market competitive. But this had to be done using the
already existing spectrum taken up by 2G and 3G. OFDM had many advantages. It
had reduced computational complexity due to use of known FFT/IFFT techniques,
frequency diversity, was robust against narrowband interference, and had graceful
degradation of performance under excess delay. Advanced multi-antenna solutions
[9] such as beamforming, spatial multiplexing, and transmit diversity in addition
to MIMO that made the spectrum efficient, increased capacity, and made a robust
link were provided by the LTE standard. The core design to support LTE is called
Evolved Packet Core (EPC) which can be seen in Fig. 3. LTE has a capability of data
rate up to 150 Mbps. LTE-Advanced has been targeted to obtain 1 Gbps downlink
and 500 Mbps uplink [10]. With such capability, the existing 2G and 3G systems
are being planned to be phased out to free the spectrum which is captured by them.
But it is not an easy task because LTE alone cannot guarantee service in every nook

Fig. 3 Evolved packet core architecture

314 V. Parameswaran and M. Shanmugasundaram

and corner of the world. 2G is the best fall back network option available in this
fast-evolving industry of wireless networks.

2.5 Fifth Generation

The world has started becoming greedy. From the smart handheld devices to more
advanced use cases such as online gaming, virtual reality, artificial intelligence, 3D
4K–8K cloud and video streaming, autonomous cars, big data [11], and the most
important one which will drive all the previous, Internet of Things (IoT), the vision
forward has been just like some sci-fi movie. The extent of people realizing tech-
nology to aid in surgery and medical tracking proves the fact that we humans have
progressed to unimaginable levels. To envisage this vision, IUT has aimed for a
standard peak data rate of 20 Gbps and latency of 1 ms for 5G systems. Basi-
cally, IUT has defined 3 categories of service. They are enhanced Mobile Broad-
band (eMBB) or handsets, Ultra-Reliable Low-Latency Communications (URLLC)
including industrial applications and autonomous vehicles, and Massive Machine
Type Communications (MMTC) or sensors. As the throughput requirements have
increased enormously, a large number of new spectrum called the 5G NR frequency
bands have been allocated which work in the GHz range and some particularly in
millimeter wave bands. In addition to this, multiple access schemes, modulation
methods, Future PHY/MAC, flexible duplexing methods, and massive MIMO have
all been under research for 5G [12]. Corporates like Qualcomm, Intel, and Huawei
are actively involved in modem technology while Nokia, Cisco, ZTE, Samsung, and
Ericsson are involved in infrastructure of 5G. Companies have left no stone unturned
to safeguard their architecture or design for the 5G systems due to which details on
the architecture are only an imagination.

3 Conclusion

Every generation of wireless mobile communication intended to raise the bars set
for that particular decade. 2020 is seen as a revolutionary year in this field where we
will be able to experience some technologies which were imagined only in movies.
The journey so far has been fantabulous and will continue to be. No generation can
be considered as a winner as all of them satisfied the needs of that decade very well
in their own way and own limitations.
Journey of Wireless Communication 315

References

1. Raychaudhuri D, Mandayam NB (2012) Frontiers of wireless and mobile communications.

Proc IEEE 100(4):824–840
2. Escher JS (1997) Wireless portable communications trends and challenges. In: 1997 IEEE
radio frequency integrated circuits (RFIC) symposium. Digest of Technical Papers, Denver,
CO, USA, pp 1–3
3. Hughes CJ, Appleby MS (1985) Definition of a cellular mobile radio system. IEE Proc F—
Commun Radar Signal Process 132(5):416–424
4. Eberspächer J (2009) GSM—architecture, protocols and services. Wiley, Chichester
5. Heine G (1999) GSM networks: protocols, terminology, and implementation. Artech House,
Boston, MA
6. Orlik PV, Rappaport SS (1999) On the hand-off arrival process in cellular communica-
tions. In: WCNC. 1999 IEEE wireless communications and networking conference (Cat. No.
99TH8466), New Orleans, LA, USA, vol 2, pp 545–549
7. Hoy J (2015) 3GPP network types. In: Forensic radio survey techniques for cell site analysis.
Wiley. [Link]
8. Chia S (2007) Mobile network evolution beyond 3G. In: 2007 IEEE radio and wireless
symposium, Long Beach, CA, pp 269–272
9. Ghosh A, Zhang J, Andrews J, Muhamed R (2011) Fundamentals of LTE. Prentice Hall
10. Abeta S (2010) Toward LTE commercial launch and future plan for LTE enhancements (LTE-
Advanced). In: 2010 IEEE international conference on communication systems, Singapore, pp
146–150
11. Huang J et al, A big data enabled channel model for 5G wireless communication systems.
IEEE Trans Big Data
12. Vannithamby R, Talwar S (2017) Massive MIMO communications. In: Towards 5G: applica-
tions, requirements and candidate technologies. Wiley
A Survey on K-Means Clustering
for Analyzing Variation in Data

Pratik Patil and A. Karthikeyan

Abstract Most of the times data for certain task seems to be varying due con-
stant changes made to method of data collection as well as due to inclusion of new
parameters related to the task. This may result in false conclusion derived from data
generated and might lead to failure in task or degradation in the standard of activity
related to that task which is being monitored from that data. Clustering is basically
the grouping of similar kind of data wherein each cluster consist of data with some
similarities. Whereas most of the data is unstructured or semi-structured, and that’s
where unsupervised K-means Clustering method plays role to convert the data into
structured one’s for clustering. This paper consist of K-means clustering method
which is being used to keep an eye on such variations which are occurring in data
generated for a task when certain changes are incorporated in technique to track this
data.

Keywords K-means · Clustering · Machine learning · Dataset · Variation ·

Analysis · Data mining · Iterations · Parameters · Eucledian · Structure

1 Introduction

Data analysis and its mining are basically related to data science field of computer
science which are the major part of nowadays technique to understand the pattern,
structure and hidden specifications of large data that is being generated by user
activities on daily basis. This Data analysis plays a major role in keeping track of this
huge amount of data and improving the application whose data is being monitored.
Clustering is the machine learning method which involves grouping of data
according to their characteristic, behaviors, and features. These data points are

P. Patil (B)
[Link] Embedded Systems, VIT, Vellore, Tamil Nadu, India
e-mail: priyanshuagri@[Link]
A. Karthikeyan
School of Electronics Engg, VIT, Vellore, Tamil Nadu, India
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 317

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
318 P. Patil and A. Karthikeyan

grouped such that, each data point in a particular group differs from other group
data point and shows some similarity with the other points in its own group. Cluster-
ing technique is an unsupervised ML method wherein input data points are clustered
into groups without any reference data about its features. These clustered data can
then be utilized to understand the behavior, pattern in these groups for better under-
standing of the problem. The clustering method of ML follows unsupervised learning
procedure for training.
Unsupervised learning is the method of training an artificial intelligence algorithm
with data that is neither labeled nor classified so as making the algorithm to work
through it without any help or guide. There are two types of clustering technique
hard clustering and soft clustering. In hard clustering, every data point in the group
either belongs to its cluster completely or partially. Whereas in soft clustering, the
likelihood of each data point to be in a particular group is assigned instead of putting
it in separate group.
Based on the task of clustering, there are different methods which can be used to
define the likelihood between the data points viz. connectivity model, centroid model,
distribution model, and density model. Connectivity model takes into account the data
space exhibited by the data points and the points which are closer, are clustered in one
group. Centroid model considers the centroids defined by user and according to data
points distance from centroids they are grouped in particular group. Distribution
model works on basis of similarity of data point probable distribution (Normal,
Gaussian), whereas density model search for areas of varied density of data points
in data space.

2 K-Means Clustering

K-means clustering is an iterative technique which involves finding local maxima

during each iteration so that data points are grouped properly. For processing the
data points, first it works with formation of groups for randomly selected centroids.
Then it performs the optimization through iterative method. Whereas the groups are
formed through calculating closest distance to the centroid using Eucledian distance.
Let say there are two points,

L = (x1(L), x2(L), . . .) and

M = (x1(M), x2(M), . . .)

Then its Eucledian distance will be,

d(L , M) = sqrt((x1(L) − x1(M))2 + (x2(L) − x2(M))2 + · · ·)

A Survey on K-Means Clustering for Analyzing Variation in Data 319

The centroids mentioned are basically the means of the points that are assigned
to it in the cluster, which can be calculated by either random initialization method or
through selection of centroid in range of dataset values.
The Algorithm is as follows:
• First we select the number of clusters k (using different methods).
• Then select random k points which are the centroids (not necessarily from your
dataset).
• Assign each data point to the closest centroid according to distance from the
centroids, this forms K clusters.
• Now compute and place the new centroid after recalculation.
• Reassign each data point to the new closest centroid. If any reassignment took
place again repeat the step four, otherwise stop as final clusters are formed.

3 Procedure for Finding Clusters

If we use the random initialization method, then we face the issue of random ini-
tialization trap wherein the clusters formed after the iterations are unexpected and
results in improper data analysis. So to solve this issue we are using the K means
Elbow method or K-means ++ method (Figs. 1 and 2).

Fig. 1 Actual clusters expected

320 P. Patil and A. Karthikeyan

Fig. 2 Cluster obtained (Random initialization trap)

Steps for K-means Elbow method to find number of clusters,

• So first start with number of clusters to be as k = 1, 2, 3… (Fig. 3).

• Calculate for each k value the Within cluster sum of squares WCSS given by
equation,

Fig. 3 Elbow method plot

A Survey on K-Means Clustering for Analyzing Variation in Data 321

  
WCSS = distance(Pi, C1)2 + distance(Pi, C2)2 + distance(Pi, C3)2 + · · ·

Pi are the distance of points in dataset

This is the distance of all points in cluster from centroid which are squared and
summed.
• Now use these WCSS values for each value of k to make the WCSS plot as shown
in fig.
• Lastly, find the optimized value of k where we have almost unvarying distortion
i.e. WCSS (check the point where there is no steep descent).

4 Analysis of Data Variation Using K-Means

Many of research fields are utilizing K-means for keeping the track of data variation
happening in the data generated by particular field, wherein the number K is decided
for the expected variation understanding. The data points from previous version of
task and its further changed version are compared to get complex clustering plot for
the task.
This data can be used to keep a complete understanding of changes incurred over a
period of time and then accordingly act to handle the errors or further improvements
in the system. From Fig. 4, we can see that the plot formed show variation in terms of
cluster with respect to some parameter related to the system under analysis. Wherein
the data points are grouped into eight clusters each with varying density. Data points
in each cluster indicates the variation, for example from cluster 6 it is clear that the
data points present in it are collected from system with different versions whose
properties are related whereas in some case the variation is unique to itself example
cluster 2 and 5 wherein it has less data points from few version of changes system
and may indicate a major change in system or some serious bugs in system. This
gives researcher’s a good understanding of the system behaviors and can utilize it to
overcome the issue.

5 Advantages and Disadvantages

5.1 Advantages

Easy implementation and being unsupervised it doesn’t require any reference output
for training, i.e., based on input it finds patterns and forms cluster. For large number
of variable, it require high computational power compared to other clustering algo-
rithms. It can change clusters with each iteration thus generating new centroid for
322 P. Patil and A. Karthikeyan

Fig. 4 Clustering data for variation in data points

perfect optimization. Clusters produced by K-means are almost accurate for each
data points. This method reduces the noise and spatial course.

5.2 Disadvantages

There is no unique solution for this method, as for different k value we get different
result being sensitive to local optima and initial values. Also for the K value as it is
not known properly for accurate results. Generates same size clusters mostly even
if input belongs to different size cluster. Rescaling the data can change the result in
other way. The way in which data is feed to K-means model also decides the results.

6 Conclusion

Clustering methods are the best one when something related to analyzing large
unstructured data comes up. But since it requires large amount of dataset for learning
the pattern it has its drawbacks for small database systems. In this paper, we mostly
talked regarding K-means clustering method which involves lots of iterations and try
to get optimal results but this is not efficient solution. Hence K-means elbow method
can be utilized which again requires some computation power. K-means clustering
A Survey on K-Means Clustering for Analyzing Variation in Data 323

being unsupervised learning doesn’t require any reference output data for analysis as
is not easily available in most system. This clustering method can further be improved
by using multiple iteration for optimized value or by utilizing K ++ method which
uses elbow criteria to find optimized k. For variation analysis of data K-means seems
to be perfect method as it is simple, easy to implement and moreover it catches the
changes in system at very cheap cost compared to other algorithms. Also being sim-
ple in clustering mechanism it ignores the too small detail which is good for some
system with some exceptions where only certain issue leads to problem.

Bibliography

1. Kodinariya TM (2013) Review on determining number of Cluster in k-means. 1(6)

2. Pattern recognition and machine learning book by Christopher Bishop
3. Gondaliya B (2014) Review paper on clustering techniques. 2(7). ISSN 2349-4476
4. Koundinya AK, Srinath NK, Anchalia PP (2013) Mapreduce design of K-means clustering
algorithm
5. Santini M (2016) Machine learning for language technology ML4LT
6. [Link]
7. [Link]
8. Theodoridis S Koutroumbas K (2003) Pattern Recognition. Elsevier Science
9. Kleinberg, J (2003) An impossibility theorem for clustering. In: Advances in neural information
processing systems. pp 463–470
10. Han J, Kamber M (2001) Data mining: concepts and techniques. Morgan Kaufmann
11. Kaufman L, Rousseuw P (1990) Finding groups in data—an introduction to cluster analysis.
In: Wiley series in probability and mathematical statistics
12. Machine Learning For Dummies Book by John Mueller and Luca Massaron
13. Ester M, Kriegel H-P, Sander J, Xiaowei X (1996) A density-based algorithm for discovering
clusters in large spatial databases with noise. Kdd 96(34):226–231
A Review on Mobile Cloud Computing
Interoperability Issues and Challenges

Tribid Debbarma and K. Chandrasekaran

Abstract Mobile cloud computing (MCC) is the convergence of two recent tech-
nologies namely “Cloud Computing” and “Mobile Computing” with wireless net-
works as a communication backbone. There are mainly three paradigms that use
the concepts of MCC, viz. edge computing, fog computing and cloudlets. Due to
the presence of various heterogeneous hardware and software platforms in MCC,
there are many interoperability issues which create vendor/services lock-in prob-
lems, it also makes data and application portability difficult. This paper studies the
different paradigms of MCC and the challenges in making them interoperable in
heterogeneous hardware and software platforms. We have summarized some of the
MCC-based research papers and their findings. Contribution of this paper is the sum-
mary of challenges and research scopes in the field of MCC where it needs to be
addressed to mitigate the interoperability issues.

Keywords Mobile cloud computing · Interoperability · Portability · MCC

1 Introduction

Cloud computing (CC) and mobile computing (MC) are the two most happening
technologies in the recent information and communication technology (ICT) and
computing world. Cloud computing has changed the ICT infrastructure in a great
way. The days of companies/organizations handling their own ICT infrastructure is
replaced with cloud data centres and cloud storage services. On the one hand, mobile
device technology is growing by leaps and bounds. Still, these technologies are not

T. Debbarma (B)
Department of Computer Science and Engineering, National Institute of Technology Agartala,
Agartala, India
e-mail: tribid@[Link]
K. Chandrasekaran
Department of Computer Science and Engineering, National Institute of Technology Karnataka,
Surathkal, Mangalore, India
e-mail: kchnitk@[Link]

© Springer Nature Singapore Pte Ltd. 2020 325

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
326 T. Debbarma and K. Chandrasekaran

without its own limitations; e.g., CC does provide distributed services accessible
from anywhere in the world, but it has limited or no mobility. On the other hand,
mobile computing devices are having limited battery, storage, processing power and
data storage capability due to its small form factor.
MCC merges two technologies and makes cloud computing services mobile. Fur-
ther, mobile devices are augmented with large cloud storage and unlimited computing
power of servers by accessing data centre infrastructure. The access to cloud server
systems can help mobile devices to save the energy consumption by different tech-
niques, e.g. by using computational offloading techniques [1, 2] and partitioning of
mobile applications at the runtime to execute the application in the mobile device
and cloud servers with dynamic/static context awareness.
Despite having many advantages, MCC has certain challenges, viz. interoperabil-
ity and portability of the cloud services in the vastly heterogeneous mobile operating
system (OS) and hardware platform, different wireless communication technology,
offloading overheads, security, privacy, etc.
This paper summarizes different MCC and cloud computing-based research works
findings and highlights the research scopes in MCC, present scenario, and challenges
that need to be addressed. Specifically, the aspects of interoperability and portability
of the MCC services with data portability to the end-users are highlighted for further
work.
The rest of the paper is organized in different sections as follows: Sect. 2 discusses
the background of MCC and motivation behind this work. Section 3 discusses CC-
and MCC-related works. Section 4 discusses the challenges in MCC, Sect. 5 discusses
the research scopes in MCC field including interoperability issues, and in Sect. 6,
conclusions of this review and future work are discussed.

2 Background and Motivation

MCC is the convergence of mobile computing and cloud computing. Mobile cloud
computing leverages cloud computing and mobile computing for a pay-as-you-use
manner.
According to Sanaei et al. [3] “Mobile Cloud Computing is a rich mobile com-
puting technology that leverages unified elastic resources of varied clouds and net-
work technologies toward unrestricted functionality, storage, and mobility to serve
a multitude of mobile devices anywhere, anytime through the channel of Ethernet
or Internet regardless of heterogeneous environments and platforms based on the
pay-as-you-use principle”.
There are different paradigms that are based on MCC concepts. Below, we dis-
cussed some of the extended MCC concepts that are being studied and implemented
at a different level in the industry and academia.
Edge Computing (EC): This architecture primarily transfers the processing, data
analysis, communication and other tasks from the edge gateway or the source devices
to the programmable automation controllers (PACs) instead of sending it to the cloud
A Review on Mobile Cloud Computing Interoperability Issues … 327

data centres. Only the large data and processing jobs are sent to the cloud systems.
This approach reduces the latency of different operations where it does not need the
service of large cloud systems [4].
Cloudlets: Computing systems connected at the edge of the Internet, which can
act as local cloud computing data centre and access point to the mobile devices
covering a small area and provides low latency computing services with a limited
area of mobility [5].
Fog Computing (FC): This term was first introduced by Cisco Systems [6]. The
concept of fog computing has been coined after edge computing. In FC, the cloud
computing is extended to the edge of the Internet. The mobile devices and IoT devices
transfer the data processing tasks into IoT gateway or fog nodes instead of directly
connecting to the cloud data centres. The availability of local gateway and fog nodes
can greatly reduce access latency of different services. The IoT gateways and fog
nodes can access centralized cloud data centres as and when additional resources are
required. OpenFog Consortium [7] formed by academic and industrial organizations
are supporting, and developing guidelines for fog computing-based systems.

2.1 Motivation

Though there are different studies which have emphasized different issues related to
cloud computing and MCC such as security, scalability, VM migration [1], portabil-
ity and interoperability [8, 9] issues. Still, there is no common industry standard till
date for MCC services interoperability and portability. The industry is yet to adapt
standard interoperable application and service frameworks. Interoperability is non-
trivial in the MCC as the hardware and application software’s of mobile devices are
vastly heterogeneous in nature. Moreover, different wireless network technologies
on which the mobile devices are connected to the cloud systems make the interoper-
ability issues more challenging and also makes it vulnerable to security and difficult
to maintain QoS.
Presently, some of the cloud services that are extended to MCC services, e.g.
Dropbox, Google Drive, etc., support storage services across different mobile OS
and device categories which end-users can access without any device/OS lock-in. On
the other hand, service like iCloud Drive mobile app-based service is not supported
outside Apple products. Voice assistant service Siri (by Apple) is supported only in an
Apple-developed operating system; this creates a lock-in situation where the users of
other devices from non-Apple products cannot access such services. Here, the service
and device lock-in create a situation where the end-user may have to compromise in
his/her requirements. MCC with interoperable applications and services is the need
of the hour.
328 T. Debbarma and K. Chandrasekaran

3 Related Works

There are various studies on CC and MCC. Some of the studies’ findings have been
summarized in Table 1.
Haile and Altmann [8] studied the impacts of portability and interoperability on
the value of cloud computing platforms and their users. Their findings give us an idea
of how closed platform lock-in is counterproductive for the cloud service providers
and can be dissatisfying to the end-users.

Table 1 Summary of different MCC studies

Studies/proposed systems Proposed solutions
CloneCloud [1] Migrates application into the computational cloud with
dynamic profiling and static analysis to optimize
energy and execution time. It is designed to offload the
tasks of android-based systems to the cloudlets
MAUI [10] Computation offloading with context awareness for
decision-making
Marmalade [11] Is a cross-platform toolkit that converts the application
code to the target platform-supported format
automatically
SAMI [3] SOA-based system to facilitate application portability
prevents energy losses by using nearest resources and
provides interoperability by using MNO
Cloudlet [5] It uses virtualization techniques to mitigate the
heterogeneity problems of mobile devices. It is a small
cloud data centre that provides low latency
compute-intensive services to nearby mobile users
MOMCC [12] MNOs are used to authorize and govern mobile users to
enhance computing capabilities in the cloud platform
Mirage [13] A cloud OS that runs on top of a hypervisor. Different
portable cross-platform and applications can be
produced in the normal OS, and they can be compiled
to run in the mobile devices
Aneka [14] Architecture and platform for distributed application
development in the cloud computing. This allows to
develop cloud applications for different cloud
infrastructures
Akherfi et al. [9] A middleware to adapt cloud Web services by
transforming SOAP to REST and XML to JSON
Green Cloudlet Network (GCN) [15] A private virtual machine executes the offloaded tasks
of user equipment. The cloudlets are run by green and
with grid power as a backup
Manjunatha et al. [16] A cloud mobile hybrid application development
systems using domain-specific language (DSL) for
auto-generation of a communication link between
mobile systems and target cloud platforms
A Review on Mobile Cloud Computing Interoperability Issues … 329

Sanaei et al. [3] proposed a three-tier architecture based on service-oriented archi-

tecture (SOA) for MCC that provides interoperability by using mobile network oper-
ators (MNOs) and portability by utilizing SOA concepts. But the use of MNOs for
arbitration process in a continuous manner adds processing overheads. CloneCloud
[1] concept tries to resolve the energy consumption and execution time issues of
mobile devices by offloading the process into the mobile device’s clone, i.e. running
in a cloud system.
Aneka [14] is a development environment for the CC that allows to develop cloud
applications which can be deployed in various cloud infrastructures.
By providing portability and interoperability, cloud service providers can give
better leverage to the users of different mobile platforms which are very much present
in the MCC scenario.
The summary of some of the MCC studies, their proposed solutions are given in
Table (1).

4 Challenges of MCC

The challenges of MCC are inherent to cloud computing and mobile computing.
Many of the issues overlap with cloud computing and mobile computing. The aug-
mentation of mobile computing systems into the cloud computing adds additional
challenges in the existing CC issues. Table 2 provides summary of different MCC
survey studies. The major issues and challenges in MCC are listed below.
Privacy and Security: Privacy is one of the most pertinent issues in MCC, where
there is a thin line between personal/private data and data which is to be shared. So
when a mobile device is connected to the CC service, its application which is keeping
the private data and shareable data within the same application’s accessibility is a
major concern of accidental and unwanted breach of those private data.

Table 2 Summary of different MCC survey studies

Studies Discussed open issues and challenges
Sanaei et al. [17] Context awareness, live VM migration, trust, security, privacy, mobile
communication congestion
Abolfazli et al. [18] Lightweight techniques, portability, interoperability, seamless
connectivity, live VM migration
Han Qi [19] Data delivery, task division, better service
Dinh et al. [23] Low bandwidth, network access management, quality of service,
standard interface, service convergence
Chang et al. [20] Privacy, SaaS design for mobile, green computing and energy saving,
mobility, architecture and infrastructure for mobile cloud
Fernando et al. [21] Mobility, mobile cloud security, incentives for surrogates
Roman et al. [22] Security issues for the edge computing
330 T. Debbarma and K. Chandrasekaran

Privacy issues of MCC [20] have been discussed as a requirement for MCC in their
review study. Fernando et al. [21] discussed the mobile security issues as challenges
in the MCC. Roman et al. [22] in their study highlighted the security challenges in
the edge computing platforms.
The underlying security issues of cloud computing are also omnipresent in MCC.
In addition to this, the use of wireless networks introduces the security vulnerabilities
of wireless communications.
Portability: Portability issues are very much inherent in different mobile cloud
services; e.g., songs purchased through iTunes service are not compatible or portable
to other players. This service is application’s platform-dependent. They do not sup-
port outside of other players out of the box. Portability of different mobile services
among different OS vendors and hardware is still a prevalent issue which is not
addressed fully.
Interoperability: The underlying technologies of MCC are heterogeneous in
many ways. The major components of MCC are MC devices with different OS ven-
dors and versions, wired and wireless networking technologies (3G, GSM, WAN,
4G, etc.) and different cloud service vendors with proprietary architectures. Interop-
erability is of major concern to the customers of MCC as the changes in MC tech-
nologies happen very rapidly and the end customers might fall in a vendor/service
lock-in situations which are not desirable. There are some open standards and APIs
and middleware’s to support the issues. But there is still a gap in the adaptability of
any common standard by different cloud service providers.
The interoperability and portability issues and challenges have been discussed in
[21, 18].
Limitation in the interoperability among cloud services is a major issue which
creates cloud vendor/services lock-in situations where the end-users/clients are
compelled to accept compromises. The interoperability among different cloud
infrastructure and services are to be resolved in many dimensions.

5 Research Scopes

The interoperability challenges need to be studied in more detail at the academia

and industry. Primarily, interoperability issue is mainly concerned to developers
and service providers for management of IaaS, PaaS, and SaaS layers of the cloud
computing. But the growing interest of MCC services among end-users who are non-
developers has increased the importance to extend the benefits of the interoperability
and portability of services and data towards them also. In [24], it interpreted the
interoperability and portability with five different scenarios. In this guide, scenarios
1–3 give guidelines on how to achieve the interoperability and address the issues
related to it. The scenario becomes more complex when these cloud services are to
be augmented with mobile devices. Vertical and horizontal heterogeneity [17] nature
of MCC needs to be considered on different aspects to achieve an MCC system that
supports the interoperability.
A Review on Mobile Cloud Computing Interoperability Issues … 331

Survey studies [17] discussed the taxonomy and open challenges issues of hetero-
geneous MCC. In their study, they have pointed out some open issues which need to
be addressed, e.g. context awareness, live VM migration, trust and security issues.
ReSTful models have been used in CC systems for achieving interoperability in
the industry and in academic studies. The ReSTful model can be integrated with
SOA into the MCC platform to achieve interoperability and data portability.

5.1 Summary

There are many studies and development of middleware and application program-
ming interface (API) taking place to provide interoperability among the services for
the management of PaaS as well as IaaS and SaaS layers, but fewer works have been
carried out for interoperable MCC applications that serve end-users of MCC services.
End-users need mobile applications that are interoperable across various mobile plat-
forms, and the ability to port the data has to be incorporated in the new architectures.
Further research works need to be carried out for the development of frameworks with
ReSTful Web services and service-oriented architecture for seamless interoperability
and portability among various MCC services.

6 Conclusions

Many research activities addressed different challenges pertaining to MCC. Still,

there is a need to develop standards and frameworks to handle the interoperability
and portability issues of MCC.
This paper discussed about the findings of different review studies and MCC
related sources. Many studies have proposed application migration techniques using
with static/dynamic profiling for optimized decision to reduce energy consumption
and augment the processing power of cloud systems. Some studies discussed on the
security and privacy issues as major challenges. But few studies have emphasized
on the portability and interoperability issues of MCC applications and services at
the end-user level. As there is still no standard industry-accepted architecture and
many cloud services vendors run homegrown platform with little interoperability
and portability to others, it has opened an opportunity for researchers to develop
semantic architectures and standards with interoperability and portability among
heterogeneous MCC technologies.
In our future work, we will develop an interoperable MCC services architec-
ture based on a combination of RESTful and SOA architecture that will support an
interoperable MCC application, irrespective of software and hardware platform.
332 T. Debbarma and K. Chandrasekaran

References

1. Chun B, Ihm S, Maniatis P, Naik M, Patti A (2011) CloneCloud: elastic execution between
mobile device and cloud. In: Proceedings ACM the European professional society on computer
systems (EuroSys’11). Salzburg pp 301–314
2. Kristensen MD (2010) Scavenger: transparent development of efficient cyber foraging appli-
cations. In: Proceedings IEEE eighth annual international conference on pervasive computing
and communications (PerCom). Mannheim pp 217–226
3. Sanaei Z, Abolfazli S, Gani A, Shiraz M (2012) SAMI: service based arbitrated multi-tier
infrastructure for mobile cloud computing. In: Proceedings of IEEE 1st international conference
on communications in China workshops (ICCC). Beijing, pp 14–19
4. Mobile Edge Computing: [Link]
5. Satyanarayanan M, Bahl P, Caceres R, Davies N (2009) The case for VM-Based cloudlets in
mobile computing. IEEE Pervasive Comput 8(4):14–23
6. Fog Computing: [Link]
[Link]
7. OpenFog Consortium: [Link]
8. Haile N, Altmann J (2017) Evaluating investments in portability and interoperability between
software service platforms. Future Gener Comput Syst
9. Akherfi K, Harroud H, Gerndt M (2016) A mobile cloud middleware to support mobility and
cloud interoperability. Int J Adapt Resilient Auton Syst (IJARAS)
10. Cuervo E, Balasubramanian A, Cho D, Wolman A, Saroiu S, Chandra R, Bahl P (2010) MAUI:
making smartphones last longer with code offload. In: Proceedings of ACM 8th annual interna-
tional conference on mobile systems, applications and services (MobiSys’10). San Francisco,
pp. 49–62
11. Marmalade: [Link]
12. Abolfazli S, Sanaei Z, Shiraz M, Gani A (2012) MOMCC: market-oriented mobile cloud
computing based on service oriented architecture. In: Proceedings of IEEE 1st international
conference on communications in China workshops (ICCC). Beijing, pp. 8–13
13. Madhavapeddy A (2010) Mirage, a cloud operating system: [Link] (2010)
14. Vecchiola C, Chu X, Buyya R (2009) Aneka: a software platform for .NET-based cloud com-
puting. In: Gentzsch W, Grandinetti L, Joubert G (eds.) High speed and large scale scientific
computing. pp 267–295
15. Xiang S, Ansari N (2017) Green cloudlet network: a sustainable platform for mobile cloud
computing. IEEE Trans Cloud Comput
16. Manjunatha A, Ranabahu A, Sheth A, Thirunarayan K (2010) Power of clouds in your pocket: an
efficient approach for cloud mobile hybrid application development. In: IEEE 2nd international
conference on cloud computing technology and science (CloudCom’10). Dayton, OH, pp
496–503
17. Sanaei Z, Abolfazli S, Gani A, Buyya R Heterogeneity in mobile cloud computing: taxonomy
and open challenges. IEEE Commun Surv Tutorials
18. Abolfazli S, Sanaei Z, Sanaei MH, Shojafar, M., and Gani, A.,: Mobile Cloud Computing: The
State-of-the-art, Challenges, and future research. Encyclopedia of Cloud Computing, wiley
(2015)
19. Qi H, Gani A research on mobile cloud computing: review. Trend and Perspectives
20. Chang RS, Gao J, Gruhn V, He J, Roussos G, Tsai WT (2013) Mobile cloud computing
research—issues, challenges and needs. Serv Oriented Syst Eng (SOSE)
21. Fernando N, Loke S, Rahayu W (2012) Mobile cloud computing: A survey. Future Gener
Comput Syst 29(1):84–106
22. Roman R, Lopez J, Mambo M Mobile edge computing. In: Fog et al (eds) A survey and analysis
of security threats and challenges. Future generation computer systems
A Review on Mobile Cloud Computing Interoperability Issues … 333

23. Dinh H, Lee C, Niyato D, Wang P (2011) A survey of mobile cloud computing: architecture,
applications, and approaches. Wireless Communications Mobile Computing
24. [Link]
[Link]
Filter Bank Modulation in Massive
MIMO Scenario

S. Sruthi and J. Dhoulath Beegum

Abstract Advancement in the field of wireless communication technologies

emerges as a research area of increased interest due to the recent developments
in 5G and Internet of things. Newer and better methodologies of modulation will
help to achieve efficiency in energy and judicious use of spectrum. Filter bank multi-
carrier modulation scheme in the context of 5G is a better alternative to any of the
currently deployed schemes. This promises better intersymbol interference and inter-
carrier -interference factors, and hence, it is ideal for massive MIMO scenarios. A
comparative study of FBMC system with the presently deployed OFDM scheme is
also analysed.

Keywords Massive MIMO · FBMC · OFDM · MU-MIMO · Impulse response ·

5G

1 Introduction

As per the Cooper’s law, the amount of wireless voice and data communication is
increasing at an exponential pace. The Ericsson Mobility Report forecasts a 42%
rise in mobile data traffic from 2016 to 2022. So we need to meet the continuously
increasing demand and satisfy the rising expectations of service quality. Researchers
need to turn every stone unturned to design new revolutionary wireless network
technologies.
Massive multi-input multi-output (MIMO) wireless communication refers to the
idea of equipping the cellular base stations with a very large number of antennas.
Massive MIMO is a potential technology which allows for orders of improvement
in parameters like spectral and energy efficiencies. Hence, better and efficient mod-
ulation schemes are the need of the hour. The commonly used modulation scheme

S. Sruthi (B) · J. Dhoulath Beegum

Department of ECE, TKM College of Engineering, Kollam, Kerala, India
e-mail: sruthisree11@[Link]
J. Dhoulath Beegum
e-mail: jdhoulathf@[Link]
© Springer Nature Singapore Pte Ltd. 2020 335
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
336 S. Sruthi and J. Dhoulath Beegum

is Orthogonal Frequency-Division Multiplexing (OFDM). But it has some inherent

disadvantages like need of a cyclic prefix code, problem of spectral leakage, etc.
Filter bank multi-carrier (FBMC) is a multi-carrier transmission approach. It pos-
sesses advantages over OFDM. FBMC does pulse shaping on each sub-carrier, and
each sub-channel is then filtered. It does not require a cyclic prefix code which
increases bandwidth efficiency [1]. It also promises good spectral efficiency and bet-
ter data rate. So the large number of users can simultaneously use the data or voice
service with minimal interference.

2 Related Work

MIMO scheme is still a widely researched arena of technology which needs advance-
ments in various aspects of performance, security and efficiency to be met. Introduc-
tion of MIMO as a wireless service scheme can drastically enhance the capacity and
reliability of wireless systems. The acceptance to Internet of Things (IoT), which is a
network of networks, brought in the need for better and efficient wireless connectivity
schemes.
In recent years, the focus has been shifted to a newer version of MIMO with
enhanced capabilities. Multi-User MIMO (MU-MIMO) systems employ multiple
antennas at the base station. The expensive communication components are needed
only at the base station, and the user terminals can use relatively cheap single-
antenna devices [2]. Also, the performance of MU-MIMO system is less sensitive to
propagation environment than that in the case of point-to-point MIMO.
This paper focuses on filter bank multi-carrier (FBMC) modulation. Basically,
filter bank technique is an improvement over direct FFT mechanism [3]. OFDM and
its different variants require the property of orthogonality to be satisfied among all the
carriers, whereas FBMC needs the property of orthogonality only for adjacent sub-
channels. Also, OFDM makes use of a particular bandwidth of frequency with other
carriers, whereas FBMC uses the transmission channel associated with a particular
bandwidth to different sub-channels [4]. To put the channel bandwidth to complete
use, sub-channels are to be modulated in such a way as to adapt with the orthogonality
constraint of the neighbours and offset quadrature amplitude modulation.

3 System Model

Massive Multiple-Input Multiple-Output (MIMO) is a promising technological

advancement that is under consideration for the fifth generation (5G) of mobile sys-
tems. But this technology requires improvement in parameters like data rate, energy
efficiency and latency in comparison with the presently deployed LTE scheme [5]. In
particular, due to the effect of considerable side-lobe interference due of sub-carriers
in OFDM, it suffers from a high spectral leakage which results in higher out-of-band
emissions (Fig. 1).
Filter Bank Modulation in Massive MIMO Scenario 337

Fig. 1 Massive MIMO

scenario

Filter bank multi-carrier (FBMC) is a 5G candidate waveform which has better

spectral properties. The requirement of synchronization in uplinks can be relaxed, and
imbibing carrier aggregation into the system becomes easy. Due to these advantages,
the technique of FBMC is being actively studied to deduct the efficiency of the
technique in practical scenarios.
In the proposed design, d m,n denotes the discrete time data symbol transmitted
over the mth sub-carrier and nth time instant. The total number of sub-carriers is taken
as M. To nullify the possible interference between data symbols and to maintain the
jθ
orthogonality, the data d m,n is corrected in phase with the term em,n , where θ m,n =
(π/2)*(m + n). Thus, each symbol will have a (±π/2) difference in phase with its
nearby neighbours in both time and frequency by means of which orthogonality is
ensured. The output is upsampled and then given as input to the filter Fig. 2. Finally,

Fig. 2 Block diagram of the

filter design
338 S. Sruthi and J. Dhoulath Beegum

Table 1 Frequency domain filter coefficients

K H0 H1 H2 H3
√
2 1 2/2 – –
3 1 0.911438 0.411438 –
√
4 1 0.971960 2/2 0.235147

the pulses are shaped using the filter p(l), which has designed to be a Nyquist pulse
which has zero crossings at every Mth intervals [6–10]. Mathematically, the technique
that is described can be depicted as follows:

+∞ M−1
 
x(l) = dm, n am,n (l) (1)
n=−∞ m=0

where
 
nM ∧
am,n (l) = pm l − e jθm,n (2)
2

Half-Nyquist filters are used at the transmitter and receiver ends [3]. The half-
Nyquist filter with the corresponding frequency coefficients for K = 2, 3 and 4,
respectively, is as in Table 1.
The equation which gives impulse response h(t) of the filter is given by


K −1  
2π kt
h(t) = 1 + 2 H k cos (3)
k=1
KT

where H k indicates the various frequency coefficients for different number of sub-
carriers [11–13].

4 Results and Discussions

The FBMC modulation technique is implemented in the massive MIMO scenario.

The OFDM technique is also carried out. A comparative study between the modu-
lation techniques of FBMC and OFDM is done on the basis of spectral leakage, and
the subsequent results are obtained as follows.
The impulse response will help to get an idea about the behaviour of the system
to various inputs. The impulse response for the given filter is obtained as in Fig. 3.
Filter Bank Modulation in Massive MIMO Scenario 339

Fig. 3 Impulse response of the filter

It is pretty evident from the frequency response of the system Fig. 4 that the FBMC
system offers a lower spectral leakage in comparison with the OFDM system. The
OFDM plot shows a persistent intrusion of sub-bands in the required output band,
whereas FBMC plot shows a gradual drop in the influence of sub-bands on the main
output.
As frequency component increases, it can be observed that the spectral leakage
drastically falls off in FBMC whereas it is not so in the case of OFDM. So FBMC
is a reasonable approach for 5G systems where a large number of users and large

Fig. 4 Frequency response of the proposed system

340 S. Sruthi and J. Dhoulath Beegum

number of devices are to be supported. Figure 5 depicts the frequency responses of

the proposed FBMC system for varying number of sub-carriers (Fig. 6).
The response of the system for different number of sub-carriers is as plotted
above. It can be observed that when K = 2, even though the spectral leakage is poor,
the unwanted sub-bands have considerable strength. When K = 3, there is a drastic
reduction in the strength of sub-bands, which can be filtered-off with a moderate
filter. But when K = 4, it shows a very good response with minimal influence of

Fig. 5 Comparative study of FBMC with varying number of sub-carriers

Fig. 6 Comparison of system with different number of sub-carriers

Filter Bank Modulation in Massive MIMO Scenario 341

sub-bands. Therefore, it can be observed that, as the number of sub-carriers increase,

the output is getting narrower (Table 2).
It can be easily observed that, though there is a minimal difference between
the leakage values of OFDM and FBMC systems in the initial samples, but as the
succeeding samples are analysed, it becomes clear that the leakage values are very
low for FBMC in comparison with OFDM. The truncation of the required frequency
band is a demerit in case of OFDM (Fig. 7).

Table 2 Comparative study

Samples OFDM FBMC
of spectral leakage values at
different sample instants for 0.0307 0.99 0.9891
OFDM and FBMC systems 0.0368 0.9857 0.984
0.043 0.9805 0.9777
0.0491 0.9746 0.9703
0.0552 0.9679 0.9614
0.0614 0.9605 0.9512
0.0675 0.9523 0.9395
0.0736 0.9434 0.9261
0.0798 0.9337 0.9111
0.0859 0.9234 0.8943
0.092 0.9124 0.8758
0.0982 0.9007 0.8555
0.1043 0.8883 0.8335
0.1104 0.8753 0.8097
0.1166 0.8617 0.7843
0.1227 0.8475 0.7574
0.1289 0.8326 0.729
0.135 0.8173 0.6993
0.1411 0.8014 0.6685
0.1473 0.7849 0.6367
0.1534 0.768 0.6042
0.1595 0.7506 0.5712
0.1657 0.7327 0.538
0.1718 0.7145 0.5047
0.1779 0.6958 0.4715
0.1841 0.6767 0.4388
342 S. Sruthi and J. Dhoulath Beegum

Fig. 7 Spectral leakage comparison for OFDM system versus FBMC system

5 Conclusions

The performance of filter bank multi-carrier (FBMC) modulation in massive MIMO

scenario is studied and analysed. FBMC system is found to be advantageous since
it does not require a cyclic prefix code before transmission. The impulse response
of the filter and the frequency response of the system are plotted to compare with
the presently deployed OFDM system. Influence of sub-bands on the required fre-
quency component can be easily observed from the plot. The comparison of FBMC
system with varying number of sub-carriers is also plotted. As frequency component
increases, spectral leakage drastically falls off, and hence, inter-symbol interference
is minimal. So FBMC is a potential candidate for 5G systems. A comparative study
of FBMC and OFDM systems was carried out, and the parameters of spectral leakage
of both systems were compared and it is concluded that FBMC system is better in
all respects over OFDM scheme. Hence, efficiency of the system in terms of energy,
spectrum, hardware, cost and resource utilization is better.
Filter Bank Modulation in Massive MIMO Scenario 343

6 Future Scope

FBMC implementation in massive MIMO is a leading research topic in the inter-

national communication research arena. One of the future works on the topic will
be to develop a scheme to combat the imminent problem of pilot contamination in
the system so as to deploy it to incorporate dense heterogeneous networks. In the
near future, the need for distributed massive MIMO will become more demanding,
and hence, a study on the same can be carried out due to the integration of Internet
of things, visible light communication and related technologies. FBMC promises to
support carrier aggregation scheme.

References

1. Farhang A, Marchetti N, Figueiredo F, Miranda JP (2014) Massive MIMO and waveform

design for 5th generation wireless communication systems. In: IEEE 5GU
2. Machrouh Z, Najid A (2018) High efficiency IEEE 802.11ax MU-MIMO and frame aggre-
gation analysis. In: 2nd international conference on recent advances in signal processing,
telecommunications & computing (SigTelCom). pp 107–110
3. Farhang-Boroujeny B (2014) OFDM versus filter bank multicarrier. IEEE Signal Process Mag
28(3):92–112
4. Farhang-Boroujeny B (2014) Filter bank multicarrier modulation: a waveform candidate for
5G and beyond. Adv Electri Eng
5. Ott D, Himayat N, Talwar S (2017) 5G: transforming the user wireless experience. In: Towards
5G: applications, requirements and candidate technologies. pp 34–51
6. Aminjavaheri A, Farhang A, Doyle LE, Farhang-Boroujeny B (2017) Prototype filter design
for FBMC in massive MIMO channels. In: IEEE ICC signal processing for communications
7. Marzetta TL (2015) Massive MIMO: an introduction. Bell Labs Tech J 11–22
8. Abuibaid MA, Colak SA (2017) Energy-efficient massive MIMO system: Exploiting user
location distribution variation. AEU Int J Electron Commun72:17–25
9. Farhang A, Marchetti N, Doyle LE, Farhang-Boroujeny B (2014) Filter bank multicarrier for
massive MIMO. In: 2014 IEEE 80th vehicular technology conference (VTC2014-Fall)
10. Bellanger M, Le Ruyet D, Roviras D, Terr´e M, Nossek J, Baltar L, Bai Q, Waldhauser D,
Renfors M, Ihalainen T et al (2010) FBMC physical layer: a primer. PHYDYAS
11. Elijah O et al (2015) A comprehensive survey of pilot contamination in massive MIMO—5G
System. IEEE Commun Surv Tutorials 18(2)
12. Zafar A, Zhang L (2018) Spectrum efficient MIMO-FBMC system using filter output
truncation. IEEE Trans Veh Technol 67(3)
13. Banelli P, Buzzi S, Colavolpe G, Modenini A, Rusek F, Ugolini A (2014) Modulation formats
and waveforms for 5G networks: who will be the heir of OFDM?: An overview of alternative
modulation schemes for improved spectral efficiency. IEEE Signal Process Mag 31(6):80–93
Autonomous Farming—Visualization
of Image Processing in Agriculture

Shivam Thakur, Sushant Bawiskar, Sachin Kumar Singh

and M. Shanmugasundaram

Abstract Nowadays, it is important to automate the processes in farming. For effi-

cient farming, we can use embedded systems and IoT by which farming could become
like a video game. Robots in control are more than a simple case study. We are demon-
strating a simple example of Ploughing in which Tractor has its own vision, through
which it can identify the boundary of the field and can plough the field without any
driver. A farmer can easily operate various farming operations on his smartphone in
just one tap. The vision is able to detect the poles on the boundary of farm, and using
image processing, camera on tractor is able to detect the colour flag on a pole, and
by detecting the colour of the pole, it is able to turn by its own and plough the field
completely and stop. We can use this concept to automate the seeding, irrigation,
weeding, harvesting, delivery, etc., depending upon the crop.

Keywords Raspberry Pi · IoT · Image processing

1 Introduction

India is growing day by day in every sector such as manufacturing and IT sector
[6]. But India is still lagging somewhere in the agricultural sector. Although the
government is doing a lot of efforts to improve productivity and so to help farmers,
they are launching new schemes, but still, somewhere we have an option to use
technology that brings us a better future. It can boost the agriculture sector by not

S. Thakur (B) · S. Bawiskar · S. K. Singh · M. Shanmugasundaram

Department of Embedded Technology, VIT, Vellore, India
e-mail: shivam.thakur2018@[Link]
S. Bawiskar
e-mail: sushant.ravindra2018@[Link]
S. K. Singh
e-mail: sachinkumar.singh2018@[Link]
M. Shanmugasundaram
e-mail: phdsundaram@[Link]

© Springer Nature Singapore Pte Ltd. 2020 345

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
346 S. Thakur et al.

Fig. 1 Different variation in lands

only making convenient to farmers, but also the agriculture department can easily
monitor productivity and can suggest the farmer in real time.
Robots represent a natural source of inspiration and a great pedagogical tool
for research and teaching in control theory [5]. Here, we have introduced with the
concept of autonomous ploughing technique in which a camera on a tractor can be
able to detect the boundary of the land and can plough the land completely without
any driver using IoT. Since lands can be of any shape, it can have variations in level
as shown in Fig. 1.
The farmer just needs to keep some poles with colour in his farm boundaries, and
then, he has to just press one button on his mobile and he can sit back and relax.
His field will be ploughed automatically by using image processing and IoT [4]. We
made a prototype which works on the same principle by interfacing Raspberry pi [9]
and IoT as shown in Fig. 3.
We can also automate next steps in farming such as seeding, irrigation, and harvest-
ing [2]. Automation in farming can increase the productivity of agricultural sector,
and we can actually save a lot of fuel by making electric tractor and install multiple
sensors which can monitor the production rate in real time and using IoT. This real-
time data can be accessed by Agriculture Department using cloud computing [3]. A
farmer can get the real-time suggestions on mobile app and can work accordingly.
Humidity variations and temperature variations can cause a severe reduction in the
growth of a crop [8]; thus, it is important to know that when and what a crop needs
actually. In summer, field dries very quickly; thus, it is important to irrigate the field
more frequently, whereas, in winter and rainy season, it needs water less frequently.
So, in this way, we can achieve automation in irrigation and can develop various
automated techniques in farming.
Autonomous Farming—Visualization of Image Processing … 347

Similarly, IoT can play a major role in connecting everything to a cloud, and in
this way, we can get the efficiency of seeds by calculating at the time of harvesting.
Also, by knowing the environmental conditions, we can improve productivity by
implementing solutions for location-based challenges [11].
An improved version of modern agriculture is with the seeding machine the ground
it covers is faster than human [10]. Best chance seeds have to rooter and grow now the
farming comes in the digital arena with the IoT-based system which gives data to the
farmers. The whole farming land is planted by a single human monitoring process
over a control dashboard laptop or tablet which gives us proper output with the
desired accuracy, which is in the favour of the farmer. Basically, the proper planting
depends on two parameters: proper depth of seeds and spacing between plants. In the
early stage, these two problems come in front of farmers; by autonomous farming,
we can control these two parameters.

2 Related Work

Today’s farmers are using traditional methods for farming which is the main reason
for less productivity and wastage of electricity; some methods are introduced to over-
come these problems; automated farming robot is also a big concept in which attempt
is done to connect robotics and automation with agriculture. The main intention is
to reduce the effort of farmers encountered in the field. The main reason for using
electric vehicles is to reduce greenhouse gasses and carbon footprints.
The robots [12] are used to assist farmers in the agriculture field; the technology
used to control the system is Raspberry Pi which provides manual controlling of the
system. The system proposed is used for ploughing, dispensing, and fruit picking
[7].

3 Proposed Method

Here, we have made a prototype which has a ploughing mechanism and also a
Raspberry Pi and USB camera. As shown in Fig. 2 using OpenCV and NumPy, we
were able to differentiate between colours, and thus, we can program the tractor such
that it ploughs the complete field in a zigzag manner by detection colours on the
poles. If it detects the red colour, then it has to turn 180° right and plough the side
lane, and if it detects the blue colour, then it has to turn 180° from the left side and
again plough the side lane and so on; it completes the complete land irrespective
of field size and shape. Farmer has to just set poles on boundaries before starting
operation. And rest the autonomous tractor will do [1].
As shown in Fig. 3 if the camera detects the yellow colour, then the vehicle needs
to stop. The farmer needs to get connected with cloud server MQTT, and then, he can
turn off the ploughing in between whenever he wants. It is completely user-friendly
348 S. Thakur et al.

Fig. 2 Ploughing path

Fig. 3 Prototype

and like a social media platform where farmer is able to interact with the need of its
crops by getting information from a bunch of sensors installed on the field.

3.1 Flowchart

Here, we have demonstrated the connection of MQTT server in which if farmer turns
on the ploughing through tractor, then it will start and take input from the camera.
The camera tells it, where to go. As shown in Fig. 4 if tractor detects yellow colour,
then it needs to stop. Colour detection: Where the camera is used to capture and
also to provide input to code. Image captured by the camera is in BGR format, i.e.
Autonomous Farming—Visualization of Image Processing … 349

Fig. 4 Flowchart

red, green, and blue. All digital devices use this type of input, but for proper colour
detection, we are using HSV format.
HSV means hue, saturation, and value where hue is colour, saturation is greyness,
ad value means the brightness of the pixel. Saturation value near means it is dull or
grey looking.
The red colour in OpenCV has hue value approximately in the range of 0–
10 and 160–180. In OpenCV, value range for three matrices is 0–179(Hue),
0–255(Saturation), 0–5(Value) or (colour), (Saturation), (Brightness).
Saturation: Saturation value represents the amount to which that colour is mixed
with white.
Value: Value represents the amount to which that colour is mixed with black.
Figure 5 shows a flowchart for image processing in which morphological transfor-
mation is done to remove small noises in the image. Morphological transformations
are some operation which is based on the shape of images; for this, we need two
inputs: one is original image and another input is the kernel. Kernel decides the nature
350 S. Thakur et al.

Fig. 5 Image processing flowchart

of the operation be performed on the image. Two very basic morphological operators
are dilation and erosion. In this, we use dilation and a 5*5 kernel matrix. If at least
one pixel under kernel is one, it will simply increase colour region and reduce noise
in the image. Dilation adds a layer of a pixel to both inner and outer boundaries of
the region.
Now, contour the image previously described means differentiating each colour
with the rectangular bounded line which is called a-a contour. A simple line connects
continuous points, having the same colour or intensity. Contour is a tool mainly used
for shape detection and analysis and object detection and recognition.

4 Conclusion and Future Scope

The prototype is able to plough the field without any human effort. It is also designed
in such a way that it follows a specific predefined path by using colour detection.
Advance enhancements can be done by installing a drone in the prototype which
can fly on the field and can time to time monitor the status of crops. The drone can
provide a real-time video feed to the farmer. This can also be used to monitor land.
These drones can also be used for theft protection.
Autonomous Farming—Visualization of Image Processing … 351

References

1. Canudas de Wit C, Siciliano B, Bastin G (eds) (1996) Theory of robot control. Springer-Verlag,
New York
2. Liu JJ, Wu L (2014) The study on autonomous agricultural machinery. Modelling and control
method. Sens Transducers
3. Goldberg K (ed) (2000) The robot in the garden: telerobotics and telepistemology in the age
of the internet. MIT Press, Cambridge, MA
4. Ulrich I, Nourbakhsh I, Appearance-based obstacle. Detection with monocular color vision.
In: The proceedings of the AAAI national conference on artificial intelligence. Austin, TX
5. Canudas de Wit C, Siciliano B, Bastin G (eds) (1996) Theory of robot control. Spinger-Verlag,
New York
6. Thankachan S, Kirubakaran S (2014) E-agriculture information management system. Int J
Comput Sci Mobile Comput 3(5)
7. Umbaugh SE (2016) Digital image processing and analysis: human and computer vision
applications. With CVIP tools. CRC Press
8. Tillett ND, Hague T, Merchant JA (1998) A robotic system for plant scale husbandry. J Agric
Eng Res 69:169–178
9. Senthilkumar G, Gopalakrishnan K, Sathish Kumar V (2014) Embedded image capturing
system using raspberry Pi system. 3(2):213
10. Srivastava UK (1989) Agro-processing industries: potential, constraints and tasks ahead. Indian
J Agric Econ 44(3):42–256
11. Dwivedy N (2011) Challenges faced by the agriculture sector in developing countries with
special reference to India. Int J Rural Stud 18(2)
12. Siegwart R (2004) Introduction to autonomous mobile robots. [Link], Cambridge, MA
Design and Development of End Effector
for Domestic Robots

V. Indu, Putchala Vinay, Narjala Rohith, Kuppili Puneeth and S. Pramod

Abstract The aim of this work is to design and develop efficient and low-cost
end effector for holding and handling of fragile objects, which is to be used in the
industrial automation and mainly for the purpose of domestic robots. The design
consists of three systems: the sensor system, the control module, and the robotic end
effector. Different objects have different stress-handling capabilities, when pressure
is applied on the object goes beyond the yield point, and then object either deforms
or breaks. So, the aim of this work is to utilize that strain being applied to the object
for holding the object and handling it carefully. This end effector works with a single
degree of freedom. The end effector is designed to have three grasping fingers out
of which one is stable with the sensor on it and other two fingers are attached to a
single joint with common motion to both fingers.

Keywords Robotic gripper · End effector · Manipulation · Strain sensor

V. Indu · P. Vinay (B) · N. Rohith · K. Puneeth

Department of Electrical and Electronics Engineering, Amrita Vishwa Vidyapeetham, Amritapuri,
India
e-mail: vinayputchala@[Link]
V. Indu
e-mail: induv@[Link]
N. Rohith
e-mail: rohith@[Link]
K. Puneeth
e-mail: puneeth@[Link]
S. Pramod
Department of Mechanical Engineering, Amrita Vishwa Vidyapeetham, Amritapuri, India
e-mail: pramods@[Link]

© Springer Nature Singapore Pte Ltd. 2020 353

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
354 V. Indu et al.

1 Introduction

Robots are used to ease human processes; they are widely used in the field of indus-
tries and also used for domestic purposes. Household robots can perform many tasks
such as cleaning your house, mowing your lawn, and many more. Nowadays we can
see the increasing importance in industrial automation where the robotic processes
form the crux of this evolution.
End effector is designed to handle various objects of varying shapes and sizes.
While designing the end effector, object handling should be taken into primary focus
as there is a chance of contortion of the object. With this, we also need to take proper
discretion in checking its domestic and industrial needs.
The usage of mechanical grippers is the most common method of making end
effectors. Here, perpetual usage of force on the object by the end effector is used to
retain the object [1]. But these may not handle fragile objects without contorting them
as mechanical grippers are not touch sensitive as our human hands [2]. The usage
of programming to hold and release various objects of different shapes and sizes
with disparate touch sensitivities is a strenuous task [3]. A five-finger robot gripper
with wire, five fingers of the robot using ultrasonic motors, and various other robot
hands are being researched. These types of end effectors are tough to build because
of the intricate structure and due to the lack of assurance in safe handling [4]. The
task of handling fragile household objects such as eggs and vegetables intact is still a
challenge in today’s world. As mentioned above, the most common end effectors are
made up of mechanical linkages which follow a very rough approach in handling but
they are not very precise. The gripper is inspired from the most efficient end effector
which is the human hand [5]. The usage of polymer strings with servo motors can
be a great help in handling delicate objects but its high cost makes it unlikely for
utilizing in domestic robots [6]. The usage of depth sensor for finding the end effector
data points through the angle and position of the end effector is set [7].
The motto of this work is to develop a flexible, cost-efficient gripper that can
handle fragile objects without deformation. So the most important design objective
in our work is to utilize the strain, as it varies from object to object.

2 Working of the Gripper

The end effector developed works on a closed-loop control system, where the mea-
sured stress value from the stress sensor acts as the loop. Initially, reference stress,
i.e., stress where an object can be manipulated, is fed to the microcontroller. Now
the microcontroller generates the required PWM pulses with respect to error which
is calculated using the difference in measured stress and the reference stress. Then
the servo motor which is the actuator converts the PWM signals to the motion, where
the motion occurs in angular motion in angles (degree). As the servo is connected to
gripper, gripper moves and applies force on the object. The force is being applied in
Design and Development of End Effector for Domestic Robots 355

Fig. 1 Block diagram

the form of stress, so stress is being applied on the object, and the stress is perpetually
being cross-checked with the reference value. Every time when the reference value is
greater than the stress, that difference is given to microcontroller as shown in Fig. 1.
The microcontroller then feeds the pertinent PWM pulse to produce angle through
servo and which in the process is converted to stress as described above. Once
reference value is equal to the stress value applied, then the microcontroller stops
giving PWM input to the servo motor which in return emulates the same zero output
due to no input and maintains the stable position where the object can be manipulated
for pick and place actions.

3 Algorithm

Initialize the control pins in the microcontroller, which is used to send or receive
the data. Calibrate the sensor to reduce the error while obtaining the measured stress
value. The reference stress value of the object that is to be handled by the gripper is
initially fed to the microcontroller.
Now as mentioned in the working of the gripper, the PWM generated is converted
into angular motion, then the gripper applies force on the object. The stress that is
being applied on the object is perpetually being cross-checked with the reference
value. Every time when the reference value or the threshold is greater than the stress,
that difference is given to microcontroller.
Microcontroller unit generates the PWM pulses so that it can increment the angular
motion of servo with the angle of one degree. And the loop continues till the error in
the stress becomes zero; i.e., both the measured value and the reference stress values
are the same (Fig 2).

4 Threshold Measurement and Tabulation

Stress values where the object can be held and manipulated are measured multiple
times and are tabulated. Using the tabulated values, the average is calculated to take
that value as the reference threshold value for each object (Table 1).
356 V. Indu et al.

Fig. 2 Algorithm flow chart

Table 1 Reference threshold

Object Stress (lbs)
Plastic jar 4.23
Small plastic container 4.81
Steel glass 4.7
Plastic water bottle 2.62

5 Experimental Results

Different objects have different manipulated stress, so to check the stable position
for holding the object the measurement was in angle (degree). So, we checked the
angular position of object with respect to stress being applied on the object. Plot of
angle versus stress is shown in Fig. 3 for plastic jar, Fig. 5 for plastic container, Fig. 7
for Steel glass, and Fig. 9 for plastic water bottle. The experiment had been repeated
multiple times to check if the stress being applied is stable. Plot of time versus stress
Design and Development of End Effector for Domestic Robots 357

Stress(lbs) 3

0
135 140 145 150
Angle(degree)

Fig. 3 Angle versus stress plot for plastic jar

is shown in Fig. 4 for plastic jar, Fig. 6 for plastic container, Fig. 8 for Steel glass,
and Fig. 10 for plastic water bottle.

5.5

5
Stress(lbs)

4.5

3.5

3
100 200 300 400 500 600 700 800 900 1000 1100
Time(ms)

Fig. 4 Stress versus time during multiple manipulation periods for plastic jar

6
5
4
Stress(lbs)

3
2
1
0
135 140 145 150 155 160
Angle(degree)

Fig. 5 Angle versus stress plot for small plastic container

358 V. Indu et al.

6
5.5
5
Stress(lbs)
4.5
4
3.5
3
100 200 300 400 500 600 700 800 900 1000 1100
Time(ms)

Fig. 6 Stress versus time during multiple manipulation periods for the small plastic container

4
Stress(lbs)

0
135 140 145 150 155 160 165
Angle(degrees)

Fig. 7 Angle versus stress plot for steel glass

5.5

5
Stress(lbs)

4.5

3.5

3
100 200 300 400 500 600 700 800 900 1000 1100
Time(ms)

Fig. 8 Stress versus time during multiple manipulation periods for steel glass
Design and Development of End Effector for Domestic Robots 359

3
2.5

Stress(lbs)
2
1.5
1
0.5
0
135 140 145 150 155 160 165 170 175
Angle(degree)

Fig. 9 Angle versus stress plot for plastic water bottle

4
3.5
Stress(lbs)

3
2.5
2
1.5
1
100 200 300 400 500 600 700 800 900 1000 1100
Time(ms)

Fig. 10 Stress versus time during multiple manipulation periods for plastic water bottle

6 Conclusion

In this paper, we introduced a simple and easy method to handle fragile objects using
their stress values. Using the reference stress values, the objects can be easily manip-
ulated to move from one position to another without deformation, and algorithm
implementation procedure has been discussed in this paper.

References

1. Sushmit AS, Haque FM, Shahriar Md, Bhuiyan SAM, Sarkar MAR (2017) Design of a gesture-
controlled robotic gripper arm using neural networks. In: IEEE international conference on
power, control, signals, and instrumentation engineering
2. Haque FM, Sushmit AS, Sarkar MAR (2017) Design of a voice-controlled robotic gripper arm
using neural network. In: International conference on energy, communication, data analytics,
and soft computing
3. Venter D, Dirven S (2017) Self morphing soft-robotic gripper for handling and manipulation of
delicate produce in horticultural applications. In: 24th international conference on mechatronics
and machine vision in practice (M2VIP)
4. Lee YC, Lim SJ, Hwang SW, Han CS (2009) Development of the robot gripper for a home
service robot. In: ICROS-SICE international joint conference 2009
360 V. Indu et al.

5. Choi MS, Lee DH, Park H, Kim YJ, Jang GR, Shin YD, Park JH, Baeg MH, Bae JH (2017)
Development of multi-purpose universal gripper. In: Proceedings of the SICE annual conference
(2017)
6. Pai UJ, Sarath NP, Sidharth R, Kumar AP, Pramod S, Udupa G (2016) Design and manufac-
ture of 3D printec myoelectric multi-fingered hand for prosthetic application. In: International
conference on robotics and automation for humanitarian applications (RAHA)
7. Megalingam RK, Vivek GV, Bandyopadhyay S, Rahi MJ (2017) Robotic arm design, devel-
opment and control for agriculture applications. In: 4th international conference on advanced
computing and communication systems (ICACCS)
Total Variation and Alternate Direction
Method for Deblurring of Digital Images

S. Rinesh, C. Prajitha, V. Karthick and S. Palaniappan

Abstract Images captured using smartphones and video cameras are recorded and
can be used anywhere and at any time. While taking a quick shot or while capturing
the moving objects, it may lead to the motion blurred images. In order to recover
the sharp images from motion blurred images, blind motion deblurring can be used.
Motion deblurring can be done by knowing both edge and non-edge of motion blurred
images. Edge and non-edge are the two methods used in total variation and alternate
direction method for deblurring of digital images. Step edges can be predicted and
detected by using edge-specific method. In non-edge method, it explores various
image statistics, such as the prior distributions and it is sensitive to statistical vari-
ation over different images. Both methods are used in large dataset images, but it
fails extremely in simple images. To overcome this problem, total variation (TV)
based regularization method is used which is followed by an iteratively reweighted
algorithm based on alternating direction method. To get higher results, LSED predic-
tion—based technique is employed, which first of all restores sharp edges and then
uses them to estimate initial kernel that traps the optimization of local minimum
corresponding to sharp images.

Keywords Image restoration · Total variation · Alternating direction method ·

LSED prediction based method

S. Rinesh (B) · V. Karthick · S. Palaniappan

Department of Computer Science and Engineering, Saveetha School of Engineering, Saveetha
Institute of Medical and Technical Sciences, Chennai, India
e-mail: [Link]@[Link]
V. Karthick
e-mail: vkarthick86@[Link]
S. Palaniappan
e-mail: [Link]@[Link]
C. Prajitha
Department of Biomedical Engineering, Saveetha School of Engineering, Saveetha Institute of
Medical and Technical Sciences, Chennai, India
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 361

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
362 S. Rinesh et al.

1 Introduction and Related Work

The vital challenge of blind motion deblurring is that, the amount of unknown is
far larger than the amount of accessible measurements. Given a blurred image, we
would like it to figure out its sharp version and also the blur kernel.
Molina et al. [1] projected dirichlet distribution to model the blurring operate
with smoothness constraints on the improved pictures to resolve blind deconvolution
downside. MAP calculator is employed in modeling the original image [1]. Samson
et al. [2] proposed the method for image classification that is mainly based on edge
preserving regularization process. Variational model is predicted on regularization
theory and mechanical action theory [2]. Likas and Galatsanos [3] proposed the
expectation maximization (EM) algorithm reaps all the benefits of a full Bayesian
model. This algorithm provides better improvement when compared to BID algo-
rithm. Due to computational complexity and convergence assessment of markov
chain it is difficult to implement BID algorithm [3]. Koschan and Abidi [4] pro-
posed, Vector-valued techniques, which are used in detecting the edges in color
images. Fergus et al. [5] proposed, conventional blind deconvolution and frequency-
domain constraints on images. This is mainly focused on kernel estimation. But
in non-blind deconvolution method, there will be some artifacts; this can also be
improved [5]. Joshi et al. [6] projected sharp edge formula, which will be more
useful to measure blur in the restricted device resolution by estimating a sub-pixel,
super-resolved PSF even for in-focus images. If there is multi peak kernel it fails
to sight edges [6]. Cho and Lee projected quick motion deblurring. Latent image
estimation and kernel estimation are mainly used in prediction of step edges [7]. Xu
and Jiaya [8] iterative support detection (ISD) kernel refinement, TV deconvolution
model is used to handle narrow structure exist in latent images. The edge-specific
theme depends on the economical detection or prediction of large-scale step edges
(LSEDs), detection-based strategies [9], assume that sharp explanations is favored
by the distributed previous for LSEDs. The detection of LSEDs will cause the gen-
eration of a pointy version of the input blurred image, this assumption holds only
tiny windows around LSEDs. Prediction—based ways adopt sharpening filters [9]
inverse random model to revive LSEDs.
LSED detection-based strategies assume that sharp explanations area unit favored
by around step edges. LSED prediction—based way first of all restore sharp step
edges and so use them to estimate a decent initial kernel, that traps the improve-
ment into local minimum corresponding to sharp resolution. The most usually used
approach to revive step edges is that the shock filter. LSED work well for images
with straight forward textures and infrequently fail to handle extremely rough texture
pictures. The performance of edge-specific theme is greatly restricted by its inability
to recover large kind of image edges.
The non-edge-specific theme on the other hand is not designed to hold out deblur-
ring based on detection or prediction of LSEDs. Adopting image measurements to
favor sharp explanations [10], however such measurements work just only for little
variety of natural images. Marginalizing the thin previous distribution [5, 11] proved
Total Variation and Alternate Direction Method for Deblurring … 363

that this approach results in actually handling the condition that the image size is far
larger than the kernel size.
To address the matter of each edge specific and non-edge-specific scheme, a unique
non-edge specific adaptive scheme [NEAS] is projected. NEAS is combination of
marginalization and LSED method. NEAS work just for large dataset images, it fails
to handle simple images, each of those strategies is unable to supply higher results
because of lack of edges, so this makes the kernel estimation unreliable. NEAS
produces top quality of results for large dataset images. To overcome such problem,
total variation (TV) and an iteratively reweighted algorithm supported by alternating
direction method (ADM) is employed.
TV has proven to be a valuable construct in reference to the recovery of pictures
that include piecewise smooth elements. TV regularization is standard in image
restoration and reconstruction because of its ability to preserve image edges.

2 Proposed Scheme

Blind motion deblurring is an important subject to image processing community.

Image deblurring could be a well-known ill-posed inverse drawback, increasing
attention from several sectors. Image degradation process can be modeled as:

y = k⊗x + n

where,
y observed blurred image,
k blur kernel,
x is the latent sharp image,
n is the image noise and
⊗ denotes the convolution operator
To cope with such deblurring problem, many regularization techniques are used.
Total variation (TV) based regularization method uses [14], non convex first and
second–order regularization is used in this proposed method and it is given by,
μ  v
min H f − g22 + ς D f v11 + (1 − ς ) D 2 f 12
2
μ > 0 is a regularization parameter. v1 v2 are hyper–laplacian distribution of first
and second order derivatives. D, D 2 are the finite difference operator of first and
second order. ς is the function which is used to preserve the image details in texture
and edge region which should be close to 1. ς is achieved based on eigenvalues of
the hessian matrix. Hessian matrix is used for kernel estimation.
364 S. Rinesh et al.

J ⎡ ⎤
Dx x (G σ ∗ f k ) Dx y (G σ ∗ f k ) ⎦
σ ( f k )=⎣
D yx (G σ ∗ f k ) D yy (G σ ∗ f k )

f k is the image with motion blur, G σ denotes Gaussian distribution function, Dx x

is the first order derivative with respect to x, Dx y is the first order derivative with
respect to xy, D yx is the first order derivative with respect to yx, D yy is the first order
derivative with respect to y.
An iteratively reweighted algorithm based on alternating direction method is fol-
lowed by large-scale step edge detection method, which uses shock filter to restore
step edges in the blurred image.

2.1 Algorithm

INPUT
Blurred image g,
Hessian matrix,
Parameters μ, β1 , β2 , β3 , v1 , v2 ,
INITIALIZATION

f 0 = u 0 = g, V0 = D f 0 , W0 = D 2 f 0 , ω0 = 0, λ0 = 0, ξ = 0

While stopping criterion is not satisfied do

  
  ςk Ψ1
(1) Compute Vk+1 = max D f k + ωβ1k  − βk
2
D f k is the first order derivative of blurred image, ωk is the initialization according to
iteration β1 is the parameter, ςk denotes the kernel estimation for first order derivative,
Ψ1 is the first order derivative of blurred image.
  
 
(2) Compute wk+1 = max D f k + ωβ1k  (1−ς β2
k)

2
(1 − ςk ) is the kernel estimation for first order derivative, Ψ2 denotes second order
derivative of blurred image

(3) Compute u k+1 = f k + βξk3

End while.
LARGE-SCALE STEP EDGE DETECTION METHOD
The LSED of xm is sharpened using the shock filter, xm = xm −
sign(xm ) γ  f γ (xm )dt.
Here  is laplacian operator dt = 0.8 [7, 15].xm denotes the convolution of
Gaussian point spread function with blurred image.  f γ (xm ) is the normalization
of gradient operation with respect to x, y.
Total Variation and Alternate Direction Method for Deblurring … 365

3 Experimental Results

The experiments on the large dataset are shown in Fig. 1. Total variation and alternate
direction method handles both large dataset and simple images. We used four images
from large dataset. In the commencement, in the first step the original image is motion

Fig. 1 In large dataset, four images are compared. From left to right are a the sharp images, b motion
blurred images (30,20), kernel estimation and c the output produced by our method TV and ADM
366 S. Rinesh et al.

Fig. 2 Results produced for simple image 1. From left to right are a the sharp images b motion
blurred images (30,20), kernel estimation and c the output produced by our method TV and ADM

blurred, and blurred image was restored using total variation and alternate direction
method (Figs. 2, 3, Tables 1, 2).

Fig. 3 Results produced for simple image 2. From left to right are a the sharp images b motion
blurred images (30,20), kernel estimation and c the output produced by our method TV and ADM

Table 1 Parameter analysis

Images in large Size PSNR (dB) PSNR (dB)
for images in large dataset
dataset [15]
Image 1 256 × 256 31.6 24
Image 2 256 × 256 26.5 24.2
Image 3 512 × 512 28.5 25
Image 4 512 × 512 30.2 26.2
Total Variation and Alternate Direction Method for Deblurring … 367

Table 2 Parameter analysis

Simple images Size PSNR (dB) PSNR (dB) [15]
for simple images
Image 1 256 × 256 33.3 24
Image 2 256 × 256 31.3 24.1

4 Conclusion

In this proposed work, total variation (TV) based regularization method was pro-
posed, which could preserve edges and the kernel estimation for motion blurred
images is done. TV method is followed by an iteratively reweighted algorithm based
on alternate direction method. TV and alternate direction method is used to handle
the images in large dataset and also on extremely simple images.

References

1. Molina R, Katsaggelos AK, Abad J, Mateos J (1997) A Bayesian approach to blind deconvolu-
tion based on Dirichlet distributions. In: Proceedings of international conference on acoustics,
speech and signal processing, pp 2809–2812
2. Samson C, Blanc-Feraud L, Aubert G, Zerubia J (2000) A variational model for image
classification and restoration. IEEE Trans Pattern Anal Mach intell 22(2):460–472
3. Likas AC, Galatsanos NP (2004) A variational approach for Bayesian blind image deconvolu-
tion. IEEE Trans Signal Process 52(8):2222–2233
4. Koschan A, Abidi M (2005) Detection and classification of edges in color images. IEEE Signal
Process Mag 22(1):64–73
5. Fergus R, Singh B, Hertzmann A, Roweis ST, and Freeman WT (2006) Removing camera
shake from a single photograph. ACM Trans Gr 25(3):787–794
6. Joshi N, Szeliski R, Kriegman D (2008) PSF estimation using sharp edge prediction. In:
Proceedings of IEEE conference on computational vision pattern recognition, pp 1–8
7. Cho S, Lee S (2009) Fast motion deblurring. In: Proceedings SIGGRAPH Asia, pp 1–8
8. Xu L, Jiaya J (2010) Two-phase kernel estimation for robust motion deblurring. In: Proceedings
of 11th European conference on computer vision, pp 1–14
9. Harmeling S, Hirsch M, Scholkopf B (2010) Space-variant single-image blind deconvolution
for removing camera shake. In: Advances in neural information processing systems. Cambridge
10. Krishnan D, Tay T, Fergus R (2011) Blind deconvolution using a normalized sparsity measure.
In: Proceedings of IEEE conference on computer vision and pattern recognition, pp 233–240
11. Levin A, Weiss Y, Durand F, Freeman WT, (2011) Efficient marginal likelihood optimization
in blind deconvolution. In: Proceedings of IEEE conference on computer vision and pattern
recognition, pp 2657–2664
12. Levin A, Weiss Y, Durand F, Freeman WT, (2009) Understanding and evaluating blind decon-
volution algorithms. In: Proceedings of IEEE international conference on computer vision, pp
1–8
13. Kundur D, Hatzinakos D (1996) Blind image deconvolution. IEEE Signal Process Mag
13(3):43–64
14. Liu RW, Xu T (2013) A robust alternating direction method for constrained hybrid variational
deblurring model. In: Proceedings of computer vision and pattern recognition, pp 1–8
15. Wang C, Yue Y (2013) Non edge-specific adaptive scheme for highly robust blind motion
deblurring of natural images. IEEE Trans Image process 22(3):884–893
368 S. Rinesh et al.

16. Krishnan D, Fergus R (2009) Fast image deconvolution using hyper-Laplacian priors. In:
Advances in neural information processing systems. MIT Press, Cambridge, MA
17. Whyte O, Sivic J, Zisserman A, Ponce J (2010) Non-uniform deblurring for shaken images.
In: Proceedings of IEEE conference on computer vision pattern recognition, pp 491–498
18. Roth S, Black MJ (2007) Steerable random fields. In: Proceedings of IEEE international
conference on computer vision, pp 1–8
19. Ayers GR, Dainty JC (1988) Interative blind deconvolution method and its applications. Opt
Lett 13(7) pp 547–549
20. Ankit G, Neel J, Larry Z, Michael C, Brian C (2010) Single image deblurring using motion
density functions. In: Proceedings of 11th European conference on computer vision, pp 171–
184
Design of MIMO Triangular Microstrip
Patch Antenna for IEEE 802.11a
Application

M. Arulaalan, K. Mahendran, P. Prabakaran and G. Manikannan

Abstract A compact multiple-input-multiple-output (MIMO) antenna for wireless

local area network (WLAN) applications with dual-band characteristics is presented.
The proposed antenna is composed of two microstrip line-fed fractal triangular
microstrip antenna. To achieve good return and high isolation, a Z-shaped stub is
added in the ground plane. To reduce the mutual coupling and ECC, a Z-shaped
stub is added to the ground plane. The proposed antenna covers the bandwidths of
WLAN 5.2 GHz (5.11–5.23 GHz) and 5.8 GHz (5.72–5.92 GHz). A high isolation
over −25 dB is achieved for both the IEEE 802.11a bands. The MIMO antenna
return loss, envelope correlation coefficient, gain and radiation characteristics are
also investigated. The results indicate that the MIMO antenna is suitable for WLAN
applications. The geometry of the proposed dual-band WLAN MIMO antenna has
an overall size of 68 × 26 × 1.6 mm3 .

Keywords Multiple-input-multiple-output (MIMO) · Envelope correlation

coefficient (ECC) · Triangular microstrip antenna (TMSA)

1 Introduction

The wireless communication medium is very complicated. The signal from the trans-
mitting antenna over a wireless communication channel undergoes severe fluctua-
tions by fading the signal level, path loss, co-channel interference, etc. The bandwidth

M. Arulaalan · K. Mahendran (B) · P. Prabakaran · G. Manikannan

CK College of Engineering & Technology, Cuddalore, India
e-mail: srimahendrancs@[Link]
M. Arulaalan
e-mail: arulaalan@[Link]
P. Prabakaran
e-mail: [Link]@[Link]
G. Manikannan
e-mail: kannan.gr18@[Link]

© Springer Nature Singapore Pte Ltd. 2020 369

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
370 M. Arulaalan et al.

limitation is a major challenge to a designer in designing a system with high quality

and high spectral efficiency at low cost in single-input-single-output (SISO) system.
The two factors, improved spectral efficiency and high quality, can be achieved by
an MIMO system which is not possible in SISO system. To achieve spatial diversity
and spatial multiplexing in a MIMO system, multiple antennas are required at the
transmitting and receiving ends to improve the reliability and data rate [1]. MIMO
antennas increase the data rate and range compared to SISO using the same radio
transmits power.
MIMO can improve all forms of wireless communication system, but the design
of MIMO system is more complicated than SISO. Antennas designed for MIMO
systems require high decoupling between antenna elements, i.e., very low mutual
coupling and low correlation coefficient. In a MIMO system, antenna is an integral
part of the system.
The return loss, gain and radiation pattern are considered for the single antenna
element, but in the design of MIMO antenna, mutual coupling and ECC are the
important factors. Lower mutual coupling between antennas ensures that reliability
and data rate of the system are improved. ECC gives the information of how radiation
pattern of two antennas differs. The ECC value is zero if one antenna is horizontally
polarized and the other one vertically polarized, and similarly, one antenna radiates
toward the sky, and the other antenna radiates in ground surface. The ECC value
given in Equation can be calculated by using S parameter [1–5] without considering
radiation pattern.
 ∗ 
S S12 + S∗ S22 2
ECC =  11
 21
 .
 1 − |S11 |2 − |S21 |2 1 − |S22 |2 − |S12 |2 

There are various methods proposed to reduce the mutual coupling. The simplest
method of reducing the mutual coupling is increasing the separation between the
antennas, but the drawback is the size of the antenna increases. The various methods
used for reducing mutual coupling are to add single negative magnetic metamaterials
between antenna elements [6–8] and adding stub in the ground plane [9–11], using
slot and stub technique [12] and placing antenna elements orthogonally.
Microstrip antenna has its radiating patch with different shapes namely rectangu-
lar, triangular, circular, square, elliptical, annular ring, etc. The TMSA consists of a
triangular-shaped radiating patch on the top of the dielectric material and the ground
plane at the bottom of the dielectric material. The triangular radiating patch can be
equilateral, right-angled triangle, etc.. Helszajn and James initially introduced trian-
gular patch antenna structures. The TMSA structures have less radiation loss, and
radiation pattern of equilateral triangular microstrip antenna (ETMSA) is relatively
broad.
The TMSA has the advantage of being smaller and occupies half of the metalized
area of the patch at a fixed frequency compared to rectangular, square or circular
microstrip antenna. TMSA has the advantage of reduced mutual coupling between
adjacent antenna elements in an array. Size compactness with high value of directivity
Design of MIMO Triangular Microstrip Patch Antenna … 371

can be achieved in ETMSA. Of the different patch shapes, triangular is the best suited
for space and place problem. The partial Koch is added to enhance the performance
of the ETMSA. The details of antenna design, simulated and measured results are
discussed in the following section.

2 Antenna Design

2.1 Antenna Configuration

The geometry of the proposed dual-band WLAN MIMO antenna, with an overall
size of only 68 × 26 × 1.6 mm3 , is shown in Fig. 1. It is designed on an FR4
substrate, with a thickness of 1.6 mm and relative permittivity of 4.4. The antenna
consists of two triangular antenna elements with Koch fractal as shown in Fig. 1,
with microstrip, fed through ports 1 and 2, respectively. The two triangular antenna
elements are printed parallel to each other with Z-shaped stub on the ground plane to
provide good isolation between the two antenna ports. The two antenna elements with
Koch fractal have identical dimensions with the side length a = 15.4 mm. The two
antennas are fed by a microstrip line with an impedance of 50 . The ground plane
of the antenna is printed on the other side of the FR4 substrate. To enhance isolation
and return loss, a Z-shaped stub is added in between the ground plane of the antenna
as shown in Fig. 1. The simulation of the MIMO antenna is carried out using ADS
for optimization of return loss, gain, −10 dB impedance bandwidth, mutual coupling
and radiation pattern. The optimized dimension of SISO for dual-band IEEE 802.11a
application is extended for MIMO capability [13].

Fig. 1 Proposed layout of

WLAN MIMO antenna with
Z-shaped stub
372 M. Arulaalan et al.

Fig. 2 Layout of WLAN

MIMO antenna without stub

2.2 WLAN MIMO Antennas

The dual-band antenna for IEEE 802.11a is extended for MIMO configuration as
shown in Fig. 2. The performance of the MIMO antenna is improved by placing the
antenna elements side by side separated by a small distance without stub.

3 Results and Discussion

3.1 Simulated Results

The two dual-band antenna elements are symmetric and have the same optimized
parameters obtained for the dual-band antenna. When port 1 is excited, the MIMO
antenna resonated at 5.191 GHz and 5.794 GHz with S11 value of −25.42 dB and
−33.950 dB, respectively, as shown in Fig. 3. The simulated dual-band −10 dB

Fig. 3 Simulated S11

parameter when port 1 is
excited without stub
Design of MIMO Triangular Microstrip Patch Antenna … 373

Fig. 4 Simulated isolation

coefficient S21 parameter
when port 1 is excited
without stub

impedance bandwidth is 160 MHz (5.08 GHz–5.24 GHz) and 20 MHz (5.72–
5.92 GHz) for IEEE 802.11a application. The simulated gains of the antenna for
two resonant frequencies are 8.7 dB and 8.2 dB, respectively. The other important
parameter in measurement of MIMO antennas is the isolation or mutual coupling
between the input ports. When port 1 is excited, the isolation value S21 is −22.38 dB
and −25.30 dB for the two resonant frequencies which are shown in Fig. 4. The
isolation value is greater than −19 dB for two bands and for the entire −10 dB
bandwidth.
The ECC plays an important role in evaluating the diversity characteristics of
MIMO system. The simulated value of ECC is 0.6. The antenna has good diversity
performance if ECC is <0.7. The simulated 2D radiation pattern for the MIMO
antenna for frequencies 5.2 and 5.8 GHz is shown in Fig. 5 which is nearly
omnidirectional.

3.2 WLAN MIMO Antenna with Stub

To enhance the isolation between the ports, Z-shaped stub is introduced in the ground
plane. The significance of the stub is compared with the MIMO antenna without stub.
There are many types of stub used for isolation, three stubs Y shaped stub, two long
protruding ground stubs and short ground strip are used for UWB application to
enhance isolation. For MIMO capability Z-shaped stub is added to the ground plane
in between the ground plane of the two antennas to further enhance isolation. The
layout of the antenna with Z-shaped stub is shown in Fig. 6. The return loss of the
antenna improved for both the bands with the help of the stub. The fabricated WLAN
MIMO antenna front and back view is shown in Figs. 6, 7, respectively.
374 M. Arulaalan et al.

Fig. 5 2D radiation pattern

a 5.2 GHz. b 5.8 GHz

3.3 Measured Results

The antenna measurements are carried by using vector network analyzer (VNA Agi-
lent N9917A). The measured value of S11 is −34 and −30 dB at 5.19 and 5.8 GHz.
The measured −10 dB bandwidth for the two bands is 120 MHz (5.11–5.23 GHz)
and 200 MHz (5.72–5.92 GHz) as shown in Fig. 8.
Design of MIMO Triangular Microstrip Patch Antenna … 375

Fig. 6 Photograph of the fabricated WLAN MIMO antenna—front view

Fig. 7 Photograph of the fabricated WLAN MIMO antenna—back view

Using this Z-stub technique, the measured S21 is −34, and −32 dB is achieved for
the two operating bands of 5.11–5.23 GHz and 5.72–5.92 GHz with center frequen-
cies of 5.18 GHz and 5.81 GHz, respectively, refer to Fig. 9. The measured values
with stub are −22.38 and −25 dB for the two bands as shown in Fig. 9. The increase
in isolation value is due to the fact that current is absorbed by the Z-shaped stub,
and thus, it enhances the port isolation between the antenna elements. The simulated
gain of the antenna with stub is 8.23 and 8.10 dB. The value of ECC is 0.4 due to
reduced mutual coupling. The 2D radiation pattern shown in Fig. 10 is nearly an
omnidirectional pattern. Table 1 gives the comparison of WLAN MIMO antenna
without using stub.
376 M. Arulaalan et al.

Fig. 8 Simulated and measured S11 parameter when port 1 is excited with stub

Fig. 9 Simulated and measured isolation coefficient S21 parameter when port 1 is excited with stub

3.4 Summary

A printed MIMO antenna for WLAN applications is designed and developed. The
antenna consists of two identical fractal-based triangular microstrip antenna etched
on the top of the substrate. The isolation between the antenna elements is enhanced
by adding a Z-shaped stub on the ground plane. The proposed arrangement results
in high isolation with values better than 20 dB. The factors such as return loss, gain
and ECC indicate good performance for MIMO system.
Design of MIMO Triangular Microstrip Patch Antenna … 377

Fig. 10 Simulated 2D
radiation pattern of the
proposed WLAN MIMO
antenna a 5.2 GHz.
b 5.8 GHz
378 M. Arulaalan et al.

Table 1 Comparison of WLAN MIMO antenna with and without stub

Parameters MIMO antenna MIMO antenna with stub MIMO antenna with stub
without stub (simulated) (fabricated)
Layout

−10 dB 129 & 215 128 & 205 120 & 200
impedance
bandwidth
(MHz)
Resonant 5.191 & 5.794 5.188 & 5.807 5.18 & 5.81
frequency
(GHz)
S11 (dB) −25.421 & −33.950 −48.352 & −46.879 −34 & −30
S21 (dB) −27.52 & −29.48 −31.041 & −36.086 −22.38 & −25
ECC 0.6 0.4 –
Gain (dB) 8.44 & 8.66 8.23 & 8.10 –

References

1. Oestgates C, Clerckx B (2007) MIMO wireless communications: from real world propagation
to space-time code design. Academic Press Ltd
2. Balanis CA, Ioannides PI (2007) Introduction to smart antennas. Morgan & Claypool Publishers
3. Blanch S, Romeu J, Corbella I (2003) Exact representation of antenna system diversity
performance from input parameter description. Electron Lett 39(9):705–707
4. Manteghi M, Rahmat-Samii Y (2005) Multiport characteristics of a wide-band cavity
backed annular patch antenna for Multipolarization operations. IEEE Trans Antennas Propag
53(1):466–474
5. Chae SH, Oh S-K, Park S-O (2007) Analysis of mutual coupling, correlations and TARC in
WiBro MIMO array antenna. IEEE Antennas Wirel Propag Lett 6:122–125
6. Karaboikis M, Soras C, Tsachtsiris G, Makios V (2004) Compact dual-printed inverted-F
antenna diversity systems for portable wireless devices. IEEE Antennas Wirel Propag Lett
3(1):9–14
7. Mavridis GA, Sahalos JN, Chryssomallis MT (2006) Spatial diversity two-branch antenna for
wireless devices. Electron Lett 42(5):266–268
8. Shin YS, Park SO (2007) Spatial diversity antenna for WLAN application. Microw Opt Technol
Lett 49(6):1290–1294
9. Wu TY, Fang ST, Wong KL (2002) Printed diversity monopole antenna for WLAN operation.
Electron Lett 38(25):1625–1626
10. Chi G, Binhong L, Qi D (2005) Dual-band printed diversity antenna for 2.4/5.2-GHz WLAN
application. Microw Opt Technol Lett 45(6):561–563
11. Wang X, Du Z, Gong K (2008) A compact wideband planar diversity antenna covering UMTS
and 2.4 GHz WLAN bands. IEEE Antennas Wirel Propag Lett 7:588–591
Design of MIMO Triangular Microstrip Patch Antenna … 379

12. Ding Y, Du Z, Gong K Feng Z (2007) A novel dual-band printed diversity antenna for mobile
terminals. IEEE Trans Antennas Propag 55(7):2088–2096
13. Arulaalan M, Nithyanandan L (2017) Development of triangular shaped dual band 802.11a
WLAN application. Asian J Inf Technol 1(16):212–217
A Hollow Core Bragg Fiber
with Multilayered Random Defect
for Refractive Index Sensing

K. Ben Franklin, R. Kumar and C. Nayak

Abstract In this paper, we present the theoretical analysis of Bragg waveguide con-
taining a random multilayer defect. Transmittance spectrum of the proposed mul-
tilayered cylindrical waveguide is obtained by employing transfer-matrix method.
Presence of random defect structure shows multiple defect peaks in the photonic
band gap of a defect-free Bragg structure. The probability of occurrence of defect
peaks around 660–690 nm is found to be very good, and these peaks may be chosen
as a sensing element. Result shows that the transmittance of the defect peak has a
good agreement with the change in the refractive index of the core.

Keywords Bragg fiber · Photonic band gap · Gaussian distribution · Defect peak ·
Transfer-matrix method

1 Introduction

Currently, there is a great interest in the fabrication of Bragg fiber waveguides due to
its application in high-power laser systems, sensing, telecommunication, medicine,
surgery, etc. The idea of Bragg fiber was first demonstrated by Yeh and Yariv in 1976
[1]. Thereafter, feasibility study of cylindrical Bragg fibers was theoretically as well
as experimentally presented for different optoelectronics applications. Such studies
show that, the hollow core Bragg fiber waveguide structure can be a suitable candi-
date for chemical gas sensors [2], strain sensors [3], biosensors [4, 5], narrowband
transmission filter [6], optical de-multiplexer [7], etc.
Bragg fiber is a multilayered waveguide which consists of a low-index core and
a periodic arrangement of alternating high and low refractive index [8]. The electro-
magnetic wave (EM wave) propagation in the core of a Bragg fiber is done through
Bragg reflection produced by the alternating arrangement of high and low refractive
index, whereas the conventional fiber uses total internal reflection for propagation

K. B. Franklin · R. Kumar · C. Nayak (B)

Department of Electronics and Communication Engineering, SRMIST, Kattankulathur, Chennai
603203, India
e-mail: 83chittaranjan@[Link]

© Springer Nature Singapore Pte Ltd. 2020 381

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
382 K. B. Franklin et al.

through the core. Optical properties of the Bragg fiber are characterized by several
parameters such as core size, refractive indices, and thickness of the cladding layers,
and hence offer a wide choice to tailor their propagation characteristics. Recently, a
design of Bragg fiber was reported, in which a liquid core Bragg waveguide showed a
photonic band gap in the transmittance spectrum, due to its alternate cladding layers
[9]. This photonic band gap is nothing but a band of wavelengths that are not allowed
to propagate through the waveguide. This band gap can be used as a sensing element
because the position of band gap and transmitted intensity depends on the core refrac-
tive index. Further studies were done by introducing a defect layer in the alternate
cladding arrangement, and a defect mode (narrow transmission band) appeared in
the band gap region [10]. Analysis shows that the transmittance of this defect mode
is more sensitive to change in core refractive index than the transmittance of band
gap. Usually, defects (randomness) in photonic structure are regarded as undesir-
able features that spoil optical quality and performances. However, they can also be
viewed as an enriching factor since, when controlled, they can be used to build up
good waveguides. In disordered structures, the wavelength region corresponding to
photonic band gap is filled with localized states (such as Anderson localization) [11],
and for such states, the magnitude of the electric field is very high. Such localized
states act as a Fabry–Perot resonator and lead to increasing transmission through the
structure.
In this work, we propose a hollow core photonic structure, where the cladding is
made by embedding multilayered random defect in the alternating cladding region.
Multilayer defects (randomness) are introduced in the cladding region by making
use of the Gaussian distribution function. The transmittance spectrum is obtained by
employing the transfer-matrix method (TMM) which was developed by Kaliteevski
et al. [12].

2 Theoretical Model

As can be seen in Fig. 1, the proposed Bragg waveguide consists of a hollow core
and the core is assumed to be filled with a liquid under observation having refractive
index, nC . The cladding region of the waveguide is mainly divided into three parts,
i.e., a random cylindrical structure sandwiched between two hollow cylindrical Bragg
dielectric structures. The first cylindrical Bragg dielectric structure, around the core
region, has N layers of unit cell. The unit cell consists of two different materials
A, higher refractive index layer (nH ) with thickness (dH ) and B, lower refractive
index layer (nL ) with thickness (dL ). Whereas, the second cylindrical Bragg dielectric
structure is similar to that of the first one, but it has M layers of unit cell. Layer width of
the sandwiched random cylindrical structure is assumed to be Gaussian distributed
with O number of layers of alternating high refractive index layer (ndH ) and low
refractive index layer (ndL ).
 A Hollow Core Bragg Fiber with Multilayered Random …

Fig. 1 Cross-sectional schematic representation of the photonic band gap waveguide having a random multilayer defect
383
384 K. B. Franklin et al.

The performance analysis of such structure can be done by employing many

numerical techniques. But, one of the most popular techniques is the transfer-matrix
method (TMM) [13].
Using the Abeles theory, the transfer matrix for the first high refractive index layer
of the first cylindrical Bragg dielectric structure with initial position y0 (inner radius)
to other point y1 (outer radius) is given by
   
V (y1 ) V (y0 )
=A (1)
U (y1 ) U (y0 )

where A is the transfer matrix of the dielectric cylinder and is given by

 
a11 a12
A= (2)
a21 a22

The individual matrix elements of the same are given by

π  
a11 = ky0 Ým (ky0 )Jm (ky1 ) − J́m (ky0 )Ym (ky1 ) , (3a)
2
π  
a21 = j ky0 P Ým (ky0 )Jm (ky1 ) − J́m (ky0 )Ým (ky1 ) , (3b)
2
π  
a22 = ky0 Jm (ky0 )Ým (ky1 ) − Ym (ky0 )J́m (ky1 ) , (3c)
2
π ky0  
a12 = −j Jm (ky0 )Ym (ky1 ) − Ym (ky0 )Jm (ky1 ) . (3d)
2 P

where Jm is a Bessel function, Ym is a Neumann function, k = ωc n is the wave

vector in a medium, c is the speed of light in free space, P = με is the characteristic

impedance, and n is the refractive index of that medium.
The final transfer matrix of the proposed Bragg structure is given by
    
V yf   V (y0 )
= ABN (D1 . . . . . . .DO ) ABM (4)
U yf U (y0 )

B, the second element of the unit cell, represents the transfer matrix of the lower
refractive index material. Dxx represents the transfer matrix of Gaussian distributed
random layers, where xx = 1, 2, . . . . . . .., O − 1, O.
    
V yf V (y0 )
= (F) (5)
U yf U (y0 )

The coefficient of reflection for the proposed waveguide can be calculated with
the help of the final transfer matrix, F.
A Hollow Core Bragg Fiber with Multilayered Random … 385
   
(2) (2) (2)
F́21 − JP0 Cm0 F́11 + JPf Cmf F́22 − JP0 Cm0 F́12
rd =     (6)
(1) (2) (2)
−F́21 + JP0 Cm0 F́11 + JPf Cmf −F́22 + JP0 Cm0 F́12

F́11 ,F́12 ,F́21 ,F́22 are the matrix elements of the inverse final transfer matrix F́ and


1,2
(1,2) Hm (k1 y1 )
Cmx = (1,2)
, x = 0, f (7)
Hm (k1 y1 )

where Hm(1) and Hm(2) are the Hankel function of the first and second kind. Using
Eq. (6), the reflectance is calculated as R = |rd |2 , and the value of transmittance,
T = (1 − R).

3 Results and Discussions

Let us now present the transmittance spectra for the proposed waveguide in 600–
725 nm region. We shall focus on the case of 0th azimuthal mode. Figure 2 shows
the calculated transmittance spectra for ten possible 32-layer Gaussian distributed
random annular defects (i.e, O = 32), whose thickness values are calculated for
mean = 250 nm and sigma = 60 nm and are listed in Table 1. Here, we have chosen,
N = M = 32, nH = 1.8, nL = 1.5, dH = 100 nm, dL = 100 nm, ndH = 3.2, and
ndL = 1.5.

Fig. 2 Transmittance spectrum for 10 different permutations with random defects

Table 1 Thickness of random multilayer (32) defect for 10 different permutations
386

S. No. Defect layers thickness (nm)

1 2 3 4 5 6 7 8 9 10 11
1 223 269 208 217 254 293 175 321 214 275 291
2 327 240 267 217 261 268 250 263 298 193 258
3 218 177 324 225 254 272 254 213 231 245 309
4 351 137 361 266 300 166 220 236 271 166 273
5 277 172 239 225 269 270 270 231 220 220 292
6 282 234 338 325 258 235 307 192 283 216 229
7 185 204 733 336 269 204 332 147 243 235 269
8 245 134 223 142 300 196 256 217 268 213 279
9 185 201 733 336 269 204 332 147 243 235 269
10 232 217 231 184 220 239 252 246 286 256 358
S. No. Defect layers thickness (nm)
12 13 14 15 16 17 18 19 20 21 22
1 302 293 250 243 225 260 264 235 255 356 249
2 170 251 327 287 249 236 271 292 250 314 217
3 215 281 262 240 207 238 246 191 257 217 246
4 189 253 282 266 304 300 217 262 202 183 296
5 157 239 263 217 273 246 203 258 236 295 214
6 216 278 211 196 277 228 268 232 263 121 273
7 268 198 248 240 287 315 316 198 254 177 183
8 294 352 238 121 199 331 185 307 257 336 132
(continued)
K. B. Franklin et al.
Table 1 (continued)
S. No. Defect layers thickness (nm)
12 13 14 15 16 17 18 19 20 21 22
9 268 198 248 240 287 315 316 198 254 177 183
10 268 358 206 281 234 286 285 118 170 163 274
S. No. Defect layers thickness (nm)
23 24 25 26 27 28 29 30 31 32
1 278 231 228 236 272 217 175 252 232 261
2 170 220 242 235 216 303 185 230 211 231
3 263 213 238 264 176 291 346 288 237 267
4 250 247 295 279 267 312 296 261 215 217
5 253 157 230 222 204 282 213 277 298 242
6 342 301 295 238 259 262 254 208 232 241
7 249 341 203 272 236 317 184 251 283 316
8 238 177 424 299 332 186 221 233 315 233
A Hollow Core Bragg Fiber with Multilayered Random …

9 249 341 203 272 236 317 184 251 283 316
10 338 230 298 282 186 273 204 340 248 348
387
388 K. B. Franklin et al.

(a)

(b) (c)

Fig. 3 Transmittance spectra as function of nC a between 600−725 nm b around 675.8 nm c relation

between nC and transmittance at 675.8 nm

One can see from Fig. 2, that a large number of defect peaks, or so-called local-
ization of light, appear within the photonic band gap of a defect-free structure. The
localization of light is obvious and it is caused due to the presence of random layers.
Moreover, the probability of occurrence and transmittance of defect peak around the
region 660−690 nm is found to be very high. Therefore, these defect peaks can be
employed as a sensing element.
To find the effect of change in core refractive index, we examine the transmittance
spectra of the first defective structure in Table 1 and is depicted in Fig. 3. It is
interesting to see from Fig. 3a that, for different core refractive index, a narrow
defect peak having different transmittance appear around 675.85 [Link] defect peak
intensity increases with increase in the core refractive index, shown in Fig. 3b. The
change in core refractive index and the transmittance of defect peak have a linear
relation, Fig. 3c. Due to good agreement between the transmittance and the core
refractive index, the proposed waveguide can be a good candidature for sensing
application.
A Hollow Core Bragg Fiber with Multilayered Random … 389

4 Conclusion

Looking for alternatives to set up a fiber sensing device, we have investigated in this
work the transmission characteristic for a hollow core Bragg fiber with multilayered
defects. The obtained result is very interesting; therefore, this proposal may be a
good material for sensing industry.

References

1. Yeh P, Yariv A (1976) Bragg reflection waveguides. Opt Commun 19:427–430

2. Shi L, Zhang W, Jin J, Huang YD, Peng JD (2010) Hollow core Bragg fiber and its application
in trace gas sensing. In: Communications and photonics conference and exhibition (ACP),
IEEE Xplore, Asia, pp 477–478
3. Hang Q, Brastaviceanu T, Bergeron F, Olesik J, Pavlov I, Ishigure T, Skorobogatiy M (2013)
Photonic bandgap Bragg fiber sensors for bending/displacement detection. Appl Opt 52:6344–
6349
4. Rowland KJ, Afshar VS, Stolyarov A, Fink Y, Monro TM (2011) Bragg waveguides with
low-index liquid cores. Opt Express 20:48–62
5. Hang Q, Skorobogatiy M (2012) Resonant bio-and chemical sensors using low refractive-
index-contrast liquid-core Bragg fibers. Sens Actuators B 161:261–268
6. Chen MS, Wu CJ, Yang TJ (2012) Narrowband reflection and transmission filter in an annular
defective photonic crystal containing an ultra-thin metallic film. Opt Commun 285:3143–3149
7. Sharma G, Kumar S, Prasad S, Singh V (2015) Theoretical modelling of one dimensional
photonic crystal based optical DE multiplexer. J Mod Opt 995–999
8. Prajapati Y, Saini JP, Chauhan DS, Singh V (2014) Effect of plasma on modal dispersion
characteristic of elliptical Bragg waveguide. Opt-Electron Rev 22(1):13–20
9. Schmidt H, Hawkins A (2008) Optofluidic waveguides: II. Fabrication and structures.
Microfluid Nanofluid 4:17–32
10. Chourasia RK, Prasad S, Singh V (2017) Opto-Electron Rev 25:215–221
11. Anderson PW (1958) Absence of diffusion in certain random lattices. Phys Rev 109:1492
12. Kaliteevski MA, Abram RA, Nikolaev VV, Sokolovski GS (1999) Bragg reflectors for
cylindrical waves. J Mod Opt 46(5):875–890
13. Born M, Wolf E (1999) Principles of optics. Cambridge, London
Generation of Multiple Key Based
on Monitoring the User Behavior

S. Palaniappan, Steward Kirubakaran, V. Parthipan and S. Rinesh

Abstract Cloud computing is the recent technology used to share and store the com-
puter resources rather than having resources in local server to maintain the applica-
tion. Though cloud is used for a large amount of storage, there is no security in cloud.
In general, all the groups have data owners and data members each should have user
name, key, and group key. If a user shifts from one group to another, they can easily
access the information from another group. It leads to a security problem. In order to
increase security and confidentiality, author generates the new group key via Email
using Diffie-Hellman algorithm. In case of a new user is added or an existing user
leaves themselves from the group. The data members have to get permission from
the data owners in case of any data updation. If the user misbehaves, i.e., (DDOS)
attack, data owner or cloud terminates the user from the group. The updated key is
sent to the users through Email. This mechanism significantly improves security in
cloud computing.

Keywords Cloud computing · Group key · DDOS attack

1 Introduction

Cloud computing has a large amount of storage such as (Drop Box, Google Drive)
[1–3] and sharing resources. But there is no security in the cloud is enforced. The
main goal of the author is to improve security in cloud. It is all about generating

S. Palaniappan (B) · S. Kirubakaran · V. Parthipan · S. Rinesh

Department of Computer Science and Engineering, KCG College of Technology, Karapakkam,
Chennai, India
e-mail: [Link]@[Link]
S. Kirubakaran
e-mail: stewartkirubakaran@[Link]
V. Parthipan
e-mail: parthipansp@[Link]
S. Rinesh
e-mail: [Link]@[Link]
© Springer Nature Singapore Pte Ltd. 2020 391
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
392 S. Palaniappan et al.

dynamic and flexible group keys based on user behavior. In general, each group
has data owners and data members. For instance, an institute might have several
departments. Each department has a separate group key and users have their user
key for login purpose. Each data members have to enter user name, key, and group
key for login. Due to modification from different users, different blocks are signed
by different users. For security reasons, when a user leaves the group or misbehaves,
this user must be revoked from the group [4]. This will provide confidentiality and
reliability. As a result, this revoked user should no longer be able to access and
modify shared data, and the signature generated by this revoked user is no longer
valid to the group [5]. In this case, new group key is sent to the corresponding data
members via Email. This can be done once the data members shift from the group or
a new data member is added to the group obviously this will improve security of the
cloud. On each updation, data members need to get approved from the data owners.
In case of any mischievous activities (DDOS) attack, the data owners or cloud itself
terminates that particular data members from that group and the updated key is sent
to the corresponding group through Email.

2 Methodology

In this article, the author categories the improvement of security into three ways.
(1) Automatic Key Generation—Data members have their user name, key, group
key for login purpose. If a new member is added to the group or an existing
member leaves from the group, the updated group key is sent automatically
through Email. This will improve the privacy of the data.
For Example: Consider a department, and it consists of various members.
If a member of that department shifts from one group to another, the group key
will be automatically changed and sent to the department through Email for
confidentiality of data.
(2) Mischievous Activity—In general, each group has data owners and data mem-
bers. Data owners have authority to update the data to the cloud server for
accessing data. Data owners appoint the data member for data utility and data
updation. In the case of any mischievous activity, a data owner has their privilege
to terminate the data members from that group and generates the new group key.
Obviously, this will improve the reliability and confidentiality.
For Example: If a data member had a privilege only to access the data, in case
if a user tries to attempt for data updation, it will be considered as mischievous
activity, then data owners had an authority to terminate that particular data
member from that group and group key is updated automatically.
(3) DDOS Attack—If a data members attempt to upload the same file for N number
of times, then data owner or cloud itself terminates the data members and updated
key is sent to that corresponding group through Email.
Generation of Multiple Key Based on Monitoring the User Behavior 393

Fig. 1 Dynamic key generation architecture

For Example: If a data members attempt to login to for more than particular
number of times, the data owners automatically terminate the data members
from the group and group key is updated.
These three ways contribute security and privacy to the cloud server. And also
increase the efficiency and integrity of data. User termination will improve the high
efficiency because the terminated user is no longer to the group; hence, accessing
data from the group is impossible. Termination can be done by data owners of the
group. Once the user is terminated, resigning is impossible (Fig. 1).
However, the above architecture is mainly focused on dynamic and flexible group
key generation. Security is the big challenge in cloud computing server other than
networking and communication. This article updates the group key in case of any
activities in the group.

3 Implementation

A user interface is implemented using Java technology or C#. Database is created

using My SQL. Here, drop box is used for cloud for sharing resource. Diffie-Hellman
algorithm is used for generating a new key in case of any updation takes place among
the group for security purpose. This will make the group secured and efficient.
394 S. Palaniappan et al.

4 Diffie-Hellman Algorithm for Dynamic Key Generation

The Diffie-Hellman key agreement protocol (1976) was the first practical method
for establishing a shared secret over an unsecured communication channel [6]. The
main goal of this algorithm is to generate the new group key in case of any changes in
the group. It shares the secret key among the groups. Every piece of information that
shares among the groups is noticed by the data owners of the group. This algorithm
is used to share the random secret key in case of any changes in the group. Diffie-
Hellman key exchange shares the key randomly for security purpose among the
groups. Although RSA algorithm is the first algorithm used to share the new key.
Secret key will be generated if an existing user shifts or a new user is added to
the group and also if a user misbehaves, the group key will be updated using the
above mechanism. By this way, data in the group will be secured. Diffie-Hellman
key exchange algorithm solves the following problem. For instance, john and mark
are belong to same group; they can share and access the data of same group. If a
new user wants to access, the information from john and mark is highly impossible,
because Diffie-Hellman generates the group key. If a user wants to access, the data
group key is mandatory. Even if john and mark went under any mischievous activity,
the group key will be updated by using above algorithm and shared in cloud server
which will share the current group key for the corresponding data members. Even
when john and mark share the group key, they may use a password-authenticate
key agreement (pk) form of Diffie-Hellman to prevent man in the middle attack [6]
(Fig. 2).

5 Secure and Efficient User Termination

In this article, we argued about the user termination; this will make data more secure
and efficient. The cloud can resign blocks that were previously signed by the revoked
user with a resigning key, while an existing user does not have to download those
blocks, recompute signatures on those blocks, and upload new signature to the cloud
[4]. User termination is secured because an existing user can access the data stored
in the group. Once the user is revoked from the group, he/she is no longer in the user
list. The resigning performed by the cloud improves the efficiency of user revocation
and saves communication and computation resources for existing users [7].

6 Module Description

To perform effective resource allocation and to achieve security in cloud, five different
modules are used. These modules are discussed below.
Generation of Multiple Key Based on Monitoring the User Behavior 395

Fig. 2 Data flow diagram

Network Construction—In this module, dynamic network is created. Nodes are

interconnected with particular group, used to share the information among them.
The network should be controlled and handled for successful transmission of data.
Network is constructed using interconnection of various nodes.
Server—All the group information and details are maintained by the server. It will
distribute the data to the client of a group. In case of any data issue, server is respon-
sible. It will instruct the data owners to change the group key in case of any changes
in the group and the updated key is sent to the data members through Email.
User Status—All the user status is to be maintained here. User can shift from one
group to another, and also they can participate in more than one group. The infor-
mation can be shared depending upon user status. For authentication purpose, all the
information about the user is to be maintained here.
396 S. Palaniappan et al.

Group Key Generation—In this module, group key as well as individual key is
created and then shared among the group via Email. Group key is updated in case of
any changes in the group. This can be done whenever any changes take place in the
group.
Data Access—The user can able to access the information of any user of same group
that can be done using group key and user key. Accessing the information of a user
of a different group is impossible. Information of a user of different groups can be
accessed only with the help of that particular group key.

7 Advantages

This paper has been framed to improve security of the cloud server. The paper has
the following advantage.
• It improves high security
• It provides integrity and confidentiality
• Dynamic and flexible key generation.

8 Conclusion

In this article, we have discussed and implemented a new mechanism for improving
security and efficiency of data. It is all about maintaining secured data among the
groups. Each group has separate group key for login purpose. If a user in the group
leaves or shifts to another group, the updation of group key will take place. Since
the user from another group can not access the data. If a new user is added to the
group, the key will be updated and sent to the corresponding data members of that
group. Group key will be changed in case of any mischievous activity in the group
and that particular user will be terminated from that group. Generation of new group
key takes place in case of any (DDOS) attack in the group. These three ways of group
key generation help us to increase the security. This will improve the confidentiality
and efficiency of data. And also this will increase security in the cloud server. In
further, any other research will produce more security in the cloud.

References

1. Drop Box (2007) On the impact of virtualization on drop box like cloud file stor-
age/synchronization services
2. Mozy (2007) A study on cloud backup technology and its development
3. Bitcasa (2011) Proof of ownership in duplicated cloud storage with mobile device efficiency
Generation of Multiple Key Based on Monitoring the User Behavior 397

4. Wang B, Li B, Li H (2013) Panda: public auditing for shared data with efficient user revocation
in the cloud. IEEE Trans Serv Comput 8(1):92–106
5. Wang B, Li B, Li, H (2012) Knox: privacy-preserving auditing for shared data with large
groups in the cloud. In: Proceedings of 10th international conference on applied cryptography
and network security (ACNS’12), pp 507–525
6. [Link] key exchange Password-
authenticated_key_agreement
7. [Link]
8. Memopal (2007) Memopal
9. Gantz JF, et al. (2006) The expanding digital universe: a forecast worldwide information growth
through 2010
10. Bowers AJ, Oprea A (2009) Proofs of retrievability: theory and implementation. In: Proceedings
of the first ACM cloud computing security workshop (CCSW 2009), USA
11. Wang Q, Wang C, Li J, Ren K, Lou W (2009) Proofs of retrievability via hardness amplification.
Saint-Malo, France, pp 355–370
12. Saikumar K, Rais MdZ, Palaniappan S (2015) Smart id card with NFC tags using elliptical
curve digital certificate authentication. Int J Appl Eng Res 10(4):3077–3079
13. Yuan J, Yu S, Proofs of retrievability with public verifiability and constant communication cost
in cloud. In: Proceedings of international workshop on security in cloud computing, Hangzhou,
China
14. Shi E, Stefanov E, Papamanthou C (2013) Practical dynamic proofs of retrievability. In:
Proceedings of ACM CCS, Berlin, Germany
15. eXo Cloud IDE (2002) [Link]
16. Albert MJ, Yovan FA, Chintalapudi M, Balaji R, Vigneshwaran R (2015) Fault-open minded
resource allocation rescheduling algorithm and expense minimization for cloud systems. Glob
J Pure Appl Math 11(6):4111–4117
17. Erway C, Kupcu A, Papamanthou C, Dynamic provable data possession. In: Proceedings of
ACM CCS, Illinois, pp 213–222
18. Kumar B, Palaniappan, Jashwanthreddy, An hybrid of RSA token and iterated hash algorithm
for secured data transfer. Int J Pharm Technol 19417–19423. ISSN: 0975-766X
Towards Convolution Neural Networks
(CNNs): A Brief Overview of AI
and Deep Learning

Preetjot Kaur and Roopali Garg

Abstract Today’s era is of cutting edge of innovations as well as technologies. One

of the major problems, researchers often face is an issue looking for an appropriate
research area. For instance, there are numerous fields these days on which research
is being carried out and to pick one out of those topics is itself a challenging task.
The major objective of this review paper is to embark upon Artificial Intelligence
(AI) that prompted the emergence of deep learning (DL) and further to convolution
neural networks (CNNs). Limitations of CNNs that led to the development of Capsule
Neural Networks (CapsNets) have been included. The significant goal of this review
paper is to discuss the latest trends in which research is on-going and is still in
progress. Also, the key challenges faced by past researchers are highlighted.

Keywords Artificial intelligence (AI) · Current research trends · Convolution

neural networks (CNNs) · Capsule Neural Networks (CapsNet) · Deep learning
techniques

1 Introduction

There are many trends in the field of computer science engineering. However, out of
these, only a few patterns are ruling this century and stand out as truly newsworthy.
Since the acquaintance of Artificial Intelligence (AI) with the market, it has become
a reason for speedy changes in the technology and business worlds. AI is the ability
of machines to do tasks in such a way as humans do in any situation. The growing
trends of computer science community are of special importance as they provide
future researchers with all the tools and equipments that they need to make progress
into this field [1]. It is very necessary to build connection with new technologies, as it

P. Kaur (B)
PhD Research Scholar, UIET, Panjab University, Chandigarh 160014, India
e-mail: preetjotkaur20@[Link]
R. Garg
Associate Professor, UIET, Panjab University, Chandigarh 160014, India
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 399

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
400 P. Kaur and R. Garg

leads to the discovery of an abundance of future developments. Today’s era is of the

cutting edge of issues in innovations and technology and wide range of opportunities
are available [2] as shown in Fig. 1.
Multidisciplinary and interdisciplinary research is trending these days. For
instance, detecting facial expressions is a major concern among researchers in AI

Fig. 1 Latest research trends in computer science (CS)

Towards Convolution Neural Networks (CNNs): A Brief Overview … 401

and the combination of LBP with CNN yields desirable results [3]. Computer vision
has been in trend since the 90s and is a vast grown field now. Despite having much
work done in this field, there is a need of robust device, which could help vehicles
in becoming autonomous and help blind persons in independent outdoor navigation.
Detecting interesting video game clips is an important task in computer vision. Many
algorithms, for example, feature extraction along with deep convolution networks
provide interesting results [4].
In healthcare industry, the computerised tests such as electrocardiogram (ECG),
electroencephalogram (EEG), EKG, etc. have proved to be very beneficial for the
patients and are tremendously growing research fields. These are important to anal-
yse the emotional behaviour of the patients so as to prevent them from depression [5].
The technology has risen to levels that it can now defeat human experts in games and
many other such fields. In almost all the games, AI machines have won, and recently,
AlphaStar, the DeepMind’s project, has successfully defeated top professional Star-
Craft’s player and set a benchmark record [6]. Most of the developed algorithms are
actually inspired from natural phenomenon of biological things, for example, ant
colony optimisation algorithm and techniques such as sailfish optimiser [7].
This review paper intends to help researchers who are in the beginning phase of
framing research problem. It is organised as Sect. 2 introduces Artificial Intelligence
and its history. Section 3 mentions the timeline of deep learning and CNNs. Section 4
discusses the challenges faced by early researchers. Section 5 presents the conclusion
and is appended with references.

2 Artificial Intelligence

AI is an encompassing umbrella that intends to make such computer programs that

are able to think, behave and take decisions just like humans. Since the acquaintance
of AI with market, it has become a reason for speedy changes in the technology
and business worlds. Ever since the development of powerful digital computers, the
capability of machines to perform various tasks have grown tremendously [8]. A
part of computer science named AI seeks in making machines that are as astute as
people. The two ultimate goals of AI are creating self-learning expert systems and
embedding human intelligence into machines. Today, AI has reached the unexpected
levels of development and almost all the major developments are being made under
it.
Neural networks (NN) are the set of algorithms that copy the way by which
human brain operates. The first model of NN was developed by McCulloch and Pitts
in 1943 [9]. By the prediction of machine learning (in 1947) and development of
Allan’s Turing test (in 1950), first, machine learning (ML) program was developed.
At the same time, several self-learning programs to play the games such as checkers
were built [10]. ML teaches machines on how to learn, by providing them with data
and without programming of any type.
402 P. Kaur and R. Garg

2.1 Story of Birth of Artificial Intelligence

with Chronological Order

Getting into the roots of any field is important, before beginning research in it. Table 1
shows the inception of AI and how has it grown over the years.

3 Deep Learning

Deep learning (DL) is a part of machine learning, which is further a subset of AI.
It is based on learning data representation method, i.e. an algorithm learns by itself
from large amount of data. The objective of DL is to enhance technologies such as
driverless vehicles, Siri in Apple’s iPhone, Amazon’s Alexa, etc.

3.1 Deep Learning History

Frank Rosenblatt set up the foundation of deep neural networks in the theory devel-
oped for perceptron in 1957 [11]. Around the same time, two types of cells—simple
and complex—were discovered in the primary virtual cortex. Most of the artificial
neural networks (ANNs) are inspired from them. The ideas of control theory, as pub-
lished by Henry J. Kelly in 1960 [12], were utilised in making ANNs. Ivakhneko,
known as father of modern deep learning, developed a famous method of inductive
statistical learning—group method of data handling (GMDH) [13]. Using GMDH,
he created an eight-layer deep network known as ALPHA in 1971. In the 1980s, K.
Fukushima proposed Necognitron that led to the development of first convolution
neural networks (CNNs) [14].
Hopfield networks are recurrent neural networks. They have become famous
tool for DL these days. An AI-based program called NETtalk was written by Ter-
rence Sejnowski and Charles Rosenberg around 1985 [15]. It was used to pronounce
English words just like a child and learns by its own. Geoffrey Hinton, the Godfa-
ther of DL, suggested many improvements in NN for word prediction, object shape
recognition, etc. By 1989, deep learning’s CNN was combined with backpropaga-
tion algorithm to make the machines read handwritten notes [16]. Supervised deep
learning algorithms started to develop by the early 1990s.
In order to retain information for longer period of time, long short-term memory
was proposed in 1997. The integration of backpropagation algorithm with gradient-
based learning became popular technique for deep learning [17]. The ImageNet
database was launched in 2009 [18]. It was used to build high level features by using
large-scale unsupervised learning [19]. Today deep learning has reached exceptional
levels [20].
Table 1 History of artificial intelligence
Timeline of history of AI
1940 1943 1950 1951 1955 1956 1958
Possibility of Artificial neurons Alan Turing test • SNARC introduced Logic theorist • Dartmouth Perceptron and
creating artificial proposed introduced • First, chess-playing conference logic introduced
brain discussed program • Birth of AI
1963 1965 1970 1972 1974 1975 1980
Grant of 2.2 Dendral expert Prologue created • WABOT-1, First AI Winter Frames created XCON (an expert
million dollars system produced intelligent robot and scripts made system) formed
given to AI by • MYCIN (expert
DARPA system)
1982 1985 1987 1988 1997 2003 2005
Hopfield nets and Funding began Second AI Winter • HiTech won a Deep blue defeated AI became rigorous Stanford robot won
backpropagation till 1993 match world champion scientific field DARPA grand
introduced • Deep thought won challenge
Towards Convolution Neural Networks (CNNs): A Brief Overview …

2007 2011 2016 2018 2019

Self-driving SUV-boss won Watson—an IBM’s question AI products marketed AI machine painted canvas (1) chatbots selling insurance
the DARPA grand challenge answering system defeated sold in October 2018 for (2) drawing robot-Ai-Da
$432,500
403
404 P. Kaur and R. Garg

3.2 Convolution Neural Networks (CNN)

CNNs—the algorithms of deep artificial neural networks (ANNs)—are used to cate-

gorise images based on the low-level features such as edges as shown in Fig. 2. CNNs
include large number of neurons in their layers, which help them in recognising new
images based on their learning from previous dataset.
Today, CNNs find numerous applications in almost all the vision imaging areas
such as human facial expression recognition [21], classifying social-network events
[22], brain tumour detection [23], age and gender recognition [24], community ques-
tion answering [25], etc. The architecture of CNN is originated from Neocognitron
[14], which is inspired from cat’s visual system [26]. The layers of CNN are discussed
below.
Convolution Layer: It is the core layer of the CNNs that do most of the computational
work. It takes an image matrix and a filter as input and performs the following
computation as shown in Eq. (1).

Activation map (Output) = (image matrix) · (Filter matrix) (1)

ReLU: The output of the above layer is passed through nonlinear activation function-
rectified linear unit (ReLU ), to make network as nonlinear. The obtained output is
shown in Eq. (2).

f (x) = max(0, x) (2)

Pooling Layer: It is also referred to as downsampling layer as it reduces the number of

excessive parameters and redundancy in an image. To reduce the amount of memory
consumed by the network, spatial pooling is applied. There are different ways of
doing so, such as max pooling, average pooling and sum pooling.
Fully Connected Layer: After many convolution and pooling layers, the output
is flattened into a vector and fed into a fully connected layer. This layer actually
categorise the image by matching the detected features with several classes of the
objects and the class that most correlate to those features are detected.

Fig. 2 Architecture of convolution neural networks (CNNs)

Towards Convolution Neural Networks (CNNs): A Brief Overview … 405

3.3 Capsule Neural Networks (CapsNet)

Due to some limitations observed while using CNNs, structures such as capsules are
added to them, called CapsNet [27]. These are listed below.
• CapsNets are invariant to viewpoint as they recognise objects even from different
angles and provide less error rate as compared to CNNs.
• CapsNet is composed of many capsules and each capsule represents a group of
neurons. Due to this, lesser parameters are required for connections between layers.
• CNN breaks on applying transformations such as rotation, scaling, etc. Also, same
viewpoint is shown even after different rotations, while CapsNet prevents this and
shows better viewpoints.
• CNN requires a large amount of data while training in contrast to CapsNet.
• CNNs suffer more loss by adversarial attacks such as fast gradient sign method
(FGSM) in contrast to CapsNet [28].

4 Problems Faced by Previous Researchers in Artificial

Intelligence

There were certain risks involved in the development of AI in the past years and
due to this, it took a long time to develop and grow. Availability of Limited Com-
puter Resources was one such reason. The simplest AI applications require at least
1000 MIPS to execute properly. But in 1976, even the Cray-1 (fastest supercomputer)
was able to achieve only 80–130 MIPS. Today, the fastest supercomputer-Sunway
Taihulight (China) delivers maximum sustained performance of 93.01 petaflops (a
quadrillion floating-point operations per second). Cost of the hardware was the sec-
ond issue that researchers faced. An average desktop computer had a cost of 3000$
approx. in the 1976s, making it a huge factor. Due to the unavailability of experimen-
tal labs, time was another crucial factor. According to Tim Cook, certain problems
demanded an exponential time to get solved [29]. ML and DL algorithms required
a gigantic amount of datasets, which were unavailable at that time. Accomplish-
ing complex tasks such as face recognition, etc. were very tough for very simple
AI systems developed in earlier times. Certain scientists (such as Dreyfus) criti-
cized that artificial thinking is based on logical deduction and humans rarely used
logics when they solve problems. Therefore, funding became major problem for AI
researchers [30, 31]. AI has become a highly emerging field now with very innovative
technologies.
406 P. Kaur and R. Garg

5 Conclusion

This paper gives an insight to the researchers, about the spectrum of the field of com-
puter science. There are several trends in which research is on-going these days and
choosing one, out of them is a challenging task. The characteristics of all areas have
been discussed with major stress on AI’s deep learning and CNNs. The limitations of
CNN and how CapsNet has improved those limitations have also been highlighted.
The key challenges faced by researchers working in AI have also been identified. This
review paper endeavours to guide the researchers to frame their research problem in
the chosen research area.

References

1. Catania BK (Dü.) (2019) Theory and practice of computer science. In: 45th international
conference on current trends in theory and practice of computer science. Springer International
Publishing, Nový Smokovec, Slovakia
2. IEEE Computer Society predicts the future of tech: top 10 technology trends for 2019, 18
Dec 2018. IEEE Computer Society. [Link]
technology-trends-2019
3. Kennedy Chengeta SV (2018) Facial expression recognition using local directional pattern
variants and deep learning, ACAI 2018. In: International conference on algorithms, computing
and artificial intelligence. ACM, China, NY, USA
4. Chao Huang LZ (2018) RGVCD: a new real-time game video clip detection system, ISB-
DAI’18. In: International symposium on big data and artificial intelligence. ACM, Hong Kong,
NY, USA, pp 172–177
5. Yu X, Qi W (2018) A user study of wearable EEG headset products for emotion analysis, ACAI
2018. In: International conference on algorithms, computing and artificial intelligence. ACM,
Sanya, China, NY, USA
6. Vinyals OA (2019) Mastering the real-time strategy game StarCraft II
7. Shadravan S, Naji HR (2019) The sailfish optimizer: a novel nature-inspired metaheuristic
algorithm for solving constrained engineering optimization problems. Eng Appl Artif Intell
80:20–34. Elsevier Ltd.
8. Birrer FAJ (1986) Artificial intelligence. In: Emerging technologies and military doctrine.
Palgrave Macmillan, UK, pp 44–52
9. McCulloch WS, Pitts W (1943) A logical calculus of the ideas immanent in nervous activity.
Bulletin Math Biophys 5(4):115–133
10. Samuel AL (1959) Some studies in machine learning using the game of checkers. IBM J Res
Dev 3:210–229
11. Rosenblatt F (1958) The perceptron: a probabilistic model for information storage and
organization in the brain. Psychol Rev 65:386–408
12. Kelley HJ (1960) Gradient theory of optimal flight paths. ARS J 30:947–954
13. Ivakhnenko AG, Lapa VG, Nikolic ZJ (1966) Cybernetic predicting devices. Purdue University
School of Electrical Engineering, Lafayette, Indiana
14. Fukushima K (1980) Neocognitron: a self-organizing neural network model for a mechanism
of pattern recognition unaffected by shift in position. Biol Cybern 36:193–202
15. Bien B (1988) The promise of neural networks. Am Sci 76:561–564
16. LeCun Y, Boser B (1989) Backpropagation applied to handwritten zip code recognition. Neural
Comput 1:541–551
Towards Convolution Neural Networks (CNNs): A Brief Overview … 407

17. LeCun Y, Boser B (1998) Gradient-based learning applied to document recognition. IEEE
86:2278–2324
18. Deng J, Dong W, Li LJ, Fei-Fei L (2009) ImageNet: a large-scale hierarchical image database.
In: IEEE conference on computer vision and pattern recognition. IEEE, FL, USA
19. Le QV (2013) Building high-level features using large scale unsupervised learning. In: IEEE
international conference on acoustics, speech and signal processing. IEEE, Vancouver, BC,
Canada
20. Dhabaleswar K, Panda DK, Awan A (2019) High performance distributed deep learning:
a beginner’s guide. PPoPP’19. In: 24th symposium on principles and practice of parallel
programming. ACM, Washington, District of Columbia, NY, USA, pp 452–454
21. Nikolaos Christou NK (2018) Human facial expression recognition with convolution neural
networks. In: Third international congress on information and communication technology.
Springer, Singapore, pp 539–545
22. Hussain A, Keshavamurthy BN, Wazarkar S (2019) An efficient approach for classifying
social network events using convolution neural networks. In: Advances in data and information
sciences, vol0 39. Springer, Singapore, pp 177–184
23. Havaei M, Davy A, Warde-Farley D, Biard A, Courville A, Bengio Y, Larochelle H (2017)
Brain tumor segmentation with deep neural networks. Med Image Anal 35:18–31
24. Kharchevnikova AS, Savchenko AV (2018) The video-based age and gender recognition
with convolution neural networks. International conference on network analysis, NET 2016:
computational aspects and applications in large-scale networks. Springer, Cham, pp 37–46
25. Wang J, Sun J, Lin H, Dong H, Zhang S (2016) Predicting best answerers for new questions:
an approach leveraging convolution neural networks in community question answering. In:
Chinese national conference on social media processing, SMP 2016, pp 29–41
26. Hubel DH, Wiesel TN (1962) Receptive fields, binocular interaction and functional architecture
in the cat’s visual cortex. J Physiol 160:106–154
27. Sabour S, Frosst N, Hinton GE (2017) Dynamic routing between capsules. In: 31st conference
on neural information processing systems (NIPS 2017). CA, USA
28. Yuan X, He P, Zhu Q, Li X (2017) Adversarial examples: attacks and defenses for deep learning
29. Cook S (1971) The complexity of theorem proving procedures. ACM Digital Library
30. Feng Z, Xiaofeng H, Lin W, Xiaoyuan H, Shengming G (2016) Inspiration for battlefield
situation cognition from AI military programs launched by DARPA of USA and development
of AI technology. In: Theory, methodology, tools and applications for modeling and simulation
of complex systems, pp 566–577
31. Mitchell JR (1987) Workshop on research directions and opportunities II: current funding
programs. In: Empirical foundations of information and software science IV. Springer, Boston,
MA, US, pp 507–517
Populating Indian GST Details into Java
Apache Derby Database Powered
by Glassfish Server

R. Sridevi and S. Srimathi

Abstract In simple words, data is the real specifics associated with any objective
for consideration. Data, in the perspective of databases, corresponds to all particular
items that are stored in a database, moreover separately or as a set. Data in a database
in general is preserved in database tables, which are ordained into columns with the
intention to enforce the data types retained therein. Database is a staged collection
of data and elaborates them as information; more exclusively, a database depicts
an electronic framework which eases data to be modestly accessed, manipulated
and updated. In other words, whenever an organisation having database it embod-
ies method of storing, managing and retrieving information and they are governed
by database management system (DBMS) to make it flexible. Generally, databases
encompass multiple tables with numerous different fields suitable to the data stored
in the table. Database management system (DBMS) let its users to retrieve infor-
mation in the database and operate data. In addition, it controls the access to the
database. This paper deals with Indian GST rates for various categorical products
into the Derby DB using NetBeans as IDE and Glassfish application server. This
system can be manipulated and enhanced with the simple SQL queries. This system
proves to be a smart way for creating a secure, standardised and transactional Derby
DB.

Keywords Indian GST system · Oracle populating · DBMS manipulation · Derby

DB · NetBeans

R. Sridevi (B) · S. Srimathi

Department of Computer Science and Engineering, K Ramakrishnan College of Engineering
Samayapuram, Trichy 621112, India
e-mail: sridevivelon@[Link]

© Springer Nature Singapore Pte Ltd. 2020 409

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
410 R. Sridevi and S. Srimathi

1 Categories of DBMS

There were four categories of database management systems available which are
listed below:
• Relational database management systems.
• Hierarchical database management systems.
• Network database management systems.
• Object-oriented database management systems.

1.1 Hierarchical DBMS

This category implies the hierarchical if the relations present in the database are with
the purpose of having one data that seems as the sub of another data, which indicates
the “parent–child” links with them. This category is appropriate for keeping data
items depicting attributes features and so on.

1.2 Network DBMS

A network information model could be an information model that enables multiple

records to be joined to a similar owner file. The multiple linkages make sure that this
permits the network information model to be terribly versatile. Moreover, the data
in network model has many-to-many relationship, so one owner file is often joined
to several member files and vice versa.

1.3 Object-Oriented DBMS

Object-oriented DBMS has significant advancement when compared to the remaining

DBMS. An Object-oriented DBMS acquire data from numerous dissimilar sources
like text, photographs and frame output in a multimedia.

1.4 Relational DBMS

In relational DBMS, the database relations are represented by table with the relations
that the data related with other in the same or with other tables which can be exactly
handled by means of merging one or more tables and processed with the help of
structured query language.
Populating Indian GST Details into Java Apache Derby Database … 411

2 Uses of Database

Uses of database systems include:

• Accumulate data for effectual and proficient management.
• Simple to know and user-friendly.
• Security and integrity of data.
• Better access to accurate data.
• Handle query processing and management in an excellent way.
• Enhanced decision-making done with the support of database.
• Data sharing and storage for additional handing made.
• Database guarantees error-free information.

3 Oracle Database

When the company develops and complaisance policy varies, it necessitates the
flexibility to manage data economically. Oracle affords a principally absolute, incor-
porated and protected database and data management solution for any exploitation.
The organisation necessitates data to stay on premises, managed in data centre or
deployed in the cloud; Oracle facilitates access to the same database technology. An
Oracle database gathers data ordered by type with associations preserved amongst
the different categories. An Oracle database is an aggregation of information per-
ceived as unit which is used for large business processing because of its flexibility
and realistic nature to handle data and software applications. Oracle has abundant
uses everywhere around the globe in various fields with options like quicker access
and speedy recovery, and it makes management and processing of knowledge in a
truthful way.
Oracle database reacts extraordinarily well for requesting situations, and the exe-
cution is performed in an extremely well mannered. It is a reliable database, and
with further highlights, those are tested through the ACID test, which is essential in
ensuring the propriety of information. This is necessary because information is the
centre of every structure in the association.
Oracle is the most reliable and highly used relational database found today. Its
common usage includes:
• Pre-defined data stored for future processing.
• Keeping the standard query language (SQL).
• Administration and handling the data.
• Storing needed information and massive data.
• Providing security options.
• Recovering database.
Oracle is utilised in most applications and one amongst is in banking. It offers pow-
erful combination of technology with wide range of integrating business applications,
412 R. Sridevi and S. Srimathi

together with practicality engineered completely in banks. Most of the databases sim-
ilar to Sybase and SQL include conveniences intended for using conditional loops,
arrays, therefore an exceeding program and in addition, services like cursors and
employee tables. However, all this can be readily used in a fancy approach that
is implausibly easy and resource-intense functions. Usually, new versions releases
serve better than earlier versions that exist in Oracle new version which also comprise
several features existing compared with prior versions, and it maintains improvement
and liberates novel products, and thus, the performance is improved to a great extent.
The advantages of newer version led to additional options and with advanced fea-
tures like this, one could run Java in the Oracle 8i itself. In its next version, 9i Oracle
overcame the drawbacks existing in 8i and added new features to assist the DBA in
handling changes. In the recent times, Oracle 10 versions have many features together
with recycle bin which enhance operations of users just like Windows recycle bin.
Information is the heart of each application or organisation that demands care-
ful maintenance. However, often, application disruption happens, and principally,
DBA maintains the explanations for this like hardware collapse, and the reasons
may include human-made errors like unintentional removal of valuable informa-
tion, removing the incorrect information or inappropriate usage of the table. By
using recent flash technology, it provides streamline management and administra-
tion processes in an efficient manner and provides many features used for providing
protection, security, maintenance, dependability and performance in a user-friendly
way.

4 Different Ways of Populating Oracle Database

SQL loader uses three different strategies to load information: conventional path,
direct path and external tables.

4.1 Conventional Path

The default loading method available in Oracle is the conventional path load. The
SQL INSERT statements were used to execute and populate tables in Oracle.
Populating Indian GST Details into Java Apache Derby Database … 413

4.2 Direct Path

In this method, the Oracle data blocks were written directly on the database files,
which are much faster than the previous method and eliminates a lot of the Oracle
database overhead by means of data formatting.

4.3 External Tables

In this, external table is created for data which is enclosed in a data file, and the
INSERT statements allow modification of the data and use SQL*loader to carry out
the tasks like loading data in network, handling multiple data files, multiple tables,
selectively load, manipulate and generating key values and loading data from disk
and tape.

5 Indian GST System

In earlier taxation, before GST, there have been several indirect taxes charged by each
state and central government. States naturally raise taxes as value-added (VAT). In
India, all state had a various set of policy. In the pre-GST system, each buyer as well as
the ultimate client paid tax on tax. The sale of products amongst interstate was taxed
by the Centre’s CST central state tax, was relevant for interstate sale of products, and
many other all possible taxes like duty taxes for travel, movie, tips and other local
taxes were the put by the respective state and the central government. There were
several indirect taxes put in the pre-GST like central excise duty, duties of excise,
additional duties of excise, additional duties of customs, special additional duty of
customs, cess, state VAT, central sales tax, purchase tax, luxury tax, entertainment
tax, entry tax, taxes on advertisements and taxes on lotteries, betting and gambling,
and all these extra taxes are currently included in CGST, SGST and IGST.
GST replaced many indirect taxes for the entire country. On 29 March 2017, the
Goods and Service Tax Act was passed in the Parliament and came into effect on 1
July 2017. The GST law came into effect on 1 July 2017 in both the Lok Sabha and
Rajya Sabha.
414 R. Sridevi and S. Srimathi

5.1 Elements of GST

In GST, there are three different elements of taxes enforced such as: CGST, SGST
and IGST.
• CGST: The tax which was charged by the central government for intra-state sale.
• SGST: The tax which was charged by the state government for intrastate sale.
• IGST: The tax which was charged by the central government on an interstate sale.

5.2 Benefits of GST

The introduction of GST mainly removes the barriers in the taxation process by elim-
inating the tax on tax, so the cost of product reduced. Usually, the GST is principally
a technological-based one, and in this, all the product-related deeds starting from
registration of product to the filing of return and application processing everything
is done only in online GST Portal.

5.3 Experimental Setup

The experiment of populating Indian GST rate for a specific product is carried out
using Java DB in NetBeans IDE. The secure and transactional Apache Derby DB
running on Glassfish application server is used to implement the system. Here, var-
ious operations like creation, populating DB, updating DB and deleting items were
performed using the create table command or using SQL queries. After creating
Derby DB, the details of various commonly used items along with Indian GST rates
are populated into the database. The software requirements include NetBeans ver-
sion 7.2 along with Java development kit 6.0 and bundled with Glassfish application
server. The following steps were involved in populating Indian GST rates into the
Derby DB.
1. Installing NetBeans IDE version 7.2 and above.
2. Configuring and adding Glassfish server into the environmental setup.
3. Starting server and establishing connection with the Derby DB.
4. Creating table under Derby DB.
5. Adding and updating Derby DB.
6. Deleting entries if necessary under Derby DB.
Populating Indian GST Details into Java Apache Derby Database … 415

5.4 Results and Discussions

The details of Indian GST rates for various categorical products are added into the
Derby DB using NetBeans as IDE and Glassfish application server. This system
proves to be a smart way of creating a secure, standardised and transactional Derby
DB. This system can be manipulated and enhanced with the simple SQL queries. The
following snapshots illustrate the implementation of populating Indian GST rates into
our system designed using Derby DB. Figure 1 shows the usage of NetBeans IDE 8.2
for various versions of Java. According to the version of Java and the compatibility
of operating system, the NetBeans IDE 8.2 can be downloaded. Figure 2 shows the

Fig. 1 Snapshot of NetBeans IDE 8.2

Fig. 2 Snapshot of Java DB (Derby) database

416 R. Sridevi and S. Srimathi

features to download and install Java DB Derby database. Figure 3 shows about how
to create Java DB database in NetBeans IDE. Figure 4 shows enabling Apache Derby

Fig. 3 Snapshot of creation of Java DB

Fig. 4 Snapshot of enabling Apache Derby network server

Populating Indian GST Details into Java Apache Derby Database … 417

network server. Figure 5 shows creating Java Derby DB and assigning username and
password for the created database. Figure 6 shows the way to establish connection in
Java Derby DB. After the connection establishment, Fig. 7 shows the way to rename
the DB. Figure 8 shows that Glass Fish server is made as a default application server
whilst establishing connection with Java Derby DB. Also, Fig. 9 shows the creation
of GST Table, and Fig. 10 shows how to create the columns of GST Table like SNO,
ITEMS and GST PER CENT. Finally, Figs. 11 and 12 show the way the details of
various products are added into the table and final populated GST Table for Indian
GST System.

Fig. 5 Snapshot of creating Java Derby DB

Fig. 6 Snapshot of connection establishment in Java Derby DB

418 R. Sridevi and S. Srimathi

Fig. 7 Snapshot of renaming DB after connection establishment

Fig. 8 Snapshot of application server made default

Populating Indian GST Details into Java Apache Derby Database … 419

Fig. 9 Snapshot of creating column list after creating GST table

Fig. 10 Snapshot of creating columns like SNO, ITEMS and GST PER CENT
420 R. Sridevi and S. Srimathi

Fig. 11 Snapshot of ADDING VALUES into the columns

Fig. 12 Snapshot of final populated table

Populating Indian GST Details into Java Apache Derby Database … 421

6 Conclusion

This paper deals with Indian GST rates for various categorical products into the
Derby DB using NetBeans as IDE and Glassfish application server. This system can
be manipulated and enhanced with the simple SQL queries. This system proves to be
a smart way of creating a secure, standardised and transactional Derby DB. In future,
the system can be enhanced by depicting the operations and accessing permissibility
and can be used as a back-end database system whilst designing a website for Indian
GST System.

Bibliography

1. Viswanathan B (2016) Goods and services tax (GST) in India. New Century Publications, New
Delhi
2. Gupta SS (2019) GST how to meet your obligations, 7th edn. Taxmann Tax & corporate laws
of India
3. Taxmann (2018) GST manual with GST law guide & GST practice referencer, 2 volumes, 10th
edn
4. Laddha V, Saxena S, Patwari P (2019) GST audit manual, 2nd edn.
5. Amutha D (2018) Economic consequences of GST in India, SSRN’s eLibrary
6. Nayyar A, Singh I (2018) A comprehensive analysis of goods and services tax (GST) in India.
Indian J Fin 12
7. Mehra G (2017) GST as major reform in taxation system of India. Int J Sci Res (IJSR) 6(9):497–
501
8. Jatin (2016) Awareness towards goods and services tax in India. Int J Inform Futuristic Res
4:5891–5896
9. Madesh HR, Kavya Dechamma KM Impact of good and service tax on various sectors in India.
IOSR J Bus Manage (IOSR-JBM)
10. Mahender P (2017) GST effect on manufacturing industry—India. Int J Manag Stud Res
(IJMSR) 5(1):28–30
11. Cnossen S (2013) Preparing the way for a modern GST in India. In: International tax and public
finance. Springer Science & Business Media, New York
12. Dash BB, Raja Angara V (2013) Intergovernmental transfers and tax collection in India: does
the composition of transfers matter? In: Public budgeting & finance, Blackwell Publishing Ltd.,
Maiden
13. Satish Kumar R (2018) Indian textile industry: opportunities, challenges and suggestions.
Trends Textile Eng Fashion Technol CRIMSON PUBLISHERS 2(3)

Web References

14. [Link]
15. [Link]
16. [Link]
Intrusion Detection System Using
WoSAD Method

R. Sridevi and N. Nithya

Abstract The network services are reducing the secure communication through pre-
ventive detection of intrusion services. Besides, different types of network transmis-
sion host the intrusion detection services by data centres that vary of server organisa-
tion mechanism. Network security topology is afflicted of one application service by
a different communication reciprocal that assume poor performance through appli-
cation services. Further, work factors that can affect the communication packets
performance by including network infrastructure component failure, configuration
issues, or damage cost of informal distinct components. The intrusion detection ser-
vices are the community around time for servicing a request of main user. This work
has been proposed to Whole of Service Anomaly Detection (WoSAD) methods,
which remain more effective than the alternative route transmission and to the whole
of service models.

Keywords Total cost of ownership · Whole of service · Anomaly detection ·

Intrusion detection

1 Introduction

The network intrusion services are generally a major problem of market distributors,
individuals, and disruptors of current firm models. The costs of network service fail-
ure stages or delay communications are important drawbacks for customer services
and emerging new marketing services. Information Technology Security Manager
(ITSM) in vast organisations is a complexity of energetic and highly solid firm condi-
tion in service operations. Various manufacturing standards for, e.g. ISO/IEC 20000,
COBIT substantiate possibility (e.g. Information Technology Infrastructure Library

R. Sridevi (B) · N. Nithya

Department of Computer Science and Engineering, K. Ramakrishnan College of Engineering,
Trichy, India
e-mail: sridevivelon@[Link]
N. Nithya
e-mail: nithyasrichithra@[Link]

© Springer Nature Singapore Pte Ltd. 2020 423

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
424 R. Sridevi and N. Nithya

(ITIL), Microsoft MOF, HP ITSM, and IBM ITPM) which naturally slowdowns the
services. Intrusion said to be the process of accessing data and computer resources
without authenticity thus causing hazard to security breach [1]. Intrusion detection
is like a weapon which monitors and analyse the activities going on in a computer or
network and to identify security breaches. Intrusion detection system (IDS) assess
the activities of servers, file systems, firewall, routers, and also tests the network
traffic activities that deviate from the security policy [2] and alerts the system or net-
work about the threat. These events should be avoided detection for security abilities.
Moreover, a number of firm multitudes their information services data centres that
feature varying degrees of information processing service centre streamlining. The
reasons for low performance is based on infrastructure or component failure, design
issues, or accidental cost of actualising distinct components. The principal measure
used to define user experience purely based on time taken for completing the request,
so require a robust communication to control any process corruption.
The effectiveness of detecting the network data inconsistency or abnormalities
from usual behaviour developed the system with time distribution known as grade
of service (GoS). This paper investigates the behaviour incompatibilities by sens-
ing abnormal events in complex real-world systems and summarising them with
consistent service authorisation frameworks like Information Technology Security
Manager (ITSM) and also it distinguishes ITSM frameworks such as ITIL.
Threshold alert set as a key for identifying delay in applications which compute
and observe performance activities and generate alerts every time when things move
beyond defined limits. The problem with the threshold-based notification system is
which can cause technical improvement assets become unsusceptible and the alarms
become too noisy and how it is rectified is a big question which can be addressed
by event filtering ability that reduces noise by multiple orders of magnitude but
categorises actual problems. Operationally, it can generate prohibitive storage details
and security on system presentation for huge data and ensures the monitoring systems
to be sensible to manage, balance size.
The challenge of real-time anomaly detection technique earlier than primary
source diagnostic detail is lost and/or rollup happen. This is effectively achieved com-
munity process and problem management ITSM process with modify process detec-
tion, the characterisation of event symptoms, formulation, and testing assumptions
and root cause justification.
The rest of the paper comprises of the proposed method, methodology, and
conclusion part.

2 Proposed Method

The Whole of Service Anomaly Detection (WoSAD) proposes to be extra efficiency

than another operations and whole of service models.
The WoSAD methodology is mainly used to detect, characterise, support root
cause identification, and determine tortuous systems operation anomaly problems.
Intrusion Detection System Using WoSAD Method 425

1. Whole of Service Measurement Method (WoSMM)

• Application portfolio monitoring points,

• Application architecture sampling points.

2. Whole of Service Profiling (WoSP)

• Application identification,
• Transaction load identification,
• Traffic mix identification.
3. Whole of service measurement used in system profile WoSP
The WoSAD based on data-driven model through which data behaviour rep-
resents the application presentation signature concept with time capture the
WoSMM data contour. Figure 1 depicts the WBAN connectivity diagram.

• AA scheme for WBANs scheme proven to be vulnerable securing and vali-

dating the result the system performed with the computation costs at a client
side.
• Our proposed scheme reduces the computation time and not only solves the
existing approach drawbacks, but also reduces the computation trouble on the
users.
• The lightweight cryptographic approach is verification of device collocation
and authentication scheme for WBANs.

Fig. 1 WBAN connectivity

426 R. Sridevi and N. Nithya

3 Methodology

3.1 Security Requirements

The scheme of WBAN mechanism performs a client and the service provider com-
municates independently. Accordingly, the scheme of authentication for WBANs
technology is vulnerable to different type of attacks. The agreement of secure com-
munication in WBANs schemes, the authentications are able to solve various types
of attacks. According to previous works, the authentication arrangement for WBANs
scarifies the following security supplies.
(1) Shared Substantiation:
It is important that the authentication schemes for WBANs services which can pro-
vide mutual authentication among the client and the application service provider and
limited the number of clients in accessing information.
(2) Obscurity:
To protect the client’s collective details it is mandatory that no one including or
modify the application service provider and network manager with client’s identity.
(3) Non-traceability:
The application service provider and the network manager were not capable to out-
line the client’s action priory and not sufficient for maintaining the client’s privacy
location to know necessary for authentication scheme.
(4) No Verification Table:
Usually, the verification table performs secure authentication systems. Lot of trouble
to manage the service provider has to report when connect to a new client or existing
client revoke from the system so as a result of requirement that no verification table
used by the system.
(5) Session Key Agreement:
To authenticate the confidentiality, integrity, and non-repudiation of secure data trans-
mission in WBANs scheme with session key management between the client and
server.
(6) Perfect Forward Secrecy:
The communication established between the client and the service provider in
WBANs scheme by using shared key while the adversary becomes their secret keys
by decrypting the received messages. To protect the client’s information, it is essen-
tial WBANs scheme supports forward secure communication and ensure network
user cannot receive the session key authentication level if the secret keys for client
and the application service provider are gain.
Intrusion Detection System Using WoSAD Method 427

(7) Attack Resistance:

In open network position, the secure validation for WBANs exposed to different
attacks like reply attack, impersonation attack, modification attack, stolen veri-
fier table attack, and includes man-in-the-middle attack. The safeguard security
mechanism is highly required for authentication schemes.
(8) Data Encryption:
The recommendation of a lightweight encryption algorithm as the use of secure
IoT (SIT) method which is depicted in Fig. 5. It performs a 64-bit block-cipher
mechanism and requires 64-bit key to encrypt the data for security. The algorithm is
a combination of feistel and identical substitution–permutation network architecture.
It used to advance secure IoT (ASIT) approach. The 1024-bit block cipher needs to
be 10,254-bit key computation to encryption of the secure data.
For the experimental purpose, the KDDCUP’99, especially meant for intrusion
detection data set used which can hold tcp dump portions. The data set contains a
total of 24 attack types that fall into four major categories: Denial of service (Dos),
probe, user to root (U2R), and remote to user (R2L). Each record is labelled either
as normal, or as an attack, with exactly one specific attack type. In this work, all the
UDP packets and their relevant attacks were taken and analysed with java platform.
Figure 2 indicates the process of training and test set selection, Fig. 3 mentioned about
the hybrid learning process used, and Fig. 4 shows the result which was detected as
anomaly (Fig. 5).

Fig. 2 Training data set selection

428 R. Sridevi and N. Nithya

Fig. 3 Hybrid learning process

Fig. 4 Anomaly detection

Intrusion Detection System Using WoSAD Method 429

Fig. 5 SIT algorithm

(9) Whole of Service Measurement Method:

The Whole of service measurement takes “black box” as key characteristic, begins
the measurement of application component reaction from time-to-time in end-to-end
manner that captures the reaction time and makes summary about complexity either
430 R. Sridevi and N. Nithya

“inductive” or “reductive” with the approach of component by component; top-to-

bottom down approach. The simple terms of these metrics are related to assign-
ment towards the system with the number of users participated in activity, and the
responsible time of result system.

4 Conclusion

In this paper, proposed to WoSAD methodology is proposed to combination of data

communication model. The performance of application signature is behavioural data-
driven denotes existent time attack of the WoSMM data profile. The GoS process
proxy service the model of data-driven associated with online performance distri-
bution patterns captured in queue models such as M/M/1 model. The GoS-based
WoSAD performance communication proxy “k” found efficient detection model;
here, substitutions are used to every transaction in acceptable score boundaries.
The issue of key with the least expeditious individual business models needs to
de-sensitise them to incomprehensibility from separate transference inconsistency.
This communication device has noise seems to be spontaneously reasonable
payable for inconsistency calculations across the many individual communication
transaction types are involved. The entire service of mean plus standard deviation
response time detection models are less efficiency. This scheme relies on statistical
metrics that clarify symmetrical distributions. These metrics were not describing any
online transaction time behaviour which is asymmetrical in setting of response. The
WoSAD techniques at entire surrounds such as all positions ends are not dependent
on specific system mechanism. This work proposed to Whole of Service Anomaly
Detection (WoSAD) methods are established to be effective than alternative route
transmission and whole of service models.

References

1. Gow R, Rabhi FA, Venugopal S (2018) Anomaly detection in complex real world application
systems. IEEE Trans Netw Serv Manage
2. Zhang T, Zhu Q (2018) Distributed privacy—preserving collaborative intrusion detection
systems for VANETs. IEEE
Blackhole Attack Implementation and Its
Performance Evaluation Using AODV
Routing in MANET

Anshu Kumari, Madhvi Singhal and Nishi Yadav

Abstract Mobile Ad hoc Network (MANET) is a self-sorted out remote system,

comprising of independent nodes. The correspondence in the MANET is of multihop
in nature because of non-attendance of any settled foundation or centralized base. An
assailant, it may encroach effectively into MANET by acting like authentic middle of
the road hub and present different kinds of security assaults on information exchange
occurring among source and goal. In this paper, we have simulated the blackhole
assault in AODV reactive routing protocol of MANET and investigated its viability
by considering various performance metrics.

Keywords MANET · Blackhole · Performance · PDR · Throughput

1 Introduction

MANET is a remote system where a gathering of mobile nodes can progressively

change the topological structure [1]. It is a self-arranging and self-designing system.
A hub can discuss specifically with its nearby hubs just which are available inside its
transmission go. The hubs communicate with one another by means of radio waves.
The hub goes about as both sender and receiver. MANETs do not utilize any type
of settled foundation or unified organization that we see by and large if there should
be an occurrence of all the more broadly utilized mobile node systems. MANETs
are anything but difficult to convey and design which results in their prevalence
in contrast with wired systems. The principle highlight of MANET is its moment
organized setup. In any case, routing in MANET is a test because of its dynamic
topology in the system as portable hubs can move toward any path in the MANET

A. Kumari · M. Singhal (B) · N. Yadav (B)

Department of Computer Science & Engineering, Guru Ghasidas University, Bilaspur, India
e-mail: madhvisinghal4@[Link]
N. Yadav
e-mail: nishidv@[Link]
A. Kumari
e-mail: anshu1997kri@[Link]
© Springer Nature Singapore Pte Ltd. 2020 431
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
432 A. Kumari et al.

[2]. MANET is useful in spots that does not have any correspondences foundation
or once that framework is extremely broken [2]. These sorts of systems have the
resulting striking qualities: dynamic topologies, data transfer capacity compelled
variable limit joins, restricted physical security and vitality-obliged activities.
There are mainly three types of routing algorithms available namely proactive,
reactive and hybrid (mixture of reactive/proactive) routing algorithms. Proactive is
a table-driven routing protocol. In this, each node keeps up the rundowns of all con-
ceivable goal hubs in a table. To keep the data in the routing table updated, each
node intermittently exchanges routing messages [3]. A change in the routing table is
informed to all other nodes in the framework through flooding technique. Distributed
Sequenced Distance Vector (DSDV) is a proactive protocol [4]. Ad hoc On-Demand
Distance Vector (AODV) and Dynamic Source Routing (DSR) are reactive routing
protocols [4] which follow on-demand routing where the routes are found only just
when required. When route is required, a type of route discovery system is utilized
which results in large control message traffic. Since these conventions expect par-
ticipation between two hubs for packet sending, a noxious hub may prompt routing
assault in the system that disturbs the ordinary routing tasks of MANET [3]. There-
fore, decentralized and dynamic nature of MANET may prompt different assaults in
the system that can degrade the working of the system [3].
The MANET is prone to various types of attacks which can be classified into two
broad categories, one is passive attacks and other is active attacks. A passive attacker
monitors the correspondence channel and snoops the data being exchanged in the
network without altering it. Thus, a passive attacker does not disrupt proper operation
of the system but it is very difficult to be detected. While an active attacker attempts
to alter the principal data being exchanged and disrupts the normal working of the
system which makes it easy to detect as compared to passive attacker. Because of
immaterial structure and quick course of action, MANETs are fitting for emergency
conditions like natural calamities rescue activity, medical clinics, war zone, meetings
and military applications. Thusly, data trade between two center points must require
security. In any case, the extremely basic assaults in MANET are blackhole attack [5–
8], wormhole attack which have stunning effect on the efficiency of the framework.
Blackhole assault is a remarkable kind of attack that generally occurs in the
reactive algorithms. A blackhole hub is the dangerous node that pulls in the packets
by incorrectly attesting that it has most short and fresh way to reach the goal, by then
drops the packets. These blackhole hubs may perform distinctive damaging exercises
on the framework. For instance, it may carry on as a source node by distorting the
route request packet, may go about as a destination center point by twisting the route
reply packet, or may lessen the quantity of hop counts, when sending route request
packet.
Blackhole Attack Implementation and Its Performance Evaluation … 433

2 Related Work

In past, there have been various attempts for countering the blackhole attacks. Some of
them that we overviewed are as follows: In [9], the author has proposed a technique to
locate the blackhole attack on AODV protocol in MANET. This technique considers
that the first route reply packet that it has received is the reaction from malicious
node and, therefore, removes that node from the network. And when the second
route reply packet it receives, is considered for the route reply saving mechanism as
it originates from the goal hub. He has named this technique as blackhole detection
system. The modified AODV with this BDS arrangement against blackhole node has
high packet delivery ratio when contrasted with the already existing AODV protocol
under blackhole attack. In [10], the proposed method handles with the multiple
blackhole nodes assault in MANET. To deal with the numerous blackhole nodes
assault, the source hub utilizes the sequence number idea to distinguish the various
blackhole nodes in MANET. The source sequence number is utilized by the source
hub to detect the blackhole assault. In [8], the authors proposed a technique that uses
the fake route request packets to detect and isolate blackhole attack in MANET. The
fundamental plan to distinguish and confine noxious hubs is which the utilization of
fake messages. Whenever the source hub needs route to goal, it will flood counterfeit
route request packets in the system. These route request packets are made counterfeit
by including the IP address of the hub which is not present in the system. So when
the hub answers with the route reply packet, it will be identified as the malevolent
node and can be segregated from the system. In [3], the authors have proposed a
clustering approach to identify the blackhole nodes in AODV routing protocol of
MANETs. In order to identify the impossible to miss contrast between the number
of information packets got and sent by the hub, each individual from the cluster will
ping once to the cluster head. On the off chance that anomalousness is seen, every
one of the hubs will obscure the malicious hubs from the system.

3 Proposed Work

In this paper, we have considered the implementation and effect of blackhole attack
on AODV routing protocol. Softwares that we have used for this are NS-2.35, NSG2.1
and gawk. NS-2.35 is a network simulator. NSG2.1 is a network simulator generator
which has been used to [Link] file which is run on NS-2.35. GAWK has been
used to read the trace (.tr) file for analyzing the performance of the network. AODV
protocol is already present in NS-2.35. For better analysis of the result, we have
considered changing number of nodes (100, 200, 300, 400 and 500). After simulation,
packet delivery ratio and throughput have been calculated considering the following
definition.
Packet Delivery Ratio (PDR): The ratio of the number of packets received by TCP
sink to the number of packets delivered by the TCP source.
434 A. Kumari et al.

Throughput: Throughput of the network is defined as the number of bits of data that
are received by the goal node per unit time. It is the ratio of number of bits received
by the goal node to the total time taken.
Packet Drop Ratio: Packet drop ratio tells about the network performance in terms
of packets that have been dropped by the nodes. It has been calculated as:

Packet Drop Ratio = (Number of packets received by node − Number of packets

forwarded by node)/Number of packets received by node

The Simulation environment is as follows (Fig. 1; Table 1):

Fig. 1 Simulation of AODV protocol in NS-2.35

Table 1 Simulation
Parameters Values
environment parameters
Simulator NS-2.35
Protocols AODV
No. of nodes 100, 200, 300, 400, 500
Topology Grid
Simulation time 200 s
Traffic type FTP
Propagation model Two ray ground
Max packet in queue 20
Number of blackhole node 1
Packet size 500 bytes
Blackhole Attack Implementation and Its Performance Evaluation … 435

Fig. 2 Addition of the “bhassaultaodv” protocol agent in the “\tcl\lib\[Link]” file

Fig. 3 Addition in the “\makefile” at the NS-2.35 directory

First of all, we analyzed the performance of already existing AODV protocol

utilizing system parameters packet delivery ratio, throughput and packet drop ratio
for each node. Then, we implemented blackhole node behavior by modifying the
existing AODV. For this, we renamed the file [Link] as [Link] and then
modified its content as per our need. To set a routing agent to the node, we included
the following lines of code in “\tcl\lib\[Link]” file (Fig. 2).
Then, compile the NS2 to create object files. Then, the “/makefile/” is added with
following lines in the root directory NS-2.35 (Fig. 3).
Then, we make a node as a blackhole node. We have made node 21 as a blackhole
node in our simulation. When a packet utilizing AODV protocol is received, recv()
is called which processes the packet according to its type. If it is an RREQ, RREP or
RERR packet, then recvAODV() is called and if it is a data packet, then it is routed to
the destination. When the data packet reaches a blackhole node, it is dropped there
and hence, does not reach to its goal.
In order to incorporate these changes in the required files, open the terminal and run
“make” command inside “ns-allinone-2.35/ns-2.35/” directory. If it run successfully,
then our new AODV protocol is included successfully. A new tcl file is created with
one blackhole node in the system to simulate the new protocol and run into the
terminal. Its performance is analyzed utilizing system parameters packet delivery
ratio, throughput and packet drop ratio for each node.

4 Simulation Result

From Fig. 4, we can conclude that in case of blackhole attack, the PDR value of
network decreases irrespective of the number of nodes. Drop in PDR value can be
seen in small as well as large network. Figure 5 reveals that throughput is not much
affected by the size of network. It remains almost same on increasing or decreasing
436 A. Kumari et al.

PDR PDR with Blackhole

0.98

0.96

0.94
Packet Delivery Ratio

0.92

0.9

0.88

0.86

0.84

0.82

0.8
100 200 300 400 500
Number of Nodes

Fig. 4 Packet delivery ratio versus number of nodes

Throughput Throughput with Blackhole

0.16

0.14
Throughput (in kbps)

0.12

0.1

0.08

0.06

0.04

0.02

0
100 200 300 400 500
Number of Nodes

Fig. 5 Throughput versus number of nodes

Blackhole Attack Implementation and Its Performance Evaluation … 437

Normal Black hole

2.5
Packet Drop Ratio

1.5

0.5

0
0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72 76 80 84 88 92 96
Node ID

Fig. 6 Packet drop ratio versus node ID

Table 2 PDR and throughput with and without attack

No. of nodes PDR PDR with blackhole Throughput Throughput with
blackhole
100 0.939671 0.901709 0.130594 0.038083
200 0.942738 0.864469 0.12708 0.02037
300 0.939394 0.891626 0.140837 0.032999
400 0.953842 0.920188 0.13592 0.031706
500 0.953842 0.920188 0.13592 0.031706

number of nodes in the network. But it gets affected by blackhole attack, and its value
is dropped to large extent on adding malicious node. Figure 6 presents the packet
drop ratio of each node in the network. From the graph, it can be seen that drop ratio
of packet has increased after adding malicious node. Its effect can be seen in whole
network (Table 2).

5 Conclusions

In this paper, we presume that because of self-configuring nature of the MANET

much kind of inside and outside assaults are conceivable which debases the system
performance. Among all the security assaults, blackhole attack is the most widely
recognized and denial of service attack. We have simulated the blackhole attack in
NS2 by modifying the current AODV protocol and results are analyzed graphically by
taking different system parameters like throughput, packet delivery ratio and packet
drop ratio of each node. We have additionally analyzed these parameters on differing
number of nodes in the system. In the wake of analyzing the result, we found that
because of single blackhole node present in the system, packet drop ratio for almost
every hub is increased which debases the PDR value and throughput of the entire
system.
438 A. Kumari et al.

Acknowledgements Working on this project on “Blackhole Attack Implementation and its Per-
formance Evaluation using AODV Routing in MANET” was a source of immense knowledge to us.
We would like to express our sincere gratitude to our guide Mrs. Nishi Yadav for her guidance and
valuable support throughout the course of this project work. We acknowledge with a deep sense of
gratitude, the encouragement and inspiration received from our faculty members and classmates.

References

1. Mirza S, Bakshi SZ (2018) Introduction to MANET. Int Res J Eng Technol 5(1)
2. Desai Piyusha P (2013) A study of secure routing in mobile ad-hoc networks. Int J Adv Res
Comput Eng Technol 2(5)
3. Rashmi AS (2014) Detection and prevention of black-hole attack in MANETS. Int J Comput
Sci Trends Technol 2(4)
4. Banwari DS, Upadhyay D (2013) Routing algorithms for MANET: a comparative study. Int J
Eng Technol 2(9)
5. Kaur G, Kaur N (2014) Review of attacks on MANETS. Int J Innov Res Comput Commun
Eng 2(6)
6. Madhuri K, Kasi Viswanath N, Usha Gayatri P (2016) Performance evaluation of AODV
under black hole attack in MANET using NS2. In: International conference on ICT in business
industry & government. IEEE. [Link]
7. Ghonge M, Nimbhorkar SU (2012) Simulation of AODV under blackhole attack in MANET.
Int J Adv Res Comput Sci Softw Eng. ISSN: 2277 128X
8. Kaur J, Singh B (2014) Detect and isolate black hole attack in MANET using AODV protocol.
Int J Adv Res Comput Eng Technol 3(2)
9. Koujalagi A (2018) Considerable detection of black hole attack and analyzing its performance
on AODV routing protocol in MANET (Mobile Ad Hoc Network). Am J Comput Sci Inf
Technol 6(2)
10. Kalia N, Sharma H (2016) Detection of multiple black hole nodes attack in MANET by
modifying AODV protocol. Int J Comput Sci Eng 8(5)
Enhancement of Security Using B-RSA
Algorithm

Aman Gupta, Saurabh Gupta and Nishi Yadav

Abstract Cryptography is a scientific art which deals with the methods for changing
over messages, data and information in some dynamically and haphazardly planned
language of characters which are mixed up and not understandable for people or
notwithstanding for a machine. While doing this, it is guaranteed that authorized
frameworks and humans can recuperate the original message by validating their
authentication. There are different calculations and strategies are proposed for cryp-
tography purposes. This paper focuses on examination of two individual crypto-
graphic calculation RSA and Blowfish, how they work and what will be the impact
on security and speed if these two are converged into one hybrid calculation with
appropriate modification. This paper presents the comparative analysis in terms of
encryption and decryption time for individual as well as hybrid algorithm.

Keywords Cryptography · RSA · Blowfish · Encryption · Decryption ·

Symmetric and asymmetric key

1 Introduction

Technical innovations are getting advanced and spreading its wide region so fastly
that at present we find ourselves doing each kind of the works by technical means.
We are utilizing web not only for surfing the website pages but also for various daily
to daily life works and needs like online shopping, cash exchange, secret military
administrations, correspondence over social web media, and so forth. That is the
reason there is have to utilize our own personal information for finishing undertakings
like check card subtleties while online transactional exchanges. Consequently, we

A. Gupta (B) · S. Gupta · N. Yadav

Department of C.S.E, Guru Ghasidas Central University, Bilaspur, India
e-mail: ag4cse@[Link]
S. Gupta
e-mail: mohitsaurav12u@[Link]
N. Yadav
e-mail: nishidv@[Link]
© Springer Nature Singapore Pte Ltd. 2020 439
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
440 A. Gupta et al.

need such security component which can guarantee that our information is secure and
is spreading over web in an incomprehensible structure so that regardless of whether
interloper gain admittance to the information by, in any case, it won’t most likely
comprehend the real data or information. Web itself don’t represent accessing our
data for utilization of danger, however, a few interlopers with expectations of mischief
go over our information by unapproved implies. At this situation idea of system or
network security in terms of cryptographic art come into the action. Cryptographic
algorithms are not just in charge of changing over message into confused structure
yet a decent and secure algorithm additionally gives the confirmation to essential
objectives of system security. Essential objectives of network security are:
(i) Confidentiality, which guarantees getting to information just by approved
elements means maintaining secrecy
(ii) Integrity, it manages the exactness part of the information and
(iii) Availability, which guarantees non-renouncement to the authentic clients.
Cryptography process revolves around certain terminologies. Original data which
is intended to be send is called plain-text given as an input to the algorithm of cryptog-
raphy. The yield of the calculation which is ambiguous coded structure information
is called cipher-text. Conversion procedure of the plain-text into cipher-text is called
encryption and switch is called as decryption. Cryptography algorithms can be fur-
ther divided on basis of principles of processing data and keys also known as cipher
keys.
Cryptographic strategies depend on two principles. On the off chance that each
littlest individual unit of plaintext is mapped with another unit, at that point, it is
called substitution and on the off chance that the units of plaintext are reordered, at
that point, it is known as standard of transposition.
For encryption and decryption algorithms uses cipher keys which are the soul for
any algorithm because methods are publicly known to everyone. The only thing that
is meant to be secret is keys which drives the whole algorithm. When both sender
and receiver use the same key for respective encryption and decryption process, it
is called symmetric or private key cryptography and when two different keys are
used by either side for encryption and decryption is known as asymmetric or public-
key cryptography. Blowfish is one of the examples of symmetric type and RSA is
asymmetric type of algorithm.

1.1 RSA

RSA is one of the broadly used asymmetric key kinds of cryptography algorithm.
RSA remains as an abbreviation for the surname initials of the makers of this algo-
rithm—Rivest, Shamir, and Adleman. RSA calculation utilizes the idea of consider-
ing factoring issue in number system as it finds as one of the basically troublesome
Enhancement of Security Using B-RSA Algorithm 441

problem. As indicated by this problem factorization of result of two huge prime num-
bers is by one way or another isn’t a bread and butter case. And thus, this algorithm
takes advantage of this problem to make itself much secure.
Steps involving in encryption and decryption of RSA algorithm are as follows.

1.1.1 Key Generation

(1) Choose any two dissimilar large random prime numbers say i and j.
(2) Calculate the value of n as product of i and j.

n=i∗ j

modulus operation of this algorithm will use the value of n.

(3) Totient function  “phi”.

(n) = (i) ∗ ( j)
(n) = (i − 1) ∗ ( j − 1) // (k) = k − 1; if k is prime.
(n) = n−(i + j−1)

(4) Select the value of public key (e) in such a way so that e is relative prime to
 (n).

gcd(e, (n)) = 1{1 < e < (n)}

Here e is our public key and pair (e, n) will use at encryption side.
(5) Compute the value of private key (d) such that congruence relation does exist.

d ∗ e ≡ 1(mod(n))

Here d is our private key and pair (d, n) will use at decryption side.
(6) Encryption: Let plain-text at sender side is M. Receiver side public key (n, e)
is known to everyone thus sender also knows. Private key d remains only up to
receiver side. It is not meant to be shared.
(7) First, convert M into a number m (m < n) by using any pre-defined standards but
is ensured that both sender and receiver is aware with that standard or protocol.
And then finally cipher-text c is computed as.

c = m e mod n

Now, this cipher-text c is sent to receiver side.

(8) Decryption: To recuperate original message from ciphertext at receiver side it
requires the private key of receiver d. Original message m is recovered as
442 A. Gupta et al.

m = cd mod n

For increasing the security of RSA one new approach was given in which four
prime numbers were used and modulus operation was performed twice unlike
conventional RSA. This method was named as D-RSA [1].

1.1.2 Key Generation in D-RSA [1]

(1) Modulus keys:

n1 = q1 ∗ q2 and n2 = r1 ∗ r2

(2) Totient function:

(n1) = (q1 − 1) ∗ (q2 − 1)

(n2) = (r1 − 1) ∗ (r2 − 1)

(3) Public keys (e1, e2):

gcd (e1, (n1)) = 1

gcd (e2, (n2)) = 1

(4) Private keys (d1, d2):

d1 ∗ e1 ≡ 1 (mod(n1))
d2 ∗ e2 ≡ 1 (mod(n2))

(5) D-RSA encryption:

 e2 
c= m e1 mod n1 mod n2

(6) D-RSA decryption:

 d1 
m= cd2 mod n2 mod n1

1.2 Blowfish

Blowfish is a kind of symmetric-key algorithm which finds its place as one of the
very fast algorithms as well as it is very compact. Blowfish is structured by the
Enhancement of Security Using B-RSA Algorithm 443

Bruce Schneier in year of 1993. This algorithm is openly accessible and unpatented
algorithm. It is a sort of block cipher algorithm which encrypts plain-message as in
64-bit square sizes. One important feature of this algorithm is that it utilizes variable-
length key up to 448 bits beginning from 32 bits. The soul of this algorithm is use of
substitution boxes also called S-Boxes.
Steps involved in cryptography process of Blowfish algorithm:
This algorithm works in two sections; one that deals with key expansion phase
and other one handles the main encryption process.
In key expansion phase subkeys are generated. For subkey generation steps:-
(1) First of all, keys are stored in 14 K—arrays (K1–K14) each of size 32 bit. In
this way, variable-length key will have upper bound to 448 bits (14*32).
(2) Initialize P-array and S-boxes.
(3) Both are initialized with the hexadecimal values of “pi”.
(4) There are 18 P-arrays each of 32 bit (P1–P18) and four S-Boxes each containing
256 entries of 32 bits.
(5) Now bitwise XOR operation is performed between P-arrays and 32-bit keys.

P1 = P1 XOR K1
P2 = P2 XOR K2
P14 = P14 XOR K14
P15 = P15 XOR K1
P18 = P18 XOR K4

(6) In this way, 64-bit plain text when passed through Blowfish encryption process,
subkeys are generated by using above method. Subkeys are generated only when
all the P-arrays and S-Boxes are replaced.

1.2.1 Encryption Phase

For encryption, plain-text is divided into block of 64 bits. Let one such block is
named as B. Further, these 64 bits are divided into two equal blocks size of 32 bits
and named as BL and BR .
First, 32 bits that is BL are bitwise XORed with 32-bit P1 array. The output let X is
now given to the F-function of Blowfish algorithm and the output of this function let
Y is now again XORed with 32 bits of BR . Now this output becomes the left part of
the algorithm which will continue its XOR with next P2 array and the output of first
XOR that is X now becomes the right part of the algorithm. This process continues
18 times as we have to XORed with all 18 P-arrays. After 18 rounds we get the
ciphertext which will be of 64 bits in size equal to the input plain-text one block size
(Figs. 1 and 2).
444 A. Gupta et al.

Fig. 1 Blowfish encryption

Fig. 2 F-function working

Enhancement of Security Using B-RSA Algorithm 445

1.2.2 Working of F-Function

The output of BL and P1 that is X which is first XOR now works as input to the F
function. This X is now split into 4 equal parts of each 8 bits now given as input to
the four S-boxes. Each S-box gets 8 bits and produces the output of 32 bit by using
its substitution process. Output of S-box S1 is XORed now with S2 and output is
again XORed now with S3 and finally their output is XORed with S4 and we get
output as Y which will become the input to the XORed with BR .

1.2.3 Decryption

For decryption phase, the reverse approach of encryption phase is applied. We start
with the cipher-text and go up to plain-text while processing the 18 steps of XOR’s
with P arrays. And finally, 64-bit ciphertext is now converted into 64 bit of plain-text.

2 Literature Survey

In this paper [1] Ekka et al. proposed a hybrid algorithm of AES and RSA pursued
by EX-OR activities. By doing this they reasoned that proposed plan takes lesser
time in encryption stage than decoding stage, so security increments.
In this paper [2] Quilala et al. utilizes the Modified Blowfish algorithm for provid-
ing security to the Electronic Medical records. Authors did appropriate modification
in original Blowfish algorithm to make EMR system more secure.
In this paper [3] Kaur and Singh utilizes the Blowfish algorithm for image encryp-
tion. Authors takes advantage of Random selective block encryption for Image
encryption. Authors compared the encryption and decryption time of their algorithm
with standard methods and find that results are better as compared to the previous.
In this paper [4] Manu and Goel used the RSA algorithm utilizing twofold encryp-
tion and twofold decoding utilizing twofold open and private keys to give security
against Brute Force assault. This paper beats the shortcoming of customary RSA
that is on the off chance that we can figure modulus its prime numbers, at that point
private key will be in risk.
In this paper [5] Vasantha and Prasad proposed a half and half security algorithm
for RSA cryptosystem. Here they processed public and private keys by utilizing four
prime numbers. After that they looked at the different procedure times and speed
with customary RSA.
In this paper [6] Panda and Chattopadhyay did their work on investigating and
planning of upgraded RSA for enhancing the security by including the element of
check at both sender and beneficiary side. They improved their work by utilizing
K-NN calculation just as numerous public keys.
In this paper [7] Mathur et al. used the Blowfish algorithm for providing security
aspects in internet of things. Authors slightly modified the original algorithm by
446 A. Gupta et al.

changing in F-function of the algorithm and found that in terms of throughput and
speed this algorithm significantly proved fruitful.
In this paper [8] Suresh and Neema used a hybrid algorithm for cloud computing.
They selected RSA and Blowfish algorithm for this work. For implementation pur-
pose, they used FPGA device. In this way authors provides one algorithm for data
security at cloud computing level.
In this paper [9] Bansal and Singh performed the analysis part on basis of imple-
mentations of various symmetric key-based cryptography like Blowfish. Their basis
for comparison involves three parameters what is the execution speed, size of block
for processing and size of cipher keys.

3 Proposed Work

RSA algorithm which is public key algorithm on one side have favorable position of
high security in view of complex factorization issue yet on other hand it gets moderate
down in calculation with regards to the matter of encryption of bigger information.
While Blowfish algorithm has preferred standpoint of one it is unpatented openly
accessible to utilize algorithm and second it is one of the quickest block cipher
algorithms in symmetric key algorithm.
Presently this paper introduces a half and half algorithm which is made by con-
solidating such two calculations with the goal that recently planned calculation will
have forces of RSA as far as mind-boggling security and on other side quickness and
unwavering quality of Blowfish calculation. This new algorithm is named as B-RSA.

3.1 B-RSA Encryption

This algorithm first takes plain-text as M. Then this message is first encrypted by
blowfish algorithm using symmetric key process of this algorithm. All the interme-
diate process of these algorithms is applied starting from subkey generation and then
splitting plaintext in 64-bit blocks and then all 18 rounds of XORed along with use
of S-Boxes in F-function. And finally, message M is now in encrypted form that is
ciphertext let C1. Now this ciphertext is used as a plain text for RSA algorithm. Here
we use RSA algorithm of 4 prime numbers and dual modulus operation. Now this
algorithm converts C1 into another ciphertext C2. And finally, original plaintext M
is now in encrypted form as C2 by B-RSA algorithm.
Enhancement of Security Using B-RSA Algorithm 447

3.2 B-RSA Decryption

For decryption phase at receiver side, the reverse approach of encryption phase is
done. Ciphertext C2 is decrypted using dual modulus RSA algorithm. As a result
of this we get the output as C1 which is still not our original plaintext. Now this
Ciphertext C1 is decrypted by Blowfish algorithm decryption phase. After completing
all the 18 rounds of XOR’s in order of P18–P1 array we get the original plaintext as
M by B-RSA algorithm.

3.3 Example

Let us take plain-text as “hello” which is meant to be encrypted. Let denote this
message as M. first of all M is given as input to Blowfish algorithm phase of B-RSA.
Blowfish Symmetric key = ??6|(??6???¿ßÉ Cipher text C1 = z]ñz?H?+
Now, this C1 is becomes input to RSA algorithm with dual modulus. Ciphertext
C1 in bytes = 12293-15122-104726343

Encrypted bytes by RSA phase = 76-16-23127-747-2121-51-71-96-45-52115-67-

96-12329-139-71-101-10-101-5918-72-442-26-3123-1185911084-28-22-43109-
646-77-2314-6995106-24108-10443-123-80-25-50-100-7248-71-120-476-31123-
1210511132-994604279-10-54-7614-33-1124-731-10399109-8420-502359-10-12-
96-3065-32-126-123-67-22313416704194-108-30-50-32-8-5-119126-22-72-399-
10165-34-9764-120-105-7342-121-84-2-112-49-29-67-29-21-341-702-43-125-10
287619973541259111-1171-5229-42100-74119749-838111469591-9-19-126-
110-23926-93-12083-86116-104-40-10910635653-42-8612537-100-62857-48-
2648113-58-96-8127-96-49-99-3286-49-71-3654-12512-56-948-60119-10782-
187815-30-37137-61113-9110192-9382-94-75-13126-149711973-54-7-664783
599109-79-10070-36

Now decryption phase starts. These encrypted bytes are now decrypted using
RSA-decryption phase.
Decrypted string C1 = z]ñz?H?+
Now this C1 is gives as input to blowfish-decryption phase and we get our original
plaintext as M.
M = “hello”
448 A. Gupta et al.

4 Simulation Work

In this paper, we discussed a new algorithm B-RSA which is hybrid of one sym-
metric and one asymmetric algorithm Blowfish and RSA respectively. Here we are
comparing the Encryption and Decryption time of Blowfish algorithm with proposed
B-RSA algorithm. We implemented these algorithms in java language and measured
the execution time of these algorithm’s encryption and decryption phase.
From Figs. 3 and 4 we can see that decoding time of B-RSA algorithm increments
as contrast with the first blowfish calculation. It demonstrates that increase in decod-
ing time will make B-RSA calculation substantially more secure towards assaults

Fig. 3 Encryption time versus number of input character

Fig. 4 Decryption time versus number of input character

Enhancement of Security Using B-RSA Algorithm 449

on the system on the grounds that such mind-boggling and difficult to unscramble
result upgrades the security as contrasted with blowfish algorithm. In same way we
find that encryption time likewise expanded when contrasted with original blowfish
algorithm. It is a direct result of the extra periods of RSA which are included into
blowfish algorithm. On the off chance that there is need of making an increasingly
secure system at expense of high handling time of encryption and decoding then
parameters of RSA algorithm can be expanded like we can go to 6 prime numbers
with triple modulus task. In this paper we utilized RSA of double modulus. Making
triple modulus in action will make framework or system significantly more secure.

5 Conclusion

From the overall estimations, we presume that proposed B-RSA algorithm have the
abilities of both RSA and Blowfish algorithm. Regardless of having perplexing and
secure mechanism like double modulus RSA calculation, B-RSA is additionally
quick block cipher calculation. Here we can say utilizing of B-RSA algorithm will
be by one way or another different and significantly more difficult for gatecrashers
to break the component of calculation of B-RSA algorithm. In this way we infer
that B-RSA algorithm is such calculation which full-fill the prerequisites of a decent
cryptographic algorithm having secure mechanism just as quick execution.

References

1. Ekka D, Kumari M Yadav N (2019) Enrichment of security using hybrid algorithm. In:
International conference on computer networks and communication technologies, Springer,
Singapore
2. Quilala TFG, Sison AM, Medina RP (2018) Securing electronic medical records using modified
blowfish algorithm. Indonesian J Electr Eng Inf (IJEEI) 6(3):309–316
3. Kaur A, Singh G (2018) A random selective block encryption technique for secure image
cryptography using blowfish algorithm. In: Proceedings of the 2nd international conference
on inventive communication and computational technologies (ICICCT 2018)
4. Manu and Goel A (2017) In: 3rd IEEE international conference on computational intelligence
and communication technology (IEEE-CICT 2017), ABES engineering college Ghaziabad, India
5. Vasantha R, Prasad RS (2017) An advanced security analysis by using blowfish algorithm. Int
J Sci Res Comput Sci Eng Inf Technol 2017
6. Panda PK, Chattopadhyay S (2017) A hybrid security algorithm for RSA cryptosystem. In:
International conference on advanced computing and communication system (ICACCS-2017),
Coimbatore, INDIA, 2017
7. Mathur S, Gupta D, Goar V, Kuri M (2017) Analysis and design of enhanced RSA algorithm to
improve the security. In: 3rd IEEE international conference on computational intelligence and
communication technology (IEEE-CICT 2017)
450 A. Gupta et al.

8. Suresh M, Neema M (2016) Hardware implementation of blowfish algorithm for the secure data
transmission in internet of things. In: Global colloquium in recent advancement and effectual
researches in engineering, science and technology (RAEREST 2016)
9. Bansal VP, Singh S (2015) A hybrid data encryption technique using RSA and blowfish for cloud
computing on FPGAs. In: Proceedings of 2015 RAECS. UIET Panjab University Chandigarh
21–22nd December 2015
Analyzing Different Multiparty
Computation Techniques

Tanmay Borade, Dhananjay Dakhane and Tushar Ghorpade

Abstract Strong encryption provides support to data privacy. Although encryption

can make the data secure in data transfer and at rest, at some point this encrypted data
definitely needs to be decrypted. Now at this very point, it so happens that the data
becomes susceptible to attacks ultimately resulting in compromised data privacy.
This is where secure multiparty computation (MPC) comes into picture; thereby, it
provides ability to calculate required values from numerous encrypted data sources
without any party compromising on their secret data. At ground level, MPC is a
very general concept that can be realized using different protocols, such as secret
sharing, in which secret data from each party is divided and then distributed ran-
domly, encrypted “shares” among the parties. This distributed data when eventually
aggregated would provide the final desired result. If anyone happens to intersect the
data at hand of any of the parties, it would prove futile. With this paper, we focus
on different algorithms or techniques that work behind the scenes in implement-
ing MPC. These techniques include homomorphic encryption, followed by RSA
combined with Paillier’s algorithm and lastly the concept of garbled circuits.

Keywords Multiparty computation · Homomorphic encryption · Garbled circuits ·

Paillier’s algorithm

T. Borade (B) · D. Dakhane · T. Ghorpade

Ramrao Adik Institute of Technology, Navi Mumbai, India
e-mail: [Link]@[Link]
D. Dakhane
e-mail: ddakhane@[Link]
T. Ghorpade
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 451

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
452 T. Borade et al.

1 Introduction

With the advent of modern world technology, today’s world revolves around data.
This data is sent over a global-wide network; hence, the network security and anal-
ysis have become very prime importance for any large organization or establish-
ment. Computers today are involved in large money transfers between banks, IT
industries, stock markets, electronics and telecommunication industries, railway and
rail services, schools and colleges, satellites and space research, etc. In such crucial
areas, security cannot be compromised. Networks could be susceptible to attacks like
Phishing, Eavesdropping, DOS and DDOS attacks, and Man-in-the-middle attack.
To prevent all of this, network security is essential. Cryptography essentially means
converting just an ordinary plain text into scrambled and clueless text and vice versa.
A. Secure Multiparty Computation
The term “multiparty computation” was coined by Yao [1], arguably regarded as
the most fundamental problem in cryptography. Being a powerful abstraction, it can
model any cryptographic task. The problem shown in Fig. 1 is defined as follows:
We have a set of n distrusting parties {P1,…,Pn}, each with its own private input
x1,…, xn. They want to compute some publicly known function f on their inputs
without disclosing their inputs. The approach used by a generic secure computation

Fig. 1 A generalized scenario of secure MPC

Analyzing Different Multiparty Computation Techniques 453

protocol is to “securely” evaluate Boolean circuit (with AND, OR, XOR gates) or
arithmetic circuit (with + or * operator) representing the function f to be computed.
By “secure circuit evaluation,” we mean that the circuit will be evaluated in such
a way that nothing other than the circuit output that represents the function output
must be revealed during the circuit evaluation.
B. Yao’s Millionaire Problem

Yao’s millionaires’ problem [2] is a secure multiparty computation problem given by

Andrew Yao which came into light in 1982. The problem involves two millionaires,
say, Alice and Bob, who want to ascertain who is richer. The condition is that they
should not reveal their true wealth. This complication is very similar to another
problem that has two numbers a and b, and the goal is to solve the inequality a > =
b without exposing the definite values of a and b.

2 Related Work

Rani et al. [3] explain the fourth challenge of big data, that is, Veracity, which means
how much trustworthy is the data, how securely is the data received, stored, pro-
cessed, and transmitted. Zhenlin et al. [4] aim at constructing a predicate encryption
having multiplicative homomorphic property for the class of inner-product predi-
cates. Yao et al. [5] propose a protocol based on homomorphic encryption which
allows function input to be encrypted with various public keys. Based on ECC for
SMC problem, Wang et al. [1] present a homomorphic encryption scheme that results
in reduced communication and computation cost. Yao [2] in his paper presents a new
mechanism for regulating the knowledge transfer process in cryptographic protocol
design. Dhakar and Gupta [6] point out the loopholes in RSA cryptosystem: factoring
huge numbers by mathematical attack and, with brute force, trying all private keys.
Xiang et al. [7] define a new operation—similar modular arithmetic which can be
used to achieve algebraic homomorphic encryption scheme on the scope of a rational.
Youn et al. [8] put forth an additive HE scheme, used for calculating statistical data
such as mean and variance. Naumann [9] in this paper has explained the function
of Yao’s original algorithm in detail. Snyder [10] in his paper gives a thorough and
detailed answer of Yao’s original protocol and its security characteristics. Cimato
et al. [11] in his paper put forth the technique of using multiple-valued logic for the
forming garbled circuits. Ehsanpour [12] in his paper discusses constructing garbled
circuit for two parties to securely evaluate a function by means of quantum gates
(QG).
454 T. Borade et al.

3 Multiparty Computation Techniques

3.1 Multiplicative Homomorphic Encryption Along

with Secure MPC

Encoding:
1. Padding of the message is done with k1 zeros to be n—k0 bits in length.
2. r is just any k0-bit string that is randomly generated.
3. G expands the k0-bits of r to n—k0 bits.
4. Evaluate X as

X = m000 . . . 0 ⊕ G(r ) (1)

5. H reduces the n—k0 bits of X to k0 bits.

6. Calculate Y as

Y = r ⊕ H (X ) (2)

7. Finally, the result is calculated as X||Y.

Decoding:

1. First, find r as

r = Y ⊕ H (X ) (3)

2. Second, recover m as

m000 . . . 0 = X ⊕ G(r ) (4)

Algorithm for Secure Cloud Computing:

At first, padding of the user’s data takes place by Optimal Asymmetric Encryption
Padding (OAEP) as shown in Fig. 2, combined with hybrid encryption technique
based on RSA algorithm (HE-RSA). Multiple parties wish to work out a func-
tion upon their personal inputs without compromising on correctness and secrecy.
Encrypted data is homomorphically encrypted. This double encrypted data is not
decrypted in cloud. The algorithm [13] incorporates the MPC along with HE that fol-
lows evaluations to be done on encrypted data without decrypting it. After combining
homomorphic encryption and multiparty computation (HE + MPC), the confiden-
tiality and integrity of the data are preserved without any compromise. The latency
is lesser than HE but more than MPC. The algorithm thereby gives medium amount
of latency based on the HE and MPC.
Analyzing Different Multiparty Computation Techniques 455

Fig. 2 Optical asymmetric encryption padding

3.2 Additive Homomorphic Encryption with Paillier’s

Algorithm

Key Generation:
1. Let p and q be any two large primes.
2. Compute the modulus as

n = p∗q (5)

3. Calculate the Carmichael function

λ = LCM( p − 1, q − 1) (6)

and

g =1+n (7)

4. Calculate

L 1 = L(g λ mod n 2 ) (8)

456 T. Borade et al.

5. Calculate
n
μ = 1 + K. (9)
L1

Thus, public key is (n, g)

private key is (λ, μ)

Encryption:

For any plaintext m, choose a random r ∈ Zn∗2 , calculate

c = g m ∗ r n mod n 2 (10)

Decryption:

1. For ciphertext c <= n 2 , calculate

L 2 = L(cλ mod n 2 ) (11)

2. Finally, calculate the plaintext from ciphertext as:

 
L2
m= mod n (12)
L1
or

m = L 2 ∗ μ mod n (13)

Actual Implementation:

Suppose a number of hospitals wish to jointly evaluate total number of people suf-
fering from a certain disease without revealing any secretive or private information
of the concerned patients. Here, security is the key factor that should definitely be
taken into consideration. In fact, information in medical field is mostly secretive kind
of data that should be kept confidential. To accomplish this, there is a need of secure
framework that will collaborate between various hospitals. The solution to this forms
the secure MPC technique.
With that in mind, after successful encryption each hospital sends their secretive
data for evaluations. Then the cloud provider does the computation work on encrypted
data and forwards the final result back to the hospitals. After this, every hospital then
decrypts this newly received encrypted data using their secret key to yield the final
result. This way the parties are aware only of the final output but not aware of the
individual inputs thus confidentiality prevails. So, this method [14] permits a number
of hospitals to together evaluate a function of their secretive inputs without sharing
patients’ crucial info to third parties.
Analyzing Different Multiparty Computation Techniques 457

In this scheme, hospital A and hospital B encrypt patients’ secretive data m1 and
m2 to yield the ciphertexts C1 and C2 using public key pk. So,

C1 = E(m1, pk)) (14)

C1 = g m 1 ∗ r 1n mod n 2 (15)

C2 = E(m2, pk) (16)

C2 = g m 2 ∗ r 2n mod n 2 (17)

Here, cloud provider evaluates the product of the encrypted data just like RSA.

C1 . C2 = E(m1, pk) . E(m2, pk) (18)

C1 . C2 = (g m1 ∗ r 1n ) . (g m2 ∗ r 2n ) mod n 2 (19)

C1 . C2 = g m1+m2 ∗ (r 1 . r 2)n mod n 2 (20)

C1 . C2 = E(m1 + m2, pk) (21)

Eventually, every medical asylum uses its secretive key to get last result, which
equals the sum of m1 and m2.

D(E(m1, pk) . E(m2, pk)(mod n 2 )) = m1 + m2(mod n) (22)

Together a number of hospitals perform addition operation on encrypted data. To

sum it up as a whole, the solution securely evaluates addition operations. Hence, it
is a effective method to assist collaborative systems in healthcare systems.

3.3 Garbled Circuits Using Quantum Gates

In Yao’s GBC, the evaluating function is depicted as a Boolean circuit made up of

binary gates. Encryption is then done of input and output wires such that the person
or machine evaluating the GBC does not find any information regarding inputs. With
GBC consisting of simple Boolean circuits, researchers found out that overall OT
payload shoots up proportionally with the number of gates and become tremendous
for complex GBCs [15]. So researchers were keenly fascinated in making a GBC
that would require a meager number of non-EX-OR gates. An idea of increasing the
458 T. Borade et al.

Table 1 Comparison of gates required

Version Comparator(4-bit) Comparator(8-bit) Adder(32-bit)
Garbled circuit 10 26 127
Quantum garbled circuit 8 16 64

efficiency is incorporating quantum gates in place of Boolean ones thereby forming

Q-GBC. The main motive behind is that the Q-GBC technique will increment the
number of XOR gates without any communication delay and computation latency
and lessen the number of non-XOR gates, thereby demanding meager interaction.

Quantum garbled circuit for comparison function:

With classical Boolean circuits, it is found that for 4-bit comparator circuit, the
number of non-XOR gates is 10, whereas in case of 8-bit comparator circuit, the
number of non-XOR gates comes out to just 26. According to RevKit result, each TR
gate has got 1 CNOT and 1 Toffoli gate [12]. So we conclude that for implementing
a 4-bit comparator circuit, we need just 8 non-XOR gates while for 8-bit millionaire
problem with QC we need just 16 TR and 16 non-XOR gates.

Quantum garbled circuit for 32-bit adder:

With classical Boolean circuit of adder, a 32-bit adder GC has 127 AND, 187 INV,
and 61 XOR gates. Peres FA gates are utilized as quantum full adder which are
implemented by two quantum Peres gate. Additionally, a 32-bit adder forms an array
of 32 PFAG. Hence, it can be said that a 32-bit adder can be realized as a GC that
needs only 64 non-XOR gates (Table 1).

4 Comparative Analysis of Techniques

See Table 2

5 Conclusion

Nowadays secure multiparty computation is the trendy topic in the field of network
security and cryptography. With secure MPC, it is no longer needed to trust the third
parties. Thus by eliminating the third party, multiple parties can come together to
evaluate some common function f over a cloud environment. The idea of having
a secure MPC is quite old but the research still continues, and there are still many
things yet to be explored in this regard. Thus, we have studied different techniques of
multiparty computation. We compared the different algorithms and techniques such
as multiplicative HE with OAEP, additive HE with Paillier’s scheme, and garbled
Analyzing Different Multiparty Computation Techniques 459

Table 2 Comparison of MPC techniques

Parameters Multiplicative HE + Additive HE + Garbled circuits
MPC Paillier’s algorithm
Confidentiality Confidentiality is Confidentiality is Confidentiality is
maintained maintained maintained
Integrity Integrity is Integrity is Integrity is
maintained maintained maintained
Overhead Overhead is Overhead is less Overhead is less
moderate
Algorithm or Multiplicative Additive Using quantum gates
technique used homomorphic homomorphic in garbled circuits
encryption and encryption and
OAEP technique Paillier’s scheme
Advantages Overhead lies in Productive technique Lesser non-EXOR
between only HE and to do computation gates employed
only MPC even in distributed which means lesser
circumstances overhead
Limitations Not fully Not fully Difficult to
homomorphic, only homomorphic, only implement with the
multiplication addition operations existing technology
operations can be can be performed
performed

circuits using quantum gates. Comparative analysis of the algorithms states limitation
and advantage of individual algorithm. Secure MPC is undisputedly the future and
would definitely be prominent among all kinds of organizations.

References

1. Hong MQ, Zhao WB, Wang PY (2016) Homomorphic encryption scheme based on elliptic
curve cryptography for privacy protection of cloud computing. In: 2016 IEEE 2nd international
conference on big data security on cloud, IEEE international conference on high performance
and smart computing, IEEE international conference on intelligent data and security
2. Yao A (1986) How to generate and exchange secrets. In: 27th annual symposium on, foundations
of computer science 1986, IEEE, pp 162–167
3. Rani PS, Vigneswari D (2016) Security and privacy in big data analytics. Int J Intell Electron
Syst 10(2) July
4. Zhenlin T, Wei Z (2015) A Predicate encryption scheme supporting multiparty cloud com-
putation. In: 2015 international conference on intelligent networking and collaborative
systems
5. Yao Y, Wei J, Liu J, Zhang R (2016) Efficiently secure multiparty computation based on
homomorphic encryption. In: proceedings of CCIS 2016
6. Dhakar RS, Gupta AK (2012) Modified RSA encryption algorithm (MREA). In: 2012 second
international conference on advanced computing & communication technologies
7. Xiang G, Cui Z (2012) The algebra homomorphic encryption scheme based on Fermat’s
Little Theorem. In: 2012 international conference on communication systems and network
technologies
460 T. Borade et al.

8. Youn TY, Jho NS, Chang KY Practical additive homomorphic encryption for statistical analysis
over encrypted data
9. Naumann F (2016) Garbled circuits. In: Seminar Innovative Internet-Technologien und
Mobilkommunikation SS2016
10. Snyder P Yao’s Garbled circuits: recent directions and implementations
11. Cimato S, Ciriani V, Damiani E, Ehsanpour M A multiple valued logic approach for the
synthesis of garbled circuits
12. Ehsanpour M (2017) Toward design of garbled circuits using Quantum gates. In: 2017 IEEE
41st annual computer software and applications conference
13. Das D (2018) Secure cloud computing algorithm using homomorphic encryption and multiparty
computation. International Symposium on Networks and Security Systems, Delhi, India
14. Marwan M, Kartit A, Ouahmane H Applying secure multi-party computation to improve
collaboration in healthcare cloud
15. Ehsanpour M, Cimato S, Ciriani V, Damiani E (2017) Exploiting Quantum gates in secure
computation. In: 2017 Euromicro conference on digital system design
Power and Delay Comparison of 7:3
Compressor Designs Based on Different
Architectures of XOR Gate

Rekib Uddin Ahmed and Prabir Saha

Abstract This paper presents the power and delay comparison of 7:3 compressor
circuit designed using three different architectures of XOR gate which are based
upon mirror circuit, 4-transistor, (4-T) and transmission gate (TG). The compressors
have been implemented in transistor level at 180 nm technology and the functionality
is verified in Cadence-spectre. Among the 7:3 compressors, the design utilizing the
TG-based XOR gate is exhibiting least power consumption and the least delay is
exhibited by the design which is based upon mirror circuit-based XOR gate.

Keywords Compressor · Delay · Mirror circuit · Power · Transmission gate

1 Introduction

In microprocessors and digital signal processors, the multipliers are used to perform
dedicated operations like convolution, correlation, and filtering. The multiplication
process involves formation of partial products followed by addition of partial prod-
ucts in order to generate the final binary result. Among the process involved in
multiplication, the addition of partial products is the main cause of the delay, area,
and power consumption in the multipliers. Conventionally, addition of partial prod-
ucts was performed by huge numbers of adders that are capable to add two or three
bits at a time [1]. One of the methods to minimize the number of adders is the use of
compressors. Compressors are the basic circuits which count the number of “ones”
present in the input vector. Different types of compressors like 3:2, 4:2, and 5:2 have
been reported in literature [2–5] which are suitable for 8 × 8 bit multiplier. Large-
sized compressors such as 7:3 and 15:4 compressors are also proposed in order to
design 16 × 16 and 32 × 32 bit multiplier [1, 6, 7]. One important component used
in the compressors is XOR gate, which is also the building block in various circuits

R. U. Ahmed (B) · P. Saha

National Institute of Technology Meghalaya, Shillong, India
e-mail: rekib@[Link]
P. Saha
e-mail: sahaprabir1@[Link]
© Springer Nature Singapore Pte Ltd. 2020 461
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
462 R. U. Ahmed and P. Saha

like adder, parity checker, etc. While implementing in XOR gate in CMOS, the main
intention is to reduce its transistor count aiming at reducing the overall delay and area
of the constituting circuit [8]. Various CMOS architectures of XOR gate have been
proposed [9] in order to obtain faster and more compact circuit structures. In this
paper, the 7:3 compressor [10] is designed by utilizing three different architectures of
XOR gate: (1) mirror circuit-based (2) 4-transistor (4-T)-based, and (3) transmission
gate (TG)-based. The compressor circuits are implemented at gate level using the
generic process design kit (gpdk) −180 nm technology node. The functionality of the
compressors is verified through simulation in Cadence-spectre and the comparison
between the designs as the function of power and delay is presented.

2 The 7:3 Compressor

A 7:3 compressor takes seven-bit input vector X = (x6 , x5 , x4 , x3 , x2 , x1 , x0 ) and

generates three-bit output vector q = (q2 , q1 , q0 ). In order to obtain the Boolean
expressions for the q, the X is partitioned into two sub-vectors X a = (x2 , x1 , x0 ) and
X b = (x6 , x5 , x4 , x3 ). The truth tables for X a and X b are given in Tables 1 and 2.
The Boolean expressions for outputs of X a are given by

qa1 = x2 x1 + x2 x0 + x1 x0 (1)

qa0 = (x2 ⊕ x1 ⊕ x0 ) (2)

Similarly, the Boolean expression for outputs of X b is as follows (The simplifi-

cation of qb1 is given in Appendix)

qb2 = x6 x5 x4 x3 (3)

Table 1 Truth table for sub-vector X a

x2 x1 x0 qa1 qa0
0 0 0 0 0
0 0 1 0 1
0 1 0 0 1
0 1 1 1 0
1 0 0 0 1
1 0 1 1 0
1 1 0 1 0
1 1 1 1 1
Power and Delay Comparison of 7:3 Compressor Designs … 463

Table 2 Truth table for sub-vector X a

x2 x2 x2 x2 qa1 qa1 qa1
0 0 0 0 0 0 0
0 0 0 1 0 0 1
0 0 1 0 0 0 1
0 0 1 1 0 1 0
0 1 0 0 0 0 1
0 1 0 1 0 1 0
0 1 1 0 0 1 0
0 1 1 1 0 1 1
1 0 0 0 0 0 1
1 0 0 1 0 1 0
1 0 1 0 0 1 0
1 0 1 1 0 1 1
1 1 0 0 0 1 0
1 1 0 1 0 1 1
1 1 1 0 0 1 1
1 1 1 1 1 0 0

qb1 = [x6 x5 + x4 x3 + (x6 + x5 )(x4 + x3 )]qb2 (4)

qb0 = (x6 ⊕ x5 ⊕ x4 ⊕ x3 )

The outputs of two sub-vectors are added through a carry-lookahead (CLA) adder
in order to obtain the final expression for the outputs of the 7:3 compressor.

q2 = qb2 ⊕ [qa1 qb1 + (qa1 ⊕ qb1 )(qa0 qb0 )] (5)

q1 = qa1 ⊕ qb1 ⊕ (qa0 qb0 ) (6)

q0 = (qa0 ⊕ qb0 ) (7)

The gate-level logic diagram of the outputs q2 , q1 , and q0 is shown in Fig. 1. The
XOR operations used in the 7:3 compressor are implemented using three different
architectures of XOR gate which are discussed below.
464 R. U. Ahmed and P. Saha

Logic block for Logic block for Logic block for CLA

Fig. 1 Logic diagram for the outputs of the 7:3 compressor circuit

2.1 Mirror Circuits-Based XOR Gate (Design 1)

Figure 2a shows the CMOS XOR gate which is based on the concept of mirror circuit.
With the two inputs A and B, the possible combinations are: AB, AB, AB, A B. The
AB and AB provide connections from the power supply to Y thereby producing the
output high (Vdd ), while AB and A B connect the Y to the ground giving output low
(0 V).

2.2 4-Transistor XOR Gate (Design 2)

The 4-T XOR gate consists of two CMOS inverters cascaded together as shown in
Fig. 3 with input A applied to the first stage inverter (inverter 1), and connected to
Power and Delay Comparison of 7:3 Compressor Designs … 465

Fig. 2 Different architectures of XOR gate: a mirror circuit-based, b 4-transistor-based, c TG-

based, and d TG

(a) (b)

Fig. 3 Bar chart showing comparison between the compressor designs: a power analysis, b delay
analysis
466 R. U. Ahmed and P. Saha

the top of second stage inverter (inverter 2). The A from inverter 1 is applied to the
bottom of inverter 2. The other input B is used to switch the MOSFET pairs M 3 and
M 4 . With B = 0, M 3 conducts and the logic level at A is transferred to the Y thus
producing the term AB. If B = 1, M 4 conducts thereby connecting A to Y , giving
AB. Combining these two gives the function: AB + AB = A ⊕ B.

2.3 Transmission Gate-Based XOR Gate (Design 3)

The XOR gate can be implemented by using one of the input variables to control the
TG. Figure 2c shows the XOR gate implementation using two TGs (T 1 and T 2 ) where
the input B is used to control the TGs with A and A as inputs. The transmission gate
(TG), sometimes referred as solid-state switch which selectively blocks or passes a
logic level from the input to the output. The TG as shown in Fig. 2d is comprised
of p-type (M 5 ) and n-type (M 6 ) MOSFETs. The gates of M 5 and M 6 are biased in
such a manner so that both MOSFETs are either ON or OFF simultaneously. When
the voltage at node C is at logic 1, the complementary logic 0 is applied to node
C, allowing both M 5 and M 6 to conduct and pass the logic level at D to Z . When
voltage on node C is at logic 1 and logic 0 is applied to node C, the both transistors
OFF thereby forcing a high impedance condition across D and Z .

3 Results and Discussion

The 7:3 compressor designs are simulated in Cadence-spectre and compiled with
gpdk 180 nm technology considering device parameters: channel length L = 180nm,
channel width W = 1.08 µm, source and drain (S/D) diffusion area = 0.972 µm2
and S/D diffusion periphery = 3.96 µm. The compressor designs are simulated under
supply voltage of 1.8 V with input pulses of amplitude 1.8 V having rise and fall
time = 10 ps of time period 20 ns.
The power dissipation and delay of the compressor designs are recorded at rising
edge (RET) and falling edge transition (FET) of input pulses. The power and delay
observed for some set of selected set of input combinations are given in Tables 3 and
4.
Figure 3 shows the pictorial representation of average power dissipation and delay
exhibited by three compressor designs. The least power dissipation is exhibited by
the Design 3 which utilized the TG-based XOR gate. As it has been stated earlier that
TGs are inherently low power consuming and they are good for designing XOR gate
[11, 12], so the Design 3 is exhibiting a minimum power dissipation of 13.84 µW
at RET and 14.09 µW at FET. On the other hand, the rise and fall time of mirror
circuit-based XOR gate are shorter [9], so the Design 1 is exhibiting a minimum
delay of 0.4701 ns at RET and 0.3698 ns at FET. While the 4-T XOR gate is useful in
reducing the transistor count but from power dissipation and delay point of view, it
Power and Delay Comparison of 7:3 Compressor Designs … 467

Table 3 Power dissipation (in µW) at different combinations of X

X Design 1 Design 2 Design 3
RET FET RET FET RET FET
1111111 16.52 17.63 28.38 103.0 12.57 15.29
1111110 13.49 14.65 26.62 45.24 11.69 14.12
1011111 17.73 17.86 85.99 158.5 16.49 16.4
1010111 14.06 13.22 111.2 129.3 16.06 14.31
1010101 11.61 12.44 100.1 159.4 14.57 13.36
1010000 7.553 7.981 21.15 98.13 8.262 7.22
1111101 14.92 16.8 61.03 133.2 12.4 14.56
1111011 15.18 16.75 53.79 133.4 11.53 13.92
1110111 18.02 17.44 82.43 158.7 17.04 16.44
1101111 18.57 17.59 81.43 159.3 16.01 14.73
0111111 18.26 17.97 84.19 159.0 15.63 14.66
Average 15.08 15.48 66.93 130.65 13.84 14.09

Table 4 Delay observed (in ns) at different combinations of X

X Design 1 Design 2 Design 3
RET FET RET FET RET FET
1111111 0.4018 0.479 0.4081 1.6563 0.3124 0.5381
1111110 0.4747 0.479 0.5041 1.1679 0.4384 0.5324
1011111 0.5268 0.2895 0.6976 2.5549 0.4678 0.4846
1010111 0.3369 0.2668 1.5546 1.6573 0.5419 0.4849
1010101 0.2956 0.2631 1.008 2.4842 0.5492 0.4892
1010000 0.5223 0.4681 0.5437 1.1789 0.4775 0.5191
1111101 0.465 0.4802 1.0076 2.4968 0.4117 0.5373
1111011 0.4542 0.4818 0.4903 2.5196 0.3659 0.5427
1110111 0.5601 0.285 0.4062 2.5176 0.4687 0.4846
1101111 0.5837 0.2858 0.7026 2.537 0.4613 0.485
0111111 0.5502 0.2895 0.7671 2.5844 0.4604 0.4851
Average 0.4701 0.3698 0.6985 2.123 0.4504 0.5075

does not seem to be advantageous. So, Design 2 is showing maximum delay (2.123 ns
at FET) and power consumption (130.65 µW at FET) as compared to Design 1 and
Design 3.
468 R. U. Ahmed and P. Saha

4 Conclusion

In this paper, 7:3 compressor designs have been presented by utilizing the mirror
circuit, 4-T, and TG-based XOR gates. The functionality of the designs is verified and
comparative study with respect to power and delay have been presented. Maximum
average power dissipation and delay is observed in case of 4-T XOR gate-based
compressor (Design 2). Least average power dissipation and delay is observed in
the compressor utilizing the TG-based XOR gate (Design 3). The performance of
Design 3 is found better among the discussed designs of 7:3 compressor.

Acknowledgements This work is supported by Ministry of Electronics and Information Technol-

ogy, Government of India, under Visvesvaraya Ph.D. Scheme.

Appendix

The Karnaugh map for the output qb1 is shown below:

x4 x3
x6 x5 00 01 11 10

00 1

01 1 1 1

11 1 1

10 1 1 1

Considering the circled minterms:

qb1 = [x4 x3 + x5 x3 + x5 x4 + x5 x4 x3 + x6 x3 + x6 x4 + x6 x4 x3 + x6 x5 + x6 x5 x3 + x6 x5 x4 ]qb2

= [x4 x3 + x5 x3 + x5 x4 (1 + x3 ) + x6 x3 + x6 x4 (1 + x3 ) + x6 x5 (1 + x3 + x4 )]qb2
= [x4 x3 + x5 x3 + x5 x4 + x6 x3 + x6 x4 + x6 x5 ]qb2
= [x6 x5 + x4 x3 + x5 (x3 + x4 ) + x6 (x3 + x4 )]qb2
= [x6 x5 + x4 x3 + x5 (x3 + x4 ) + x6 (x3 + x4 )]qb2 (8)

qb1 = [x6 x5 + x4 x3 + (x3 + x4 )(x5 + x6 )]qb2 (9)

Power and Delay Comparison of 7:3 Compressor Designs … 469

References

1. Dandapat A, Goshal S, Sarkar P, Mukhopadhyay D (2010) A 1.2 ns 16 × 16 bit binary multiplier

using high speed compressors. Int J Elect Electron Eng 4: 485–490
2. Menon R, Radhakrishnan D (2006) High performance 5:2 compressor architecture. In: IEE
Proc. Circuits, Devices, Syst. vol 153, pp 447–452
3. Veeramachaneni S, Krishna KM, Avinash L, Puppala SR, Srinivas MB (2007) Novel architec-
ture for high-speed and low power 3-2, 4-2, and 5-2 compressors. In: IEEE Int. Conf. VLSI
Design, pp 324–329
4. Siliveru A, Bharati M (2013) Design of compressor based multiplier using degenerate pass
transistor logic. Int J Eng Trends Technol 4:896–900
5. Lakshmi GS, Fatima K, Madhavi BK (2016) Compressor based 8 × 8 bit vedic multiplier using
reversible logic. In: Int. Conf. Devices, Circuits, Syst. pp 174–178 (2016)
6. Ravi N, Prasad TJ, Umamahesh M, Rao TS (2010) Performance evaluation of high speed
compressors for high speed multipliers using 90 nm technology. In: Recent Adv. Space Technol.
Serv. Climate Change. pp 189–193 (2010)
7. Marimuthu R, Rezinold YE, Mallick PS (2017) Design and analysis of multiplier using
appropriate 15-4 compressor. IEEE Access 5:1027–1036
8. Kumar S, Kumar M (2014) 4-2 Compressor design with new XOR-XNOR module. In: Int.
Conf. Adv. Comput. Comm. Technol. pp 106–111 (2014)
9. Uyemura JP (2002) CMOS logic circuit design. Kluwer Academic Publishers
10. Multiple Operand Addition. [Link]
11. Navi K, Maeen M, Foroutan V, Timarchi S, Kavehei O (2009) A novel low-power full-adder
cell for low voltage. Integration, VLSI J 42:457–467
12. Shams AM, Darwish TK, Bayoumi MA (2002) Performance analysis of low-power 1-bit CMOS
full adder cells. IEEE Trans VLSI Syst 10:20–29
Survey of Big Data Warehousing
Techniques

Jaspreet Kaur, Rajashree Shedge and Bharti Joshi

Abstract There is a growing need in the industry toward the development of new and
sophisticated tools for storing the exponentially growing volume, velocity and variety
of data, which is collectively referred to as big data. There has been a paradigm shift
from traditional data warehousing techniques to inclusion of NoSQL technology in
order to fulfill the requirements of big data. While Hadoop has powerful features,
which is not a replacement to Data Warehouse, rather it is a complement. Data
Warehouse is already good at processing structured data so when used in conjunction
with Hadoop, it becomes a winning combination. Hadoop can be considered as one
of the back ends of Data Warehouse for handling unstructured data. Hence there is
research on enhancing existing Data Warehouse with new features that have been
successful at handling big data, and most popular one among them is MapReduce.
We discuss the different tools and techniques used for improving Data Warehouse
by adding these features and discuss the limitations associated with them.

Keywords Data warehousing · Hadoop · Unstructured · MapReduce

1 Introduction

Data storage is the recording (storing) of the data in a storage medium. The storage
medium could be as diverse as DNA, RNA, phonographic recording, magnetic tape,
and optical disks. Organizations typically store their data in a Data Warehouse which
is a large store of data collected from heterogeneous sources within the company
that allows the management to take informed decisions. With the advent of social
platforms and various online portals for consumers to put their views, it becomes

J. Kaur (B) · R. Shedge · B. Joshi

Ramrao Adik Institute of Technology, Navi Mumbai, India
e-mail: jaspreetseera@[Link]
R. Shedge
e-mail: [Link]@[Link]
B. Joshi
e-mail: [Link]@[Link]
© Springer Nature Singapore Pte Ltd. 2020 471
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
472 J. Kaur et al.

crucial for organizations to process such data, which is mostly present in a textual
format. Moreover, the volume of data has increased multiple folds and ways to
process such huge information is challenging.
The mining of big data produces many interesting results. However, researchers
are facing lot of challenges for extracting useful information from such huge chunks
of data. Problems arise in the areas of data capture, storage, searching, sharing,
analysis, management, and visualization [1]. Moreover, the rate at which the data
is growing is exponential and this will surpass the current storage capacity in near
future.
Traditional Data Warehouses were built on Online Transactional processing sys-
tems by using suitable tools and architecture which is 15–20 years old. Previously,
business analysts were satisfied with analyzing the small chunks of data and pro-
vide daily, fortnightly or monthly reports to the top management. There was no
proper foresight at that point of time on how big data would completely encapsulate
and change the data and storage requirements in future. However, we cannot ignore
the time and investment spent on building Enterprise Data Warehouse (EDW) and
must benefit from new technologies, products and approaches to modernizing the
currently inflexible Data Warehouses. Moving toward big data warehousing would
require undergoing the following changes [2, 3].
• Offloading the ETL process to a Hadoop cluster that uses inexpensive, industry-
standard hardware and processes data quickly.
• Using Parallel processing.
• Capitalizing on in-database analytics.
• Adding unstructured data metrics to existing Data Warehouse.
• Use Data Federation to extend the Data Warehouse.
• Creating Data repository.
• Adopting Lambda architecture.
Section 2 describes the existing research work in the area of big data warehousing
techniques. Different techniques are discussed in Sect. 3. An endeavor to compare
different techniques briefly in Sect. 4. Section 5 concludes the survey of the different
techniques.

2 Related Work

Authors Ramkrushna et al. [4] explained the advantages of Apache Spark over
Hadoop MapReduce and analyzed real-time data using time series analysis. Chen
et al. [5] discussed Octopus which is a hybrid big data integration engine whose focus
is to minimize the data movement by pushing queries to the respective back ends and
getting it processed there rather than fetching the table in memory like Spark does.
The authors also showed how Octopus [5] performed better than Spark in terms of
query running time. Zhou et al. [6] explained that the requirements of real-time pro-
cessing and large capacity of big data can be fulfilled by hybrid storage. The authors
Survey of Big Data Warehousing Techniques 473

propose a preference model to weight the storage performance imbalance when data
is stored on various devices. Based on this calculation, data is distributed on devices
by matching the device performance with data access characteristics. Bhat [7] dis-
cusses that there is an ever-increasing gap between the volume of digital data being
created and the available storage capacity of current devices. The author discusses
the working of Optical, DNA and Holographic storage technologies and compares
them on the basis of storage density, throughput and lifetime and implications of
adoption.
Pticek and Vrdoljak [8] discussed that lack of schema in NoSQL databases makes
them less comprehensible and difficult for integration and analysis. Hence the authors
suggested that the use of semantics would make the contents more understandable
and easy for integration. Dehdouh et al. [9] developed an aggregation operator called
Columnar NoSQL Cube which can build data cubes over Data Warehouses stored in
column-oriented NoSQL database management system using HBase engine. Hav-
ing the CN-Cube operator solves the drawback of absence of OLAP operators for
column-oriented database. Pticek and Vrdoljak [10] discussed MapReduce research
in warehousing of big data wherein they have highlighted that for big data ware-
housing solution, integrate one or more MR-based technologies like Hadoop and
Hive and combine it either NoSQL or traditional Data Warehouse. Herodotus et al.
[11] explain about Starfish which is a self-tuning system for big data analytics. It is
built on Hadoop while adapting to user needs and system workloads to provide good
performance automatically. Abello et al. [12] discuss the possibility of having data
in a cloud by using BigTable to store the hierarchical data and using Map Reduce to
deploy cubes in ad hoc Data Marts.

3 Big Data Warehousing Techniques

The key requirements of big data storage are its ability to handle large amounts of data,
scalability to handle growth and real-time data processing for analysis. The typical
approach has been to provide an architecture that spans several storage products in
order to provide the required performance and capacity at a reasonable price. A new
generation of storage tools has been developed specialized for big data, and efforts
have been made to augment these systems with traditional Data Warehouses. In this
section, we do an exhaustive study about the recent tools developed for storage of
big data and discuss their performance, capabilities and limitations. It investigates
the challenge of storing data in a secure and privacy-preserving way.

3.1 Apache Spark

Apache Spark is a cluster computing technology for large scale data processing. It
does not use MapReduce as an execution engine, rather it uses its own distributed
474 J. Kaur et al.

runtime for executing work on cluster. [4] It is built on top of Hadoop MapReduce
and extends the MapReduce framework to incorporate many other computations.
The main drawback of Hadoop is that it can be used for batch processing and not
real-time computation. But with the growing needs of the industries, having real-
time processing is becoming a necessity. Spark is 100x faster than Hadoop for large
scale data processing. It provides easy interface for users and is polyglot, i.e., can be
programmed in Scala, Python, R and Java. In addition, Spark is flexible in the sense
that it can be deployed using Mesos, Hadoop via Yarn or by using its own cluster
manager.
Spark creates a Spark Context object, which gives information about accessing
the cluster. Using the attributes defined in the Spark Context, it connects to a Cluster
Manager whose task is to allocate resources across various applications. The cluster
manager is in turn connected to the worker nodes/data nodes where the execution
takes place. These worker nodes also have a cache memory which makes Spark
fault-tolerant. The executor is space in RAM where all the blocks reside for execution.
Spark also supports streaming data, SQL queries, and complex analytics such
as graph algorithms and machine learning using components like Spark SQL, Spark
Streaming, GraphX, Spark Core Engine and, MLlib. Resillient Distributed databases
(RDD) which is an immutable distributed collection of objects, are the heart of every
Spark program. Spark provides two categories of operations on RDDs: transforma-
tions and actions. A transformation generates a new RDD from an existing one. An
action triggers a computation on an RDD and produces the result into an external
storage.

3.2 Octopus

Nowadays, organizations maintain huge amount of data in multiple backend systems

which include traditional warehouses and more recently, big data systems. Integrat-
ing, querying and analyzing such disparate data become a difficult task. Octopus is a
hybrid data processing engine to integrate backend systems. It divides the task into
subqueries and pushes them into the respective backend system where it is processed.
It comprises of two main components, namely parser and optimizer. Parser generates
a logical operator tree based on user programs. Based on the logical operator tree
generated by the parser, the optimizer generates optimal query plans for execution
in the respective backend [5].
We consider an input query q and S backend systems where S i (1 ≤ i ≤ S), D
datasets Dj (1 ≤ j ≤ D) and Q subqueries generated by Octopus where Qk (1 ≤ k
≤ Q). Three binary indicators are introduced to understand the relationship among
the sets. Given a query task q to access whole data sets Dj , generation of subqueries
Qk that will be pushed down to backend systems S i with the goal of minimizing the
overall query processing time., we measure the running time in the following criteria.
Survey of Big Data Warehousing Techniques 475

Tmi,j of system S i to move data Dj needed by Qk

Tpi,k of system S i to process query Qk .

3.2.1 Dependencies Among Queries

Tpi,k depends solely on the processing power of system S i . Tmi,j of system depends
on whether the data set Dj required by subquery Qk is present locally or needs to
be fetched from other systems. In the previous case, time Tmi,j is dominated by
the time required by system Si to load Dj. In the second case, the time Tmi,j is
dominated by the time required to fetch dataset Dj from remote system. This task
of assigning subqueries to backend systems is a job scheduling NP-hard problem;
hence, a heuristic solution has been proposed by leveraging the data locality principle
which minimizes the data movement.

3.3 Preference Aware Hadoop Distributed File System

Hadoop Distributed File System (HDFS) is the primary data storage used by applica-
tions of Hadoop. Due to its distributed storage, scalability, reliability, fault tolerance,
high availability and replication, it is a suitable storage medium for big data. However,
HDFS also faces certain limitations, one of them being that it distributes data evenly
across all storage devices under the assumption that all storage devices have the
same I/O performance. Different devices have different I/O performance, asymmet-
ric read/write performance as well as asymmentric read/write access characteristics
of data. Hence, a preference model was designed (PAHDFS) to calculate the weight of
the storage performance and distribute data based on matching of performance with
data access characteristics [6]. Data is divided into blocks. The following equations
explain the calculation of preference degree.

Si
TRm (i) = + TSm (1)
PRm

Read Time of Data Block Bi when it is distributed on Dm is equal to the sum of

read transmission time and seek time.
Similarly, write transmission time of data block Bi when it is distributed on Dm
is equal to the sum of write transmission time and seek time.

Si
TWm (i) = + TSm (2)
PWm

Hence overall access time for Data Block Bi is calculated as

Cm(i) = TRm ∗ FRi + TWm ∗ FWi (3)

476 J. Kaur et al.

Hence, preference degree of block Bi to device Dm is equivalent to the difference

in the access time of Bi on device Dn and device Dm

Pmn (i) = Cn(i) −Cm(i) (4)

The greater the value of Pm, n (i), the more block Bi prefers to reside in device
Dm. Hence deriving the final equation
    
1 1 1 1
Pmn (i) = Si ∗ − ∗ FRi + − ∗ FWi + (TSn − TSm ) ∗ (FRi + Fwi )
PRn PRm PWn PWm
(5)

The blocks are sorted in descending order of access frequency. The algorithm
traverses the blocks from high frequency to low frequency and determines whether
migration is necessary. Higher preference degree of a block in a particular device
implies that the particular block will display higher performance if it resides on
that device. The algorithm traverses in the Check Device set in descending order of
preference degree to ascertain if migration is required. To overcome the problem of
excessive migrations, the following conditions are imposed
• The benefit of migration should exceed the overhead incurred.
• Migration should not pull block with higher preference out of the target device.

3.4 Optical Data Storage (ODS)

According to a report published by International Data Corporation, the amount of

data is forecasted to reach 44 zettabytes by the year 2020, out of which 13 zettabytes
is supposed to be critical data. However, it is estimated that the current storage
technologies will be able to hold only 15% of 44 zettabytes resulting in a data capacity
gap of 6 ZBs. Optical Data Storage (ODS) [7] which first emerged as CDs and later
progressed to DVDs and Blu-ray disks recorded information in diffraction-limited
region within a layer beneath the surface of the disk. Due to this, the capacity was
limited to few tens of Gigabytes(GBs) The optical spot which records the digital bits
is directly proportional to the wavelength of Laser beam and inversely proportional
to the numerical aperture of the optical head. The diffraction limit places a limit on
the storage capacity to few terabytes per disk which is not enough to overcome the
data capacity gap. Nanophotonic approaches have paved the way for increasing the
storage capacity by breaking or circumventing the diffraction barrier.
Survey of Big Data Warehousing Techniques 477

3.5 DNA Data Storage (DDS)

DNA has the potential to have huge storage capacity because it is incredibly dense,
is volumetric rather than planar and is fairly stable. DNA strand or Oligonucleotide
is a linear sequence of four nucleotides namely Adenine (A), Cytosine (C), Guanine
(G) and Thymine (T ). Two DNA strands can bind to each other and form a double
helix. A in one strand can combine with T in another strand and likewise for C and
G. Probability of a nucleotide to bind to the existing partial strand at each step in
the process is as high as 99% and is referred to as coupling efficiency. There are
four different base pairs namely AT, GC, TA and CG and they are used to encode
four Quaternary digits which are 00,01,10 and 11. For reading data stored in DNA,
nucleotide pairs are decoded into Quaternary digits and are further translated to binary
data. However, the drawback associated with DNA storage is that it is vulnerable
to a wide variety of errors during synthesis and sequencing. The error rate could be
around 1% per nucleotide. Also, nucleotide sequences degrade when stored leading
to data integrity issues [12].
Though DNA, data storage offers huge rewritable storage capacity and sustain-
ability, but still is an expensive proposition because of the cost of sequencing and
synthesis. Hence, we look at another technology known as Holographic Data storage
in the next section.

3.6 Holographic Data Storage (HDS)

Holographic data storage (HDS) overcomes the limitation of ODS by storing data
throughout the volume of the media rather than only on the surface. HDS employs
multiplexing for recording multiple holograms in the same volume of data. Light
from a laser source is split into two beams. The first is a signal beam which is passed
through a Spatial Light Modulator to encode binary data. The second is a reference
beam which is recombined with the modulated signal beam before entering the
recording media. The pattern produced by the overlapping beams is imprinted as a
hologram. The differentiating feature of the hologram is the unique volume address
provided by reference beam. For retrieving the data, the reference beam that was
used for recording the hologram is used. However, signal beam is independent from
the rest of the holograms [7].

3.7 Semantic Web Technologies

NoSQL, which stands for Not Only SQL, is specially designed for large sets of dis-
tributed data. It is basically schema-less which allows data of any form to be stored.
However, the diversity and lack of clear schema of NoSQL database present inte-
gration problems with relational databases. This drawback led to the development
478 J. Kaur et al.

of semantic web technologies that can compensate the lack of schema in NoSQL
databases and boost the design of the Data Warehouse. NoSQL Databases are classi-
fied as document, column, key-value stores and graph. Though they differ in features,
they share the schema-less model and have good fault tolerance and consistency. The
role of semantics in the semantic-based integration scenario has been discussed under
the following heads [8].
• Data Source schema extraction.
• Integration of data source schemas.
• Schema reduction.
• Outlining the multidimensional concepts.
The advantages of having Ontologies are that they can be easily updated, can solve
heterogeneity conflicts, can detect multidimensional concepts, can discover hidden
semantics and explore large unstructured data sets. Hence using ontologies can help
in automating the process to a certain extent.

3.8 Columnar NoSQL Cube

Column-oriented NoSQL systems unfortunately have no OLAP operators. To over-

come this limitation, CN-CUBE was developed which is an aggregation operator
specially designed for column-oriented NoSQL databases. After identifying the
attributes which would form the basis of dimension and facts, a query is formed
as a view is created as a result. CN-CUBE uses this view hence reducing disk access.
To form all the dimension combinations required to build the cube, it uses value posi-
tions and hash tables and extracts data that satisfies the requirements of the query.
Because of their suitability for big data, HBase DBMS and Hadoop platform were
chosen. Column-oriented databases store data by column rather than rows. Such an
approach resulted in faster aggregation, compression and data retrieval and was suit-
able for big data where the volume of data is very high. The data, which is stored
in key-value format, had an additional attribute namely timestamp. This allows us
to distinguish the recent data. Hence, the combination of key, column name and
time-stamp forms the coordinates of the value. The time-stamp ensures that the most
recent version of data is to be taken while retrieving the results of the query.

4 Comparative Analysis of Techniques

Based on the detailed analysis covered in the previous section, we do a comparative

analysis of the various techniques based on various performance parameters. This
helps us in quickly understanding where we stand presently with respect to these
techniques and how they need to mature further (Table 1).
Survey of Big Data Warehousing Techniques 479

Table 1 Comparative analysis of techniques

Technique Parameter Advantages Limitation
Apache spark 100X times faster than • User-friendly APIs • No support for real-time
[4] MapReduce • Capable of near real-time processing
processing • Expensive
• Fault tolerance by the use • Lacks a file
of RDDs management system
Octopus [5] Octopus outperforms spark • Fully integrate the power Limited to one module of
version 1.4.0 by having a of backend systems spark and cannot fully
5.25 times faster running • Optimizes the amount of replace it unless other
time data movement modules of spark such as
machine learning are
integrated
PAHDFS [6] Read throughput improves • Unburden name node for • Not suitable for small
by 87.3%, write throughput scalability files
improves by 15.6%, mixed • Same interface as HDFS • Hadoop security model
throughput by 70.6% • Devices adequately is complex and disabled
utilized by default
Optical data • Storage density-hundreds • Reliability under extreme • Absence of rewritable
Storage [7] of petabytes atmospheric conditions storage support
• Throughput • Low maintenance cost • Limited to low latency
10GBps(High) • Energy efficient applications
DNA data • Storage density-214 PB • Huge and rewritable • Low throughput. And
storage [7] per gram of DNA storage capacity high cost of sequencing
• Low throughput • Sustainability under and synthesis of DNA
extreme weather strands
conditions • Limited to huge
volumes of data having
high latency
Holographic • Storage density- tens of • Better than ODS and DDS • Changes in recording
data storage terabits percm3 in cost/TB basis technique may make it
[7] • Throughput 400MBps • Lifetime is best among all incompatible
(Low) devices • Intermediate solution
till other technologies
pick up
Semantic Semi automation is possible • Bridge the gap between • Lack of standardization
web since the veracity factor of traditional and NoSQL in ontology
technologies big data requires databases • Considers big data
[8] supervision and hard to • Global ontology to derive warehouse design from
automate multidimensional table scratch
Columnar CN-CUBE has lesser • Reduces input-output Though column-oriented
NoSQL computation time than flows by using hash tables datastores are popularly
cube [9] HIVE CUBE operator • Outperforms the HIVE used for big data, it has no
15 node cluster records the CUBE operator OLAP operators. Limited
best time to only aggregation.
Support for new operators
to be added
480 J. Kaur et al.

5 Conclusion

For an IT industry which was just getting used to handle terabytes of data in their
Data Warehouses, handling unstructured data in petabytes becomes a daunting task.
According to a research, the growth of data in the next five years shall be more than
data collected in the last fifteen years. Data Warehouse alone is incapable of handling
such volume and variety of data. Hence, data storage vendors have been working
toward incorporating big data and analytics tool such as Hadoop into their infrastruc-
ture. The initial hype of big data storage techniques focussed on only Hadoop and
MapReduce but now many other players have entered the market like SAS and SAP
HANA. Storage is a necessity but it is not so easy to address. Cloud storage providers
such as Azure and Google offer hardware, processing and storage for current and
future data growth. Multi-cloud data management solutions are being built which
use a mix of services from cloud service providers as well as private clouds.
We analyzed recent research in big data storage technologies and did a compar-
ative study of these technologies. Some of them are in nascent stage and need to
mature further in order to satisfy the purpose while other tools such as Hadoop and
MapReduce technologies are in developed stage and further improvements are being
suggested through the use of add-ons or separate technologies developed using the
proven technologies. Hence, there are plenty of solutions in the market but not a
single leader which can replace all others.

References

1. Oussus A, Benjelloun FZ et al (2018) Big data technologies: a survey. J King Saud Univ
30(4):431–448
2. Big Data: Five Tactics to Modenize your Data Warehouse. [Link]
emc-perspective/[Link]
3. Extract, Transform and Load Big Data with Apache Hadoop. [Link]
dcd9/[Link]
4. Maheshwar RC, and Haritha D (2016) Survey on high performance analytics of bigdata with
apache spark. In: international conference on advanced communication control and computing
technologies, pp 721–725
5. Chen Y, Xu C, Rao W, Min H and Su G (2015) Octopus: hybrid big data integration engine. In:
IEEE 7th international conference on cloud computing technology and science, pp 462–465
6. Zhou W, Feng D, Tan Z, Zheng Y (2018) Improving big data storage performance in hybrid
environment. J Comput Sci 26:409–418
7. Bhat WA (2018) Bridging data-capacity gap in big data storage. Future Generation Comput
Syst 87:538–548
8. Pticek M, Vrdoljak B (2018) Semantic web technologies and big data warehousing. MIPRO
2018:1214–1219
9. Dehdouh K. et al. (2014) Columnar NoSQL CUBE. In: IEEE international conference on
systems, man and cybernetics, October 2014, pp 3828–3833
10. Pticek M, Vrdoljak B (2017) MapReduce research on warehousing of big data. MIPRO
2017:1361–1365
11. Herodotou H. et al. (2011) Starfish: a self tuning system for big data analytics. In: 5th biennial
conference on innovative data systems research (CIDR2011), 2011, pp 261–272
Survey of Big Data Warehousing Techniques 481

12. Abello A, Ferrarons J, Romero O (2011) Building cubes with MapReduce. In: proceedings of
the ACM 14th international workshop on data warehousing and OLAP(DOLAP11), 2011, pp
17–24
13. Lee S, Jo S, Kim J (2015) MRDataCube: data cube computation using MapReduce. In:
international conference on big data and cloud computing, 2015, pp 95–102
14. Sokolov I, Turkin I (2018) Resource efficient data warehouse optimization. In: 9th IEEE
international conference on dependable systems, services and technologies, 2018, pp 491–495
Detecting Denial-of-Service Attacks
Using sFlow

Shivaraj Hublikar, Vijaya Eligar and Arun Kakhandki

Abstract This paper addresses how to detect denial-of-service attacks using sFlow.
Denial-of-service (DoS) attack is a critical security challenge in software-defined
network (SDN). In DoS attack, the network bandwidth is acquired by disrupting
the services of the server by abruptly increasing the traffic and making the server
unavailable for other users. The most challenging problem of DoS attack is to detect
the attack almost instantly and in a precise manner. This paper presents the detection
of DoS attacks by using sFlow analyzer, a SDNs flow monitoring tool. In the event of
any attack, sFlow collects sample packets from network traffic, analyzes suspicious
behavior and creates handling rules which are then sent to the controller. Implemen-
tation of DoS attack is carried out by emulating a typical network in Mininet and
integrating this with sFlow analyzer. Through the simulated results, the potential
DoS victims and attackers are quickly found.

Keywords Bandwidth detection · DoS attack · SDN · sFlow

1 Introduction

In a traditional data network, devices are structured into data-plane and control-
plane which are local. If there are ten devices in a network, then each device has
its own data-plane and control-plane that are having all the relevant information
regarding forwarding tables. The data-plane comprises of switches, while the control-
plane comprises of controllers of different types. Networking device will get the

S. Hublikar · V. Eligar (B)

KLE Technological University, Hubballi, India
e-mail: vijayaeligar@[Link]
S. Hublikar
e-mail: shivaraj@[Link]
A. Kakhandki
KLSs VDRIT, Haliyal, India
e-mail: bvbarun@[Link]

© Springer Nature Singapore Pte Ltd. 2020 483

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
484 S. Hublikar et al.

Fig. 1 Traditional network and SDN [11]

packets which are decided by forwarding tables. Since the demand of traditional
networks is increasing, working on topology is complex [5]. Even though traditional
networks are global and very popular, they have various drawbacks. Firstly, there is no
scope to extend the network if a new feature or a protocol is to be added. Secondly,
any new command cannot be accepted to improve the functionality of traditional
networks since it is not programmable. Thirdly, cost of the network is effectively high
since each device contains both data-plane and control-plane. Different topologies
of the network are not possible since the traditional network is configured with
set of predefined rules during the manufacture. Furthermore, the physical network
infrastructure cannot be fully utilized since arranging the traditional network with
predefined polices becomes complicated and error prone.
Recent trends such as machine learning, artificial intelligence, cyber security,
internet of things (IoT) and mobile traffic have heavy traffic which cannot be managed
by the network. SDN manages the overall network programmatically. SDN is very
unique from the traditional networks which provide entire central control over the
network by separating the control-plane and the data-plane as shown in Fig. 1.
SDN manages the network by centrally controlling the devices which greatly
utilize and improve the network management. This network has data-plane within the
device for sharing the forwarding data, whereas the control-plane is connected with
separate device called controller which handles the information. SDN is categorized
into three parts, controllers, southbound application program interfaces (APIs) and
northbound (APIs). Controller, which is the programmable central system, controls
the entire network which has the information of all the resources (like switches
and routers) connected to this network. Switches and routers are the information
devices which require southbound APIs as the medium for the controllers to transfer
the data packets. OpenFlow is standard protocol used in southbound APIs. SDN
uses northbound APIs to manage the traffic which is monitored by the network
administrators to communicate with applications shown in Fig. 2.
Detecting Denial-of-Service Attacks Using sFlow 485

Fig. 2 SDN controller

Fig. 3 Open switch

architecture

1.1 OpenFlow

OpenFlow is a standard protocol that provides communication and interface between

control-plane and data-plane in SDN. The data packets are transferred between
devices which have to maintain traffic routing flows that is managed by the Open-
Flow protocol. Each SDN device needs to maintain the set of Flow-tables that is
controlled by OpenFlow switch. The incoming packets are controlled by the switch
which are installed by the control-plane that maintains communication channel and
also contains the Flow-table rules [1]. Figure 3 presents the logical structure of an
OpenFlow switch. When there is a mismatch in the Flow-tables that does not match
with any existing flow rules, the mismatch flows try to trigger forwarding plane to the
controller which reduces the bandwidth and memory. The limited communication
bandwidth between the control and data-planes could be a bottleneck of the whole
network, and lead to security problems. Todays commercial OpenFlow switches
only support cable connection to the controller. The practical connection bandwidth
is tested to be less than 10 Mbps.
Research in SDN and DoS attacks is predominantly focused on detection of the
attack. In July 2001, Internet infrastructure was hit worldwide by DoS attack worms
named Code Red and NIMDA. Network scanning was used by Code Red Worm to
attack the network at a rapid propagation rate to detect and exploit Internet Informa-
tion Server (IIS) automatically. This DoS attack consumed huge bandwidth which
affected lot of network devices. The attack was complex since the attack came from
various sources of IP addresses for which packet analyzing was a big task. In order
to mitigate or limit the damage to network resources, content filtering of some type
were performed.
486 S. Hublikar et al.

In [2], the authors propose a mechanism that avoids the overloading of switch
TCAMs along with controller and control channel bandwidth and a solution called
SLICOTS, which mitigates a type of flooding attack TCPSYN in SDN. It is built on
top of the controller in order to monitor the network traffic related to TCP requests.
This paper is organized as follows. Section 2 discusses DoS attacks on SDN.
Section 3 presents the DoS detection method based on sFlow tool. Section 4 demon-
strates the method to detect DoS attacks. The simulation results and discussion are
presented in Sect. 5 followed by conclusion in Sect. 6.

2 DoS Attack

DoS attack causes serious impact on the computing system. DoS attacks deny services
to valid users by completely consuming the target resources. DoS attacks are usually
initiated by an individual or group of individuals exploiting aspects of the Internet
Protocol to deny other users from legitimate access to systems and information. The
router hosts are disconnected if forwarding packets are stopped by router. The recent
applications which are targets to these attacks are Web servers, mail servers and other
services [6].
In [9], the authors explain SDN-aimed DoS attacks (data-to-control-plane satura-
tion attacks). The attacker first sends the packets which do not match the packets in
the Flow-tables by generating table-miss packets without the order with some or all
fields, which does not match with existing flow rules on the target switch as shown
in Fig. 4. Then, the attacker launches DoS attacks on the SDN network by flooding
with large amount of table-miss packets. These table-miss packets will target massive
packet in messages from the switch to the controller, and consume their communi-
cation bandwidth, CPU computation and memory in both control- and data-planes
[2].

Fig. 4 DoS attack

Detecting Denial-of-Service Attacks Using sFlow 487

DoS attacks can be of various types:

• Destructive: Attacks which destroy the device to prevent the proper operation of
function, such as deleting or changing properties or information and power supply.
• Resource consumption: Attacks which try to reduce the ability of the device to
perform effectively by opening many simultaneous connections to a single device.
• Bandwidth consumption: Attacks which attempt to affect the bandwidth capacity
of the network device.
This paper aims to concentrate on bandwidth attack which is done using sFlow
tool. When any DoS attack is detected, sFlow generates flow rules by analyzing
samples of packets collected from the network traffic to be sent to the controller. In
this work, security services are developed to protect networks against different type
of network attacks for third parties like servers. A DoS attacker attacks the network
by providing enormous flooding traffic in a short time to a server by increasing flow
so that the server gets disconnected.

3 sFlow

sFlow helps to monitor the network, that develops various ways of handling the traffic
flows, and to improve the performance of the network which consist of switches
and routers. sFlow [12] is an open-source sampling tool used for measuring the
traffic which is compatible with OpenFlow network. It consists of sFlow agents and
collector. The sFlow agent captures traffic statistics from the device which is under
observation that uses sampling technology. sFlow datagrams immediately forward
the sampled traffic statistics to a sFlow collector for analysis. The main task of the
two modules is listed below.
1. sFlow Collector: It is a server where sFlow datagrams are collected and stored.
2. sFlow Analyzer: It provides real-time overview of the network traffic flow by
analyzing the received datagrams by analyzing the irregularities of the network
parameters and detailed information. sFlow agents send a stream of sFlow sam-
plings continuously to the collector where they are analyzed to supply a real-time,
network-wide view of traffic flows.

3.1 Integration of Mininet and sFlow

Network simulators play an important role by evaluating network topologies of dif-

ferent types over a small scale. Many network simulators like Mininet, GNS3 and
EsiNet are available. Mininet is used here which is an open-source network simulator
used to generate traffic and analyze its flow.
Mininet [8] is an open-source network emulator and is a command line interface
(CLI) and enabled simulator which supports the use of analysis tools like the sFlow,
488 S. Hublikar et al.

NetFlow and RMON. Mininet creates virtual switch, hosts, links and controllers on
a single Linux kernel with a single command. Custom topologies are created using
Mininet. It simulates a real machine and can create different hosts. The limitation of
Mininet is that it does not have OpenFlow controller and it runs on slower links (10
or 100 Mbps).
According to [7], these simulators provide a platform to set a network topology
as a replication to the real-world environment about analysis and detection of the
attacks using Mininet and sFlow. The analysis is done in a number of steps in the
tool, which are listed here.
1. [Link] will run the shell script command to run the sFlow application.
2. In another window, Mininet topology will execute. The ping command is run to
check the connectivity between the hosts of the topology created. A zero percent-
age drop depicts the complete connectivity between the hosts.
3. For accessing the sFlow trend GUI, a local host is created using the command:
localhost: 8008.
4. A typical command in Mininet for detection of attack is given as:
[Link] It provides an approach to run
a SDN Controller along with it.

4 Implementation

In this section, the implementation of the network is discussed. Initially, sFlow-RT

is created to receive a continuous flow of data that is sent from the network devices
and converted into metrics. As soon as the flow reaches certain predefined metric
level, it is sent to an analyzer. Next, using the Mininet command, a topology is built
with link bandwidths of 10 Mbps. Finally, the output is the link between two hosts.
In order to make it easier to get started, the latest release of sFlow-RT includes
a Mininet helper script [Link] that automates sFlow configuration. The following
example shows how to use the script and build a simple application in Python.
The various steps to detect the flows in a Mininet topology created in Linux
environment using sFlow-RT are listed here and shown in Fig. 5.
1. sFlow-RT: sFlow-RT is an open-source tool that has an embedded OpenFlow con-
troller, allowing monitoring and flow insertions to OpenFlow supporting switches.
It analyzes certain events of interest, raise triggers and apply traffic handling rules
to a particular controller. In order to analyze sFlow-RT flows and react on traffic
changes, it has to be configured to work together with the existing network.
2. For Creating Mininet: In a second terminal, add the—custom argument to the
Mininet command line. The following command builds a depth ‘2’ tree topology
with link bandwidths of 10 Mbit/s.
cd sflow-rt
sudo mn –custom extras/[Link] –link tc, bw=10 –topo tree, depth=2, fanout=2.
The response of the tool after creating the topology is shown in Fig. 6. The [Link]
Detecting Denial-of-Service Attacks Using sFlow 489

Fig. 5 Starting sFlow

Fig. 6 Creating topology

script extends Mininet, automatically enabling sFlow on each of the switches in

the topology, and posting a JSON representation of the Mininet topology using
sFlow-RT.

5 Results and Discussion

Whenever the network is simulated without any attack, bandwidth between the two
hosts maintained at 9.63 Mbps. When a DOS attack is initiated on the network, the
bandwidth falls to 30 kbps. The bandwidth has reduced from Mbits to kbits due to the
DoS attack which has increased the datarate. This resulted in the decreased system
performance. The GUI output of the DoS detection is shown in Fig. 7. Here, sFlow
is initiated in the local host to detect the DoS attack when host h1 pings h2. As more
number of packets are sent, DoS attack is detected.
490 S. Hublikar et al.

Fig. 7 DoS attack detected

6 Conclusion

DoS attack was detected by acquiring network bandwidth and disrupting the services
of the server by abruptly increasing the traffic by making the server unavailable for
other users. DoS attack was detected using the sFlow tool. In case of any attack,
sFlow collects sample packets from network traffic, analyzes suspicious behavior
and creates handling rules which are then sent to the controller. Implementation of
DoS attack is carried out by emulating a typical network in Mininet and integrating
this with sFlow analyzer. The results indicate efficient identification of DoS attack
by using sFlow.

References

1. Ambrosin M, Conti M, De Gaspari F, Poovendran R (2017) Lineswitch: tackling control plane

saturation attacks in software-defined networking. IEEE/ACM Trans Netw (TON) 25(2):1206–
1219
2. Dridi L, Zhani MF (2018) A holistic approach to mitigating DOS attacks in SDN networks.
Int J Netw Manag 28(1):e1996
3. Jyothirmai P, Raj JS, Smys S (2017) Secured self organizing network architecture in wireless
personal networks. Wirel Pers Commun 96(4):5603–5620
4. Nugraha M, Paramita I, Musa A, Choi D, Cho B (2014) Utilizing OpenFlow and sFlow to
detect and mitigate SYN flooding attack, 17(8):988–994
5. Ombase PM et al (2017) Survey on DOS attack challenges in software defined networking. Int
J Comput App 975:8887
6. Othman RA (2000) Understanding the various types of denial of service attack. Bus Week
Online. Accessed 12 Feb 2000
7. Peter: Mininet flow analytics. [Link]
Accessed 10 Jan 2019
8. Scarlato M. Network monitoring in software defined networking (thesis). Accessed 30 Jul 2014
Detecting Denial-of-Service Attacks Using sFlow 491

9. Shang G, Zhe P, Bin X, Aiqun H, Kui R (2017) Flooddefender: protecting data and control
plane resources under SDN-aimed DOS attacks. In: INFOCOM 2017-IEEE conference on
computer communications. IEEE, pp 1–9
10. Sridhar S, Smys S (2016) A hybrid multilevel authentication scheme for private cloud environ-
ment. In: 2016 10th international conference on intelligent systems and control (ISCO). IEEE,
pp 1–5
11. Stallings W (2015) Foundations of modern networking: SDN, NFV, QoE, IoT, and Cloud.
Addison-Wesley Professional
12. Swapna AI, Reza MRH, Aion MK (2016) Security analysis of software defined wireless net-
work monitoring with sFlow and FlowVisor. In: International conference on communication
and electronics systems (ICCES). IEEE, pp 1–7
Area and Power Efficient Multiplier-Less
Architecture for FIR Differentiator

Sanjay Eligar and R. M. Banakar

Abstract Digital FIR filters have been used for various signal processing tasks in
embedded systems. This paper presents a novel architecture for implementation of
an FIR differentiator in FPGA designed for a specific application. The designed
component is to estimate the velocity from the suspension displacement in a semi-
active suspension controller. The architectures of FIR are modified based on the
direct and transposed form, with incremental changes in the manner in which the
constant-coefficient multiplication is executed. The recoding of constant FIR filter
coefficients using Canonic Signed Digit representation is explored. Three architec-
tures are designed and compared for efficient usage of area of implementation in
FPGA. It is observed that the architecture using CSD requires 16% lesser area, and
the optimized architecture is an additional 5% more area-efficient and 16% lesser in
complexity of design because of sharing of resources. An overall reduction in power
consumption by 3% is observed and the computation of the FIR filter convolution
sum is faster by 17%.

Keywords FIR differentiator · Constant multiplication · Canonic Signed Digit ·

Design validation

1 Introduction

Linear phase FIR filters are a category of filters which exploit the property of sym-
metry or anti-symmetry to reduce the number of multiplications in the computational
algorithm. In the FIR convolution, each output sample (y[n]) is obtained as a sum
of weighted samples of the input signal of finite number and the delayed input sam-
ples (x[n], x[n − 1]...x[n − k]) [6]. A close observation of computation reveals that

S. Eligar (B) · R. M. Banakar

BVB College of Engineering and Technology, Hubli, India
e-mail: eligar@[Link]
R. M. Banakar
e-mail: banakar@[Link]

© Springer Nature Singapore Pte Ltd. 2020 493

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
494 S. Eligar and R. M. Banakar

there is a continuous multiplication of input samples by constants (bk ). This condi-

tion which is further explored here, and opens up enormous possibilities to optimize
in an FPGA. The convolution sum can be implemented using different types of FIR
structures. The most simplest approach is the direct form FIR digital filter structure
in which the filter coefficients of the transfer function is nothing but the multiplier
coefficients themselves.
In [4] an FIR filter is designed as a differentiator for the specifications of a velocity
estimator for semi-active suspension controller. Velocity estimation is one of the
primary requirements for the application discussed here, and which is a controller
for semi-active suspension system using magneto-rheological dampers fitted to a
passenger car. The controller chosen is a variable structure controller of type Sigma-
1 and the computation requires estimation of velocity from linear displacement of
suspension deflection [3]. An FIR of Type III is chosen for implementation with the
specifications as: pass-band frequency ( f p = 400 Hz), stop-band frequency ( f st =
500 Hz) and sampling frequency ( f s = 1 kHz). Using equiripple method for design
yielded a structure of length M = 7 [4] and the filter coefficients are as shown in
Table 1. Because of the anti-symmetric behaviour of the odd length filter designed
here, three pairs of filter coefficients are equal in magnitude, but opposite in sign
and the centre coefficient (b3 ) is 0. The computation of an output sample y[n] needs
6 multiplications and 5 additions apart from the delay elements. This is definitely
not an optimal architecture since the anti-symmetry property of the linear phase FIR
filter is not exploited. Accordingly, the optimized expression for computation of the
convolution sum is shown in Eq. (1).

y[n] = b0 (x[n] − x[n − 6]) + b1 (x[n − 1] − x[n − 5]) + b2 (x[n − 2] − x[n − 4])
(1)
As seen the number of product terms is reduced from 6 to 3, but the number of
additions remains same at 5. This is a huge savings in computation resources (50%),
since multiplication of two 16-bit numbers in Q15 fixed-point format consumes lot
of resources. This optimization works very well if the implementation of the system
is done using DSPs, which have dedicated multiply-and-accumulate (MAC) units. In

Table 1 FIR filter coefficients for M = 7

bk Values Q15 Non CSD Non
0’s 0’s
b0 0.15175991026808974 0001001101101101 8
b1 −0.41604328754013736 1100101010111111 11
b2 0.94234994673082473 0111100010011111 10
b3 0 0000000000000000 0
b4 −0.94234994673082473 1000011101100001 7 1̄00010001̄01̄00001 5
b5 0.41604328754013736 0011010101000001 6 0101̄010101000001 6
b6 −0.15175991026808974 1110110010010011 9 0001̄01̄0010010101̄ 6
Area and Power Efficient Multiplier-Less Architecture for FIR Differentiator 495

the application chosen here, the filter coefficients are fixed, and so the multiplier or
MAC unit need not be programmable to take different values of coefficients. Thus,
the focus of optimization should now be towards multiplication of an input variable
by a constant coefficient. This leads to further exploration in optimal multipliers
starting from the shift-and-add algorithm which is discussed in the next section.
In Sect. 2, the specific case of multiplication of a constant with a variable is dis-
cussed along with the need to use the transposed form architecture of FIR. Section 3
explores the possibility of reducing the number of adders through Canonic Signed
Digit representation of the filter coefficients. Section 4 explores the possibilities of
using common sub-expressions in multipliers and other techniques to minimize the
hardware requirements. In Sect. 5, the top-level architecture of the design is presented
along with validation of design and the results of synthesis. Finally, the concluding
observations are presented in Sect. 6.

2 Constant-Coefficient Multiplication

Constant-coefficient multiplication optimization is one of the important domains of

low power and area-efficient designs of FIR filters. Literature in this domain has
extensive research on two types of problems: single constant multiplication (SCM)
[9] and multiple constant multiplication (MCM) [10]. In literature related to digital
signal processing, the minimization of area and computation time is seen as major
criteria. In [1], the authors propose an algorithmic optimization which discovers
the common sub-expressions that exist in the computations of various products. In
[5], the authors propose a multiplier-less FIR filter using a CSD notation which
minimizes the number of non-zero values in the constant coefficient. Apart from
the algorithmic changes, some optimizations are also done at the architecture level.
In [2], Duraiswamy et al., propose an area-efficient FIR architecture that uses the
anti-symmetry property to compute only half of the overall products needed to be
computed. The remaining products are evaluated by using NOT gates and setting the
carry-in of the remaining tap adders to ‘1’.
In the design problem chosen in this paper, the focus is to arrive at the trade-off in
filter architectures in choosing between direct versus transposed form FIR structure,
Q15 versus optimal CSD implementation and finally searching for common sub-
expressions or/and shifters in the design which can be shared across all the product
terms and not just a single multiplication. Since the length of the FIR filter is small,
manual sub-expressions with a mix of Q15 and CSD representation are used to
discover common sub-expressions.
The most basic form of constant-coefficient multiplier is a shift-and-add multi-
plier, in which the number of add operations required is one less than the number
of non-zero bits in the constant coefficient [6]. The initial design of the FIR filter is
implemented using 16-bit adders and shifters. In order to exploit the anti-symmetry
property of the design, and use a simple shift-add architecture, an equality of the
form b0 x[n] = −b6 x[n] should exist. This is not possible if a direct form of FIR
496 S. Eligar and R. M. Banakar

structure is used as in this case the multiplication by anti-symmetric coefficients is

with different input samples, b0 x[n] and b6 x[n − 6]. To summarize, we need to store
the original result of b0 x[n] and use it later for computation of b6 x[n − 6] = −b0 x[n].
This scenario could be avoided if a transposed structure of FIR is used. In this case,
the multiplication needed is with the same input sample across the structure at any
given point of time. The assignment b0 x[n] = −b6 x[n] in digital logic is easily
achieved by means of an inversion and addition by ‘1’, since Q15 notation uses 2’s
complement notation.
Multiplication by −1 can be simply achieved by using NOT gates for inversion
and setting the carry input to ‘1’ in the tap adders wherever needed. The constant-
coefficient data is represented using Q15 format and the corresponding values are
shown in Table 1 in the third column. In the design option, there is a choice to either
use Option 1: {b0 , b1 , b2 } or Option 2: {b4 , b5 , b6 }, to exploit the anti-symmetry
property. The choice is then decided based on the total number of non-zero bits in
the sets of three coefficients. As seen from the fourth column in Table 1, Option 1 has
29 non-zero bits as compared to Option 2 which has 22 non-zero bits. Also the first
adder in Option 1 would need inputs as {b6 x[n], b5 x[n]} and an additional adder
to add ‘1’ despite using carry-in as ‘1’ for one of the adders. To summarize, Option
1 needs 7 + 10 + 9 + 1 + 5 = 32 adders while Option 2 needs 6 + 5 + 8 + 5 = 24
adders, which is a reduction in requirement of adders by 25%.
Thus, multiplication using Q15 notation for filter coefficients requires 13 unique
shifters and 20 shifters in all. The transposed form of the FIR differentiator now
need not use 6 multipliers but uses only 3 multipliers. The remaining 3 product terms
are implemented using ‘multiply by 1’ logic, which uses NOT gates and the binary
‘1’ is added in the tap adders at next stage as discussed earlier in this section. The
entire FIR structure can now be rearranged and is as shown in Fig. 1. This is the
most basic architecture which does not optimize the multiplication operation. This
architecture is referred to as ‘A1’ in this paper. It uses Q15 representation of constant
coefficients and a shift-and-add architecture for multiplication. All the data lines in
the architecture are of 16-bit width, except for the carry-in of the last three tap adders

Fig. 1 FIR differentiator using Q15 (A1)

Area and Power Efficient Multiplier-Less Architecture for FIR Differentiator 497

which is 1-bit width. The control lines which are used for clock, reset and enable are
not shown in the figure. The next section explores the possibility of further reduction
in the number of adders required for computing the convolution sum.

3 Canonic Signed Digit Multiplication

The constant filter coefficients can be recoded such that they contain the fewest num-
ber of non-zero bits which can be accomplished using ‘Canonic Signed Digit’ (CSD)
representation. CSD recoding enables the reduction in the number of partial products
during multiplications [7, 8]. The filter coefficients in the design discussed here can
now be recoded using the 16-bit CSD representation for the three coefficients b4 , b5
and b6 as shown in last two columns of Table 1. The architecture which uses optimal
CSD representation for constant coefficients and shift-add method for multiplication
is referred to as ‘A2’ in this paper. Architecture ‘A2’ uses 19 adders, 15 shifters (11
unique shifters) and 48 NOT gates, and is shown in Fig. 2. The number of adders
required in ‘A2’ is now 4 + 5 + 5 + 5 = 19, which is a reduction of 21% as com-
pared to ‘A1’. The architecture presented here is similar to [2] in which Jason Thong
et al., demonstrate the usage of NOT gates and tap adders with carry-in set to ‘1’
to implement the negation operation of half of the product terms in anti-symmetric
FIR architecture. In this paper, we propose a further optimization of the architecture
where in common sub-expressions are explored along with shift operations which
can be shared across all the product terms. Once the number of adders required for
overall implementation is minimized, the next level of optimization is explored in the
shifters used for multipliers. The next section explores various ways of optimizing
the shift operation in the CSD multiplier.

Fig. 2 Optimized architecture of FIR differentiator using CSD (A2)

498 S. Eligar and R. M. Banakar

4 Optimization of CSD Multiplier

The computation of convolution sum involves using 3 multiplications which are

executed concurrently, namely b4 x[n], b5 x[n] and b6 x[n]. These product terms are
implemented using a series of shift-and-add operations. The next process of opti-
mization is to focus on minimizing the overall number of computations. Since in all
the multiplication operations one of the inputs is the same input sample x[n], there is
a scope to share some of the addition or/and shift operations. This gives rise to possi-
bilities of re-examining the choice of using CSD for all multiplications. The first task
is to identify the occurrence of any common sub-expressions which may appear in
common across 2 or 3 multiplications in the design. In that case, because of concur-
rency in the computation, these sub-expressions can be shared across the multipliers,
thereby reducing the quantum of resources consumed. There could be a scenario
where a direct Q15 representation would yield more common sub-expressions as
compared to the CSD format. Another possibility is a combination in which 1 or
2 multiplications are in Q15 and the other/s are in CSD format, may yield more
common sub-expressions.
The final selection of the data representations of constant coefficient happens
based on that combination which leads to least number of adders. In the design of
FIR differentiator used here, and the final choice is CSD implementation of all the
three coefficients as there is no common sub-expression. This choice is definitely not
a universal choice, since the optimizations are heavily dependent on the number of
filter coefficients and their values. The next level of optimization is to choose that
option which needs least number of unique shifters and also maximum number of
shifters which are shared. In the next part of this section, the possibility of sharing
the shifters is explored further.
The logical structure of a shifter if implemented in an FPGA is hard wiring
between the bit registers. For example, a y = (x [n ]  1) shifter, which is a right
shift by 1, would simply connect bit x[15] to y[14], x[14] to y[13] and so on till
the connection x[1] to y[0]. The most significant bit of y[15] is then cleared to ‘0’.
Thus, there is no need to have any underlying logic like OR/AND gates in between,
irrespective of the amount of shifts needed. The only additional logic needed for the
shifter would be the control and reset logic which decides when the shifter should be
enabled and under what circumstances it should be reset. Since the shift operations
are concurrent for all the three multiplications encountered here, the common shift
operations can be shared which results in an alternate architecture which uses only 11
shifters as compared to 16 in ‘A2’, other resources being the same. This architecture
is optimal in that it uses minimum number of adders and shifters as compared to ‘A1’
and ‘A2’, and is shown in Fig. 3. A seen the last three tap adders in the architecture
have an additional carry input of ‘1’, which is needed to implement the negation
expression for the coefficient product terms discussed in Sect. 2. All the data lines
in the architecture are of 16-bit width except the carry-in bits of last 3 tap adders
which are single-bit width. Again the control lines which are used for clock, reset
and enable are not shown in this figure.
Area and Power Efficient Multiplier-Less Architecture for FIR Differentiator 499

Fig. 3 FIR differentiator using shared CSD shifters (A3)

5 Results and Discussion

The prototyping of the FIR differentiator is done on a FPGA development board

using multiple architectures as discussed here. All 3 designs use the multiplier-
less architecture in a transposed structure of FIR filter. All 3 designs implement
3 coefficient products, and the remaining 3 product terms are arrived at by using
anti-symmetry property of filter coefficients using NOT gates. The design A1 is
implemented using Q15 notation for constant coefficients. In design ‘A2’ optimal
CSD notation is used, which minimizes the number of adders in the architecture.
Finally, in ‘A3’, the shift operations are shared between concurrent multiplications
yielding an optimal architecture.
This paper presents the synthesis results for a Spartan6 FPGA device. The results
of synthesis of all the 3 architectures ‘A1’, ‘A2’ and ‘A3’ are listed in Table 2. The
results of number of ‘adders’ have already been established earlier through the design
of FIR architectures. Both ‘A2’ and ‘A3’ have 21% lesser number of adders as
compared to ‘A1’. The overall area of the implementation is estimated based on the
sum total of slice registers and LUTs used in the design. Based on the synthesis results
it is seen that ‘A2’ is 16.2% area-efficient than ‘A1’, while ‘A3’ consumes 21.5%
lesser area than ‘A1’. This reduction demonstrates the benefits of using optimal CSD
representation of the filter coefficients instead of notation in Q15 format.
The number of latches used in ‘A3’ is the highest at 528, while ‘A3’ is better than
‘A2’ by 16%. This optimization is due to sharing of shifters among all the product
terms in the FIR implementation. The reduction in complexity for ‘A3’ is 33.3% when
compared to ‘A1’. This proves that the architecture ‘A3’ is the most area-efficient
as compared to [2]. The reduction in area obviously reduces the power consumption
of the system as seen in Table 2. It is seen that ‘A3’ consumes the least power at 2.4
mW, which is an additional power savings of 1.24 and 2.95% as compared to ‘A2’
and ‘A1’, respectively.
500 S. Eligar and R. M. Banakar

Table 2 Comparison of FIR differentiator architectures

Parameter A1 A2 A3
Slice registers 511 430 417
Slice LUTs 446 372 334
LUT FF pairs 564 479 463
Adders 24 19 19
Latches 528 419 352
Datapath delay (ns) 13.823 11.273 11.462
Power (Logic+Signal) 3.4 2.43 2.40
(mW)

The area and power advantages in ‘A3’ come at the cost of increased data path
delay of 11.462 ns, which is 1.7% slower than ’‘A2’. But it is definitely faster than
‘A1’ by 17.1%.

6 Conclusion

The design and implementation of an FIR differentiator to estimate the velocity

from suspension displacement are presented. The advantage in optimizing hardware
resources by using the transposed form of FIR filter structure is demonstrated. It
allows to convert the problem of single constant multiplication to multiple constant
multiplication, thereby allowing for sharing of resources by means of common sub-
expressions and shifters. An alternate representation of Q15 filter coefficients using
Canonic Signed Digit is presented, which yields an area optimization of 22% for the
chosen design specifications. The number of adders needed for implementation of
the FIR convolution sum is also reduced by 21%. The design complexity in terms
of latches and multiplexers used reduces by 33% as compare to the standard Q15
representation. The proposed design is also power efficient by 3% and computes
the sum faster by 17%. The future work is to explore implementation of the FIR
differentiator is VLSI where further optimizations can be done at circuit-level of
abstraction.

References

1. Chen MC, Chen TT (2014) Minimizing design costs of an FIR filter using a novel coefficient
optimization algorithm. Math Probl Eng 1–9. [Link]
2. Duraiswamy P, Bauwelinck J, Vandewege J (2011) Efficient implementation of 90 phase shifter
in FPGA. EURASIP J Adv Sig Proc 2011(1):32
Area and Power Efficient Multiplier-Less Architecture for FIR Differentiator 501

3. Eligar S, Banakar RM (2014) A model based approach for design of semiactive suspension
using variable structure control. Int J Tech Res Appl. e-ISSN 2320–8163
4. Eligar S, Banakar RM (2018) Optimization of control algorithm for semi-active suspension
system. In: 2018 proceedings of the international conference on intelligent computing and
sustainable system
5. Illa A, Haridas N, Elias E (2016) Design of multiplier-less FIR filters with simultaneously
variable bandwidth and fractional delay. Eng Sci Tech Int J 19(3):1160–1165
6. Mitra SK, Kaiser JF (1993) Handbook for digital signal processing. Wiley, New York
7. Parhi KK (2007) VLSI digital signal processing systems: design and implementation. Wiley,
New York
8. Proakis JG (2001) Digital signal processing: principles algorithms and applications. Pearson
Education, India
9. Thong J, Nicolici N (2010) A novel optimal single constant multiplication algorithm. In:
Proceedings of the 47th design automation conference, pp 613–616. ACM
10. Thong J, Nicolici N (2011) An optimal and practical approach to single constant multiplication.
IEEE Trans Comput-Aided Des Integr Circ Syst 30(9):1373–1386
Optimized Prediction Model to Diagnose
Breast Cancer Risk and Its Management

Athira Vinod and B. R. Manju

Abstract Breast malignancy is the second biggest disease that results in fatal con-
dition for women population. Research endeavors have revealed with expanding
affirmation that the support vector machines (SVMs) have more noteworthy pre-
cise conclusion capacity. In this paper, breast disease determination is dependent on
a SVM-based technique that has been proposed. Investigations have been directed
on various preparing test allotments of the Wisconsin breast malignancy dataset
(WBCD), which is generally utilized among scientists who use machine learning
strategies for breast disease conclusion. The working of the technique is assessed
by utilizing characterization precision, particularity positive and negative prescient
qualities, collector working trademark bends, and perplexity lattice. The outcomes
demonstrate that the most elevated grouping precision (99%) is achieved for the
SVM.

Keywords Breast cancer classification · Support vector machine (SVM)

1 Introduction

Breast disease is the most widely recognized harm among women, representing
almost 1 of every 3 tumors analyzed among women in the USA, and it is the second
driving reason for malignancy demise among women. Breast cancer happens because
of anomalous cell developing in the breast tissue, generally alluded to as a tumor.
A tumor does not mean disease—tumors can be considerate (not harmful), pre-
threatening (pre-dangerous), or dangerous (malignant) [1, 2]. Tests such as MRI,

A. Vinod (B)
Amrita Center for Wireless Networks & Applications (AWNA), Amrita School of Engineering,
Amrita Vishwa Vidyapeetham, Amritapuri 690525, India
e-mail: athiravinodathu@[Link]
B. R. Manju
Department of Mathematics, Amrita School of Engineering, Amrita Vishwa Vidyapeetham,
Amritapuri 690525, India
e-mail: manjubr@[Link]

© Springer Nature Singapore Pte Ltd. 2020 503

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
504 A. Vinod and B. R. Manju

mammogram, ultrasound, and biopsy are commonly used to diagnose breast cancer
performed.
According to the studies, breast cancer is the most common type of cancer for
women regardless of race and ethnicity. Although we may not be aware of all the
factors contributing to develop breast cancer, certain attributes such as family history,
age, obesity, alcohol, and tobacco use have been identified from research studies on
this topic [3]. Like the classification of different features of breast cancer, there are
mechanisms to find the faults which occur in variable wind turbine blade using J48
algorithm [4].
The previous research work carried out in the domain of breast cancer: In [5], a
method was proposed to work on a set of images which were extracted from MIAS
and DDSM databases based on local binary pattern (LBP).
In [6], a method is proposed to detect mammograms, which in turn is used to
determine the breast cancer. Here, the algorithms like logistic regression, decision
tree, and Hoeffding tree is used and a comparative analysis is done to validate the
accuracies. But the method uses, neglecting missing values and working over the
remaining dataset, which is not feasible due to the reason that every data is important
and must be handled in a way such that no data is lost.
In [7], classification of breast cancer into benign and malignant is done using SVM
classifier which produces an accuracy of 92.15%. Hence, the following proposed
model is for optimizing the existing SVM classifier and gives a best accurate model.
In [8], discovery of breast malignant growth utilizing (a) marker-controlled level
set division of anisotropic dispersion sifted preprocessed picture versus and (b)
segmentation utilizing marker-controlled level set on a Gaussian-separated picture.
This is reasonable for highlight extraction in PC supported breast malignant growth
finding.
In [9], the results show that the level set method and improved edge map are
suitable for the accurate segmentation of breast regions from thermal images. The
capability of anisotropic filter on breast thermograms is determined and validated by
extracting the tissues using level sets.

2 Objective

The marks in the information are discrete, and the expectation falls into two classi-
fications (e.g., dangerous or favorable). In machine learning, this is an arrangement
issue. In this way, the objective is to arrange whether the breast disease is amiable
or threatening and foresee the repeat and non-repeat of harmful cases after a specific
period. To accomplish this, the model utilized machine learning characterization
strategies to fit a capacity that can foresee the discrete class of new info.
Optimized Prediction Model to Diagnose Breast Cancer Risk … 505

3 Proposed System

This work focuses on investigating the probability of predicting the type of breast
cancer (malignant or benign) from the given characteristics of breast mass computed
from digitized images. The cases provided are cases diagnosed with some type of
tumor, but only some of them (approximately 37%) are malignant. Here, the available
data is examined and attempted to predict the possibility that a breast cancer diagnosis
is malignant or benign based on the attributes collected from the breast mass.
The proposed methodology consists of the following steps [10]:
1. Identifying data sources.
2. Data preprocessing.
3. Data visualization (exploratory data analysis).
4. Machine learning prediction.
Proposed methodology of this prediction model is depicted in Fig. 1.

3.1 Identifying data sources

In [11], the dataset is accessible from machine learning store kept up by the University
of California, Irvine. The dataset contains 569 examples of harmful and considerate
tumor cells.

Fig. 1 Proposed methodology

506 A. Vinod and B. R. Manju

• The initial two sections in the dataset store are one of cancers (M = malignant, B
= benign), separately.
• The Sects. 3–3.2 contain 30 genuine esteem includes that have been registered from
digitized pictures of the cell cores, which can be utilized to assemble a model to
anticipate whether a tumor is amiable or dangerous.

3.2 Data Preprocessing

Data preprocessing [12] is an urgent advance for any information investigation issue.
Usually, a smart thought to set up the information in such approach to best uncover
the structure of the issue to the machine learning calculations that you expect to
utilize. This involves a number of activities such as:
• Assigning numerical values to categorical data;
• Handling missing values; and
• Normalizing the features (so that features on small scales do not dominate when
fitting a model to the data).
The objective of the data preprocessing is to locate the most prescient highlights
of the information and channel it. Along these lines, it will upgrade the prescient
intensity of the examination demonstrate.
In this work, the accompanying advances are taken consideration:
• Allocate highlights to a NumPy cluster X and change the class marks from their
unique string portrayal (M and B) into numbers.
• Split data into training and test sets.
• Institutionalize the information.
• Acquire the eigenvectors and eigenvalues from the covariance lattice or connection
grid.
• Sort eigenvalues in plunging request and pick the k eigenvectors that relate to the
k biggest eigenvalues where k is the quantity of measurements of the new element
subspace, k ≤ dk ≤ d.
• Build the projection lattice W from the chose k eigenvectors.
• Change the first dataset X by means of W to get a k-dimensional component
subspace Y.
It is common to select a subset of features that have the largest correlation with
the class labels. The effect of feature selection must be assessed within a com-
plete modeling pipeline in order to give you an unbiased estimated of your model’s
true performance. Hence, in the next section you will first be introduced to cross-
validation, before applying the PCA-based feature selection strategy in the model
building pipeline.
Optimized Prediction Model to Diagnose Breast Cancer Risk … 507

3.3 Data Visualization

Exploratory data analysis (EDA) is an imperative advance which happens after ele-
ment designing and obtaining information and it ought to be done before any demon-
strating. This is on the grounds that it is essential for an information researcher to
almost certainly comprehend the idea of the information without making suspicions.
The aftereffects of information investigation can be very helpful in getting a handle
on the structure of the information, the appropriation of the qualities, and the nearness
of extraordinary qualities and interrelationships inside the informational index.
The purpose of EDA is:
• To utilize rundown measurements and representations to all the more likely com-
prehend information, discover pieces of information about the inclinations of the
information, its quality and to detail suppositions, and the speculation of our
hypothesis.
• For data preprocessing to be effective, it is fundamental to have a general image of
your information. Basic factual portrayals can be utilized to distinguish properties
of the information and features (which features esteems ought to be treated as com-
motion or exceptions). Next step is to explore the data. There are two approaches
used to examine the data using:
• Descriptive statistics is the way toward gathering key attributes of the informa-
tional index into straightforward numeric measurements. A portion of the regular
measurements utilized is mean, standard deviation, and relationship.
• Visualization is the way toward anticipating the information, or parts of it, into
Cartesian space or into dynamic pictures. In the information mining process, infor-
mation investigation is utilized in a wide range of steps including preprocessing,
modeling, and interpretation of results.
At the end, we get the following conclusions, which are shown in the following
Fig. 2:
• Mean estimations of cell span, edge, territory, conservativeness, concavity, and
sunken focuses can be utilized in characterization of the malignancy. Bigger
estimations of these parameters will in general demonstrate a connection with
dangerous tumors.
• Mean estimations of surface, smoothness, and symmetry do not demonstrate an
inclination of one conclusion over the other.
• In any of the histograms, there are no recognizable vast anomalies that warrants
further cleanup.

3.4 Machine Learning Prediction

Support vector machines (SVMs) [13] learning algorithm will be used to build the
predictive model. SVMs are one of the most popular classification algorithms and
508 A. Vinod and B. R. Manju

Fig. 2 Breast cancer feature correlation matrix

have an elegant way of transforming nonlinear data, so that one can use a linear
algorithm to fit a linear model to the data.
Kernelized support vector machines [14] are ground-breaking models and perform
well on an assortment of datasets.
• SVMs take into consideration complex choice limits, regardless of whether the
information has just a couple of highlights.
• They function admirably on low-dimensional and high-dimensional information
(i.e., few and numerous highlights) and however do not scale great with the quantity
of tests.
• Running a SVM on information with up to 10,000 examples may function
admirably yet working with datasets of size at least 100,000 can wind up testing
regarding runtime and memory use.
• SVMs require cautious preprocessing of the information and tuning of the param-
eters. In this way, these days the vast majority incline toward tree-based models,
Optimized Prediction Model to Diagnose Breast Cancer Risk … 509

for example, irregular woodlands or angle boosting (which require practically no

preprocessing) in numerous applications.
• SVM models are difficult to investigate; it very well may be hard to comprehend
why a specific forecast was made, and it may be dubious to disclose the model to
a non-expert.
Machine learning prediction has these following steps:
1. Split data into training and test sets.
2. Classification with cross-validation.
3. Model accuracy: receiver operating characteristic (ROC) curve.
4. Optimizing the SVM classifier.

3.4.1 Split Data into Training and Test Sets

This method to compare the performance of a machine learning algorithm is to use

different training and testing datasets.
• Divide the available data into a training set and a testing set (70% training, 30%
test),
• Train the algorithm on the 70% data,
• Make predictions on the 30% data, and
• Compare the predictions results against the expected results.
The extent of the split can rely upon the size and particulars of your dataset, despite
the fact that usually to utilize 67% of the information for preparing and staying 33%
for testing.

3.4.2 Classification with Cross-Validation

Split the data into test, and training set is crucial to avoid overfitting. This allows
generalization of real, previously unseen data. Cross-validation extends this idea
further. Instead of having a single train/test split, we specify so-called folds so that
the data is divided into similarly sized folds.
• Training occurs by taking all folds except one—referred to as the holdout sample.
• On the completion of the training, you test the performance of your fitted model
using the holdout sample.
• The holdout sample is then thrown back with the rest of the other folds, and a
different fold is pulled out as the new holdout sample.
• Training is repeated again with the remaining folds, and we measure performance
using the holdout sample. This process is repeated until each fold has had a chance
to be a test or holdout sample.
• The expected performance of the classifier, called cross-validation error, is then
simply an average of error rates computed on each holdout sample.
510 A. Vinod and B. R. Manju

This process is demonstrated by first performing a standard train/test split and then
computing cross-validation error. The three-fold cross-validation accuracy score for
this classifier is 0.97.

3.4.3 Model Accuracy: Receiver Operating Characteristic (ROC)

Curve

In statistical modeling and machine learning, a commonly reported performance

measure of model accuracy for binary classification problems is area under the curve
(AUC). To understand what information the ROC curve conveys, consider the so-
called confusion matrix that essentially is a two-dimensional table where the classifier
model is on one axis (vertical), and ground truth is on the other (horizontal) axis, as
shown below. Either of these axes can take two values (as depicted).
In an ROC curve plotted, “true-positive rate” on the Y-axis and “false-positive rate”
on the X-axis, where the values “true positive,” “false negative,” “false positive,” and
“true negative” are events (or their probabilities) as described above. The rates are
defined according to the following equations:
• True-positive rate (or sensitivity)}:

tpr = tp/(tp + fn) (1)

• False-positive rate:

fpr = fp/(fp + tn) (2)

• True-negative rate (or specificity):

tnr = tn/(fp + tn) (3)

In all definitions, the denominator is a row margin in the above confusion matrix.
Thus, one can express
• The true positive rate (tpr) is defined as the probability that the model says “+”
when the real value is indeed “+” (i.e., a conditional probability). However, this
does not tell you how likely you are to be correct when calling “+” (i.e., the
probability of a true positive, conditioned on the test result being “+”).
Figure 3 shows the receiver operating characteristic example.
• To interpret the ROC correctly, consider what the points that lie along the diagonal
represent. For these situations, there is an equal chance of “+” and “−” happening.
Therefore, this is not that different from making a prediction by tossing an unbiased
coin. Put simply, the classification model is random.
Optimized Prediction Model to Diagnose Breast Cancer Risk … 511

Fig. 3 Receiver operating

characteristic example

• For the points above, the diagonal, tpr > fpr, and the model say that you are in a
zone where you are performing better than random. For example, assume tpr =
0.99 and fpr = 0.01, Then, the probability of being in the true-positive group is
(0.99/(0.99 + 0.01)) = 99%.
• Furthermore, holding fpr constant, it is easy to see that the more vertically above
the diagonal you are positioned, the better the classification model.

3.4.4 Optimizing the SVM Classifier

Machine learning models are parameterized, so their conduct can be tuned for a given
issue. Models can have numerous parameters, and finding the best mix of parameters
can be treated as an inquiry issue. In the code built up, the point is to tune parameters
of the support vector machine (SVM) classification show utilizing scikit-learn.
We can tune two key parameters of the SVM algorithm:
• The value of C
• And the type of kernel.
The default method for SVM (the SVC class) is to use the radial basis function
(RBF) kernel with a C value set to 1.0. C value which indicates range of relax-
ation of margin. Like with K-nearest neighbor (KNN), a grid search using 10-fold
cross-validation with a standardized copy of the training dataset is performed. The
demonstration of SVC with RBF kernel is shown in Fig. 4. For the RBF kernel, the
value of gamma is set to be 0.7 for the optimization. This predictive model used
512 A. Vinod and B. R. Manju

Fig. 4 SVC with linear

kernel

Fig. 5 SVC with RBF

kernel

number of simpler kernel types like linear kernel and polynomial kernel, with C as
regularization parameter (Fig. 5).
• C values < 1 indicate less bias.
• C values > 1 indicate more bias.
The predictive model with kernel types like linear, polynomial (degree 3), and RBF
is creating decision boundaries for the classifiers by considering the mean values of
radius and texture.
Python scikit-learn provides two simple methods for algorithm parameter tuning:
• grid search parameter tuning.
• random search parameter tuning.

4 Results

There are two possible predicted classes: “1” and “0.” Malignant = 1 (indicates
presence of cancer cells) and benign = 0 (indicates absence).
• The classifier made a total of 174 predictions (i.e., 174 patients were being tested
for the presence of breast cancer).
Optimized Prediction Model to Diagnose Breast Cancer Risk … 513

Fig. 6 SVC with

polynomial kernel

• Out of those 174 cases, the classifier predicted “yes” 58 times and “no” 113 times.
• In reality, 64 patients in the sample have the disease, and 107 patients do not
(Fig. 6).
Rates as computed from the confusion matrix.
1. Accuracy: Overall, how often is the classifier correct?
• (TP + TN)/total = (57 + 106)/171 = 0.95
2. Misclassification Rate: Overall, how often is it wrong?
• (FP + FN)/total = (1 + 7)/171 = 0.05 equivalent to 1 minus Accuracy also
known as “Error Rate.”
3. True-Positive Rate: When it is actually yes, how often does it predict 1?
• TP/actual yes = 57/64 = 0.89 also known as “Sensitivity” or “Recall.”
4. False-Positive Rate: When it is actually 0, how often does it predict 1?
• FP/actual no = 1/107 = 0.01.
5. Specificity: When it is actually 0, how often does it predict 0? Also known as
true-positive rate.
• TN/actual no = 106/107 = 0.99 equivalent to 1 minus false-positive rate.
6. Precision: When it predicts 1, how often is it correct?
• TP/predicted yes = 57/58 = 0.98.
7. Prevalence: How often does the yes condition actually occur in our sample?
• actual yes/total = 64/171 = 0.34.
The optimized SVM classifier accuracy is 0.99. The fivefold cross-validation
accuracy score for this classifier is 0.97.
Performance measure in terms of precision, recall, f1-score, and support is tabu-
lated in Table 1. The demonstration of predicted versus actual values can be seen in
Fig. 7.
514 A. Vinod and B. R. Manju

Table 1 Performance measure of the classifier

Precision Recall f1-score Support
0 0.94 0.98 0.96 107
1 0.97 0.89 0.93 64
Avg/total 0.95 0.95 0.95 171

Fig. 7 Predicted values

versus actual values

5 Conclusion

This work exhibits the displaying of breast cancer as arrangement assignment utiliz-
ing support vector machine joined with highlight choice. The SVM performs better
when Wisconsin breast cancer dataset (WBCD) is institutionalized, with the goal
that all characteristics have a mean estimation of zero and a standard deviation of
one. We can figure this from the whole preparing dataset and apply the equivalent
change to the info properties from the approval dataset. The execution of the model
is assessed as far as exactness, false-positive rate, false-negative rate, specificity, pre-
cision, prevalence, misclassification rate, and ROC. The outcomes demonstrate that
the SVM display that contains five highlights is advanced execution for the model.

References

1. Dheeba J, Tamil Selvi S (2011) Classification of malignant and benign microcalcification using
SVM classifier. In: 2011 international conference on emerging trends in electrical and computer
technology, Nagercoil, 2011. pp 686–690
Optimized Prediction Model to Diagnose Breast Cancer Risk … 515

2. Karahaliou N et al (2008) Breast cancer diagnosis: analyzing texture of tissue surrounding

microcalcifications. IEEE Trans Inf Technol Biomed 12(6):731–738
3. Bardou D, Zhang K, Ahmad SM (2018) Classification of breast cancer based on histology
images using convolutional neural networks. IEEE Access 6:24680–24693
4. Manju BR, Joshuva A, Sugumaran V A data mining study for condition monitoring on wind
turbine blades using Hoeffding tree algorithm through statistical and histogram features. Int J
Mech Eng Technol (IJMET) 13(1):102–121
5. Král P, Lenc L (2016) LBP features for breast cancer detection. In: 2016 IEEE international
conference on image processing (ICIP), Phoenix, AZ, 2016. pp 2643–2647
6. Manju BR, Amrutha VS (2018) Comparative study of datamining algorithms for diagnostic
mammograms using principal component analysis and J48. ARPN J Eng Appl Sci
7. Souza JC et al (2017) Classification of Malignant and Benign tissues in mammography
using dental shape descriptors and shape distribution. In: 7th Latin American conference on
networked and electronic media (LACNEM 2017)
8. Witten IH, Frank E, Hall MA, Pal CJ (2016) Data mining: practical machine learning tools and
techniques. Morgan Kaufmann
9. Gopakumar S, Sruthi K, Krishnamoorthy S (2018) Modified level-set for segmenting breast
tumor from thermal images. In: 2018 3rd international conference for convergence in
technology (I2CT)
10. [Link]
11. UCI machine learning repository, breast cancer Wisconsin (Diagnostic) dataset, [Link]
[Link]/ml/datasets/Breast+Cancer+Wisconsin+%28Diagnostic%29
12. Data Preprocessing, [Link]
13. Mandhala VN, Sujatha V, Devi BR (2014) Scene classification using support vector machines.
In: 2014 IEEE international conference on advanced communications, control and computing
technologies, Ramanathapuram, 2014. pp 1807–1810
14. Support Vector Machines and Kernel Methods, 2004
Smart Cane Design for Indoor
and Outdoor Navigation:
A Cost-Effective Guide

Vandana Mansur and R. Sujatha

Abstract In this paper, the distinction between prototype and product is clearly
described. During product design, there are many processes involved which pave a
way for an operational product. The discussion on different types of prototypes gives
the designer a better insight into the product deployment and testing phase. The
in-depth knowledge of the testing involved in different types of prototypes assures
that all the technical specifications and product specifications are considered. A
questionnaire is developed to design the product that gives the detailed description
of the design flow which is to be followed during the development phase. The field
testing steps to be followed are known in advance if a proper design flow is adopted.
The concept of “Engineering, Design and Process” is discussed in detail for the
product design approach.

Keywords Smart cane · Product design · Microcontroller

1 Introduction

The visually challenged are those who are deprived of the ability to visualize the
world. They face many difficulties and challenges. According to recent statistics,
there are around 217 million people with visual imparity. Statistics also show that
80% of individuals aged over 50 are visually impaired. It is good and of more use,
if comfortable and user-friendly assistance is provided in terms of cost-effective
devices.
In [1–3], a blind stick model is developed using Arduino Uno, infrared sensor,
ultrasonic sensor, LCD display, playback module and voltage regulator. The ultra-
sonic sensor detects the object ahead of the person by measuring the distance from
the object to the stick. To detect the objects in left and right, IR sensors are used.

Vandana Mansur (B) · R. Sujatha

School of Electronics Engineering, Vellore Institute of Technology, Vellore, Tamil Nadu, India
e-mail: vandanacmansur@[Link]
R. Sujatha
e-mail: sujatha.r@[Link]
© Springer Nature Singapore Pte Ltd. 2020 517
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
518 Vandana Mansur and R. Sujatha

The objects appearing very close are also detected. IR sensor is used in order to
avoid calculation problem which may be caused by using many ultrasonic sensors. To
assist the blind person to reach the destination, a voice playback module is used and
that is controlled through commands generated by microphone. The intention here is
to develop a friendly sensing device for the blind. Power bank of 5 V is supplied for the
circuit that maintains a constant level of power supply for its working. The software
coded is embedded on to the Arduino board and the model works accordingly [4, 5].
In [6], the authors describe modeling of the sensor-assisted stick done by using
Pro/E Creo 5.0 software. The PRO/Engineer is 3D CAD/CAM/CAE-integrated soft-
ware developed by Parametric Technology Corporation (PTC). It is one of the world’s
leading CAD/CAM/CAE softwares that provide a wide range of integrated solutions
for various aspects of product designing and manufacturing. This application was
the first to market that provides a feature for solid modeling, drafting and assem-
bly modeling, finite element analysis and NC and tooling function for mechanical
engineers.
The microcontroller used here is ATmega8. It is based on AVR RISC architecture
which is low-powered CMOS 8-bit microcontroller. The sensor consists of distance
measurer comprising of position-sensitive detector, infrared emitting diode (IRED)
and signal processing unit. Triangulation method is opted for distance detection, and
this works more or less like a proximity sensor. The piezo buzzer used produces
sound that is a warning or indication to the user. It works on reverse process of the
piezoelectric effect. A playback voice recorder is used which speaks out the voice
once there is any obstacle detected. Suppose if the top-side sensor detects an obstacle,
then the recorder speaks out as “Top, Top,…”. This helps the user to detect obstacle
and change his path to reach his destination.
In [7], the authors have developed an intelligent walking stick especially for the
blind by using radio-frequency identification (RFID) technology, which tracks the
objects using electromagnetic fields that automatically identify the tags attached
to the object. This technology also uploads the person’s location data which helps
the family members of the user to locate the blind person and monitor his or her
movements. It is very difficult for the visually impaired people to travel alone in
unknown places as they do not have any information regarding their location. In
order to make their mobility easier, the different technologies like GSM and GPRS
are used for their navigation. This system uses GPS for outdoor navigation and GSM
or GPRS to send the information to their family member [8, 9].
The RFID tag is used for indoor navigation. The RFID sensor is installed on the
walking stick of the visually impaired. The tags provide a landmark to the blind
person. Each tag will be fed with the information of the direction and location so
that the current location is determined precisely. For the safety purpose, the RFID
tag is covered with a protective shield. The only drawback of using RFID could be
the cost-effectiveness. As two technologies are used for both indoor and outdoor
navigations, this is an effective solution for the blind [10, 11].
The working of the above solution is as follows. The obstacle is detected by the
sensor and distance is calculated; the distance is matched with the code, and then, the
buzzer gets activated. The GPS is interfaced with GSM model to locate the person’s
Smart Cane Design for Indoor and Outdoor Navigation: A Cost … 519

position; the location is updated in the server and sent to the android application
which gives the account of the travel route of the blind [12–15].

2 Prototype and Product: An Overview

Prototype can be called as “original form” of the end vision product. Prototypes are
the sample products or model built to replicate a design or test a concept learnt. Gen-
erally, a prototype is used to evaluate and enhance the design, accuracy and precision
of the product concept used. Prototype serves as a real-time working system provid-
ing the specifications rather than a theoretical one. Prototyping can be considered as
a bridge or a step between the formation and visualization of an idea or a concept
and the evaluation of the same. Thus, the process of prototyping can be called as
materialization. Prototype can also be called as the “primitive form” [16, 17].
Prototype is a proof of concept for a new idea or product that proves the viability
and illustrates the applications of the new technology. There are many types of
prototyping like electronics prototyping, software prototyping, data prototyping and
scale modeling [18, 19].
There are two types of prototype: alpha and beta prototypes. Alpha prototype
is used to check whether the product works as expected. It is a similar product in
size, shape, material and features as the final product will be after production. It is
a modeled product. Beta prototype is made to see the reliability of the product to
be manufactured. The prototype is checked through remaining bugs and mistakes
in the final product. This model product is given to customers for testing in the real
environment. Beta prototype is modeled with the actual parts, material and process
that are to be followed during production.
A product can be described as a design that is the final produce of a process
with proper efforts. It is a physical quantity that can be seen, touched, felt and
used. A product is a bundle of objects which consists of different features, benefits,
services, style, color, quality, grade and structure that fulfills the consumer needs
and requirements. A product has all the characteristics concerned to the customers.
Thus, the companies need to keep changing the product design which could be the
prototype for the new product that remains competitive in this rising consumerist
culture. The two classifications of a product are tangible and intangible. Tangible
product is referred as a physical object seen or touched. An intangible product can
only be felt like an insurance policy or a service.
A product is offered for sale but a prototype cannot be sold. The prototype is
just a basic proof of a design, but a product is added up with many more features
and factors such as manufacturing, packing, transportation, power source, external
appearance, color, style and also effectiveness. A prototype may fail to prove the idea,
but a product should be relevant and have an immediate use . It should functionally
520 Vandana Mansur and R. Sujatha

perform for what it is designed. Quality also plays an important role in the materials
used for the product. Products need to be marketed, wherein advertising and ‘brand
building’ come to entity. The product needs to stand out and a brand name could
make this possible.

3 Product Design Steps

Requirement analysis for the product design is done in this work. It is the societal
need for the visually challenged to lead a quality life. The basic requirement analysis
of a compact power supply, handy device, proper concealing of the controller unit
and aesthetics of the device is important.
There are six steps to be followed for the process of product design.
1. Idea generation;
2. Screening ideas;
3. Feasibility study;
4. Preliminary design;
5. Pilot runs and testing;
6. New product launch.
The idea generation is a process that begins with understanding the needs and
services of the customers. Ideas evolve from various sources both from inside and
outside of the field. It can be the employees, marketing team, research team or
sales force and reverse engineering. The external idea generation can be from the
consumers, environmental factors, technological strategies and other factors. The
product developers need to set a benchmark to find the best in the product that
meets the performance required. Reverse engineering is also important in inspecting
the competitor’s products and can incorporate new design features to improve the
product.
Screening of ideas is necessary for eliminating those that do not have high poten-
tial, so as to avoid the losses which could be incurred in future stages. The product
development committee sets certain criteria that need to be met in the product, and
each member gives a proposal supported by sample models, graphics, outline mod-
els and technical specifications. This helps to avoid the degree of overlap with the
products and services that already exist. For a better progress, each dimension of the
idea can be scaled on 0–10 and then proceed further.
Feasibility study is to analyze the market, economic and technical strategies.
The market survey gives an analysis whether the proposed product is in demand
by the customers so that the product making can be continued. If the demand is
met, then economic analysis is to be made which aims to establish a cost-effective
product in comparison with the present estimated sales volume. Finally, the technical
specification is concerned with technological viability of the product, production
and manufacturing, requirement of the chemical materials. If this analysis passes the
study, then product making is continued.
Smart Cane Design for Indoor and Outdoor Navigation: A Cost … 521

In preliminary design, the engineer builds a prototype followed by its testing and
retesting till the design meets the expected results. The physical appearance, size,
shape, color and style are taken care off. The brand identification, market image,
finishing, etc., will also be seen through. Pilot runs and testing before finalizing the
design adjustments are made in the product. To check the consumer acceptance of
the product, market testing is also carried out, which helps in launching the product
in the market.
Consistent production and marketing abilities are required by the company to sell
the product. The company should develop the confidence in their ability of production
and marketing as the volume increases and should work in a coordinated manner to
launch a product successfully.
Figure 1 shows the product design flowchart. In this one notice testing at each
stage is an important step. It is testing that ensures the design flow used is correct and
optimal. In industry scenario, 70% of the time is spent in testing. Planning the design
platform and deploying it play a significant role in achieving a working product.
During design planning stage, product specifications need to be arrived at. In this
work, an organized approach to obtain the design specifications is presented. This is
illustrated below.

Fig. 1 Product design

flowchart Idea generation

Screening Ideas

Feasibility study

Preliminary design
Testing and Retesting of prototypes
Design specifications

Pilot run and final testing

Final design approval

New product launch

522 Vandana Mansur and R. Sujatha

A series of questions assist in arriving at the product specification.

Customer-related questions:
What is the main objective of the product?
What is the user experience that needs to be translated as the product requirement?
What is the risk of the product not being acceptable by the customer?
Are there any device constraints from the user?
What is the voice of the customer?
What is the societal value associated with the product design?
What is the Wish list of the customer?
Hardware-related questions:
What are the hardware models to be developed?
What are the interface design units?
What are the components needed?
From whom should the electronic components be purchased?
To develop a sample design what are the mechanical parts needed?
What is the cost of the spares?
What is the computing unit that has to be used?
Which controller to be used for design deployment?
What is the voltage required for the design implementation?
Should the design have a battery as a power supply unit?
Should the design have 5 V power supply designed or use one which is available in
the market?
Should the design have a voltage regulator?
What is the type of the buzzer required for the audio output?
What are the tools like wire cutter, etc., that are needed?
Any specific safety measures to be used during the design and test phase?
Is there any special training needed for the designer to proceed further?

Embedded software questions:

What is the computational model?
Is the software available as open access?
What are the design modules to be developed?
What are the mathematical models to be implemented for proper computation and
simulation?
What is the hardware and software requirement together for the design flow?

4 Design Schedule

The design schedule followed for this work is depicted in the form of graphical
representation using Gantt chart. This representation helps the designer to work in
an organized way. If any task is not done in time, the chart acts as a reminder to
complete the task. Table 1 shows the Gantt chart for the work taken up.
Smart Cane Design for Indoor and Outdoor Navigation: A Cost … 523

Table 1 Gantt chart

Days 10 20 30 40 50 60 70 80
Brain storming
Selection of
sensor
Component
collection
Design Skill
learning
Design
development
Interface the
Circuit
Software
development
Testing

5 Smart Cane Design and Testing

The design commenced with the Wish list of the user.

1. Lightweight;
2. Clear audio output and good proximity response;
3. Less hassles with power/battery.
The product design needs to follow certain distinct approach. There are three key
words “Engineering, Design, Process”. The key words itself indicate a disciplined
approach to be followed when a product needs to be deployed. Dule Savic, Design
Engineer [20], Grundfos Company limited, Belgrade, Serbia, has shared his view in
FAQ’s Quora forum.
Process is a set of tasks to be followed which eventually give the required outcome.
Design is a representation of the appearance and functionality in the form of
block diagrams, flowcharts, interfacing features and the software design platform to
be implemented.
Engineering indicates that during the design phase the fundamentals of science
principles that are to be used in order to achieve the design goals. Science principles
may be related to physics, chemistry or mathematics.
From the above Wish list, it is observed that it provides information in a very
Layman’s way. It is the task of the designer to transform the Wish list into proper
design specifications. Analyzing the first Wish list of the designer, a “light weight
but yet sturdy cane” is the need of the user.
524 Vandana Mansur and R. Sujatha

Coming to the second one, it evidently indicates two design features.

a. Clear audio output
Now mapping this Wish list to a design specification, there can be the following
forms of audio output.
• Buzzer;
• Sound card;
• Voice response.
From these, the buzzer is finalized for the design of this work. The reason to choose
this is it provides a clear audible response without complicated design interface.
b. Good proximity response
To translate the designers’ understanding of this Wish list “Good Proximity Respon-
se”, it means that what is the design interface that is actually required to achieve this.
Here, it is the task of the designer to precisely meet this Wish list and to make sure
that the module is exactly as required—no more, no less. The questionnaire will be
of immense help to transform the Wish list to technical specification.
From this Wish list, the designer gets insight into two very fundamental technical
rules to be investigated. They are the physics of the device to be interfaced and the
mathematics involved in computing the response.
“Good Proximity Response” is normally the distance measurement. The word
proximity translates to distance in our design approach.

Distance = Speed∗ Time/2

The time for the transmission of the signal to the target and receiving it back is
taken into consideration.
There are two modes of sending information mainly,
a. Sound wave;
b. Light wave.
If sound wave is chosen, ultrasonic sensor is the choice. If light wave is chosen, IR
sensor is the choice. In this work, sound wave transmission is chosen for proximity
response. The speed of sound is 340 m/s. This gives the information of the sensor
to be used. This is added to the component list. A major review of this gives the
design unit needed. One can decide between a microcontroller and a processor. In
the present design, Uno ATmega is chosen as the processing unit.
Analyzing the mathematical model needed is extremely important. It assists the
designer in developing the software or the algorithm to be developed. In the present
design case, there is a clear functionality need to model the design for distance
measurement using appropriate software.
Tangential information from this is to freeze the software platform to be used in
the design. The mathematical model is used to solve a real-world system of interest.
It assists in understanding the system better through the mathematical model.
Smart Cane Design for Indoor and Outdoor Navigation: A Cost … 525

Building a mathematical model gives an insight into different computing aspects

related to the system. Solving a mathematical model and interpreting its solution in
the real-world situation are of great importance to simulate and predict the future
outcome. Mathematical model simplifies the problem by making certain assumptions
and taking them into design consideration. It provides a basis to describe, formulate,
simplify and simulate the problem statement. Mathematically describing features
can be done in many ways:
• Formulating equations;
• Defining variables;
• Data acquisition;
• Plotting graphs to find the nature of system;
• Probability calculation.
Simulating is a distinct task during product design phase. It is necessary to know
what the system output is considered into the decision making.
The mathematical model can be implemented here with the program development
in Arduino IDE using Embedded C. The hardware pins of the ultrasonic sensor are
invoked by the software commands, and the mathematical formula is implemented
in the code.
If the condition given for the buzzer is matched, then the buzzer gives the audio
output. Speed of sound is 340 m/s. Then 1/340 = 29.1. This knowledge is needed
for the simulation model to work.
Designing the Smart Cane:
The Arduino Uno microcontroller board is based on microchip ATmega 328P devel-
oped by [Link]. This board is very feasible for interfacing as it is based on
object-oriented programming in C language for code implementation. There are 14
digital pins and 6 analog pins. The controller is the brain of the system carrying out
the functions. The Arduino accepts input voltage 7–20 V, but the operating voltage
is 5 V.
This work is an integrated blind stick with ultrasonic sensor, ATmega Uno micro-
controller, a power bank for power supply on the stick to function and a buzzer to
give an indication of the obstacle. The ultrasonic sensor measures the distance of the
obstacle detected as per the range specified in the software code embedded onto the
Arduino Uno board. If the given condition of target detection is matched, then the
buzzer keeps giving an indication to the blind person. This is a simple mechanism
which is easy to handle and also cost-effective. There is a need to make sure that the
design provides all the functionalities with right interface and is also cost-effective.
The product is shown in Fig. 2. The basic feature of ultrasonic sensor is to measure
the distance of an object using ultrasonic waves. A transducer is used in the sensor
to send ultrasonic pulses and receive the echo. Ultrasonic sensor is cost-effective
and can be used in both indoor and outdoor environments. This can be referred as a
proximity sensor as it is a non-contact sensor. Ultrasonic sensor HC-SR04 is used in
this work. The measuring range is 2–400 cm with an accuracy of 3 mm. The voltage
and current rating are 5 V and 15 mA, respectively.
526 Vandana Mansur and R. Sujatha

Fig. 2 Smart cane

The main objective is to detect the obstacle. The Arduino and the ultrasonic sensor
are connected to the pins according to the pins assigned in the software.
The power management unit is a standalone power bank which normally the user
may already have. Multipurpose usage of this for the user’s mobile and navigation
stick saves cost. The cost of the stick can also be saved in the proposed product
design, if the user is already having his/her stick. Then, the electronic bay product
cost of the gadget will be the minimal cost to be borne by the user.
Another benefit of this product design is the height adjustment ability to detect
the obstacles. It can be adjusted to knee height for indoor navigation and above knee
height for outdoor navigation. The sound buzzer alert provides a quick reflex to the
person who is using this.

6 Conclusion

The smart cane appears as a user-friendly guide for the visually impaired people.
The system provides enhanced mobility and also ensures safety. It is a cost-effective
and feasible device. This paper throws light on the product design methodology. The
difference between product and prototype is discussed which helps the designer to
design an efficient product that has all the requirements and provides the expected
Smart Cane Design for Indoor and Outdoor Navigation: A Cost … 527

services. Documentation and questionnaire are important records in designing or

developing a product. The designer has to see through this questionnaire and find
solutions to these questions during product development.
Preparation of Gantt chart will help the developer to schedule the tasks accordingly
and follow up the work in a planned manner in order to complete it in the assigned
time frame. Learning the physics concepts involved in the working of the device is
required to know the output accuracy.
The knowledge of mathematical modeling is required for formulation and sim-
ulation of the software. Thus, this paper leverages an overall view on the product
design development, physics and mathematics involved in it.

References

1. Dhanuja R, Farhana F, Savitha G (2018) Smart blind stick using Arduino. Int Res J Eng Technol
(IRJET) 5(3):2–6
2. Kumar M, Kabir F, Roy S (2017) Low cost smart stick for blind and partially sighted people.
Int J Adv Eng Manag (IJOAEM) 2(3):65–68
3. Tekade A, Sonekar M, Ninave M, Dongre P (2018) Ultrasonic blind stick with GPS tracking
system. Int J Sci Eng Res (IJSER) 8(3):3–5
4. Badoni Manoj, Semwal Sunil (2011) Discrete distance and water pit indicator using AVR
ATmega8 in electronic travel aid for blind. Int J Disaster Recovery Bus Continuity 2:59–61
5. Gbenga DE, Shani AI, Adekunle AL (2017) Smart walking stick for visually impaired people
using ultrasonic sensors and Arduino. Int J Eng Technol (IJET) 9(5):3442–3447
6. Prasanthi G, Tejaswidho P (2016) Sensor assisted stick for the blind people. Trans Eng Sci
3:12–16
7. Kher Chaitrali S, Dabhade Yogita A, Kadam Snehal K, Dhamdhere Swati D, Deshpande Aarti
V (2015) An intelligent walking stick for the blind. Int J Eng Res General Sci 3(1)
8. Ambudkar B, Patil P, Bonage K, Gire B (2015) Voice navigation stick for blind. Int J Adv
Foundation Res Sci Eng (IJAFRSE) 1(9)
9. Gayathri G, Vishnupriya M, Nandhini R, Banupriya M (2014) Smart walking stick for visually
impaired. Int J Eng Comput Sci 3(3):4057–4061
10. Shanmugam M, John V, Gupta M, Saravanakumar K (2017) Smart stick for blind people. Int
J Trend Res Develop (IJTRD) 1:30–32
11. Sourab BS, D’souza S (2015) Design and implementation of mobility aid for blind people. Int
J Electronic Electr Eng 1:3–5
12. Lin CH, Cheng PH, Shen ST (2014) Real-time dangling objects sensing: a preliminary design
of mobile headset ancillary device for visual impaired. Int J Electronic Electr Eng 1:1–3
13. Sukhija Nitish, Taksali Shruti, Jain Mohit, Kumawat Rahul (2014) Smart stick for blind man.
Int J Electronic Electr Eng 7(6):631–638
14. Gaikwad D, Baje C, Kapale V, Ladage T (2017) Blind assist system. Int J Adv Res Comput
Commun Eng 6(3):442–444
15. Singh R, Succena A, Singh N (2016) PIR based walking stick. Int J Comput Sci Mobile Comput
5(4):486–490
16. Elverum CW, Welo T, Tronvoll S (2016) Prototyping in new product development: Strategy
considerations. In: 26th CIRP design conference sciencedirect. Elsevier, pp 117–122
17. Vaziri R, Mohsenzadeh M (2012) A questionaire based data quality methodology. Int J Database
Manag Syst (IJDMS). 4(2):28–31
18. Oliva R, Kallenberg R (2003) Managing the transition from products to services. Int J Serv Ind
Manag 14(2):1–5
528 Vandana Mansur and R. Sujatha

19. Parry G, Newnes L, Huang X (2011) Goods, products and services. Springer J 21–25
20. Savic D (2018) Difference between engineering, design and process. Design Engineer,
Grundfoss Company limited, Belegrade, Serbia, Quora forum, pp 1–2
Triclustering of Gene Expression
Microarray Data Using Coarse-Grained
Parallel Genetic Algorithm

Shubhankar Mohapatra, Moumita Sarkar, Anjali Mohapatra

and Bhawani Sankar Biswal

Abstract Microarray data analysis is one of the major area of research in the field
computational biology. Numerous techniques like clustering and biclustering are
often applied to microarray data to extract meaningful outcomes which play key
roles in practical healthcare affairs like disease identification, drug discovery, etc.
But these techniques become obsolete when time as an another factor is considered
for evaluation in such data. This problem motivates to use triclustering method on
gene expression 3D microarray data. In this article, a new methodology based on
coarse-grained parallel genetic approach is proposed to locate meaningful triclusters
in gene expression data. The outcomes are quite impressive as they are more effective
as compared to traditional state-of-the-art genetic approaches previously applied for
triclustering of 3D GCT microarray data.

Keywords Triclustering · Parallel genetic algorithms (PGAs) · Coarse-grained

PGAs (CgPGAs) · Mean-square residue (MSR) · Gene expression microarray data

1 Introduction

In microarray research, finding groups of genes exhibiting similar expressions, clus-

tering and biclustering techniques are more commonly used in gene expression anal-
ysis [9, 14]. However, these techniques become inefficient when the influence of the
time as a factor affects the behavior of expression profiles [7]. Now, these types of

S. Mohapatra (B) · M. Sarkar · A. Mohapatra · B. S. Biswal

DST-FIST Bioinformatics Laboratory, IIIT Bhubaneswar, Bhubaneswar, India
e-mail: B114042@[Link]
M. Sarkar
e-mail: B114066@[Link]
A. Mohapatra
e-mail: anjali@[Link]
B. S. Biswal
e-mail: c114002@[Link]
© Springer Nature Singapore Pte Ltd. 2020 529
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
530 S. Mohapatra et al.

Fig. 1 Illustration of a tricluster

longitudinal experiments are gaining interest in various areas of molecular activities

where the evaluation of time is essential. For example, in cell cycles, the evolution
of diseases or development at the molecular level is time based as they consider
time an important factor of evaluation [1]. Hence, triclustering appears to be a valu-
able mechanism as it allows evaluation of the expression profiles under a block of
conditions along with under a subset of time points.
A coherent tricluster is defined as a set of genes that pursues either coherent values
or behaviors. These clusters might have useful information that identify significant
phenotypes or potential genes relating to the phenotypes and their regulation relations
[17]. The computational complexity of triclustering algorithms is more expensive
than the biclustering algorithms (which are already NP hard), so heuristic-based
algorithms are an upstanding resemblance for triclustering (Fig. 1).
Genetic algorithms (GAs) are search-specific algorithms and are motivated by
the characteristics of genetics and natural selection [10]. GAs usually undergo
some important phases like reproduction, mutation, fitness evaluation and selec-
tion. Sequential GAs are competent in many applications as well as in different
domains. However, there exist some problems in their utilization of problems like
triclustering. For example, the fitness evaluation in sequential GAs is usually very
time-consuming. Also, sequential GAs may get trapped in a sub-optimal region of
the search space thus becoming unable to find better quality solutions. So parallel
GAs (PGAs) seem to be a better alternative to the traditional sequential GAs with the
adoption of parallelism. The static subpopulations with migration parallel GAs have
a key characteristic of applying multiple demes along with the presence of a migra-
tion operator. Coarse-grained parallel genetic algorithms (CgPGA) follow the same
general terms for a subpopulation model having a fairly small number of demes with
many individuals. Very often coarse-grained parallel GAs are treated as distributed
GAs as in general, their implementation is carried out on distributed memory MIMD
computers. This appeal can also be well configured with heterogeneous networks.
In this paper, an algorithm based on coarse-grained parallel genetic algorithms
(CgPGA) approach is proposed. This algorithm finds genus of similar patterns for
genes on a three-dimensional space, where genes, conditions and time factor are
taken into consideration.
The rest of this paper is organized as follows: A review of the literature is presented
in Sect. 2. The proposed methodologies along with the details of the fitness functions
Triclustering of Gene Expression Microarray Data Using … 531

and the genetic operators used are described in Sect. 3. The simulation results with
their GO term validation are discussed in Sect. 4.4. Finally, Sect. 5 presents the sum-
mary and the research findings of the proposed scheme and prospects for the future
work.

2 Related Work

Zhao and Zaki introduced tricluster algorithm in 2005 [21]. In this work, the patterns
are discovered in three-dimensional (3D) gene expression data along with a set
of matrices for the quality measure. A contemporary approach that finds coherent
triclusters which contain the regulatory relationships among the genes is stated in
[20] and subsequently, extract time-delayed clusters in [18].
LagMiner, in [19] introduced a new technique to detect time-lagged 3D clusters.
The evolutionary computation in the form of a multi-objective algorithm has also
been employed in the search for triclusters in [13]. Bhar Anirban et al. in 2012
presented δ-TRIMAX algorithm [2]. Again in 2013, the same authors applied the δ-
TRIMAX algorithm in estrogen-induced breast cancer cell datasets which provides
insights into breast cancer prognosis [3]. David et al. presented a novel tricluster
algorithm called as TriGen in 2013 [8]. The novelty of this TriGen algorithm lies
upon the use of the genetic approach to mine three-dimensional gene expression
microarray data. In 2015, Ayangleima et al. applied coarse-grained parallel genetic
algorithm (CgPGA) with migration technique to mine biclusters in gene expression
microarray data [12]. In the year 2016, Kakati et al. presented a fast gene expression
analysis that uses distributed triclustering and parallel biclustering approach [11].
In her work, the initial bicluster finding is performed by parallel or shared memory
approach and then the triclusters are extracted by a distributed or a shared-nothing
approach. Premalatha et al. in 2016 presented TrioCuckoo [16] which implemented
triclustering using the famous cuckoo search technique.

3 Proposed Methodology

In this section, the reported algorithm has experimented on the standard yeast cell-
cycle dataset (Saccharomyces cerevisiae) [15]. Then, the biological validation pro-
cess is initiated with a tool called GO Term Finder (version 0.83) [4] to get the
functional annotations of the genes resulted in the output tricluster.

3.1 Encoding of Individuals

Every individual in the population encodes a tricluster. Triclusters are represented

in the form binary strings of G + C + T length, G being the genes (rows), C being the
532 S. Mohapatra et al.

conditions (columns) and T being the times (height) of the 3D expression matrix.
If the bit in an individual is 1, it indicates that the respective row, column or height
have a place in the tricluster.

3.2 Fitness Function

Here, a fitness function has been implemented to select the best candidates, which is
conceptualized up on the three dimensions aspect of the mean square residue measure
(MSR) which has been an all-time effective biclustering measure for gene expression
analysis [5]. It is named as Fmsr now onward. As Fmsr is a minimization function, we
expect better results with smaller values.

Fmsr (TC ) = MSR − Weights − Distinction

The function is defined for every tricluster (TC). It is minimizing and thus, lower
values are favorable.

3.3 Weights

The weights term is defined as:

Weights = G l ∗ wg + Cl ∗ wc + Tl ∗ wt

where wg , wc and wt are weights for the number of genes, conditions and times in
a tricluster solution, respectively. High values of weights are favorable.

3.4 Distinction

The distinction term is defined as:

Distinction = CDNg /G l ∗ wdg + CDNc /Cl ∗ wdc + CDNt /Tl ∗ wdt .

where, CDNg (Co-ord Distinction no. of g), CDNc (Co-ord Distinction no. of c) and
C D Nt (Co-ord Distinction no. of t) are, respectively, the number of genes, conditions
and time coordinates in the tricluster that are absent in the tricluster being evaluated,
and wdt , wdg and wdc are the distinction weights of the genes, conditions and
times, respectively. Distinction is a measure for the uniqueness of the tricluster being
currently evaluated. With increased value of distinction, non-overlapping solutions
compared with results previously found can be found. Where,
Triclustering of Gene Expression Microarray Data Using … 533

• G: Tricluster gene coordinates subset.

• C Tricluster condition coordinates subset.
• T : Tricluster time coordinates subset.
• Tl : No. of time co-ord of the tricluster
• Cl : No. of condition co-ord of the tricluster.
• G l : No. of gene co-ord of the tricluster.
• TCv (t, g, c): Expression value of gene g under condition c at time t from the
expression matrix.

3.5 Tri-CgPGA

Tri-CgPGA is based on coarse-grained genetic algorithms which come under par-

allel genetic algorithm family. So like coarse-grained algorithms, this evolutionary
algorithm takes several steps to execute which are illustrated in the flowchart and
pseudo-code below (Fig. 2).
Algorithm 1: Tri-CgPGA Pseudo Code
Input: Expression Matrix
Output: Coherent Triclusters
1 Load the expression matrix
2 Specify the number of cores to be used in parallel
3 for tricluster number I =1 to maximum_triclusters do
4 Initialise the initial population
5 Evaluate the population
6 for generation number J=1 to maximum_generations do
7 selection of parents
8 crossover each parent to generate offsprings
9 mutation of generated offsprings
10 evaluate the new individuals
11 select the individuals with better fitness
12 if migration_interval =count then
13 choose the best individual of the best deem and replace with the
worst individual of the other parallel deems

14 select best individuals from all deems

15 select the final individual from best_indiduals
16 add final tricluster to output_set
17 return output_set
534 S. Mohapatra et al.

Fig. 2 Tri-CgPGA
Start
algorithm workflow

Load expression Matrix

Specify Number of Cores

Initialize Initial Population

Evaluation of Population

Genetic Operators

Is mi-
gration
interval =
Count ?

Yes

Choose the best individual and

replace with the worst individual of other parallel deems

Select the best individual from other deems

Select the final individuals from the best individuals

Final tricluster to the output set

Stop

4 Experimental Results and Discussions

All the computational simulations are performed in general conditions on a multi-

processor machine with four processors Intel Core i7 3.60 GHz with 4 GB RAM and
Windows 8.1 64 bit operating system memory. The yeast cell-cycle dataset (Sac-
charomyces cerevisiae) [15] is used for establishing the efficacy of the proposed
algorithm. This dataset contains 6179 genes, 4 conditions and 14 time points. The
experiment is performed on the above-mentioned dataset along with its two synthetic
versions but only reported for the former.
Triclustering of Gene Expression Microarray Data Using … 535

50 Generations 100 Generations

200 Generations 400 Generations

Fig. 3 Fitness value plots

4.1 List of the Parameters

During execution, some parameters have been set up like the crossover probability Pc ,
mutation probability Pm , weights: wg for genes, wc for conditions and wt for times,
distinction weights: wdg , wdc and wdt for genes, conditions and times, respectively.
The details of them are available in Table 1. As the algorithms are designed for gene
filtration (to obtain the solution with a minimum number of genes), the value of wg
is set to 0.8 so that maximum number of genes can participate in the solution. While
setting up the parameters for the distinction term, a higher value is being provided
for the genes to cover up as much space as possible in this dimension.

4.2 Results on Yeast Dataset

The simulation results are analyzed from the perspective of the different generations.
Analyzing across different generations, it indicates as the number of generations is
increased, the values also increase. So for bigger generations, better homogeneity
among the genes is obtained which is presented in the following graphs (Fig. 3)

Table 1 Values of the parameters taken during algorithm execution

Pc Pm wg wc wt wdg wdc wdt
0.8 0.5 0.8 0.1 0.1 1.0 0.0 0.0
536 S. Mohapatra et al.

Table 2 Detailed information about triclusters found by Tri-CgPGA algorithm

Gene size Avg. MSR Avg. volume Avg. No. of Avg. No. of Avg. No. of
genes conditions time
1000 493.35 5124.65 616 1.35 6.5
3000 1322.88 33,889.5 1651.37 3 7
6178 2669.086 67,798.75 3334 2.65 7.65

4.3 Comparitive Study

The results obtained from the execution of the algorithm are quite impressive in
terms of time and the volume of the output triclusters. As the fitness function is
minimizing, lower the value of MSR the better is the fitness of the tricluster. Further,
the results of the Tri-CgPGA algorithm is compared with the results obtained by the
trigen algorithm [8]. The comparison has been done on the basis of computational
time taken by the proposed algorithm to execute the codes and to derive the output.
In the case of Tri-CgPGA algorithm, it took 30 s approximately to run for 1000 genes
for 50 generations to deliver the output, whereas the trigen algorithm [8] requires
118 s to do the same. Hence, exploring parallelism with the genetic approach on
triclustering of gene expression microarray data is preferable against the traditional
GAs as it reduces the computation time for the algorithm execution. Other relevant
information regarding the results obtained from the algorithms Tri-CgPGA algorithm
is presented in Table 2.

4.4 GO Term Analysis

The validation of the results obtained is carried out with the Gene Ontology project
(GO) [6]. This analysis renders the ontology of terms which describes gene product
annotation data along with its characteristics. The ontology describes attributes like
molecular functions, cellular component and the relevant biological processes. The
queries associated with the associated genes are addressed in GO using the GO Term
Finder (version 0.83) [4]. The findings of the GO Term analysis are presented in
Table 3.
Triclustering of Gene Expression Microarray Data Using … 537

Table 3 GO for yeast cell-cycle results

Cluster ID Biological process Molecular function PI (P-value = < 0.01)
0044699 Single-organism process Only one organism is being involved 3.02E−10
0016043 Cellular component Assembling or de-assembling of a 4.87E−08
organization cellular component constituent parts
0065007 Biological regulation Biological process regulation of 0.00725
quality or function
0080090 Single-organism cellular Cellular-level activity, occurring 1.61E−06
process within a single organism
0060255 Single-organism process Only one organism is being involved 1.91E−06
0019222 Single-organism process Only one organism is being involved 0.00019
0044763 Single-organism cellular Cellular-level activity, occurring 2.69E−06
process within a single organism
0050789 Single-organism process Only one organism is being involved 4.10E−05
2000112 Single-organism cellular Cellular-level activity, occurring 4.89E−06
process within a single organism
0010556 Single-organism cellular Cellular-level activity, occurring 0.00315
process within a single organism
0071840 Cellular component Biosynthesis of constituent 0.00753
organization macromolecules, assembly,
arrangement of constituent parts, or
disassembly of a cellular component
0051171 Cellular component Biosynthesis of constituent 0.00026
organization or macromolecules, assembly,
biogenesis arrangement of constituent parts, or
disassembly of a cellular component
0006996 Organelle organization Cellular-level assembly, 6.00541
arrangement of constituent parts, or
disassembly of an organelle within a
cell
0010468 Organelle organization Cellular-level assembly, 0.00563
arrangement of constituent parts, or
disassembly of an organelle within a
cell
0032774 Biological regulation Biological process regulation of 0.00939
quality or function

5 Conclusion

A new framework Tri-CgPGA, based on the coarse-grained parallel genetic approach

(CgPGA) to generate the triclusters from gene expression database is proposed in our
work. The results of the suggested framework are compared with another state-of-the-
art technique called as Trigen algorithm. As the comparison justifies the proposed
scheme’s efficiency over the existing schemes considering the computation time,
538 S. Mohapatra et al.

hence it is preferable to adopt parallel GAs over traditional GAs in the triclustering
of gene expression 3D microarray data. There exist number of future directions which
might further improve this framework: (1) The acquisition of large-scale databases
from other standard datasets to measure the performance of the frameworks (2) To
further improve the coherence and the computation time, other competent evalu-
ation measures with the suggested or other existing versions of PGAs should be
investigated to obtain more meaningful triclusters.

References

1. Bar-Joseph Z (2004) Analyzing time series gene expression data. Bioinformatics 20(16):2493–
2503
2. Bhar A, Haubrock M, Mukhopadhyay A, Maulik U, Bandyopadhyay S, Wingender E (2012)
δ-TRIMAX: extracting triclusters and analysing coregulation in time series gene expression
data. In: International workshop on algorithms in bioinformatics. Springer, pp 165–177 (2012)
3. Bhar A, Haubrock M, Mukhopadhyay A, Maulik U, Bandyopadhyay S, Wingender E (2013)
Coexpression and coregulation analysis of time-series gene expression data in estrogen-induced
breast cancer cell. Algorithms Mol Biol 8(1):9
4. Boyle EI, Weng S, Gollub J, Jin H, Botstein D, Cherry JM, Sherlock G (2004) Go: termfinder-
open source software for accessing gene ontology information and finding significantly
enriched gene ontology terms associated with a list of genes. Bioinformatics 20(18):3710–
3715
5. Cheng Y, Church GM (2000) Biclustering of expression data. ISMB 8:93–103
6. Consortium GO (2004) The gene ontology (GO) database and informatics resource. Nucl Acids
Res 32(Suppl 1):D258–D261
7. Gómez-Vela F, Martínez-Álvarez F, Barranco CD, Díaz-Díaz N, Rodríguez-Baena DS, Aguilar-
Ruiz JS (2011) Pattern recognition in biological time series. In: Conference of the Spanish
association for artificial intelligence. Springer, pp 164–172
8. Gutiérrez-Avilés D, Rubio-Escudero C, Martínez-Álvarez F, Riquelme JC (2014) Trigen:
a genetic algorithm to mine triclusters in temporal gene expression data. Neurocomputing
132:42–53
9. Hartigan JA (1972) Direct clustering of a data matrix. J Am Stat Assoc 67(337):123–129
10. Holland J, Goldberg D (1989) Genetic algorithms in search, optimization and machine learning.
Addison-Wesley, MA
11. Kakati T, Ahmed HA, Bhattacharyya DK, Kalita JK (2016) A fast gene expression analysis
using parallel biclustering and distributed triclustering approach. In: Proceedings of the sec-
ond international conference on information and communication technology for competitive
strategies. ACM, p 122
12. Laishram A, Vipsita S (2015) Bi-clustering of gene expression microarray using coarse grained
parallel genetic algorithm (CGPGA) with migration. In: 2015 Annual IEEE India conference
(INDICON). IEEE, pp 1–6
13. Liu J, Li Z, Hu X, Chen Y (2008) Multi-objective evolutionary algorithm for mining 3D
clusters in gene-sample-time microarray data. In: IEEE international conference on granular
computing, 2008. GRC 2008. IEEE, pp 442–447
14. Rubio-Escudero C, Zwir I, et al (2008) Classification of gene expression profiles: comparison
of k-means and expectation maximization algorithms. In: Eighth international conference on
hybrid intelligent systems. IEEE, pp 831–836
15. Spellman PT, Sherlock G, Zhang MQ, Iyer VR, Anders K, Eisen MB, Brown PO, Botstein
D, Futcher B (1998) Comprehensive identification of cell cycle-regulated genes of the yeast
saccharomyces cerevisiae by microarray hybridization. Mol Biol Cell 9(12):3273–3297
Triclustering of Gene Expression Microarray Data Using … 539

16. Swathypriyadharsini P, Premalatha K (2018) Triocuckoo: a multi objective cuckoo search

algorithm for triclustering microarray gene expression data. J Inf Sci Eng 34(6):1617–1631
17. Tchagang AB, Phan S, Famili F, Shearer H, Fobert P, Huang Y, Zou J, Huang D, Cutler A, Liu
Z et al (2012) Mining biological information from 3d short time-series gene expression data:
the optricluster algorithm. BMC Bioinform 13(1):54
18. Wang G, Yin L, Zhao Y, Mao K (2010) Efficiently mining time-delayed gene expression
patterns. IEEE Trans Syst Man Cybern Part B Cybern 40(2):400–411
19. Xu X, Lu Y, Tan KL, Tung AK (2009) Finding time-lagged 3d clusters. In: IEEE 25th interna-
tional conference on data engineering, 2009. ICDE’09. IEEE, pp 445–456
20. Yin Y, Zhao Y, Zhang B, Wang G (2007) Mining time-shifting co-regulation patterns from
gene expression data. In: Advances in data and web management. Springer, pp 62–73
21. Zhao L, Zaki MJ (2005) Tricluster: an effective algorithm for mining coherent clusters in
3d microarray data. In: Proceedings of the 2005 ACM SIGMOD international conference on
Management of data. ACM, pp 694–705
Ensemble Feature Selection to Improve
Classification Accuracy in Human
Activity Recognition

Nivetha Gopalakrishnan, Venkatalakshmi Krishnan

and Vinodhini Gopalakrishnan

Abstract Real-time data with redundant and irrelevant features can degrade the
performance of the classifier. Dataset with more number of features also increases
the noise of the data and increases the time complexity of the learning algorithm.
Feature selection is a solution for such problems where there is a need to reduce the
dimensions of the data. In existing feature selection methods, the resultant feature
sets can lead to local optima in the space of feature subsets. In this paper, ensemble-
based feature selection approach is proposed to reduce size of the dataset and to
improve classification accuracy. Results show that the proposed ensemble approach
enhances the classifier performance, with reduced number of features.

Keywords Sensor · Feature · Classification · Learning

1 Introduction

Extracting useful knowledge from the raw sensed data, in general, has provided use-
ful information in various domains. Machine learning and data mining techniques
are effective than the classical mathematics and statistical techniques in extracting
knowledge from data. Data classification is the process of categorizing data into labels
or classes. The performance of the classifier model is described by the percentage of

N. Gopalakrishnan (B)
Department of Electronics and Communication Engineering, University College of Engineering
Panruti, Panruti, Tamilnadu, India
e-mail: [Link]@[Link]
V. Krishnan
Department of Electronics and Communication Engineering, University College of Engineering
Tindivanam, Tindivanam, Tamilnadu, India
e-mail: venkata_krish@[Link]
V. Gopalakrishnan
Department of Computer Science Engineering, Annamalai University, Chidambaram, Tamilnadu,
India
e-mail: vinodhini.g.t@[Link]

© Springer Nature Singapore Pte Ltd. 2020 541

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
542 N. Gopalakrishnan et al.

accuracy of correctly predicting the labels or classes. The significant specification

of the classifier model depends on the number of training data and the adequacy of
the features used in the analysis. Feature selection (FS) is defined as the process of
the selection of the features in the analysis in order to maximize the performance of
the resulting model. Feature selection is usually performed as a data pre-processing
stage before the classification process. Feature selection retains only useful features
or attributes via removing irrelevant features [1, 2]. This means that the full informa-
tion content can be obtained from minimum number of features in the dataset with
maximum distinction information about the classes. Hence, by eliminating the irrel-
evant features, the amount of data can be reduced which will improve the classifier
performance [3].
The rest of the paper is organized as follows: Sect. 2 outlines the literature survey
and motivation; Sect. 3 explains the proposed methodology and experimental setup;
Sect. 4 gives the experimental results obtained, and Sect. 5 concludes the paper with
possible future scope.

2 Literature Survey

In recent years, ensemble learning is applied to many data mining tasks such as outlier
detection [4], classification [5], and few ensemble-based feature selection methods
[6]. A hybrid evolutionary ensemble feature selection method with two variants
of genetic algorithms (GA) has been proposed [7]. In the first stage, the features or
attributes are selected using GA’s. The features selected by a feature selection method
are combined together in the second stage through averaging method. Another feature
selection method is proposed for breast cancer classification with ensemble size
of five feature selection methods such as CHI2 discretization, information gain, 1
rule, relief, and RMean methods, and feature subsets are aggregated using mean
function. Another method is proposed with ten feature selection methods such as
ReliefF, ReliefF-W, SVM-ONE, and SVM-RFE with three subset aggregators mean,
median, and exponential for biomarker discovery from high-dimensional genomic
data. Authors Guru et al. in [8] proposed ensemble of three FSM such as chi-square,
IG and likelihood ratio and three aggregation methods such as intersection, union,
and average-based aggregations for text classification. Another method was proposed
using ensemble of five feature selection methods like symmetric uncertainty, gain
ratio, information gain, relief and chi-square and rank-based combiner is used to
aggregate the feature subset.

3 Proposed Work

In this proposed work, we opt to use ensemble with variants in ensemble size. The
proposed ensemble uses different FSM with the same training data as shown in
Ensemble Feature Selection to Improve Classification Accuracy … 543

Fig. 1 Ensemble feature

FSM 1
selection

FSM 2
Data set Feature
subsets
FSM 3 Aggregator

FSM N

Fig. 1. In spite of the extensive number of FSMs available for feature selection, there
is no existing work investigating the advantages of the EFSM after outlier detection
and removal. With this drawback, in this proposed work, an analytical study of
alternative ways of ensemble feature selection methods resulted by multiple FSMs.
The proposed ensemble feature selection framework consists of two steps as shown
in Fig. 1. The first step is ensemble formation, and the second is the aggregation
of ensemble outputs. Moreover, as the main goals of ensemble feature selection
is to improve classification accuracy of feature selection. Feature aggregators like
intersection lead to very limited number of features sets, even leads to the null set of
features, and in practice, it does not tend to produce best results. On the contrary, the
union operator consists of combining those features subsets selected by weak feature
selectors. In this case, the final set of features subsets contains all the features that
had been considered important by any selector, leads to selection of all features in the
dataset. This approach tends to produce better results than the intersection, but at the
expense of a lower reduction in the number of the features. Statistical aggregators like
mean, median can also be used as an aggregation function. In this proposed approach
in Fig. 2, outlier-removed HAR dataset is applied to ‘M’ feature selection methods
(FSM), features selected from ‘M’ FSM are FS1, FS2, FS3, … FSM and these subsets
are combined using feature aggregator method. In this proposed approach, feature
subsets are aggregated using frequency of occurrence (FOO) of features or attributes.
Then the aggregated subset is ranked in descending order, top k attributes are labeled
as most significant relevant features (MSRF), and last ‘K’ attributes are labeled as
least significant labeled features (LSRF), and LSRF attributes are removed from the
dataset. Other than top K features, other features are to be analyzed. As depicted in
Fig. 3, divide all intermediate features into blocks of size S, as ‘B1, B2, and B3 …
BS.’ Add the attributes of first block B1 to MSF and perform classification and find
the classification accuracy. Next, add the attributes of block B2 with top k attributes
and B1 attributes and find classification accuracy and repeat this procedure for other
blocks. Finally, add attributes of last block BS with top k + B1 + B2 + B3 + ··· BS −
1 and perform classification. Now, sort all classification accuracies with feature block
544 N. Gopalakrishnan et al.

Start

Input: Outlier Removed Data set DS of m features and C classes;

Feature Selectors =N; Threshold Value = K; Output = Final Feature
Subset;

Find Feature sub Sets using ‘N’ Feature selectors and Aggregate all
subsets using frequency of occurrences to form ensemble features

Sort Rank aggregated features in descending order

Select top K ranking features as most relevant feature and Remove last K
features least relevant features.

Perform Intermediate Blocks Feature Selection (IBFS) for intermediate

features

Evaluation Metrics

Stop

Fig. 2 Workflow of the proposed ensemble feature selection (ENFS) method

Top K features Intermediate Features (B1, B2, B3, ….BS) Last K

Features
Select Most = Total number of features - Top K Remove Least
Significant features features - Last K Features Significant
(MSF) Features (LSF)

Fig. 3 Intermediate blocks feature selection (IBFS)

Ensemble Feature Selection to Improve Classification Accuracy … 545

Table 1 Ensemble
S. no. Ensemble size (N) Feature selection methods
combinations
1 2 IGR, MRMR
2 3 IGR, MRMR, PCA

sets and find the maximum classification accuracy. The set of blocks with maximum
classification accuracy is the final ensemble attributes.

4 Experimental Setup

To evaluate this proposed ENFS method, an experimental evaluation is carried

using human activity recognition datasets from offline and online environments. The
datasets used for this experimental evaluation are dataset-1 (DS1) collected from
offline environment, and the second dataset is the online dataset-2 (DS2). These
datasets are preprocessed for outlier detection and then passed to ensemble feature
selection (ENFS) for further processing.
1. MRMR: Peng et al. [9] proposed the mutual information-based method called,
minimum redundancy maximum relevance (MRMR). It selects features accord-
ing to the maximal statistical dependency criterion.
2. Information gain: The importance of an attribute relating to the class is evaluated
using the information gain measure [10].
3. PCA: This method calculates weight using principle components analysis [11].
We used the implementation of the feature selection algorithms provided by the
rapid miner software. The performance of the ensemble approach is evaluated using
support vector machine (SVM) classifier. In this study, the SVM classifier used a
Gaussian radial-basis-function (RBF) with values C = 1 and gamma = 0.01 (default
values for both parameters in rapid miner). In order to prove the efficiency of ensemble
feature selection methods, features selection with various ensemble sizes is taken.
First, we choose ensemble size (N) as two and combinations of two standard FS
methods IGR and MRMR are used. Feature selection is done individually using IGR
and MRMR, and the results obtained are aggregated using the proposed aggregation
algorithm. Finally, activity classification is done with ensemble feature subsets using
SVM classifier. The same procedure is repeated for ensemble size three using feature
selection algorithms as shown in Table 1.

5 Results and Discussion

In this section, we evaluated the performance of this proposed ensemble feature selec-
tion (ENFS) with existing three individual feature selection methods such as IGR,
546 N. Gopalakrishnan et al.

Table 2 Number of features

S. no. Feature selection-methods Number of features
selected
selected
DS1 (46) DS2 (46)
1. IGR 36 36
2. MRMR 36 36
3. PCA 36 35
4. ENFS2 29 29
5. ENFS3 29 28

MRMR, and PCA, method in terms of number of features selected and classification
accuracy.

5.1 Number of Features Selected

The total number of features selected using ensemble feature selection is shown in
Table 2. The total number of features selected by standard existing methods for DS1
are IGR = 36, MRMR = 36, and PCA = 36. Total number of features selected by
standard existing methods for DS2 are IGR = 36, MRMR = 36, and PCA = 35.
For ENFS = 2, the number of feature subsets obtained from two standard methods
IGR = 36 and MRMR = 36 are aggregated using frequency of occurrence, and the
final feature subset is selected using aggregation algorithm. The number of features
selected by ENFS2 is 29 for DS1 and 29 for DS2, whereas the number of features
selected by ensemble methods for DS1 and DS2 are ENFS3 = 29 and ENFS3 =
29. From the results shown in Table 2, the number of features selected by ensemble
methods is less when compared to other ten standard feature selection methods.

5.2 Classification Accuracy

In order to prove the efficiency of this proposed ensemble feature selection meth-
ods, classification accuracy is typically treated as the main goal. Feature subsets
selected using ensemble feature selection are classified using SVM classifier. The
classification accuracies to classify six human activities for the datasets DS1 and DS2
with outlier removal (WOR) are calculated. For DS1, the classification accuracy for
the three standard feature selection methods, and the five proposed ensemble feature
selection methods are shown in Table 3. The proposed ensemble methods ENFS2 and
ENFS3 have achieved higher classification accuracies than standard feature selection
methods for DS1 and DS2. It is clear that the ensemble methods with reduced num-
ber of features or attributes can achieve high-classification accuracies than existing
feature selection methods.
Ensemble Feature Selection to Improve Classification Accuracy … 547

Table 3 Classification
S. no. Feature selection-methods Accuracy
accuracy of DS1, DS2, and
DS3 DS1 DS2
1 IGR 92.8 92.6
2 MRMR 92.3 91.89
3 PCA 92.04 91.99
4 ENFS2 95.87 95.24
5 ENFS3 96.54 96.32

Thus from the experimental results obtained, the proposed (ENFS) ensemble fea-
ture selection methods improve classification accuracy of human activity recognition
(HAR).

6 Conclusion

In this paper, to increase the classification accuracy the irrelevant features of the
activity datasets are removed using ensemble feature selection method ENFS. Two
variants of the proposed ensemble methods such as ENFS2 and ENFS3 are proposed.
The ensemble combinations of IGR, MRMR, and PCA are used for analysis. The
proposed methods yield an optimal feature subsets. From the results obtained, the
number of features selected is less for the proposed ensemble methods. It can be
seen that the proposed ENFS methods achieves high-classification accuracy due to
elimination of irrelevant features. Therefore, the overall performance of the proposed
ENFS has been found to be excellent for the two datasets.

References

1. Chandrashekar G, Sahin F (2014) A survey on feature selection methods. Comput Electr Eng
40:16–28
2. Yaqub M, Javaid MK, Cooper C, Noble JA (2011) Improving the classification accuracy of the
classic RF method by intelligent feature selection and weighted voting of trees with application
to medical image segmentation. In: Suzuki K, Wang F, Shen D, Yan P (eds) Machine learning
in medical imaging. lecture notes in computer science. Springer, Berlin, Heidelberg, 7009
3. Quiroz JC, Banerjee A, Dascalu SM, Lau SL (2017) Feature selection for activity recognition
from smartphone accelerometer data feature selection for activity recognition from smartphone
accelerometer data. Intell Automation Soft Comput 1–9
4. Zimek A, Campello RJGB, Sander J (2014) Ensembles for unsupervised outlier detection.
ACM SIGKDD Explorations Newsletter 15:11–22. [Link]
5. Rokach L (2010) Ensemble based classifiers. Artif Intell Rev 33:1–39. [Link]
s10462-009-9124-7
6. Pérez RR, Silva AF (2015) Ensemble features selection method as tool for breast cancer
classification Isabel Ramos. Int J Image Mining 1:224–244
548 N. Gopalakrishnan et al.

7. Goh J, Thing VLL (2015) A hybrid evolutionary algorithm for feature and ensemble selection
in image tampering detection. Int J Electron Secur Digit Forensics 7:76–104
8. Guru DS, Suhil M, Pavithra SK, Priya GR (2018) Ensemble of feature selection methods
for text classification: an analytical study. In: Abraham A, Muhuri P, Muda A, Gandhi N
(eds) Intelligent systems design and applications 2017. Advances in Intelligent Systems and
Computing. Springer, 736, 337–349
9. Peng H, Long F, Ding C (2005) Feature selection based on mutual information criteria of max-
dependency, max-relevance, and min-redundancy. Pattern Anal Mach Intell 27:1226–1238.
[Link]
10. Lee C, Lee GG (2006) Information gain and divergence-based feature selection for machine
learning-based text categorization. Info Process Manag 42(1):155–165
11. Song F, Guo X, Mei D (2010) Feature selection using principal component analysis. In: Inter-
national conference on system science, engineering design and manufacturing informatization,
IEEE society, 27–30
An Exploration on Cloud Computing
Security Strategies and Issues

Amrita Raj and Rakesh Kumar

Abstract Cloud computing is revolutionizing many ecosystems by providing orga-

nization with computing resources that feature easy connectivity, deployment,
automation, and scalability. In the recent past, the attractive features of cloud com-
puting fuel the consolidation of cloud environment in the industry, which has been
consequently motivating research on the related technologies by both the academia
and industry. Regardless of its advantages, computing paradigm raises security con-
cerns in transition phase, which have subjected several studies. Cloud computing
tends to offer scalable on-demand services to the consumer with greater flexibility
and lesser infrastructure investment. Since cloud services are delivered using classi-
cal network protocols and formats over the Internet, implicit vulnerabilities existing
in these protocols as well as threats introduced by newer architectures have raised
many security and privacy concerns. In this paper, we have focused on the data
security issues found in the cloud computing. In addition to this, we discovered an
appropriate solution and a private cloud domain.

Keywords Cloud computing on-demand services · Virtualization · Security · Data

security · Threads · Attacks

1 Introduction

In the distributed computing framework implementation, the client has the features
of high adaptability and quality [1]. The assets on the demand and the client do not
know any information about the place of assets. Client can equip their application
and information from an unspecified area. The distributed computing is considered
as the valuable difference in data industry as well as a more effective improvement of

A. Raj (B) · R. Kumar

Department of Computer Science and Engineering (CSE), M.M.M. University of Technology,
Gorakhpur, U.P. 273010, India
e-mail: amritaraj4501@[Link]
R. Kumar
e-mail: rkiitr@[Link]

© Springer Nature Singapore Pte Ltd. 2020 549

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
550 A. Raj and R. Kumar

data technology for the general society. Most of the cloud computing framework now
consists of reliable services which have been sent through information center built on
the servers with various levels of virtualization technologies. The cloud computing
is the outcome of various factors like conventional computing, communication tech-
nology, and business approach [2]. The cloud computing can ensure the information
security and client will not secure the information without anyone else’s input [3].
So the distributed computing will be the process for storing information in the cloud
framework. Numerous organizations support the distributed computing stages, for
example, IBM, Amazon, Microsoft, Google, VMware, and EMC which involves its
common element that has been done by the cloud computing [4]. In spite of the fact
that distributed computing and their advantages are huge, security and protection
concerns are the essential obstructions toward their wide appropriation [5].

1.1 Measured Service

Although computing assets are combined and shared by more than one consumer
(i.e., multi-tenancy), cloud framework is to utilizing a suitable system for measuring
the usage of these assets for required person. The National Institute of Science and
Technology (NIST) had defined cloud computing as “Cloud computing is a model
for enabling ubiquitous, convenient, on demand network access to a shared pool of
configurable computing resources (e.g., networks, servers, storage, applications, and
services) that can be rapidly provisioned and released with minimal management
effort or service provider interaction.” The services given by the cloud are very
striking because of there intrinsic feature present in on-demand service. Due to this
feature, the client is required to pay to remunerate for service that they have used [6].

1.2 On-Demand Self-service

Consumer can individually calculate computing skills, such as server time and net-
work storage automatically, if needed, without the need for human communication
with all the service providers [7].

1.3 Broad Network Access

Capabilities are accessible over the network and access through the standard com-
ponent that promotes use by heterogeneous stages (e.g., cell phones, laptops, and
PDAs).
An Exploration on Cloud Computing Security Strategies and Issues 551

1.4 Resource Pooling

In the attempt of using multiplex consumers or virtualization, the cloud service

provider is pooled with computing resources; models are dynamically assigned with
defibrillators, resources and redistributed the need of user demand. In this way, the
motivation to establish a pool-based computing design is contained in the two-impact
factor.

1.5 Economics of Scale an Expertise

The outcome of a pool-based model is that the physical computing resources become
“invisible” to the customers, who do not normally have the information of controller
on the place, formation, and ownership of these resources.
Example: Databases, Central Processing Unit etc. It is easy to access the
information as it is stored in the cloud.

1.6 Rapid Elasticity

Computing resources for customers develop fast rather than constant: There is no
upfront and agreement because they can increase them as much as they want and
leave them after scaling them. Apart from this, the resources provision seems to
be infinitely mated for them, and consumption can increase rapidly to complete the
time-peak imports [9] (Table 1).

Table 1 Comparison of data

Parameter Data processing center
processing centers
Traditional Virtual Cloud based
On-demand service No No Yes
Wide network access Yes Yes Yes
Elasticity No Yes Yes
Measured pooling No Yes Yes
Resource pooling Yes Yes Yes
552 A. Raj and R. Kumar

2 Cloud Model Provides Three Types of Services

2.1 Software as a Service

The SaaS provides the customer a platform or application to manage a cloud infras-
tructure continuously. In general, SaaS allows to use software applications as a service
to end users. Example: Google Applications.

2.2 Platform as a Service

It is the ability that has been given to a customer to deploy onto the cloud infrastructure
his own applications without the need of installing any platform or tools on the local
machine. PaaS states towards providing platform layer resources, in addition to it
providing the operating system support and software development frameworks that
can be deployed to craft higher-level services [10].

2.3 Infrastructure as a Service

It provides the customer with the ability to plan prepare, storage, network, or dif-
ferent basic computing assets, or all pass the customer to deploy and fast arbitrary,
which can include software and applications. Furthermore, it enables the customer
to send and run subjective programming, which can incorporate working framework
and applications. Consumers have limited authorized over operating systems and
applications, storage deployment applications, and potentially selected networking
components [8]. Be that as it may, client can deal with his information put away
on cloud and applications which he has sent. Gmail and Dropbox are a few uses of
distributed computing administrations [11].

3 Attacks on Cloud Computing

3.1 Zombie Attacker

On the Internet, the aggressor challenges to increase the victim by transfer demands
from the honest host on the system in this way and these hosts are known as zombie.
In the cloud, the demand for virtual machines (VMs) is approachable by every client
by the Internet. An aggressor zombie may flow largest multiple calls. Such attacker
cloud software helps to improve the performance of the cloud.
An Exploration on Cloud Computing Security Strategies and Issues 553

In order to serve the large number of requests the cloud becomes overloaded and
finally becomed exhausted causing the DoS (Denial of Service) or DDoS (Distributed
Denial of Service) to the servers. The results of above becomes adverse as the cloud
is not able to serve valid user’s request due to the flooded attacker’s requests. Thus
better authentication and authorization & IDS/IPS can give better protection against
such attacks. In case of flooding or zombie attack, the cloud provider provides more
computing power to serve the huge number of requests (which includes zombie
requests too). Attacker Service can become a lead offs insufficiency of the attack
you are not able to do so, you cannot use any information about any service that is
used in the search result [12]. With the gadgets, these methods enable an aggressor
to screen the simple qualities of intensity supply and interface associations; along
these lines, they can be utilized to get to the chip surface straightforwardly, so we can
watch, control, and meddle with the gadget. Owing to these truths, employers are
extra relying on cloud-based information processing to achieve a large information
amount. The employers are ignorant of the strength data processing methods of the
cloud-based features and want that the existing network is protected and reliable
adequate to inhibit any unapproved approach to the information [8]. This type of
attack is DDOS attack. Thus the existing network must be protected & reliable
enough to prevent any unapproved accessing of the information [8] (Table 2).

3.2 Man-In-The-Middle Cryptographic Attack

This attack is complete when places himself between two clients. Whenever attackers
put themself in the information way, there is the feasibility that they can interrupt
changes information [13].

3.3 Side-Channel Attack

These methods enable an aggressor to screen the qualities of intensity supply and
interface associations; along these lines, they can be utilized to get to the chip surface
straightforwardly, so we can watch, control, and meddle.

3.4 Service Level Agreement

In numerous regards, cloud computing speaks to redistributing of calculation and

capacity to outside specialists co-op. Such redistribution has been administered ser-
vice level agreement (SLA) that indicates least dimensions of execution that the client
can anticipate. Although, classically there exists 99.99% system availability per year
yet SLA have not covered security aspects such as confidentiality and integrity.
554 A. Raj and R. Kumar

Table 2 Analysis on security problem and solutions directive

Synopsis of threads into cloud and solution instruction
Threads Issues Causes cloud Resolution
services instruction
Modification into Decline of modify IaaS, PaaS, or SaaS Access control or
economical class over cloud database checking framework
infrastructure on offered
framework administrations
Insulting utilization The intruder gives PaaS and SaaS Strong enrollment
of Cloud computing the signup or when and verification of
due to the strong comprehensive
attackers the absence monitoring of system
of approval, benefit traffic
minimum
Unsafe interfaces and Poses threads like SaaS, PaaS or IaaS Establish secure
APIS clear-content verification as well as
validation, provide modify
transformation to component among
content: advance coded
verification communication
Malicious insiders Insider malicious PaaS, IaaS, or SaaS Access clarity as
movement debate security and
from firewall and management system,
pass around usage deference
protection model broadcasting as well
as breath information
Shared technology Allow one client to IaaS For solid certification
problem interface other access control
client’s services by component and for
compromise managerial task.
hypervisor Inspection and
vulnerability as well
as structure
Information damage Confidentiality IaaS, PaaS, or SaaS Uses protected APIs
and leakage information can be encrypting, keys,
removed and apply information
modified detention, or
substitute policy
Service hijacking Customer file and PaaS, IaaS, and SaaS Uses security
service instances strategies, solid
cases could thus validation
make another new mechanism, or
base for attacker movement
monitoring
(continued)
An Exploration on Cloud Computing Security Strategies and Issues 555

Table 2 (continued)
Synopsis of threads into cloud and solution instruction
Risk profiling Interior protection SaaS, IaaS, or PaaS Uncover incomplete
strategies, security logs, information and
compliance, infrastructure detail.
structure, solidifying Use observing and
fixing inspecting or alerting framework
log might be ignored for information
breaks

In a cloud computing seller market, it is sensible to expect that not all suppliers
will be capable, or willing, to give some level of security to their customers.
Besides, a given cloud supplier may offer administrations with differing dimension
of security relying upon how much the client will pay for the administration [14].
In spite of the fact that cloud consumer does not command over the fundamen-
tal computing assets, they do need to guarantee the assets when purchasers I have
related their center business capacities onto their depended cloud. In other words,
it is important for client to get provides on service delivery. Normally, these are
given through service level agreement (SLA) consulted between suppliers and pur-
chasers. The first issue is the definition of SLA specifications in such a way that
has an appropriate level of granularity is the tradeoffs between expressiveness and
complicatedness such that they can cover most of the consumer desires and relatively
simple to be verified and evaluated. Also the different cloud offerings (IaaS, PaaS,
SaaS & DaaS) will certainly need to define different SLA meta-specifications.
This likewise raises various usage issues for the cloud provider. For instance, assets
administrator needs to have exact and restored data on the assets use at specific
time inside the cloud. By refreshed data, we mean any adjustment subscribed to
by. The assets administrator is in continuous assessment and modification for SLA
fulfillment. The assets administrator needs to utilize quick useful choice model and
streamlining algorithm to do. SLAs cannot be completed when resource requests may
be required to be dismissed. All of these need to be done complete “self- service”
in the cloud computing. Apart from this, there is a need to consistently include user
feedback and evaluation features in advanced SLA evaluation framework [9].

3.5 Application-Level Security

The application level security refers to the usage of software and hardware resources
for providing security to the applications such that the attacker is not able to get the
unauthorised access to the application and make the desirable changes to its format.
Now, some day the attack begins and it tries to access as the trusted user and it easily
allows the full access to the attacker and this makes the client suffer. The major
reason behind this is; old network level security policies. With the latest technological
556 A. Raj and R. Kumar

progress, it is quite possible to duplicate a trusted user and contaminate the entire
data without seeing it. Subsequently, it is important to introduce large amount of
security checks to limit these dangers.
Traditional ways to deal with increased security problem have been developed to
develop a work-oriented ASIC device that can handle the specific work that provides
higher level of security with greater performance. Be that as it may, with application-
level dangers being dynamic these shut frameworks have been seen to case back in
contrast with the open finished framework.
The abilities of a shut framework and additionally the flexibility of an open finished
framework have been consolidated to build up the security stages dependent on Check
Point Open Performance Architecture utilizing Quad-Core Intel Xeon processors.
Indeed, even in the virtual condition, organizations like VMware and so forth are
utilizing Intel Virtualization Technology for better execution and security base. It has
been seen that all the more frequently sites are anchored at the system level and have
solid safety efforts; however, there might be security provisos at the application level
which may permit data access to unapproved clients. The dangers to application-
level security incorporate XSS assaults, cookie poisoning, hidden field control, SQL
infusion assaults, DoS assaults, backdoor and debug options, CAPTCHA breaking,
and so forth coming about because of the unapproved utilization of the applications
[12].

3.6 Data Security

Data Security is the prime concern for any technology, but it still remains as a major
challenge when SaaS users have to depend upon their providers for proper security
[15].
The organisational data is often processed in plaintext & stored in the cloud in
SaaS. Moreover the data backup becomes a prime topmost aspect in order to carry out
recovery in case of any disaster but this imparts security questions as well [16]. Rotat-
ing administration applications & databases having sensitive data about the Cloud
Service Provider (CSP) which have no controller of their own information have vari-
ous weaknesses [17]. Numerous clients have weaknesses in the information security
model and this increases an unauthorized access to information. The following valu-
ations validate the security of the enterprise that collects the information at the SaaS
vendor [15]:
• Cross-site scripting [XSS].
• Access control weaknesses.
• OS and SQL injection flaws.
• Cross-site request forgery [CSRF].
• Cookie manipulation.
• Hidden field manipulation.
• Insecure storage.
An Exploration on Cloud Computing Security Strategies and Issues 557

• Insecure configuration.
In SaaS, organizational information is often managed in plaintext and stored in
the cloud. The SaaS supplier is responsible for the security of the information in the
way it is processed and stored. Additionally, data backup is very important feature
for ceasing the recovery cause of disaster; however, it brings security burden as well
[16].

3.7 Security Concerns with the Hypervisor

Distributed computing lays essentially on the idea of virtualization. In a virtualized

world, the hypervisor is characterized as a controller prominently known as virtual
machine director (VMM) that enables various working systems to be kept running on
a system at any given moment, giving the assets to each working system to such an
extent that they do not meddle with each other. As the digit of operating systems going
on the hardware unit increases, however, the problem of security increases, the need
to consider the new operating system. Since different working frameworks would
keep running on a solitary equipment stage, it is beyond the realm of imagination to
expect to monitor all, and consequently, keeping all the working frameworks secure
is troublesome. It might happen that a visitor framework attempts to path a malignant
encryption on the feaster framework and fetch the framework low or take complete
control of the framework and square access to other company working frameworks.
It cannot be denied that there are dangers related with having the equivalent phys-
ical foundation among a lot of numerous employers; even one existence vindictive
can make dangers the other utilizing the equivalent organization, and consequently,
security as for hypervisor is of extraordinary worry as all the visitor frameworks are
controlled by it. On the off chance that a programmer can deal with the hypervisor,
he can make changes to any of the visitor working frameworks and oversee every
one of the information going through the hypervisor.
Different kinds of attacks can be propelled by focusing on various parts of the
hypervisor. In the hypervisor architecture, an advanced cloud security system can be
developed on the basis of the behavior of different components, which monitors the
interval between guest VM activities and various hybrids.

3.8 Virtualization Level Security Issues

Of cloud runs simultaneously on the host computer applying OSs in the virtual-
ized (multi-inhabitant) condition. Actual weaknesses in a VM that are circulated all
through the physical and simulated venture assets permit digital attacker, malware,
or different dangers to remotely misuse.
558 A. Raj and R. Kumar

Fig. 1 Structure of TCPS

The security danger also increases with the help of VMS. As soon as the amount
of visitor operating framework increases in the hypervisor, security concerns with
that new visitor OS also increases. Since it is impossible to hope to screen all OSs,
keeping up the security of those OSs is troublesome. There can be chances that a
visitor framework attempt to execute harmful code upon the host structure as well
as cleave lower hold the full control of a framework plus square approach to further
visitor OSs. There are dangers related to having equivalent real framework among
lot of various clients yet one being malevolent. On the off chance that a hacker can
deal with hypervisor is that they can make changes to any of the visitor OSs and
deal with every one of the information going through hypervisor. Isolation between
two VMs is not completely adequate by current virtual machine monitors (VMMs).
By compromising the lower layer hypervisor vulnerabilities, the attacker can get
access over installed VMs. Example of some attacks include Blue pill, Subvert &
DKSM on the virtual layer. This is still an open challenge to prevent such threats.
Latest Generation of rootkits that benefit from the processor technology that allows
an attacker to insert an additional hypervisor between the hardware and the software.
The hypervisor takes control of the system & transforms the original OS into a virtual
guest on the fly. As regards the software based virtualization is considered, this kind
of hijacking does not require restart & this makes it all the way more difficult to
detect the intrusion [11].

3.9 Sharing of VM images in Cloud Introduces Security Risk

The proprietor of a picture is worried about privacy (e.g., unapproved access to

the picture). The client of picture is worried about privacy, for example, a malig-
nant picture that is able to debasing or theft the clients own individual information.
For sample, instances working on Amazon’s EC2 platform simply compromise by
performing different attack, related the mark wrapping attack short scripting (XSS)
attack, or DOS attack. This enables attackers to make, modify, and erase VM pictures
and change administrative passwords and settings that are put into setting with EC2
for S3 get to. This is a threat of infringement (e.g., working unrestricted software
An Exploration on Cloud Computing Security Strategies and Issues 559

and programming with expired licenses). The manager of cloud is worried about the
security and consistency of the cloud all in all and the trustworthiness of the pictures.
There is a danger of harms caused by malware contained in any picture put away in
the repository.
There ought to be standard machine for finding out integrity of visitor VMs for
effective capability or avoid interface of computing, information damage, and abuse
of assets.
Figure 1 depicts Manager based clear cloud protection system (TCPS) that screens
honesty of cloud parts. TCPS is put among visitor’s OS and the virtualization layers,
which screens visitor VMs and secures them opposite to interlopers and assaults. It
also addresses clear issues in the cloud [11].

4 Cloud Security Problems

The cloud framework is functioning in online network, and the security issues in
online network also can be detected in the cloud classification. The cloud structure
is not different from the traditional structure in the PC, and it can fit other impor-
tant and unique security issues. The large outfit around cloud computing remains
security and privacy [18]. The information resources the cloud classification. The
cloud requirement provides information control framework for the employer. The
information security analysis too may be deployed in the cloud system. The cloud
framework can deploy in various cloud bud. The different area has various rules, so
the security managing can face the rule problem. Distributed computing service is
necessary to enhance permissible security [19]. It has appeared as a major issue for
online network clients, and they have to face the problem of managing large multiple
of records and identification, which helps the clients using password and identifi-
cation management scheme that decreases security of their personal information.
Besides, Web site-driven web experience issues in managing Internet client record
and individual content allocation [18, 19].
For large and fast processing of the information, a CSP may utilize assets that are
accessible around the activity. This factor reveals the client information over whole
net which may result in major security menace. To fix this problem, an intrusion
finding system (IDS) component is mostly in the cloud paradigm [20].

4.1 Trust Chain in Clouds

Trust has a significant influence in pulling in more customer by depending on cloud

players, due to loss of control (as discussed earlier) cloud costumers depend upon the
cloud providers using trust system as on alternate into giving customers transparent
power their information and cloud resources (Fig. 2).
560 A. Raj and R. Kumar

Fig. 2 Management layer

Therefore, the cloud provider assures the client that the operation of the provider
is following organizational safety measures and standards.
Concern of identification management includes various measurements; amid oth-
ers, they consist of clients having to share their characteristics to many service
providers, dealing with individual data of the client being conciliated while execut-
ing a federated identity. Rodriguez et al. exhibited Federated Identity Architecture
(FIA) as a method for resolving weakness and there are three designs for enforcing
security problem in FIA including WS-Federation, Shibboleth, and Liberty Alliance
[21]. Other than attack, permissible compliance and protection guaranty are other
softy matter in federated identity. The recent FIA has no decent way to protect client’s
data. Stage for Privacy Perform.33 acnes Project or P3P (founded by W3C) has been
introduced as a standard and an assignment for developing FIA, by integrating P3P
into the FIA [22]. Network to human resources (HR) is hard because HR is the only
master source for team identification. Ability to succeed federated identities does not
exist in maximum administrations. There are no authoritative data sources to find
identities in partner associations. Most associations have no capability to convey
with individuals directly in extra institution. These problems and the need of provi-
sioning standards emphasize the importance for decent and comprehensive method
to manage how identity properties, accounts, and development management system
of all entity-types will take action in the cloud co-framework [23] (Fig. 3).

5 Conclusion Direction

This paper explains the security problem of the cloud computing such as low-cost,
platform, independent scalability, elasticity, as well as reliability. The security cloud
An Exploration on Cloud Computing Security Strategies and Issues 561

Fig. 3 You and Jun

proposed mode [23]

computing manages various field of information management along with its services.
The problems included data security in cloud problem also discuss problems in
the cloud system are discussed. The cloud computing are on the accelerated pace
in their development which have good prospect along with great potential. In this
script presented the number of attacks cloud authentication although our main in
used to magnified the theft concern because some of the issues are partially solved
but identity theft requires further thought. The various customers which degrade
mistrust as well as privacy of cloud computing which do not want to move the data
into Cloud computing. The various method which are used to protect the security in
order to make it effectively or solved this problem are issues are check by the cloud
computing provide. Developing the cloud computing as well as security issues is the
core problem.

References

1. AmazonElastic Compute Cloud, [Link] National Institute of Standards

and Technology. The NIST definition of cloud computing; 2011. [retrieved 14.04.11]
2. Google App Engine, [Link] Google AppEngine, [Link]
[Link]
3. Microsoft, [Link]
4. Ren K, Wang C, Wang Q (2012) Security challenges for the public cloud. IEEE Internet Comput
16(1):69–73
5. Vieira K, Schulter A, Westphall C, Westphall C (1989) Intrusion detection techniques for
Grid and Cloud computing environment. IT Professional 12(4):38–43. Young M (1989) The
technical writer’s handbook. University Science, Mill Valley, CA
6. Mell P, Grance T (2011) The NIST definition of cloud computing
7. Zissis D, Lekkas D (2012) Addressing cloud computing security issues. Fut Gener Comput
Syst 28(3):583–592
8. Dillon T, Wu C, Chang E (2010) Cloud computing: issues and challenges. In: 2010 24th IEEE
international conference on advanced information networking and applications (AINA). IEEE,
pp 27–33
9. Hashizume K, Rosado DG, Fernández-Medina, E, Fernandez EB An analysis of security issues
for cloud
10. Veeramachaneni VK (2015) Security issues and countermeasures in cloud computing environ-
ment. Int J Eng Sci Innovative Technol 4(5)
562 A. Raj and R. Kumar

11. Modi C, Patel D, Borisaniya B, Patel A, Rajarajan M (2013) A survey on security issues and
solutions at different layers of Cloud computing. J Super Comput 63(2):561–592
12. Singh A, Shrivastava DM (2012) Overview of attacks on cloud computing. Int J Eng Innovative
Technol (IJEIT) 1(4)
13. Rong C, Nguyen ST, Jaatun MG (2013) Beyond lightning: a survey on security challenges in
cloud computing. Comput Electr Eng 39(1):47–54
14. Bhadauria R, Chaki R, Chaki N, Sanyal S (2011) A survey on security issues in cloud computing.
arXiv preprint arXiv:1109.5388
15. Hashizume K, Rosado DG, Fernández-Medina E, Fernandez EB (2013) An analysis of security
issues for cloud computing. J Internet Serv Appl 4(1):5
16. Kumar SN, Vajpayee A (2016) A survey on secure cloud: security and privacy in cloud
computing. Am J Syst Software 4(1):14–26
17. Subashini S, Kavitha V (2011) A survey on security issues in service delivery models of cloud
computing. J Netw Comput Appl 34(1):1–11
18. Cloud Securit Alliance: [Link]
19. Dean J, Ghemawat S (2008) MapReduce: simplified data processing on large clusters. Commun
ACM 51(1):107–113
20. Sun ST, Pospisil E, Muslukhov I, Dindar N, Hawkey K, Beznosov K (2011) What makes users
refuse web single sign-on?: an empirical investigation of OpenID, p. 4
21. Gharooni M, Zamani M, Mansourizadeh M, Abdullah S (2011) A confidential RFID model to
prevent unauthorized access. pp 1–5
22. Rodriguez UF, Laurent-Maknavicius M, Incera-Dieguez J (2006) Federated identity architec-
tures
23. Archer DCJ, Puhlmann N, Boehme A, Kurtz P, Reavis J (2011) Security guidance for critical
areas of focus in cloud computing v3.0. Cloud Secur Alliance
A Hybrid Approach-Based Energy
Aware Cluster Head Selection for IOT
Application

A. Kannammal and S. Suresh

Abstract Internet of Things environment has collection of communication sys-

tem of various devices with few heterogeneous and homogeneous characteristics,
where each device has unique identification address for activate action over network.
The main purpose of gathering information is to process the communications and
evolution. IOT applications face several challenges such as energy efficiency, reli-
able communication, latency awareness, etc. The most emerging requirement in IOT
application is to implement an effective energy conscious communication protocols
and clustering techniques. These techniques are reducing energy utilization of nodes
and increase the network lifetime and scalability. A robust clustering technique is
essential for self-organizing sensor networks. Proposed work on this paper is hybrid
method of LEACH with firefly technique for optimal cluster head selection. The pro-
posed hybridization methods used for best clustering with cluster head also enhance
the energy level of node and improved lifetime of networks. Simulation results have
shown the improved performance of proposed task.

Keywords WSNs · IOT · Energy-aware clustering · LEACH · Firefly algorithm ·

Hybrid approach

1 Introduction

During last decade, the research community focuses on the development of smart sen-
sors networks with small size, less expensive [1]. The current research works revival
in WSN and Internet of Things (IOTs). Internet of Things was first proposed for

A. Kannammal (B)
Department of Computer Science and Engineering, Jayam College of Engineering and
Technology, Dharmapuri, TamilNadu, India
e-mail: [Link]@[Link]
S. Suresh
Department of Computer Science and Engineering, P.a College of Engineering and Technology,
Pollachi, Coimbatore, TamilNadu, India
e-mail: [Link].72@[Link]

© Springer Nature Singapore Pte Ltd. 2020 563

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
564 A. Kannammal and S. Suresh

the purpose of automatically recognizing and tracing the flow of goods. Nowadays,
IOT can be developed in current Internet environment to connecting wide variety
things to interoperate. An IOT integrating the current emerging research technolo-
gies such as sensing, communication, networking, and cloud computing. An IOT
application faces the many challenges on today research work. An IOT environment
offers quick deployable solution with low cost in various applications. All nodes in
IOT architecture have the following components such as sensor to sense and monitors
the environment, computation unit for processing information, memory unit for data
storage, transmission unit for transmitting these data to a base station and a power
management unit for managing lifetime of battery.
IOT networks can be categorized with following constraints factors such as a
amount of energy consumed; computing and storage capacity; communication range
and bandwidth also some problems in networks architectures like security and fault
tolerance, etc. [2]. In IOT application, the energy utilization relay in the following
criteria such as design management on network and deployed environments. An IOT-
network can be adapted in different applications like environment observing, army
sector, healthcare system, etc. [3, 4].
The researches in IOT application have several challenges remain open; the chal-
lenges are energy level management, area of coverage and synchronization data and
security [5]. An energy-conscious communication among all the devices is the main
challenge. These articles focus on these challenges by selecting head node using
clustering technique. Clustering techniques can group sensors into different subset.
In each individual cluster, a CH is chosen as group coordinators for the purpose of
producing a transmission schedule, gathering sensors data, and transmitting assem-
bled data to the base station (BS). Cluster head can communicate with the base
station directly or through other CHs called intra-cluster routing. The network life-
time is essential in IOT research, lifetime of network can be enhanced with effective
energy-aware mechanism [6–8].
This paper reviews the energy-aware technique and dispute factors of clustering
in IOT systems. The rest of this work can be described as follows. Section 2 gives
overview of some of cluster head selection algorithm. Section 3 describes some of
the preliminary work such as protocol and network model, and the proposed works
are described in Sect. 3.1. Section 4 shows the simulation results. The proposed work
is concluded in Sect. 5.

2 Related Work

Researches in IOT mainly focus on developing clustering protocols for energy-saving

purpose. Younis et al. [9] describe hybrid energy efficient distributed (HEED) tech-
nique for clustering IOT architecture. This protocol mainly focuses on balancing
load among the group. The key parameters to decide on cluster head are residual
energy of respective node, degree of node, and distances to neighbors. Heinzelman
et al. [10] describe EEDUC protocol gives a way to producing distributed cluster.
A Hybrid Approach-Based Energy Aware Cluster Head Selection … 565

In this scheme, the grouping of node done based on distance between node and clus-
ter head, distance within a certain unit of cluster head can be groped in same cluster
and node with m units of distance from the cluster head may consider other cluster.
Yang et al. [11] implemented most effective optimal cluster head methods by using
various hybrid meta-heuristic algorithms. This scheme has to focus global optimum
with improved solutions. From literature GA has good global search characteristics,
convergence is poor. Representation of weights in PSO is done arbitrarily and hence
search is limited to either global or local space. In this work, it is proposed to hybrid
the firefly meta-heuristic with LEACH, which finds optimal global solution with fast
convergence.
Faisal et al. [12] SEP protocols mainly focus on heterogeneous architectures.
SEP selects cluster heads based on the probabilities that is weighted on each node.
Zonal-Stable Election Protocol (Z-SEP) which impart zonal heterogeneous, it is more
applicable in IOT multi-region real environment.

3 Protocol Description

3.1 Low Energy Adaptive Clustering Hierarchy

The LEACH protocol [13–15] was the most frequently adapted protocol. This pro-
tocol mainly focus on homogeneous sensor networks because of all the nodes have
equal energy. In this method, the head nodes are selected depend on system-defined
probability values. The cluster head then processes the data and communicates it with
the base station. This technique includes two-step processes such as setup phase and
steady state phase. Main task of setup phase is nominating CH also clusters are
formed based on chosen CH. In steady phase information transmission performed.
A random number will be chosen in range from 0 to 1 for each node. The node has
random number which is below its threshold value can be selected as cluster head,
the random value greater than threshold can be consider as normal node
⎡ ⎤
ρ
t (n) = ⎣    ⎦ , i f n N else t (n) = 0 (1)
1 − ρ Rmod ρ1

where the value of ρ stands for CH probability, R describes the current round number
and N stands group of normal node rather than cluster heads in the last1/ ρ rounds.
Clustering can be framed based on energy, the energy of available CH can be matched
with other actual node, if any non-CH node has higher value than current CH, then the
non-CH node can be chosen as new CH. The transmission energy can be calculated
using first order radio model. A new optimal scheme projected as hybridization of
LEACH and firefly methods to diminish energy utilization and enhance the network
lifetime.
566 A. Kannammal and S. Suresh

3.2 Firefly Algorithm

Firefly heuristic is based on the light intensity produced by fireflies. The intensity
of light produced is mapped to the objective function, and hence, fireflies with low
intensity are attracted toward fireflies with higher light intensity. The intensity of the
light helps a firefly swarm move to brighter and attractive locations which can be
mapped to finding best [16].
Firefly algorithm standardizes based on firefly characteristics and can be listed as
follows:
• Each firefly can be attracted to another irrespective of their sex.
• The brightness produced by the firefly is directly proportional to its attractiveness
and between two fireflies; the firefly with higher brightness attracts the one which
has lower brightness. A firefly moves randomly if it is not able to find a brighter
neighboring firefly.
• In the mathematical model, firefly’s brightness is based on the objective function.
Firefly meta-heuristic is chosen for its capability of providing optimal solutions
for multi-objective problems.

3.3 Firefly Procedure

1. Define an objective function g(A), A = (A1 , A2 , … AD )

2. Initializes fireflies Ai (i = 1, 2, …, N)
3. Calculate light intensity I i at Ai is using objective function g(Ai )
4. Compute light absorption coefficient γ
5. if t < M
i. for x = 1: N
ii. for y = 1: i
iii. if (I y > I x ) then
iv. Reassign firefly x: y
End if
6. Based on distance, the attractive of brightness varies r via exp[− γ r]
7. Calculate new solutions then revise light intensity
Scope of loop y and x
8. Pick best rank solution
9. Repeat step from 5 to 8 until the condition becomes false.
A Hybrid Approach-Based Energy Aware Cluster Head Selection … 567

4 Proposed Hybrid Algorithm

This proposed scheme is used hybrid-based approach of LEACH and firefly search.
Enhancing lifetime and reducing energy conception of an IOT node by clustering
the network with suitable methods. The clustering process can be done in IOT two
steps such as CH election and cluster formation.
i. CH Election:
Cluster head is acting as coordinator of cluster, while selecting CH suspicious
analysis needs. The CH can be selected using three ways such as deterministic
with help of preset fixed location, random scheme based on probability value
and adaptive based on energy. The last method is more preparable.
ii. Cluster Formation:
Cluster head broadcasting own information to all other sensors, each individ-
ual sensor selects its own CH based on hop count, distance, and size of cluster.
LEACH protocol gives better result in the initial steps then followed iteration
the firefly methods execute, thus enhance the network energy level and lifetime.
The evaluation of the proposed work was carried using MATLAB. Simulations
were carried out using LEACH, firefly. The proposed hybrid LEACH with fire-
fly algorithm minimized the packet loss rate. The IOT-network performance is
measured based on energy and lifetime node.
Figure 1 demonstrates that the result of hybrid scheme with individual protocol
such as LEACH and LEACH-Monkey Search algorithm for energy optimization.
This hybrid schemes performing better than other schemes because of joins the
searching ability of FA with strength of LEACH. This proposed hybrid method
increases the lifetime and efficiency of node.

Fig. 1 Network throughput of hybrid schemes

568 A. Kannammal and S. Suresh

5 Conclusion

Current research widely focus on sensors application like IOT in which clustering
process is main issues to saving energy level of sensors node, in concern this, a pro-
posed hybrid scheme of LEACH and firefly search methods performance is offered.
The result of hybrid approach for the clustering nodes illustrates enhancement in the
performance parameters such as energy level, lifetime of IOT environments.

References

1. Labraoui N, Gueroui M, Aliouat M, Petit J (2011) RAHIM: robust adaptive approach based
on hierarchical monitoring providing trust aggregation for wireless sensor networks. J UCS
17(11):1550–1571
2. Arampatzis T, Lygeros J, Manesis S (2005). A survey of applications of wireless sensors
and wireless sensor networks. In: Intelligent Control, 2005. Proceedings of the 2005 IEEE
international symposium on Mediterrean conference on control and automation. IEEE, pp. 719–
724
3. Palumbo F, Ullberg J, Štimec A, Furfari F, Karlsson L, Coradeschi S (2014) Sensor network
infrastructure for a home care monitoring system. Sensors 14(3):3833–3860
4. Malinowski J, Geiger EJ (2014) Development of a wireless sensor network for algae cultivation
using ISFET pH probes. Algal Res 4:19–22
5. Xu L, Collier R, O’Hare GM (2017) Survey of clustering techniques in WSNs and consideration
of the challenges of applying such to 5G IoT scenarios. IEEE Internet of Things J 4(5)
6. Huang J, Meng Y, Gong X, Liu Y, Duan Q (2014) A novel deployment scheme for green
Internet of things. IEEE IoT J 1:196e205
7. Jelicic V, Magno M, Brunelli D, Paci G, Benini L (2013) Context-adaptive multimodal wireless
sensor network for energy-efficient gas monitoring. IEEE Sens J 13:328e38
8. Sun X, Coyle EJ (2012) Quantization, channel compensation, and optimal energy allocation
for estimation in sensor networks. ACM Trans Sens Netw 8:1e25
9. Younis O, Fahmy S (2004) Heed: a hybrid, energy-efficient, distributed clustering approach
for ad hoc sensor networks. IEEE Trans Mob Comput 3(4):366–379
10. Li C, Ye M, Chen G, Wu J (2005) An energy-efficient unequal clustering mechanism for
wireless sensor networks. In: Proceedings of the 2nd IEEE international conference on mobile
Adhoc and sensor systems, 597–604
11. Yang X-S, Deb S, Fong S (2014) Metaheuristic algorithms: optimal balance of intensification
and diversification. Appl Mathe Inform Sci (AMIS) 8(3):977–983
12. Faisal S, Javaid N, Javaid A, Khan MA, Bouk SH, Khan ZA (2013) Z-SEP: zonal-stable election
protocol for wireless sensor networks. J Basic Appl Sci Res 3(5):132e9
13. Tyagi S, Kumar N (2013) A systematic review on clustering and routing techniques based upon
LEACH protocol for wireless sensor network. J Netw Comput Appl 36:623e45
14. Heinzelman W, Chandrakasan A, Balakrishnan H (2000) Energy-efficient routing protocols
for wireless microsensor networks. In: 33rd Hawaii International conference system sciences
15. John A, Babu KV (2017) Two phase dynamic method for cluster head selection in wireless
sensor network for internet of things applications. IEEE WiSPNET 2017 conference
16. Fister Jr I, Yang X-S. Brest J (2013) A comprehensive review of firefly algorithms. Swarm
Evol Comput 13:34–46
Modified Dive and Rise Technique
Incorporating Enhanced Weighted
Centroid Localization Algorithm
with Ocean Current Mobility Model
in Underwater Acoustic Sensor Networks

R. Bhairavi and Gnanou Florence Sudha

Abstract Underwater Acoustic Sensor Networks (UASN) have a wide variety of

applications such as oil platform monitoring, prediction of natural disasters, monitor-
ing the pollution levels, study of aquatic life, etc. The next step after the deployment
of sensor nodes in the subsea environment is their localization. Localization is a vital
requirement to utilize the sensed data effectively and it is necessary for trailing of
nodes and detection of the target. In this paper, drifting of unlocalized nodes caused
by changes in ocean currents or other factors that are modelled by using Meandering
Current Mobility (MCM) Model and the unlocalized nodes computes their posi-
tional coordinates by the Modified Dive and Rise Technique (MDRT) incorporating
Enhanced Weighted Centroid Localization (EWCL) algorithm. Simulation results
indicate that the proposed MDRT with EWCL algorithm outperforms the existing
MDRT with DV Hop localization algorithm in terms of packet delivery ratio, delay,
localization ratio, and coverage.

Keywords Localization · Modified dive and rise technique · Enhanced weighted

centroid localization · DV hop localization

1 Introduction

In recent times, Underwater Acoustic Sensor Networks is evolving as an emerging

technology for the exploration of the underwater aquatic environment. UWASN
comprises of ample number of distributed sensor nodes, Autonomous Underwater
Vehicle (AUV), surface buoys, base station, and some application-specific devices
[1]. In the subsea environment, the propagation of electromagnetic waves is extremely
worse. At low frequencies, Radio waves can propagate only brief distances. These

R. Bhairavi (B) · G. F. Sudha

Department of Electronics and Communication Engineering, Pondicherry Engineering College,
Puducherry 605014, India
e-mail: bhairavi@[Link]
G. F. Sudha
e-mail: gfsudha@[Link]

© Springer Nature Singapore Pte Ltd. 2020 569

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
570 R. Bhairavi and G. F. Sudha

characteristics make the radio waves unsuitable for communication in the aquatic
environment.
Optical signals, although utilized in the blue-green region in the range of 500 nm,
undergo attenuation. Irrespective of high bandwidths, the propagation of optical
signals is not more than 100 m. The acoustic signal travels around 1500 m/s in
underwater. This range is approximately a magnitude of five orders lesser than elec-
tromagnetic signals. Propagation speed of acoustic signal is a versatile parameter that
depends on the following factors temperature, salinity and pressure (due to depth)
of water.
Underwater acoustic nodes with sensing, processing and communication capabil-
ities need to locate themselves for various applications since the information about
the sensed event without its positional information is of no use. The process of
finding the positional coordinates of the underwater nodes localization. Thus, pre-
cise localization of sensor nodes is a primary concern in Underwater Acoustic Sensor
Network. Underwater Acoustic Communication suffers from high propagation delay,
low bandwidth, Doppler shift, high bit error rate. These factors, in turn, makes the
localization of nodes in UWASN a challenging task.
The structure of the paper is organized as follows: In Sect. 2, the overview of
various localization techniques in Underwater Acoustic Sensor Network is summa-
rized. In Sect. 3, the proposed MDRT with Enhanced Weighted Centroid Localization
(EWCL) algorithm for efficient localization of nodes in subsea environment by con-
sidering the ocean drift model is described. In Sect. 4, the Aquasim results of the
proposed MDRT with EWCL and its comparison with the existing Dive and Rise
technique is presented. Finally, the conclusion is made in Sect. 5.

2 Related Works

Several research works related to the computation and estimation of positional coor-
dinates of the underwater acoustic nodes in the subsea environment are discussed in
this section. M. TalhaIsik et al. proposed in [2], a Three-Dimensional Underwater
Localization (3DUL) scheme. 3DUL comprises of 2 phases, Ranging and projec-
tion and dynamic trilateration phase. In the initiative phase, the unlocalized acoustic
node computes its distance to the neighbouring anchor nodes. In the Projection and
Dynamic Trilateration Phase, each sensor node utilizes the computed distance and
the (z coordinate) depth information from the initiative phase and creates a robust
virtual quadrilateral plane by projecting any 3 anchor node onto its horizontal level
to execute three-dimensional localization.
Erol-Kantarci et al. in [3, 4] studied the design principles, architectural depen-
dencies, pros and cons of various localization techniques. A comprehensive study
of centralized and distributed localization techniques along with the two subcate-
gories namely, estimation based and prediction based and their performances are
discussed. Zhou et al. in [5] proposed Unit Ball Fitting (UBF) algorithm and Iso-
lated Fragment Filtering (IFF) algorithm for localization and accurate or high fidelity
Modified Dive and Rise Technique … 571

boundary detection in sensor network. Unit Ball Fitting algorithm finds a group of
latent boundary nodes, which is succeeded by the refinement IBF algorithm, that is
responsible for eradicating the nodes those misconceived as boundary nodes by Unit
Ball Fitting technique. For each 3D boundary, a locally planarized triangular mesh
surface is constructed.
Anil et al. in [6] made a comparative study of various techniques of localization that
are anchor-based and anchor free schemes in underwater acoustic sensor networks.
Mukesh Beniwal et al. in [7] proposed an extended version of Dive and Rise scheme
incorporating time synchronization free localization technique and gave a detailed
analysis of energy efficiency which is a prime concern for UWASN is discussed. In
this work, the acoustic nodes whose positional coordinates are unknown are assumed
to be static and the speed of propagation of acoustic signal in water is assumed to be
constant. In real-time considerations, speed of propagation of acoustic signal in water
depends on various parameters like temperature, pressure, salinity, gravity variation,
specific volume and other oceanographic forces like tides, water currents etc., To
beat out this limitation, in the proposed paper drift of the sensor nodes caused due to
oceanographic forces is taken into account by incorporating the Meandering Current
Mobility (MCM) Model in the subsea environment.
Anjana Das et al. proposed in [8], a positional estimation algorithm based on
a single anchor node. In SAS (single anchor node support) sensor node estimates
its location by considering the TOA and AOA (Time and Angle of Arrival) upon
the reception of a packet from an anchor node. Location is estimated by using the
equi-rectangular approximation method for projecting the latitude and longitude
coordinates of sensor node into a Cartesian plane. Simulations and testing of SAS
technique is carried out in several locations in Arabian ocean and which depicted
that for sensor nodes of short-range, SAS technique exhibited better performance.
HanjiangLuo et al. proposed in [9] localization for double-head maritime sensor
networks (LDSN). LDSN has three phases. In the initial phase, self-moored node
localization (SML) algorithm is utilized by the supergroup nodes to compute the
positional coordinates of the moored subsea nodes. The moored nodes that are local-
ized in the former step turn into anchor nodes which aid the unlocalized underwater
moored nodes to compute their locations by USD (underwater sensor localization)
algorithm. In the terminating phase, FLA (Floating Node localization Algorithm) is
used to compute the positional coordinates of free floating sensor nodes.
ZhuSanfeng et al. proposed in [10] a Dual-Hydrophone Localization (DHL)
method with an objective to overcome three main issues and provide high robust-
ness and better performance by reducing the effects of location and angular error
in underwater nodes. DHL converts localization problems into half-plane inter-
section issues. Synchronized Dual-Hydrophone is used to meet the requirement of
time synchronization which is required to overcome the long latency in underwater
communication.
DV (Distance Vector) Hop algorithm proposed in [11] comprises of three stages.
In the initial stage, each anchor node forwards their positional coordinates and their
value of hop count to its one hop neighbouring nodes. The nodes upon the reception
of these packets will add one to the hop count value and sequentially forwards
572 R. Bhairavi and G. F. Sudha

the packets to its one hop nodes. Thus, almost each and every acoustic node in
the entire sensor network will perceive the minimum hop count from each anchor
node and the positional information of every anchor node [12]. In the succeeding
phase, each and every anchor node computes the average hop distance. Finally,
each unlocalized nodes estimates its positional coordinates (x, y) by performing
multilateration technique. Underwater nodes are fitted with pressure sensor so the
depth (z coordinate) can be calculated.
Kayalvizhi et al. proposed in [13], a node localization technique where the posi-
tional coordinates of the underwater nodes are estimated by Dive and Rise (DNR)
localization incorporated with DV hop technique. Beacon node dive and rise above
the water surface and periodically broadcast its location coordinates to the unlocal-
ized nodes. Upon the reception of DNR messages, each unlocalized node estimates its
positional coordinates by performing DV Hop localization. The main disadvantages
of DV Hop technique is that the unlocalized nodes utilizes the average hop distance
(AHD) estimated by the anchor node as a replacement for the actual distance, which
induces localization error. Thus localization accuracy is an indispensable problem
in the DV hop technique. The computational complexity of the DV hop algorithm is
also comparatively high.
To overcome the disadvantages of existing works, in this paper, Modified Dive and
Rise Localization (MDRT) incorporating Enhanced Weighted Centroid Localization
(EWCL) algorithm [14] is proposed so that the unlocalized nodes compute their
positional coordinates accurately.

3 Proposed Modified Dive and Rise Technique

Incorporating Enhanced Weighted Centroid Localization
(EWCL) Algorithm

3.1 Drift Model

In most works, the acoustic nodes whose positional coordinates are unknown are
assumed to be static and the speed of propagation of acoustic signal in water is
assumed to be constant In real-time considerations, speed of propagation of acoustic
signal in water depends on various parameters like temperature, pressure, salinity,
gravity variation, specific volume and other oceanographic forces like tides, water
currents etc., To beatout this limitation, in the proposed paper drift of the sensor nodes
caused due to oceanographic forces is taken into account by designing and incorpo-
rating the Meandering Current Mobility (MCM) Model in the subsea environment
[15].
Meandering Current Mobility Model for oceanology renders a satisfactory degree
of accuracy in modelling coastal shallow and deep water ocean currents. MCM
validates the drift of the sensor nodes in the subsea environment caused by waves,
hydro currents, tides and other oceanographic forces [16]. The mobility of nodes in
Modified Dive and Rise Technique … 573

meandering current mobility model is defined by Eqs. (1) and (2),

[y − A(t) sin(M(x − St))]

ϕ(x, y, t) = − tan h  (1)
1 + M 2 A2 (t)cos2 (M(x − St))
A(t) = W + α cos(t) (2)

where W is width of the meanderer, S is phase speed, M is no. of meanders,  is

the frequency, and α is the amplitude. For this work, the following values have been
assumed, W = 1.2, S = 0.2, M = 2π /7.5,  = 0.4, and α = 5. The current field of
the ocean is taken as the sum of the net response of the tidal field of ocean and the
residual current field. The spatially unvarying oscillating ocean current in an unitary
direction is adopted in order to deceive as the tidal field of ocean, while the boundless
sequence of clockwise and anticlockwise rotating eddies is the residual current field.
The velocity field without any dimension in the kinematical model can be judged
tentatively or taken close to

(3)

where Vx is the speed in x-axis, Vy is the speed in y-axis, K 1 , K 2 , K 3 , K 4 , v and

ň are variables which are closely related to environmental factors, such as tides,
etc. These parameters will change in different environments. Using Eqs. (1), (2) and
(3), the drift of acoustic sensor nodes by oceanographic forces is modelled. After
the implementation of the drift model, the randomly deployed underwater nodes at
distinct depth begin to locomote. Hence the node position changes to (x  , y  , z  ).
Thus, to compute the geographical coordinates of an unlocalized acoustic node,
MDRT with EWCL algorithm is proposed.

3.2 Dive and Rise Technique

Dive and Rise Localization method comprises of mobile anchor nodes also known
as DNR Beacons with inbuilt GPS to compute their geographic coordinates while
floating on the surface. These DNR beacon nodes have a predefined mobility pattern
which dive and rise in vertical direction. While diving, these beacon nodes period-
ically broadcast their location coordinates to the unlocalized nodes that are in the
region of coverage. Upon rising after one cycle, the DNR Beacon nodes recompute
its geographic coordinates and dives again. Unlocalized nodes upon reception of
these messages from the DNR Beacon nodes computes its positional coordinates by
enhanced weighted centroid localization algorithm.
574 R. Bhairavi and G. F. Sudha

3.3 Modified Dive and Rise Technique (MDRT)

Enhanced weighted centroid localization computes weight by taking into account

of several factors such as effect of different anchor nodes, radius of communication
and the near anchors of the unlocalized node. EWCL algorithm comprises of three
phases:
• Computation of minimum hop count (hcount ) of each node from every anchor node
in the network
• Computation of AHDanchor(m)
• Computation of positional coordinates of unlocalized node.
Computation of minimum hop count (hcount ) of each node from every anchor
node in the network:
In the existing DVHop localization algorithm, to localize a node the vital requirement
of large number of anchor nodes leads to higher localization error. The proposed
MDRT with DNR localization overcomes this issue by making the anchor node
broadcast only within m hops instead of broadcasting to all the nodes in the network.
The range of m is 2 to hmax .
Initially each anchor node broadcasts <x k , yk , hcount > information where <x k , yk >
represents its two-dimensional positional coordinates and hcount its hop count value
in the form of packets to the unlocalized nodes within m hops. The value of hcount
is initially assigned to 0. Each unlocalized node (q) maintains a table containing
the following information <k, x k , yk , hqk > for each anchor node k within m hops.
Each unlocalized node upon the reception of a packet from an anchor node checks
its own table. If the hqk value stored in its own table is less when compared to the
received hqk and value of hqk is less then m, then the received hqk value is ignored
by the unlocalized node else the value of hqk is incremented by 1 and the new hqk is
subsequently stored by the unlocalized node in its table. After the updating the table,
it broadcasts the table to its neighbouring unlocalized nodes. Thus at the end of this
initiative phase, each unlocalized node computes its minimum hqk value from every
anchor node k that are within m hops and maintains an updated hop count table.
Consider a network with 6 Anchor nodes (here nodes 10, 11, 12, 13, 14, 15) and
9 unlocalized nodes (nodes 1, 2, 3, 4, 5, 6, 7, 8, 9) as shown in the Fig. 1. Let the
value of m be 2 hops. The 2-dimensional positional coordinates of anchor nodes is
specified in the Table 1.
In the initial phase, anchor nodes in the network broadcast its geographic coordi-
nates to all the unlocalized nodes within m hops. In the network depicted in Fig. 1,
for the unlocalized node 3, all the anchor nodes within m hops (anchor nodes 10, 11,
12, 13, 14) are considered and anchor node 15 (since its hop count is greater than 2)
is ignored. At the end of this phase, the unlocalized node 3 maintains an updated hop
count table as shown in Table 2. Similarly, all the other unlocalized node maintain
their updated hop count table containing <x k , yk , hcount > information of the anchor
nodes within m hops.
Modified Dive and Rise Technique … 575

Fig. 1 Deployed anchor and

unlocalized nodes

Table 1 Positional
Anchork xk yk
coordinates of anchor nodes
10 436 287
11 124 237
12 381 264
13 394 605
14 347 412
15 460 407

Table 2 Table maintained by

Anchork xk yk hk3
node 3 for each anchor node
within m hops 10 436 287 2
11 124 237 1
12 381 264 1
13 394 605 2
14 347 412 1

Computation of AHDanchor(k):
Each anchor node computes its average hop distance AHDanchor(k) using the following
Eq. (7).
t 
k=0;l=0 (xk − xl )2 + (yk − yl )2
AHDanchor(k) = (4)
hk
576 R. Bhairavi and G. F. Sudha

where t is the aggregate number of anchor nodes in the UWASN, l denotes all other
anchors and h k is the number of hops between anchor node k and anchor node l,
(xk , yk ) and (xl , yl ) represents coordinates of anchor nodes k and l, respectively.
Each anchor node after estimating its average hop distance AHDanchor(k) , sends
the computed value to all the unlocalized nodes within m hops. The unlocalized
node stores only the packet from the nearest anchor whose packet with information
AHDanchor(k) while discards all the other packets. Each unlocalized node estimates
its distance from the anchor node Ak using the Eq. (5),

dk = AHDanchor(k) ∗ h k (5)

Computation of positional coordinates of unlocalized node:

In terminating stage, the unlocalized nodes determine their location coordinates < x r ,
yr > exploiting Eq. (6). Weight used for estimating the coordinates of the unlocalized
node is computed using the following Eq. (7)
t t
k=1 wk x k wk yk
X r = t , yr = k=1
m (6)
k=1 wk k=1 wk

where Wi = h1qk is the weight of each anchor node i, hqk is minimum hop count value
of node q from anchor k and t represents the total number of anchor nodes.
 t  AHDanchor
r
h qk l
wi = k=1
(7)
t ∗ h qk

where, AHDanchor(l) is the average hop distance of its nearest anchor node l to the
unknown node q and r is the communication radius of node.
The weight factor is inversely proportional to number of hops [14]. This has been
used to give more weight age to nearest anchor. The anchor with less number of hops
is closer to the given node, thus has more impact in determining the location of given
node (Fig. 2).
The computed weight factor relies on the distance between the anchor node and
the unlocalized node. The near anchor node to the unlocalized node have higher
impact factor than the far anchor nodes.
Depth of Sensor nodes
The z coordinate which represents the depth of the sensor node is computed from
the following parameters: Variation of gravity (GR) [15], specific volume (V ).

Q = Sin(Lat/57.29578);

The gravity variation and the specific volumes [16, 17] are calculated by,
Modified Dive and Rise Technique … 577

Start

Each AN broadcasts <xk,yk,hcount> to

unlocalized nodes within m hops

Each unlocalized node compares the

hqk(received) with its hqk(stored)

No Yes
If hqk(received)
< hqk(stored)

hqk(received) = hqk(received) +1 hqk(received) is ignored

Initialization of Φ2

Each AN computes its AHDanchor(k) and

broadcasts within m hops

Each unlocalized node computes its distance

from Ak

Computation of weight and thereby

computation of location coordinates

End

Fig. 2 Flowchart for EWCL algorithm

   
GR = 9.780318 ∗ 1.0 + 5.2788e−3 + 2.36e−5 ∗ Q 2 ∗ Q 2 + 1.092e−6 ∗ P
(8)
  
specificvolume = −1.82e−15 ∗ P + 2.279e−10 ∗ P − 2.2512e−5
∗ P + 9.72659) ∗ P (9)

where P = 10,000 decibars

578 R. Bhairavi and G. F. Sudha

Depth(Z coordinate) = specific volume/GR (10)

By using Eq. (10), depth (Z coordinate) can be calculated. Thus the coordinate of
node (x, y, z) can be calculated using Eqs. (6) and (10).

4 Performance Evaluation

In this section, the performance of the proposed Modified Dive and Rise Technique
with EWCL algorithm and the existing DV hop localization algorithms were eval-
uated using Aquasim (NS 2.30). Aquasim can efficiently configure and simulate
the Real Underwater Acoustic Channel incorporating the object-oriented design of
NS-2. The communication range of the underwater acoustic nodes are 200 m. 50
unlocalized nodes and 7 Dive and Rise Beacon nodes are randomly deployed in
1000 m × 650 m monitored space with varied depth. The mobility of the unlocal-
ized acoustic sensor nodes are restricted within a range of 50 m. Table 3 lists the
simulation parameters of NS 2.30.

4.1 Performance Metrics

The performance of the proposed MDRT with EWCL algorithm is evaluated in terms
of localization ratio, packet delivery ratio, transmission range versus localization
ratio, transmission range versus coverage and error. The proposed technique is studied
taking into considerations the drifting of sensor nodes by incorporating MCM. The

Table 3 Aquasim simulation

Parameter Value
parameters
Number of nodes 57
Interface type Phy/underwater phy
MAC type 802.11
Queue type Queue/drop tail/pri queue
Antenna type Omni-antenna
Propagation type Underwater propagation
Transport agent UDP
Application agent CBR
Meandering current mobility Mean: A = 1.2, c = 0.2,
model settings K = 7.5
2π
, ε = 5, ω = 0.4
Ocean current parameter K 1 , K 2 = π , K 3 = 2π , K 4 ,
settings K 5 = 1, v = 1, ň = 0.5–3.0
Modified Dive and Rise Technique … 579

existing Dive and Rise technique with DV hop localization technique is compared
with the proposed MDRT with EWCL algorithm.
Nodes versus localization ratio
The localization ratio (Loc_Ratio) of the UWASN is the ratio of total number of
localized underwater nodes to the total number of deployed underwater nodes. The
Loc_Ratio is determined as follows

Loc_Ratio% = (NLoc /NTot) × 100% (11)

where N Loc is the number underwater nodes whose positional coordinates are known
and N Tot is the total number of deployed underwater nodes.
From Fig. 3, it is perceived that when the number of deployed sensor nodes in the
network is 25, the proposed MDRT with EWCL algorithm achieves Loc_Ratio of
96%. As Loc_Ratio is inversely proportional to the total number of deployed sensor
nodes in the network, it decreases as the number of nodes increases. When the
total number of nodes is increased to 55, the proposed modified MDRT algorithm
still maintains a localization ratio of 89%. Thus it is observed that the proposed
MDRT with EWCL algorithm has higher Loc_Ratio compared to the existing DVhop
irrespective of the increase in the deployed sensor nodes.
The performance of the proposed modified MDRT algorithm is next validated
by analyzing the localization ratio while varying the transmission range in steps of
20 m. Initially, the transmission range of sensor nodes is 100 m. From the Fig. 4, it
is observed that when the transmission range is increased, the localization ratio of
the proposed modified MDRT scheme with EWCL approach also increases when
analyzed with the existing localization techniques. Consider the network scenario
where the transmission range of sensor node is 150 m, the proposed DNR with EWCL
has a localization ratio of 90% whereas it is 88% for existing DV hop localization.
Localization error percentage
The performance of the modified EWCL algorithm is next analyzed in terms of
localization error percentage. Localization error percentage is computed using the
difference between the real and the estimated coordinates of the node by using the
Eq. (11),

Fig. 3 Localization ratio 98

versus number of nodes 96
Localization Ratio

94
92 DNRL with DV
90 Hop Algorithm
88
86 Proposed DNRL
84 with EWCL
82 Algorithm
25 30 35 40 45 50 55
No. of Nodes
580 R. Bhairavi and G. F. Sudha

Fig. 4 Localization ratio 120

versus transmission range
100

Localization Ratio
80 DNRL with DV
60 Hop Algorithm

40
Proposed DNRL
20 with EWCL
Algorithm
0
100 120 140 160 180 200 220 240
Transmission Range


(X − X  )2 − (Y − Y  )2
%Localization error = ∗ 100 (12)
n∗R
(X, Y ) = Actual coordinates of the sensor nodes
 
(X , Y ) = Estimated coordinates of the sensor nodes

As shown in the Fig. 5, initially when the deployed nodes is 10, the modified
EWCL algorithm has an error percentage of 39.46% whereas for DV hop it is
46.187%. The proposed modified EWCL algorithm shows considerable reduction
in the localization error percentage than that of the existing DV hop localization
technique. The localization error percentage for the proposed algorithm is 50.311%
and for existing DV hop algorithm is 51.49% when the total number of deployed
nodes is 55. Thus, from the Fig. 5, it is inferred that when the deployed nodes
increases, the localization error percentage is also increased.
Transmission range versus coverage
Coverage (C R ) is defined as the extent at which each point of the deployed network
is under the vigilance of the sensor node. Let T R be the transmission range of DNR
Beacon node. The travelling distance of the DNR Beacon node before broadcasting
its geographic coordinates is denoted as Dbeacon . Then the CR for the DNR Beacon
node is computed from Eq. (12) as,

Fig. 5 Localization error

Localization Error Percentage

60
percentage
50
40
DNRL with DV
30 Hop Algorithm

20
Proposed DNRL
10 with EWCL
Algorithm
0
10 15 20 25 30 35 40 45 50 55
No. of Nodes
Modified Dive and Rise Technique … 581

1.2

0.8
Coverage (%)

0.6
0.10 Beacon Interval
0.4 0.20 Beacon Interval

0.2

0
100 120 140 160 180 200 220 240
Transmission Range (m)

Fig. 6 Transmission range versus coverage

C R = ((22/7)/12)(4TR + Dbeacon )(2TR − Dbeacon )2 (13)

The performance of the proposed MDRT with EWCL is then validated by varying
the beacon interval in two cases initialized as 0.10 and 0.20 s in terms of transmission
range versus coverage. Transmission range of underwater nodes is gradually changed
from 150 to 250 m with a fixed interval of 20 m. From Fig. (6), it is inferred that when
the beaconic message interval is initialized as 0.10, the overall coverage percentage
increases when compared with the beacon interval 0.20. It is clearly observed that
when the transmission range of the sensor increases, then there is a subsequent
increase in the coverage. The reason for achieving coverage to such an extent is that
when the beacon interval is set small with higher transmission range of sensor nodes,
then many sensor nodes are able to receive multiple messages.

5 Conclusion

In this paper, the Dive and Rise Localization technique is modified using the
Enhanced Weighted Centroid localization algorithm for the estimation of node posi-
tion in subsea environment. EWCL algorithm comprises of three phases: Determin-
ing minimum number of hop counts by each node from every anchor, Determining
average distance per hop by each anchor, Determining location of unknown node.
The results validate that proposed MDRT incorporated with EWCL algorithm shows
improvement in the localization performance when compared with the existing DV
hop localization algorithm evaluated in terms of localization ratio, localization error
percentage, transmission range and coverage.
582 R. Bhairavi and G. F. Sudha

References

1. Ong KG, Yang X, Mukherjee N, Wang H, Surender S, Grimes CA (2004) A wireless sensor
network for long-term monitoring of aquatic environments: design and implementation. Sens
Lett 2(1):48–57
2. TalhaIsik M, Akan OB (2009) A three dimensional localization algorithm for underwater
acoustic sensor networks. IEEE Trans Wirel Commun 8:4457–4463
3. Erol-Kantarci M, Mouftah HT, Oktug S (2010) Localization techniques for underwater acoustic
sensor networks. IEEE Commun Mag:152–158
4. Erol-Kantarci M, Mouftah HT, Oktug S (2011) A survey of architectures and localization
techniques for underwater acoustic sensor networks. IEEE Commun Surv Tutorials 13:487–507
5. Zhou H, Xia S, Jin M, Wu H (2014) Localized and precise boundary detection in 3-D wireless
sensor networks. IEEE/ACM Trans Netw:1–14
6. Anil CB, Mathew S (2014) A survey and comparison study of localization in underwater sensor
networks. Int J Comput Sci Mob Comput 3:23–29
7. Beniwal M, Singh RP, Sangwan A (2016) A localization scheme for underwater sensor networks
without time synchronization. Springer Wirel Pers Comm 88:537–552
8. AP Das, SM Thampi (2015) Single anchor node based localization in mobile underwater
wireless sensor networks. Springer Int Publ:757–770. Springer, Switzerland
9. Luo H, Wu K, Gong Y-J, Ni LM (2016) Localization for drifting restricted floating ocean sensor
networks. IEEE Trans Veh Technol 65:9968–9981
10. Sanfeng Z, Naigao J, Lei W, Xueshu Z, Shuailing Y, Ming Z (2016) A novel dual-
hydrophone localization method in underwater sensor networks. IEEE/OES China ocean
acoustics symposium
11. Kaur A, Gupta GP, Kumar P (2017) A survey of recent developments in DV-hop localization
techniques for wireless sensor network. J Telecommun Electron Comput Eng 9:69–71
12. Khurana M, Payal A (2011) Analysis of DV-hop localization algorithm in wireless sensor
networks. In: Proceedings of the 5th national conference, India, pp 1–4
13. Kayalvizhi C, Bhairavi R, Sudha GF (2018) Localization of nodes with ocean current mobility
model in underwater acoustic sensor networks. In: Proceedings of international conference
on computer networks and inventive communication technologies (ICCNCT 2018). Springer,
India
14. Kaur A, Kumar P, Gupta GP (2017) A weighted centroid localization algorithm for randomly
deployed wireless sensor networks. J King Saud Univ Comput Inf Sci:1–11
15. Caruso A, Paparella F, Vieira LFM, Erol M, Gerla M (2008) The meandering current mobility
(MCM) model and its impact on underwater mobile sensor networks. In: Proceedings of 27th
IEEE infocommunication, USA, pp 771–779
16. Che X, Wells I, Dickers G, Kear P, Gong X (2010) Re-evaluation of RF electromagnetic
communication in underwater sensor networks. Commun Mag IEEE 48:143–151
17. Grosso VAD (1974) New equations for the speed of sound in natural waters (with comparison
to other equations). J Acoust Soc Am 93(4):10841091
Tree-Based Approaches for Improving
Energy Efficiency and Life Time
of Wireless Sensor Networks (WSN):
A Survey and Future Scope for Research

Pranesh and Santhosh L. Deshpande

Abstract Wireless Sensor Networks (WSN) are characterized by highly

application-specific nature, stringent resource constraints, self-organizing, spatio-
temporal traffic, and large dynamic topology with several contradicting design goals.
Of these design goals, network life time and energy efficiency are considered as of
paramount importance. Many research works from the past have dedicated them-
selves in extending the network life time and achieving energy efficiency of WSN
through various techniques, including that of the application of Tree as a data struc-
ture. This article attempts to present a detailed survey of the existing research works
with the application of different variants of Trees. Further, the paper tries to analyze
the performance implications of application of variants of trees, advantages, and dis-
advantages. The paper mentions possible feasibility of the application of Red Black
Trees (RBL) in WSN and the potentials for future research while giving algorithmic
comparison of RBL with other tree data structures.

Keywords Wireless sensor network · Tree-based approaches · Red black tree ·

Network life time · Energy efficiency

1 Introduction

A wireless sensor network is constituted of large number of sensor nodes deployed

randomly around phenomenon of interest Some important applications of WSN are:
home automation, industrial applications, monitoring applications for environment,
traffic, & wildlife, health care applications, defence applications involving security &
surveillance, and some special research applications like study of sand bar formations

Pranesh (B)
VTU-RRC, Visvesvaraya Technological University (VTU), Belgaum, Karnataka, India
e-mail: praneshvkallapur@[Link]
S. L. Deshpande
Department of Post Graduate Studies, Visvesvaraya Technological University (VTU), Belgaum,
Karnataka, India
e-mail: sld@[Link]

© Springer Nature Singapore Pte Ltd. 2020 583

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
584 Pranesh and S. L. Deshpande

or related to oceanography, etc. Due to these wide varying applications WSN are
evolving as a new paradigm for information processing [1, 2]. Although WSN are
similar to Adhoc Networks, there are number of fine differences. Some important
differences between WSN and Adhoc network are: the number of sensor nodes
in a sensor network will be large, densely deployed, prone to variety of frequent
failures, the topology of a sensor network changes very frequently, and have stringent
constraints for various resources. In this regard, a novel architecture & design of
WSN, its protocol stack, layer-wise open issues have been discussed in detail, along
with description of some of the existing WSN specific protocols as solutions to some
of the open issues, in [1]. The need for self-organization and some solutions for the
same have been presented elaborately in [3]. Thus, with the above general background
about WSN, the motivation of this paper is to attempt to present a detailed survey of
the various existing works which proposed application of Tree as a data structures
and Chain concept from discrete mathematics for achieving energy efficiency and
extended the network life time. Further, the organization of this paper is as follows:
Sect. 2 presents the detailed literature survey of research works, the implications of
application of tree data structures on the overall performance of WSN is analyzed
in Sect. 3 and Sect. 3.4 identifies the potentials for application of Red Black Trees,
and the paper will be concluded in Sect. 4 mentioning the scope for future research
using Red Black trees.

2 Literature Survey

Wendi et al. have proposed a work involving clustering for energy-efficient com-
munication resulting in seven times efficiency in energy [4]. Huang et al. proposed
an energy-efficient routing scheme for WSN based on clustering with the use of
minimum spanning trees degree constrained (CMST-DC) [5]. Z. Han et al. pro-
posed an efficient routing by constructing routing tree (GSTEB) and mention about
achieving increase in life time anywhere in between 100 and 300% depending on
various scenarios while comparing with PEDAP [6]. Weighted Spanning Tree vari-
ant of LEACH (WST-LEACH) was proposed in [7] and authors claim that it per-
forms better than LEACH. Geographic and Energy Aware Routing (GEAR) algo-
rithm is proposed in [8]. Krishnamachari et al. [9] have discussed in detail about the
influence of source-destination placement, communication network density, and the
energy costs on each other. Boulis [10] proposed distributed estimation algorithm to
explore energy-accuracy subspace for subclass of periodical data collection problems
and presents results with five-fold improvement in energy efficiency. Yu et al. [11]
explores energy-latency trade off to a great deal. Tan and Korpeoglu [12] proposed
a new work consisting of two new algorithms, with power efficiency as their major
design goals, called as “Power Efficient Data gathering and Aggregation Protocol
(PEDAP)”. These two algorithms were relying on the usage of near-optimal mini-
mum spanning trees. A near-optimal chain-based protocol with energy-efficient as
Tree-Based Approaches for Improving … 585

its primary design goal, called Power-Efficient Gathering in Sensor Information Sys-
tems (PEGASIS), was proposed by [13]. Kulik et al. [14] proposed Sensor Protocols
for Information via Negotiation (SPIN) family of protocols, claiming 75% energy
efficiency but incur lot meta-data usage. He et al. [15] proposed a research work by
using feedback between sensor nodes, and reported 30–50% energy efficiency when
compared to non-feedback based fixed scenarios but incurs 2 byte extra overhead for
each packet for dealing with feedback. Tabassum et al. [16] proposed a work which is
energy-aware version of periodical data collection. Further the same authors present
two more improved works in [17, 18] which are based on ‘chain’ concept and claim
to achieve 15–30% extended network life time and 90–95% energy efficiency. Du
et al. [19] also present a similar chain-based improved algorithm for data gathering,
called ‘Chain Oriented SEnsor Network’ (COSEN). Heuristics based aggregation
tree construction for data gathering had been discussed in detail (called EADAT) in
[20]. Kalpakis et al. [21] attempted to propose a novel solution for extending life
time of network for data gathering purposes. Another work intending to extend the
life time of network for data aggregation is presented by Hong and Prasanna [22].
Sadagopan et al. [23] focused on maximizing the data collected than so far con-
sidered parameters. However, Ordonez et al. [24] presented a more interesting work
concerning the same maximization of data collection but amidst of energy constraints
while providing flexibility to choose trade-off depending upon the design require-
ments. Another research work intending to extend the life time of the network for
data aggregation was presented by Xue et al. [25]. Kim et al. [26] proposed an idea of
construction of trees for the purpose of data collection, which was termed by authors
as Tree-Based Clustering (TBC). Parthasarathy and Karthickeyan [27] presented a
research work to improve the life time of the network for data aggregation purposes.
The approach of this work involved both trees and cluster concepts as well. Salam
and Ferdous [28] present a detailed survey of Tree-based data aggregation works for
WSN in detail. Dreef et al. [29] focused on the application of Tree as a data structure
to achieve improved security. Hussain et al. [30] proposed a work for hierarchical
clusters operating in distributed fashion.

3 Analysis of Implications of Application of Tree Data

Structures on Overall Performance of WSN

It may be observed from the literature survey presented in Sect. 2 that, most of the
existing works have considered mainly spanning-tree variant as a data structure, with
one or two applying general tree approaches. Also, few works were based on chain
concept.
586 Pranesh and S. L. Deshpande

3.1 Disadvantages of Minimum Spanning Trees (MST)

and Binary Search Trees (MST)

Main advantage of MST and BST variants of trees is that they are easy and simple
to implement. But, at the same time, they have several disadvantages. In case of
minimum spanning trees, they have varying path lengths along with many instances
being applicable. Most frequently followed algorithms for constructing minimum
spanning tree have been shown to run with complexity of O(m log n) where ‘m’ is
the number of edges in the resultant spanning tree and ‘n’ is the number of nodes.
Any other variants of binary trees are also having varying depths and operational
costs (time complexity) depending upon the circumstances. For example, simple
binary search tree may often show the performances for best case as O(log n) and
for worst-case can degrade to O(n) when it becomes unbalanced, where ‘n’ is the
number of nodes in the tree. Thus, the time taken to perform operations is less if the
height of the search tree is small; but if its height is large, their performance may be
no better than with a linked list.

3.2 Disadvantages of Self-balancing Variants of Trees

Now let us consider the case of self-balanced tree variants those could be considered
for application in WSN. AVL trees are a kind of self-balancing trees. But they are
good only if lookups dominate the insert/delete operations. In the case of frequent
insert/delete operations, even though its depth is at most ~1.44 * lg(n + 2), AVL
tree performance will be slower requiring as many as (log n) rotations to maintain
balance in an n-node tree. Further, AVL trees impose rigid balance on the tree structure
leading to slow and costly operations.

3.3 Reasons for Worst-Case Performance of MST, BST,

and Self-balancing Variants of Trees When Applied
in WSN

But, all the above-discussed variants of tree data structure are not well suited for appli-
cation in WSN. Major reasons for such infeasibility come from two aspects. First
aspect of such infeasibility is that most often applied tree variants in the existing works
are going to demonstrate the worst performance in case of frequent insert, delete and
lookups in trees with arbitrarily longer depths. Second aspect of such infeasibility
stems from some important unique features of WSN like highly dynamic topologies
with very large number of sensor nodes. This unique nature may also imply that
the length of path/depth/number of levels in WSN with respect to Sink/Base sta-
tion/Gateway may be a serious concern. Added to these limitations, spatio-temporal
Tree-Based Approaches for Improving … 587

nature of traffic in WSN also expects that path discovery and maintenance activities,
which are local to a particular part of the WSN at any time, should be attempted in
an efficient way in order to achieve the overall better performance and specifically
the energy efficiency and extended network life time.

3.4 Red Black Trees (RBL): Feasibility and Potentials

for Application in WSN

At first, let us glance through the properties of Red Black Trees(RBL). Red Black
Trees are a variant of self-balancing trees. The nodes in RBL are differentiated as
red and black nodes. Further, every path in RBL from root to leaf has same number
of black links. In RBL at most one red link in-a-row or path from root to leaf is
permissible. Height of tree in case of RBL is less than 2 log (n + 1). Additionally,
following are the advantages of RBL trees:
• Red-black trees perform insert, delete, and lookup with the Best- and Worst-case
complexity guaranteed to be always O(log(n)).
• Particularly useful when inserts and/or deletes are relatively frequent.
• Relatively low constants in a wide variety of scenarios.
• All the advantages of binary search trees are also available.
The only disadvantage is that they are comparatively difficult to be implemented.
More details on these Red Black Trees can be found in from references like [31].
The brief summary of implementation costs in case of various tree variants has been
described in [31] and repeated in Table 1 for convenient reference.
With the above-mentioned advantages of RBL trees, the potentials for being
applied in WSN are very high. With this motivation, the authors of this paper have
been progressing in their research work, with a hope, to design algorithm(s) for

Table 1 Summary of implementation costs of various data structures [31]

Worst case scenario Average case scenario
Approach Search Insert Delete Search Insert Delete
Sorted array log N N N log N N N
Unsorted list N 1 1 N 1 1
Hashing N 1 N 1a 1a 1a
Binary search tree N N N log Nb log Nb log N b
Randomized binary search tree log Nc log Nc log Nc log N log N log N
Splay tree log N d log N d log N d log N d log N d log N d
Red-Black tree log N log N log N log N log N log N
Legend a based on random hash map for all keys; N b number of nodes ever inserted; c probabilistic
guarantee; d amortized guarantee
588 Pranesh and S. L. Deshpande

applying RBL trees exploiting the previously mentioned properties of WSN and to
take the benefit of advantages of RBL tree specifically to improve the energy effi-
ciency and extend the network lifetime. Authors are hoping to publish the outcomes
of the same in their future research publications.

4 Conclusions

WSN have found widespread application in different domains. The WSN are also
characterized by peculiar features and conflicting design goals making them different
from other types of networks. Various techniques like application of data structures
etc. have been considered in the earlier research works to improve the overall perfor-
mance and specifically energy efficiency and network life time. In this regard, this
article made an attempt to present, as far as possible, a detailed survey of existing
research works which applied tree data structures for improving the energy efficiency
and extending the network life time of WSN. Further, an analysis of implications of
applying different types of tree data structures on the overall performance and specif-
ically towards the energy efficiency and network life time was also presented. Also,
this article presented a brief discussion on the possible potentials for application of
Red Black Tree as data structures for improving energy efficiency and network life
time while comparing the same with other variants of trees with a hope to establish
the feasibility of application of Red Black Trees as a data structure in WSN.

References

1. Akyildiz F, Su W, Sankarasubramaniam Y, Cayirci E (2002) Wireless sensor networks: a survey.

Comput Netw 38(4):393–422
2. Zhao F, Guibas LJ (2007) Wireless sensor networks: an information processing approach.
Elsevier Publications
3. Sohrabi K, Gao J, Ailawadhi V, Pottie GJ (2000) Protocols for self organization of a wireless
sensor network. IEEE Pers Commun 7(5):16–27
4. Heinzelman WR, Chandrakasan A, Balakrishnan H (2000) Energy-efficient communication
protocol for wireless micro sensor networks. In: Proceedings of international conference system
sciences
5. Huang Y, Lin J, Liang C (2008) An energy efficient routing scheme in wireless sensor net-
works. In: 22nd international conference on advanced information networking and applications
workshops. IEEE, pp 916–921
6. Han Z, Wu J, Zhang J, Liu L, Tian K (2014) A general self-organized tree-based energy-balance
routing protocol for wireless sensor network
7. Chen P, Gong S, Zhang H (2010) Weighted spanning tree clustering routing algorithm based on
LEACH. In: 2nd international conference on future computer and communication (ICFCC’10).
IEEE, pp V2–223–V2-227
8. Yu Y, Govindan R, Estrin D (2001) Geographical and energy aware routing: a recursive data
dissemination protocol for wireless sensor networks
Tree-Based Approaches for Improving … 589

9. Krishnamachari B, Estrin D, Wicker S (2002) The impact of data aggregation in wireless sensor
networks. In: Proceedings of 22nd international conference on distributed computing systems
workshops, pp 575–78, July 2002
10. Boulis A, Ganeriwal S, Srivastava MB (2003) Aggregation in sensor networks: an energy-
accuracy tradeoff. In: 1st IEEE international workshop on sensor network protocols and
applications, USA, May 2003
11. Yu Y, Krishnamachari B, Prasanna VK (2004) Energy-latency tradeoffs for data gathering in
wireless sensor Networks. IEEE INFOCOM
12. Tan HO, Korpeoglu I (2003) Power efficient data gathering and aggregation in wireless sensor
networks. SIGMOD Rec 32(4):66–71
13. Lindsey S, Raghavendra C, Sivalingam KM (2002) Data gathering algorithms in sensor
networks using energy metrics. IEEE Trans Parallel Distrib Syst 13(9):924–935
14. Kulik J, Rabiner W, Balakrishnan H (1999) Adaptive protocols for information dissemination
in wireless sensor networks. In: Proceedings of 5th ACM/IEEE mobicom conference, Seattle,
WA, Aug 1999
15. He T, Blum BM, Stankovic JA, Abdelzaher T (2004) AIDA: adaptive application—independent
data aggregation in wireless sensor networks. ACM Trans Embed Comput Syst 3(2):426–457
16. Tabassum N, QEK Mamun, Urano Y (2007) An energy aware protocol for periodical data
collection in wireless sensor networks
17. Tabassum N, Mamun QEK, Haque AKMA, Urano Y (2006) A chain oriented data collection
protocol for energy-aware and delay constrained WSN. Afr J Inf Commun Technol 2(3):126–
136
18. Tabassum N, Mamun QEK, Urano Y (2006) COSEN: a chain oriented sensor network for
efficient data collection. In: Third international conference on information technology: new
generations (ITNG’06)
19. Du K, Wu J, Zhou D (2003) Chain-based protocols for data broadcasting and gathering in
sensor networks. In: International parallel and distributed processing symposium, Apr 2003
20. Ding M, Cheng X, Xue G (2003) Aggregation tree construction in sensor networks. 2003 IEEE
58th Veh Technol Conf 4(4):2168–2172
21. Kalpakis K, Dasgupta K, Namjoshi P (2003) Efficient algorithms for maximum lifetime data
gathering and aggregation in wireless sensor networks. Comput Netw 42(6):697–716
22. Hong B, Prasanna VK (2004) Optimizing system lifetime for data gathering in networked
sensor systems. In: Workshop on algorithms for wireless and ad-hoc networks (A-SWAN),
Boston, Aug 2004
23. Sadagopan N (2004) B Krishnamachari (2004) Maximizing data extraction in energy-limited
sensor networks. INFOCOM 3:1717–1727
24. Ordonez F, Krishnamachari B (2004) Optimal information extraction in energy-limited wireless
sensor networks. IEEE J Sel Areas Commun 22(6):1121–1129
25. Xue Y, Cui Y, Nahrstedt K (2005) Maximizing lifetime for data aggregation in wireless sensor
networks. In: ACM/Kluwer mobile networks and applications (MONET) special issue on
energy constraints and lifetime performance in wireless sensor networks, Dec 2005, pp 853–864
26. Kim KT, Lyu CH, Moon SS (2010) TBC for energy efficient WSN. In: Proceedings of
international conference advanced information networking and application workshop
27. Parthasarathy P, Karthickeyan R (2014) Tree based data aggregation algorithm to increase the
lifetime of wireless sensor network. Int J Innovative Res Sci Eng Technol (IJIRSET) 3(1)
28. Salam MA, Ferdous T (2012) Tree-based data aggregation algorithms in wireless sensor net-
works: a survey. In: Proceedings of the 2012 international conference on industrial engineering
and operations management Istanbul, Turkey, 3–6 July 2012
29. Dreef D, Sun B, Xiao Y, Wu K (2006) Secure data aggregation without persistent cryptographic
operations in wireless sensor networks. In: 25th IEEE international performance, computing,
and communications conference (IPCCC’06), IEEE, pp 635–640
30. Hussain S, Yang L, Gagarin A (2009) Distributed search for balanced energy consumption span-
ning trees in wireless sensor networks. In: International conference on advanced information
networking and applications workshops, IEEE, pp 1037–1042
590 Pranesh and S. L. Deshpande

31. [Link]
32. Duarte Melo EJ, Liu M (2003) Data-gathering wireless sensor networks: organization and
capacity. Comput Netw Int J Comput Telecommun Netw 43(4)
33. Vaidhyanathan K, Sur S, Narravula S, Sinha P (2004) Data aggregation techniques sensor
networks. In: Technical report, OSU-CISRC-11/04-TR60. Ohio State University
Automated Cyber Threat Intelligence
Generation from Honeypot Data

Kumar Sanjeev, B. Janet and R. Eswari

Abstract The evolving of advance cyber threats requires the cyber security specialist
and system analyst to detect, analyse and timely react against such kind of cyber
attacks. In real practical scenario, the timely dissemination of attack information is
a challenge and that is not possible without cyber threat intelligence with inclusion
of deep analysis of attack features and attack contextual information. In this paper,
automated proactive approach for cyber threat intelligence generation is presented
integrated with standard data sharing formats that can act as attack indicator for
the security defence mechanism put in place in an organization such as SIEM. The
strength of Honeypot-based approaches for cyber threat intelligence is proven with
well-defined use cases. The capabilities of Honeypots to detect zero-day attacks
can be benefited if and only if the attack events are timely digested by the security
solutions and that is only possible by sharing the attack events in standard data sharing
languages. The developed system is fully automated that include captured attack
data is processed by various automated analysis engines, augmenting the contextual
information and applying deep learning models for later threat prediction. Finally,
we propose a system design incorporating deep learning neural network-based cyber
threat intelligence generation for cyber threat prediction. To achieve all these, cluster
of VM Honeypots are deployed in a public IP4 network.

Keywords Cyber threat intelligence · Cyber security · Honeypots · Malware ·

Deep learning

K. Sanjeev (B)
Centre for Development of Advanced Computing (C-DAC), Mohali, India
e-mail: sanjeev@[Link]
B. Janet · R. Eswari
National Institute of Technology, Tiruchirappalli, India
e-mail: janet@[Link]
R. Eswari
e-mail: eswari@[Link]

© Springer Nature Singapore Pte Ltd. 2020 591

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
592 K. Sanjeev et al.

1 Introduction

In present technological domain, everybody is connected and utilizing the technol-

ogy without knowing the advantage and disadvantages of these technologies. The
same philosophy is applying in Internet-connected cyber space. The users of the
Internet are growing exponentially. In the current technological world where people
are virtually connected and sharing information over the Internet, it exposes them to
the attackers. In recent time, the attacks performed by the attacker are highly targeted
and sophisticated that exploit the vulnerabilities in processes adopted by the peoples
connected over the Internet [1].
Recently, cyber criminals are becoming more skilled, and it is seen that complex
attacks are being launched by adopting the advanced tactics, techniques and proce-
dures (TTPs) which are becoming very difficult to handle by major defensive security
solutions and it is very hard to investigate and remediate [2]. The TTPs adopted by
the attacker is becoming less detectable, more persistent and sophisticated. Many
organizations being already effected by such category of incidents like attacker used
to deploy the ransomware and then demand for the payments to unlock and access
the critical data and assets of the organizations. For example, the latest WannaCry
ransomware attack targeted over 150 countries and infected millions of the com-
puters [3]. To protect against these latest attacks, organizations and companies are
spending huge amount of budget in state of the art security tools such as intrusion
detection and prevention systems, firewalls, corporate antivirus solutions, end point
protection systems, web application firewall(WAP), unified threat management sys-
tem, content filtering, etc. But due to limitation in their detection approach, these
security tools solutions sooner or later will be failed to provide security against the
latest sophisticated and complex attacks.
Cyber threat intelligence has gained popularity and attention by the researchers
and security analyst in recent years, and it has been considered as a quick remedial
solution to counter the latest complex attacks. Organizations have started to download
and incorporate the threat intelligence feed in their security mechanism from open
sources intelligence (OSINT) databases as well as commercial partners [4]. One of
the definitions of cyber threat intelligence is “evidence-based knowledge, including
context, processes, indicators as well as actionable advise about the current state of
the cyber connected system” [5].
This paper discusses the current state of art proactive technologies that can lead
to generation of threat intelligence. In first section of the paper, the proactive tools
helpful to generation of cyber threat intelligence are discussed including technology
benefit to cyber security. Then in Sect. 2, the solution is proposed and designed that
will lead directly as indicator for various state of art defensive security solutions. Also
the experimental results are discussed in subsequent section. The main contribution
of this research is highlighted as:
• Capabilities of Honeypot technology are addressed by the researchers to detect
zero-day attacks however but most of the time it leads to offline analysis of captured
data. In this paper, an automated proactive approach for attack data capturing,
Automated Cyber Threat Intelligence … 593

collection and processing is presented without loss of the contextual information

of attack data.
• In-built attack data classification and labelling mechanism in the system that
depicts the unclassified data further require the deeper investigation of attack
data to reduce the false alarms. Honeypot is a machine designed by the human
analyst, there may be possibility of false alarms. This problem is addressed by
machine-based learning models.
• Detection of unknown and latest attack events through regular updating of
Honeypots.
• Cyber threat intelligence generation and data sharing in standard data exchanged
formats.
• Timely dissemination and integration of threat information to defensive security
solutions.

2 Proactive Tools for Cyber Threat Intelligence

2.1 Cyber Threat Intelligence

In recent years, cyber threat intelligence (CTI) is playing a vital role in information
security but there is a lack of standardization and literature review on clarifying
the standard definition of it, every company is using his own definition of it [6].
Due to lack of standardization, it is a confusion term among many users of it. It is
can be defined as threat patterns that as evidences to support that with severity of
threat associated with it [7]. The definition of cyber threat intelligence states that
the companies can convert their processes into actions at any level of security by
considering those evidences and information as contextual records [8].

2.2 State of Art Proactive Technology

Here in this section, the proactive technology is introduced and evolution of proactive
technology in cyber security field is highlighted. Lance Spitzner, founder of Honeynet
Project, defined Honeypots as “a security resource whose value lies in being probed,
attacked or compromised [9]”.
Figure 1 depicts the evolution of Honeynet technology started from GenI to GenIII
Honeynet. The development of Honeypots began in the year of 1999 [9]. GenI (Gen-
eration I) Honeynet technology is introduced that was simple to implement and also
considered as proof of concept in Internet security. GenI Honeynet only had basic
requirements in the form of data capture and data control. Reverse firewall technique
was implemented to control the infection propagation. This firewall was simple to
set up as only design criteria is: (i) allow all inbound connection to the Honeypot
594 K. Sanjeev et al.

Fig. 1 Data control

and (ii) control outbound connection originating from Honeypot. The data capture
mechanism in GenI Honeypot was performed with IDS engine which has to monitor
the traffic traversing through the firewall to Honeypot machine.
Then, GenII (Generation II) Honeypot developed was started in the year 2002. The
main aim of Gen II Honeynet is to include the real and high-interaction Honeypot to
catch the real malwares. As this generation of Honeypots provided the real resources,
their risk increased; thereby, the data control was crucial component. This class of
system basically consists of inline firewall and an Intrusion Prevention System (IPS).
In Generation III Honeynet, the standards are introduced in the form data capture,
data control and data analysis mechanisms. The combinations of Honeypots low- and
high-interaction Honeypots are implemented with strong data control, collection and
data analysis mechanism provided by Roo-Honeywall. Figure 2 describes the various
generation of Honeynet in an abstract form.
Figure 3 depicts the broad overview and evolution of proactive technologies and
their detection vector in a general way, and it may not be complete study but to
provide the abstract overview of Honeynet technology and their detection attacks
indicators over the period of time [10].

• Honeypot as Proof of Cocept

• Low Interaction
Gen I Honeynet • Simple Data Capture
• Simple Data Control
• Real Resources
• Data Capture
Gen II Honeynet • Data Control as Reverse
Firewall
• Hybrid Honeypots
• Data Capture
Gen III Honeynet • Data Control
• Data Analysis

Fig. 2 Generation of Honeynet

Automated Cyber Threat Intelligence … 595

• Proactive
1999-2002 Technologies-
Gen I, Gen II
• Virtual
Honeynet
2005-2008 • Generation III
Honeynet
• Active Honeypot
• Client Side Exploits
2009-2012 • Automated Botnets
• Advance Botnets
• Adaptive Honeypots
2013-2017 • IoT Honeypots

• Advance
Threats
2018- • Cyber Threat
Intelligence
onwards
• Deep learning

Fig. 3 Evolution of Honeynet and their detection vector

3 Proposed Solution

3.1 Honeypot as Detection Indicators for SIEM

In security field, a new technology is introduced in recent years known as SIEM

which stands for Security Information and Event Management [11]. It is a very
helpful product having capabilities to digest security event’s raw information from
multiple sources such as syslog, firewall, IDS/IPS, analyse them and apply correlation
mechanism to them and provide the unified picture of threat to security and incident
analyst. The broad block diagram of SIEM is depicted in Fig. 4.
One scope to enhance the detection approach of SIEM or other related security
defensive solutions is use of Honeypots. Honeypots is an environment where the
vulnerabilities are intentionally introduced in order to catch the latest and complex
attacks who are trying to attack the asset of the organizations. Moreover, Honeypot
had shown the capabilities of zero-day attack detection.
The evidences captured on the Honeypot are treated as attacks infection with
low false positive and also these evidences can be used as contexts of the attacks.
These information can lead to generation of threat intelligence as detection indicator
to SIEM. By deployment of a variety of Honeypots covering different geographi-
cal locations, different types of organizations (banking, telecommunications, health,
academic, etc.) shall give a real-time visibility of the attack infection life cycle.
596 K. Sanjeev et al.

Fig. 4 Honeypot integrated with SIEM

4 Experimental Results

Table 1 describes the first level of identified threat feeds that can be generated and
converted into standard formats for data sharing platforms. The first column of
the table indicates the parsed and normalized data sets of different attack classes
such as (i) Port Scan, (ii) Connections, (iii) Bot CnC IP, (iv) Infection Source IP,

Table 1 Threat feed categories

Parsed threat feed name Threat feeds from various Honeypots
Indicator Source of feeds
Port scan IP and geo-details Passive Honeypots
Connections IP and geo-details Passive Honeypots
Bot CnC IP Command and control IP Botnet analysis applied on
Honeypot data
Infector source Bot source IP Processed data of Honeypot
Egg download IP Secondary infection IP in Bot Processed data of Honeypots
infection life cycle
Exploit IP and geo-details Passive Honeypots
Malicious URL URLs Active Honeypots
Malware Malware hash All Honeypots
Malware domain Domains Behaviour analysis
DGA domains DGA malicious domains ML unsupervised model for
DGA Botnet analysis
Automated Cyber Threat Intelligence … 597

Fig. 5 Malware distribution

Table 2 Honeypot categories

Honeypot class Types of Honeypots
Category of Descriptions
Honeypot
Dionaea Passive Honeypot Emulate ports and
services
Window 7—home Passive Honeypot Real resources
base edition

(v) Egg Download IP, (vi) Exploit IP, (vii) Malicious URLs, (viii) Malware, (ix)
DGA Domains, (x) Malware Domains and (xi) Indicator of Compromise. The attack
classes are in standard data sharing models such as STIX 2.0 [12] formats so that it
can be directly given to security solutions such as SIEM.
Figure 5 depicts the distribution of malware attacks that occurred during
01/07/2018 to 04/02/2019. These are the brief categories of malwares hashes labelled
by popular antivirus scanners. Table 2 highlights the category of Honeypot sensors
deployed during this period. Certain malware hashes have not labelled by popular
antivirus engines; hence, they need further deeper investigation to strengthen the
scope of cyber threat intelligence.
Table 2 depicts the classes of Honeypot deployed in IPv4 networks including low-
and high-interaction Honeypot sensors. First column is the class of Honeypot second
is category like passive or active Honeypot and third column depicts the descriptions
like what are the ports and services running on that Honeypot sensor.

5 Conclusion and Future Work

In this paper, automated framework for cyber threat intelligence generation using
Honeypot Technology is presented. Honeypot is a resource where the vulnerabilities
are intentionally configured to catch the latest targeted attacks; hence, there is a
598 K. Sanjeev et al.

low false positives. But we cannot take everything as low hanging fruit because it
needs further deeper investigation to avoid any mislead information to the customer,
organizations, incident response team, etc. Analytical techniques are applied on both
class of attacks (i) labelled attack detected by signature-based and heuristic-based
approaches and (ii) unlabelled attacks, i.e. not detected or missed by these techniques.
In the end, we propose a research and development of Deep Neural Network-based
cyber threat intelligence that will automatically analyse the attack data based on the
feature matrix given to them and can be fine-tuned at later stage.

Acknowledgements We would like to thank to Centre for Development of Advanced Computing

(C-DAC), Mohali for considering National Institute of Technology (NIT), Tiruchirappalli as a model
deployment location under threat intelligence generation project.

References

1. Ernst and Young Global Limited (2014) Cyber threat intelligence—how to get ahead of
cybercrime. Insights on Governance, Risk and Compliance
2. Watkins K-F (2017) M-Trends 2017: a view from the front lines, vol 4. Premier Outlook
3. Kaur Sahi Asst S. A Study of WannaCry Ransomware Attack. Int J Eng Res Comput Sci Eng.
2017;4(9):79
4. [Link]
5. [Link]
6. Boeke S, van de BDP J (2017) Cyber threat intelligence—from confusion to clarity; an
investigation into cyber threat intelligence
7. Qiang L, Zeming Y, Baoxu L, Zhengwei YJJ (2017) Framework of cyber attack attribution
based on threat intelligence. ICST Inst Comput Sci Soc Inform Telecommun Eng 190:92103
8. Alsmadi I (2019) Cyber threat analysis. In: The NICE cyber security framework. Springer,
Cham
9. Spitzner L (2003) The Honeynet project: trapping the hackers. IEEE Secur Priv Mag 1(2):15–23
10. Kumar Sanjeev et al (2012) Distributed Honeynet system using Gen III virtual Honeynet. Int
J Comput Theory Eng 4(4):537–541
11. Ailianos T (2014) SIEM optimization using Honeypots. Master Thesis
12. [Link]
Analysis of MPTCP Packet Scheduling,
The Need of Data Hungry Applications

Neha Rupesh Thakur and Ashwini S. Kunte

Abstract Multihomed devices are common in today’s environment but are underuti-
lized. Uninterrupted application requirements have leap bounds in terms of through-
put requirements. Multipath TCP (MPTCP) is a recent and successfully built standard
at transport layer, to achieve the above requirement using multipathing. Long-lived
flows carry heavy payload and short-lived flows look for quick response. Schedul-
ing algorithm should consider these requirements and accordingly implement vary-
ing strategies to fulfill these needs. Long-lived flows need MPTCP, to get maxi-
mum throughput. Short-lived flows can perform with TCP or with slow subpath of
MPTCP. To distinguish between short- and long-lived flows and distribute their traf-
fic on appropriate subflow of MPTCP, an intelligent packet scheduling algorithm is
required. Research is climbing toward building optimum scheduler for MPTCP. Many
packet scheduling algorithms are investigated in this paper for proper path selection,
increased throughput, energy efficiency, bandwidth aggregation and receiver buffer
optimization, by which issues are listed for them to develop better strategy using
newer and advanced algorithms.

Keywords Multihomed · Multipath TCP · Packet scheduling · Short- and

long-lived flows

1 Introduction

Why multipathing: TCP uses only one interface for single path communication. If
it comes across the range of powerful and cheap network-like WI-FI, then it has to
break its previous network path (4G/LTE), and it has to establish a new one. Switching
between two networks may introduce unnecessary delay and breaks. This degrades

N. R. Thakur (B) · A. S. Kunte

Department of Electronics and Telecommunication, Thadomal Shahani Engineering College,
Bandra, Mumbai, India
e-mail: neha.thakur81@[Link]
A. S. Kunte
e-mail: askunte@[Link]

© Springer Nature Singapore Pte Ltd. 2020 599

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
600 N. R. Thakur and A. S. Kunte

the ongoing application’s performance. This requirement introduces the concept of

concurrent multipath communication. Today’s multihomed devices, modern radio
access technologies and advanced wireless communication technologies are able
to provide seamless and fast communication to data hungry applications. Seamless
data reception needs multipath transmission rather than single path transmission.
This paper is mainly written to focus packet scheduling in multipath TCP. In first
section, it gives information about what is MPTCP, its evolution and how it works.
This paper gives small description of congestion control in MPTCP also, and finally
detailed study of its packet schedulers is given.

1.1 Exploration of MPTCP

Work has been done at various layers on multipath transmission. Application layer
protocol XFTP (modified File Transfer Protocol) [1] provides multiple sockets to sin-
gle path for multipath communication. Microsoft gave virtual WI-FI technology for
multipath communication. At transport layer Transmission Control Protocol (TCP)
[2], enhancement is done since last decade to give best throughput, resilience and
better performance feel in modern applications, requiring fast response. Next to TCP
other protocol built is SCTP (Stream Control Transmission Protocol) [3]. SCTP uses
TSN (Transmission Sequence Number) for flow control of multipaths. It is IETF
(Internet Engineering Task Force) standard. SCTP is UDP protocol and not compati-
ble with TCP. It uses round-robin path scheduling technique. LS-SCTP (Load Sharing
SCTP) [4, 5] uses intelligent path scheduling for multipath transmission. PATTHEL
(Parallel TCP transfer Helper) [6] gives better feel of multipath transmission but it
does not determine total number of paths available between the end hosts. MPTCP
[7] is recent and successful work done at transport layer. MPTCP, the standard of
IETF is purely built in Linux kernel. MPTCP is always restricted by middleboxes
whereas MPLEX [8] is a software architecture used for mobile multipath, which
manipulates proxy. It is transparent to server and client application. MPLEX selects
multipaths using mobile settings. In this three-way handshake is reduced to one-way
handshake thus bandwidth is properly utilized. “Table 1” shows small description of
multipath protocols in chronological order.

1.2 MPTCP

TCP uses only one IP address and network interface at the end hosts. Multihomed
devices require concurrent paths transmission for good throughput and efficient per-
formance. Multipathing is implemented mostly at transport layer since multipathing
protocols do not modify lower layer protocols and these are transparent to upper
layers. MPTCP is mainly invented for large network capable data centers. Due to the
Analysis of MPTCP Packet Scheduling … 601

Table 1 Exploration of MPTCP

Multipathing protocols Author Method/use
TCP/IP [9] Ramaboli et al. Packet switching
TCP [10] Dr. Maxemchuck Multipathing at transport layer
IETF TCP [11] Huitems, Christian TCP multipathing
TCP [12] Key, Peter et al. Stable and balance congestion
control for multipath TCP
TCP [13] Shakkottai, Srinivas et al. Benefit of multipathing for
multihomed gaming and good
for Internet service providers
SCTP [3] Stewart, Randall Multipathing without
considering fairness with TCP
W-SCTP [14] C. Casetti and W. Gaiotto Bandwidth aggregation for
proper traffic allocation
CMT-SCTP [15] R. Iyengar, Janardhan et al. Fast and less retransmission
using delayed acknowledgment
for congestion control
CMT-SCTP [16] Liu, Jiemii et al. Avoidance of receiver buffer
bloating by using RTX-LCS
retransmission policy
LS-SCTP [17] Abd, Ahmed, Tarek Saadawi Flow and congestion control
et al. using association and path
sequence number uses
(cwnd/RTT) factor to find weight
of each subpath
CMP-SCTP [18] Liao, Jianxin et al. Flow and congestion control
using association and path
sequence number uses
bandwidth of each subpath to
find weight of each subpath
WIMP-SCTP[19] Huang, Chung-Ming et al. Wireless multipath SCTP,
increases throughput by using
two modes of transmission
PTCP [20] Hsieh, Hung-Yun et al. Micropath switching using
bandwidth factor
M-TCP [21] Chai, Jiwei, Kaixin Xu et al. Sender side multipathing in
heterogeneous lossy networks
MELOT [22] Sharma, Vicky et al. Multipath loss tolerant protocol
heterogeneous lossy networks
TCP-MPTCP [23] Raiciu, Costin, et al. Improved throughput, aggregated
bandwidth for data center
environment
MPTCP [24] Raiciu, Costin et al. Fairness to TCP
MPTCP-MA [25] Lim, Yeon-sup et al. Improvement in MPTCP for
wireless communication
602 N. R. Thakur and A. S. Kunte

Fig. 1 TCP versus MPTC

multihoming features of devices, MPTCP is further implemented for wired and wire-
less communications. MPTCP is built on TCP and not on UDP [26]. Figure 1 shows
the network layer architecture of [26] TCP versus MPTCP. MPTCP uses many paths
with different interfaces having different IPs between two hosts. Each path of MPTCP
is single path TCP (SPTCP). Host to host connection of each subpath is formed by
three-way handshake as that of TCP. Multipaths combine into a single socket at end
host. Number of paths in MPTCP are equal to number of interfaces available at
present. To avail multipath facility multihomed devices should be MPTCP capable.
“Figure 2” shows environment of MPTCP. Multipaths give high-quality end to end
(e2e) performance and throughput as well as it can be used as backup. MPTCP uses
two types of sequence numbers subflow and data sequence numbers [27]. Sequence
numbers are helpful in two ways, for loss detection, retransmission and reassembly
of out of order packets at receiver side. Currently, MPTCP is used in mobile IOS built
in applications like SIRI and Korean GIGA LTE which is able to give Gbps through-
put for local mobile communication. It is also used in other applications for single

Fig. 2 MPTCP environment

Analysis of MPTCP Packet Scheduling … 603

and multifile download. These applications are Voice over IP (Skype, Google Hang-
out), Web browsing (Google search), Video streaming (YouTube, Netflix) and Instant
Messengers (Facebook, Google, Hangout, WhatsApp). MPTCP is built on TCP. TCP
header is not changed for MPTCP. Among all subflows of MPTCP, each individual
subflow behaves exactly same as SPTCP. MPTCP works with cycle of three things
establish connection, maintain connection and end up connection. MPTCP connec-
tion establishment is shown in “Fig. 3.” It establishes connection like SPTCP. It
uses SYN/ACK (request/response) for each subflow connection establishment. If
two network interfaces are available, then MPTCP connection can have two sub-
flows between server and user [26]. Considering two subflows as AB and CD. Firstly
server sends MP_CAPABLE signal. Saying server is MPTCP capable from end A
to end B (user) along with request of connection (SYN) and encrypted key of A.
On the other hand, after receiving these information from A, B understands A is
requesting connection and if B is free and not MPTCP compatible, it continues as
SPTCP; otherwise, it sends MP_CAPABLE, SYN and ACK (acknowledgment) of
B and encrypted key of B toward A. On the reception of these information from
B, A sends ACK and encrypted keys of A and B. This way AB connection estab-
lishes. Additional connection can be added from any side but usually server initiates
new connection by ADD_ADDR. Server sends MP_JOIN, nonce (random number
to avoid replay attack), IP of C, security key of C and flag to distinguish primary
path or backup path for establishing additional path CD. D then sends MP_JOIN,
security key of D, authentication message and random number to C. C after receiv-
ing these sends ACK and its authentication message to D. Finally, D sends ACK
to C and establishes the additional connection of different interface network. While
establishing additional connection, user can discard the IP address by using NATs.
MPTCP maintains connection by using packet scheduling and congestion control

Fig. 3 MPTCP connection establishment

604 N. R. Thakur and A. S. Kunte

Fig. 4 MPTCP congestion control framework

algorithms. Packet scheduling distributes packets on different subflows. Congestion

control diverts packets from congested subflows to less congested or free subflows.
MPTCP connection ends with FIN flag, when the entire data transfer on that subflow
finishes. It can also be interrupted and ended with RST (reset flag). Cascading buffers
are mediators between application and transport layer [28]. These buffers are useful
in communication.

1.3 MPTCP’s Requirements

Congestion control. Congestion control/load balancing is required in MPTCP to

increase the effectiveness of multipathing. MPTCP’s sender congestion window
(cwnd) size should be less than receiver cwnd size. Process is shown in “Fig. 4.”
TCP NewReno [29] based, TCP subflow congestion control framework [30] works in
three steps, first is Slowstart (ss), in this cwnd size is blindly doubled for each round
trip time when TCP just starts building connection. Flow control is not done here. It
persists for a very short time. It ends when cwnd size climbs to its maximum limit,
slow start threshold (ssthresh) parameter, or when very first packet is lost. In second
step, congestion avoidance is done. It starts just after first step. Flow control is done
using different congestion control methods. Third step is fast retransmit/immediate
recovery. In this step, cwnd size is freezed. New packet insertion is also stopped.
Only retransmission of lost packets is carried out.
MPTCP behavior with short-lived and long-lived flows. Short-lived flows carry
lightweight data. For long-lived flows, MPTCP is blessing but for short-lived flows
it is a curse. Google search (web browsing), WhatsApp messaging, SMS (Short
Analysis of MPTCP Packet Scheduling … 605

Messaging System) and other mobile applications generate short-lived flows. Short-
lived flows are delay sensitive rather than bandwidth and throughput sensitive thus
lowest RTT path is selected. When it is unavailable next lowest RTT, path is chosen
but it may not always be accurate. If one connection has two subflows of 10 and
100 ms with congestion window size of 10 but 11 packets are queued at the sender
buffer, then 10 packets will be transferred on 10 ms path and others will be sent on
100 ms path. Total delay is 100 ms. Delay can be reduced to 20 ms if we wait for
10 ms subflow instead of using 100 ms path. These two are heterogeneous paths
and may generate many out of order packets. Out of order packets require more
retransmissions and more receiver buffer. This phenomenon is called as head of line
blocking. But after timeout, entire data unit may be discarded due to missing packets
and the throughput decreases.
Packet scheduling. MPTCP uses concurrent diverse paths for transmission. These
subflows differ in path characteristics like RTT, cwnd, delay and packet loss rate.
Due to path diversity, small sequence number packets may reach receiver after high
sequence number packets. They lead to many out of order packets at the receiver,
many retransmission, many data units rejection and ultimately streaming quality may
not be good. Due to delay factor, many packets are discarded with late successful
retransmission. Receiver buffer is limited in size, so buffer overflow or head of line
blocking may occur at the receiver. To avoid this problem of buffer bloating [31] and
head of line blocking, smart and intelligent packet scheduler is required. MPTCP
packet scheduling is packet oriented. MPTCP IETF standard uses two packet sched-
ulers, default and round-robin. Default scheduler always opts minimum RTT subflow
whereas round-robin scheduler gives data transmission chance to every subflow in
round-robin manner. Some of the packet schedulers are surveyed and issues are listed
in “Table 2.”

2 Limitations of Existing MPTCP

Long and short flows need to be discriminated properly. Delay should be considered
in heterogeneous multipathing and retransmissions. Stringent reliability in video
streaming and flexible multimedia MPTCP Architecture is needed and receiver buffer
overflow should be managed.

3 Conclusion

MPTCP is capable of providing maximum throughput and uninterrupted transmis-

sion for huge data-carrying applications. In huge data-centric applications like video
streaming applications, delay as well as performance is crucial. Discrimination of
packet types is essential to distribute them on different appropriate subflows for
concurrent transmission in MPTCP. Default scheduling methods will not use much
Table 2 Survey of multipath packet scheduling
606

S. No. Problem definition Method Advantage Scope

1 Bandwidth aggregation in mobile Mobile IP is used to achieve Bandwidth optimization and Path losses are not considered
wireless technology [32] objective. Equisize packets are proper packet scheduling on
scheduled on link until idle link multipaths are achieved. Path
gets packet. Scheduler uses overheads and retransmissions are
weighted queue striped on reduced
channels
2 Efficient multipath transmission Finds scores of paths. Minimum Packet transmission is done Lack of intelligence in
[14] score path is selected for considering width of congestion transmitting specific packet on
transmission. Multiple send windows and bandwidth of specific path from queued sender
buffers are used subpath buffers
3 Delay aware throughput For each subflow, it measures Along with increased throughput, It is good for packet scheduling
optimization [33] combined delay of sender TCP this algorithm can find path failure but throughput is below ideal
queue and network delay. From and do recovery detection. throughput because of heavy
delay, it estimates arrival time of Missing packets are also packet loss
packets at receiver and recovered
accordingly schedules new
packets at sender
(continued)
N. R. Thakur and A. S. Kunte
Table 2 (continued)
S. No. Problem definition Method Advantage Scope
4 Augmentation of SCTP to Packet reordering is done at Along with increased throughput, Although concurrent multipathing
improve performance in sender. It splits fast retransmit. this algorithm can do path failure is used, it uses only one path for
multihoming to serve concurrent Virtual queues are formed at and recovery detection traffic and others are used for
multipath situations [15] sender for retransmission queue. backup or retransmission
Two variables are used one is to
save ACK of packet and other is
to store destination of packet (sent
for first time). Packet sent after for
missing packet to same
destination is acknowledged
newly
Analysis of MPTCP Packet Scheduling …

5 Multipathing using UDP protocol Transport address is IP address Multipathing is achieved using the Even if it is multipath protocol, it
at transport layer [3] and SCTP port number. User data UDP algorithm uses only single path for
is fragmented and chunks of data transmission at a time. It is not
are built. Packet validation is done compatible with TCP
for path management. Two types
of sequence numbers one for data
and other for subflows are used.
Congestion avoidance is done
using receiver acknowledgment
6 Efficient transfer of data on Data is split into chunks and Without doing any change in the Losses on wireless links are not
multipaths using TCP [6] spread on different TCP channels protocol stack, the algorithm is considered
based on different active channels. implemented
Channel is assigned to active
interfaces
(continued)
607
Table 2 (continued)
608

S. No. Problem definition Method Advantage Scope

7 Reduced out of order packets at Predicts how many data packets Smart scheduler compared to During path selection, actual path
the receiver with concurrent flows should be placed on each subpath SCTP and MPTCP quality is not considered
at the transport layer [34] based on delay
8 Building real-time multimedia Parallel subpaths are formed Real-time multimedia traffic has Real-time competition and rate
environment using TCP [35] between a distributed server vast bandwidth variations on changes of TCP flows cannot be
replicas and multimedia client. subpaths, still TCP-ROME managed
RTT is used to control and increases quality of experience
schedule multimedia streams on
different paths. Coordination
module coordinates content and
rate. Expected video streaming
time is estimated to control
real-time traffic
9 Estimation of RTT by using Estimation of RTT is done by Improves multipathing efficiency This algorithms are applied to
machine learning [36] many experts. Finally, average of TCP by 40% and TCP but can be implemented for
RTT is taken and correction in retransmission also reduces by MPTCP also
new RTT is done by considering 30%
current RTT
10 Concurrent transmission on Uses two types of packet In multihoming scenario, MPTCP All path characteristics are not
multiple paths (multiple schedulers: default: uses lowest achieves throughput at least that considered for path scheduling
interfaces) to get better RTT path in round-robin fashion. of normal TCP
throughput [7] Round-robin: subflows are
selected randomly for data
transmission in round-robin
manner
(continued)
N. R. Thakur and A. S. Kunte
Table 2 (continued)
S. No. Problem definition Method Advantage Scope
11 Improvement of throughput and Even if congestion window is not If packets are not over sent at It is not good for bursty traffic.
receiver buffer efficiency in available on subflows, subflows receiver, then there is Receiver buffer needs to be used
dynamic scenarios like gaming to are used to schedule the data. improvement in the throughput as carefully than default
reduce undesirable delay [37] Scheduling time is calculated for compared to MPTCP default
each path. Scheduled data packet scheduler
is not sent immediately on path. If
congestion window get available,
scheduled packet is sent on the
path
12 (N + 1)th packet is estimated for (N + 1)th packet is scheduled on Gives more robustness in lossy Bursty path losses are not
slow, eligible subflow. under scheduling subflow. network considered
Analysis of MPTCP Packet Scheduling …

Decision-making is done for each Concurrently, N packets are

packet on different available scheduled on other eligible
subflows [38] subflows. Loss indication is done
by triple duplicate ACKs. Loss
rate of packet is different for each
packet
13 Throughput optimization and Adaption of congestion window is Decrease in out of order packets at Recovery of missing and delayed
congestion window matching for done by using path delay, RTT the receiver. Receiver buffer size packets is not coordinated. Out of
all subflows [39] and congestion window of each is saved order packets are also decreases
subflow, expected throughput is
calculated for each subflow. If
adaption time is exceeding
maximum adaption time, the path
is blocked for some time. Pull and
push method is used
(continued)
609
Table 2 (continued)
610

S. No. Problem definition Method Advantage Scope

14 Feedback SACK (selective Estimation of N number of More adaptive in wireless Path losses and bursty traffic are
acknowledgment) of out of order packets on faster subflows. Two networks not considered
packet is used for scheduling of different situations are handled, N
current packets [40] is too small and N is too large
15 High-quality mobile streaming in Uses coupled congestion. FEC Arranges out of order packets into Energy consumption and
heterogeneous network using code adjusts packet size and order. Multimedia interrupted retransmission are not considered
coding and rate allocation. redundancy at rate allocator. FEC output is avoided by hiding of
Forward error correction (FEC) is coder checks path status (RTT, frame copy error at application
used for coding [41] delay) for window adaption. FEC layer. Path loss and delay on
packets wrap encoded video subflows are reduced
packets. Receiver has informative
feedback unit, which provides
RTT, bandwidth, packet loss of
subflows to scheduling algorithm
at sender
16 Improvement of existing Finds maximum RTT path. Improvement over MPTCP In scheduler, all path
scheduler and building scheduler Checks path as fresh or characteristics are not considered
for high performance in multipath retransmitted. Retransmission is
applications [42] done on different subflow.
Congestion window is checked
for each path. Path transmission
time and number of packets on
subflows will be calculated
(continued)
N. R. Thakur and A. S. Kunte
Table 2 (continued)
S. No. Problem definition Method Advantage Scope
17 Serving of short flows within less If the RTT difference between two It uses default MPTCP Slow path is unused until
time by considering delay and subflows is considerably more, round-robin strategy. Data congestion window of fast path
bandwidth [27] then it freezes subflow having delivery is always done on fastest gets saturated
large RTT for short-lived flows. path and always gives throughput
This is done regardless of their at least equal to SPTCP
congestion window
18 Enhancement of MPTCP for It considers satellite network is Packets are forwarded to Automotive computing is not
multimedia services in always available. Along with application layer after RTO used. Slow start and loss of
multihomed vehicular networks to MPTCP default round-robin without waiting for missing packets cause poor performance
get sustained throughput when scheduling, deep packet packets. Efficient handoffs and
bandwidth is changing [43] inspection by group of pictures is reduced jitter are achieved
Analysis of MPTCP Packet Scheduling …

also used. Selective discarding

and partial reliability of packets
are used
19 MPLEX design for multipath Middlebox proxy programming is Bandwidth is saved. Instead of Every middlebox should be
transmission [8] done for multipath selection three-way handshake only modified
one-way handshake is required.
63% improvement over
short-lived flows
(continued)
611
Table 2 (continued)
612

S. No. Problem definition Method Advantage Scope

20 Maximization of video data Integer Linear Program is Algorithm is independent of Not that efficient as optimal
received on time at the receiver executed at scheduler. Four applications. Uses same MPTCP schedulers
[25] constraints are used. Video block at transport layer
is received only when exact
number of different packets are
obtained. Packets can send data of
same video block. Decoding
happens only if video block is
received on which it is dependent.
Video block is transmitted by
packets if it is received before
decoding deadline
21 Implementation of solution in Designed by using simple or The best analysis is done for Algorithm is implemented using
wireless heterogeneous radios, machine learning algorithms. better understanding and many algorithms thus it is
using concurrent subpaths, which Keeps track on interface experimentation complex
is throughput aware and energy utilization and checks for
efficient [44] utilization above threshold. It uses
five different algorithms for
prediction
22 When Router’s buffer size is Routers buffer size, RTT, Ack are Routers buffer consideration for Packet losses and retransmissions
small, throughput of MPTCP analyzed for transmission of new path with other path are not considered for path
should not drop at least as that of packets on the subflow characteristics
SPTCP [45]
(continued)
N. R. Thakur and A. S. Kunte
Table 2 (continued)
S. No. Problem definition Method Advantage Scope
23 Implementation of earliest It considers RTT, bandwidth, bit Multimedia transmission is done Negligible degradation of quality
completion first algorithm for rate, congestion window and in less time and with less error at of experience thus algorithm is
packet scheduling using buffer size of connection. If the the receiver. Web download and not 100% perfect
maximum path characteristics and fastest subflow is not available, browsing also became fast
not only RTT [46] packet is scheduled on second
fastest path
24 Better video streaming by using Three algorithms are used: (1) Algorithm works for video Scheduler uses dynamic path
utilization of network short packet delay, (2) largest streaming over MPTCP. It is rare characteristics. Random losses are
performances measures [29] packet credit and (3) largest topic not considered
estimated throughput
25 Optimization of throughput by Use of quality interpreter and High-level API is built. Seven Scheduler execution time is
Analysis of MPTCP Packet Scheduling …

using scheduling of efficient run time environment. novel scheduler objectives are challenging. At application layer,
application-oriented packets on API—per packet scheduling is achieved less number of schedulers are
time at the receiver [47] done. Differentiate small and tested
large data requests. Scheduler
selection in 100 ns. Load
balancing in scheduler by less
overheads. One scheduler with
different modes with registers.
Before stoping path, it is flushed
(continued)
613
Table 2 (continued)
614

S. No. Problem definition Method Advantage Scope

26 Efficient multipathing to increase Optimum path characteristics are Access point interference is Efficient, adaptive and more
throughput by using learning [48] chosen. Optimization of WI-FI reduced. Algorithm is accurate, realistic scheduling algorithm can
path. According to feedback, stable and reduces overfitting. be built
result path is managed or aborted More predictive, uses less
overhead
27 Packet scheduling in MPTCP It estimates packet on each Algorithm considers packet losses Small probabilities are neglected
considering packet losses and concurrent subflows. It considers and path characteristics (RTT, thus retransmission packets and
feedback from the receiver [49] feedback from the receiver to cwnd). Receiver feedback is packets after timeout are not
change the estimation of packets considered for proper packet flows considered
on the master subflow. Dynamic
path characteristics are
considered. Modeling time of
master subflow is estimated.
Estimation is improved by error
offset
N. R. Thakur and A. S. Kunte
Analysis of MPTCP Packet Scheduling … 615

intelligence for packet scheduling on different subflows also they do not consider
various path losses and different environmental conditions. Using more intelligence
and machine learning module, path scheduling can be studied in more better way.

References

1. Allman M, Kruse H, Ostermann S (1996) An application-level solution to TCP’s satellite

inefficiencies. In: Proceedings of the first international workshop on satellite-based information
services (WOSBIS), (XFTP)
2. Duke M et al (2006) RFC 4614: a roadmap for transmission control protocol (TCP) specification
documents. IETF Internet Standard
3. Stewart R (2007) Stream control transmission protocol. No. RFC 4960
4. Abd A, Saadawi T, Lee M (2004) LS-SCTP: a bandwidth aggregation technique for stream
control transmission protocol. Comput Commun 27(10):1012–1024
5. Amer PD, Ekiz N, Natarajan P, Becke M, Tuexen M, Dreibholz T, Stewart RR, Iyengar J (2019)
Load sharing for the stream control transmission protocol (SCTP)
6. Baldini A, De Carli L, Risso F (2009) Increasing performances of TCP data transfers through
multiple parallel connections. In: 2009 ieee symposium on computers and communications.
IEEE 2009
7. Bonaventure R et al (2012) An overview of multipath TCP. Login 37(5):17
8. Nikravesh A et al (2016) An in-depth understanding of multipath TCP on mobile devices:
measurement and system design. In: Proceedings of the 22nd annual international conference
on mobile computing and networking. ACM 2016
9. Ramaboli AL, Olabisi EF et al (2012) Bandwidth aggregation in heterogeneous wireless net-
works: a survey of current approaches and issues. J Netw Comput Appl 35(6):1674–1690
(TCP/IP packet scheduling)
10. Maxemchuk N (1975) Dispersity Routing in Store-and-Forward Networks
11. Huitema C (1995) Multi-homed TCP. Internet Draft IETF
12. Key P, Massoulié L, Towsley D (2006) Combining multipath routing and congestion control
for robustness. In: Conference on information sciences and systems
13. Shakkottai S, Altman E, Kumar A (2006) The case for non-cooperative multihoming of users
to access points in IEEE 802.11 WLANs. In: Proceedings IEEE INFOCOM 2006. 25TH IEEE
international conference on computer communications. IEEE 2006
14. Casetti C, Gaiotto W (2004) Westwood SCTP: load balancing over multipaths using bandwidth-
aware source scheduling. In: IEEE 60th vehicular technology conference 2004. IEEE, vol 4,
VTC2004-Fall
15. Iyengar JR, Amer PD, Stewart R (2006) Concurrent multipath transfer using SCTP multihoming
over independent end-to-end paths. IEEE/ACM Trans Netw 14(5):951–964
16. Liu J et al (2008) Rethinking retransmission policy in concurrent multipath transfer. In: 2008
International conference on intelligent information hiding and multimedia signal processing,
(RTX-LCS), IEEE 2008
17. Abd A, Saadawi T, Lee M (2004) Improving throughput and reliability in mobile wireless
networks via transport layer bandwidth aggregation. Comput Netw 46(5):635–649
18. Liao J, Wang J, Zhu X (2008) cmpSCTP: an extension of SCTP to support concurrent multi-path
transfer. In: 2008 IEEE international conference on communications. IEEE 2008
19. Huang C-M, Tsai C-H (2007) WiMP-SCTP: multi-path transmission using stream control trans-
mission protocol (SCTP) in wireless networks. In: 21st International conference on advanced
information networking and applications workshops (AINAW’07), IEEE, vol 1
20. Hsieh H-Y, Sivakumar R (2002) pTCP: an end-to-end transport layer protocol for striped
connections. In: 10th IEEE international conference on network protocols 2002. Proceedings
IEEE
616 N. R. Thakur and A. S. Kunte

21. Chen J, Xu K, Gerla M (2004) Multipath TCP in lossy wireless environment. In: Proceedings
of IFIP third annual Mediterranean ad hoc networking workshop (Med-Hoc-Net’04)
22. Sharma V et al (2008) MPLOT: a transport protocol exploiting multipath diversity using erasure
codes. In: IEEE INFOCOM 2008-the 27th conference on computer communications. IEEE,
178
23. Raiciu C et al (2010) Data center networking with multipath TCP. In: Proceedings of the 9th
ACM SIGCOMM workshop on hot topics in networks. ACM 2010
24. Raiciu C et al (2011) Improving datacenter performance and robustness with multipath TCP.
ACM SIGCOMM Comput Commun Rev 41(4)
25. Lim Y-S et al (2014) Cross-layer path management in multi-path transport protocol for mobile
devices. In: IEEE INFOCOM 2014-IEEE conference on computer communications. IEEE 2014
26. Van Der Pol R et al. (2012) Multipathing with MPTCP and OpenFlow. In: 2012 SC companion:
high performance computing, networking storage and analysis. IEEE 2012
27. Hwang J, Joon Y (2015) Packet scheduling for multipath TCP. In: 2015 Seventh international
conference on Ubiquitous and future networks. IEEE 2015
28. Corbillon X et al (2016) Cross-layer scheduler for video streaming over MPTCP. In:
Proceedings of the 7th international conference on multimedia systems. ACM 2016
29. Allman M, Paxson V, Stevens W (1999) TCP Congestion control IETF RFC 2581, (TCP
NewReno)
30. Matsufuji R et al (2017) Multipath TCP path schedulers for streaming video. In: 2017 IEEE
Pacific Rim conference on communications, computers and signal processing (PACRIM). IEEE
2017
31. Polese M et al (2018) A survey on recent advances in transport layer protocols. In: arXiv
preprint arXiv:1810.03884
32. Hasegawa Y et al (2005) Improved data distribution for multipath TCP communication.
GLOBECOM’05 IEEE Global Telecommun Conf 1:5
33. Mirani FH, Boukhatem N, Tran MA (2010) A data-scheduling mechanism for multi-homed
mobile terminals with disparate link latencies. In: 2010 IEEE 72nd vehicular technology
conference-fall. IEEE 2010
34. Park J-W, Karrer RP, Kim J (2011) TCP-ROME: a transport-layer parallel streaming protocol
for real-time online multimedia environments. J Commun Netw 13(3):277–285
35. Nunes BAA et al (2011) A machine learning approach to end-to-end RTT estimation and
its application to tcp. In: 2011 Proceedings of 20th international conference on computer
communications and networks (ICCCN). IEEE 2011
36. Yang F, Wang Q, Amer PD (2014) Out-of-order transmission for in-order arrival scheduling for
multipath TCP. In: 2014 28th International conference on advanced information networking
and applications workshops. IEEE 2014
37. Ni D et al (2014) Fine-grained forward prediction based dynamic packet scheduling mechanism
for multipath TCP in lossy networks. In: 2014 23rd International conference on computer
communication and networks (ICCCN). IEEE 2014
38. Bhat PA, Talmale G (2014) MPTCP combining congestion window adaptation and packet
scheduling for multi-homed device. In: International conference for convergence for technol-
ogy. IEEE 2014
39. Ni D et al (2015) OCPS: Offset compensation based packet scheduling mechanism for multipath
TCP. In: 2015 IEEE international conference on communications (ICC). IEEE 2015
40. Wu J et al (2016) Streaming high-quality mobile video with multipath TCP in heterogeneous
wireless networks. IEEE Trans Mob Comput 15(9):2345–2361
41. Popat KJ, Raval JA, Johnson S et al (2015) An efficient scheduling scheme of multipath TCP
for MPI. Int J Sci Eng Technol Res (IJSETR) 4(6):2123–2126
42. Rene S et al (2015) Multipath TCP architecture for infotainment multimedia applications in
vehicular networks. In: 2015 IEEE 81st vehicular technology conference (VTC Spring). IEEE
2015
43. Saputra Y et al (2017) E-MICE: energy-efficient concurrent exploitation of multiple Wi-Fi
radios. IEEE Trans Mob Comput 16(7):1870–1880
Analysis of MPTCP Packet Scheduling … 617

44. Kim H, Choi S (2016) The effect of routing path buffer size on throughput of multipath TCP.
In: 2016 International conference on information and communication technology convergence
(ICTC). IEEE 2016
45. Lim Y-S et al ECF: an MPTCP path scheduler to manage heterogeneous paths. In: Proceedings
of the 13th international conference on emerging networking experiments and technologies.
ACM 2017
46. Frömmgen A et al (2017) A programming model for application-defined multipath TCP
scheduling. In: Proceedings of the 18th ACM/IFIP/USENIX Middleware conference. ACM
2017
47. Chung J et al (2017) Machine learning based path management for mobile devices over MPTCP.
In: 2017 IEEE international conference on big data and smart computing (BigComp). IEEE
2017
48. Xue K et al (2018) DPSAF: forward prediction based dynamic packet scheduling and adjusting
with feedback for multipath TCP in lossy heterogeneous networks. IEEE Trans Veh Technol
67(2):1521–1534
49. Ferlin S et al (2018) MPTCP Meets FEC: supporting latency-sensitive applications over
heterogeneous networks. IEEE/ACM Trans Netw (TON) 26(5):2005–2018
Voice Controlled Home Automation
Using Blynk, IFTTT with Live Feedback

M. Rajasekhar Reddy, P. Sai Siddartha Reddy, Suthapalli Akhil Sri Harsha

and D. Vishnu Vashista

Abstract As people life becomes easier, people are habituated to the comforts in
order to meet their needs. One such invention is home automation. This area has
many opportunities which are emerging every day. The proposed technique will give
home automation with a cost-effective implementation. We can automate the lights,
fans, refrigerator, etc. The requirements are NodeMcu, relay module, smartphone
with Blynk and IFTTT apps, and proper Wi-Fi source.

Keywords Home automation · IFTTT · Blynk · IoT · Analog extender · NodeMcu

1 Introduction

IoT is one of the latest technologies emerging in the modern world [1, 2]. The recent
development on the Internet creates many opportunities for people to come up with
new ideas. One such idea is the home automation. There are several such automation
techniques where people require more and easier ways to use it in their day-to-day
life. In order to bring comfort to the user, the Google Assistant is used as a translator.
Google Assistant is integrated with the circuit by using the If This Then That (IFTTT),
and it acts as the intermediate platform.

M. Rajasekhar Reddy (B) · P. Sai Siddartha Reddy

School of Computing, SASTRA Deemed to Be University, Thanjavur, India
e-mail: rajasekharmanyam04@[Link]
P. Sai Siddartha Reddy
e-mail: siddusiddartha1999@[Link]
S. Akhil Sri Harsha · D. Vishnu Vashista
School of Electrical and Electronics Engineering, SASTRA Deemed to Be University, Thanjavur,
India
e-mail: akhilsuthapalli@[Link]
D. Vishnu Vashista
e-mail: vishnuvasista99@[Link]

© Springer Nature Singapore Pte Ltd. 2020 619

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
620 M. Rajasekhar Reddy et al.

The proposed system will purely come under the Internet of Things (IoT), in which
the different modules (such as lights, fans, refrigerators, etc., with the smartphones)
can be connected.
The overview of the paper is as follows: Sect. 2 describes the prerequisites.
Section 3 describes the past and present existing techniques which are being used.
Section 4 explains the proposed method with complete analysis. Section 5 describes
the results and analysis. Section 6 concludes this paper.

2 Related Work

2.1 NodeMcu

It is an open-source platform which is mainly used for IoT. It has a Soc ESP8266
manufactured by Espressif Systems [3]. The main features of NodeMcu are it’s an
open source and Wi-Fi enabled. The term MCU stands for microcontroller unit on a
single chip. It has many processors along with memory and several programmable
input/output pins. These i/o pins also called as general purpose input/output (GPIO)
are used to interface many sensors and other electronic modules to achieve a particular
task. The code is created in Arduino IDE with pre-installed libraries. Later that code is
compiled and uploaded to the NodeMcu. NodeMcu has the following specifications:
• Maximum output voltage: 3.3 V.
• Wi-Fi Direct (P2P), soft-AP.
• Analog to digital: one input with 1024 step resolution.
• Built-in USB connector (Fig. 1).

2.2 Relay Module

A relay is an electromagnetic switch which can turn ON/OFF a much larger electric
current when a relatively small current is applied to its terminals [5]. The main
working component of a relay is an electromagnet, which is a coil of wire that
excites and creates a temporary magnetic field when current is passed.
The following are the specifications of a relay module in Fig. 2.
• High-voltage side current 10 A and 250 V AC or 30 V DC.
• Source voltage 5–12 V.
• For each channel, the control signal is 3–5 V.
• For each channel 1- normally open, 1-normally closed and 1-common.
Voice Controlled Home Automation Using Blynk, IFTTT with Live … 621

Fig. 1 Pin configuration of NodeMcu [4]

Fig. 2 Relay module [6]

2.3 Sensing Circuit

The sensing circuit works as a feedback circuit wherein the power status of the
appliances connected to the relay is detected. The circuit mainly consists of a potential
divider along with a diode and a capacitor as shown in Fig. 3.
Working. The input to the circuit is a 230 V AC supply wherein the drop across
the 4Kohm is brought nearly to 5 V AC which is passed through a diode and a
capacitor so as to generate voltage level of 5 V which is passed to the analog input of
NodeMcu. A high-level voltage indicates switch ON condition of the appliance. The
NodeMcu is programmed in such a way that the data from analog input is processed
and the notification of whether the device is ON or OFF.
622 M. Rajasekhar Reddy et al.

Fig. 3 Sensing circuit

2.4 Analog Pin Extender (ADS1115)

The number of analog pins in NodeMcu is only one. In order to connect more devices,
an external device must be used. ADS1115 acts as an analog pin extender increasing
the number of analog pins on which NodeMcu can work to 4. ADS1115 performs
communication with NodeMcu and transfer the data when triggered.

2.5 Blynk

Blynk is a platform with iOS and Android apps to control Arduino, Raspberry Pi,
NodeMcu and many other devices over the Internet [7]. It is a digital dashboard where
you can build a graphic interface for your project by simply dragging and dropping
widgets. For android mobiles, it is available in Google Play Store. The auth_token
generated gives the user, protection from others controlling the NodeMcu. In the
project window, we are given with a toolbox where we can add several widgets like
buttons, terminal, value display, notification, etc. Upon adding the required widgets,
the user needs to configure the respective widget as to which pin of NodeMcu is to
be controlled using that widget. There is a play button which starts the communica-
tion between Blynk and NodeMcu. And the operation can be stopped by clicking a
rectangular stop box on top.

2.6 IFTTT

‘If This Then That’ (IFTTT) is a free web-based service to create chains of simple
conditional statements, called applets [8]. An applet is triggered by changes that occur
within other web services such as Gmail, Facebook, Instagram, Google Assistant,
Voice Controlled Home Automation Using Blynk, IFTTT with Live … 623

etc. In this paper, Google Assistant accesses the Blynk cloud. Since Blynk has no web
service associated with it, webhooks are used with the IP address of Blynk followed
with a particular auth_token.

3 Existing Techniques

3.1 Past Techniques

Bluetooth. Android apps are used to send data via Bluetooth, and the received data
is analyzed in Arduino which responses accordingly. Those apps could be created
using various app development software like MIT app inventor, Android studio, etc.
The drawbacks are limited range and sync rate [9].
ZigBee. The above problem is overtaken if we use ZigBee for long range and it
has its own limitations as it could not provide access worldwide [10].
DTMF. In ‘Dual Tone Multi-Frequency’ the numbers can be sent from one device
to other over a phone call. On receiving the digit, pressed the decoder detects the
frequency of the digit and sends relative signals to control devices.

3.2 Present Techniques

Privatization. Many companies are now developing their own software and
controller circuits to control home appliances. This technique is costly.
Blynk. It is a cloud service provided by Blynk. The microcontroller is connected
to its server with a unique auth code. It has features like GPS, notification alert, live
data transmission, etc. The code needed reduces enormously.
IoT. The data can be trans-received worldwide and can be used in various places
which have Internet access using IoT [11]. The rapid growth in IoT is due to the
in-hand operation, reliability, extendability, and connecting a range of devices.

4 Proposed Technique

Keeping the circuit ready, this cost-effective technique provides a solution to home
appliances control, fetching data from Blynk app as well as from various platforms.
To work with those platforms, a service connector known as IFTTT integrates to
perform various actions when triggered. IFTTT, Blynk, and Google Assistant are
integrated such that IFTTT gets data from Google Assistant and sends to Blynk cloud.
The Blynk server transmits data to NodeMcu. Then, microcontroller controls various
appliances connected to it via a relay module. A sensing circuit is adopted which
624 M. Rajasekhar Reddy et al.

Fig. 4 Overall architecture of the proposed method

basically detects the voltage applied, ensuring perfect feedback on the appliance
status. The overall architecture of the proposed method is shown in Fig. 4. The
procedure can be written as an algorithm.

Algorithm: Home_Automation()

1. Install Arduino software with NodeMcu, Blynk, ADS1115 libraries in laptop

2. Install Blynk app in the smartphone

2.1. Select new project and set it to NodeMcu

2.2. Burn the code (Sect. 4.1) with network credentials and auth_token

3. Create an IFTTT account

3.1. Create a new applet inside IFTTT as specified.
3.2. Connect Google Assistant with Blynk server.

4. Connect the relay and sensing circuit as shown in Fig. 4.

5. Place an internet hotspot access point.
1. Create a new applet inside IFTTT as specified.
2. Connect Google Assistant with Blynk server.
Voice Controlled Home Automation Using Blynk, IFTTT with Live … 625

End Home_Automation

This work is split into five steps to set up the automation. They are as follows:
(a) software code (b) Blynk app configuration (c) IFTTT setup (d) hardware (e)
simulation of the sensing circuit.

4.1 Software Code

The programming is done in Arduino IDE Software, which is an open-source applica-

tion. In Fig. 5, the necessary packages which are required for a NodeMcu to connect
with the Blynk server [12] and the analog extender ADS1115 are included. The cre-
dential details of the internet hotspot device along with the auth_token are declared
(as shown in Fig. 6) to connect the NodeMcu to the hotspot. Declare a function
called ‘powerstatus()’ (as shown in Fig. 7), which gives the current status of the
device (ON/OFF). In the ‘setup(),’ serial monitor is used for debugging. Declare D0 ,
D1 pins on the NodeMcu board as output pins. D0 , D1 are referred to 16, 5 which
are Arduino equivalent. Set a timer object which runs the block ‘myTimerEvent()’
for every second and initiate the Blynk connection. Figure 8 gives the setup details.
Virtual pins V 1 , V 2 are used to control D0 , D1 respectively as shown in Fig. 9. The
‘loop()’ block which runs infinitely, starts Blynk as well as a function ‘powerstatus()’
is called, which senses the present state of the device using the sensing circuit, as
shown in Fig. 10.

Fig. 5 Packages

Fig. 6 Credential
declarations
626 M. Rajasekhar Reddy et al.

Fig. 7 Power status

Fig. 8 Setup

Fig. 9 Assigning virtual

pins
Voice Controlled Home Automation Using Blynk, IFTTT with Live … 627

Fig. 10 Loop

4.2 Blynk Configuration

1. Create an account in Blynk using Google or Facebook.

2. Click on NEW PROJECT option as shown in Fig. 11.
3. Given the project title and select the device in the ‘choose device’ option as per
requirement. The screenshot is in Fig. 12.
4. Auth token is generated and will be sent to your Gmail.
5. A dashboard having 2000 energy points will be provided.
6. The required components are dragged and dropped on the workspace and are
configured as shown in Fig. 13.

4.3 IFTTT Setup

Working Algorithm:
1. Login into the IFTTT using Gmail credentials, and grant access to it.
2. Go to MY APPLET and create a new applet.
3. Click on ‘+this’ button and select the web service as Google Assistant.
4. Choose trigger as ‘say a phrase,’ fill trigger fields as shown in Fig. 14
5. After creating a trigger to turn on a device, click on ‘+that’ button.
6. Select the action service as ‘WEBHOOKS.’
7. Now make a web request as shown in Fig. 15.
8. Fill the URL field as ‘HTTP://188.166.206/Auth_token/update/pin.’
9. Choose the method as ‘PUT’ and content type as ‘application/json.’
10. Fill the body as [“1”] and click on create action.
11. Click on the finish button to finish the applet creation.
12. Create various applets for ON/OFF of different appliances separately.
628 M. Rajasekhar Reddy et al.

Fig. 11 Start-up screen

4.4 Hardware

The hardware components, pin configuration, and specifications are explained in

Sect. 2. The connections are made to integrate NodeMcu, relay module, sensing
circuit. The connections are shown in Fig. 16.

4.5 Simulation of Sensing Circuit

See Fig. 17.

WORKING. Now when we say a particular frame of the sentence to Google
Assistant, it will pass on the information to IFTTT and the latter will perform the
action of updating the Blynk cloud with specified data. The updated data is received
by the NodeMcu and performs the actions.
Voice Controlled Home Automation Using Blynk, IFTTT with Live … 629

Fig. 12 New project

5 Results

Using the proposed method, the accessibility is increased and the person operating
this need not be in the specific or region bounded. The proposed method costs very
less when compared to other techniques [13]. The cost of components used in the
proposed method is listed in Table 1.

6 Conclusion

The proposed method is cost-effective for home automation with easy maintenance
and security. Home automation remains as a growing field in twenty-first century.
This paper aims at providing people an easier access for operating the devices at
home from any part of the world at a very low cost.
630 M. Rajasekhar Reddy et al.

Fig. 13 Workspace
Voice Controlled Home Automation Using Blynk, IFTTT with Live … 631

Fig. 14 Trigger creation

632 M. Rajasekhar Reddy et al.

Fig. 15 Making a web

request
Voice Controlled Home Automation Using Blynk, IFTTT with Live … 633

Fig. 16 Circuit

Fig. 17 Sensing circuit simulation

Table 1 Cost of components

S. No. Components Cost
used in the proposed method
1 NodeMcu Rs. 350
2 Relay Rs. 180
3 Analog input extender Rs. 450
4 Blynk and IFTTT apps Rs. 0
Total Rs. 980
634 M. Rajasekhar Reddy et al.

References

1. Singh H, Pallagani V, Khandelwal V, Venkanna U (2018) IoT based smart home automation
system using sensor node. In: 4th international conference on Recent Advances in Information
Technology (RAIT) IEEE
2. Lee Y, Jiang J; Underwood G; Sanders A, Osborne M (2017) Smart power-strip: home
automation by bringing outlets into the IoT. IEEE
3. Taştana M, Gökozanb H (2018) An Internet of Things based air conditioning and lighting
control system for smart home. Am Sci Res J Eng Technol Sci (ASRJETS) 50(1):181–189
4. [Link]
5. Rout K, Mallick S, Mishra S (2018) Design and implementation of an Internet of Things based
prototype for smart home automation system. Researchgate
6. [Link]
7. Durani H, Sheth M, Vaghasia M, Kotech S (2018) Smart automated home application using
IoT with Blynk App. In: Second International Conference on Inventive Communication and
Computational Technologies (ICICCT)
8. Vorapojpisut S (2015) A lightweight framework of home automation systems based on the
IFTTT model. Researchgate
9. Mandula K, Parupalli R, Murty CHAS, Magesh E, Lunagariya R (2015) Mobile based
home automation using Internet of Things (IoT). In: International Conference on Control,
Instrumentation, Communication and Computational Technologies (ICCICCT) IEEE
10. Vivek GV, Sunil MP (2015) Enabling IOT services using WIFI—ZigBee gateway for a
home automation system. In: IEEE International Conference on Research in Computational
Intelligence and Communication Networks (ICRCICN)
11. Mohamed Imran M (2016, April) Intelligent home control and monitoring system via Internet.
Int J Sci Dev Res (IJSDR). 1(4). ISSN: 2455-2631
12. Altabatabaie KF (2017) Remote automation system control using Arduino board. Researchgate
13. Alvarez J, Arturo A, Gutierrez S, Rodrigo PM, Lay-Ekuakille A (2018) A low-cost presence
detection system for smart homes. In: International conference on Research in Intelligent and
Computing in Engineering (RICE), IEEE
AD-LIB: Automated Library System

Dattatray Sawant and Rohan Patil

Abstract Automation is increasing rapidly, and intelligence in applications emerges

as a new form of automation. The impact of automation is observed in the software,
hardware, and machine layer. Due to automation, human intervention is reduced
in a number of areas such as manufacturing, transport, utilities, defense, facilities,
operations and lately, information technology. With this view in mind, we have
developed library system for renewal and submission of books. To develop this
system, we have used a microcontroller, a barcode scanner, and conveyor. First, we
developed graphical user interface with MATLAB and then with LabVIEW for this
system. Developed library system allows students to renew and submit in 24/7 basis.

Keywords Graphical user interface · Virtual instrument · Acquisition · Database ·

Barcode · Library management

1 Introduction

Robotization and innovation have turned out to be an aid for digital operations.
It has made the human to work simpler. The developed system demonstrates the
understudy subtitles, the book of the week; it additionally demonstrates the visual
history of the understudy; it makes a database of the restoration and accommodation
and furthermore contains a criticism area. The procedure begins with client entering
one of a kind ID given to every client. When the client enters a substantial ID, the
drove on the GUI turns green. On the off chance that the client enters an invalid
ID, the drove will turn red. On the off chance that the ID entered is substantial, the
photograph of the client will be shown alongside his/her different details shown,

D. Sawant (B)
Department of Mechatronics Engineering, MPSTME, NMIMS University, Mumbai, India
e-mail: [Link]@[Link]
R. Patil
Automation Department, Jain Irrigation Systems Ltd., Jalgaon, India
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 635

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
636 D. Sawant and R. Patil

which are name, sexual orientation, degree, stream, year, age, and the name of the
last book issued by him/her.
Once signed in, one can look over the numerous activities given and the one he
wishes to do. He can check his issual history, present the book, re-establish the book,
check the book of the week, or look for any book. On the off chance that the client
taps on the issual history catch, he can check the quantity of books which he directly
has issued on his name. The issual history demonstrates the quantities of books he
has issued. It demonstrates all the book subtitles that are the Book ID, Title of the
book, Publication Company, Name of the Author, Genre, and its time of distributing,
ISBN no. also, the book version for each issued book. The following choice is present
in the book.
On the off chance that the client needs to restore the book they obtained, they
should examine the standardized tag given on the book by holding it under the
constant standardized tag scanner. Once examined, they should press the submit
catch. At the point when the catch is squeezed, the transport line framework turns
on, and the belt activates. The book goes to the opposite side and falls into a book
holding holder. On the off chance that the client re-establishes the book, they should
filter the book by holding it under the standardized tag scanner. When the book is
examined snap on the re-establish catch. The date will be reached out by a week or
the quantity of days indicated by the library as per its framework. The book of the
week catch will show the image of the book of the present week. This book changes
each week and should be refreshed by the curator physically.
Usually, librarian has to enter data manually in the system, and then give it back
to the user while returning or re-issuing a book. During examination, there are long
queues while returning of books. To overcome this problem, we developed library
system where submission and re-issual of books are done automatically. The portal is
a model of a system which can be introduced in libraries to automize the procedures
with the goal that less time is taken for the accommodation and re-issual of books.
The developed system primarily consists of three parts, viz. the GUI screen, the
conveyor line, and the barcode identification scanner. It comprises of sensors and
LEDs to show what activities are going on. With the help of sensor, barcode scanner
gets activated. Every user will be issued a unique ID and password to sign in. After
user signs in, detailed information about his library transactions will be displayed.
The user may choose to either submit or renew the particular book. This procedure
will along these lines lessen the time taken for the entire accommodation process.
Likewise, supervision is not required for the task of these portals.

2 Related Work

Using pick and place robotic arm, the library management system is developed
[1]. Anita et al. created robotization in library for transporting of books in faster
manner [2]. Mobile robotics is introduced as a motivating platform in courses with
the help of combination of LEGO NXT mobile robots with LabVIEW [3, 4]. By the
AD-LIB: Automated Library System 637

use of an RGBD sensor, the pose of a known object is detected in the shelf using
visual data [5]. Rahul Pol et al. designed and developed a low cost superior four
degrees of freedom (DoF) robot ARM [6]. Modern database-driven Web applications
are developed by ISIS family member [7]. Author [8] investigated the problems
encountered during automation in two Nigerian university libraries. The significant
improvement in service and use of library facilities are observed with the help of
proposed library information system [9]. The mechanical and technical issues and
problems were investigated while designing line follower robot [10]. In the paper
[11], the library inventory management system (LIMS) is demonstrated using line
following robot (LFR). Based on National Instruments LabVIEW controlled PXI RF
hardware, Pavel and Rao [12] developed a UHF radio frequency identification tag
test and measurement system.

2.1 Motivation and Scope

In the present scenario, the reader has to physically come to the library specifically
to submit or renew the book; the librarian will then manually scan the barcode on
the book. Once the computer receives the code on the screen, then the librarian
will search for the excel sheet belonging to the SAP ID of the student. Since the
information has to be typed by the librarian inside the excel sheet, there is a margin
for the human error, as well as it is time-consuming since the supervisor has to check
for the librarian’s mistakes at the end of the day. Since all students get free from their
lectures about the same time, there is a heavy rush, and the staff is under pressure to
complete the process as soon as possible, which may lead to further errors. Due to
long queue, it is very difficult for physically disabled user.

2.2 Salient Contribution

The idea of automated library is yet to be implemented in India. AD-LIB is an

integration of microcontroller, Barcode Scanner, National Instruments LabVIEW.
• Strong acrylic sheet covering to withstand the shocks and simple encasing of the
framework inside the wall.
• Black conveyor belt for easy scanning of the barcodes placed on the books.
• Secured Logging via a password-based GUI 24/7 without fail.
• Accessing and printing the database through Microsoft Excel Sheet. Arrangement
of enrollment of new user at portal and additionally at the organizational work
area.
• Unique barcode assigned to every book and each student. The barcodes are useful
in calling out the functions from the spreadsheet database.
638 D. Sawant and R. Patil

• National Instruments LabVIEW GUI for cost sparing and easy to understand user-
friendly interactive system, LabVIEW is utilized for interfacing barcode scanner
with the GUI.
• National Instruments VISA tool is used to establish the connection between
Arduino and the GUI.

3 Methodology

Automation and technology have proven to be a boon for us. It has made human work
easier. The portal which we have made aims at reducing the workload on the library
staff by automating the renewal and submission process. Figure 1 shows the flow
chart of the whole process of renewal and submission of books. It also has additional
features, viz. It shows the student details, the book of the week, it also shows the
issual history of the student, and it creates a database of the renewal and submission
and also contains a feedback section. The process starts with the user entering the
unique ID given to each user. Once the user enters a valid ID, the led on the GUI
turns green. If the user enters an invalid ID, the led will turn red. If the ID entered is
valid, the photo of the user will be displayed along with his/her other details which
are name, gender, degree, stream, year, age, and the name of the last book issued
by him/her. Once logged in, one can choose from the many actions given, the one
he wishes to do. He can check his issual history, submit the book, renew the book,
check the book of the week, or search for any book. If the user clicks on the issual
history button, he can check the number of books which he presently has issued on
his name. The issual history shows the numbers of books he has issued. It shows all
the book details that are the Book ID, Title of the book, Publication Company, Name
of the Author, Genre, its year of publishing, ISBN no., and the book edition for each
issued book. The next option is submit the book.
If the user wants to return the book they borrowed, they will have to scan the
barcode given on the book by keeping it under the continuous barcode scanner. Once
scanned, they will have to press the submit button. When the button is pressed, the
conveyor belt system turns on and the belt actuates. The book goes to the other side
and falls into a book holding container. For renewal, user will have to scan the book
by barcode scanner. After the successful scan, click on the renew button. The date
will be extended by a week or the number of days specified by the library according
to its system. The book of the week button will display the picture of the book of
the current week. This book changes every week and needs to be updated by the
librarian manually.
AD-LIB: Automated Library System 639

Fig. 1 Flowchart of AD-LIB

4 Softwares

The main purpose of AD-LIB is to carry out the library operations as desired by
the user, so we require an intermediate platform that would take the inputs from
the end-user, which in our case would be students, and these inputs would give
specific commands to the controller. A Graphic User Interface (GUI) functions as
the median platform, which has been designed and programmed to perform the
640 D. Sawant and R. Patil

functional requirements. The inputs given by the user would be captured via an
event structure, initialized during mouse click.
A GUI has been the Centralized Unit of the System; it comes in handy while
interfacing microcontroller, barcode scanner, database (spreadsheets), Arduino UNO
so that the user can trace back the problem or the run-time errors when the system
breaks down. The GUI build up was initially started on MATLAB, but later, it was
shifted to LabVIEW as the LabVIEW programming has a specific architecture in
the form of state machines. We found the MATLAB programming a little sturdy and
rectifying error consumed lot of time.
LabVIEW comes as an alternative for Supervisory Control and Data Acquisition
(SCADA), so getting hands on experience on LabVIEW for future aspects was a
good start for any Mechatronics Engineer. Whereas, having a little knowledge of
MATLAB programming made us aware of the application development protocols.
Overall, all the interfacing with the different components made us aware of the real-
time working of the subsystems from a single centralized unit that supervises the
decision and control of the modules attached to it.

4.1 MATLAB Guide

As we can see in Fig. 2, the initial GUI contained combinations of Push Buttons,
a Header Text Box, and an ACTIVE X control developed by the Adobe Reader.
You can likewise code the format and conduct of your application total utilizing
MATLAB capacities.

Fig. 2 Initial MATLAB GUIDE screen

AD-LIB: Automated Library System 641

Fig. 3 MATLAB GUI algorithm

In this approach, you make a customary figure and place intuitive parts in that
figure automatically. These applications bolster similar kinds of designs and intelli-
gent parts that GUIDE underpins, and also selected boards. Utilize this way to deal
with manufacture complex applications with numerous reliant parts that can show
any kind of plot.
Every time there is a mouse click event on any one of the push buttons placed on
the GUI screen, it generates a variable value assigned to the function named as event
data. Therefore, upon a clicking, an event is called out of particular function name
to which the push button has been added. An insight of the working algorithm has
been shown in Fig. 3. If a user clicks on push button 4, a function named push button
4 call-back function is called, and its event data is then transfered to the edit text (set
(handles.edit1, String, SAP4)) window above to show which push button has been
pressed. Radio Buttons, Navigation Panel, and User Feedback had been added to the
final GUI on MATLAB GUIDE as shown in Fig. 4.

4.2 National Instruments LabVIEW

LabVIEW structures are called virtual instruments in light of the fact that their
appearance and activity mimic physical instruments, for example, oscilloscopes and
multimeters. LabVIEW contains an extensive arrangement of devices for securing,
breaking down, showing, and putting away information and also devices to enable
you to investigate code you compose.
In LabVIEW, you fabricate a VI, or front panel, with controls and markers. Con-
trols are handles, push buttons, dials, and other information systems. Pointers are
642 D. Sawant and R. Patil

Fig. 4 Final MATLAB GUI

charts, LEDs, and other yield shows. After you construct the front pane as shown
in Fig. 5, you include code utilizing VIs and structures to control the front panel
objects. The square chart contains this code.
One advantage of LabVIEW over other advancement situations is the broad
help for getting to instrumentation equipment. Drivers for a wide range of sorts
of instruments such as ARDUINO, Barcode Scanners, DAQs, Motors, and LEDs are
accessible for consideration. These present themselves as graphical hubs.

Fig. 5 Front panel of AD-IB screen in LabVIEW

AD-LIB: Automated Library System 643

Fig. 6 Block diagram of the GUI in LabVIEW

Once we had created the front panel, we moved on to developing logic of the block
diagram as shown in Fig. 6 which consists of the control of the front panel objects.
The square outline contains this graphical source code. The terminals provide the
information to the control after execution of the VI. The block diagram shows up as
symbol or feed information to be written in the database (spreadsheet).
Terminals are passage and communication ports that trade data between the front
board and block diagram. Upon initializing the inputs or providing events in the form
of mouse clicks, the GUI performs tasks when a VI runs. The use of event structures
monitors the execution of functions, .jpeg image reading, writing data to the excel
sheets, and a communication channel between Arduino UNO and Front Panel Push
Button (Submit a Book) in content-based programming architecture as shown in
Fig. 7.
Combination of push buttons and clusters are graphical portrayals in the silver
squares, and case structures give selective content-based programming decision to
the Front Panel. The output of the clusters purely depends upon the unique passwords
of the students and the information saved in the case structures.
Whenever a transaction occurs through the GUI, its information is saved in the
excel sheets with the help of the write to measurement file function palettes; the
librarian can send a command to print the monthly data as per the administration
requirements. This eliminates the data discrepancies and human error during data
logging, and thus AD-LIB becomes more reliable and accurate in such scenario.
644 D. Sawant and R. Patil

Fig. 7 Final AD-LIB GUI screen in LabVIEW environment

5 Testing

Initially, testing was done in two parts. The conveyor and the GUI were both tested
separately. The conveyor was first controlled using PIC microcontroller. There were
two sensors attached at the start and the end on the conveyor to sense when the book
is kept. When the book was placed, it would get sensed by the sensor, and a red led
on the sensor circuit would glow to show that the book has been placed, the conveyor
would then actuate in clockwise direction. When the book would reach the sensor
2 it would sense, a red LED on the circuit would glow indicating that the book has
passed; this would give the microcontroller instruction that the book has passed and
the conveyor has to be stopped.
There were times when the conveyor did not stop even after the book passed from
the sensor and fell in the basket. This was because there was a problem in the range
of the sensor. The range was adjusted by the POT given on the IR sensor circuit.
The conveyor could also be moved counter clockwise by reversing the connections
to the microcontroller. There were two switches which were used to start or stop the
conveyor manually when the sensor did not pick up the signal accurately. When we
bought the barcode scanner (INTEX IN-101), the logic was to have a sensor which
gives command to the scanner to actuate the scanning, but at the later part of the stage
we realized that the sensor is getting lot of the disturbances from the surrounding and
continuous change in the direction of rotation of the motor to pick up the barcode
would be difficult task, so the discontinuous scanner was converted to the continuous
mode through the barcode command, given in the manual of the INTEX IN-101.
AD-LIB: Automated Library System 645

6 Conclusion

There is very little automation in India’s library system and a lot of scope to increase
the library system efficiency and improve the data-keeping methods. This paper
surely helps them in understanding the microcontrollers, actuation system, HMIs,
Bio-Metric systems, and data logging to improve their knowledge. The developed
system provides effective management of library as well as reduces human interven-
tion. The GUI build up was initially started on MATLAB, but later it was shifted
to LabVIEW as the LabVIEW programming has a specific architecture in the form
of state machines. We found that LabVIEW software shortens the design cycle and
simplifies the design process as compared to MATLAB. Finally, since the complete
project will be built in India and will be used for Indian people, the project meets the
Make in India Initiative.

Acknowledgements In Fig. 7, we have used the author photo. The authors thankful to the NMIMS
Deemed to be University’s Mukesh Patel School of Technology Management and Engineering,
Mumbai, India for providing the necessary facilities for carrying out this work.

References

1. Gade A, Angal Y (2017) Automation in library management using LabVIEW. In: IEEE
international conference on computing communication control and automation, pp 1–5
2. Gade A, Angal Y (2017) Development of library management robotic system. In: IEEE
international conference on data management, analytics and innovation, pp 254–258
3. Gomez-de Gabriel JM, Mandow A, Fernandez-Lozano J, Garcia-Cerezo AJ (2011) Using
LEGO NXT mobile robots with LabVIEW for undergraduate courses on mechatronics. In:
IEEE Trans Educ 54(1):41–47
4. Gomez de Gabriel JM, Mandow A, Fernandez-Lozano J, García-Cerezo A (2015) Mobile robot
lab project to introduce engineering students to fault diagnosis in mechatronics systems. IEEE
Trans Educ 28(3):187–193
5. Rennie C, Shome R, Bekris KE, De Souza AF (2016) A dataset for improved RGBD-based
object detection and pose estimation for warehouse pick and place. IEEE Robot Autom Lett
1(2):1179–1185
6. Pol RS, Giri S, Ravishankar A, Ghode V (2016) LabVIEW based four DoF robotic arm. In:
IEEE international conference on advances in computing, communications and informatics,
pp 1791–1798
7. De Smet E (2010) Some ISIS-software history and technical background on the new FOSS
integrated library system ABCD. Liber Q 324–335
8. Adegbore AM (2010) Automation in two Nigerian university libraries. Libr Philos Pract 1–13
9. Jibia MS, Mubaraka CM, Michael O (2013) Integrating ICT in library management: design and
development of an automated library management system for Cavendish University Uganda.
In: Innovative systems design and engineering, pp 1–11
10. Pakdaman M, Sanaatiyan MM, Ghahroudi MR (2010) A line follower robot from design
to implementation: technical issues and problems. In: 2nd IEEE international conference on
computer and automation engineering, pp 5–9
646 D. Sawant and R. Patil

11. Thirumurugan J, Kartheeswaran G, Vinoth M, Vishwanathan M (2010) Line following robot for
library inventory management system. In: IEEE international conference on emerging trends
in robotics and communication technologies, pp 1–3
12. Nikitin PV, Rao KV (2009) LabVIEW-based UHF RFID tag test and measurement system. In:
IEEE Trans Ind Electron 56(7):2374–2381
Design and Integrate IoT Sensors to RO
Water Purifiers for Remote Monitoring
and Allowing Users to Pay Per Usage
on the Rented RO System

H. D. Sundresh and D. Priya

Abstract Internet of Things (IoT) has given promising chances to make better condi-
tion for humankind. One of its applications is smart metering. In today over-polluted
world drinking water is the most precious and one of the costliest resources. This
paper discusses the approach to provide access to clean drinking water for everyone
cost effectively, by renting the smart water purifier and allowing the user to pay as
per their usage, without worrying about the quality of water and maintenance of the
purifier. This approach is win-win situation for the user as well as for the company.
This system will be very economical.

Keywords IoT · Water purifier · Billing

1 Introduction

Internet of things (IoT) had made a revolution on how we see the real-world objects.
It is essentially empowering the littler frameworks to get associated with the web,
giving them the capacity to think and impart without expecting human to human or
human to machine connection. IoT is a rising worldwide innovation, in which things
can be associated with the web and controlled remotely from anyplace around the
world [1] (Fig. 1).
Lately, a wide scope of IoT applications has been created and sent in businesses
just as in local territories, for example, robotized observing, control, management,
and maintenance [2]. Sensors and actuators are ending up increasingly groundbreak-
ing, less expensive, and littler, which make their utilization unavoidable. The data
gathered by sensors are changed into savvy data, therefore engrafting knowledge
into our encompassing condition. IoT empowers billions of gadgets to report their
area, personality, and history over remote associations. With the utilization of new

H. D. Sundresh (B)
RV College of Engineering®, Banglore, India
e-mail: sundreshhd@[Link]
D. Priya
Department of Information Science, RV College of Engineering®, Bangalore, India
© Springer Nature Singapore Pte Ltd. 2020 647
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
648 H. D. Sundresh and D. Priya

Fig. 1 Overview of IoT

advancements, personal satisfaction is improved just as it lessens the effect of human

exercises and utilization designs on condition.
From the urbanization, there has been an exponential increase of demand for
clean drinking water [3]. As the population is growing in cities, cost for management
of water transmission, storage, and treatment and distribution for consumption are
serious issues in underdeveloped and developing countries. With the rapid changes
in lifestyle, there is an impact on usage of water and related overheads on sewerage
requirements.
In upcoming years, water, the need of life, is conceivably to present the most
noteworthy test of its expanded utilization [4, 5] with populace rise, mechanical
advance ment, and diminishing supplies because of contamination and over abuse.
India is evaluated to end up water focused [6] on the country by 2020.
Brilliant water framework will make clients mindful of their utilization conduct.
A wide scope of water sensors is accessible for various modern and residential
applications, which measure distinctive substances like water quality, temperature,
and spillage location [7, 8].
This paper proposes the implementation of different sensors for water quality
control and maintenance of the water purifiers, thus providing access to clean drinking
water to the users.

2 Objectives

• Simplifying the safe drinking water challenge.

• Ease of monitoring shelf life of different filters in the RO system for the company
that rents it, by integrating IoT-enabled TDS sensor at water inlet and outlet.
• Creating tension-free environment for the user without need for worrying about
the quality of water.
• Enabling location tracking with the GPS sensor, thus company can track the device
if the user steals it.
• To provide day by day measure of water usage.
• Service provider of water purifier can charge the user based on water usage instead
of the amount charged for purifier.
Design and Integrate IoT Sensors to RO Water Purifiers … 649

3 Literature Survey

In the present days, there are a lot of water purifiers available in the market with a
price tag of around 10,000 INR. Even after buying the purifier the user needs to spend
for maintenance, i.e., changing filters time to ensure safe quality drinking water.
If we do a cost-benefit analysis between the proposed system and the regular
purifiers, the purifier of an established brand costs around 15,000–35,000 INR and
of non-established brands costs around 5000–15,000 INR but in the proposed system
the purifier is installed at free of cost in the user space. Also the annual maintenance
cost comes around 6000 INR for an established brand’s purifier and around 3000
INR for non-established brand’s; in the proposed system, it is done free of cost to the
user. The water quality and the filters’ life are monitored in real-time in the proposed
system. So if we consider cost per liter of water, with the established brands it comes
around 2.5 INR per liter and non-established brands cost around 1 INR per liter.
But with the proposed system, the company can offer a flat rate per month up to
certain liters and then it can charge according to the usage of the customer (say 250
INR for the first 250 L then 1 INR per liter). A normal household of four people
consumes around 500 L in a month, if they purchase bottled water then that costs
around 750 INR (considering 20 L bottle at cost of 30 INR per bottle), or if they
install any established brands purifier then that will cost around 1000 INR but in the
proposed system it costs around 500 INR which is less than any of the mentioned.
Also company can recover its investment within 3 years (Table 1).
Thus, it is a win-win situation for both company and consumer.

4 Methodology

The proposed system will be implemented in different modules.

• One module is where we integrate sensors like TDS sensor, flow rate sensor, and
water cut-off valve on the purifier, which are connected to a central server using a
GSM module. Periodically, the data from the purifier device is transferred to the

Table 1 Cost-benefit analysis

Comparison Proposed system Purifier of Purifier of
established brand non-established
brand
Purchasing cost Zero 15,000–30,000 INR 5,000–15,000 INR
Annual maintenance Zero Around 6000 INR Around 3000 INR
cost
Water quality and Real-time updates Need to be done by Need to be done by
filter life from provider user user
Cost per liter Around 1 INR Around 2.5 INR Around 1 INR
650 H. D. Sundresh and D. Priya

server. Also, the company can send signals to the purifier to stop the water flow if
the bill amount is pending [9–11].
• Another module is where the company can access the data collected on the central
server, analyze the data, and update the customer about their usage and billing.
The company will keep an eye on the quality of input water and on the output
water quality to analyze whether the filters are properly working or not; in case of
any abnormality is seen, then it will alert the customer not to use the purifier and
will send the service team to check the device and repair it. The customer can also
login to the portal and check their usage behavior [12, 13].

5 Conclusion

A novel system for implementing an economic and reliable smart water sensor on
the water purifier using IoT-based hardware is discussed. The features of system and
the benefits are discussed. The proposed system overcomes the burden of extra costs
such as annual maintenance and service charges paid by the user for regular service.

Acknowledgements The authors also acknowledge the help received from the institution—RV
College of Engineering, Bangalore.

References

1. Lokhande DG, Mohsin SA (2017) Internet of things for ubiquitous smart home system. In: 2017
1st International Conference on Intelligent Systems and Information Management (ICISIM),
5–6 Oct 2017
2. Atzori L, Iera A, Morabito G (2010) The Internet of Things: a survey. Comput Netw
54(15):2787–2805
3. Mudumbe MJ, Abu-Mahfouz AM (2015) Smart water meter system for user-centric consump-
tion measurement. In: 2015 IEEE 13th International Conference on Industrial Informatics
(INDIN), 22–24 July 2015. [Link]
4. Smart water metering networks an Intelligent Investment? [Online]. Available [Link]
[Link]/articles/wwi/print/volume-26/issue-5/regulars/creative-finance/smart-water-
[Link]
5. Tzortzakis G, Katsiri E, Karavokiros G, Makropoulos C, Delis A (2016) Tethys: sensor based
aquatic quality monitoring in water ways. In: 2016 at 17th IEEE international conference on
mobile data management
6. Gupta N, Shukla D (2016) Design of embedded based automated meter reading system for
real time processing. In: IEEE Students Conference on Electrical, Electronics and Computer
Science (SCEECS), 5–6 Mar 2016. [Link]
7. Hsia1 S-C, Hsu S-W, Chang Y-J (2012) Remote monitoring and smart sensing for water quality
and leakage detection. IET Wirel Sens Syst 2(4):402–408
8. Suresh M, Muthukumar U, Chandapillai J (2017) A novel smart water-meter based on IoT
and smartphone app for city distribution management. In: 2017 IEEE region 10 symposium
(TENSYMP), 14–16 July 2017
Design and Integrate IoT Sensors to RO Water Purifiers … 651

9. Shingote KS, Shahane P (2016) Microcontroller based flow control system for canal gates in
irrigation canal automation. In: 2016 IEEE 6th international conference on advanced computing
10. Mashhadi SKM, Yadollahi H, Mash AM (2016) Design and manufacture of TDS measurement
and control system for water purification in reverse osmosis by PID fuzzy logic controller
with the ability to Compensate effects of temperature on measurement. Faculty of Electrical
Engineering-Iran University of Science and Technology, Tehran, Iran
11. Cloete NA, Malekian R, Nair L (2016) Design of smart sensors for real-time water quality
monitoring, published in 19 July 2016
12. Umbaugh HJ (1967, June) Billing frequency and customer relations. J Am Water Works Assoc
59(6):669–674
13. Britton TC, Stewart RA, O’Halloran KR (2013) Smart metering: enabler for rapid and effective
post meter leakage identification and water loss management. J Clean Prod 54(1):166–176
MPTCP a Solution Over Crunch
in Bandwidth Requirement
for Multimedia Application

Vidya Kubde and Sudhir Sawarkar

Abstract Multimedia real-time applications like surveillance systems and video

conferencing, where latency plays an important role along with the bandwidth.
Today’s network advancements are able to resolve bandwidth constraints, but in
terms of real-time applications issues related to latency impacts the performance.
Multipath TCP (MPTCP) is observed to solve this issues and also able to give the
better efficiency in network utilization. In this paper, we have attempted to compare
and analyse the growth of MPTCP in multimedia applications and identify the related
issues, where by solving these, performance of MPTCP will improve along with the
increase in overall efficiency in multihomed or IoT devices and also on real-time
multimedia applications.

Keywords Bandwidth aggregation · MPTCP · Multimedia applications · Video

streaming

1 Introduction

The use of multimedia has been increasing exponentially over the last decade. Cur-
rently the 80% of the bandwidth has been used by the multimedia application on
Internet [1]. It is predicted in next decade the requirement of the bandwidth will
be multifold as more application areas like education, surveillance, vehicular net-
work and industrial Internet are yet to adopt in full fledge. It has been observed by
researchers [2] that bandwidth requirement is growing at the rate of 42% per year.
The need of future applications like multimedia applications is saturating the Internet
and will drive data rates to levels far beyond the capability of current Internet [3].

V. Kubde (B)
Department of Information Technology, Datta Meghe College of Engineering, Navi Mumbai, India
e-mail: vidyakubde@[Link]
S. Sawarkar
Department of Computer Engineering, Datta Meghe College of Engineering, Navi Mumbai, India
e-mail: sudhir_sawarkar@[Link]

© Springer Nature Singapore Pte Ltd. 2020 653

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
654 V. Kubde and S. Sawarkar

Many researchers have already attempted to meet this huge bandwidth requirement
by using aggregation of multiple heterogeneous networks.
MPTCP allows to make them work together and get the benefits of the two at the
same time. Multipath TCP allows applications to use all the available interfaces at
the same time to improve quality of service (QoS) and quality of experience (QoE).
Multipath TCP does not require any additional infrastructure for its implementation.
Understanding the need of future demands and use of existing infrastructure
efficiently, we here attempted to study and analyse in depth the issues faced by
researchers while implementing the MPTCP which will help to proceed the research
on various issues related to MPTCP. This paper describes the basics of MPTCP in
Sects. 2 and 3 and further analyse the impact of MPTCP on performance, energy
consumptions, and in the case of video streaming applications in Sects. 4 and 5.

2 MPTCP and Related Work

MPTCP is a multipath solution implemented at the transport layer. It is IETF standard;

experimental at this stage, MPTCP [4] is designed with goals like, MPTCP should
work along with TCP; in case of failure of MPTCP, it should fall back to the TCP,
or in case of poor performance then also it should fall back to TCP. Multipath TCP
should not harm other flows i.e. it should not take more capacity on any of its path
than if it was a single path using only that route.

2.1 MPTCP Working

Application layer access connects to TCP/MPTCP layer by single socket and it

appears to the application layer with normal TCP layer. MPTCP establishes con-
nection initially with one of its interface as per the standards of TCP and further it
dynamically adds another interface to the existing connection. “Figure 1” describes
the details of the establishment of the connection process in MPTCP.
Connection in MPTCP is established with three way handshake signals in “Fig. 1”.
At start, M1 confirms that M2 supports MPTCP, when M2 confirms than M1 acknowl-
edges and primary flow is established. The secondary flow is joined to this connection
with again three way handshake signals.

3 MPTCP Progress

MPTCP as is a extension of TCP launched in 2011 with the aim of utilizing all the
available interfaces concurrently at the same time to improve bandwidth and pro-
vide redundant connection with added performance and load balancing strategies as
MPTCP a Solution Over Crunch in Bandwidth Requirement … 655

Fig. 1 Connection establishment in MPTCP

shown in Fig. 2. The use cases of MPTCP were smartphones and datacenters. 2012
added the 3G cellular networks to the communication which was not sufficient for
multimedia applications, and this deficiency made researchers to use MPTCP for
improving quality of service in multimedia applications. Apple, a leading innovative
and manufacturer of mobile and desktop, deployed MPTCP in iOS for its siri digital
assistant application in 2013. Apple has been using MPTCP since its iOS 7 release
understanding the benefit of MPTCP with handover capabilities, where MPTCP has
ability to seamlessly handover from one interface to another. In 2014, research was
focused on improvement in QoE at the end user side for the multimedia applications
like video on demand and high definition (HD) videos over the cellular network and
WiFi network through MPTCP. The emphasis was to minimize the delay variation,
packet loss, thus enhancing QoE in video streaming applications in heterogeneous
networks. In 2015, researchers targeted unsolved issue of bandwidth aggregation for
concurrent video transmission in heterogeneous access networks where the problem
of mobile video delivery using MPTCP in heterogeneous wireless networks with
multihomed terminals was put forth. This issue is addressed by considering path
asymmetry in different access networks to achieve the optimal quality of real-time
video streaming. From 2016 onward, smartphone vendors have started to deploy
Multipath TCP, but its performance with real smart phone applications has to be
focused. Multipath TCP is a new TCP extension that has strong potential on smart-
phones, as shown by its adoption by Apple and Korea Telecom. Apple’s deployment
656 V. Kubde and S. Sawarkar

Fig. 2 Evolution history of MPTCP

focused on a single use case and very little work was done in interactions between
real smartphone applications and Multipath TCP.
2017 focused mainly on the energy efficient applications on multihomed devices
operated on battery. Several MPTCP schemes for reducing energy consumption and
improving user-perceived video quality were developed in this era.

4 Issues and Analysis

In the presents year, bandwidth was not the main concern as cost of the bandwidth
decreasing day by day, but many multimedia and video conferencing applications
which are latency sensitive applicants, needs less delay path where MPTCP is the only
solution. Especially, researchers are focusing on use of MPTCP in all the multihomed
devices like smartphones, laptops, tablets or IoT devices based on raspberry pi to
improvement in QoS parameters like jitter, delay, energy and latency which will cater
the need of the users for multimedia based applications. The architecture components
like scheduler, buffer management, congestion control and many more has to be
revised with application perspective. “Table 1”, is the outcome of our analysis which
we have carried doing in-depth survey of the development and research till today.
The analysis is carried out keeping the focus on the solving issues and improving
performance considering the QoS and QoE.
Considering the MPTCP design goals and solving issues appeared in past we have
categorized and identified blocking points of MPTCP as follows.
MPTCP a Solution Over Crunch in Bandwidth Requirement … 657

Table 1 Classifications of issues

Issues Methodologies Evaluation metrics
Subflow path scheduling Comparison of different Delay
designs for multipath
schedulers ranging from round
robin to optimize multipath
scheduler [5]
Implemented three different Throughput
schedulers considering
sending rate, window space
and time/space and compared
them with default MPTCP
scheduler [6]
In-depth performance analysis (1) Goodput (2) Delay
considering heavy data transfer
and application constraints [7]
Estimating packet receiving
time at the receiver and
scheduling packet [8]
A new scheduling policy Throughput
considering congestion control
and speed as decision metrics
[9]
A scheduling scheme that Throughput
assigns data packets to
subflows, by estimating
out-of-order packets caused by
difference in the subflows [10]
Segregating the data on Throughput
different links based upon
quality requirements or type of
data [11]
Path heterogeneity Forward error correction (1) Throughput
(FEC) coding at transport layer
[12]
Forward error correction (1) PSNR
(FEC) coding [13] (2) Delay (3) Goodput
Context aware and partially PSNR
reliable, MPTCP extension
[14]
A loss-aware disabling Goodput
mechanism that temporarily
switch a lossy wireless
interface to a backup mode
[15]
(continued)
658 V. Kubde and S. Sawarkar

Table 1 (continued)
Issues Methodologies Evaluation metrics
Algorithm, to handle the Throughput
different path characteristics
[16]
Congestion control The different methodologies of Throughput
congestion control are studied
and analysed [17]
Coupled congestion control –
[18]
An congestion control –
algorithm is designed
considering energy, round trip
time and path loss rate [19]
The measurement results for –
congestion control
performance in three
multihomed, real world
Internet scenarios that are
evaluated [20]
Bidimensional multipath –
congestion control approach to
deal with TCP fairly in shared
bottleneck [21]
(1) Traffic shaping (2) Traffic (1) Throughput (2) Energy
offloading [19] consumption
Delay based traffic shifting Throughput
[22]
Fast coupled retransmission Queuing delay, TCP timeouts
[23]
A congestion control –
algorithm for MPTCP, using
packet queuing delay as
congestion signals [24]
Gentle slow start scheme Round trip time
(GSAM) for which finds the
appropriate threshold to
smooth the aggressive slow
start behaviour [25]
Algorithm to reduce flappiness Throughput
across paths [24]
(continued)
MPTCP a Solution Over Crunch in Bandwidth Requirement … 659

Table 1 (continued)
Issues Methodologies Evaluation metrics
Adaptive congestion window Goodput
i.e. minimizing the difference
between the subflows and
proactive scheduling algorithm
to determine the packet
sending sequence to each path
[26]
Energy consumption Modified MPTCP to minimize Goodput, delay
the impact on application
delay [27]
Designed and implemented a Throughput, energy
variant of MPTCP, reducing consumption
power consumption without
affecting download latency
[28]
Compared the energy cost of Energy consumption
streaming on a device with two
interfaces and device with one
interface [29]
(1) Flow rate allocation (1) PSNR (2) Inter packet
algorithm (2) Dropping of less delay (3) Jitter
priority frames [30]
Flow rate allocation algorithm (1) Energy (2) PSNR (3)
[31] Goodput
Path switching mechanism MPTCP with path awareness Throughput
can dynamically shift the
traffic to a single path and vice
versa [12]
Short flow/long flow [32] (1) Throughput
(2) RTT
Multi-attribute aware path –
selection approach for MPTCP
[33]
Out-of-order packets Analysis of four solutions for (1) Link utilization
out-of-order packets (1) (2) Throughput
D-SACK, (duplicate selective
acknowledgement)
(2) Eifel algorithm (3)
TCP-DOOR (detection of
out-of-order and response) (4)
Forward Retransmission Time
Out Recovery (F-RTO)
algorithm [34]
Scheduling mechanism [35] (1) Throughput
(2) Packet loss rate
(continued)
660 V. Kubde and S. Sawarkar

Table 1 (continued)
Issues Methodologies Evaluation metrics
Network coding [36] Throughput
MPTCP in video streaming Challenges of using MPTCP in –
video streaming [37]
Explored whether MPTCP –
always benefits video
streaming [38]
Survey on existing literatures –
on QoE of video streaming
[39]
User perception Of video (1) Peak Signal To Noise ratio
quality over LTE network and (PSNR)
different protocols [40] (2) Packet loss
(3) (OWD) One Way Delay
(4) Inter Packet Delay (IPD)
Qos improvement in (1) Peak Signal to Noise ratio
multimedia applications (PSNR)
through partial reliability [41] (2) Loss rate
(3) End to end delay

It describes the ability to use multiple paths with different characteristics to reach
the destination. The difference in the characteristics of different paths like round
trip time, packet loss ratio etc. is going to cause variation in the throughput. The
difference in delay of the paths will lead to out-of-order packets at the receiver.
One of the solutions for this problem is FEC coding [13]. QoE is effected by path
heterogeneity; in [14], the author has given a context aware solution for this issue.
Loss due to heterogeneous networks can be reduced by switching between full mode
and backup mode MPTCP [15]. To use MPTCP in latency sensitive applications,
[11] has suggested FEC coding as a solution for heterogeneous paths. The negative
aggregation caused by the lossy path can be bypassed, by switching to single path
MPTCP [12].
Congestion control in MPTCP switches the traffic from more congested path
to less congested path. The TCP fairness design goal states the controlling of sub-
window increase on each path as important issue. At start, congestion control of single
path TCP was used on each path of MPTCP. This has not satisfied the TCP fairness
goal, so later on the control of the sub-window was done collectively at centralized
manner. All the subflows congestion windows were coupled together [18]. More
deliberate design of multipath congestion control algorithms has used a queuing
delay [24] as a congestion signals. The challenging issue of MPTCP congestion
control is TCP friendliness. One of the approaches for this problem is to apply the
weight [21] to individual congestion control of each subflow, so that a bundle of sub
flows can have the same aggressiveness as one TCP flow. In [19] congestion control,
algorithm with new approach is proposed, which considers energy consumption, RTT
MPTCP a Solution Over Crunch in Bandwidth Requirement … 661

and packet loss rate. Congestion control along with energy efficiency is suggested
by [19, 22] where they have focused on the energy consumption at the receiver side.
Throughput of MPTCP is primarily dependent upon the scheduler. In [42] the
scheduling of the packet depends upon congestion on that path, RTT of that path
and its energy consumption. Estimation-based scheduling is proposed by [8]. The
various scheduling strategies are analysed by [5]. Constraint based proactive solution
is proposed by [10], where additional path is only used when round trip time is
acceptable. The wrong scheduling decision leads out-of-order packets [35] at the
receiver. A novel scheduler [11] segregates the data, into control information and the
data. The scheduling of this both is done on different paths, depending upon their
characteristics.
The multi-interfaces remaining active all the time may result in more energy con-
sumption. However, multipath transfer consumes more energy to maintain all the
active interfaces. Analysis of MPTCP energy consumption [27] along with handover
mechanism in various operational modes is proposed. In [29], authors have deter-
mined the energy cost of using MPTCP. An energy aware variant of MPTCP called
as eMPTCP [28] is designed and implemented. To minimize the energy consump-
tion in quality constrained video [30], dropping of less priority frames and flow
rate allocation algorithm is used. Congestion control along with energy consumption
perspective is analysed [22].

5 Conclusion

MPTCP has been proposed to implement from 2011, to overcome the bandwidth
aggregation. Since then the researchers have modified and improved the MPTCP
in various stages, but still looking to the future need of bandwidth and multimedia
application TCP is not fully replaced by MPTCP. New areas are emerging into the
like IoT and tactile Internet where multiple interfaces are available on such devices
capable of handling MPTCP, but because of issues like congestion control, path
heterogeneity, energy consumption, scheduler stated in the analysis, the efficient
utilization of bandwidth and low latency applications are yet to be used with MPTCP.

References

1. Index, Cisco Visual Networking (2015) Cisco visual networking index: forecast and method-
ology 2015–2020. White paper, CISCO
2. [Link]
index-vni/[Link]
3. [Link]
[Link]
662 V. Kubde and S. Sawarkar

4. Raiciu C et al (2012) How hard can it be? Designing and implementing a deployable multipath
{TCP}. In: 9th {USENIX} Symposium on Networked Systems Design and Implementation
({NSDI} 12)
5. Singh A et al (2012) Performance comparison of scheduling algorithms for multipath transfer.
In: 2012 IEEE Global Communications Conference (GLOBECOM). IEEE
6. Kimura BYL, Lima DCSF, Loureiro AAF (2017) Alternative scheduling decisions for multipath
TCP. IEEE Commun Lett 21.11:2412–2415
7. Paasch C et al (2014) Experimental evaluation of multipath TCP schedulers. In: Proceedings
of the 2014 ACM SIGCOMM workshop on capacity sharing workshop. ACM
8. Kim HA, Oh B-H, Lee J (2012) Improvement of MPTCP performance in heterogeneous
network using packet scheduling mechanism. In: 2012 18th Asia-Pacific Conference on
Communications (APCC). IEEE
9. Yang F, Amer P, Ekiz N (2013) A scheduler for multipath TCP. In: 2013 22nd International
Conference on Computer Communication and Networks (ICCCN). IEEE
10. Oh B-H, Lee J (2015) Constraint-based proactive scheduling for MPTCP in wireless networks.
Comput Netw 91:548–563
11. Mondal A et al (2018) PPoS: a novel sub-flow scheduler and socket apis for Multipath TCP
(MPTCP). In: 2018 Twenty Fourth National Conference on Communications (NCC). IEEE
12. Ferlin S et al (2018) MPTCP meets FEC: supporting latency-sensitive applications over
heterogeneous networks. IEEE/ACM Trans Netw (TON) 26.5:2005–2018
13. Wu J et al (2016) Streaming high-quality mobile video with multipath TCP in heterogeneous
wireless networks. IEEE Trans Mobile Comput 15.9:2345–2361
14. Cao Y et al (2016) PR-MPTCP+: context-aware QoE-oriented multipath TCP partial reliability
extension for real-time multimedia applications. In: 2016 Visual Communications and Image
Processing (VCIP). IEEE
15. Nguyen K et al (2016) An enhancement of multipath TCP performance in lossy wireless
networks. In: 2016 IEEE 41st conference on Local Computer Networks Workshops (LCN
Workshops). IEEE
16. Nguyen SC et al (2011) Evaluation of throughput optimization and load sharing of multipath
TCP in heterogeneous networks. In: 2011 Eighth international conference on wireless and
optical communications networks. IEEE
17. Xu C, Zhao J, Muntean G-M (2016) Congestion control design for multipath transport
protocols: a survey. IEEE Commun Surv Tutorials 18(4):2948–2969
18. Raiciu C, Handley M, Wischik D (2011) Coupled congestion control for multipath transport
protocols. No. RFC 6356
19. Wang W, Wang X, Wang D (2018) Energy efficient congestion control for multipath TCP in
heterogeneous networks. IEEE Access 6:2889–2898
20. Fu F et al (2015) Performance comparison of congestion control strategies for multi-path TCP
in the NORNET testbed. In: 2015 IEEE/CIC International Conference on Communications in
China (ICCC). IEEE
21. Honda M et al (2009) Multipath congestion control for shared bottleneck, vol 357. In:
Proceedings of PFLDNeT workshop
22. Zhao J, Liu J, Wang H (2017) On energy-efficient congestion control for multipath TCP. In:
2017 IEEE 37th International Conference on Distributed Computing Systems (ICDCS). IEEE
23. Hwang J, Walid A, Yoo J (2018) Fast coupled retransmission for multipath TCP in data center
networks. IEEE Syst J 12(1):1056–1059
24. Cao Y, Xu M, Fu X (2012) Delay-based congestion control for multipath TCP. In: 2012 20th
IEEE International Conference on Network Protocols (ICNP). IEEE
25. Dong P et al (2019) Tuning the aggressive slow-start behaviour of MPTCP for short flows.
IEEE Access 7:6010
26. Zhou D, Wei S, Minghui S (2013) Goodput improvement for multipath TCP by congestion
window adaptation in multi-radio devices. In: 2013 IEEE 10th Consumer Communications and
Networking Conference (CCNC). IEEE
MPTCP a Solution Over Crunch in Bandwidth Requirement … 663

27. Paasch C et al (2012) Exploring mobile/WiFi handover with multipath TCP. In: Proceedings of
the 2012 ACM SIGCOMM workshop on Cellular networks: operations, challenges, and future
design. ACM
28. Lim Y-S et al (2015) Design, implementation, and evaluation of energy-aware multi-path
TCP. In: Proceedings of the 11th ACM conference on emerging networking experiments and
technologies. ACM
29. Kaup F et al (2015) Can multipath TCP save energy? A measuring and modeling study of
MPTCP energy consumption. In: 2015 IEEE 40th conference on Local Computer Networks
(LCN). IEEE
30. Wu J, Bo C, Ming W (2016) Energy minimization for quality-constrained video with multipath
TCP over heterogeneous wireless networks. In: 2016 IEEE 36th International Conference on
Distributed Computing Systems (ICDCS). IEEE
31. Wu J et al (2017) Quality-aware energy optimization in wireless video communication with
multipath TCP. IEEE/ACM Transactions on Networking 25.5:2701–2718
32. Deng S et al (2014) Wifi, lte, or both?: measuring multi-homed wireless internet performance.
In: Proceedings of the 2014 conference on internet measurement conference. ACM
33. Zeng J et al (2017) Multi-attribute aware path selection approach for efficient MPTCP-based
data delivery. J Internet Serv Inf Secur 7.1:28–39
34. Alheid A, Kaleshi D, Doufexi A (2014) Performance evaluation of MPTCP in indoor het-
erogeneous networks. In: Proceedings of the 2014 first international conference on systems
informatics, modelling and simulation. IEEE Computer Society
35. Xue K et al (2018) DPSAF: forward prediction based dynamic packet scheduling and adjusting
with feedback for multipath TCP in lossy heterogeneous networks. IEEE Trans Veh Technol
67.2:1521–1534
36. Ageneau P-L, Nadia B, Mario G (2017) Practical random linear coding for MultiPath TCP:
MPC-TCP. In: 2017 24th International Conference on Telecommunications (ICT). IEEE
37. Priyadarshini MS, Rekh AS (2016) Streaming video with MultiPath TCP in wireless networks.
IEEE Trans Mobile Comput 15.4:2345–2361
38. James C et al (2016) Is multipath TCP (MPTCP) beneficial for video streaming over DASH? In:
2016 IEEE 24th international symposium on Modeling, Analysis and Simulation of Computer
and Telecommunication Systems (MASCOTS). IEEE
39. Su G-M et al (2016) QoE in video streaming over wireless networks: perspectives and research
challenges. Wireless Netw 22.5:1571–1593
40. Uppu P, Kadimpati S (2013) QoE of video streaming over LTE network
41. Diop C et al (2012) QoS-oriented MPTCP extensions for multimedia multi-homed systems.
In: 2012 26th international conference on advanced information networking and applications
workshops. IEEE
42. Wu J et al (2017) Quality-aware energy optimization in wireless video communication with
multipath TCP. IEEE/ACM Trans Netw 25.5:2701–2718
Design of AYUSH: A Blockchain-Based
Health Record Management System

A. V. Aswin, K. Y. Basil, Vimal P. Viswan, Basil Reji and Bineeth Kuriakose

Abstract We are living in a world where data is considered to be the next fuel; also,
we know how much value the data is. Considering this over the health records on
which a patient deals with whenever he/she goes to a hospital, the present scenario
lags in securing the patients’ health record management system in order to provide
more transparency over patients’ past health data. If the same was available, then
data sharing between hospitals would have been much easier. If the hospital gets to
know the past health history like the amount of drugs the patient is consuming as
part of medication, then the doctor could make a stern decision over the disease and
could also satisfy the patient with a better treatment over his symptoms. The basic
problem to be dealt with these would be the privacy consideration of the patient when
he/she is sharing the data. In order to tackle that problem, we are using a patient-
centric health record management system under the distributed network architecture.
In this paper, we are proposing a solution that makes use of the emerging blockchain
(permissioned) technology to achieve this goal.

Keywords Blockchain · EHR · Hyperledger Fabric · Health care · Ethereum

A. V. Aswin · K. Y. Basil · V. P. Viswan · B. Reji · B. Kuriakose (B)

Muthoot Institute of Technology and Science, Ernakulam, India
e-mail: bineethbinz@[Link]
A. V. Aswin
e-mail: aswinavofficial@[Link]
K. Y. Basil
e-mail: basilky145@[Link]
V. P. Viswan
e-mail: vimal974ever@[Link]
B. Reji
e-mail: basilreji3@[Link]
B. Kuriakose
Oslo Metropolitan University, Oslo, Norway

© Springer Nature Singapore Pte Ltd. 2020 665

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
666 A. V. Aswin et al.

1 Introduction

The advancement and impact of technology can be found in every field of the soci-
ety. Considering this over the health records which a patient deals with whenever
he/she goes to a hospital, the present scenario lacks a secure patient’s health record
management system to provide more transparency over patients’ past health data.
Also, there is an absence of a widely accepted digital standard for sharing EHR data.
If the hospital gets to know the past health history of a patient like past diseases, the
amount of drugs the patient is consuming as part of medication, medical test results,
allergies, etc., then the doctor could make a stern decision over the disease and could
also satisfy the patient with a better treatment. This can also benefit patient by getting
a better treatment at lower cost and reduced time. Unnecessary medical tests can be
avoided if the past history of patient is available.
In this paper, we are proposing a solution to these problems in healthcare sector
by applying emerging blockchain technology. In the AYUSH network, there will be
patients and service providers like hospitals, doctors, and laboratories. Patients can
securely create an account on AYUSH, and all the upcoming medical records of the
patients will be added to the AYUSH platform. Easier user interface and APIs will
be provided by the platform for easier access for participants.
Saving the history of patients in the blockchain can provide many advantages.
Since there is no central administrator, patient will have full control over his/her
entire health records and he/she can determine for whom permission should be given
to access their data. Built-in transparency how the blockchain system works can
bring trust between participating entities. Existing symmetric and asymmetric cryp-
tography principles can be used along with the blockchain system to provide privacy
control for patients. All the data of patients can be encrypted by their private key,
and needed records can be selectively decrypted for sharing. The distributed record
keeping nature of blockchain can ensure that patient data is not lost by a single point
of failure. Health records in blockchain are series of records logically connected
using the hashes of records. This provides easier traceability to health records on
blockchain, along with resistance toward tampering of digital records. Immutabil-
ity of blockchain records can bring easier claim adjudication. The transaction rates
of Hyperledger, the blockchain platform we are using, are also higher. Patients can
benefit even if they have a long history of records. Researchers are also the benefi-
ciaries of the platform, since they can avail large amount of anonymous health data
for finding new insights.

2 Blockchain in Health Care

Blockchain [1] is a distributed and decentralized chain of records or ledgers. The

immutability of blockchain records ensures trust between participating entities. This
has brought many applications for blockchain including health care.
Design of AYUSH: A Blockchain-Based Health Record … 667

2.1 Problems in Healthcare Sector

Current healthcare infrastructure lacks proper mechanisms for the exchange of

healthcare information. There is no accepted standard for sending, receiving, and
managing information between EHR systems. Most of the cases, patient has to carry
all of his past medical records to the new hospital or has to do the previous medi-
cal tests again. Lack of health history of a patient may lead to improper treatment
also. Another problem is the lack of patient-centric longitudinal list of records. Thus,
patient data gets scattered across multiple records at multiple hospitals. The current
health data keeping systems does not provide any resistance toward tampering and
cannot be used for claim adjudications on insurances. Currently, there are chances
that healthcare service providers may use the patient’s data for analytics, without con-
sidering their privacy. The health data is unreliable since it is not stored at multiple
locations.

2.2 Advantages of Using Blockchain

Blockchain is perfectly suited for health care because of its distributed, immutable
implementation. Blockchain can ensure trust between participating entities. Privacy
and security of data can be preserved by using cryptography principles. The transac-
tion speed of blockchain is much faster, and transactions are traceable from beginning.
Distributed nature of blockchain ensures data reliability also.

2.3 Challenges in Implementation

Even though there are good reasons behind applying blockchain on healthcare sector,
there are many challenges which have to be solved for proper implementation of
blockchain over health care. There are limited successful blockchain models and the
success of blockchain on health care, and the cost that will incur is uncertain. Health
data is huge, and replicated storage of data will create storage capacity problems.
The platform should be scalable as more service providers are added, and they should
not be considerable delay. Different standards for keeping and sharing EHR are also
a major challenge. Various rules like HIPAA and unwillingness to share health data
by service providers are also a major problem.
668 A. V. Aswin et al.

3 Familiarization of Tools

3.1 Hyperledger Fabric

Hyperledger Fabric [2] is an open-source enterprise-grade permissioned distributed

ledger technology (DLT) platform designed for use in enterprise contexts that deliv-
ers some key differentiating capabilities over other popular distributed ledger or
blockchain platforms. Hyperledger Fabric has a highly modular architecture which
is configurable. It enables innovation, versatility, and optimization for a broad range
of applications including banking, finance, insurance, health care, human resources,
supply chain, and even digital music delivery.
The taxonomy of blockchain includes public, private, consortium blockchains,
where public blockchain is completely distributed in terms of its architecture,
whereas consortium blockchain is considered to be partially decentralized and private
blockchain is said to be a centralized one. The Fabric platform provides a permis-
sioned network. In a permissioned network, the resources that are being shared in
the network are permissioned; that is, only authorized persons get access to it. This
means that while the participants may not fully trust one another (they may, e.g., be
competitors in the same industry), a network can be operated under a governance
model that is built off of what trust does exist between participants, such as a legal
agreement or framework for handling disputes [2].

3.2 InterPlanetary File System (IPFS)

InterPlanetary File System (IPFS) [3] is a new hypermedia distribution proto-

col addressed by content and identities. IPFS enables the creation of completely
distributed applications. It aims to make the Web faster, safer, and more open.
IPFS is a distributed file system that seeks to connect all computing devices with
the same system of files. In some ways, this is similar to the original aims of the
Web, but IPFS is actually more similar to a single BitTorrent swarm exchanging git
objects. After a file is successfully added onto the IPFS, it computes a hash of the
file and now other systems will be able to access this file through this calculated
hash value. Consider the situation of a PDF file. Let the file be stored on the IPFS
network. If you run your own node, the file would only be stored on your node and
available for the world to download. If someone else downloads it and sends it, then
the file is stored on both your nodes, and so on [3].
Design of AYUSH: A Blockchain-Based Health Record … 669

4 Proposed Work

In the proposed system, we are applying the emerging blockchain technology in

creating transparent health record chain. When the patient goes to a new hospital
after similar visit in yet another hospital, he/she opts the data to be shared to the
new hospital. If any new health data is produced at the new hospital, they first add
the information (clinical reports or other scan reports) into the IPFS file system. The
hash of this file is added to the blockchain.
The proposed solution is a patient-centric one; that is, the patient has got the power
to give access to his data. Hyperledger Fabric is used since the system requires a
permissioned network.

4.1 Architecture

This section will discuss the architecture of the proposed system, AYUSH.

4.1.1 API Module

API module acts as an interface between AYUSH platform and end users (patients,
hospitals, and doctors). It abstracts implementation details of AYUSH platform, and
users can interact using simple UI and HTTP requests (Fig. 1).

4.1.2 AYUSH Platform

AYUSH platform abstracts implementation details of Hyperledger Fabric and IPFS

and provides services to API module. Various modules present in AYUSH network
are:
1. Fabric client;
2. IPFS interface;
3. Auth module;
4. Membership module;
5. Application logic and database.
IPFS Interface
InterPlanetary File System (IPFS) [3] interface is used to communicate with IPFS to
upload or fetch from IPFS network.
Auth Module
Auth module is used to recognize a user’s identity, allowing only legitimate users to
interact with the platform.
670 A. V. Aswin et al.

Fig. 1 AYUSH architecture

Membership Module

Membership module is a component that aims to offer an abstraction of a membership

operation architecture. In particular, it abstracts away all cryptographic mechanisms
and protocols behind issuing and validating certificates and user authentication. It
issues certificates which are used by peers to join the Hyperledger Fabric network.

Application Logic and Database

This module will be responsible for overall operation of the platform, and it has a
local database also.
Design of AYUSH: A Blockchain-Based Health Record … 671

5 Methodology

Every patient in the AYUSH network will have a public key and a private key, which
are created at the time of registration. All the data of patients are encrypted using their
private key. To share the data, decryption is needed using their private key, which is
encrypted using their passwords. So, patient decrypts his private key and selects the
health records which have to be shared. Then, the selected data is decrypted at client
side. The decrypted data is again encrypted using the public key of service provider,
to provide additional security to ensure that only required service provider receives
the data. The integrity of the shared file can be verified at the receiver end just by
re-calculating the hash of the file and comparing it with the original hash which is on
the AYUSH blockchain network. After receiving the shared files at service provider,
they analyze the past data to give better treatment for the patient. After necessary
observation, the textual details like symptoms are added to the blockchain. The
treatment records of the patient may be of huge size, which is difficult to put in the
blockchain since it will increase the time required to calculate the hashes. This might
also cause reduced transaction rates. To avoid this, the heavy records are stored on
the IPFS, after encrypting the records with public key of the patient. The hash of the
original field is added to the blockchain to later check the integrity of the file, if it gets
shared with any other service provider. The hash of the encrypted file is generated
by IPFS, after uploading the file into IPFS. This helps to uniquely identify files in
the IPFS system and to prevent duplicate upload of files. The record is now spread
to all the nodes in the network from the node on which it was uploaded. This makes
the data reliable and immutable to the IPFS system. No one other than patient can
decrypt this data, since only he has the matching private key pair. The newly added
record becomes a part of the personal health records of the patient. He/she can later
share these files with other service providers by following the similar process. All
the encryption, decryption, and key generation process are performed at the client
side for improving the security.

6 Conclusion

We are now able to witness the revolutionary changes that technologies have brought
to various sectors in the society which makes human life much more simpler. Even
though we have achieved a lot in various sectors, the healthcare sector decades back in
terms of technological advancements. The blockchain is considered to be one of the
promising technologies with its application over many sectors. Bringing blockchain
into healthcare sector could enhance the trust between participating entities which
contributes to the betterment of healthcare record management system while consid-
ering the privacy of patients’ data. We hope the proposed design of AYUSH which
can solve the data sharing and privacy issues faced by health industry today.
672 A. V. Aswin et al.

References

1. Blockchain: The Chain of Trust and its Potential to Transform Healthcare—Our Point of View,
IBM Global Business Services Public Sector Team 6710, 8 Aug 2016
2. Chen Y, Li H, Li K, Zhang J (2017) An improved P2P file system scheme based on IPFS and
blockchain. 2017 IEEE Int Conf Big Data (Big Data). [Link]
3. Ekblaw A, Azaria A, Halamka JD, Andrew Lippman MD (2016) A case study for blockchain in
healthcare: “MedRec” prototype for electronic health records and medical research data, MIT
Media Lab
Support Vector Machine-Based Focused
Crawler

Vanshita R. Baweja, Rajesh Bhatia and Manish Kumar

Abstract The Internet is an immense source of information. People use search

engines to find desired web pages. All these web pages are gathered from the search
engine by using web crawler. In traditional crawler, the information retrieval was
based on the occurrence of keywords in a document due to which many irrelevant web
pages were also retrieved. For the effective classification of web pages, support vector
machine (SVM)-based crawler model is proposed in this paper. Various features of
URL and web page are used for effective classification. SVM is trained by using
these features and further tested. The proposed model is analyzed using precision
and recall metrics. The experimental results exhibit optimized results by using this
proposed approach.

Keywords Support vector machine · Focused crawler · Feature extraction · Web

page classification · Uniform resource locator

1 Introduction

Web pages are the source of information exchange in today’s world. For example,
many firms can publicize their stocks and sales through web pages. If any person
has interest in football match, they can watch matches live on web pages. All the
web pages are gathered by the help of web crawler [1]. It is a tool that searches for
a specific subset of the web rather than exploiting all regions of the web.
There are many search engines provided to users for searching the set of keywords
like Google. The search engine serves as a tool that provides many desired web pages.
Typical search engines use only keyword searching for web page classification. They

V. R. Baweja (B) · R. Bhatia · M. Kumar

Computer Science, PEC University of Technology, Chandigarh, India
e-mail: vanshitarbaweja@[Link]
R. Bhatia
e-mail: rbhatiapatiala@[Link]
M. Kumar
e-mail: manishkamboj3@[Link]
© Springer Nature Singapore Pte Ltd. 2020 673
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
674 V. R. Baweja et al.

Fig. 1 Taxonomy of supervised classification methods [2]

may give irrelevant web pages and unrelated links. Many machine learning methods,
supervised and unsupervised exist in literature. Supervised algorithms are used to
train machine learning task for every input with corresponding outputs, whereas
unsupervised algorithms are performed over machine learning task with a set of
inputs which in turn gives relationships between different inputs. Figure 1 represents
the classification of supervised learning algorithms.
This paper presents a novel web page classification framework that uses SVM
as a classifier. SVM classifies the samples using an optimal hyperplane. An optimal
hyperplane is determined by the samples closet to it and not bothering any other
sample called support vectors.
The rest of this paper is organized in the following manner. Section 2, presents an
overview of SVM framework and its types. Section 3 presents the literature survey.
Section 4 explains the proposed model of SVM web page classification model. The
SVM classification method is explained in Sect. 5. In Sect. 6, experimental results
are presented. Then, conclusions are explained in Sect. 7.
Support Vector Machine-Based Focused Crawler 675

2 Support Vector Machine

2.1 Overview

Support vector machine is a supervised machine learning algorithm and can be used
as classification and regression model. SVM was introduced by Vapnik and Chervo-
nenkis [3] in 1963. In 1992, Vapnik suggested nonlinear classification with the help
of kernel trick [3].
SVM can be used for classification or regression by forming a hyperplane or a set
of hyperplanes. During the training of SVM quadratic problem arises, this problem
can be solved by sequential minimal optimization (SMO). Therefore, SMO is used
for training of SVM.

2.2 SVM Working

In learning a classifier for binary classification, given a set of training samples in the
form

( p1 , q1 ), . . . , ( pn , qn )

The input features pi ∈ R are usually d-dimensional vectors describing the prop-
erties of the input example, qi are either −1 or 1 each of them are indicating a class
which belongs to pi . Equation (1) represents the hyperplane that can be used to
distinguish the sample sets in SVM where w is the optimum weight and b is the
excursion.

w∗ p+b =0 (1)

Figure 2 demonstrates the linearly separable case where samples of one category
are portrayed as “z” and samples of other category is portrayed as “•” [4]. These data
points are separated by a hyperplane. There is the number of possibilities of drawing
a hyperplane but according to SVM methodology, only one optimal hyperplane is
selected. An optimal hyperplane lies between maximum margin, i.e. d1 + d2 and
determined by the samples closest to it and not bothering any other sample. As
support vectors determine the optimal hyperplane, the maximal margin can be found
by minimizing 21 w2 , as shown in Eq. (2)
 
1
min w2 (2)
2

An optimal hyperplane can be configured by minimizing (2) under the constraint

of (3)
676 V. R. Baweja et al.

Fig. 2 Optimal separating hyperplane [4]

qi ∗ (w ∗ pi + b) ≥ 1, ∀i (3)

By the use of kernel functions, hyperplane can be optimized for nonlinear cases
too. SVM can be used for nonlinear cases, by implementing the kernel functions. The
implicit mapping (φ(.)) of the input samples ( pi , p j ) to high-dimensional feature
spaces by the use of kernel function is illustrated in Fig. 3. Hence, the kernel function
is given by Eq. (4):
 
K ( pi , p j ) = φ( pi ).φ( p j ) (4)

Fig. 3 Mapping of input space to high-dimensional feature space [4]

Support Vector Machine-Based Focused Crawler 677

3 Related Work

There are many machine learning methods for classification purpose and SVM is
popularly used for different kinds of classification and documented as the best clas-
sification algorithm [2, 5, 6]. The best training characteristic of SVM is implemented
by SMO that helps in finding an optimal hyperplane and hence results in highly
accurate classification in most applications. In this paper, SVM is used for web page
classification and features are extracted from the content of web page.
Joachims [5] used the technique in which documents are transformed to strings
of characters. Information gain is the criterion used to select subsets of features. It
analyses the properties of learning with text. SVM attains the considerable enhance-
ments over the best-performing methods and results in a vigorous manner over a
variety of many learning tasks.
Basu et al. [6] compared different algorithms used as text classifiers, classifying
news items and identifying a reduction in feature set that helps to improve the results.
It is observed that artificial neural network is more complex in terms of computation
than SVM algorithm.
Kan et al. [7] used URL features like length, orthographic features, sequential
Bi-, Tri-, 4-grams, URI components, entropy reduction for web page classification.
Resulting feature is used in supervised maximum entropy modelling. The research
says that lower the entropy more meaningful URL will be found.
Chen et al. [2] proposed a web page classification model which consists of two
methods, i.e. latent semantic analysis (LSA) and web page feature selection. These
methods are used to extract semantic and text features. To improve the performance
factor, they have used feature selection criteria and a similarity measure and its evalu-
ation. It works for both small and large data sets. The problem of linear inseparability
can be resolved by selecting an appropriate kernel function.
Wang et al. [8] proposed a focused crawler based on naïve Bayes to find the
relevant pages according to the research. They use reinforcement learning to train
the crawler and naïve Bayes classifier is used for evaluating the class of new web page.
The importance of the word in link context is calculated using TF-IDF vectorization.
It is observed that it has better performance than PageRank crawler. But this focused
crawler can only be applied on the small datasets.
Lee et al. [4] proposed a text document classification framework that uses SVM
in training phase and Euclidean distance function in the classification phase. This
framework has a very low impact on the implementation of different kernel func-
tions and parameter c. C is the parameter that controls the influence of each support
vector. The classification decision is based on the category that has the lowest aver-
age distance (Euclidean distance) between support vectors and unknown data point.
Parameter C does not have a great impact on SVM because each feature is unique.
Selvakumar et al. [9] proposed technique that increases the efficiency of the
crawler by using the anchor text of links and by checking the revisit policy. This
technique is able to manage large spaces of features and high generalization ability.
678 V. R. Baweja et al.

In result, this makes the SVM more complex that in turn demand for high time and
memory ability.
Shafiabady et al. [10] describe a methodology that uses self-organizing maps
(SOM) and alternatively does the clustering by using the correlation coefficient
(CorrCoef). They are eliminating the effect of curse of dimensionality. SOM and
CorrCoef are used for organizing and labelling the unlabelled data. The proposed
clustering technique is able to match the actual human data with a high degree for
accuracy.
Several research gaps observations have been made through the literature review,
few of them are discussed here. Traditional crawling methods require huge amounts
of time and cost. In web page classification, to reduce the dimensionality of feature
space, there is always a need to select best feature subsets. Most of the researchers
have worked on small datasets. The classified model should work in such a way that
it categorizes the entailed web pages in an effective manner such that the desired
web pages can be extracted time efficiently. This optimizes the existing models
and classifies web pages effectively. Therefore, the main objective of this paper is
to propose a framework for web page classification that uses SVM-based focused
crawler model.

4 Proposed SVM Web Page Classification Model

Various Indian universities associate with Indian academicians working for univer-
sities across the globe. Generally, when searching for the names of these Indian aca-
demicians from a domain, users are engrossed in particular web pages/topics within
that domain, such as Harvard University domain has web pages of faculty, Ph.D., or
alumni, events, curriculum and student details [11]. The users are mostly interested
in faculty, Ph.D. and alumni web pages. However, searching specific names from
aforementioned web pages takes a lot of time. Thus, to lessen the time, proposed
approach classifies the desired topics/web pages in a particular domain.
The proposed approach aims the web page classification using SVM-based
focused crawler. The workflow of the proposed approach is shown in Fig. 4. It is
divided into two modules:

4.1 URLs Screening

Crawled URLs of particular university are taken as input. All the URLs are being
preprocessed and learned under the algorithm. Features are extracted from URLs’
text by finding TF-IDF values and treated as training input for SVM classification
system. Then, the classification decision-making modules classify the URL into two
categories, i.e. relevant URLs and irrelevant URLs. Relevant URLs are related to
Support Vector Machine-Based Focused Crawler 679

Fig. 4 Workflow for the proposed model

certain domains like department, faculty, alumni, Ph.D. scholars, whereas irrelevant
URLs are the remaining ones. Algorithm 1 illustrates the algorithm of the URL
screening approach.
680 V. R. Baweja et al.

Algorithm1. Algorithm for URL Screening

Input: URLs list of university u, Labelled data lb
Output: list of URLs related to faculty
for each URL u in the given list do //Preprocessing Phase
Tokenize (u);
Stemming (u);
Final_token = Remove_stopwords (u);
If (final_token in Bag_of_words) do //Labeling
Label_URL = 1
Else do
Label_URL = 0
For each Labelled data ld in the list of URLs do //Learning Phase
Apply filtered classifier (ld)
Apply SVM classifier (ld)
X_train, X_test, y_train, y_test = cross_validaƟon.train_test_split(X, y, test_size=0.2)
Classifier = SMO
[Link] (ld)
[Link](X_train, y_train)
Train_URLs(ld)
for each new URL u in the given list do //PredicƟon Phase
Repeat the preprocessing phase for u
Use the saved model for the predicƟon.
Algorithm 2. Algorithm for Web page classificaƟon
Input: list of filtered URLs u, list of names fetched_names, cross validaƟon data
Output: list of URLs containing Indian names
for each URL u in the given list do // Preprocessing Phase
Tokenize (u);
final_token = Tagging (u);
if (final_token == “PERSON”) do
create list as fetched_names
for each fetched_names in list do
if (fetched_names NOT in dicƟonary) do
create list as modified_names
for each modified_names in surnames_list do // Labelling
if (modified_names in surnames_list) do
Label_URL = 1
Else do
Label_URL = 0
For each Labelled data ld in the list of URLs do // Learning Phase
Apply filtered classifier (ld)
Apply SVM classifier (ld)
X_train, X_test, y_train, y_test = cross_validaƟon.train_test_split(X, y, test_size=0.2)
Classifier = SMO
[Link] (ld)
[Link](X_train, y_train)
accuracy = [Link](X_test, y_test)
Train_URLs(ld)
for each new URL u in the given list do // PredicƟon Phase
Repeat the preprocessing phase for u
Use the saved model for the predicƟon.
Use the saved model for the predicƟon.
Support Vector Machine-Based Focused Crawler 681

4.2 Web Page Classification

All the screened URLs are considered as seed URLs for further module. Now, seed
URLs are being preprocessed and learned under the SVM algorithm. Firstly, web page
content is extracted and this content is tagged so as to find names from the web page.
Then, extracted names are matched with list of Indian surnames. All these filtered
names are processed and converted into numeric by finding each TF-IDF values.
Then, these are considered as features and they are trained under SVM classification
system. Afterwards, the classification decision-making modules classify the URL
into two categories, i.e. relevant pages and irrelevant pages. Relevancy of a page is
considered when Indian name is found in a web page. Algorithm 2 illustrates the
algorithm of the web page classification approach.
The proposed approach involves SVM classification system that consists of three
phases: preprocessing phase, learning phase and prediction phase. These phases are
described in further section. Section 5 elaborates the working of all phases in brief.

5 SVM Classification System

The proposed workflow of SVM classification is shown in Fig. 5. It is divided into

three phases:

5.1 Preprocessing Phase

Before extracting features of webpage, preprocessing of web page should be per-

formed. Preprocessing phase consists of removal of HTML tags, tokenizing of words
and stemming.
Removal of HTML tags. Mostly, web pages are formulated in HTML. HTML
has open and closed tags represented by ‘<’ and ‘/>’. They create the hindrance in
analysis. Therefore, HTML tags were removed to prevent interference and text is
retained where needed.
Tokenization. Tokenization of words is done over the text extracted from webpage
content. After tokenizing, the following operations are performed:
• Some of the repetitive words are considered as stop words. Stop words like a, an,
the, of, is, am, are and many more. They increase the data size and increase the
processing time. To avoid this, stop words are removed where required.
• It reduces the inflected words as many of them have the same meaning and are
repeated and increases the data set size. To avoid this, stemming of words are
performed where and when required. For example, graduates, graduating reduces
to graduate, and applies, applying, apply, applied all reduces to apply.
682 V. R. Baweja et al.

Fig. 5 SVM classification system framework

Tagging. After tokenizing, all words are tagged as proper noun, singular (NNP),
predeterminer (PDT), adverb (RB), verb (VB) and many more. Tagging is performed
to find names easily which comes under person tag.

5.2 Learning Phase

After processing of data, these tokens are used for labelling. Labelling is done by
comparing them with some bag of words. Bag of words contains important words
related to department, faculty, alumni, post-graduates and many more. Then, system
gathers web pages’ content for features. These features include number of keywords,
frequency of a word in a document, TF-IDF vector of these words. This processed
data is given to filtered classifier as input. Under filtered classifier, SVM classifier is
applied to train the processed data.
Support Vector Machine-Based Focused Crawler 683

5.3 Prediction Phase

When a new document has to be predicted, again its features are extracted accord-
ingly. Using the prediction model (estimator algorithm), prediction is performed.

6 Result Analysis

This section narrates the design experiment of the model to test the performance of
SVM web page classification. The experiments are reported below:

6.1 Experiment Environment

The experiment uses Core i5 2.50 GHz computer with 4 GB RAM. Input is generated
using python language on Windows 8.1 Professional operating system. The lexicon
is stored in text format. A tool, Weka, is used for classification. Waikato Environment
for knowledge Analysis (Weka) has a group of machine learning algorithms. SVM
is implemented using this tool.

6.2 Dataset

Crawled URLs are classified as relevant and irrelevant. These URLs are of foreign
universities, i.e. Harvard and Lancaster. The University domain has different inter-
linked domains like faculty, student, alumni or events related. There is need to classify
them according to the need of domains. For the cross validation, there is a random
selection of training set and testing set from samples.

6.3 Performance Evaluation

In the experiment, dataset is having four possible outcomes as shown in Table 1.

Precision and recall values are analyzed to evaluate the performance of the
experiment.
684 V. R. Baweja et al.

Table 1 Four possible outcomes of classification

Predicted class
System does not classify System classify to
to correct category correct category
Actual class Does not belong to True negative False positive
category
Belongs to category False negative True positive

6.4 Result

The foreign university URLs are treated as dataset and used for training and testing
purpose. The datasets are processed through the vectorization step that transforms
each URL or URL content into numerical format. This representation of dataset in
numerical form fulfil the requirement of SVM, since all the machine learning meth-
ods require the vector space model and accept only numerical data to perform their
tasks. Apart from data transformation, preprocessing steps also reduces the dimen-
sionality of the dataset. The output details of URL screening process is mentioned in
Table 2. Although, the final goal is to minimize the intervention of humans in training
and classification process. After the screening of some URLs, all the relevant URLs
of this classification are the input of next classification module, i.e. web page classi-
fication. In this experiment, cross validation is performed using SVM with ten-folds.
Confusion matrices are obtained during the web classification process and shown in
Table 3 with respect to their datasets. Confusion matrix illustrates the performance
of a classification model on the dataset.
At last, web page classification is performed to find URLs containing Indian
names or not. The results are shown in Table 4.

Table 2 Four possible

Dataset Samples Class
outcomes of classification
Train Relevant URLs Irrelevant URLs
Harvard 10,000 1766 8234
Lancater 10,000 5337 4663

Table 3 Details of cross

Dataset Folds Confusion matrix
validation  
Harvard 10 2628 34
7 7336
 
Lancaster 10 1347 242
58 3434
Support Vector Machine-Based Focused Crawler 685

Table 4 Details of web page

Dataset URLs Relevant URLs Irrelevant URLs
classification
(Indian names (Indian names not
found) found)
Harvard 1766 361 1404
Lancaster 5337 1309 4028

Table 5 Comparison
Dataset Machine learning Precision Recall
analysis
Harvard SVM 0.938 0.846
Naïve Bayes 0.862 0.828
Lancaster SVM 0.94 0.94
Naïve Bayes 0.82 0.78

6.5 Comparative Analysis

Comparative analysis is done over different supervised machine learning algorithms.

The best-supervised algorithm is SVM for classification purpose. SVM gives better
results compared to naïve Bayes for two datasets as shown in Table 5.

7 Conclusion

This paper discusses the need for creating a database of Indian academicians extracted
from foreign universities. For this, SVM classification approach is reported to have
good accuracy compared to others. In web page classification, there is a need to
select best subsets of features to reduce dimensionality of feature space that will
improve the time and efficiency of the classifier. To solve this issue SMO has been
implemented on dataset using weka. This gives a fair binary classification in finding
web pages having Indian names. The experiments show, this classification approach
gives a satisfying precision value. To have more accuracy in the results, there is a
need to give large training data set to the algorithm.

References

1. Rungsawang A, Angkawattanawit N (2005) Learnable topic-specific web crawler. J Netw

Comput Appl 28(2):97–114
2. Chen RC, Hsieh CH (2006) Web page classification based on a support vector machine using
a weighted vote schema. Expert Syst Appl 31(2):427–435
3. Wikipedia, Support Vector Machine. [Link]
686 V. R. Baweja et al.

4. Lee LH, Wan CH, Rajkumar R, Isa D (2012) An enhanced support vector machine classification
framework by using euclidean distance function for text document categorization. Appl Intell
37(1):80–99
5. Joachims T (1998) Text categorization with support vector machines: learning with many
relevant features. Mach Learn ECML 98:137–142
6. Basu A, Walters C, Shepherd M (2003) Support vector machines for text categorization. In:
Proceedings of the 36th annual hawaii international conference on system sciences. IEEE
(2003)
7. Kan MY, Thi HON (2005) Fast webpage classification using URL features. In: Proceedings
of the 14th ACM international conference on Information and knowledge management. ACM
(2005)
8. Wang W, Chen X, Zou Y, Wang H, Dai Z (2010) a focused crawler based on naive bayes
classifier. In: Third international symposium on intelligent information technology and security
informatics. Jinggangshan, IEEE 2010, pp517–521
9. Selvakumar M, Vijaya A (2014) Design and development of a domain specific focused crawler
using support vector learning strategy. Int J Innov Res Comput Commun Eng 2(5)
10. Shafiabady N, Lee LH, Rajkumar R, Kallimani V, Akram NA, Isa D (2016) Using unsupervised
clustering approach to train the support vector machine for text classification. Neurocomputing
211:4–10
11. Kumar M, Bhatia R, Ohri A, Kohli A (2016) Design of focused crawler for information retrieval
of Indian origin academicians. In: 2016 international conference on advances in computing,
communication, and automation (ICACCA) (Spring), IEEE 2016, pp 1–6
Indoor Object Tracking Using Wi-Fi
Access Points

Vinit Asher, Hardik Thakkar, Suyog Tambe and Prasenjit Bhavathankar

Abstract With the rise of new analytical methods to track and predict various aspects
of our lives, gathering data is as important as ever. The way an object moves indoors
is an important aspect in terms of crowd flow management, retail marketing, and
security fields. Wi-Fi has become a ubiquitous standard with almost every indoor
space having some form of Wi-Fi device. This existing infrastructure is put into
service to track an object based on received signal strength values. Received signal
strength values from different access points form a unique fingerprint which can be
used to recognize a location on the map. In order to make the system adaptive to
change and drift in fingerprints, deep learning methods are used to convert the system
from a static to a dynamic system. The proposed neural network model has the error
of 1.3 m for predicting the x-coordinate and 1.8 m for the y-coordinate in the given
experimental setup.

1 Introduction

Global positioning system (GPS) has become the de facto standard for providing
navigation and tracking services. GPS works well when there is a direct line of sight
between sender and receiver. The accuracy of GPS increases as the number of satel-
lites, it gets its signal from, increases. The ultra-high frequency (UHF) signals used
by GPS also face disturbances due to other UHF sources. Also, GPS struggles indoors
due to the construction materials like concrete not being conducive to transmission

V. Asher (B) · H. Thakkar · S. Tambe · P. Bhavathankar

Sardar Patel Institute of Technology, Mumbai, India
e-mail: vinitasher1997@[Link]
H. Thakkar
e-mail: hthakkar8@[Link]
S. Tambe
e-mail: suyog.tambe52@[Link]
P. Bhavathankar
e-mail: p_bhavathankar@[Link]
© Springer Nature Singapore Pte Ltd. 2020 687
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
688 V. Asher et al.

of radio signals. Hence, to get a viable solution indoors, alternatives such as cell
towers, Bluetooth beacons, ZigBee, and radio frequency identifier (RFID) have been
proposed for indoor tracking. But, they each have their own drawbacks and have not
managed to find a uniform and consistent way of implementation. However, Wi-Fi
is an exciting prospect to fill this gap, as it is already present in most places and
characteristics of various Wi-Fi bands are standardized and well defined.
With the advent of data analysis, various insights about object movement inside
buildings can be analyzed. More the data is collected about movement patterns inside
building, the more improvement can be brought in how the building is designed.
Especially, commercial spaces can have huge advantages from a system to track
people indoors. Navigation inside commercial spaces has potential of giving better
visibility to all shops regardless of their location. How people move inside building
can also give a better idea on which shop locations are more lucrative, and which
advertising locations can get most exposure. As an attempt to make the surroundings
more inclusive to differently abled people, an indoor positioning system can help
them navigate easily inside large buildings. Assets can also be tracked inside the
building, to prevent their theft and easy location of the asset. Home automation is
also a use case for indoor tracking system.
The amount of variation in type of indoor environments which can be encountered
is one of the hurdles in designing such a system. Indoor environments are always
changing, density of crowd changes depending on time of the day; similarly, the
layout of the floor can be changed over time too. Since behavior of Wi-Fi signals varies
in different environments, with signal attenuation and multi-path fading playing a
major role in defining the signal strength at a particular location, it is difficult to
predict how these changes will affect the signal received at a indoor location.
Along with accuracy of the indoor tracking system, effort must also be taken to
make it as unobtrusive as possible and as intuitive as possible.
Machine learning has been applied to areas where an inherent pattern can be
explored and used in efficiently solving a problem. A case can be made for using
machine learning techniques to classify current location of object from the signal
values reported or to use a neural network to predict the current location.
This paper is divided into sections as follows, Sect. 2 gives a concise idea about
related research which has been performed for indoor object tracking. Section 3
contains the proposed system where the problem at hand is formulated. Section 4
contains the results and Sect. 5 concludes the paper.

2 Related Work

Object tracking problem has been approached in many different ways, with methods
differing in terms of how they approach the three major components of any tracking
system, technology used for tracking, representing the environment in the system,
and calculating current location.
Indoor Object Tracking Using Wi-Fi Access Points 689

Choice of technology depends on environment as well as cost factors. LAND-

MARC proposed in [1] used active RFID tags as reference points while calculating
current location. It tries to overcome the problems of traditional RFID reader by
performing scans at eight power levels. However, the latency is traded for accuracy
with the system requiring more time to detect the accurate location. Also, RFID suf-
fers from high disturbances in case of environments with large quantities of metal.
Bluetooth is another choice due to its popularity in consumer electronics. In [2, 3],
authors have made a case for low-energy Bluetooth beacons to be used for indoor
location tracking given their popularity and low-energy needs. Bluetooth requires
lower power than Wi-Fi at the mobile node. However, Bluetooth requires dedicated
beacons which only serve one purpose. It is a trade-off between power and versatil-
ity. [4–6] show the variety in which Wi-Fi-based indoor localization system can be
made. Not only are the aforementioned systems accurate, they do so while utilizing
the existing hardware. Wi-Fi has a higher range than most other alternatives. [7] for-
mulates the number of reference points required for a stable RSS reading depending
on obstructions which are expected to be present in environment. Among findings
in [8], it is shown that multiple APs working on same frequency do not have a neg-
ative effect on the reading. Hence, having a dense network of APs is possible. The
existing infrastructure availability is the major advantage Wi-Fi has over competing
techniques.
Generally, to represent the environment, two categories of solution can be used
either a fingerprint-aided system or a purely model-based system. Fingerprinting, a
process by which a unique value is associated with a location on a map, has been
widely used in indoor tracking systems. The location fingerprints are collected by
performing a site survey of the received signal strength (RSS) from multiple access
points (APs). The entire area is covered by a rectangular grid of points. The RSS
is measured with enough statistics to create a database or a table of predetermined
RSS values on the points of the grid. The vector of RSS values at a point on the
grid is called the location fingerprint of that point [9]. In [10], authors have surveyed
a collection of various mathematical models which can be used to track an object.
Features of received signal like angle of arrival, time of flight, and time difference
of arrival help determine where the object is currently.
K-nearest neighbors (KNN) is a popular algorithm which relies on the online RSSI
to obtain the ‘k’ nearest matches (on the basis of off-line RSSI measurements stored
in a database) of the known locations using root mean square error (RMSE) [10]. But
the good performance of KNN is highly dependent on the metric used for computing
pair ware distances between data points [11]. Trilateration determines the position of
node from intersection of 3 circles of 3 anchor points that are formed based on distance
measurements between its neighbors [12]. However, trilateration requires more than
3 anchors for getting a confident prediction. More recently, 802.11mc Wi-Fi standard
has added round trip time (RTT) support which is based on finding distance from APs
based on time required for a packet to complete one trip between AP and mobile.
Artificial neural networks (ANN) are also fit for this task as shown in [13, 14].
With availability of deep learning libraries for deployment on mobile devices, neural
networks have potential of providing high accuracy in dynamic environments.
690 V. Asher et al.

3 Proposed System

In the following section, the proposed object tracking system to find the location of
user based on the RSSI values from the available access points is described.

3.1 System Block Diagram

Figure 1 shows the block diagram of proposed system. A Wi-Fi enabled object is at
the core of the system. The object collects the RSSI values from various APs. The
object is connected to either Intranet or Internet from where it can fetch the map data.
For the initial call, to select the map, current RSSI readings are sent to the server.
Once the map is downloaded with the pre-trained model of the floor, the object itself
can calculate its current position. Current position is calculated by passing the current
RSSI readings into the neural network model.

3.2 Off-line Fingerprinting and Database Description

The off-line phase is very crucial step of the process as it is required to collect the data
for training the system to predict the actual location of the user in real time. In the
off-line fingerprinting phase, the system has to be calibrated with respect to a single
AP considering it as an origin. The RSSI values from different AP’s are recorded
automatically and the coordinates (x, y, z) are stored with respect to the point which
is calibrated as origin. This process of taking reading has to done continuously till the
complete area of interest is covered at least once. This process is very tedious and to
amplify the performance, the readings at a certain point are taken multiple times by
taking the readings from the AP’s automatically. These extra scans for RSSI values at
the same point help in training the model well and reduce the error in predicting the
position of object accurately. The Dataset is created by all these RSSI values which

Fig. 1 System block diagram

Indoor Object Tracking Using Wi-Fi Access Points 691

are collected by fingerprinting this dataset is then used to predict the coordinates of
the receiver in real time. The dataset has the following attributes—
X—The x-coordinate based on the grid
Y—The y-coordinate based on the grid
Z—The z-coordinate (basically the floor number)
AP’s—RSSI from different access points.
The data from all the AP’s is saved in separate columns.
Some preprocessing is required on this dataset as there are various factors which
affect the RSSI values. One common issue which is faced is the RSSI values are
negative values which converge toward 0 as the distance between the AP and the
receiver decreases. Also, during fingerprinting, if the AP is far away or signal strength
is reduced due to obstacle between the receiver and AP it is possible that the receiver
saves 0 for such values. This causes problem if no preprocessing is done. Ideally,
even if the Euclidean distance between the AP and the receiver is 0 the RSSI never
becomes 0 it is always negative. Thus, to clean the dataset, zeros in the dataset are
replaced by 100. This value is arbitrary and signifies that no signal was received at
that location from a particular AP [15]. These steps have to be completed before
proceeding to feed the neural network with the RSSI values from all the AP’s.

3.3 Regression Using Deep Neural Network

The main idea is to implement regression on the data and predict the actual position
of the user with respect to the grid at given point of time. The received RSSI values
at any location are used to predict the x, y, and z coordinates based on the grid in
real time. After the data is preprocessed, the RSSI values from the access points are
feeded to the neural network as shown in Fig. 2. The neural network shown gives
the coordinate value as the output and thus there are three similar networks for all

Fig. 2 Neural network used in the model

692 V. Asher et al.

the axes. All the axes are trained separately with the RSSI values and thus it is the
network with 1 input layer, 5 hidden layers, and 1 output layer for each axis. The
deep neural network used for regression has variable number of input layers. This
number is dependent on number of access points used to design the system (This
varies from system to system based on requirement). The output layer consists of
the 1 output neuron with linear function. This structure is same for all the axes. Each
hidden layer is a dense layer and thus all the neurons in the layer get the input from
all the previous layers. There are a total of 5 hidden layers with 32 and 64 neurons
alternatively. The final hidden layer has 70 neurons. The output layer has a linear
activation (as regression is done) and a single neuron to predict the value of the
coordinates.
All the hidden dense layers have the advanced leaky rectifier linear unit
(‘LeakyReLU’) as the activation function which helps in mapping between the input
and the output variables.

3.3.1 Mathematical Analysis Deep Neural Network

Activation function—LeakyReLU or leaky rectified linear unit is used for the sys-
tem. It is the advanced version of the rectifier linear unit. Both these functions are
explained in detail later in the section. The primary purpose of the activation function
is to add nonlinearity to the model and learn from previous results. The process of
backpropagation [15] is possible because activation functions update the weight and
the bias depending on the output of the previous losses.
Mathematically, rectified linear unit (Fig. 3) is given by—

F(a) = max(0, a) (1)

Fig. 3 Rectifier linear unit

(RELU)
Indoor Object Tracking Using Wi-Fi Access Points 693

Fig. 4 Leaky rectifier linear

unit (LeakyReLU)


0, if a < 0.
F(a) = (2)
a, otherwise.

Range:[0 to ∞)

F(a) is the RELU Function.

Along with the simplicity, there is also a drawback of this function—it may
sometime render a neuron dead. There could be a certain weight update that could
possibly cause the neuron to deactivate forever and thus a better version of RELU,
that is, the LeakyReLU is used in the system. Figure 4 represents the LeakyReLU.
Mathematically, Leaky rectified linear unit is given by—

F(a) = max(δa, a) (3)


δa, if a < 0.
F(a) = (4)
a, otherwise.

F(a) is the LeakyReLU function

Delta (δ) here is a small constant ( 0.001). This small value increases the range
for rectified linear unit to
Range : (−∞ to ∞).

LeakyReLU has proved to be much reliable in recent times. Also, RELU or its
variants are used because they generally converge faster [15].
Optimization function—This function generally help to minimize or maximize
the error function which is a function that depends on the models parameter use to fit
the training data to receive the line of best fit (considering the regression problem in
this system). The proposed system uses the ‘RMSprop’ optimizer, this is very similar
694 V. Asher et al.

to the gradient descent with momentum but the calculation of the gradient varies in
the two approaches. The RMSprop reduces the vertical fluctuations, which helps us
to reach the global minima faster.
The updates done by RMSprop for each parameter are done by the following
equations.
vt = λvt−1 + (1 + λ) ∗ G 2t (5)

γ
Δwt = − √ ∗ Gt (6)
vt + 

wt+1 = wt + Δwt (7)

λ is hyperparameter generally set to 0.9, It has to be tuned as needed.

γ is the initial learning rate.
wt is the weights at time t.
G t is gradient at time t.
vt is exponential average of square of gradients. The epsilon () is used so that
the equation is never accidentally divided by zero. Also, as seen RMSprop indirectly
works like simulated annealing.
The loss or error function—In case of regression the error function generally used
is ‘MSE’ which stands for mean squared error. A quality of a machine learning model
is determined by the value returned by the loss function. If the value is high means
there is a lot of deviation in the predicted value when compared to actual value. There
are many types of loss functions which can be used. The proposed system uses mean
squared error (MSE) and mean absolute error (MAE).
As the name suggests the mean squared error is the mean of squares of difference
between actual value and predicted value.
The mean absolute error is the mean of absolute values of the difference between
the predicted values and actual values.

4 Results

The regression algorithm used in system for predicting the position and metrics like
MSE and MAE were used to check the validity of the model the results of which are
tabulated in Tables 1 and 2.

Table 1 Mean square error – Validation loss Training loss

X-coordinate 1.3 m 0.82 m
Y-coordinate 1.8 m 1.85 m
Indoor Object Tracking Using Wi-Fi Access Points 695

Table 2 Mean absolute error – Validation loss Training loss

X-coordinate 0.96 m 0.67 m
Y-coordinate 1m 0.922 m

Fig. 5 Floor plan

These results are based on readings taken in Sardar Patel Institute of Technology,
Mumbai. The complete details of the experiment are further explained.
The RSSI values were collected by using an android device with a personalized
application which had the features of scanning for access point and storing the basic
service set identifier (BSSID) for getting the RSSI values from it at the different
locations. The experiment was done with a total of five access points. One access
point was considered as the origin and all the values for the position of the user
were taken with respect to the origin. The floor was divided into the grid of 0.5 m x
0.5 m squares and fingerprints from these five access points were recorded at several
different locations within the area of concern. Figure 5 illustrates the example of the
floor plan (Area of concern).
The data was then collected manually by going to different coordinates of the grid
(yellow dots in Fig. 5) and take readings for RSSI values at those points. The final
dataset generated had a total of eight columns—X, Y, Z*, AP1, AP2, AP3 and about
400 tuples. The number of columns is generally variable as it is based on the number
of access points that are configured to determine the location of the user. Here, five
APs were used to find the position, this number is arbitrary and may vary depending
on the size of the area to be covered and accuracy expected for determining the
position of the device in space.
Initially, dataset contained many null RSSI values due to weak signal strength and
also due to the incapacity of the receiver to continuously scan for the access points.
Both these issues were solved to some extent by the idea of automatically taking a
number of readings at the same location and preprocessing the dataset as explained
previously in 3.2. Figure 6 shows the structure of the dataset. In the final step of the
preprocessing stage, the data is split into train and test set in the ratio of 8.5:1.
696 V. Asher et al.

Fig. 6 Dataset

Fig. 7 Loss function

(x-coordinate)

*Note—z was kept constant throughout the experiment as it signifies the floor
number and experiment was performed on a single floor. And thus, all the values of
column z are 0. The model was not trained for z-coordinate in the experiment.
After the split, the model was compiled and fitted to get the best curve to predict
the location. In the training process, hyperparameter tuning resulted in different
parameters for training the neural network for x- and y-coordinate. The training of
x-coordinate was done with 1000 epochs and batch size of 21 to get the best loss
value for the testing and the training dataset. Figure 7 shows the error reaching to 0
for both test and train set at 1000 epoch.
For the y-coordinates, the parameter is different only 195 epochs with batch size
18 were needed to reach the least error value. Figure 8 shows the error function.
The testing data along with the predicted data is plotted in Fig. 9. The orange dots
are the points where data was collected. Green points are the testing data and the
blue points are the points predicted by the model.
The MSE and MAE which were calculated using this network are being used as
evaluation metrics. The validation loss is the equivalent loss obtained on the test set
when the best loss value is calculated on the train set. Whereas, loss on the training
Indoor Object Tracking Using Wi-Fi Access Points 697

Fig. 8 Loss function

(y-coordinate)

Fig. 9 Actual versus

predicted location

set is also mentioned in the table these are the losses found when the best weights
are found for the network.

5 Conclusion

As described in the table above, this model can be used to predict the location. The co-
relation between the number of AP, their position, and accuracy is still uncertain and
can be found in the future experiments. The system can be modified for various use
cases blind navigation, crowd flow management, warehouse management, etc. The
system further will be updated to have a grid-based agent trained by reinforcement
learning.
698 V. Asher et al.

References

1. Ni L, Liu Y, Lau YC, Patil A (2003) LANDMARC: indoor location sensing using active
RFID. In: Proceedings of the first IEEE international conference on pervasive computing and
communications. (PerCom 2003). IEEE Computer Society
2. Faragher R, Harle R (2015) Location fingerprinting with bluetooth low energy beacons. IEEE
J Sel Areas Commun 33(11):2418–2428
3. Noertjahyana A, Wijayanto IA, Andjarwirawan J (2017) Development of mobile indoor posi-
tioning system application using android and bluetooth low energy with trilateration method.
In: 2017 international conference on soft computing, intelligent system and information tech-
nology (ICSIIT). IEEE
4. Kumar S, Gil S, Katabi D, Rus D (2014) Accurate indoor localization with zero start-up cost. In:
Proceedings of the 20th annual international conference on mobile computing and networking
- MobiCom ’14. ACM Press
5. Vasisht D, Kumar S, Katabi D (2016) Decimeter-level localization with a single wifi access
point. In: 13th USENIX symposium on networked systems design and implementation (NSDI
16). USENIX Association, Santa Clara, CA, pp 165–178
6. Xiong J, Jamieson K (2013) Arraytrack: a fine-grained indoor location system. In: Presented
as part of the 10th USENIX symposium on networked systems design and implementation
(NSDI 13). USENIX, Lombard, IL, pp 71–84
7. Husen MN, Lee S (2016) Design guideline of wi-fi fingerprinting in indoor localization using
invariant received signal strength. In: 2016 international conference on information and com-
munication technology (ICICTM). IEEE
8. Kaemarungsi K, Krishnamurthy P (2004) Properties of indoor received signal strength for
WLAN location fingerprinting. In: The first annual international conference on mobile and
ubiquitous systems: networking and services. MOBIQUITOUS 2004. IEEE
9. Kaemarungsi K, Krishnamurthy P (2004) Modeling of indoor positioning systems based on
location fingerprinting. In: IEEE INFOCOM 2004. IEEE
10. Zafari F, Gkelias A, Leung KK (2017) A survey of indoor localization systems and technologies.
CoRR
11. Liang X, Gou X, Liu Y (2012) Fingerprint-based location positioning using improved knn.
In: 2012 3rd IEEE international conference on network infrastructure and digital content, pp
57–61
12. Asmaa L, Hatim KA, Abdelaaziz M (2014) Localization algorithms research in wireless sensor
network based on multilateration and trilateration techniques. In: 2014 third IEEE international
colloquium in information science and technology (CIST), pp 415–419
13. Hwang R, Hsu P, Cheng J, Chen C, Chang C, Huang H (2011) The indoor positioning tech-
nique based on neural networks. In: 2011 IEEE international conference on signal processing,
communications and computing (ICSPCC), pp 1–4 (2011)
14. Tuncer S, Tuncer T (2015) Indoor localization with bluetooth technology using artificial neural
networks. In: 2015 IEEE 19th international conference on intelligent engineering systems
(INES), pp 213–217
15. Torres-Sospedra J, Mendoza-Silva GM, Montoliu R, Belmonte O, Benitez F, Huerta J (2016)
Ensembles of indoor positioning systems based on fingerprinting: simplifying parameter selec-
tion and obtaining robust systems. In: 2016 international conference on indoor positioning and
indoor navigation (IPIN). IEEE
Crime Analysis and Prediction Using
Graph Mining

A. G. Sreejith , Alan Lansy , K. S. Ananth Krishna , V. J. Haran

and M. Rakhee

Abstract Crime investigation and counteractive action is a deliberate methodology

for distinguishing and examining examples and patterns in crime. Our framework
can foresee regions which have a high likelihood for crime event and can predict
crime-prone regions. With the expanding approach of mechanized frameworks, crime
information investigators can help the law authorization officers to accelerate the
way toward identifying violations. Utilizing the idea of information mining, we can
extract beforehand, uncertain valuable data from unstructured information. Crimes
are a social aggravation and cost our general public beyond all doubt in a few different
ways. Any study that can help in explaining crime quicker will pay for itself. About 10
percent of the criminals carry out about half of the violations. Here we utilize graph
mining techniques for gathering information to distinguish the crime instances and
accelerate the way toward enlightening crime. Graph mining is done with the help
of identifying the structure of the graph to obtain frequent patterns of information.
With the help of graph database, we could store the past criminal records and infer
important information from it. Our project aims to store the data in a graph database
and try to determine the important patterns on the graph which can be used to predict
the regions which have a high probability of crime occurrence and can help the law
enforcement officers to enhance the speed of the process of solving crimes.

Keywords Graph mining · Crime analysis · Graph database · Neo4j

1 Introduction

Graph mining is employed to find helpful patterns from a piece of graph-structured

information. The graph contains a set of nodes interconnected using edges which
might be used to show the relation between the information and entities. Graphical
illustration of any information provides an additional convenient and effective way

A. G. Sreejith · A. Lansy · K. S. A. Krishna · V. J. Haran (B) · M. Rakhee

Muthoot Institute of Technology and Science, Kochi, India
e-mail: haranvj1997@[Link]
URL: [Link]

© Springer Nature Singapore Pte Ltd. 2020 699

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
700 A. G. Sreejith et al.

to analyze the relationship between the data and entities using graph mining. We
extract patterns from the graph through which we can analyze the contents and get
useful data from it. For performing the graph mining, we first need to create a graph
database of existing records. For creating the graph database, we can use many graph
database platforms. We are using Neo4j as it is easy to use and provide drivers for
many languages and have very good cypher query language. After creating the graph
database, we need to extract the data from it using the API of the database to perform
the analysis. Here, we are using R programming language for graph analysis as it is
simple and easy to use and have many predefined packages for graph analysis like
igraph in which we can find out frequent patterns and perform different centrality
measurements for finding frequent crimes, crime-prone areas, and major people
influencing in the crimes.

2 Crime Analysis and Graph Mining

Graph mining is the process of finding out frequent subgraphs present in the input
graphs and finding useful information from it. Graphs are easy to understand, and
mining using graphs is more efficient and faster than normal data normal statistical
mining methods. Analysis is done by finding the frequent patterns and by analyzing
different centrality measurements. Crime data is best suited to be represented as
a graph as it contains different relationships among the data entities like people,
incidents, vehicles, etc., and relations are there in the data which forms a complex
graph patterns when represented as a graph, and we can perform analysis on it easily
for getting the useful data out of it.

2.1 Advantages of Using Graph Mining

The techniques based on graph mining are effective and efficient in comparison with
the other statistical data applications. It has a wide range of applications in the regions
of law requirement, e-advertising, health and medical industry, educational training,
bio-informatics, agriculture, and so on.

2.2 Disadvantages of Using Graph Mining

With greater advantages of graph mining, it also have some disadvantages like as the
data grows, the mining algorithms should be as efficient as to quickly process the
data, graph mining is in its early stages and the algorithms developed needs more
refinement for it to be used with large data, and with datasets with very less number
of relationships, graph mining is not very efficient.
Crime Analysis and Prediction Using Graph Mining 701

3 Familiarization of Tools

3.1 Neo4j

Neo4j is a graph database management system developed by Neo4j, Inc. It is the

most popular graph database according to DB-Engines ranking and is the 22nd most
popular database overall. [12] In Neo4j, data is stored as nodes and relations between
the data as an edge. Each edge and node can be given as many numbers of attributes
as needed. Both edges and nodes can be labeled. These labels help to reduce the
number of searches. The graph data of Neo4j is stored as record files on the disk.
To increase the retrieval speed, it uses the two-layer mechanism of caching, namely
file system cache and object cache. Every node holds only their first reference to the
relationships. Transaction modules contain a transaction log and transaction manage-
ment which ensures that transactions are ACID. The system clustering capabilities
are characterized by a module called HA module [12].

3.2 R Programming Language

R is a programming language and environment used for statistical computing and

graphics. It provides a variety of statistical (linear and nonlinear modeling, classical
statistical tests, time-series analysis, classification, clustering) and graphical tech-
niques and is very much extensible. It is a free software. The RStudio IDE is used
for R [13].

4 Related Work

Crime data mining uses graph mining techniques for crime analysis. Crimes can
be subdivided into many types based on different principles. There are eight crime
categories like traffic violations, sex crimes, theft, fraud, arson, drug offenses, cyber
crimes, and violent crimes [10].
Various crime data mining techniques are available currently. Some of the
most commonly used techniques include entity extraction, clustering techniques,
association rule mining, and sequential pattern mining and classification [10].
Criminal analysis methods like hot-spot detection, crime comparison, and visu-
alization of crime patterns are important [10]. A time-series is drawn between the
crime frequency and the time in crime pattern visualization, and some interesting
crime trends were identified from this. In addition to these steps, some other analysis
steps such as crime clock, outbreaks detection, and nearest police station detection
were also used [10].
702 A. G. Sreejith et al.

An intelligent crime detection system can be used to predict likely suspects for
a given crime. Five types of agents, namely message space agent, gateway agent,
prisoner agent, criminal agent, and evidence agent, were used for the same [10]. In
countries like England, the Police Department of Cambridge has performed a similar
one named Series Finder for determining the patterns in crimes [11].

4.1 Overview of Popular Graph Database

Talking about the graph databases in recent years, there has been a large number of the
high-performance graph database environment, such as Neo4j, Infinite, Graph, DEX,
InfoGrid, HyperGraphDB, and Trinity. Among all of them, Neo4j is the considered
mainstream of a Java-based open-source software. Its kernel is a very fast graphics
engine. It supports the recovery, provides two phases of the submission, and supports
for XA transactions and other database product features. Neo4j is a network-oriented
database, that is, an embedded, disk-based, fully transactional Java persistence engine
which stores data structured in networks rather than in tables [12].

5 Proposed Work

Graphs can be used to effectively analyze the data, and we intend to use the graph
database to store the criminal data and analyze it to check for different patterns in
the graph to show the regions having the high chance of occurrence of crime and
visualize the areas having a high crime rate. The first step would be to create a graph
database using the existing crime records and then analyze it to show the relevant
details. Our system could be used to easily visualize the entire database, identify
regions having a high crime rate, identify major criminals or gangs and also, the
crime patterns and similar crimes. Finally, we can generate a community graph of
criminals and identify their network.

5.1 Architecture Diagram

Figure 1 shows the system architecture, in which the user gives the crime data or
queries which then processed by the modules makes the graph database. A graph
database is used for further graph mining processes which then yield the necessary
output which is then processed by the output processing unit which produces the
output which the user needs, which can be shown as graphs or charts.
User inputs the data to the system as a CSV file of the records and can create
a database also users can give queries the system to get useful information, data
conversation system is used to convert the data to the form which is optimized for
Crime Analysis and Prediction Using Graph Mining 703

Fig. 1 System architecture

graph analysis. A graph database is used to store the data and to support the graph
mining module by supplying data through its API. Graph mining modules are used
to analyze the created graph database. Graph mining module will provide the inputs
to the output modules which will show the outputs to the user in the form of graphs,
charts, or the output to the user query.

5.2 Methodology

A graph database can be created using an open-source platform like Neo4j which is
very powerful and has its own query language called cipher to query graph database.
Figure 2 shows the data flow diagram of the system in which the input module accepts
the input in CSV format which should be processed by the input module to make it to
format which can be accepted by the Neo4j for creating the database. With the help
of the Neo4j API, we can directly load the data to create the graph database. Here
the query from input module is taken by query management unit which converts the
normal query to Neo4j’s cipher query, queries the graph database, and gets the pattern
graph which is then given to subgraph mining unit, which mines the subgraph with
pattern matching approach which shows the crimes with the given pattern from the
database. Here Neo4j platform is the main module which is used to generate, query,
and control the graph database. Clustering is done with the help of R programming
language. With the help of clustering, the connection between the crimes and criminal
communities can be identified. Centrality measurements consist of mainly two types,
namely node centrality and betweenness centrality, which will tell the major type
of crimes and who are the major criminals in the area. The result processing unit
processes the results and generates the result according to the user query, and output
is shown. Graph mining techniques like centrality measurements which include the
degree centrality measures can be used to find out the node with maximum direct
704 A. G. Sreejith et al.

Fig. 2 Data flow diagram

links. Betweenness centrality is the number of times a node occurs on the shortest
path between a pair of nodes. In closeness centrality measure, the center of attention
is on the closeness of a node in the graph to all the other nodes in the graph. These
centrality measurements can be used accordingly to find out the major criminals
or gangs. Clustering techniques can be used to cluster the data and show criminal
community graph and networks. Subgraph mining can be used to identify the pattern
of a crime and find out similar crimes in the database.

6 Conclusion

The number of crimes is increasing day by day, and at the moment, we do not have
a mechanism to predict the most possible crimes happening in an area. Each crime
has a pattern. Mining patterns from the graph show the foremost possible crime in
a region. The techniques used for the processing as well as the prediction of data
can improve the performance, speed, and accuracy in the crime prediction process.
These procedures will successfully decide regularities in the crimes by breaking
down present and past criminal data and foresee future violations.
Crime Analysis and Prediction Using Graph Mining 705

7 Future Scope

To increase the quality of crime prediction system, one of the enhancements that
may be created is to develop the system as a mobile application. This might enable
the user to report the crime remotely as shortly as he observes one and eliminates
the requirement for a desktop system to report a crime. In future, system can enable
the user to register complaints online. Users will read the progress of their criticism
online. By the long-run technology, the user will read the case details and progress
of the complaints on their mobile phones. Techniques belonging to the informa-
tion mining and computing field might be incorporated into the system to build the
statement method even more practical and economical.

References

1. Kang H-W, Kang H-B (2017) Prediction of crime occurrence from multi-modal data using
deep learning
2. Shrivastava S, Pal SN (2009) Graph mining framework for finding and visualizing substructures
using graph database. In: 2009 IEEE international conference on advances in social network
analysis and mining
3. Tang D, Tan Y (2011) Graph-based bioinformatics mining research and application. In: 2011
IEEE fourth international symposium on knowledge acquisition and modeling
4. Sarvari H, Abozinadah E, Mbaziira A, McCoy George D (2014) Constructing and analyzing
criminal networks. 2014 IEEE security and privacy workshops
5. Bogahawatte K, Adikari S (2013) Intelligent criminal identification system. 2013 IEEE 8th
international conference on computer science education
6. Xu Y, Mingyang L, Ningning A, Xinchao Z (2012) Criminal detection based on social network
analysis. In: 2012 IEEE eighth international conference on semantics, knowledge and grids
7. Ho L-Y, W J-J, Liu P (2012) Distributed graph database for large-scale social computing. In:
2012 IEEE fifth international conference on cloud computing
8. Lu H, Hong Z, Shi M (2017) Analysis of film data based on Neo4j. In: IEEE/ACIS 16th
international conference on computer and information science (ICIS)
9. Jain R, Iyengar S, Arora A (2013) Overview of popular graph databases. 2013 IEEE
fourth international conference on computing, communications and networking technologies
(ICCCNT)
10. Jayaweera I, Sajeewa C, Liyanage S (2015) Crime analytics: analysis of crimes through
newspaper articles. Available: 2015 Moratuwa Engineering Research Conference (MERCon)
11. Sathyadevan S, Devan MS, Gangadharan SS (2014) Crime analysis and prediction using data
mining. Available: 2014 first international conference on networks soft computing (ICNSC
2014)
12. Huang H, Dong Z (2013) Research on architecture and query performance based on distributed
graph database Neo4j. Chongqing University of Posts and Telecommunications
13. R programming [Online]. Available [Link]
A Semi-supervised Approach to Detect
Malicious Nodes in OBS Network
Dataset Using Gaussian Mixture Model

Muhammad Kamrul Hossain Patwary and Md. Mokammel Haque

Abstract In this study, we have followed a semi-supervised approach for the classi-
fication of optical burst switching (OBS) network traffics generated by the network’s
edge nodes. We used expectation maximization (EM) technique for Gaussian mix-
ture model (GMM) to obtain a probabilistic classification of the OBS nodes. For
this purpose, we used a trustworthy OBS network dataset from UCI machine learn-
ing repository. Preprocessing and principal component analysis were applied to the
dataset for arranging the data so that GMM can play its role fairly. Only 1% (10
samples) of labeled data from OBS dataset was used to initialize the parameters of
GMM and the rest 99% was used for testing performance of the model. We found
a maximum accuracy of 69.7% on the test data using just 1% labeled data with the
tied covariance type of the constructed GMM. The significance of this result is that
it shows the GMM can be used in classification of OBS networks and other similar
networks in semi-supervised way when one has very few labeled data and when
labeling a huge dataset is not feasible.

Keywords Gaussian mixture model (GMM) · Optical burst switching (OBS)

network · Clustering · Semi-supervised learning · Principal component analysis

1 Introduction

Optical burst switching (OBS) network [1] is a relatively new networking technology.
It combines the optical circuit switching and optical packet switching in one system
of data transportation. In summary, in an OBS network, a sending terminal sends a
control packet ahead of the actual data packet. The control packet travels through the
network and allocates necessary resources for the upcoming data. Once the process

M. K. H. Patwary (B) · Md. Mokammel Haque

Department of Computer Science and Engineering, Chittagong University of Engineering and
Technology, Raozan, Chittagong, Bangladesh
e-mail: muhammadkamrulhossain@[Link]
Md. Mokammel Haque
e-mail: mokammel@[Link]

© Springer Nature Singapore Pte Ltd. 2020 707

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
708 M. K. H. Patwary and Md. Mokammel Haque

is complete, the data packet follows the path reserved by the control packet and
reaches to the destination. This system of all optical WDM data transportation, that
has opened a new door of multi-gigabit bandwidth utilization is prone to various
attacks. One notable attack is burst header packet (BHP) flooding attack. It is a
kind of denial of service attack. It happens when an edge node of an OBS network
maliciously transmits numerous control packets without sending any associated data
packet (also called data burst). As a result, the control packets (also called burst
header packets) reserve significant amount of resources which causes a denial of
service for other nodes of the network. That is why it is important to detect those
malicious nodes who are sending BHP without any associated data burst. Detecting
a malicious node in a network by analyzing network traffic is a very rich field of
research. Machine learning has been used to classify the traffic of various networks
[2]. Since the concept of OBS network is relatively new and it is not widely deployed
infrastructure, very few notable works exist regarding classification of OBS network
nodes/traffic using machine learning approaches.
Machine learning procedures can be divided into three types: supervised, semi-
supervised and unsupervised. In supervised machine learning, a large set of labeled
data is used to train and predict the classes. In unsupervised machine learning,
unlabeled dataset is trained and distinct clusters/groups are discovered. In semi-
supervised machine learning, a relatively small amount of labeled data and a large
amount of unlabeled data is used to train and predict the classes/groups of data.
Clustering is a type of unsupervised machine learning. Among the available cluster-
ing algorithms, Gaussian mixture model [3] is one. In many cases, the distribution
of the data in a dataset cannot be represented using a single Gaussian distribution.
Rather, a mixture of several Gaussian can be used to represent such distribution.
Gaussian mixture model(GMM) is a probabilistic model that attempts to find a mix-
ture of finite number of Gaussian distributions that can model an input dataset. The
parameters of that mixture are unknown. GMM uses the expectation maximization
algorithm [4] for finding the model’s parameters. After learning from train data, it
can assign each test sample to the most probable Gaussian it belongs to. GMM has
various covariance type, for example, diagonal, tied, spherical, full covariance etc.
A Gaussian distribution’s covariance matrix and mean determines the length and
direction of the axes of its ellipsoid density contours. Four covariance types of mix-
ture model are illustrated in Fig. 1 using a 2D example. In these contour plots of

Fig. 1 Plots of the mixture of Gaussians using GMM for different covariance types [5]
A Semi-supervised Approach to Detect Malicious Nodes … 709

Fig. 1, two components (or clusters) are located at coordinate (0, 0) and (5, 6) with
different weights. The diagram shows different covariance types: full, diagonal, tied
and spherical. ‘Full’ means the components are free to adopt any shape and posi-
tion independently. ‘Tied’ means that components may have similar shape where the
shape is not predefined. The ‘Diagonal’ means that the axes of contour are along the
coordinate axes. ‘Spherical’ is like diagonal type but with circular contours.
The reason for choosing GMM for this work is that it is one of the fastest method
for learning mixture models. Besides, GMM maximizes only the likelihood, hence
chances are negligible that it will bias the means towards zero. Also it is less prone
to bias the clusters to have specific structures.
In this study, we used an OBS network dataset, available at UCI machine learn-
ing repository [6] for classification using GMM. This dataset was build by some
researchers using a simulation environment. Network expert assigned the class labels
for each node’s traffic. There are four different types of nodes in the dataset. The
‘Class’ attribute is the target attribute which holds four possible labels for the cor-
responding node, namely: Block, No-Block, NB-No-Block, and NB-Wait. We took
the dataset, performed necessary cleaning and feature selection task. Then we did
principal component analysis [7] to make the dataset fit for GMM. A semi-supervised
approach was taken by using very few labeled data to precompute mean and covari-
ance for the GMM. After that GMM was trained based on few training samples. Then
the model predicted the class label for test data. Finally we computed the accuracy
of the model for different GMM covariance types.

2 Related Work

Very few studies exist on the malicious behavior of optical burst switching (OBS)
network and fewer on the application of machine learning to classify OBS nodes.
In [8], authors endeavored to filter misbehaving burst header packets (BHP) at the
optical layer. They made use of optical codewords to verify whether a BHP came
from a legitimate source or not by comparing bit pattern which represents a sequence
of pulses that decide the authenticity of source nodes.
In [9], authors demonstrated a firewall to prevent BHP flooding attack. It made
use of offset time included in the BHP. The offset time is the time difference between
a BHP arrival and its associated data burst arrival. So the firewall detects malicious
behavior by comparing the specified delay in the BHP and actual delay of related
data burst.
In [10] authors developed a method to classify an OBS network traffic into three
distinct categories: Trusted, Suspicious and blocked, by analyzing various parameters
related to resource utilization by the nodes. The authors set some rules and thresholds
to decide between the classes.
In [11], authors discussed about data burst loss in OBS network. Loss can occur
due to contention which does not necessarily mean that there is a congestion in the
network. To differentiate between loss due to contention and due to actual congestion,
710 M. K. H. Patwary and Md. Mokammel Haque

the authors employed both supervised and unsupervised methods. The authors found
that the measure of observed losses derived from the number of bursts between
failures follow a Gaussian distribution which has different characteristics for different
types of losses.
In [12], authors used exactly the same dataset as we have done in this work.
The target attribute of the dataset has four probable labels for each row: Behaving-
No-Block, Misbehaving-Block, Misbehaving-No-Block and Misbehaving-Wait. The
authors followed supervised classification approach using decision tree algorithm to
predict four labels of data. The decision tree is an effective classifier which can pro-
duce simple if-then rules that classifies the data. In that study, some if-then rules were
generated using decision tree algorithm that can fit any new input to the appropriate
label. The model showed high accuracy when trained with sufficiently large labeled
dataset.
In [13], authors used K-means clustering algorithm to classify an OBS network
nodes using a semi-supervised training method. A tiny amount of labeled data from
an OBS dataset was chosen as training samples and then an initial clustering was
done with those samples to find the probable cluster centers. With these assumed
labels and cluster centers, the test dataset was clustered. The accuracy on the test
dataset was 41.61% and 53.72% with 1% and 2% training data respectively.
In this study we are using EM based GMM in a semi-supervised approach to
classify the OBS network dataset which, we believe, to the best of our knowledge,
is among the earliest to be proposed and implemented.

3 Proposed Work

Semi-supervised learning (SSL) [14] is a kind of machine learning method which is

useful when we have a large quantity of unlabeled data as compared to the labeled
data. Practically it is difficult to find labeled dataset for a particular problem because
labeling is usually done by a human hand. Going to an expert to label a large dataset
is difficult, costly and time consuming task. In such situation, SSL methods help to
label the unknown classes. Usually a SSL algorithm starts with the available labeled
data and adjusts its parameters using them. Then using sophisticated probabilistic or
other mathematical models it labels the unlabeled data. SSL models also make use
of the unlabeled data’s information to update its parameters for a better prediction.
In this work, after performing necessary dataset preprocessing, we took few sam-
ples (1%) from the OBS dataset as training set to initialize the parameters of GMM.
The reason for choosing 1% data for training is that, labeled data are hard to find. The
OBS dataset has 1075 records, so taking 1% yields just 10 samples. We condition
that those 10 sample will have members from all the four classes that we want to
find. It can be stated without exaggeration, that it is easier to find 10 labeled data, or
to get 10 sample data labeled than to label thousands of data.
A Semi-supervised Approach to Detect Malicious Nodes … 711

Fig. 2 Steps of SSL approach using GMM on OBS data

After the split, mean and covariance of the Gaussians to be discovered were
initialized using the sample data from the training set. Then the model was trained
and tested against four of GMM covariance types. The process is illustrated in Fig. 2.

3.1 Dataset Cleaning and Feature Selection

The chosen dataset has 22 attributes [6]. The names are as follows:
Node, Utilised_Bandwith_Rate, Packet_Drop_Rate, Full_Bandwidth, Aver-
age_Delay_Time_Per_Sec, Percentage_Of_Lost_Pcaket_Rate, Percent-
age_Of_Lost_Byte_Rate, Packet_Received__Rate, Amount_of_Used_Bandwidth,
Lost_Bandwidth, Packet_Size_Byte, Packet_Transmitted, Packet_Received,
Packet_lost, Transmitted_Byte, Received_Byte, 10-Run-AVG-Drop-Rate, 10-
Run-AVG-Bandwith-Use, 10-Run-Delay, Node_Status, Flood_Status, Class. The
last attribute, ‘Class’ is our target attribute. It has four distinct labels/classes. There
are some missing or empty values which need to be processed. The missing values,
or empty values were replaced by the mean of the corresponding attribute values.
Also there are some attributes that can be ignored, like ‘Packet_Size_Byte’ attribute
which has constant packet size value. Also the ‘Node’ attribute can be removed as it
symbolizes node ID. The ‘Node_Status’ attribute can also be removed as it is a target
attribute which we are not considering in this work. The OBS dataset has very irregu-
lar shape when plotted in a graph. This makes the task of clustering harder. Clustering
712 M. K. H. Patwary and Md. Mokammel Haque

Fig. 3 Scatter plot of three attributes against other three from OBS dataset

algorithms can find the clusters easily when they are well separated from each other
and have a regular pattern/shape. A scatter plot of a few pairs of attribute from the
OBS dataset is shown in Fig. 3. Four colors symbolizes four distinct classes. And in
Fig. 4, distribution of a few attributes of OBS dataset is plotted. We can see that the
clusters are not spherical/circular in shape. They are irregular and non-spherical. So
the clustering algorithms will find it difficult to do a clear grouping of the classes.

3.2 Normalization

We performed normalization [15] on the OBS dataset so that each individual sample
will have unit norm. Each row of the dataset was rescaled using L2 norm so that the
data points come closer and make a finer grouping of different labels.
A Semi-supervised Approach to Detect Malicious Nodes … 713

Fig. 4 Distribution plot of six attributes from OBS dataset

3.3 Principal Component Analysis

We performed principal component analysis (PCA) [7] to reduce the number of

attribute down to only two. After plotting the two dimensions in a scatter plot we
found that the four groups are more clearly separable than before. Figure 5 shows
the scatter plot and Fig. 6 illustrates their distribution.
In Fig. 5, the four different colors represent four class labels of the target attribute
‘Class’. We can see the outliers in each group. The green is the leftmost color in
the plot, which has some blue as outliers. Then the blue group which crosses its
boundary and falls into the red group as outlier. The rightmost group is yellow color
which touches the red group in the boundary.
714 M. K. H. Patwary and Md. Mokammel Haque

Fig. 5 Scatter plot of two components from PCA showing four groups of data in colors

Fig. 6 Histogram of two components given by PCA on OBS dataset

3.4 Selection of 1% Data from OBS Dataset

We worked with the OBS dataset which has a target attribute called ‘Class’ having
four likely sub-classes, i.e. No-Block, Block, NB-No-Block, and NB-Wait. Their
meaning is as follows: Misbehaving node and will be blocked (Block), Behaving
node and will not be blocked (No-Block), Misbehaving node but will not be blocked
instead will be in waiting state (NB-Wait), and Misbehaving node but will not be
A Semi-supervised Approach to Detect Malicious Nodes … 715

Table 1 Dataset split into train and test set

OBS data % Amount of labeled samples Amount of block, NB-no-block,
no-block, NB-wait
Total 100 1075 120, 500, 155, 300
Train 1 10 3, 3, 2, 2
Test 99 1065 117, 497, 153, 298

blocked (NB-No-Block). We separated 1% data randomly from OBS dataset where

samples from all four classes appears. As the selection is random, the number of
sample from each class varies in every experiment. We found that the result is stable
even with fluctuations in the sample frequency of each class. In Table 1, a selection
is shown in detail.

3.5 Gaussian Mixture Model

Gaussian mixture model is a probabilistic model that considers a distribution con-

sists more than one Gaussian. It calculates the joint probability for a data point to
determine the most probable cluster (i.e. component). GMM employs the expectation
maximization (EM) algorithm to maximize the likelihood function for a data point
to be a member of a cluster. EM works in two step. In first step, called E step, the
expectation or the responsibility of a data point for every component is calculated.
Then in second step, called M step, the maximization of likelihood function is done
by re-estimating the parameters using existing responsibilities.

Algorithm of EM for GMM

Step 1 Initialization step

Using the labeled training data, initialize the means μk , variances  k for each of four
components.

1 
n
μk = wik xi
Nk i=1

1 
n
Σk = wik (xi − μk )xi (xi − μk )T
Nk i=1

where,
μk means of k-th gaussian
 k variances of k-th gaussian
x data point for dataset X, i.e. (x p X)
716 M. K. H. Patwary and Md. Mokammel Haque

n total data points in dataset X

k mixture components
wik probability
n that point x i is generated by the k-th Gaussian
Nk w
i=1 ik i.e. the effective number of data points assigned to k-th Gaussian
Then initialize the mixing coefficients π , and evaluate the initial value of the log
likelihood L() which is given by:
 

n 
K
L(Θ) = ln πk F(xi |Θk )
i=1 k=1

where,
k (μk ,  k )
πk prior probability (weight) of k-th gaussian
F(x i |k ) probability distribution of observation x i , parameterized on 

Step 2 Expectation step

Evaluate weights:

πk F(xi |k )
wik =

K
π j F(xi | j )
j=1

Where, wik is the probability that point x i is generated by the k-th Gaussian

Step 3 Maximization step

Re-evaluate parameters:

1 
n
μnew
k = wik xi
Nk i=1

1     T
n
Σknew = wik xi − μnew
k xi xi − μnew
k
Nk i=1
Nk
πknew =
N
Then, evaluate L(new ) and stop if converged.
The algorithm is repeated for each of four different covariance types, i.e. full,
diagonal, tied and spherical.
A Semi-supervised Approach to Detect Malicious Nodes … 717

4 Result Analysis

We choose Python [16] programming language environment (version 3) for imple-

menting the classification model. We used EM based GMM implementation from
Scikit learn machine learning library [17]. After necessary cleaning and prepro-
cessing of the OBS dataset, the EM for GMM algorithm was implemented for four
covariance types. The accuracy of the model was tested against the testing data (99%).
It was found that each of four covariance types had different accuracy as shown in
Fig. 7a–b. The maximum performance on the test data was given by tied covariance,
that is 69.7% accuracy.
From the above figures we see that the model best performs with ‘tied’ covariance
type, i.e. gives 69.7% accuracy. This output is significant. It clearly shows how semi-
supervised approach helped gain a good accuracy with just 1% (10 samples) of data.
The results also showed that ‘tied’ covariance suits best to the distribution of the
OBS dataset. The accuracy of the model can be improved by increasing samples in
the training dataset.
It should be noted that, we found far better performance when we used 1% labeled
data for the initialization of GMM’s mean and covariance, and then fitted the GMM
on the rest 99% unlabeled data. This approach is valid because SSL allows using
any unlabeled data to gain a better insight. We tested the fitted GMM and found that
the performance was more stable than the previous one. So, we suggest to use this
approach.

Fig. 7 a Performance of GMM with tied (left) and spherical (right) covariance type on test data,
accuracy 69.7% and 24.6% respectively. b Performance of GMM with diagonal (left) and full (right)
covariance type on test data, accuracy 49% and 59.3% respectively
718 M. K. H. Patwary and Md. Mokammel Haque

5 Conclusion

In this work, we constructed a Gaussian mixture model (GMM) classifier to predict

the behavior of traffics of an Optical burst switching (OBS) network. For this we used
an OBS network dataset from UCI machine learning repository. The dataset consists
22 attributes and the target attribute has four distinct classes of traffic that indicates
the behavior of the corresponding node, i.e. the source of the traffic. We followed a
step-by-step procedure to clean and preprocess the dataset to make it suitable for the
GMM classifier. As the distribution of the data is very irregular and nor-spherical,
we performed some scaling and transformation on the dataset to fit the data for
our classifier. A semi-supervised approach was followed, that is, we considered
very few (1%) labeled data samples from the dataset for training and initializing the
classifier. The rest (99%) of the data was used to test the performance of the classifier.
Experiments showed that the built model displays very good performance (almost
70% accuracy) in predicting the class label of unlabeled data, even when trained with
tiny amount of training samples. The accuracy of the model can be increased further
by introducing new method of choosing samples for initializing the GMM because
GMM is very sensitive to that. Besides, if available unlabeled data can be used to
find some insights to the distribution of the Gaussian components, it can help further
improve the performance of the model.

Acknowledgements The dataset used in this work was collected from UCI Machine Learning
Repository, Irvine, CA: University of California. Originally by Adel Rajab, Department of Computer
Science and Engineering, University of South Carolina SC, USA. We used the python implemen-
tation of Ron Weiss (ronweiss@[Link]) and Gael Varoquaux for the graphical representation
of different covariance types.

References

1. Qiao C, Yoo M (1999) Optical burst switching (OBS)–a new paradigm for an optical networks.
J High Speed Netw 8(1):69–84
2. Wang M, Cui Y, Wang X, Xiao S, Jiang J (2018) Machine learning for networking: workflow,
advances and opportunities. IEEE Netw 32(2):92–99
3. Reynolds D (2015) Gaussian mixture models. Encycl Biometr 827–832
4. Moon TK (1996) The expectation-maximization algorithm. IEEE Sign Process Mag 13(6):47–
60
5. Plots of the actual mixtures. [Link]
6. BHP flooding attack on OBS Network Data Set. [Link]
Header+Packet+%28BHP%29+flooding+attack+on+Optical+Burst+Switching+%28OBS%
29+Network%23
7. Olive DJ (2017) Principal component analysis. In: Robust multivariate analysis. Springer,
Cham, pp 189–217
8. Sliti M, Boudriga N (2015) BHP flooding vulnerability and countermeasure. Photon Netw
Commun 29(2):198–213
9. Sliti M, Hamdi M, Boudriga N (2010) A novel optical firewall architecture for burst switched
networks. In: 12th international conference on transparent optical networks, pp. 1–5
A Semi-supervised Approach to Detect Malicious Nodes … 719

10. Rajab A, Huang CT, Al-Shargabi M, Cobb J (2016) Countering burst header packet flooding
attack in optical burst switching network. In: International conference on information security
practice and experience. Springer, Cham, pp 315–329
11. Jayaraj A, Venkatesh T, Murthy CSR (2008) Loss classification in optical burst switching
networks using machine learning techniques: improving the performance of tcp. IEEE J Sel
Areas Commun 26(6):45–54
12. Rajab A, Huang CT, Al-Shargabi M (2018) Decision tree rule learning approach to counter burst
header packet flooding attack in optical burst switching network. Opt Switch Netw 29:15–26
13. Patwary MKH, Haque MM (2019) A semi-supervised machine learning approach using
K-means algorithm to prevent BHP flooding attack in optical burst switching network.
Unpublished (2019)
14. Chapelle O, Scholkopf B, Zien A (2009) Semi-supervised learning. IEEE Trans Neural Netw
20(3):542
15. Perronnin F, Sánchez J, Mensink T (2010) Improving the fisher kernel for large-scale image
classification. In: European conference on computer vision. Springer, Berlin, Heidelberg, pp
143–156
16. Python programming language. [Link]
17. Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, Vanderplas J (2011)
Scikit-learn: machine learning in Python. J Mach Learn Res 12:2825–2830
Apache Spark Methods and Techniques
in Big Data—A Review

H. P. Sahana, M. S. Sanjana, N. Mohammed Muddasir and K. P. Vidyashree

Abstract Major online sites such as Amazon, eBay, and Yahoo are now adopting
Spark. Many organizations run Spark in thousands of nodes available in the clusters.
Spark is a “rapid cluster computing” and a broader data processing platform. It has
a thirsty and active open-source community. Spark core is the Apache Spark kernel.
We discuss in this paper the use and applications of Apache Spark, the mainstream
of popular organization. These organizations extract, collect event data from the
users’ daily use, and engage in real-time interactions with such data. As a result,
Apache Spark is a big data next-generation tool. It offers both batch and streaming
capabilities to process data more quickly.

Keywords Apache Spark · Big data · Data processing

1 Introduction

Apache Spark [1] has stolen the focus on iterative learning jobs for machines, inter-
active, and batch data analysis. It is based on Hadoop MapReduce and extends the
MapReduce model to more computing types, including interactive queries and stream
processing, efficiently. Description of a scheme not only measures water consump-
tion per second but also generates enough electricity (~4 W) to run on the Apache
Spark data aggregation network. Apache Kafka is a distributed LinkedIn publica-
tion subscribe message system. Both Spark and Kafka have a distributed process-
ing environment [2]. JA-BE-JA is a distributed k-way solving problem algorithm.
A balanced graph partitioning has many relevant applications, including biological
networks, parallel programming, and online analysis of social networks, and can
be used to reduce communication costs and balance workload. The abstraction of
the resilient distributed data set is the key concept behind the Spark calculations
[3]. Apache Spark is a novel solution for processing large-scale production data

H. P. Sahana (B) · M. S. Sanjana · N. Mohammed Muddasir · K. P. Vidyashree

Department of Information Science and Engineering, Vidyavardhaka College of Engineering,
Mysuru, Karnataka, India
e-mail: hpsahana98@[Link]

© Springer Nature Singapore Pte Ltd. 2020 721

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
722 H. P. Sahana et al.

to solve the disadvantages of Hadoop. Evolutionary undersampling was developed

with special attention to the class imbalance problem as an extension of evolutionary
prototype selection algorithms. Spark’s flexibility enabled the proposed model to be
implemented easily and fully automatically [4].
SAC is built on Spark to increase the performance using the computer in-memory
model. Spark itself provides a Web user for monitoring the resource use of the entire
cluster, the status of the task, and detailed information for each stage of the task,
which emphasizes the application profile and the time of execution [5].
Apache Spark is an open-source computer cluster with APIs in Scala, Java, Python,
and R. Spark can cache working data or intermediate data in memory to minimize data
latency so that Spark performs much better than Hadoop in iterative workload types
such as machine learning algorithms and interactive data mining [6]. Deep learning
offers solid models of learning for MBD analysis. Apache Spark is an open-source
cluster computing platform for scalable MapReduce. Spark program RDDS natively
supports fault-tolerant executions and recovers worker node program operations [7].
The Halvade framework is the solution to the scalability challenge of the GATK
pipeline. GATK is a commonly used tool: Intel Realignment, Haplotype Caller, and
Base Recalibration. The output of this tool set is a VCF (variant format) file with
sequenced DNA variations [8].
APRACK is designed to calculate some of the user-specified features, such as
those of the largest real part of the largest magnitude, with k own values. It gives
the calling program control by multiplying the matrix-vector request. TFOCS is a
state-of-the-art numeric solver, formally, a convex solver of the first order [9]. Big
data analysis converges with HPC in order to unleash big data’s full potential. The
RDMA shuffle engine is designed to bypass JVM to prevent the overhead of JAVA
communication sockets [1].
Apache Spark and Apache Storm are designed to perform high-speed heteroge-
neous traffic data analysis in real time [10]. Clustering algorithms on static GPS data
have been evaluated using the Apache Spark processing architecture. SPARQLGX
is a direct assessment. After preprocessing RDF data, it evaluates SPARQL queries
using Apache Spark [11]. It depends on a compiler of a conjunctive SPARQLX query
that generates Scala code executed by the Spark infrastructure. Spark enables stream
processing with large input data and handles only a small amount of flying data [12].
Clustering of k-means is a nonhierarchical approach of grouping items into different
groups [12].

2 Comparison of Different Methods in Apache Spark

Table 1 gives us the idea of how the different methods and techniques used in Apache
Spark help to increase the result in the big data for the organization.
Table 1 Comparison of different Apache Spark methods and techniques in big data
Reference No. Approach Method Parameter Result Advantage
1 Detailed design for Systematic performance Spark APIs, Intel HiBench RDMA-based Spark design is A combined version of
high-performance evolutionary on chameleon sort and Terasort implemented as a pluggable RDMA Spark+
RDMA-based Apache Spark SDSA comet and an in-house model “RDMA-HDFS” achieved the
on high-performance cluster with cutting-edge best performance with up to
networks InfiniBand technologies 82% improvement on SDSC
comet cluster
2 Description of a smart city, a Stand-alone remote Apache Domestic water pressure Generation of 4 W electric A future smart city will have
self-powered water Spark streaming as a levels, utility electric power power as UK domestic water innovative features as
monitoring solution that uses distributed alternative to the pressure levels high-frequency monitoring of
low-cost water turbines conventional hierarchical water flows and automated
smart grid leak detection
3 An adaptation to the Implementation of JA-BE-JA BSP abstraction, the second An experiment conducted is Spark version adopting the
JA-BE-JA algorithm to make over a modified version of version of Apache Spark, similar to the ones effectively relaxed consistency models
it efficiently computed by peers to comply with the BSP Spark version of BSP achievable using Spark runs significantly faster
BSP model
4 Provide a new big data Multiple parallel operations, Divide and conquer based on Spark used as parallelization It enabled the softening of
scheme based on new in-memory operations MapReduce paradigm by technology helped in the lack of density issue
emerging technology Apache provided by Apache Spark dividing data into multiple goodness of evolutionary presented in the extremely
Spark to tackle highly subsets undersampling model imbalanced problems
imbalance data
Apache Spark Methods and Techniques in Big Data—A Review

5 To answer if Apache is Typical seismic data Traditional HPC, big data Customized seismic data Performance analysis of
scalable to process seismic processing algorithms used analytics platform distribution in Spark, several typical seismic
data with its in-memory for the study of performance extraction of templates for processing algorithms
computation and data locality and productivity seismic data processing
features algorithms
6 TCP-H benchmark as JVM log such as GC and JIT, Java heaps, Java virtual Achieves 30–40% speedup Improve Spark core runtime,
optimization with perspective system utilization, hardware machine, operating systems on average and up to 5x faster to develop a more advanced
logs events from PMO, SMT than the naive configuration algorithm about JVM
technology
(continued)
723
Table 1 (continued)
724

Reference No. Approach Method Parameter Result Advantage

7 A context-aware activity MapReduce computing on MBD analytics, scalable Enables tuning of deep This approach is 63% faster
recognition application with a many Spark works, master learning framework over models with many hidden than Hadoop
real-world data-out deep models Apache Spark layers and millions of MapReduce-based solutions
containing millions of parameters on a cloud cluster
samples to validate the
framework
8 A framework used for Apache Spark, GATK, DNA Harvard, a Hadoop Reduction of execution by This approach is 63% faster
implementation of an analysis pipeline MapReduce solution, keeping data active in than Hadoop
in-memory distributed Churchill, an HPC memory between MapReduce MapReduce-based solutions
version of GATK cluster-based solutions steps
9 Description of the distributed Separating matrix operations Spectral and convex Provides a comprehensive set This has been commercially
and local matrix from vector operations for optimization problems benchmark on accessing supported by a slew of
computations available in single node usage, JVM hardware level optimizations companies that provide
Apache Spark for matrix computation from further services
JVM
10 Big data platform for Apache Flume is used for Virtualization, data Load balancing for all Spark Flume-integrated message
low-latency traffic flow ingesting real-time data to aggregation, and lightweight executors leads to uneven brokers such as active MQ or
analysis based on real-time Spark-streaming analysis protocols such as COAP and results Kafka for more advanced
and high-velocity data MQTT message routing
11 Implementation of a SPARQL queries on Spark executable code, RDF SPARQLGX outperforms SPARQLGX represents an
distributed RDF datastore distributed RDF (resource data, SDE several state-of-the-art interesting alternative in
based on SPARQLGX description framework) data Hadoop-reliant systems several scenarios
sets
12 Comparison of Hadoop A standard machine learning Performance analysis using Spark is a very strong Spark will be the de facto
MapReduce and Apache for algorithm for clustering the k-means algorithm contender and would bring a framework for a large number
analysis big data (k-means) change of use cases involving big
data processing
H. P. Sahana et al.
Apache Spark Methods and Techniques in Big Data—A Review 725

3 Conclusion

This paper outlines the main features and the importance of Apache Spark. Apache
Spark is a cluster computing platform that is designed to be fast and extends the popu-
lar MapReduce model to support more computing types, including interactive queries
and stream processing. It includes the full functionality of the successful hydrogen
generation scheme demonstration with a power of ~4 W. The experiment was car-
ried out on the BSP version of peers by using a distributed framework such as Spark.
Apache Spark offers multiple parallel operations that benefits in-memory opera-
tions. Profiling Spark applications include deep data partition analysis memory and
network usage. The in-memory function of Spark and the generation of immutable
objects significantly reduce GC overhead from 30 to 10% and by optimizing JVM
counts. The Spark-based learning framework for mobile big data analysis offers a
promising learning tool to add value from raw mobile big data.
Apache Spark uses the multi-node cluster efficiently, increasing the efficiency
and flexibility of the framework. By separating matrix operations from vector oper-
ations, a large number of traditional algorithms intended for single node use can be
distributed. The default IP Spark over Infiniband (IpoIB) achieves a performance
improvement of up to 79%. Spark streaming and Spark batch involve high per-
formance in experiments with the Spark executor, which helps to set RDF data
specifically. Spark helps to simplify the computer-intensive and challenging tasks of
processing high volumes of real-time or achieved data, both structured and unstruc-
tured, integrating relevant complex capabilities such as machine learning and graph
algorithms. Spark brings mass processing of big data.

Acknowledgements The authors express gratitude toward the assistance provided by Accendere
Knowledge Management Services Pvt. Ltd., in preparing the manuscripts. We also thank our mentors
and faculty members who guided us throughout the research and helped us in achieving the desired
results.

References

1. Lu X, Shankar D, Gugnani S, Panda DKDK (2016) High-performance design of Apache Spark

with RDMA and its benefits on various workloads. In: Proceedings of 2016 IEEE international
conference on big data, Big Data 2016, pp 253–262
2. Domoney WF, Ramli N, Alarefi S, Walker SD (2016) Smart city solutions to water man-
agement using self-powered, low-cost, water sensors and Apache Spark data aggregation. In:
Proceedings of 2015 IEEE international renewable and sustainable energy conference, IRSEC
2015
3. Carlini E, Dazzi P, Esposito A, Lulli A, Ricci L (2014) Balanced graph partitioning with Apache
Spark. In: Euro-Par 2014: parallel Processing workshop, pp 129–140
4. Triguero I, Galar M, Merino D, Maillo J, Bustince H, Herrera F (2016) Evolutionary under-
sampling for extremely imbalanced big data classification under Apache Spark. In: 2016 IEEE
congress on evolutionary computation, CEC 2016, pp 640–647
726 H. P. Sahana et al.

5. Yan Y, Huang L, Yi L (2015) Is Apache Spark scalable to seismic data analytics and compu-
tations? In: Proceedings of 2015 IEEE international conference on big data, IEEE Big Data
2015, pp 2036–2045 (2015)
6. Chiba T, Onodera T (2015) Workload characterization and optimization of TPC-H queries on
Apache Spark. IBM Research—Tokyo, Japan, pp 1–12 (2015)
7. Alsheikh MA, Niyato D, Lin S, Tan H-P, Han Z (2016) Mobile Big data analytics using deep
learning and Apache Spark. IEEE Netw 31:21–29
8. Mushtaq H, Al-Ars Z (2015) Cluster-based Apache Spark implementation of the GATK DNA
analysis pipeline. Proceedings of 2015 IEEE international conference on bioinformatics and
biomedicine, BIBM 2015, pp 1471–1477
9. Zadeh RB, Meng X, Staple A, Yavuz B, Pu L, Venkataraman S, Sparks E, Ulanov A, Zaharia
M (2016) Matrix computations and optimization in Apache Spark. In: KDD’ 16, pp 31–38
10. Maarala AI, Rautiainen M, Salmi M, Pirttikangas S, Riekki J (2015) Low latency analytics
for streaming traffic data with Apache Spark. In: Proceedings of 2015 IEEE international
conference on big data, IEEE Big Data 2015, pp 2855–2858
11. Graux D, Jachiet L, Genev P, Graux D, Jachiet L, Genev P, Graux D, Jachiet L, Genevès P,
Layaïda N (2016) SPARQLGX in action: efficient distributed evaluation of SPARQL with
Apache Spark. In: 15th international semantic web conference
12. Gopalani S, Arora R (2015) Comparing Apache Spark and Map Reduce with performance
analysis using K-means. Int J Comput Appl 113:8887
Artificial Intelligence Applications
in Defense

C. Sinchana, K. Sinchana, Shahid Afridi and M. R. Pooja

Abstract Artificial intelligence (AI) is a rapidly growing field of computer science,

which involves computer programming and offers tremendous advantages in the
military expert system, human intelligence development, and support. This paper
describes the use of artificial intelligence in the military field. It shows how the
natural language of processing, ontology, and a system based on knowledge is used
to create a unified military system. It also shares an insight into the use of AI in the
military field for autonomous weapons, land analysis, and aircraft carrier landing.

Keywords Artificial intelligence (AI) · Military · Ontology · Knowledge-based

system · Autonomous weapons · Terrain analysis · Aircraft carrier landing

1 Introduction

Artificial intelligence is part of informatics that is booming wildly in the field of the
military system. Advances in artificial intelligence (AI), i.e., the natural language of
processing, ontology, knowledge-based system, land analysis, and improved aircraft
carrier landing performance enable new military capabilities that have a significant
impact on military strategies. In this paper, artificial intelligence mainly talks about
intelligence, surveillance, recognition of the balance of offense/defense, and, most
importantly, about the autonomous arms system. Autonomous weapons are those
that have the ability to make their own decisions based on programming constraints,
and artificial intelligence incorporates these weapons.
Many military programs, including systems that help combat management in real
time, predict and defeat enemy war plans, plan air strikes, fly hazardous pursuits,
eliminate operations, and implement AI systems. Military engagement, promises
on private industry resources, creates discussion platforms, publishes newsletters

C. Sinchana (B) · K. Sinchana · S. Afridi · M. R. Pooja

Department of Computer Science and Engineering, Vidyavardhaka College of Engineering,
Mysuru, Karnataka, India
e-mail: sinchanachandan24@[Link]

© Springer Nature Singapore Pte Ltd. 2020 727

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
728 C. Sinchana et al.

dedicated to military AI systems, thus creates online databases for AI information,

and creates a multi-command committee.

2 Ease of Use

2.1 Ontology

Ontology can be defined as the data model that represents knowledge and the rela-
tionship between these concepts as assets of concepts within a domain [1–4]. Today,
people have access to more data than people have had in the last decade in a single
day. It is therefore important to represent the relationship between the data, as the
data are available in many ways. The two standards governing ontology construction
are:
(1) web ontology language
(2) resource description framework.
Ontology components are classes and relationships. The main benefit of ontology
is that new relationships can easily be added to the existing model. Ontology is now
used in many fields, such as knowledge representation, knowledge sharing, integra-
tion of knowledge, reuse of knowledge, and information retrieval. With ontology, it
is possible to accurately represent a certain military domain, for example, operations,
intelligence, logistics, etc.
In certain military domains, we can identify and define entities and events. Rela-
tionships between entities and events can be identified and defined. This improves the
ability to reason over a particular domain. It contributes to the development of tactics,
techniques, and processes. All of these improve efficiency. The ontological methods
used by the war planners lead to situational awareness, understanding of the situation,
and a common operational picture. These are all necessary for decision-making. The
ontology is having good scalability and can be next enriched and improved.

2.2 Knowledge-Based System and AI Supportability

Artificial intelligence is a rapidly growing military field that offers tremendous advan-
tages in the development and support of weapons systems. The study area generally
accepted for the artificial intelligence field includes problem-solving, logical reason-
ing, language, programming, learning, robotics, and vision, system, and language
[5].
The knowledge-based system is a computer program that uses methods of deter-
mination and knowledge to solve problems that require important human expertise
to solve them. Knowledge based on the knowledge of the system comprises truths,
heuristics, and patterns. The air force can provide each technician with the expertise
Artificial Intelligence Applications in Defense 729

Fig. 1 Structure of a knowledge-based system [6]

of the best air force maintainers through the use of knowledge-based systems. It
helps to avoid delays in the need for expertise, retains knowledge of decisions and
actions taken in times of crisis, and ultimately helps to retain corporate knowledge
by trapping and coding knowledge.
By making the knowledge base more accessible, greater flexibility should be
achieved, uniform knowledge should be used to minimize the number of mechanisms
required to handle knowledge leading to a more transparent system, and these are
the ways in which we can offer greater support through the use of AI.
Figure 1 is a diagram of the two-part knowledge system, the inference engine,
and the knowledge base.

2.3 Autonomous Weapon

Autonomous weapons are able to make their own decisions on the basis of pro-
gramming limitations. These arms contain “AI” This includes a drone or a UAV
(unmanned aerial vehicle) usually refers to a pilotless aircraft operating through a
combination of technologies, including computer vision, AI, object prevention tech-
nology, and others. However, drones can also be autonomous ground or sea vehicles.
Autonomous and automated guns’ difference: Automatic weapons are those that do
not have the ability to learn and cannot change their objectives, and they can be
referred to as anti-material weapons, because they could cause civil damage. In this
type of weapon, the output is always the same for each input. There is no ability
to reason in automated weapons. However, autonomous weapons have the ability to
determine the best results possible for a set of inputs. It can modify its objective or
deploy sub-goals. Once activated, this weapon does not require further intervention
730 C. Sinchana et al.

of the operator. It can have many capabilities, such as target identification, tracking,
and firing, all of which have different levels of human interaction and input.
Working:
These weapons use technologies such as light detection and ranging (LIDAR). Radars
and other devices are used to create a map or “world” around the gun using light
pulses. Once the device is complete, it can navigate independently. In practice, these
weapons can encounter many obstacles. The programmer or an engineer cannot write
code for each obstacle to rely on “machine learning.”
Knowledge based on Reasoning:
Logical action based on knowledge occurs if the rule of the established rules does
not require the current situation. The skill-based tasks are easier to automate, for
example, in self-driven car navigation, the goal is the destination, and the best route
can be determined by applying traffic rules and vehicle dynamics. But the uncertainty
is much higher in the case of knowledge-based reasoning. It is possible to approxi-
mate human intelligence using the concepts of natural language processing, image
processing, machine learning, and profound learning, but there are many drawbacks
since machine learning is data-driven, and it will not be possible to recognize scenar-
ios or patterns that are slightly different from data. Autonomous military weapons:
Several military experts considered autonomous arms to be morally acceptable and
ethically better.
Lieutenant Colonel Douglas A. states that, on behalf of robots, moral benefits
may be to remove people from high-pressure combat zones. Prairie USR: It focuses
on neurological research, referring to the nerve circuits that cause self-control when
stressed [7]. There are currently no autonomous arms matching human intelligence.
However, AI will have a huge impact on the military in the future.

2.4 Terrain Analysis for the Development of the Mission Plan

Here, we see how the hybrid expert system analyzes the terrain, which develops mis-
sion plans that are part of the logical part of the computer-generated force automation
system. Figure 2 shows an analysis of the terrain. Land analysis plays an important
role in most military operations and applications. It is an integral part of preparing
the battlefield intelligence (IPB) [8]. Terrain analysts built many databases in the
early days for all important areas of operation. They form a basic basis for tactical
decisions, intelligence, and tactical operations. They also help to plan and carry out
most other battlefield functions. Since terrain features are constantly changing on
the surface of the earth, databases need to be constantly revised and updated. The
entity behaviors resulting from the above system are efficient enough to provide
an acceptable training value in a distributed interactive simulation for living human
opponents. It examines the problem of 3D representation of high-resolution terrain
in simulations and offers suggestions and advice for polygons to maintain a high
Artificial Intelligence Applications in Defense 731

Fig. 2 Terrain analysis [9]

degree of obedience in significant terrain in the military while reducing the number
of terrains used to represent no use.
The investigation of current developments in the automated target recognition is
carried out. This technology is relatively new in the field of tactical intelligence and
can be used in military simulations. Installation of knowledge-based systems (KBS)
is finally carried out for testing imitation material development systems. An ongoing
debate is being conducted on the virtual testing ground, which captures test expertise
and provides a test analysis function that helps the army fulfill the promise of using
simulation technology to facilitate the procurement process.

2.5 Multiple Artificial Intelligence Techniques Used

in an Aircraft Carrier Landing Decision Support Tool

An aircraft carrier is a warship that operates as a coastal airfield with a full aircraft
deck and aircraft transport, arming, deployment, and recovery facilities. Balloons
were used in the early twentieth century to deploy nuclear warships from wooden
732 C. Sinchana et al.

vessels carrying many fighters, helicopters, strike planes, and other aircraft. A heavier
aircraft such as fixed-wing gunships and bombers was launched by aircraft carriers
but cannot land at the moment. Aircraft carriers focus on modern action vessels with
their tactical and diplomatic power, their autonomy, their mobility, and their many
methods. This has shifted strategically to the main role of a fleet. One of the best
advantages of navigating international waters is that it does not interfere with regional
sovereignty and increases the power of airlines from third countries, reducing flight
time and distance of transport and thus increasing time. Availability in the zone of
action. Some of the recently completed projects that improve the decision of landing
signal officers (LSOs) under the guidance of aircraft landing in aircraft carriers have
been discussed here. Using multiple artificial intelligence (AI) techniques, auxiliary
solutions have been developed. The project developed pilot trends and also flight
prediction techniques and improved the user interface of LSO by applying cognitive
psychology decision-centered design methods.
Researchers have identified the parameters of important flight modes and devel-
oped the future of the neuro-fuzzy aircraft system [10–12]. Researchers used pilot
trend techniques and logical and obscure logic based on case studies. In addition
to many LSOs, they decided on the best performance options and had the correct
performance logic of the information provided by the pilot trending module, which
resulted in the design and implementation of the LOSO interface. Two specific areas
of the AI application, the data fusion part of the pilot trend system and the flight path,
are presented in Fig. 3.

Fig. 3 Landing signal

officers at work [9]
Artificial Intelligence Applications in Defense 733

Table 1 Comparative analysis of various fields in defense

AI in different fields of Definition/meaning Uses in defense
defense
1. Ontology The data model used for • To represent operations,
knowledge representation and intelligence, logistics, etc.
representation of • To define entities and
relationships between various events
concepts in a domain • To define tactics,
techniques, and processes
2. Knowledge-based system Computer program that uses • Used by air force to provide
and AI supportability methods of determination and expertise in air force
knowledge to solve problems maintainers for the
technicians
3. Autonomous weapons Weapon system which once • Various drones, advanced
activated can select and act robots, shooters, etc. are
on targets without the further designed and used in order
intervention of a human to act in specific situations
operator during a war
4. Terrain analysis Analysis of land, terrain • Forms the basis for any
before conducting any tactical decisions,
military operation operations, and intelligence
5. Aircraft carrier landing A warship which is provided • New pilot trends, flight
with full aircraft deck and prediction techniques, and
aircraft transport, arming improved user interface in
used during wars an aircraft carrier

3 Conclusion

In Table 1, we have shown the subject of AI and some of its applications in the military,
as it is very important for the improvement of defense technology performances. In
this paper, we see how ontology works and how AI has improved the performance of
land analysis and the use of multiple AI techniques in aircraft carrier landing. There
is a long way to apply AI in the military field.

Acknowledgements The authors express gratitude toward the assistance provided by Accendere
Knowledge Management Services Pvt. Ltd. in preparing the manuscripts. We also thank our mentors
and faculty members who guided us throughout the research and helped us in achieving desired
results.

References

1. Jia M, Yang B, Zheng D, Sun W, Liu L, Yang J (2009) Automatic ontology construction
approaches and its application on military intelligence. In: 2009 Asia-Pacific conference on
information processing, pp 348–351
734 C. Sinchana et al.

2. Jia M, Bing-Ru Y, De-Quan Z, Wei-Cong S (2009) Research on domain ontology construction

in military intelligence. In: 2009 third international symposium on intelligent information
technology Application, pp 116–119
3. Neches R, Fikes R, Finin T, Gruber T, Patil R, Senator T, Swartout WR (1991) Enabling
technology for knowledge sharing. AI Mag 12:36–56
4. Gómez Pérez A, Benjamins VR (1999) Overview of knowledge sharing and reuse compo-
nents: ontologies and problem-solving methods. In: Proceedings of the IJCAI-99 workshop on
ontologies and problem-solving methods (KRR5), pp 1–15, Stockholm, Sweden
5. Carlton KA (1988) Artificial intelligence supportability (Air Force application). IEEE Aerosp
Electron Syst Mag 3:25–32
6. Max Tegmark, Open letter on autonomous weapons—future of life institute. [Link]
org/open-letter-autonomous-weapons/
7. Cummings ML (2017) Artificial intelligence and the future of warfare. International Security
Department and US and the America’s Programme, pp 1–18
8. Campbell L, Lotmin A, DeRico MMG, Ray C (1997) The use of artificial intelligence in
military simulations. In: 1997 IEEE international conference on systems, man, and cybernetics.
Computational cybernetics and simulation, vol 3, pp 2607–2612
9. Richards RA (2002) Application of multiple artificial intelligence techniques for an aircraft
carrier landing decision support tool. In: 2002 IEEE world congress on computational intel-
ligence. 2002 IEEE international conference on fuzzy systems. FUZZ-IEEE’02. Proceedings
(Cat. No. 02CH37291), vol 1, pp 7–11
10. Zhihong D, Shiwei T, Dongqing Y (2001) An intra-algebra in ontology. J Comput Eng Appl
37:7–8
11. Pao Y-H (1989) Adaptive pattern recognition and neural networks. Addison-Wesley
12. Weiss NA (2012) Introductory statistics. Pearson Addison-Wesley
Hybrid Approach to Enhance Data
Security on Cloud

Pratishtha Saxena, Samarjeet Yadav and Neelam Dayal

Abstract Cloud computing is a label for the delivery of hosted services on the Inter-
net. The first challenge in the cloud computing is security, and the second challenge
is the large size of the file stored on the cloud. Today, cloud is in young age and used
too much for storing the data. So, security is a major concern. Security endures a
primary concern for businesses regarding cloud selection, usually public cloud selec-
tion. Public cloud service providers share their hardware infrastructure among the
numerous customers, as the public cloud has an environment of multi-tenant. Multi-
tenancy is an important feature of the cloud computing but also is prone to several
vulnerabilities. From the end-user aspect, cloud computing looks very insecure from
the perspective of privacy. In this paper, a technique REM (RSA algorithm, Elgamal
algorithm, MD5 technique) is proposed by which we can provide better security to
our data over the cloud. Unlike the previous techniques, which maintained only one
pillar of security, this paper maintains two pillars of security, i.e., confidentiality and
integrity along with reducing the size of the file.

Keywords Cloud computing · Data security · RSA algorithm · MD5

1 Introduction

Various definitions of cloud computing are available in which one of the squares is,
“a network solution for providing inexpensive, reliable, easy, and simple access to
computing resources [1].” Cloud computing relies on sharing of resources often over
the Internet to achieve consistency. Cloud computing is a service-based framework

P. Saxena (B) · S. Yadav · N. Dayal

Department of Computer Science & Engineering, Centre for Advanced Studies, AKTU,
Lucknow, India
e-mail: 17mcs010@[Link]
S. Yadav
e-mail: 17mcs11@[Link]
N. Dayal
e-mail: neelamdayal@[Link]
© Springer Nature Singapore Pte Ltd. 2020 735
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
736 P. Saxena et al.

which pursues at providing everything as a service. The cloud is a storehouse of

services and resources that are provided by cloud service providers. Based on the
user demand, cloud computing emphasizes on service model and a trade model based
on pay as per your use. The main concern in cloud computing is data security. The
purpose of security architecture is to maintain the system’s policy, confidentiality,
availability, and integrity. In the systematic way to understand the fundamentals of
cloud computing and securing the stored data over the cloud, a number of resources
have been considered. For the user, it is not possible to get control of his data and
computing applications until a strong security measure and privacy guarantee are
not in place. So, the user never gives priority to flexibility and economic availability
over its privacy and the security of his personal data. Nowadays, the users of the
cloud are in the incremental phase. Consumers of the cloud are growing enormously.
So, there is a need to increase the security level of cloud. Security of the data is
the main concern for the cloud storage. This paper proposes a technique to improve
confidentiality and integrity of user’s data on cloud by combining RSA, Elgamal, and
MD5 techniques. To improve the security of data, this paper proposed a technique
by which the confidentiality and the integrity of user’s data are improved.
The remaining sections of the paper are arranged as follows: Sect. 2 presents
related work which provides a detailed study on the essential data security. Section 3
gives a brief overview of cloud architecture which describes the cloud service and
deployment models. Section 4 represents the methodology of the proposed tech-
nique. Section 5 represents the results. Finally, Sect. 6 represents the future work and
concludes the paper.

2 Related Work

Somani et al. [2] presented the challenges of cloud security and also presented the
method to access cloud security. They presented the methodology for data security
in the cloud by the implementation of the digital signature with the RSA algorithm.
Subhashini and Kavitha [3] in 2010 presented a survey of different security issues
that have emerged due to the nature of service delivery models of cloud computing.
They present different security issues, threats to the cloud which is already presented
in the cloud, they discuss specific security threats of the different service delivery
models of cloud.
Albugmi et al. [1] in 2016 presented the data protection method. They proposed an
approach which is used throughout the world to ensure high data security by reducing
risk and threats. It discussed the threats to data in the cloud and their solution adopted
by several service providers’ protection for data.
Mishra et al. [4] in 2017 presented the security model for cloud computing.
Authors present some approaches to ensure data integrity and security of customers’
data which is stored on the cloud by using cryptography techniques. In this paper, they
show the economic impact of data loss and proposed the cryptographic algorithms.
Jayalekshmi M. B. and Krishnaveni S. H. [5] in 2018 discussed the different
techniques to secure the data on the cloud, and also, they presented different issues
of data storage on cloud.
Hybrid Approach to Enhance Data Security on Cloud 737

Negi et al. [6] in 2018 discussed the fully homomorphic encryption (FHE) scheme
in the optimized form in which the encryption algorithm is applied to improve the per-
formance of FHE schemes and Diffie–Hellman algorithms applied to secure channel
establishment in the fully homomorphic encryption.
Rohini and Sharma [7] in 2018 enrapt cloud computing security issues in their
paper. In this paper, for securing data over cloud, the author used the cryptographic
algorithm encryption of the data, and to maintain the integrity of the data, they use
hash code.
Gupta et al. [8] in 2018 analyzed existing encryption methods such as RSA,
KP-ABE, CP-ABE, and AED. The comparisons among them were on the basis
of computational cost and storage cost. The author also proposed an improvement
scheme to increase the speed of RSA algorithm using multi-threading concept on
the latest multi-core CPU.

3 Cloud Architecture

Cloud computing has three different services and deployment models in which cloud
service providers provide services and resources to its users from an assorted set of
models. Cloud computing elevates its users by enabling them to reduce infrastructure
cost and helps in small business to scale faster without sinking money in infrastructure
and services. Cloud architecture has different delivery models for the cloud service
providers and different deployment models for the users.

3.1 Delivery Model

The cloud delivery model delineates a discrete, pre-packaged combination of infor-

mation technology resources, which are presented by any cloud service provider.
There are three different cloud delivery models, which are as follows
• Platform-as-a-service (PaaS)
• Infrastructure-as-a-service (IaaS)
• Software-as-a-service (SaaS).
Infrastructure-as-a-Service (IaaS)
This is the lowermost layer of the cloud stack. This model provides various hardware
facilities to the consumers [9]. It provides an infrastructure of cloud-like storage
space, network and server capacity, etc., to the end user on a rental basis for increasing
organization capabilities [10]. The resources are easily scaled up depending on the
demand from the user and the service provider charge for the service on a pay per
use basis [11].
738 P. Saxena et al.

Platform-as-a-Service (PaaS)
In the cloud stack, architecture platform as a service layer is the middle and above
layer of IaaS. This model provides the development option to the customers. It is a
model in which the user can deploy own applications on a platform like a system
software, middleware, operating system, or database, etc., provided by cloud [11].
Software-as-a-Service (SaaS)
This is the uppermost layer of the cloud computing stack. At this layer, complete
software or applications are hosted that user or consumers can use. SaaS provides
the services to the end user. SaaS can provide business applications like clientele
relationship management, accounting, and corporation resource planning [12].

3.2 Deployment Model of Cloud Computing

The cloud deployment model delineates a particular environment of cloud which is

primarily differentiated by ownership, access, and size. Delivery model of the cloud
is narrated as below:
• Private cloud
• Public cloud
• Community cloud
• Hybrid cloud.
Public Cloud
A Public cloud is subordinate or dependent on the cloud service provider such as
Google, Amazon, Microsoft etc. In other words, we can say that its services are open
to all users. The user can use these resources on rent basis and can generally scale
their resource consumption up to their requirements. Rackspace, Google, Microsoft,
Amazon, and Salesforce are examples of the public cloud providers. Public clouds
are built on the norm of the standard model of cloud computing. In the standard
model, cloud service provider fabricates resources such as applications delivered,
data storage, and servers to the client by the Internet [4].
Private Cloud
Private clouds are built on the norm of standard cloud computing model. In this
deployment model of cloud, it offers resources such as applications, servers, and data
storage, but in this deployment method, these services are proprietary and mostly
customized for single user [4]. The aim of the public cloud regarding organization
is to deliver services to various corporations. Private cloud is mainly used for small
group of organizations. It has the same computing model like a public cloud in terms
that it provides resources to users, but the only difference is that public cloud is open
for all, which means many organizations can use public cloud, but the private cloud
is framed for a specific corporation.
Hybrid Approach to Enhance Data Security on Cloud 739

Table 1 Cloud service deployment model

Deployment Infrastructure Infrastructure Infrastructure Access and
model management ownership location consumption
Public cloud Third-party Third-party Off-premise Untrusted
provider provider
Private/community Corporation or Corporation or On-premise or Trusted
cloud a third-party a third-party off-premise
provider provider
Hybrid cloud Both Both Both Trusted and
corporation and corporation and on-premise and untrusted
third-party third-party off-premise
provider provider

Hybrid Cloud
The third deployment model is a hybrid cloud. It is a combination of two or more
clouds such as public and private cloud, public cloud and community cloud, and so
on. They work separately, but they are bounded together to provide services for a
common purpose [4].
Community Cloud
Community cloud is the last deployment model. It has an infrastructure service which
is shared by different corporations that serve a specific community. It has common
agitates like medical, military, etc. [4] (Table 1).

4 Methodology

In this paper, we proposed the methodology of REM technique. In this technique,

there are three modules as shown in Fig. 1. The first module describes the process
before sending the data over cloud. In this module, the input data (user data) is
encrypted by Elgamal cryptosystem. After that, the encrypted data is again encrypted
by the RSA algorithm for securing data and improving the confidentiality of data.
We calculate the hash of again encrypted data using MD5 hashing technique. In the
second module, the encrypted data and hash are stored on cloud. The third and last
modules of technique are to retrieve the data over cloud.
To provide data security over cloud, we are using REM (RSA algorithm, Elga-
mal algorithm, MD5 technique) technique. We are using the combination of these
technologies to design an algorithm in which the Elgamal algorithm will be used for
key sharing and encryption between the user and cloud server securely. It uses the
asymmetric key encryption technique for communication between the server and the
client. Elgamal is more secure because it is based on discrete logarithm. Security
740 P. Saxena et al.

Fig. 1 Methodology of REM technique

of Elgamal depends on the difficulty of computing discrete logs in a large prime

modulus. For enhancing data security, we are using double encryption technique.
Thereafter, RSA algorithm will be used for the encrypting the data second time.
RSA is also the asymmetric key encryption technique. Security of the RSA depends
on the difficulty of factoring large integers. Both the encryption techniques are used
to maintain the confidentiality of the data. Further, we will generate the hash value
of encrypted data by using the MD5 technique. A cryptographic algorithm MD5 is
used that takes an input and produces a 128-bit-long message digest. This message
digest is called “hash code” or “fingerprint” of the input. Lastly, the encrypted data
and hash code combined together are stored over the cloud storage. The flow diagram
shown in Fig. 2 represents the data flow diagram of REM technique.
Further, we discuss encryption and decryption processes of our techniques.

Algorithm 1: Encryption Algorithm

Step 1: User uploads the data to the cloud, and a unique data ID will be generated
in correspondence to every data item uploaded.
Step 2: Sharing key is generated by Elgamal algorithm.
Step 3: Data will be encrypted using asymmetric encryption algorithm, i.e., Elgamal
algorithm.
Step 4: The encrypted data is again encrypted using an asymmetric algorithm such
as the RSA encryption algorithm.
Step 5: The hash code of the encrypted data is generated using the MD5 technique
which is stored on the local user.
Step 6: The encrypted data and the hash code are stored on the cloud.
Hybrid Approach to Enhance Data Security on Cloud 741

Fig. 2 Data flow diagram of

REM technique

Algorithm 2: Decryption Algorithm

Step 1: Received data is stored in the cloud to begin decryption.

Step 2: Decrypt the data to ensure the confidentiality of the data using RSA.
Step 3: Generate the hash code of decrypted data to ensure the integrity of the data.
Step 4: Check the hash code of the data with the stored hash code on the local system.
Step 5: Decrypt again the encrypted data using the Elgamal algorithm.

5 Result

This technique improves the security of cloud storage. In this paper, we used different
parameters to analyze the performance of REM technique, which are as follows:
I. Execution time: Execution time refers to the total time taken for the execution
of algorithm.

Execution time = end of time − the start of time

742 P. Saxena et al.

Table 2 Execution time of

Data size (bytes) Execution time (s) Hash generation time
RSA and MD5 algorithm
(s)
1024 0.532 0.024
5120 2.109 0.041
10240 3.821 0.096

Table 3 Used space of

Data size (bytes) Space used (bytes) Hash space used
encrypted file and hash
(bytes)
1024 759 40
5120 738 40
10240 903 40

II. Space used: Space utilization refers to the size of the file used by the algorithm.

Space used = time ∗ buffer used per unit time

III. The probability of attacks: It is measured by how many numbers of

vulnerabilities are there in the algorithm.
The implementation of the algorithm is done, and it is tested for different size of
files as given in Tables 2 and 3. Table 2 represents the execution time of encrypting
the data files and the execution time of the hash code generation. We tested this on
different data size sets. Table 3 represents the space used. We test this on different
sizes of data, i.e., 1024 bytes, 5120 bytes, and 102400 bytes. This large file size space
is reduced which is represented in Table 3. The RSA encryption technique gives a
result with less execution time and takes very much less space size for storing the
data. As this result is checked on hashing technique MD5 which is taking very less
time in generating the hash of big data file and takes less space as given in Table 2.

6 Future Work and Conclusions

This paper proposed the methodology to improve the security of data over the cloud.
In cloud computing, security plays a very important role because it stores the user’s
confidential data and provides the confidentiality of data, which is our prime concern.
This paper proposed a hybrid approach to alleviate security challenges. We did this
work at the storage level and security authentication level. The implementation of
this approach will be done over the CloudSim simulator.
Hybrid Approach to Enhance Data Security on Cloud 743

References

1. Albugmi A, Alassafi MO, Walters R, Wills G (2016) Data security in cloud computing. In:
Fifth international conference on future communication technology, October 2017, pp 55–59
2. Somani U, Lakhani K, Mundra M (2010) Implementing digital signature with RSA encryp-
tion algorithm to enhance the data security of cloud in cloud computing. In: 1st international
conference on parallel, distributed and grid computing, pp 211–216
3. Subashini S, Kavitha V (2011) A survey on security issues in service delivery models of cloud
computing. J Netw Comput Appl 34(1):1–11
4. Mishra N, Sharma TK, Sharma V, Vimal V (2018) Secure framework for data security in cloud
computing. Adv Intell Syst Comput 583:61–71
5. Jayalekshmi MB, Krishnaven SH (2015) A study on data storage security issues in cloud
computing. Indian J Sci Technol 8(24):128–135
6. Negi A, Goyal A (2018) Optimizing fully homomorphic encryption algorithm using RSA and
Diffie-Hellman approach in cloud computing. Int J Comput Sci Eng 6(5):215–220
7. Rohini, Sharma T (2018) Proposed hybrid RSA algorithm for cloud computing. In: 2nd
international conference on inventive systems and control, ICISC, pp 60–64
8. Gupta P, Verma DK, Singh AK (2018) Improving RSA algorithm using multi-threading
model for outsourced data security in cloud storage. In: 8th international conference on cloud
computing, data science & engineering, pp 163–169
9. Kaur J, Garg S (2015) Survey paper on security in cloud. Int J Appl Stud Prod Manag 1(3):27–32
10. Gill RD, Kapur N, Singh H (2018) Increase security of data with respect to both confidentiality
and integrity over cloud. Int J Appl Eng Res 13(10):7388–7391
11. Bhadauria R et al (2014) A Survey on security issues in cloud computing. Acta Tehnica
Corviniensis—Bull Eng
12. Rong C, Nguyen ST, Gilje M (2012) Beyond lightning : a survey on security challenges in
cloud computing q. Comput Electr Eng
Security Threats of Embedded Systems
in IoT Environment

Sreeja Rajendran and R Mary Lourde

Abstract The Internet of things (IoT) technology has reached great heights with the
integration of efficient networking strategies, application protocols and VLSI. With
the unlimited number of devices connected to the IoT, security has become a major
concern. Mainly security is ensured by network and application protocols. But with
rising threats, there is a need to lay emphasis on security from a hardware design point
of view. Smart devices such as sensors and actuators are connected at the node of IoT
and work on an embedded processing platform. Hence, embedded processors become
an integral constituent of an IoT network. Selection of processors and identification
of critical modules in the processor architecture play a key role in developing a
secure design. Memory design will play a pivotal role in embedded system security.
Memory attack accounts for loss or alteration of information which can impair the
growth of IoT. This paper attempts to highlight security from a hardware perspective
along with the potential threats likely to affect the critical modules.

Keywords Hardware security · Denial of service · Processor architecture ·

Memory authentication · IoT

1 Introduction

The Internet of things opens up a new realm of technology which provides con-
nectivity to any device irrespective of its location. With advancements in VLSI and
network technology along with application requirements, IoT has been able to expand
its reach to intelligent control, intelligent transportation, precision agriculture, etc.
[1]. Evolution of micro-electro-mechanical systems, wireless communication and
digital electronics has led to the development of miniature devices called nodes [2].
These devices which can sense, compute and communicate wirelessly over short

S. Rajendran · R. Mary Lourde (B)

Department of EEE, BITS Pilani, Dubai Campus, Dubai, UAE
e-mail: marylr@[Link]
S. Rajendran
e-mail: sreejamanojnair@[Link]
© Springer Nature Singapore Pte Ltd. 2020 745
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
746 S. Rajendran and R. Mary Lourde

distances are interconnected to form wireless sensor networks. These sensor nodes
play a crucial role in IoT [3].
As interesting as it may sound, the implementation of IoT also brings along with
it many challenges. The most important among these is security. Through IoT, small
networks will be connected to larger ones which make it even harder to ensure
security. The security requirements of an IoT network can be broadly classified into
security tasks and security design metrics. Security tasks include authentication and
tracking, data and information integrity, etc. Design metrics include cost, size, energy
requirements, etc. which places constraints on security solutions [4].
The behavior of IoT depends on the data and information it collects. The more
the information shared and stored, greater will be the chances of security breaches.
Processing and storing of information on smart devices are handled by embedded
systems. Embedded system is the integration of hardware with software that is usu-
ally designed for a specific application. It consists of processor, memory, memory
controller, communication buses, etc. Embedded systems are an inevitable segment
of IoT since they are present in most of the devices connected at the nodes of the IoT.
Therefore, the potential growth and implementation of IoT rely on the development
of efficient embedded systems. An embedded system mainly consists of a proces-
sor on which the software runs. The performance of embedded systems is greatly
enhanced by on-chip features like data and instruction caches, programmable bus
interfaces and higher clock frequencies.
Security of a system is broadly classified into data security and physical security.
Data security can be ensured through cryptographic algorithms and also through
more robust design of hardware. Since an adversary can obtain the secure key and
the information through timing analysis, power analysis, etc. it is high time that
focus be laid on security through hardware design. When dealing with security from
a hardware perspective, emphasis needs to be devoted to the embedded system design.
There is an urgent need to bring about modifications in the design so that it becomes
tamper resistant. Physical security requires ensuring that only authorized users can
access the device and the services. Physical separation between entities becomes an
unresolvable issue due to resource sharing. The physical hardware components in
an IoT working device are illustrated in Fig. 1. The intentional change in hardware
inflicted by an adversary is generally referred to as a hardware Trojan. These hardware
Trojans can cause the circuit to fail under critical conditions. Hardware Trojan is
extremely difficult to detect since they are generally dormant and are designed to
get activated under very rare conditions. This paper tries to elaborate the potential
threats associated with the hardware connected to IoT.

2 IoT Security

Development of IoT would require strong security features at every level. There is
a need for information security, physical security and network security. The more
we increase the applications of the Internet of things, more will be the requirement
Security Threats of Embedded Systems in IoT Environment 747

Fig. 1 Components of an IoT device

for stringent security measures. The threats on an IoT network can be on software
or hardware. Hardware attacks generally involve hardware Trojans tampering the
processor. For any device linked to the IoT, data will be stored in memory, and
unused memory space is ideal site for malicious attack. For example, in an embedded
processor flash ROMs are in common usage. Flash memory is usually embedded on
the chip. It finds widespread use in electronic and computer devices since it is fast
and efficient as compared to EEPROM. The main drawback of flash memory is the
limit on the number of times data can be written on it. Flash ROMs are predominantly
used in embedded systems storing large amount of data in non-volatile memory like
digital cameras, cell phones, etc. These ROMs are rarely fully utilized. If an attacker
is able to place the Trojan on the unused memory space, it can have access to the
entire bus.
Network security is guaranteed through network protocols and standards. In spite
of having strong network security features, the network layer is still susceptible to
attacks like denial of service (DoS). This type of attack can arise due to lack of
sufficiently high bandwidth communication networks, which can result in data con-
gestion. Application layer is highly vulnerable to attacks due to its open-ended nature.
A major threat is poor security design of the basic function of an application. Also,
applications with no access control are targets for unauthorized use. Undesirable
network traffic is generally detected by intrusion detection systems (IDS). Network
security and applications security threats are kept at bay by communication, transfer
control and Internet protocols and intrusion detection systems. Security measures
can be taken to the next level by laying emphasis on hardware security. The first step
in designing for security involves the identification of the attacks which need to be
prevented. Design for hardware security should be done keeping in mind that any
additional circuitry added to prevent hardware attacks should itself be resistant to
attacks especially DoS. The architecture of embedded system is shown in Fig. 2.
748 S. Rajendran and R. Mary Lourde

Fig. 2 Architecture of embedded systems in IoT

3 Processor

The processor forms an integral part of any application linked to the IoT. It is very
much necessary to have a secure processor in an IoT environment. The security of
the processor is dependent on the architecture it is built upon. There has always been
a constant effort made by the designers to make the system completely secure. A
result of these efforts is shown in the different architectures that have been developed
in order to provide more security to the system. The design space available to the
designers provides a brief idea about the different parameters and features that need
to be addressed. The architectural design space of a processor is represented as shown
in Fig. 3.
The first row provides the different architecture that can be implemented such as
an ASIC or an embedded general-purpose processor with FPGA or an embedded
processor with graphic processing capabilities or a secure embedded processor. The
second row provides an idea about the different architectural parameters that need to

Fig. 3 Architectural design

space
Security Threats of Embedded Systems in IoT Environment 749

be considered such as the instruction cycle and word size. The third row articulates
security processing features that should be considered for design. The fourth row
involves selection of attack resistant features in the embedded system design [5].
Many designs have been implemented which try to focus on each of the parame-
ters and features suggested by this architectural design space. A modern-day embed-
ded application such as secure integrated cards commonly referred to as smart cards
employs a bi- processor architecture. In general, the first CPU is configured to perform
tasks that do not require utilization of sensitive information. A second CPU which
can be referred to as a master CPU is configured to perform tasks that involve manip-
ulation of sensitive information. Both CPUs communicate through a secure interface.
Though this architecture provides an enhanced overall security of the embedded sys-
tem, it does have some disadvantages. Primarily, the overall power consumption of
the system will be very high, and as a result, the cost of the application also increases
[6].
Internet of things requires the embedded systems to be miniaturized. As a result,
architectures employed need to be smaller, efficient and more secure. The Aegis
processor [7] provides an increased efficiency and reduction in power consumption
as it makes use of a single processor only. The Aegis processor architecture helps build
computing systems that are secure against both software and physical attacks with
minimum performance overhead for typical embedded applications. However, there
is a scope for improvement in this architecture. Firstly, the performance overhead
can be reduced for applications which do not require more memory which will help
in employing efficient encryption and system verification. Secondly, the architecture
is prone to physical attacks. Protecting the IC from side-channel analysis and other
hardware Trojans is required. Figure 4 shows the architecture of an Aegis processor.
The state of the art processors employed in IoT is mostly RISC processors due
to its simple architecture and low power consumption. RISC processors help to
accelerate the execution of simplified instructions. As against CISC architecture,

Fig. 4 Aegis processor architecture [7]

750 S. Rajendran and R. Mary Lourde

RISC methodology utilizes a complex instruction set to achieve high code density
and save costly memories. In IoT network, there is a limitation on the memory space
due to area and power constraints. In such cases, CISC processor is an advisable
solution. The scalable and reconfigurable micro-coded multi-core processor is a
proposed architecture for a CISC processor. The main advantage of this architecture
is its increased efficiency [8].
The recent trends in processor architecture for embedded systems are a generalized
architecture which requires lesser programming, high energy efficiency and enhanced
security aspects. There are few processors available today which do provide the
required security along with low power consumption and high performance. The
most widely used processor is the ARM processor. Figure 5 shows the architecture of
ARM Cortex M4 processor. However, this processor is also not completely secure.
Researches show that the information in such processors can be leaked through
radio frequency, optical, power and timing side channels and by interfaces such as
JTAG and RS232. The most prominent locations for hardware Trojans are memory
units and clocking units. In order to ensure hardware security of a processor, it is
necessary to identify the critical components that could fall prey to such malicious
intrusions at the design stage. In a processor, the vital information is always stored
in the memory. So an attacker would always attempt to obtain access to the memory.
Hence, memory and communication buses are the most critical components to be
taken into account while designing for security and memory authentication plays a
crucial role in safeguarding against threats.

Fig. 5 ARM Cortex M4 processor [9]

Security Threats of Embedded Systems in IoT Environment 751

4 Secure Embedded Processor Design Challenges

Since embedded systems need to fulfill the security requirements for the application
they are designed for, a number of challenges are faced at the design stage. These
include processing gap, flexibility, battery limitation, tamper resistance and cost [5].
• Complexity of security protocols and increased data rates make it difficult for the
embedded processor to keep up with computational demands involved.
• The embedded systems need to be flexible to support multiple security mechanisms
and protocols.
• Battery operated systems face a limitation of power available for performing
computations.
• The only mechanism that can guarantee foolproof security is through resistance
implementation into the hardware design.
• Implementation of security at the hardware level results in a sharp rise in cost
due to the additional circuitry involved. Therefore, there always exists a tradeoff
between security and cost.
• There also arises the need for designers to be well informed of the security impli-
cations of the design so that the design cycle can be modified to evaluate security
at the different stages.
When linked to the IoT, there is a huge amount of data storage and transfer
operations involved. Hence, safeguarding memory from potential attacks is a prime
challenge. Establishing security between endpoints in a network can only be carried
out after taking into account the cryptographic schemes, networking protocols and
communication standards.

5 Memory Attacks

Ability to verify that the data read from a specific memory location by a processor is
the same as the one it stored at the location during last write is referred to as memory
authentication [10]. The main assumption in past works was that the processor chip
alone is trusted. One technique involves storing the cryptographic digest of external
memory on a trusted area of chip and comparing the data read from the external
memory with the digest each time a read is performed. This is a tedious method as
all data from external memory should be fetched to perform integrity checking for
every read. For every write operation, the data needs to be updated as well. Another
solution suggested involves keeping separate cryptographic digests for various mem-
ory blocks. This helps to bring down the memory bandwidth problem in the previous
method but increases the cost due to requirement for additional hardware. An opti-
mal solution using tree-based structures has also been proposed. In this approach,
the leaves are the memory blocks to be shielded from attacks, and the root node com-
putes the current state of the memory block by applying an authentication primitive
to tree nodes starting from leaves.
752 S. Rajendran and R. Mary Lourde

Fig. 6 An attack model of

the memory [10]

Prevention of attacks to the memory is the primary goal of memory authentication.

Spoofing attacks, splicing attacks and replay attacks are the three main categories
of active attacks that an adversary resorts to. Figure 6 shows a model of a memory
attack. In spoofing attacks, adversary replaces a memory block with a malicious one
by obtaining access to the data bus by means of a control switch. In splicing attacks,
the adversary keeps a copy of the original data in a malicious memory. The activation
of the switch command by the adversary forces the processor to read data from the
malicious location. Replay attacks involve the replacement of memory content with a
previous data stored there. The adversary copies data from an address location to the
malicious memory to be used later to alter the contents of the same address location.
Multiprocessor technology is widely used in high-end computing platforms. In
multiprocessor systems, we require to include security aspects of shared memory
also. In symmetric multiprocessors in order to ensure cache coherence, a processor
sends the same data to all the cores along with the intended core. An attack that
comes into play in this context is message dropping where one of the cores may be
temporarily cut off from the bus. So the need arises for bus transaction authentication
on cache to cache transfers. Figure 7 depicts message dropping attack on symmetric
multiprocessor platform (SMP).
Elbaz et al. [10] discuss a technique for data encryption and authentication on
processor-memory bus wherein after adding cipher to the data, a data integrity
check is also performed to confirm the authenticity of information. This technique is
applied to guarantee secure data communication between the System on Chip (SoC)
and external memory. External memory usually holds sensitive information, and an
adversary would try to attack the data bus to acquire it. In order to prevent central pro-
cessing unit (CPU) modification, hardware security measures are inserted between

Fig. 7 Message dropping attack on SMP

Security Threats of Embedded Systems in IoT Environment 753

the cache memory and memory controller. An additional tag is appended to the plain-
text before encryption. The CPU processes read-only and read/write data. The tag
used for each of these is different. Since a read-only data stored once in external
memory is not modified during program execution and also since the encryption key
is the most significant address bits of the encrypted block as tag (T). In the event of a
splicing attack, the address used by processor to fetch the data block from the exter-
nal memory and the address used by the integrity checker to generate tag (T’) will
not match the stored tag (T). Figures 8 and 9 show the diagrammatic representation
of the process.
From a broader angle, malicious attacks on the smart grid could impact devices and
users across continents since IoT technology is involved. Data from the smart meter
could be used by an adversary as an information side-channel to study customer
behavior or habits. Smart grid technology is one of the many applications in IoT
which highlights the vulnerabilities involved in the system. These vulnerabilities
and challenges arise as a result of the integration of cyber and physical systems [11–
13]. It draws our attention to the other side of the coin where the challenges like
privacy and security of the users, as well as data, need to be safeguarded through
effective hardware design and analysis [14].
The need for development of hardware security also stems from the fact that most
of the existing security solutions were developed keeping software concerns in mind.
These solutions as described in the paper by Babar et al. [15] reveal that the software
attacks are well taken care of while most of the hardware attacks are unguarded. In
order that an entirely secure system be developed, both the hardware and software
needs to be immune to attacks.

Fig. 8 Tag insertion and encryption during write

Fig. 9 Decryption and tag matching during read

754 S. Rajendran and R. Mary Lourde

6 Conclusion

High level of security in an IoT system can be achieved by addressing all possible
threats that can affect both hardware and software of the system. The design of
embedded systems, which are an essential building block of IoT, will help safeguard
the system against hardware attacks. Memory unit of a general-purpose embedded
processor needs a secure design to thwart malicious attacks as they are programmed
as per user applications. Developing algorithms for memory protection and Trojan
detection will prove as a milestone in secure design of embedded systems for IoT
applications.

References

1. Zhao K, Ge L (2013) A survey on the internet of things security. In: 9th international conference
on computational intelligence and security (CIS). IEEE, pp 663–667
2. Gubbi J, Buyya R, Marusic S, Palaniswami M (2013) Internet of Things (IoT): a vision,
architectural elements, and future directions. Future Gener Comput Syst 29(7):1645–1660
3. Atzori L, Iera A, Morabito G (2010) The internet of things: a survey. Comput Netw
54(15):2787–2805
4. Xu T, Wendt JB, Potkonjak M (2014) Security of IoT systems: design challenges and oppor-
tunities. In: Proceedings of the 2014 IEEE/ACM international conference on computer-aided
design. IEEE Press, pp 417–423
5. Kocher P et al (2014) Security as a new dimension in embedded system design. In: Proceedings
of the 41st annual design automation conference. ACM
6. Kaabouch M, Le Cocquen E (2011) Bi-processor architecture for secure systems. U.S. Patent
No. 7,984,301. 19 July 2011
7. Suh GE, O’Donnell CW, Devadas S (2005) AEGIS: a single-chip secure processor. Inf Secur
Tech Rep 10(2):63–73
8. Ma N et al (2014) A hierarchical reconfigurable micro-coded multi-core processor for IoT
applications. In: 9th international symposium on reconfigurable and communication-centric
systems-on-chip (ReCoSoC). IEEE
9. ARM Processors, [Link]/lit/er/spmz637/[Link]
10. Elbaz R et al (2009) Hardware mechanisms for memory authentication: a survey of exist-
ing techniques and engines. In: Transactions on computational science IV. Springer Berlin
Heidelberg, pp 1–22
11. Khurana H, Hadley M, Lu N, Frincke DA (2010) Smart-grid security issues. IEEE Secur Priv
8(1):81–85
12. Liu J, Xiao Y, Li S, Liang W, Chen CP (2012) Cyber security and privacy issues in smart grids.
IEEE Commun Surv Tutor 14(4):981–997
13. Simmhan Y, Kumbhare AG, Cao B, Prasanna V (2011) An analysis of security and privacy
issues in smart grid software architectures on clouds. In: IEEE 4th international conference on
cloud computing. IEEE, pp 582–589
14. McDaniel P, McLaughlin S (2009) Security and privacy challenges in the smart grid. IEEE
Secur Priv 1(3):75–77
15. Babar S et al (2011) Proposed embedded security framework for internet of things (IoT). In:
2nd IEEE international conference on wireless communication, vehicular technology, informa-
tion theory and aerospace & electronic systems technology (Wireless VITAE), Feb 28. IEEE,
pp 1–5
Decentralized Bagged Stacking Ensemble
Mechanism (DBSEM) for Anomaly
Detection

S. L. Sanjith and E. George Dharma Prakash Raj

Abstract Intrusion detection has become a major need for the current networked
environment due to the high usage levels and the mandatory security that is needed,
as sensitive information are being shared in the network. However, there exist several
intrinsic issues in the network data that complicates the detection process. Further,
real-time detection is also required due to the high velocity of data flow that can
be expected in the domain. This paper presents an ensemble-based intrusion detec-
tion model to handle data imbalance and noise. Further, the entire approach has been
decentralized to enable parallelized detection. The proposed model utilizes a BAgged
Stacking Ensemble (BASE) as the detection model. The ensemble architecture ini-
tially creates data bags, enabling distributed processing. The bags are processed by
multiple heterogeneous base learners. Prediction results from the base learners are
passed to a stacked classifier for final predictions. This ensemble model is distributed
over the network to enable decentralized processing. Experiments were performed on
the NSL-KDD data and the results were compared with recent models. Comparisons
with state-of-the-art models indicate the effectiveness of the proposed model.

Keywords Intrusion detection · Ensemble · Stacking · Bagging · Decentralization

1 Introduction

A huge development in the computing and networking environments have been wit-
nessed in the current decade. However, awareness in terms of technology and threats
to technology are still uncalled for. Further, even the improvements in protection tech-
nologies are still remaining to be enhanced for want of better protection schemes.
This has resulted in the network intrusion domain moving towards evolving security

S. L. Sanjith (B)
Indian Institute of Management Tiruchirappalli, Tiruchirappalli, Tamilnadu, India
e-mail: sanjithsl@[Link]
E. George Dharma Prakash Raj
Bharathidasan University, Tiruchirappalli, Tamilnadu, India
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 755

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
756 S. L. Sanjith and E. George Dharma Prakash Raj

based mechanisms [1]. This is assumed to provide better detection of the security
threats.
Although several intrusion detection mechanisms such as malware prevention
modules, firewalls, authentication mechanisms, and data encryption schemes are
available, they are still not sufficient for the intrusion detection process [2]. The
current network models seems to mandate additional defense mechanisms for want
of additional security.
Artificial intelligence and artificial intelligence-based systems have become most
used approaches for detection of intrusions in networks [3, 4]. The improvements
in artificial intelligence-based models has made this domain one of the sought out
domains in the network intrusion detection scenario. Further, artificial intelligence-
based technologies can also work well with legacy intrusion detection models. Hence
they are used in conjunction with existing mechanisms for providing better security
for the user.
Intrusion detection models are generally classified into two categories [5]. First
is the misuse based detection models, where packet signatures are analyzed and the
model specifically concentrates on handling anomalous signatures. Second is the
anomaly-based detection models that aims to learn the signatures of normal packets
to identify the records that exhibits deviations when analyzed from the perspective
of normal packets are to be flagged as anomalous.
Domain analysis further reveals that the intrusion detection domain suffers from
the issue of data imbalance and noisy data [6]. Data imbalance refers to the existence
of a large number of instances in one class usually termed as the majority class and
relatively less number of instances of another class usually referred to as the minority
class [7]. This results in the majority classes getting overtrained, while the minority
classes remain undertrained. An example for data imbalance is presented in Fig. 1.
This work presents a Decentralized Bagged Stacking Ensemble (DBSEM) based
model that can effectively enable faster and more accurate predictions. Bagging was
observed to effectively reduce the issue of data imbalance to a large extent. Further, it
was also observed that stacking the ensemble was able to reduce the effects of noise.
Experimental results and comparisons indicate demonstrates the effectiveness and
efficiency of the proposed model.

2 Literature Review

Intrusion detection has gained significance due to the increased reliance on networks
and network-based communication. Though there is a huge amount of literature
corresponding to this domain, there is still a need for a decentralized, flexible and
reliable intrusion detection model. This section discusses some of the most recent
and most prominent works in the domain of intrusion detection.
An intrusion detection model using lazy learning methodology was provided by
Chellam et al. [8]. Lazy learning was observed to exhibit high computational com-
plexity levels. This model proposes a weighted indexing technique, to reduce the
Decentralized Bagged Stacking Ensemble Mechanism … 757

Fig. 1 Data imbalance—a graphical view

computational capacity of last learning models. An imbalance handling model that

is a variant of the RIPPER algorithm was presented by Cieslak et al. [9]. This is
a clustering-based approach that handles data imbalance by artificial generation of
minority classes. An anomaly-based intrusion detection model was presented by
Wang et al. [10]. This method trains on regular network behavior data and can
effectively predict anomalous behaviors. The model is based on Principal Compo-
nent Analysis (PCA) as the initial preprocessing technique. A J48 based intrusion
detection model specifically designed for high dimensional data was presented by
Shahbaz et al. [11]. This technique is based on feature extraction and feature selection
for highly improved detection of intrusions.
A real-time model for intrusion detection in high-speed networks was presented
by Rathore et al. [12]. This is a Hadoop-based model that operates on REPTree and
J48 algorithms. The model selects nine best parameters for the processing archi-
tecture enabling highly effective predictions. The research directions have further
moved forward from Hadoop and has now started using spark-based models for pre-
diction. A spark-based technique for network intrusion detection on streaming data
was presented by Dahiya et al. [13]. This technique is based on two feature reduc-
tion algorithms; Canonical Correlation Analysis (CCA) and the Linear Discriminant
Analysis (LDA). These algorithms enable faster and more accurate predictions on
the streaming network data. A parallelized anomaly detection model using Spark was
presented by Yafei et al. [14]. This model is based on HOTSAX intrusion detection
algorithms. The HOTSAX algorithm has been parallelized using the Spark archi-
tecture to enable faster processing. Other similar streaming data-based intrusion
detection models includes Big Data-based model by Juliette et al. [15], sensor data-
based model by Michael et al. [16], loss based model by Bamakan et al. [17] and
metaheuristic based model by Tamer et al. [18].
758 S. L. Sanjith and E. George Dharma Prakash Raj

Although metaheuristic models were employed for their speed and effective pre-
dictability levels, several applications involve metaheuristic models only for the pre-
processing phase. A Firefly algorithm based intrusion detection model was presented
by Selvakumar et al. [19]. This technique uses Firefly algorithm for the process of
Feature Selection on network data. A similar PCA based network intrusion detection
model was presented by Salo et al. [20]. Artificial neural network-based intrusion
detection model was presented by Shenfield et al. [21]. This technique is based on
fine-tuning the artificial neural network for effective prediction of network data.
Although the model proves to be effective with high prediction accuracy levels, the
computational complexity of neural networks is high, hence making the model less
suitable for real-time data analysis. Extreme learning machines have been gaining
prominence in the current periods due to the flexibility they provide. An extreme
learning machine based model was presented by Wang et al. [22]. This model uti-
lizes the equality constrained optimization technique to improve the detection lev-
els. The model has also been designed as an adaptive technique to enable effective
results. Other prominent ELM based models include approximation based model
[23], random search based model [24] and error minimized ELM [25].

3 Decentralized Stacked Ensemble Model for Anomaly

Detection

Anomaly detection in network environments is usually centralized. All the gener-

ated network transactions are initially passed to the centralized server. The anomaly
detection model is usually deployed in the centralized server. The model performs
predictions to identify if the particular transaction is normal or anomalous. The
decision-making process is always performed on the centralized server. Hence the
server usually handles higher loads, hence the prediction process is usually slow. A
decentralized architecture that can handle distributed workloads can provide faster
predictions. This work proposes a decentralized ensemble architecture for handling
distributed workloads in the network.

3.1 Decentralization of Architecture

Decentralization in the architecture is achieved by deploying the anomaly detection

model on multiple models. Network transmission data is analyzed by these nodes
and anomalies are detected. The major advantage is that every node analyses its own
data, hence prediction is faster and is closer to the source, making the predictions
real-time effective.
Decentralized Bagged Stacking Ensemble Mechanism … 759

3.2 Decentralized Bagged Stacking Ensemble (DBSEM)

for Anomaly Detection

Every node in the network is embedded with an intrusion detection model. The
intrusion detection model is composed of the data preparation phase, the process of
bag creation, creation, and training of the heterogeneous base learners and finally
the stacked meta-model for final prediction. The architecture for DBSEM model is
shown in Fig. 2. The algorithm for the proposed model is shown below.
Algorithm
1. Input training data
2. Data preprocessing and data cleaning
3. Data sampling to create multiple bags
4. Pass bags to heterogeneous ensemble models
5. For each obtained data bag b
a. Apply training data b to obtain the trained model
b. Apply test data to model
c. Obtain predictions
6. Aggregate predictions from ensemble models
7. Append actual predictions with the generated predictions to form the input data
for meta-model
8. Pass the data to meta-model for training
9. To perform final prediction pass validation data to ensemble model
10. Aggregate predictions and pass it to the meta-model
11. Obtain final predictions from the meta-model

Data Preparation Input data is composed of network packets that pass through
the network. The packet data consist of several internal transfer details. Some data
sets might contain missing values, while others may contain erroneous values. These
data must be imputed or cleared to make the data fit for processing. This is done
during the data preparation or preprocessing phase. The input data is analyzed and
the instances containing missing values and erroneous values are deleted. Further,
machine learning models operate only on equally distributed double values. Hence,
data containing integers are normalized using

x − xmin
Norm = (max − min) + min (1)
xmax − xmin

where x is the actual value to be normalized, x max and x min are the minimum and
maximum values in the attribute, max and min are the interval ranges between which
the normalization is to be performed. This work uses a value of 0 and 1 for x min
and x max , respectively. The cleaned data is passed to the stacked ensemble model for
analysis.
760 S. L. Sanjith and E. George Dharma Prakash Raj

Fig. 2 DBSEM architecture

Bag Creation Phase The input data is composed of several instances, and the train-
ing data is usually composed of the entire data. However, the proposed work uses
multiple learning models. Passing the entire data for all the models will result in cre-
ation of same decision rules. Hence, the input training data is divided into multiple
training data bags, where each data bag is composed of 60% of the entire training
data. This process of data sampling results in random distribution of data to each
of the heterogeneous learners, and at the same time, results in some similarity in
Decentralized Bagged Stacking Ensemble Mechanism … 761

the data patterns between the various base learners. This process of data distribution
results in highly effective predictions.
Creation of Heterogeneous Base Learners Heterogeneity in base learners is main-
tained in the base learners, hence enabling varied rules in-order to create a robust
classifier model for prediction. This work utilizes Decision Tree and Random For-
est as base learners for the process. Both the base learners are intentionally set
as tree-based learners. Tree-based learners are effective in creating decision points
with maximum information gain. Hence using them during the ensembling process
enables discovery of varied rules. The rules with highest information gain can then be
retained for the future processes. Another major advantage of using ensemble models
is that the overall model can be tuned effectively such that it avoids overfitting.
All the created bags are passed through the base learners. Every bag is passed
through both the base learners and the base learner training is performed. Every
base learner is trained using different training data, hence their decision rules are
also different. The test data, when passed through each of the base learner provides
predictions. However, it results in multiple predictions. A single prediction for each
base learner. These predictions are not directly combined, instead, it is passed to a
second level meta-learner for final predictions.
Stacked Meta-Model for Final Prediction Predictions from the previous level
are passed to the stacked meta-model for the final prediction. All the predictions are
combined, along with the actual labels to obtain the training data for the meta-model.
This phase is constructed with logistic regression as the prediction model. The
major advantage of using logistic regression as the preferred model is that it utilizes
a linear model for fitting the objective function. As the required predictions are to be
performed from the previous predictions, utilizing a simple linear model is observed
to exhibit effective predictions. Further, Logistic Regression is the de facto model
that is usually utilized in stacking models.
The predictions, along with the actual labels are passed to the Logistic Regression
model. Logistic Regression is used as the meta prediction model. The main func-
tionality of this model is that it learns the issues of the previous model to provide the
final predictions. Results from this model is used as the final prediction.

3.3 Prediction Analysis

Intrusion detection in network transmission data is usually modeled as a binary

classification problem. The predictions are usually provided to mark transactions
as normal or anomalous. Prediction analysis is performed by initially filtering the
anomalous predictions and the transactions that correspond to these predictions.
Anomalous transactions exhibit higher significance compared to normal transac-
tions. Hence prediction of such transactions gains considerable significance over the
prediction of the normal classes.
762 S. L. Sanjith and E. George Dharma Prakash Raj

The major reason for performing this process is that it is mandatory for the models
to reduce False Positive predictions. Hence re-analyzing the anomalous transactions
becomes a necessary component of the domain.

3.4 Secondary Prediction of Anomaly Data

The prediction process explained in the previous sections is performed in each node
of the network. The anomalous data that has been filtered in the previous phase
is passed to all the other nodes in the network. This transmission has been done
for two purposes; first, every node must have updated anomalous signatures. This
maintains that all the nodes are up-to-date with the anomalous signatures. Further,
the anomalous signatures should be confirmed to make sure that they are actually
anomalous.
The anomalous signatures identified from each node is passed to all the other
nodes. Predictions are obtained from these nodes and these predictions are trans-
ferred back to the origin node. The origin node performs voting to identify the final
prediction. The final prediction is transferred to all the available nodes to be saved
as an anomalous signature.

4 Result Analysis

The proposed DBSEM model has been implemented using Python. NSL-KDD
dataset is used to measure the significance of the proposed model. The NSL-KDD
data is composed of 14 attributes. Details about the attributes are shown in Table 1.
Experiments were performed using 10-fold cross-validation, and the results were
obtained. Results have been analyzed in terms of standard classifier performance
metrics.
The Receiver Operating Characteristics (ROC) curve for the proposed model is
shown in Fig. 3. The graph is created by plotting FPR in the x-axis and TPR in the
y-axis. The area of the obtained curve provides the prediction efficiency of the model.
Higher area represents better performance. It could be observed that the below curve
exhibits the highest area, exhibiting the efficiency of the proposed model.
The Precision Recall (PR) curve presents the prediction efficiency of the proposed
models. High precision and high recall are the requirements of the classifier model. It
could be observed in Fig. 4 that the proposed model exhibits very high precision and
recall values that are close to 1, exhibiting the effectiveness of the proposed DBSEM
architecture.
The table depicting the performance of the proposed DBSEM model is shown in
Table 2. It could be observed that FPR and FNR, shows values closer to 0, while
all the other metrics shows values closer to 1. This elucidates the significance of the
prediction mechanism.
Decentralized Bagged Stacking Ensemble Mechanism … 763

Table 1 NSL-KDD
No. Attribute
attributes
1 src_bytes
2 service
3 dst_bytes
4 flag
5 diff_srv_rate
6 same_srv_rate
7 dst_host_srv_count
8 dst_host_same_srv_rate
9 dst_host_serror_rate
10 dst_host_srv_serror_rate
11 dst_host_diff_srv_rate
12 serror_rate
13 logged_in
14 Attack

Fig. 3 ROC curve ROC Curve

1.2

0.8
TPR

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1 1.2
FPR

PR Curve
1.2

0.8
Precision

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1 1.2
Recall

Fig. 4 PR curve
764 S. L. Sanjith and E. George Dharma Prakash Raj

Table 2 Performance metrics

Proposed model (DBSEM)
of the proposed model
TPR 0.98
FPR 0.01
TNR 0.99
FNR 0.02
Accuracy 0.99
Precision 0.98
Recall 0.98
F1 Score 0.98

A comparison of the proposed DBSEM model with the RampLoss model [25]
and the previous work of the authors (Reinforcement based Heterogeneous Ensemble
Model-RHEM) is shown in Fig. 5 and Table 3. It could be observed that all the models
exhibit almost similar predictions. However, it should be observed that the proposed
model is decentralized and operates with just a part of the data, while RampLoss
model and RHEM are centralized approach utilizing the entire spectrum of available
information. Further, the time requirements of the proposed model is much less
compared to the RampLoss model and RHEM due to the decentralized nature of the
proposed DBSEM architecture.

Fig. 5 Comparative analysis

Table 3 Comparison of metrics

Proposed RampLoss RHEM—reinforcement-based
model—DBSEM (Centralized) model (Centralized) Sanjith
Bamakan et al. et al.
Accuracy 0.99 0.99 0.99
FPR 0.01 0.01 0.00
Precision 0.98 0.98 0.99
F1 score 0.98 0.99 0.98
Decentralized Bagged Stacking Ensemble Mechanism … 765

5 Conclusion

Intrusion detection has been one of the significant areas of research in the current
decade, due to the increasing necessity for security in network transmissions. This
work proposes an effective intrusion detection model that has been developed in
a decentralized fashion to enable faster and more effective detection of anomalies.
The Decentralized Bagged Stacking Ensemble Model (DBSEM) architecture has
been designed to provide faster and more accurate predictions in a decentralized
manner. Input data is converted into data bags and are passed to the base learners.
The base learners train on the input data and generates predictions. These predictions
are collected and passed to the secondary algorithm for meta-learning. This stacking
structure enables improved predictions. Experiments were performed using NSL-
KDD data and the results were compared with state-of-the-art models. The results
indicate the effectiveness and the high performing nature of the proposed DBSEM
architecture. The proposed architecture has been found to be highly scalable and
tends well to parallelization. Future enhancements will be towards implementing the
model on streaming architectures like SPARK.

References

1. Pontarelli S, Bianchi G, Teofili S (2013) Traffic-aware design of a high-speed fpga network

intrusion detection system. IEEE Trans Comput 62(11):2322–2334. [Link]
TC.2012.105
2. Garcia-Teodoro P, Diaz-Verdejo J, Maciá-Fernández G, Vázquez E (2009) Anomaly-based
network intrusion detection: techniques, systems and challenges. Comput Secur 28(1–2):18–28
3. Tang Y, Chen S (2007) An automated signature-based approach against polymorphic internet
worms. IEEE Trans Parallel Distrib Syst 18(7):879–892
4. Tan Z, Jamdagni A, He X, Nanda P, Liu RP (2014) A system for denial-of-service
attack detection based on multivariate correlation analysis. IEEE Trans Parallel Distrib Syst
25(2):447–456
5. Bhuyan MH, Bhattacharyya DK, Kalita JK (2014) Network anomaly detection: methods,
systems and tools. IEEE Commun Surv Tutorials 16(1):303–336
6. Akila S, Srinivasulu Reddy U (2016) Data imbalance: effects and solutions for classification
of large and highly imbalanced data. Proc ICRECT 16:28–34
7. Akila S, Srinivasulu Reddy U (2017) Modelling a stable classifier for handling large scale data
with noise and imbalance. In: IEEE international conference on computational intelligence in
data science
8. Chellam A, Ramanathan L, Ramani S (2018) Intrusion detection in computer networks using
lazy learning algorithm. Procedia Comput Sci 132:928–936
9. Cieslak DA, Chawla NV, Striegel A (2006) Combating imbalance in network intrusion datasets.
In: GrC, pp 732–737
10. Wang W, Battiti R (2006) Identifying intrusions in computer networks with principal component
analysis. In: Proceedings of the first international conference on availability, reliability and
security, pp 270–279
11. Shahbaz MB, Wang X, Behnad A, Samarabandu J (2016) On efficiency enhancement of the
correlation-based feature selection for intrusion detection systems. In: 2016 IEEE 7th annual
information technology, electronics and mobile communication conference (IEMCON), pp
1–7
766 S. L. Sanjith and E. George Dharma Prakash Raj

12. Rathore MM, Paul A, Ahmad A, Rho S, Imran M, Guizani M (2016) Hadoop based real-time
intrusion detection for high-speed networks. In: 2016 IEEE global communications conference
(GLOBECOM), pp 1–6
13. Dahiya P, Srivastava DK (2018) Network intrusion detection in big dataset using spark. Procedia
Comput Sci 132:253–262
14. Wu Y, Zhu Y, Huang T (2015) Distributed discord discovery: spark based anomaly detection
in time series. IEEE
15. Dromard J, Roudière G, Owezarski P (2015) Unsupervised network anomaly detection in
real-time on big data. Springer, Berlin
16. Hayes MA, Capretz MAM (2015) Contextual anomaly detection framework for big sensor
data. Springer, Berlin
17. Bamakan SMH, Wang H, Shi Y (2017) Ramp loss K-support vector classification-regression;
a robust and sparse multi-class approach to the intrusion detection problem. Knowl-Based Syst
126:113–126
18. Ghanem TF, Elkilani WS, Abdul-kader HM (2014) A hybrid approach for efficient anomaly
detection using metaheuristic methods. J Adv Res
19. Selvakumar B, Muneeswaran K (2019) Firefly algorithm based feature selection for network
intrusion detection. Comput Secur 81:148–155
20. Salo F, Nassif AB, Essex A (2019) Dimensionality reduction with IG-PCA and ensemble
classifier for network intrusion detection. Comput Netw 148:164–175
21. Shenfield A, Day D, Ayesh A (2018) Intelligent intrusion detection systems using artificial
neural networks. ICT Express 4(2):95–99
22. Wang CR, Xu RF, Lee SJ, Lee CH (2018) Network intrusion detection using equality
constrained-optimization-based extreme learning machines. Knowl-Based Syst 147:68–80
23. Huang G-B, Chen L, Siew C-K (2006) Universal approximation using incremental constructive
feedforward networks with random hidden nodes. IEEE Trans Neural Netw 17(4):879–892
24. Huang G-B, Chen L (2008) Enhanced random search based incremental extreme learning
machine. Neurocomputing 71(16):3460–3468
25. Feng G, Huang G-B, Lin Q, Gay R (2009) Error minimized extreme learning machine with
growth of hidden nodes and incremental learning. IEEE Trans Neural Netw 20(8):1352–1357
The Process and Application of Dual
Extruder Three-Dimension Printer

Kunal Sekhar Pati and A. Karthikeyan

Abstract The three-dimension printer means the three-dimensional printing which

is also known as additive manufacturing and it is the advanced manufacturing tool
used for producing the most complex shape geometries. We can produce complex
geometry shapes without using any kind of computer tools. The three-dimensioned
printer is used in many fields in the industries because of its most creating and
functioning prototype with minimum human resources and time. 3D printer is used
in automobile, architecture, engineering, education, medical industries, and civil
engineering. It is most reliable, cost-effective and real-time application. The three-
dimension printer is widely used and it is a very interesting technology to look out
for.

Keywords Three-dimension printing · Additive manufacturing · Printing layer

1 Introduction

3D printing or additive manufacturing is a process of making a three-dimensional

object of any shape from digital model with different shapes. Traditional machining
techniques based on cutting the materials layer by layers. Layering technique is used
when the material is cutting each layer until the complete object is manufactured.
In this way, three-dimension D printer moves away from large amount of human
resources and we can modified our product according to user or anyone. We can
make and customize any product from home or any other place.
First time three-dimension printer is introduced on 1980s where a pattern is sub-
merged in liquid polymer and that would be traced by computer. The traced pattern

K. S. Pati (B)
Department of Embedded Technology, School of Electronics Engineering, VIT University,
Vellore, India
e-mail: Kunalpati4@[Link]
A. Karthikeyan
Department of Embedded System Technologies, VIT, Vellore, India
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 767

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
768 K. S. Pati and A. Karthikeyan

is hardened into layer. This is the way built an object out of plastic. After tremendous
process additive manufacturing methods is introduced. In additive manufacturing
process three-dimension objects are adding layer upon layer of materials. Materials
vary from technology to technology. The basic concept of additive manufacturing
is using of computer for three-dimension modeling software. First thing we have to
create is a three-dimension sketch then AM device reads the CAD file from com-
puter and build a structure layer by layer. The material may be plastic, liquid, powder
filament.
Now the cost of acquiring three-dimension printing has been decreasing with
advancement of technology. Nowadays the usage of three-dimension printer has been
raised with average cost ranging on market. For commercial usage three-dimension
printers have been increased in every sector on market, for example, aerospace, spare
parts manufacturing, automobile, etc. In fact it is required skilled and expertise person
to do both software and final printing work. In production industries three-dimension
printers are most emerging product on market.

2 How It Works?

There are different types of three-dimension printers’ technology in additive

manufacturing. We are using fused deposition modeling (FDM).
Fused Deposition Modeling
Fused Deposition Modeling (FDM) technology was developed by Scott Crump in
1980s. In this method, we can print not only prototype but also know the concepts of
modeling and user product. FDM three-dimension printing technology based on the
thermoplastics materials so the product is very excellent with mechanical, chemical
and thermal.
On FDM technology the three-dimension printing machine build the objects layer
by layer from bottom to top heating and extruding the thermoplastics filaments.
First design a three-dimension CAD layer by layer model on software and printer
extruder calculate it layer by layer. Thermoplastic material is supported by printer
extruder. Then the printer heats the thermoplastic material at 220 °C on it melting
point and extrude it through nozzle on base. A computer of three-dimension printer
translates the dimension of X, Y, and Z coordinate and control the nozzle. The base
also moves according to the nozzle it means that it coordinate the nozzle. When one
layer completed its printing then the base or bed goes down and start printing the
next layers. This process is keep on going until the complete object is not printed.
Printing time depends on size and complexity of the object. Small object print fast as
compare to large object. After complete printing of an object it is take time harden
the thermoplastic materials (Fig. 1).
The Process and Application of Dual Extruder … 769

Fig. 1 Fuse deposition modeling process

3 Components

ATmega328 Arduino UNO

• High performance, low power AVR 8-bit microcontroller

• 32 × 8 general-purpose working registers
• 1 kB EEPROM
• 2 kB internal SRAM
• 23 programmable I/O lines
• 2.7–5.5 operating voltage.
770 K. S. Pati and A. Karthikeyan

Ramp 1.4 3D Printed Control Board

• Expandable to control other component

• 3 MOSFETs for heater and fan
• All MOSFETs are hooked in to PWM
• Option to connect 2 motors on Z
• I2C and SPI pins are also available.

Limit Switch

It is a mechanical block is detected when it is activated. Its help for the alignment
between bed and extruder.
Proximity Sensor
It is used for auto leveling and going to need a probe, which can accurate measure
the distance of bed at various point.
Servo Motor
Limit switch is used for bed, which is not contact with bottom of the extruder. So
servomotor is used for movement of the limit switch, which is fixed besides of the
nozzles.
The Process and Application of Dual Extruder … 771

4 Application of Three-Dimension Printers

Product Development

If we need small product then do it as soon as possible. A three-dimension printer is

fast, cost effecting, and modified the product as soon as possible at no additional cost
with minimum amount of time. It is the best machine to design early-stage design
product.
Construction Field
On civil sector, if we construct a huge bridge, building or any other plan so at that time
we need first prototype of that structure. Three-dimension printer means early stage
of evaluation of the complex and rough design of any construction or architecture.
Medical Field
Hearing aid has been made using three-dimension printing technology. Those brave
soldiers who lost his legs or hands-on war field for them three-dimension printer can
able to manufacture artificial legs or hands for them. Nowadays doctors are using
artificial heart for pharmaceutical testing.
End-Use Parts
Three-dimension printer can produce low volume, customized end-use product. This
gives more flexibility to small industries as compare to large batch industries. The
cost effecting to this end-use products are low.

5 Advantages

• Three-dimension printer product cost is low as compare to other technology

product.
• Three-dimension printer technology saves more time for both human and product
those can produce from other technology.
• Three-dimension printers are more efficient, faster and easier.
• Quality of the product more strong and durable.

6 Conclusion

Now the world is changing with the help of three-dimension printer. The use of
three-dimension printer in medical sector is going to beyond the level and nobody
knows what is in future. The additive manufacturing process helps people in solving
many problems. Three-dimension printer is limitless and till now it solves only the
772 K. S. Pati and A. Karthikeyan

scratched, there is still more to be uncovered. Now three-dimension printer is still

new technology and continuously improving. It already enhances the life of many
patients around the world.

References

1. [Link]
2. [Link]
technologies-overview/
3. [Link]
4. [Link]
applications/
5. [Link]
[Link]
6. [Link]
aerospace/
7. [Link]
8. [Link]
9. [Link]
10. [Link]
11. [Link]
reviewed-article-ORR
12. [Link]
Face Detection and Natural Language
Processing System Using Artificial
Intelligence

H. S. Avani, Ayushi Turkar and C. D. Divya

Abstract The objective of this paper is to look at a system based on artificial intel-
ligence that recognizes the faces and the system that processes the natural language.
The process of face recognition has three phases: face representation, removal of
features, and classification. But the most important phase-out of every phase is
extraction. At this stage, the image is extracted from the unique features of the face
image. However, variation in appearance remains one of the main factors affecting
the accuracy of face recognition systems.

Keywords Natural processing system · Face · Artificial intelligence

1 Introduction

Face recognition technology can verify or identify an individual from a digital image
or video source (video frames). Face recognition system works using various meth-
ods. Generally, the system compares the selected facial features in the given face
image database. Face recognition systems are used to enforce the law, entertainment,
monitoring, access to security systems, banking, personal identification, etc. Multi-
plier reception (MLP), face localization, neural network, Natural Language Process-
ing (NLP), principal component analysis (PCA), hierarchical-PEP, LDA, principle
component analysis with RBF, Graphics processing unit, etc. are the most popu-
lar face recognition techniques. The face recognition process has three phases, one
is faced representation, feature extraction, and classification. Face representation is
nothing but a face model and determines the successive algorithm that detects and
identifies the face image. The face recognition at the entrance level determines that
the given image represents the face. In the next phase, which is the extraction of
the function phase, the most unique and useful features (properties) are extracted

H. S. Avani (B) · A. Turkar · C. D. Divya

Department of Computer Science and Engineering, Vidyavardhaka College of Engineering,
Mysuru, Karnataka, India
e-mail: avanirao0811@[Link]

© Springer Nature Singapore Pte Ltd. 2020 773

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
774 H. S. Avani et al.

from the face images. The face image is compared to the images in the database for
all the features obtained. This is done in the classification phase. The output from
the classification phase gives the identity of the given face image from the database
together with the highest matching score (i.e. the smallest differences from the input
face image).

2 Literature Review

Face detection and expression recognition using the neural network approach, where
the author uses the artificial neural network, the basic propagation algorithm, the mul-
tiplier reception (MLP), the principle component analysis, and the RBF and graphics
processing unit for the face image process. The basic unit of communication such
as body movement facial expressions and physiological reactions are the parameters
and the final system should be able to detect facial expression in the same way as the
human brain [1].
A face recognition system approach uses artificial neural networks propagation
using a face localization or a localized face method. They used biometric, pixel,
segmentation, and neuron image processing. First, they check the matching of iso-
density lines of subject faces, which means comparisons of sizes or relative distance
of facial characteristics such as nose and mouth position of the eyes. The system
should recognize the human face at a very high precision density [2].
Using image processing and neural networks, a face recognition system approach
based on MATLAB uses methodologies such as discrete cosine transformation, self-
recognition map, and neural network. The face image is given as a parameter and
has an 81.36% recognition rate for 10 trials [3].
Recognition of the human emotional state by the detection of facial expression is
an approach that takes different types of images as the parameter and methodology
used is the principle of image pre-processing. And the system is tested with a database
of 30 different people with different expressions, the proposed PCM method is more
consistently accurate [4].
The processing of natural languages using artificial intelligence uses the NLP
approach. Speech, printed text or handwriting are passed to the system as parameters.
With this system, we can direct the robot that talks to the computer and nobody need
to work as a translator [5].
The artificial intelligence scope and challenges in India and the artificial intel-
ligence revolution. As an artificial intelligence, the computer machine that makes
it so intelligent to solve the computer problem that can only be solved by people.
Artificial intelligence is generally classified as a large AI and a narrow AI. AI-based
games are popular in recent days. The use of artificial intelligence helped to make
better use of time and resources. But India lags behind in the area of AI development
compared to other countries such as China and the USA. In order to benefit from AI,
the government must take the current advances in AI [6].
Face Detection and Natural Language Processing … 775

A face recognition approach using the main component analysis in MATLAB

follows the main component analysis methodology. Parameters are the removal of
some basic parts of the face, such as eyes, nose, mouth, and chin, with the database.
Increasing the number of face images in the database increases the system recognition
rate. The recognition rate, however, begins to saturate following a definite increase
in own value [7].
An effective face recognition approach using a modified center-symmetric local
binary pattern (MCS-LBP). This approach follows the methodology of the Local
Binary Modified Center-Symmetric Pattern (MCS-LBP). The face representation,
extraction of features, and classification are the three main parts of the process. They
have results with an accuracy of 88.7% for LBP and 92.8% for CS-LBP and 96.3%
for MCS-LBP [8].
The hierarchical-PEP model for real-world face recognition takes the structure of
the face parts as a parameter and the hierarchical PEP model method is adapted and
91.10% accuracy with SIFT [9] is achieved.
Face recognition based on a deep neural network is an approach that uses tradi-
tional neural network methodology. It takes face images to extract the feature vector
as a parameter. The resulting system will have a mean recognition accuracy of more
than 97.5% in LFW marches steadily towards human performance [10].
A classification approach using the radial base function uses the Fusion, LDA,
neural networks, and PCA method. The system sets video frames or photo frames as
parameters. The system accounts for 98.5% of the recognition rate [11].

3 Proposed Methodology

Artificial neural networks, also known as connection systems, are computer systems
that are similar to the animals’ biological neural network. The idea of AI neural net-
works is inspired by the neural biological network. The neural network is a framework
for other algorithms such as the algorithm for machine learning. It helps algorithms
cooperate and process complex data entries. These systems “learn” to perform tasks
by looking at examples, generally without any task-specific rules being programmed.
The methodology of the Principal component analysis also referred to as PCA.
It is a statistical approach that uses an orthogonal transformation to change a set of
possible related variables (different numerical values are taken by each entity) into
a set of values of linearly unrelated parameters known as the principal components
as shown in Figs. 1 and 2.
776 H. S. Avani et al.

Fig. 1 Generic representation of a face recognition system

Fig. 2 Overview of proposed system

Face Detection and Natural Language Processing … 777

4 Result Analysis

Ref No. Approach Methods Parameters Result

1 Face detection and Artificial neural The basic unit of Take a decision
expression network, basic communication like the human
recognition using propagation (Body movement, brain to detect
neural network algorithm, facial expression, facial expression
approaches multiplier physiological
reception (MLP), reactions)
principle
component
analysis, RBF,
graphics
processing unit,
etc.
2 Face recognition Face localization Matching of Performs human
system using or localized face, iso-density lines of face recognition at
backpropagation they used image subject faces a very high density
artificial neural processing, which means of accuracy
networks biometric, pixel, comparisons of
segmentation, sizes or relative
neuron distance of facial
features like the
position of eyes,
nose, and mouth
3 A MATLAB based Discrete cosine Face image Recognition rate
face recognition transform, 81.36% for 10
system using self-recognizing trials
image processing map, neural
and neural network
networks
4 Human emotional The principle of Different types of The system is
state recognition image images tested on a
using facial pre-processing is database which
expression used consists of 30
detection [Pre-processing different persons
component with different
analysis (PCA)] expressions, the
proposed system
PCM method have
greater accuracy
and consistency
(continued)
778 H. S. Avani et al.

(continued)
Ref No. Approach Methods Parameters Result
5 Natural language Natural Language Speech, printed Robot get a
processing using Processing (NLP) text or handwriting direction which
artificial helps to do
intelligence conversation with
a computer
without the
interference of a
person who works
as a translator
6 Artificial Null Null The government
Intelligence must adopt the
revolution and current
India’s artificial advancements of
intelligence artificial
development intelligence to
challenges and make use of its
scope benefits
7 Face recognition Principal Extraction of some We can increase
using principal component basic parts of a the recognition
component analysis face such as eyes, rate of our system
analysis in nose, mouth, and by increasing the
MATLAB chin, with the one number of images
stored in the of faces in the
database given database.
But the
recognition rate
will start
saturating when a
definite sum of the
increase in
eigenvalue
8 An efficient Modified Face With method LBP
approach to face Centre-Symmetric representation, we get 88.7%
recognition using Local Binary feature extraction, With method
a Modified Pattern and classification CS-LBP we get
Centre-Symmetric (MCS-LBP) 92.8%
Local Binary With method
Pattern MCS-LBP we get
(MCS-LBP) 96.3%
9 Hierarchical-PEP Hierarchical-PEP Face part as System achieved
model for model structures 91.10% accuracy
real-world face with a feature of
recognition SIFT
(continued)
Face Detection and Natural Language Processing … 779

(continued)
Ref No. Approach Methods Parameters Result
10 Face recognition Conventional Feature extraction Mean recognition
based on deep neural network from face images accuracy on LFW
neural network to get the feature marches steadily
vector towards the human
performance of
over 97.5%
11 Face recognition Fusion, LDA, Video frames or Gives 98.5% of the
by classification neural networks, photo frames recognition rate
using radial basis and PCA
function

5 Conclusion

When looking at previous research papers, artificial neural networks, and pre-
processing component analysis (PCA) are the most commonly used methodology.
The maximum accuracy for the combination of Fusion, LDA, Neural Networks, and
PCA methodologies is 98.5%. And we also observed that the methodology of the
neural network always works well. In the future, we will build a face recognition and
NLP system that detects or does not lie to a person.

Acknowledgements The authors express gratitude towards the assistance provided by Accendere
Knowledge Management Services Pvt. Ltd. In preparing the manuscripts. We also thank our mentors
and faculty members who guided us throughout the research and helped us in achieving desired
results.

References

1. Nisha SD (2015) Face detection and expression recognition using neural network approaches.
Glob J Comput Sci Technol F Graph Vis 15(3)
2. Revathy N, Guhan T (2012) Face recognition system using back propagation artificial neural
networks. Int J Adv Eng Technol 3:321–324
3. Nagi J, Ahmed SK, Nagi F (2008) A MATLAB based face recognition system using image
processing and neural networks. In: 4th international colloquium on signal process and its
applications, pp 83–88
4. Vasani GB, Senjaliya RS, Kathiriya PV, Thesiya AJ, Joshi HH (2013) Human emotional state
recognition using facial expression detection. Int J Eng Sci 2:42–44
5. Dhavare U, Kulkarni U (2015) Natural language processing using artificial intelligence. Int J
Emerg Trends Technol Comput Sci 4:203–205
6. Singh H (2017) Artificial intelligence revolution and India’s AI development: challenges and
scope. Int J Sci Res Sci Eng Tech 3(3):417–421
780 H. S. Avani et al.

7. Singh P, Sharma A (2015) Face recognition using principal component analysis in MATLAB.
Int J Sci Res Comput Sci Eng 3:1–5
8. Alapati A, Kang D (2015) An efficient approach to face recognition using a Modified Center-
Symmetric Local Binary Pattern (MCS-LBP). Int J Multimed Ubiquit Eng 10:13–22
9. Haoxiang L, Gang H (2015) Hierarchical-PEP model for real-world face recognition. In: Pro-
ceedings of the IEEE conference on computer vision and pattern recognition, 07–12 June, pp
4055–4064
10. Xinhua L, Qian Y (2015) Face recognition based on deep neural network. Int J Signal Process
Image Process Pattern Recogn 8:594–598
11. Mandhala VN, Bhattacharyya D, Kim TH (2015) Face recognition by classification using radial
basis function. Int J Multimed Ubiquit Eng 10:33–40
A Review on Detection of Online Abusive
Text

Bhumika Jain, Chaithra Bekal and S. P. PavanKumar

Abstract Presently, online networking have become a part and parcel of almost
everybody’s life. Also, it has become major medium of personal and commercial
communication. The current popularity of web technologies and social networking
has made mandatory for a person to be active on these sites. Therefore, people
became closely attached and find a medium to express their feeling, opinions, and
emotions through these sites and often share information without bothering what they
are sharing and with whom. Such context has become an avenue for cyberbullying.
Thus, Internet is a platform for using abusive text or image, which may lead to
many problems. Hence, it is important to place an effective system in order to put
an end to such activities, through text mining techniques, machine learning, and
natural language processing. In this paper, we compare different method of prototype
implemented in the detection of abusive content.

Keywords Abusive · Detection · Social networking · Text

1 Introduction

There are a lot of areas to express one’s opinion about any social aspects. One
can put out their opinion about something through social media. Social media like
Facebook, Twitter, etc. provide us the platform to put in our views about an issue
through comment boxes, through posting things to reach out people. Some posts
or comments might lead to bullying, harassment or even assassinating somebody’s
personality or character. This can be called as cyberbullying and is done to annoy or
hurt somebody intentionally. The comments or messages which are rude and said to
irritate somebody also come under this and are of major concern. And, it is a prime
important aspect these days that these kinds of abusive texts have to be detected

B. Jain (B) · C. Bekal · S. P. PavanKumar

Department of Computer Science and Engineering, Vidyavardhaka College of Engineering,
Mysuru, Karnataka, India
e-mail: bhumikajain94@[Link]

© Springer Nature Singapore Pte Ltd. 2020 781

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
782 B. Jain et al.

Fig. 1 Framework for the detection of abusive text [2]

automatically and reported. There can be hateful comments on women or religions

which are violating their constitutional rights.
A 2007 Pew Research study found 32% of teens have been victims of some type
of cyberbullying. And, it is still the same as per 2016 research study. Many children
were said to be attempting suicide due to this problem between 2008 and 2016 [1].
In abusive text detection, the framework processes as shown in Fig. 1 [2]. Like these,
many research studies have revealed that cyberbullying has been of major concern
these days and this problem has to be solved. There are many research studies on this
problem. There are many artificial intelligence techniques like text mining, natural
language processing, etc. used to solve the problems which have been discussed
below.

2 Literature Review

Many researchers have come up with different prototype and systems for detecting
offensive material. A lot of people including A. Mahmud, K. Z. Ahmed, and M.
Khan [3], Vandersmissen, F.D.T, Baptist., Wauters, T [4] and many more have come
up with their methods to detect abusive text, images, etc. In Kansara and Shekokar
[2] proposed a framework for detecting negative online interactions in terms of
abusive content carried out through text messages as well as images. In abusive
image detection, Local Binary Point (LBP) is used to identify interest points and
these points or features are then mapped to the existing visual word in vocabulary.
Further, support vector machine (SVM) helps in classifying whether image is abusive
or not, based on decision planes that define decision boundaries.
Chen et al. [5] have used the concepts of text mining and classifier techniques to
detect the abusive texts in any social media. They have taken 8 data sets from social
media sources that have already been reported as cyberbullying. Bag of words and N-
grams are used for text representation in text mining, whereas in the case of classifiers,
they have used two types namely Naïve Bayes (NB) classifier and support vector
machines (SVMs). The graphs being plotted according to performance obtained from
A Review on Detection of Online Abusive Text 783

the above-stated classifiers and text mining algorithms the results were obtained. The
results show the comparisons in different data sets by observing chi-square values.
In Nobata et al. [6] from Yahoo labs, NLP can be used as a method to analyze the
abusive texts in user comments. They have the Vowpal Wabbit’s regression model.
They have basically decided to divide into four classes, namely N-grams, Linguistic,
Syntactic, and Distributional Semantics. In the first-three processes, they have done
few mild preprocessing to remove disturbances and in the fourth process, comments
are represented as low-dimensional vectors and are jointly learned with distributed
vector representations of tokens using a distributed memory model. The experiments
conducted for various data sets using the four processes will produce results.
Yenala et al. [7] have conducted experiments and worked on the same aspect of
abusive texts detection but through different methodologies. Deep learning method
plays a major role in their work. In this paper, they have taken query completion
suggestions in web search and user conversations in messengers. Taking the data sets
from these two areas, they have used the method of convolutional neural networks
(CNN) and bi-directional LSTMs (BLSTM) to get results for various models and
derive conclusions on how it works for solving the problem, i.e., detection of abusive
texts.
Spertus [8] put forward a prototype system to automatically recognize flames,
called Smokey. It combines natural language processing and sociolinguistic obser-
vation to identify messages that not only contain insulting words but use them in an
insulting manner [5]. In this prototype, the text is passed through a parser in some
common format and then the output is converted to Lisp s-expression by sed and
awk scripts. A feature vector is produced for each message as these s-expressions are
processed through rules. Further, a decision tree generator C4.5 is used to generator
simple rules to evaluate the feature vector [8].
Another automatic flame detection method was given by Razavi et al. [9]. This
method extracts features at different conceptual levels and applies multilevel classi-
fication for flame detection. They have used machine learning algorithm, i.e., they
ran a three-level classification, using the Insulting and Abusing Language Dictionary
(IALD). The Complement Naïve Bayes classifier, the Multinomial Updatable Naïve
Bayes classifier, and Decision Table/Naïve Bayes hybrid classifier (DTNB) are used,
respectively, in the first, second, and third levels. Finally, the decision of whether the
text is okay or flame is based on the current instance.

3 Comparison of Different Detection Method for Abusive

Text

Table 1 gives us the picturesque idea of the detection techniques used by various
authors in the field of anomaly detection techniques using data warehouse. It also
gives the list of recommendations which we thought could have been implemented
in the system in future.
Table 1 Comparison of different detection methods for abusive text
784

Reference No. Approach Method Parameter Advantage Disadvantage Result

[2] A frame for Bag of visual words Abusive image or Both image and text are Concentrates on only Detects bullying content
cyberbullying detection (BoVW) concept with Abusive text considered; it blocks the image and text and not and either alerts or
in social network SVM classifier k-means abusive contents before on videos and audios message blocking
algorithm broadcasting
[5] Deep learning for Convolutional 8 data sets from various Most of the conversations Errors are encountered BLSTM significantly
detecting inappropriate bi-directional LSTM social medias from the datasets give such as misspellings, outperforms both
content in text (C-BiLSTM) technique correct output lack of semantic pattern-based and other
which combines the understanding, hand-crafted
strengths of convolution inadequate training data, feature-based baselines.
neural networks (CNN) borderline cases, and C-BiLSTM also proved
and bi-directional noise to be better than other
LSTMs (BLSTM) individual deep
learning-based
architectures CNN,
LSTM, and BLSTM
[6] Abusive language NLP features are used Primary data sets and This model outperforms Only character N-grams Character N-grams alone
detection in online user which measure different temporal data sets deep learning- based fare well in noisy data fare very well in these
content aspects of the user model sets noisy data sets
comment, Vowpal
Wabbit’s regression
model 5 in its standard
setting with a bit rate of
28
[7] Harnessing the power of In supervised machine Input text data sets from Large variety of datasets Concentrates more Studied impacts on
text mining for the learning, they have social media message are considered abusive text only but content detection
detection of abusive compared alternative text boards cyberbullying can be accuracy
content in social media representations and through audio or videos
dimension reduction as well
approaches, including
features election and
feature enhancement
(continued)
B. Jain et al.
Table 1 (continued)
Reference No. Approach Method Parameter Advantage Disadvantage Result
[8] Smokey: automatic Smokey: a prototype Abusive texts, flames Checks for syntactic Fails to detect sarcasm, Automatic flame
recognition of hostile system message constructs, avoid grammar, mistakes, detection
messages classification misclassifying innuendo, etc.,
[9] Offensive language Three-level classification Flames 16% better accuracy than Focus mainly on abusive Efficient flame detection
detection using with machine learning baseline, uses IALD, text software
multilevel classification algorithm as a software software is not very
sensitive to punctuation
and grammatical
mistakes
A Review on Detection of Online Abusive Text
785
786 B. Jain et al.

4 Fututre Work

In the future work, we could develop a tool that checks for abusive content on
social media. This tool should be able to check the text semantically as well for
any obnoxious content. We can implement this tool using deep learning, language
processing techniques.

5 Conclusion

Social networking is an integral part of most people. And, cyberbullying is a major

concern, as the number of online users is increasing. Many techniques and frame-
works have been proposed in order to detect offensive content. All these proposed
systems consider different variety of data sets and a very few have attempted to com-
pare these; in this paper, we have done it through a review table. Few methodologies
use NLP, classifier, bag of words while some use machine learning techniques. All
the different methodologies have their own advantages and disadvantages.

Acknowledgements The authors express gratitude toward the assistance provided by Accendere
Knowledge Management Services Pvt. Ltd. in preparing the manuscripts. We also thank our mentors
and faculty members who guided us throughout the research and helped us in achieving the desired
results.

References

1. Cook S. Cyberbullying facts and statistics for 2016–2018. [Link]

internet-providers/cyberbullying-statistics/
2. Kansara KB, Shekokar NM (2015) A framework for cyberbullying detection in social network,
vol 5, pp 494–498
3. Mahmud A, Ahmed KZ, Khan M (2008) Detecting flames and insults in text. In: Proceedings
of the Sixth International Conference on Natural Language Processing
4. Vandersmissen B, Baptist FDT, Wauters T (2012) Automated detection of offensive language
behavior on social networking sites, pp. xiv, 81 p.: ill
5. Chen H, Mckeever S, Delany SJ, Chen H, Mckeever S, Delany SJ (2016) Harnessing the power
of text mining for the detection of abusive content in social media. In: 16th UKCI, pp 1–19
6. Nobata C, Tetreault J, Thomas A, Mehdad Y, Chang Y (2016) Abusive language detection
in online user content. In: Proceeding of 25th international conference World Wide Web—
WWW ’16, pp 145–153
7. Yenala H, Jhanwar A, Chinnakotla MK, Goyal J (2017) Deep learning for detecting
inappropriate content in text. Int J Data Sci Anal
8. Spertus E (1997) Smokey: automatic recognition of hostile messages. In: AAAI’97/IAAI’97
proceedings of the fourteenth national conference on artificial intelligence and ninth conference
on innovative applications of artificial intelligence, pp 1058–1065
A Review on Detection of Online Abusive Text 787

9. Razavi AH, Inkpen D, Uritsky S, Matwin S (2010) Offensive language detection using Multi-
level classification. In: Farzindar A, Kešelj V (eds) Advances in artificial intelligence. AI 2010.
Lecture notes in computer science, vol 6085. Springer, Berlin, Heidelberg
10. Xiang G, Fan B, Wang L, Hong J, Rose C (2012) Detecting offensive tweets via topical feature
discovery over a large scale twitter corpus, pp 1980–1984. [Link]
2398556
Security of an IoT Network: A VLSI
Point of View

Sreeja Rajendran, Azhar Syed and R. Mary Lourde

Abstract Third-party IP cores, outsourcing of IC fabrication to untrusted foundries,

have increased the vulnerabilities in IC’s and reduced the trust factor of a designer
on the manufactured chips. These vulnerabilities are a consequence of malicious
modifications of the original design, which have the potential to cause catastrophic
damage to the system which uses these IC’s. IoT networks require the least vulnerable
and highly trustworthy IC’s. We present a detailed study of such malicious insertions.
Next, we discuss the methods for their identification and we also propose some
countermeasures from a VLSI aspect.

Keywords IoT · Hardware Trojans · Security · Processor

1 Introduction

Internet of Things (IoT) is a paradigm which links various realms of technology for
applications ranging from smart homes to smart healthcare and even smart agricul-
ture. The Internet of Things (IoT) has a lot to offer and presents us with countless
opportunities. Imagine systems that can be controlled and monitored over the Inter-
net, self-driven cars, factories that communicate with warehouses, smart homes with
digitized infrastructure. With the impact of Internet in our lives, the applications of
IoT are countless. However, with the advancement of such an impactful technology,
there is a downside to it. The widespread success of IoT depends on the data and
information it collects. The more information shared and stored, greater will be the
chances of security breaches, which will increase the risk of theft and manipulation.

S. Rajendran · A. Syed · R. M. Lourde (B)

Department of EEE, BITS Pilani Dubai Campus, Dubai, UAE
e-mail: marylr@[Link]
S. Rajendran
e-mail: sreejamanojnair@[Link]
A. Syed
e-mail: azharsyed18@[Link]

© Springer Nature Singapore Pte Ltd. 2020 789

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
790 S. Rajendran et al.

Fig. 1 Diagrammatic
representation of a
centralized IoT network

The more we increase the applications of the Internet of Things, more will be the
requirement for stringent security measures.
The traditional network layout for an application of the IoT is as shown in Fig. 1. It
represents a simple centralized structure. However, in recent times, the IoT has moved
from a simple structure to a complex network of decentralized devices. The IoT
network is highly dependent on different semiconductor technologies which include
microprocessors, sensors, and low power devices. IoT helps to provide network
connectivity to embedded systems, sensors, and actuators.

2 Security Threats

It is extremely important that security is taken into consideration from the very
beginning of the design process. There is a need to focus on security aspects of all
devices connected on the network. The security threat classification in the Internet
of Things network is shown in Fig. 2.
Physical attacks include reverse engineering and microprobing. Due to the expense
involved, it is believed that these attacks are seldom resorted to by adversaries. Tim-
ing analysis, power analysis, electromagnetic radiation analysis, etc. fall under the
category of side-channel attacks. Cryptanalysis threats include ciphertext attack,
plaintext attacks, etc. which attempt to break the encryption in order to obtain the
encryption key. Virus attacks, logic bombs, and denial of service fall under the cat-
egory of software attacks. Due to the broadcast nature of the transmission medium,
Security of an IoT Network: A VLSI Point of View 791

Fig. 2 Attack on IoT systems

wireless systems become vulnerable to network attacks. Denial of service, node cap-
ture, routing attacks, monitoring, and eavesdropping can be grouped under network
attacks category [1, 2].

3 Hardware Security Threats

Hardware security attacks are extremely difficult to identify but possess the capability
to impair a system resulting in devastating effects [1, 3–5]. Attacks carried out on the
hardware are generally in the form of Trojans which are specifically called Hardware
Trojans. It is defined as an unwanted but intentional alteration of a logic circuit or
design resulting in undesired circuit behavior. A clear understanding of Hardware
Trojans is quintessential for developing ICs used in highly secure applications.

3.1 Hardware Trojan Taxonomy

The Hardware Trojans (HT) can be inserted either in the design phase or during
fabrication. These inserted HT remain dormant during the normal course operation
of the IC for a long time, until they are triggered. Once HT is triggered, it will perform
the malicious action they are designed for. The classification of Trojans based on their
trigger mechanisms and their payloads is shown in Fig. 3 [3, 4, 6].
The Hardware Trojans as classified by Rajendran et al. in Fig. 4 provide us with a
good understanding of the effects of Trojans, the different stages of IC manufacturing
process, and the different locations of the IC which can be infected by Trojans [6].
The HT can be activated through internal or external mechanisms. Combinational and
sequential triggers belong to the class of internal triggers. The occurrence of a certain
set of values at internal nodes in the circuit simultaneously activates a combinational
trigger. A particular state of a set of internal registers or a specific word on the address
bus can be also used as a trigger. Sequential triggers are activated by a series of events.
The simplest trigger for sequential circuits is a counter. The HT is triggered when the
792 S. Rajendran et al.

Fig. 3 Trojan taxonomy based on trigger and payload mechanisms

Fig. 4 Hardware Trojan taxonomy

counter reaches its highest count. The external trigger mechanisms are applied from
outside the system. Activation of external triggers depends on user inputs like a reset
button or a power switch. An externally connected memory device can also act as an
external trigger. There is yet another set of Trojans which do not require any trigger
mechanism for activation. They are also known as ‘always on’ Trojans. Always on
Trojans function by making elusive changes to system specification, timing, etc.

4 Identification of Trojans

The insertion of a Trojan cannot be completely prevented. However, it can be limited

to a certain extent. The Hardware Trojan Detection techniques are classified as shown
in Fig. 5 [3]. It is very difficult to achieve a high degree of assurance that the probabil-
ity of a Hardware Trojan being present is zero. Generally, reverse engineering is the
Security of an IoT Network: A VLSI Point of View 793

Fig. 5 Trojan detection techniques

most preferred method to identify the Trojan; however, reverse engineering becomes
extremely difficult and time-consuming when we use it on complex modern IC. The
most common method used for identification of the Hardware Trojan is to compare
it with a golden reference. Though this method will not be of any significant help if
the Trojan is added prior to IC fabrication, in certain scenarios when the Hardware
Trojan is inserted in just a few IC’s, reverse engineering can be used. As a part of the
fingerprinting method, the reverse engineering process is used to identify a good IC
and using the details such as power consumption, temperature and electromagnetic
profiles, a reference is generated which is called a fingerprint. This fingerprint is then
used against all the IC of the batch as a comparison parameter to find Trojan. This
method sounds a feasible option to an extent to identify Trojans in some instances
[3, 5].
In order to avoid damaging the IC, techniques which leave the design unaltered
are preferred. These methods have better accuracy in Trojan detection. One of these
methods involves the insertion of dummy flip flops into the design to increase the
Hardware Trojan activity and therefore making it easier to detect the activity of the
Hardware Trojan. Adding extra gatekeeper logic along with some modified software,
which can monitor all writes to memory will help to identify the illegal writes on
which some action can be taken.
There are some suggested and used techniques which do not change the design
of the IC while detecting the Hardware Trojan. Adding reconfigurable logic
(DEFENSE) to the functional design is a real-time security monitor [7]. Once the
IC’s are fabricated, this logic is programmed with details regarding how the device
should behave. Any deviations can then be easily detected. Side-channel analysis is
another method which is used for Trojan detection. In this method, it is assumed that
the Trojan triggers itself and alter some important features of the IC like the power
consumed, the heat produced in different parts of the IC, the delay time (path delay).
Side-channel analysis is said to have the best rate of Hardware Trojan Detection as
it does not require the Trojan to be activated [3].
Another modification of the side-channel analysis is the current integration
method to identify the Trojans. This technique depends on the amount of current
794 S. Rajendran et al.

drawn in by the different part of the circuits. Some currents drawn can be so small
that they are assumed as an envelope of noise and therefore not detected during
measurement. However, this can be avoided by measuring the current locally and at
multiple ports in the IC design. These measurements are then compared to a golden
chip which is used as a reference and any deviations are used in the detection of
Hardware Trojans [5].

5 Basic Architecture of a Secure Processor

The traditional method used to improve security was accomplished with the help of
two processors. An embedded security processor is used in conjunction with main
processor to provide security as shown in Fig. 6. The embedded security processor
is controlled by the main processor. The work of the embedded security processor is
to evaluate the data it receives from the network interface before allowing external
network to access main processor [8].
The Internet of Things (IoT) network requires processors of small dimension with
low power consumption. In order to achieve this requirement, only one processor
is preferred instead of the traditional method of using two processors. The most
commonly used processors are ARM processors. The major reason for selecting
ARM processors is its performance and low power consumption [9].
These processors also face the risk of been affected by Hardware Trojans. The
ARM Cortex M4 processor architecture is shown in Fig. 7. The researches show

Fig. 6 Basic architecture of a secure processor

Security of an IoT Network: A VLSI Point of View 795

Fig. 7 ARM M4 cortex architecture [10]

that information in such processors can be leaked through radio frequency, optical,
power, and timing side channels and by interfaces such as JTAG and RS232. The most
prominent locations for detecting Hardware Trojans are memory units and clocking
units. As illustrated in Fig. 4, the processor is vulnerable to Hardware Trojans which
can be inserted at the design phase and abstraction level. Attack resistant features like
advanced memory management unit (MMU), redundant circuitry to thwart power
attack analysis and circuitry for fault detection need to be incorporated into the
design.

6 Implementation Issues

Basic implementation of hardware security is achieved through cryptographic algo-

rithms. Since key generation involves a great deal of computation, use of a low power
embedded device could be challenging. Battery capacity and storage limitations are
also issues that need to be tackled in order to develop a complete security solution.
Resource sharing which greatly helps to cut the hardware cost is yet another rea-
son which prevents physical separation between modules or partitions. Most of the
existing solutions are application-specific, and hence, there is no one golden solution
applicable to all systems.
796 S. Rajendran et al.

7 Counter Measures

Security solutions are always a trade-off between security, cost, performance, flex-
ibility, and power consumption. Solutions generally look into the optimization of
basic security functions like confidentiality, integrity, authentication, etc. and also
countermeasures against the security attacks [2]. The semiconductor manufacturing
requires a large investment; the importance of foundries has grown drastically, thus
increasing the exposure to thefts of masks, insertion of Hardware Trojans, and unau-
thorized excess fabrication. In order to avoid Trojan insertion at the fabrication level,
the authors in [11] propose a system in which the IP consumer needs to provide both
hardware specifications and a list of security-related properties. This has to be agreed
upon by the IP Consumer and the IP Producer. This idea is similar to the software
process proof-carrying code [12].
The designer also plays a vital role in avoiding insertion of Hardware Trojans by
making necessary architectural changes. A method developed by Hicks et al [13] uses
a Trojan countermeasure called BlueChip. It is a hybrid combination of hardware and
software. It includes an Untrusted Circuit Identification Algorithm along with a tool
set which identifies Trojan circuits during the verification or testing of the design.
Using a hybrid approach helps maximize processing efficiency while meeting the
design constraints. Deng et al. [14] proposed a method of using a hardware checksum
that is based on checking the authenticity of trusted hardware. As per the concept,
the hardware needs to send back a checksum within a stipulated time frame. If it does
not receive the expected checksum, it assumes the presence of a Hardware Trojan.
This method ensures the absence of Trojan post-fabrication process. However, this
process does not guarantee that no Trojans have been inserted during the specification,
design, verification, or manufacturing stages. Reconfigurable Architectures are of
great help in countering the Hardware Trojans. The best example of reconfigurable
architectures is FPGA. Implementation of secure processor is a novel method of
reduction of Hardware Trojans. As seen, the Trojans are inserted at various stages
such as design and fabrication of the IC. In FPGA, the hardware implementation
and design implementation are completely different. In FPGA systems, the design
is implemented at a later stage using a configuration bitstream. This process assures
us that the design can be implemented independently and it provides a high degree
of security from Hardware Trojans.
With the existing methods of Hardware Trojan identification, we still do not get
a completely secure IC. Another method which can be implemented to increase
the identification efficiency is by using combinations of identification techniques.
The basic concept of combinational identification technique is to use two or more
identification techniques together [15].
Security of an IoT Network: A VLSI Point of View 797

8 Conclusion

Internet of Things (IoT) with a strong network of hardware devices can be of utmost
importance for new economic business models. A perfectly secure solution is an
ideal requirement. However, such security systems are not easy to develop. Each and
every aspect of the network should be totally secure in order to achieve total security.
The algorithms available are very much secure from the cryptography point of view.
More emphasis is required on other elements of the network. In the IC level, more
attention needs to be given to the design, fabrication, and verification stages of the
IC. If these things are well addressed, then an ideally secure Internet of Things (IoT)
network can be achieved.

References

1. Syed A, Mary Lourde R (2016) Hardware security threats to DSP applications in an IoT
network. In: 2016 IEEE international symposium on nanoelectronic and information systems
(iNIS). IEEE, pp 62–66
2. Babar S, Stango A, Prasad N, Sen J, Prasad R (2011) Proposed embedded security frame-
work for internet of things (IoT). In: IEEE international conference on wireless communica-
tion, vehicular technology, information theory and aerospace & electronic systems technology
(Wireless VITAE), pp 1–5
3. Chakraborty RS, Narasimhan S, Bhunia S (2010) Hardware Trojan: threats and emerging
solutions. [Link]
4. Chakraborty RS, Bhunia S (2009) Security against hardware Trojan through a novel application
of design obfuscation. In: Proceedings of the 2009 international conference on computer-aided
design, pp 113–116
5. Wang X, Salmani H, Tehranipoor M, Plusquellic J (2008) Hardware Trojan detection and
isolation using current integration and localized current analysis
6. Rajendran J, Gavas E, Jimenez J, Padman V, Karri R (2010) Towards a comprehensive and
systematic classification of hardware Trojans. In: Proceedings of 2010 IEEE international
symposium on circuits and systems (ISCAS), pp 1871–1874
7. Abramovici M, Bradley P (2009) Integrated circuit security: new threats and solutions. In:
Proceedings of the 5th annual workshop on cyber security and information intelligence research:
cyber security and information intelligence challenges and strategies, CSIIRW ’09, ACM, New
York, NY, USA, pp 55:1–55:3
8. McKelvey MA (1999) Embedded security processor. International Business Machines Corpo-
ration. Patent Number: 5,896,499 20, Apr 1999
9. Suh G, O’Donnell CW, Devadas S (2005) AEGIS: a single-chip secure processor. Inf Secur
Tech Rep 10(2):63–73
10. ARM Processors. [Link]/lit/er/spmz637/[Link]
11. Love E, Jin Y, Makris Y (2011) Enhancing security via provably trustworthy hardware
intellectual property. In: IEEE international symposium on hardware-oriented security and
trust
12. Necula GC (1997) Proof-carrying code. In: Proceedings of the 24th ACM SIGPLAN-SIGACT
symposium on principles of programming languages, pp 106–119
13. Hicks M, Finnicum M, King ST, Martin MMK, Smith JM (2010) Overcoming an untrusted com-
puting base: detecting and removing malicious hardware automatically. In: IEEE symposium
on security and privacy, pp 159–172
798 S. Rajendran et al.

14. Deng DY, Chan AH, Suh GE (2009) Hardware authentication leveraging performance limits
in detailed simulations and emulations. In: Proceedings of the 46th annual design automation
conference, DAC ’09, ACM, New York, NY, USA, pp 682–687
15. Beaumont M, Hopkins B, Newby T (2011) Hardware Trojans-prevention, detection, counter
measures (a literature review) (No. DSTO-TN-1012). Defence Science and Technology
Organisation Edinburgh (Australia) Command Control Communications and Intelligence Div

Sreeja Rajendran is currently doing her Ph.D. at BITS Pilani

Dubai Campus (BPDC), UAE. She has a Masters degree in
Microelectronics from BPDC (2015) and [Link] in Electron-
ics and Instrumentation from Cochin University of Science and
Technology, India. Her research interests include microelec-
tronic circuits, low power circuit design and FinFET analysis
and testing.

Azhar Syed has [Link] in Electronics and Telecommunica-

tions from Mumbai University (2013) and Master of Engineer-
ing from BITS Pilani, Dubai Campus in Microelectronics in
2017. His current research interests include VLSI Design, Secu-
rity of processors and Internet of Things.

Dr. Mary Lourde, R did [Link] in Electrical Engineering

from Kerela University, India in 1983 and M. Tech in Electron-
ics from Cochin University of Science & Technology, India in
1987. She is awarded the Ph.D. Degree in Electrical Engineer-
ing from Indian Institute of Science, Bangalore, India for the
thesis “Design and Analysis of Digital Controllers for high per-
formance Sensorless PMSM Servo drives” in 1998. She was a
faculty at Department of Electronics, CUSAT, India since 1990.
Presently with Birla Institute of Technology & Science Dubai
Campus (BPDC), working as Associate Professor in EEE. Her
current area of research includes VLSI Desgin, Signal Process-
ing and its applications and Power Electronics & Drives. She is
an MIEEE, LMIETE, LMISTE.
Anomaly Detection Techniques in Data
Mining—A Review

K. N. Lakshmi, N. Neema, N. Mohammed Muddasir and M. V. Prashanth

Abstract Detection is one of the biggest threats to the organization. The detection of
abnormal behaviors is one of the most difficult tasks for administrators of information
systems (IS). Anomaly behavior is defined as any behavior that deviates from normal
within or outside the organization IS, including insider attacks and any behavior that
threatens the confidentiality, integrity, and availability of information systems for
organizations. The detection of anomalies is extremely important to prevent and
reduce illegal activities and to provide an effective emergency response.

Keywords Anomaly · Detection · Data mining · Intrusion detection

1 Introduction

The amount of data being generated and stored in growing exponentially, due in
huge part to the continued advances in technology. Data mining is often accustomed
to extracting helpful information from the data that surround us. Data mining is
the method of analyzing data from totally different views. Data warehouse could
be a relative database that is designed for question and analysis instead of dealings
process. In further to the current, it additionally includes extraction, transportation,
transformation, and loading (ETL) solution. Online analytical processing (OLAP)
engines consumer analysis tools and different applications that manage the method
of gathering knowledge and delivering it to users. It additionally involves subject
orienting, integrated, nonvolatilizable, and time variant characteristics. Forms of
data processing techniques or strategies are helpful to search out information that is
simply understood in large data sets [1].

K. N. Lakshmi (B) · N. Neema · N. Mohammed Muddasir · M. V. Prashanth

Department of Information Science and Engineering, Vidyavardhaka College of Engineering,
Mysuru, Karnataka, India
e-mail: lakshmikn63@[Link]

© Springer Nature Singapore Pte Ltd. 2020 799

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
800 K. N. Lakshmi et al.

Intrusion detection using data mining (IDDM) techniques aims to see the fea-
sibility and effectiveness of knowledge data mining techniques in real-time intru-
sion detection and produces solutions for this anomalies behavior, so as to develop
defense computing networks. To safeguard networks, intrusion detection systems
aim to acknowledge attacks with two primary needs high detection and low warning
rate. As attacks manifest themselves in two classes, those who are noted and peo-
ple that are seen antecedently. Data warehousing is recognized as a great tool for
extracting regularities in knowledge and so been the target of some investigations
for its use in intrusion detection [2].
The IDDM focuses on the utilization of data mining within the context, by manu-
facturing descriptions of network data using this information for derivation analysis.
Variety of existing technologies is offered for this purpose a number of that are eval-
uated as a part of the project. This method is performed by meta-mining techniques
that characterize amendment between network data descriptions; the system will
manufacture applicable notices. The end result of the IDDM project is the talents to
characterize network knowledge and to sight variations in these characteristics over
time. Combining this capability with tools that either acknowledge existing attack
patterns or operate equally to IDDM, it strengthens the power to intrusion detection
professionals to acknowledge and probably react to unwanted violations to network
operations [2]. Developing the anomaly detection system needs various knowledge
for training and testing functions.

2 Literature Review

In [2], Abraham has proposed intrusion detection using data mining techniques.
Combining data mining algorithms with technologies earned close to real-time oper-
ation is the basic result. Increase in a number of alarms generated due to changes in
rules to set overtime has adversely affected rules set stability is the application.
Dasgupta et al. [3] suggested multidimensional data anomaly detection using a
negative selection algorithm. The result is that PCA reduces a set of five-dimensional
data to two-dimensional and one-dimensional. The main advantage is the PCA’s data
reduction. It usually destroys some information within the method, which is why the
system has not detected all the anomalous data records.
Catterson et al. [4] proposed anomaly detection for transformer monitoring in
multi-variable data. A type of collective anomaly detection has been applied to a
single parameter that could improve the data on engineering anomalies. The main
advantage is that the CAD technique was used for online transformer monitoring to
detect common behaviors.
Patch and Park [5] proposed a summary of the techniques of detection. The result
is anomaly detection systems, a set of intrusion detection systems, which model
the network behavior which enables them to find both known and unknowns very
efficiently. It is often the main advantage that anomaly detection systems and hybrid
intrusion detection systems are widely used.
Anomaly Detection Techniques in Data Mining—A Review 801

Ye et al. [6] proposed that probabilistic intrusion detection techniques have been
proposed and supported by computer audit data. As a result, a series of studies on
the probabilistic properties of activity data are presented in an information system to
detect intrusions into the information system. It demonstrates the advantage of the
frequency properties of multiple events.
Anna and Krzysztof [7] proposed the detection of anomalies in data storage:
the petrol station simulator. This paper introduces the simulator’s foundations with
the result. This discussion presents future work in the field of data collection and
materialization in the warehouse data stream.
Radon et al. [8] proposed that alarm reduction has been proposed in maritime
detection of anomalies. AIS technology offers an enormous amount of ship movement
information center for maritime research and experimentation (CMRE) technology
used in close to real time. It also encourages misalarms to be reduced.
Siraj et al. [9] proposed that a smart intrusion detection system architecture has
been proposed in the data fusion intrusion sensor. The decision engine effectively
fused information in the detection database and the results of the decision-making
were also correctly reported in the alert database.
Siaterlis and Maglaris [10] proposed multi-sensor data fusion for detection of
DOS. The main advantage is that the data fusion rate increases, and the false alarm
rate decreases.
Balakrishna and Rama Ganesh [11] proposed the detection of anomalies, and SQL
prepares data sets for analysis of data mining. The data manipulation pivot of the
operator is easy to work out for a wide set of values.
Rameshkumar et al. [12] proposed that intrusion detection has been proposed as
a text-based approach to mining. This makes the computation even in binary form
accurate. The similarity values vary from 0 to 1.
In [13], Bebel et al. have proposed a formal approach to modeling a multi-
version data warehouse. Development of prototype MVDW system is the result.
The application involves a multi-version data warehouse that is composed of set of
its versions.

3 Comparison of Different Anomaly Detection Techniques

Table 1 gives us the picturesque idea of the detection techniques used by various
authors in the field of anomaly detection techniques using data warehouse. It also
gives the list of recommendations which we thought could have been implemented
in the system in the future.
802 K. N. Lakshmi et al.

Table 1 Comparison of different anomaly detection techniques

Reference Approach Method Parameter Result Advantage
No.
[2] Intrusion Data mining Data mining Usage of data Increase in no.
detection technique parameters mining of alarms
technique algorithms generation
[3] Immunity-based Negative Multidimensional A result set The data
approach selection data obtained from reduction by
algorithm five-dimensional PCA
data reduce by
PCA to two
dimension
[4] CAD approach Conditional Partial Enhance the CAD
anomaly discharged data information technique
detection about applied to
anomalies. online
monitoring
[5] Survey of the Network-based Intrusion Model the Hybrid
anomaly on anomaly detection system network intrusion
detection detection behavior which systems are
system. enables the widely
effective adopted
detection
[6] Pattern Occurrence, Anomaly Presents a series It shows the
detection frequency detecting of studies on frequency
approach parameters probabilistic property of
properties multiple
events
[7] Detection Data extraction Data Simulator Sensor
approach technique warehousing produces data miscalibration,
parameters sets with detection of
specified anomalies
anomalies
[8] Center for Automatic Kinematic data Verification for Reduces the
maritime identification false alarm false alarms
research and system reduction and motivates
experimentation technique in false alarm
approach detection
[9] Intelligent Computer Fuzzy cognitive Intrusion Effective
intrusion research maps, fuzzy rule detection database
detection security base parameters detection
approach technique
[10] DoS approach Methods of Quantitative To detect Increases the
data fusion measurement flooding attacks DoS detection
and to use data rate and
fusion in a decreases the
detection false alarm
technique rate
[11] SQL code PCA technique SQL parameters Anomaly To compute
generation detection for wide sets of
approach data mining values for data
analysis manipulating
operator
(continued)
Anomaly Detection Techniques in Data Mining—A Review 803

Table 1 (continued)
Reference Approach Method Parameter Result Advantage
No.
[12] Mining-based Intrusion IDS parameters Results in Makes the
approach detection reducing computation
technique training testes accurate even
in the binary
form
[13] Schema data Online Schema Developed a Multi-version
versioning analytical evaluation prototype data
approach processing MVDW system warehouse
methods composed of
set of its
versions

4 Conclusion

Data fusion can improve the performance and reliability of systems in the art and
science of data storage. In addition, the fusion process and data mining operations are
discussed. To protect the availability, confidentiality, security, and integrity of critical
information infrastructures, IDS is designed. The current state of the art of ID systems
is relatively primitive in relation to the recent explosion in computer communications,
cyberspace, and electronic equipment. Organizations are fully aware that technology
detection can be very important information that flows with optional and inhibitory
technical factors. The approach to anomaly detection requires the combination of
different disciplines, such as artificial intelligence and statistics. It is applied directly
to detections of intrusion and attack.
It also discusses the research problems to be addressed in the detection of intru-
sions in different processing techniques. It also tells us about the sequence of steps to
be taken to improve the efficiency and performance of the mechanism for intrusion
detection. Anomaly detection systems are now widely used in all fields due to their
efficient performance.

Acknowledgements The authors express gratitude toward the assistance provided by Accendere
Knowledge Management Services Pvt. Ltd. in preparing the manuscripts. We also thank our mentors
and faculty members who guided us throughout the research and helped us in achieving the desired
results.

References

1. Mahi S (2017) Introduction to data mining and data warehouse. Int J Adv Res Comput Sci
8(4):398–400
2. Abraham T (2001) IDDM: intrusion detection using data mining techniques 35
804 K. N. Lakshmi et al.

3. Dasgupta D, Sumi Majumdar N, Majumdar NS (2002) Anomaly detection in multidimensional

data using negative selection algorithm. In: Proceedings of the 2002 congress on evolutionary
computation. CEC’02, pp 1039–1044
4. Catterson VM, McArthur SDJ, Moss G (2010) Online conditional anomaly detection in
multivariate data for transformer monitoring. IEEE Trans Power Deliv 25(4):2556–2564
5. Patcha A, Park JM (2007) An overview of anomaly detection techniques: existing solutions
and latest technological trends. Comput Netw 51(12):3448–3470
6. Ye N, Li X, Chen Q, Emran SM, Xu M (2001) Probabilistic techniques for intrusion detection
based on computer audit data. IEEE Trans Syst Man, Cybern Part A Syst Humans. 31(4):266–
274
7. Anna G, Krzysztof P (2015) Anomaly detection in data streams: the petrol station simulator.
In: International conference: beyond databases, architectures and structures, pp 727–736
8. Radon AN, Wang K, Glasser U, Wehn H, Westwell-Roper A (2015) Contextual verification for
false alarm reduction in maritime anomaly detection. In: 2015 IEEE International Conference
on big data. IEEE Big Data 2015, pp 1123–1133
9. Siraj A, Vaughn RB, Bridges SM (2004) Intrusion sensor data fusion in an intelligent intru-
sion detection system architecture. In: Proceedings of the 37th annual Hawaii international
conference on system sciences, 2004, pp 1–10
10. Siaterlis C, Maglaris B (2004) Towards multisensor data fusion for DoS detection. In:
Proceedings of the 2004 ACM symposium on applied computing—SAC’04, p 439
11. Balakrishna PV, Rame Ganesh B (2014) Anomaly detection and SQL prepare data sets for data
mining analysis. Int J Comput Sci Inf Technol 5:6551–6555
12. RajeshKumar G, Mangathayaru N, Narsimha G (2016) Intrusion detection—a text mining
based approach. Special issue on Computing Applications and Data Mining. Int J Comput Sci
Inf Secur 14:76–88
13. Bȩbel B, Królikowski Z, Wrembel R (2006) Formal approach to modelling a multiversion data
warehouse. Bull Polish Acad Sci Tech Sci 54(1):51–62
MAIC: A Proficient Agricultural
Monitoring and Alerting System Using
IoT in Cloud Platform

A. Srilakshmi, K. Geetha and D. Harini

Abstract The Internet of things plays a crucial role in all the fields such as agri-
culture, health care, wearable’s, industrial IoT (IIoT), retail, smart city and smart
home. In every field, sensors are used to monitor the device values and provided to
alert to the decision-making process. In the proposed Monitoring and Alerting using
IoT in Cloud platform (MAIC) work, we predict whether the sensor is sending val-
ues based on the predefined time schedule. Changes in the agriculture environment
have to be captured by these sensors and should be updated in a regular phase. If
not, something would have happened to sensors, and such anomaly can be detected
easily with the help of ThingSpeak open-source IoT cloud and its apps. In addition,
sensor values (like temperature, moisture, humidity, etc.) need to be validated to
reflect the correct measures as per the environmental circumstances. If one of these
anomalies is detected, smart alerts could be sent to mobile or Twitter or an e-mail
account possessed by the individual involved in agriculture activities. The sensor
values captured in cloud are processed, and if any drastic change is detected, instant
alert is sent to the individuals. These alert messages are authenticated by performing
appropriate tweet analysis using R. The MAIC perms an intelligent monitoring of
sensor values, and it gives entire belief to the experts who are involved in the system.

Keywords Internet of things · MAIC · Sensors · Thingspeak · Twitter

1 Introduction

The Internet of things is supported by many cloud platforms such as Amazon cloud,
IBM bluemix, Microsoft Azure, ThingSpeak, Thingsworx, dream grove Google
cloud, Firebase, Salesforce and many different clouds. All the cloud platforms are
used for performing data analytics, monitoring, performance analysis and Bot ser-
vice, etc. For experimental purpose, here, ThingSpeak is chosen. The sensors used
here are supported by dexter industries. The grove shield support is given for Arduino

A. Srilakshmi (B) · K. Geetha · D. Harini

SASTRA Deemed to be University, Thanjavur, India
e-mail: srilakshmi@[Link]

© Springer Nature Singapore Pte Ltd. 2020 805

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
806 A. Srilakshmi et al.

as well as Raspberry Pi, and this GrovePi can be connected with multiple sensors
without breadboard and many external wires. If there is a cloud intervention, first step
is to connect any device such as Arduino and Raspberry Pi to the cloud and analyze
and act accordingly. The streaming data can be captured on the cloud, and smart
alerts are sent. The cloud has capability to store tera and petabytes of data streaming
every day. The output of the streaming data can be further exported either on hourly
basis or by day basis. With the data set, further analytics can be done where the user
need not sit in front of the device until hardware-related anomaly is detected.
Nowadays, this technique is used in almost all the fileds such as automobile,
smart cities, health care, bio-informatics, sensing system in educational fields and
retail. The sensors are used because more accurate and efficient results could be taken
based on the threshold limit. The sensor’s communication can be made more efficient
with the help of Message Queuing Telemetry Transport (MQTT) and Constrained
Application Protocols (COAP). Even the IoT data which is called smart data can
be secured with the help of cloud security, where each cloud offers good security
authentication, authorization, encryption, reduces cyber risk, etc. So, the users can
trust the data which will be secured on the cloud by supporting trust ability. Timely
data can be visualized and no need to retain separate record as every sensed smart
data can be exported in the form of either JSON or CSV format. This format can be
used for further analytics and visualization purpose.

2 Related Work

The study has been made before implementing MAIC with the following list of
papers. Corbellinib et al. [1] implemented a cloud-based sensor monitoring for envi-
ronment. The environmental factors are monitored, and sensed data is stored in
mobile app through Cloudino and Azure cloud [2]. The wearable sensors are used to
monitor patients health, and data are simply stored in cloud. The sensor is embedded
in human’s shirt; the values are collected and stored to own cloud through WiFi. Kim
et al. [3] shows the application of IoT in smart city how IoT plays a major role with
traffic light monitoring, minimizing the pollution smart route finder applications etc.,
and it shows the basic applications in smart city as a survey report [4]. A detailed
survey of all the cloud platforms is done and its domains related to application, device
system management, data management and analytics feasibilities are detailed [5].
The smart data in a huge volume is collected, and that streaming data is sent to cloud
database, and fuzzy rules are used for performing analysis [6]. A decision-based
system has been designed to detect light blight disease on potato crop. The complete
weather information is monitored. It uses Xively cloud to display the sensor values.
It uses Ubidots to display and monitor the sensor values [7]. The IoT detects the
machine status, and it is monitored for the unused resources which can be used later.
It uses IoT technologies and cloud to display sensor data. It also used RFID tag for
communication [8]. The status of chronic skin wound is automated, and it is moni-
tored in a wireless manner. The sensed values are displayed in a monitor as an output
MAIC: A Proficient Agricultural Monitoring and Alerting System … 807

without any cloud intervention. Udawant et al. [9] discusses the implementation of
heart rate sensor and blood pressure. ECG sensors are used to monitor patient health
and to send SMS to hospital database through cloud. This is done to avoid traffic
jams caused during peak time to save patient’s life. The messages are sent through
GPRS to the cloud [10]. The water level is monitored for overflow, and immediate
alerts are sent to the cloud. No specific cloud is mentioned [11]. Agriculture plays
a vital role in all farmers field, and it is important for Indian economy. Irrigation is
a major problem in agriculture. Because farmers waste lot of time and do not know
how much water to irrigate and at what time, this issue has been fixed using sensors.
The sensors sense the weather condition very precisely so that the irrigation is done
at right time [12]. The Iot-based forest monitoring system has been done to remotely
monitor the cutting of high-value trees in forest using WSN technique. Finally, it
is equipped with GPRS, and data is collected and sent to the mobile app [13]. An
Iot-based decision is made to automate irrigation using machine learning techniques.

3 Proposed Architecture

The MAIC architecture well suits for any applications. This is done for agricultural
purpose. It uses three different clouds supported for IoT. Each cloud has different
types of services to support the user.
The proposed architecture shows the MAIC implementation that has been done
in this paper. The incoming sensor values are collected by the cloud, and alerts are
sent using IFTTT maker service. It enables the device to communicate with each
other through a maker service usually webhooks are used for if condition.

4 Intelligent Monitoring

4.1 Connecting Sensors

In this implementation, few sensors are connected to grove shield which is placed at
the top of the Raspberry Pi board
Connecting sensor to grove shield is too easy and simple. Connecting any sensor
to grove shield and simply specifying the port number in the code will make the
code to run. In Fig. 1, the topmost part is the grove shield which is connected to
Raspberry Pi. Once the connection has been made, sensor starts reading the value.
Sensors should be fixed to correct port according to the code that has been written.
Once sensor starts displaying the values, monitoring process could be started. Any
type and number of sensors can be connected to either Arduino or Raspberry Pi with
grove. The temperature sensor for, for example, will be operating at the range of −40
to 125 °C. The accuracy range of the sensor at its operating level is 1.5 °C. Similarly,
808 A. Srilakshmi et al.

Fig. 1 MAIC architecture

Table 1 Sensor values of

Sensor_Value Condition
moisture
0 Open air
50–250 Dry soil
250–600 Humid soil
600 and above Floating in water

moisture sensor is connected to another port of grove kit. Then, the minimum, typical
and maximum moisture levels are displayed according to the soil condition. The
below function senses and displays the sensor values (Table 1).
Sensor_value = 0
analogRead(Sensor_value);
The sensor values may not be constant for all the time; it will be changing based
on the environmental conditions.
The sensors have to be covered with waterproofs to sense precisely.
By monitoring the sensor values, any type of anomaly can be detected such as
false alerts or if something happened to sensor. Anamoly here means, whether the
sensor has no update for long time, sensor may also be disconnected from the port.
Even if it is disconnected, it sends the value as 0. This 0 is different from moisture
sensor’s (Sensor_value’s 0).

4.2 ThingSpeak Monitoring Service

The sensor values are monitored via ThingSpeak which directly supports MAT-
LAB for performing analysis. It is an open-source application used for tracking any
devices, and smart alerts are sent. First an account has to be created in ThingSpeak
MAIC: A Proficient Agricultural Monitoring and Alerting System … 809

and login using MATLAB account. The first step is the channel creation with rel-
evant fields. Based upon the number of sensors, same number of fields has to be
created. Figures 2 and 3 show the values of temperature and soil moisture sensor.
After creating a channel, the API keys can be used to read and write data into the
channel. Both Figs. 2 and 3 show the private view chart of the channel. Here, the
channel is not updated frequently. For every twenty minutes, the channel is updated.
The below figure shows the specifications of temperature and moisture sensors
(Figs. 4 and 5).
The temperature level which is displayed in Fig. 1 is normal, whereas the moisture
level is high. If the moist is between 0 and 300, then the soil is dry. Specifically, if the
moist level is 0, 1, 2, 3, etc., then the sensor is in free space. For the range between
300 (min) and 700 (max), the soil is humid and enough for irrigating plants.
As shown in Fig. 2, if the range exceeds 700, then the sensor is floating in water.
In this case, it is 761. Once the chart has been displayed, MATLAB visualization can
be made to analyze the temperature and moisture variation.
Now based on the temperature, analysis can be done on one of the following.
Case (i): Sending smart alerts from ThingSpeak to Twitter, or by short messaging
service or to e-mail account.
Case (ii): If there is no update from the sensor for some time period.

Fig. 2 Raspberry Pi GrovePi

S. No Specifications Temperature Moisture

1. Voltage 3.3 ≥ X ≤5V 3.3 ≥ X ≤5V
2. Current 0 ≥ X ≤ 35mA 0 ≥ X ≤ 35Ma

Fig. 3 MAIC specifications

810 A. Srilakshmi et al.

Fig. 4 Temperature values

Fig. 5 Moisture values

Case I

A ThingTweet app is created from ThingSpeak, refer Fig. A, and it is linked to

Twitter account. By authorizing the app, an API key is generated which is later used
for any other purposes. Next React app is created, and it is set to react on particular
condition, i.e., if temperature is above 35 °C then the phrase too hot is tweeted. The
React app is linked to ThingTweet by adding the twit account into it. The condition
type of React app is set to numeric, and test frequency is set to data insertion because
if the condition is once met the alert can be sent immediately. Based upon the sensor,
the condition type is chosen as either numeric or string or status, etc. Figure 2.1
shows the alert message of Twitter. Once, it is greater than 35 °C and moderate in
the evening which is 28 °C (Fig. 6).

Case II

In the second case, if there are no updates from the sensor, it is identified and notifi-
cation is sent to the user. If the channel or channel ID has no update for last 15 min
it tweets that there is no update. Note that the same user ID is used for performing
future Twitter analysis.
MAIC: A Proficient Agricultural Monitoring and Alerting System … 811

Fig. 6 MAIC alert system 1

Figure 7 shows the Twitter message when there is no update from the sensor for
ling time. The React app is made to run to take action for ThingTweet.

ThingHTTP_APIKEYS =
struct('To_Trigger_TalkBack','LOO6RKV2VZWPH5
7Z','To_Trigger_ThingTweet','K2BDUJ30Y7AI37J L');% Your ThingHTTP app API keys
url =
'[Link]
end_request?api_key=HBVE4CS981Z9XR5O';
Trigger_TalkBack =
webread(url,'LOO6RKV2VZWPH57Z',ThingHTTP_API
KEYS.To_Trigger_TalkBack) %Trigger TalkBack
via ThingHTTP
Trigger_ThingTweet =
webread(url,'K2BDUJ30Y7AI37JL',ThingHTTP_API KEYS.To_Trigger_ThingTweet)
%Trigger ThingTweet via ThingHTTP.

Fig. 7 MAIC alert system 2

812 A. Srilakshmi et al.

4.3 Sending Alerts Using IFTTT Maker Service

Similar alerts could be sent using the If This Then That (IFTTT) platform. An applet
service is created for if condition by choosing a Web hooks app and getting the event
name. For the else condition, the alert app can be chosen from the one listed in the
applet page.

E:g A if{Web hook

{event = the Temperature is too cool} 1
}
then
alert using either
{
Messages (sms) or twitter or face book 2
}

Once a maker service has been created, it generates a unique URL for every
request that has been created. For example, below is the URL for this maker service.
This is saved for further use in ThingSpeak.
[Link]
qzeocpEXGgkVixGLlvgtxtnpoM.
To make the above code to work, the URL is configured in ThingSpeak ThingHttp.
The HTTP method used here is GET method which accesses the specified URL.
Now, when the React app is configured with ThingHTTP, it starts sending message.
The below Fig. 8 shows the messages which are received by the maker service, and
the event named TooCool is executed.
Similar alert could be sent to Facebook and e-mail simply by editing the example
A which is named number 2.

Fig. 8 MAIC alert system 2

MAIC: A Proficient Agricultural Monitoring and Alerting System … 813

5 Monitoring Using Firebase Cloud

Firebase is a mobile and Web application platform which is supported by Google. It

performs many cloud activities and API functions. It also plays a role in Internet of
things for displaying the sensor values as real-time database. The real-time streaming
database values are displayed in the database, and it is stored separately.
The configuration between Firebase and Raspberry Pi goes like this:

{
config=’apikey’,’authDomain’,’databaseURL’, ’storageBucket’
config = {
"apiKey":
"AIzaSyD2bULaxHyZA0RSxksWkz0FJTmShM5eQdM",
"authDomain": "[Link]",
"databaseURL":"[Link]
storageBucket": "[Link]"
}

Once the configuration information is verified, the Python code will be success-
fully connected to Firebase. The pyrebase. initialize_app(config) will be initialized
to a variable, and this variable is required to update the data bucket (Fig. 9).
The difference between ThingSpeak monitored values and Firebase values is
shown in Fig. 10, i.e., the ThingSpeak displays the sensor values at every second as
you can see in Fig. 4, the sensor values starting at the time of 10.07 and continuously
printing the values. But in Firebase, the streaming values are updated only if there
is change in sensor value as you can see in Fig. 4 that none of the moisture or
temperature value is constant either one is changing and hence the data is uploaded

Fig. 9 Querying MAIC

814 A. Srilakshmi et al.

FIREBASE CLOUD THINGSPEAK

Fig. 10 Comparison between firebase and ThingSpeak

to the cloud. It is is a real-time database; it takes just a fraction of second to update

the streaming data to cloud whereas ThingSpeak takes 10–11 s to upload the data
in the graph after running the code. The sensor data can be exported in the form of
JSON data in Firebase and CSV/JSON/XML in ThingSpeak. The Firebase cloud can
display the streaming view of values using stream view. In the future, it is easy to
retrieve the moisture data with database options, i.e., to print in either ascending or
descending sensor values. It is also easy to retrieve by field using view by field. The
below diagram shows the sensor value in the collection field.
The value 250 represents that the soil is humid. Based on the value and the type
of plantation, the expert decides when and how much amount of water has to be
irrigated. Irrigation is the main source for any agricultural fields
The below table shows the comparison between two cloud techniques. The Fire-
base shows the streaming data, whereas the ThingSpeak detects the value only after
20 s from connection. Firebase has no inbuilt support to MATLAB analysis and
MAIC: A Proficient Agricultural Monitoring and Alerting System … 815

visualization as ThingSpeak does. But it has good machine learning kit, and it can be
easily connected to android app. For every activity, an Android app is created so that
user can easily monitor from the app itself. IoS is also supported for IoS users. The
cloud-based code for Raspberry Pi should be connected and kept running, hence to
see the values in cloud. If the Pi device is disconnected, the user or expert may have
difficulty in interpreting. Even if the disconnection of sensor happens, such alert can
also be sent to the user, so there is no chance to get any incorrect value or missing
sensor values. The smart sensor data is always evolving from the device.
In Firebase, it is possible to add and remove fields at run time. The Firebase does
not support MATLAB, and it helps in performing the machine learning algorithms
and statistical analysis can be done. The Firebase shows the updates only if there is
a change in the sensing value of at least one sensor connected to cloud.

6 The [Link] Cloud

The [Link] cloud also displays the streaming data similar to Firebase.
#define USERNAME “ggg@[Link]”
#define DEVICE_ID “temp1”
#define DEVICE_CREDENTIAL “12345678”
#define SSID “Agri”
#define SSID_PASSWORD “12345678” ThingerWifi thing(USERNAME,
DEVICE_ID,
DEVICE_CREDENTIAL);
The above one connects to the [Link] cloud. Similar tweets can also be sent
to [Link] cloud. The [Link] is a public and open-source cloud platform for
IoT and free of cost. It enables the user to connect the IoT devices and monitor the
sensor values. The cloud also sends alert to Twitter based on the condition. It also
displays the number of devices connected to its dashboards and data buckets. It gives
a complete statistics of the implementation and works efficiently with Arduino and
Sigfox.

7 Twitter Analysis

The sensor device is intended to send the sensor data if there is no update from the
device for past x minutes or n hour. This itself detects the anomaly from the device,
and if there is throughout same values for more than x minutes, the next anomaly is
cached and the response is sent either in the form of tweets or mobile SMS.
To ensure that the anomaly is exactly from ThingSpeak or Firebase, the tweet anal-
ysis has been made. The security parameters are accessed from the Twitter account
such as consumer key, Consumer secret, Access token and Access secret.
816 A. Srilakshmi et al.

The OAuth information is sent to setup_twitter_oauth() function as arguments.

Then, it is possible to access the Twitter account regarding the number of tweets and
its origination from the Web page. The R tool is used for performing the analysis, and
the tweets are for a particular user. The user account which is created for ThingSpeak
uses tweet account of any person to whom the alert has to be sent. As shown in the
example code, the user account ranjith85914681 is also there in ThingSpeak Twitter.
The user Id is displayed in R code to ensure the identity of the user.

consumer_key <- '53Udm77XRtcVgDkWGay7UQT'

> consumer_secret <-'X2FydARqOxxxxxxxybjuWg1qQ0nFtrH4RHkI3vMddKIF h'
> access_token <- '99675463453567557-271k2HFLaeZBOClF3wNVcvblCeoHKbf'
> access_secret 'CqoZNPQMs8o6CZSDXGDFXFGDFWMkicr1bAFU5yR eOYOLlf4LBJC'
> setup_twitter_oauth(consumer_key, consumer_secret, access_token, access_secret)

Once the information are authentical, the tweets can be collected. The
head([Link]) gives the top tweets of the user. It requires packages like Twitter,
ROAuth and the same libraries.

Thus, the sample tweets are shown, and this is needed only to ensure that the
tweet alerts are only from the device connected with cloud-like ThingSpeak or
[Link] or Firebase. Even a normal person may also end a casual message saying
that it is too hot or cold in my place and to differentiate from that tweet analysis is
done.
MAIC: A Proficient Agricultural Monitoring and Alerting System … 817

8 Conclusion and Future Work

The MAIC implementation helps in effectively monitoring the sensor device

remotely. Sometimes, the sensor may be disconnected then also it is able to send
alerts to the system with zero values. The expert might not have any knowledge
about this. If the temperature of the agriculture field or moisture of the field is zero,
then such contiguous zero is detected and written to ThingSpeak and IFTTT saying
that if the temperature is zero throughout or same for long time or no update for long
time or too cool or too hot or too moist alerts could to send to either mail or SMS
or tweets based on the expertise’s source usability. The MAIC can be implemented
with any other cloud technologies. Similarly, MQTT message queuing could be used
to manage communication between IoT devices. It is lightweight protocol which is
able to publish and subscribe data between IoT devices where two devices do not
know the identity of each other.
In the future, the same alert could be sent via other standard clouds like Microsoft
Azure, IBM, Amazon Google Cloud and Salesforce to find different monitoring
techniques and perform various machine learning techniques and analytics and finally
combining everything and making as an Android app for farmers.

Acknowledgements The work projected in this paper was implemented and tested using Raspberry
Pi with grove kit provided at Technology Business Incubator (TBI), Sastra Deemed to be University.
We would like to acknowledge the facility extended by TBI for IoT.

References

1. Corbellinib S, Di Franciaa E, Grassinia S, Iannuccib L, Lombardob L, Parvisb M (2018) Cloud

based sensor network for environmental monitoring. Measurement 118:354–361
2. Leu F–Y, Ko C–Y, You I, Raymond Choo K–K, Ho C–L (2018) A smartphone-based wearable
sensors for monitoring real-time physiological data. Comput Electr Eng
3. Kim T, Ramos C, Mohammed S (2017) Smart city and IoT. Future Gener Comput Syst 76:159–
162
4. Ray PP (2016) A survey of IoT cloud platforms 1(1–2):35–46
5. Kumar PM, Lokesh S, Varatharajan S, Babu GC, Parthasarathy P (2018) Cloud and IoT based
disease prediction and diagnosis system for healthcare using Fuzzy neural classifier. Future
Gener Comput Syst 86:527–534
6. Foughali K, Fathallah K, Frihida A (2018) Using cloud IOT for disease prevention in precision
agriculture Procedia Comput Sci 130:575–582
7. Zhong RY, Wang L, Xu X (2017) An IoT-enabled real-time machine status monitoring approach
for cloud manufacturing. Procedia CIRP 63:709–714
8. Pal A, Goswami D, Cuellar HE, Castro B Kuang S, Martinez RV (2018) Early detection
and monitoring of chronic wounds using low-cost, omniphobic paper-based smart bandages.
Biosens Bioelectron 117:696–705
9. Udawant O, Thombare N, Chauhan D, Hadke A, Waghole D (2017) Smart ambulance system
using IoT. In: International Conference on Big Data, IoT and Data Science (BID), pp 171–176
10. Perumal T, Sulaiman MN, Leong CY (2015) Internet of Things (IoT) enabled water monitoring
system. In: IEEE 4th global conference on consumer electronics (GCCE), pp 86–87
818 A. Srilakshmi et al.

11. Nageswara Rao R, Sridhar B (2018) IoT based smart crop-field monitoring and automation
irrigation system. In: 2nd international conference on inventive systems and control (ICISC),
pp 478–483
12. Sharma Ak, Ansari MFR, Siddiqui MF, Baig MA (2017) IoT enabled forest fire detection and
online monitoring system. Int J Curr Trends Eng Res (IJCTER), 3(5):50–54
13. Navarro Hellin H, Martínez-del-Rincon J, Domingo Miguel R, Soto Valles F, Torres Sanchez
R (2016) A decision support system for managing irrigation in agriculture. Comput Electron
Agric 121–131
Open Set Domain Adaptation
for Hyperspectral Image Classification
Using Generative Adversarial Network

S. Nirmal, V. Sowmya and K. P. Soman

Abstract Hyperspectral image (HSI) classification attracted lots of attention due

to its complexity in dealing with large dimensions. In recent years, the techniques
for dealing with the HSI have been evolved, ensuring the increase in efficiency to
some extent in classification and other perspectives. Domain adaptation is a well-
established technique for using any trained classification model, when the feature
space from target domain is a subset of feature space from source domain. The objec-
tive of this paper is to create an efficient and effective model for HSI classification by
implementing open set (OS) domain adaptation and generative adversarial network
(GAN). This has advantages in quite few ways, such as creating a single training
model that deals with various HSI data set with common classes, classifying the
features in any data to specific trained classes and unknown (to be labelled) making
it easy to annotate. The proposed open set domain adaptation for HSI classification is
evaluated using Salinas and Pavia. The proposed method resulted in the classification
accuracy for unknown classes as 99.07% for Salinas and 81.65% for Pavia.

1 Introduction

In recent years, machine learning and deep learning techniques depict a remarkable
performance in different fields like speech processing, forecasting, computer vision,
machine translation, prediction and health care [1, 2]. Hyperspectral remote sensing
is one of the areas, where deep learning is applied due to its inexplicable efficacy
[3–6]. Hyperspectral remote sensing uses the electromagnetic spectrum to identify

S. Nirmal · V. Sowmya · K. P. Soman (B)

Center for Computational Engineering and Networking (CEN), Amrita School
of Engineering, Amrita Vishwa Vidyapeetham, Coimbatore, India
e-mail: kp_soman@[Link]
S. Nirmal
e-mail: nirmalsenthilnathan@[Link]
V. Sowmya
e-mail: v_sowmya@[Link]

© Springer Nature Singapore Pte Ltd. 2020 819

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
820 S. Nirmal et al.

and classify the objects, based on spectral response. Hyperspectral images (HSI)
consist of spatial and spectral information. Spectral information is given in the form
of bands and number of bands varies depending on the sensor.
Generative adversarial network (GAN) [7–9] is a classification technique, which
comprises of two portions as follows: a generator and a classifier. Generator in the
network considers random data and continuously tries to imitate the input data. the
classifier in the network is trained to classify the original data from the generated
data. Because of this continuous training between the generator and classifier in GAN
network, high accuracy rates can be obtained with less data samples.
Now, there is a high chance that various data sets may or may not contain common
classes. In such case, a model trained for one data set can be used in other data set
with necessary modifications. This technique is represented as transfer learning (TL)
[10–12]. Domain adaptation is a part of TL, where a features of a target data set are a
subset of features of source data set [13, 14]. There are two types of domain adaptation
namely: closed set (CS) and open set (OS) domain adaptation. CS domain adaptation
is used when the number of classes in both source and target domain are exactly the
same. Whereas, OS domain adaptation is used when target domain contains unknown
classes that are not presented in source domain [15, 16].
In the proposed work, we present a novel approach for dealing with HSI classifi-
cation using OS domain adaptation and GAN [17]. In practical, the unknown class
is unlimited, which makes it difficult to address with high precision, when training
source domain. To overcome this issue, we make use of OS domain adaptation by
backpropogation [17, 18], for training unknown class samples, as well as increasing
the individual class accuracy. Each pixel in the HSI represents the spectral signature
of the specific class. This also means that the class labelling has to be done for every
pixel, which is very difficult in real-life scenario. Also, pre-processing is done to
extract 1D pixel data with class information from 3D HSI image, to train the model.
The methodology is described in detail in Sect. 2. The experiment is carried out
on two sets of HSI data after pre-processing, as mentioned in Sect. 3. Experimental
results and conclusion are discussed in Sects. 4 and 5.

2 Methodology

In this paper, we have considered the architecture used in [17] to implement OS

domain adaptation for HSI. The data xs and label ys from source (X s , Ys ) and data xt
for target (X t ) are considered. The general architecture used in the work is described
in Fig. 1.
The class dimension for source data is K (known class) and for target data is K+1
(known + unknown class). Generally, the number of input and output classes in a
model must be same. Unlike other open set domain adaptation models [16], we do
not provide any data to separately train for unknown classes. To make up for the
missing class data, the generated data from generator, which fails to pass through the
Open Set Domain Adaptation for Hyperspectral Image … 821

Fig. 1 Open set model architecture [17] mapped for hyperspectral image classification

classifier, will be used to train the unknown class in source domain. Hence, there is
no need to provide data separately for unknown class during training.
When input data (x) are passed into the network, the generator will try to increase
the error rate by generating images. These generated images G(x) are then passed as
input to the classifier. The classifier will try to get a boundary between known and
unknown target classes. There the common class data (xs , ys ) present in the target
will be classified as known and remaining data as unknown class.
The output of classifier C(G(x)) will have K+1 logits. The softmax function is used
to convert the logits into probabilities [17]. Thus, we get K+1 dimensional output
for K-dimensional input. The classifier is trained to give output of p(y = K + 1|xt )
= t, where 0 < t < 1. In our experiment, the value of t = 0.5 is set as a boundary
between known and unknown class [3].

3 Data Set and Model Description

This section provides the intuitive understanding of HSI data set and convolution
neural network model used in the experiment.
Salinas is HSI data set collected over Salinas Valley, California, using AVIRIS
sensor [19]. The spatial resolution for this data set is 3.7-metre pixels. The dimension
of the data is 512 × 217 × 224 with a total of 16 classes.
Salinas-A is one of the extracted subscene of Salinas image, which comprises
86 × 83 × 224 pixels with 6 classes [19]. For one of our experiment, we consider,
Salinas-A data set with 6 classes (source) and Salinas data set with 8 classes (target)
is used. The class and sample for each class is mentioned in Table 1.
PaviaU is one of the two data sets obtained from Pavia, Northern Italy, using
ROSIS sensor [19]. Pavia University is comprised of 610 × 610 × 102 pixels repre-
senting nine classes, after discarding few samples with no information.
For next part of the experiment, 5 classes were selected from PaviaU data set with
1000 samples in each class, which is used for source training. Then, the complete
822 S. Nirmal et al.

Table 1 Data set description for Salinas-A(source) and Salinas(target)

Class Features Source data set Target data set
(Salinas-A) (Salinas)
0 Brocoli_green_weeds_1 391 2008
1 Corn_senesced_green_weeds 1343 3278
2 Lettuce_romaine_4wk 616 1068
3 Lettuce_romaine_5wk 1524 1927
4 Lettuce_romaine_6wk 675 916
5 Lettuce_romaine_7wk 799 1070
6 Unknown (Brocoli_green_ 0 5702
weeds_2 and Fallows)
Total samples 5348 15969

Table 2 Data set description for PaviaU with 5 classes (source) and PaviaU with 9 classes(target)
Class Features PaviaU (source PaviaU (target
domain—5 classes) domain—9 classes)
0 Asphalt 1000 6631
1 Bare Soil 1000 5029
2 Gravel 1000 2099
3 Trees 1000 3064
4 Painted metal sheets 1000 1345
5 Unknown (Meadows, 0 24607
Self-Blocking Bricks
and Shadows)
Total samples 5000 42775

PaviaU data set with around 42,775 samples with 9 classes is used as target. The data
set details are mentioned in Table 2.
The convolutional neural network (CNN) is used for training the model in the
architecture. The details regarding layers of CNN used in the experiment is shown in
Fig. 2. 1D-batch normalization is used with convolution and fully connected layer.

4 Experimental Results

The parameters which gives the best results for our model during the experiment
are mentioned below. In the proposed work, the network is trained for 2000 epochs
and batch size of 128. Adam optimization method with learning rate of (0.001) and
cosine ramp-down is used for training the network [17].
Open Set Domain Adaptation for Hyperspectral Image … 823

Fig. 2 Convolution neural network used in the proposed open set domain adaptation for hyper-
spectral image classification

4.1 Salinas Data Set

The main purpose of the work is to identify the model effectiveness in classifying the
unknown class. So, precision, recall (classwise accuracy) and f1-score are calculated
with contingency values obtained using confusion matrix (CM) and are used to
evaluate the model. Table 3 shows the results for first experimental set-up using
Salinas-A (source—6 classes) and Salinas (target—8 classes). The receiver operating
characteristics (ROC) curve is a plot between true positive rate and false positive
rate, in which the area under the curve (AUC) represents the accuracy of various
classifier. The average accuracy of the known class is 95.10% and AUC is 98%. By
using confusion matrix (CM) in Fig. 3, we can interpret that from unknown class
with 5702 samples collected from multiple classes, 53 samples are misclassified into
824 S. Nirmal et al.

Table 3 Classwise accuracy for Salinas-A (source) and Salinas (target) obtained using the proposed
method
Class Features Precision Recall F1-Score
0 Brocoli_green_weeds_1 1.00 0.93 0.96
1 Corn_senesced_green_weeds 1.00 0.93 0.96
2 Lettuce_romaine_4wk 0.96 0.93 0.95
3 Lettuce_romaine_5wk 0.94 1.00 0.97
4 Lettuce_romaine_6wk 0.97 0.97 0.97
5 Lettuce_romaine_7wk 0.96 0.94 0.95
6 Unknown (Brocoli_green_ 0.95 0.99 0.97
weeds_2 and Fallows)
Average 0.97 0.96 0.96

Fig. 3 Confusion matrix obtained for the proposed method for Salinas-A (source) and Salinas
(target) data set

other classes and classification accuracy of 99.07% and AUC of 0.98%is obtained.
We can interpret from above results that our classifier model works well in both
known and unknown classes.
Open Set Domain Adaptation for Hyperspectral Image … 825

Table 4 Classwise accuracy for PaviaU with 5 classes (source) and PaviaU with 9 classes (target)
obtained using the proposed method
Class Features Precision Recall F1-Score
0 Asphalt 0.78 0.98 0.93
1 Gravel 0.91 0.59 0.72
2 Trees 0.61 0.87 0.72
3 Self-Blocking Bricks 0.69 0.87 0.71
4 Bare Soil 1.00 0.99 0.99
5 Unknown (Meadows, Painted 0.88 0.83 0.85
metal sheets and Shadows)
Average 0.81 0.86 0.83

4.2 PaviaU Data set

For the second experiment, PavaiU data set is considered to check the authenticity
of OS domain adaptation for HSI in classifying unknown class data. Table 4 shows
the precision, recall and f1-score results for PaviaU (source— class—5000 samples)
and PaviaU (target—9 class—42,775 samples). The average accuracy rate of known
class is 85.97%. On further study using CM in Fig. 4, around 4400 samples out of
24,607 unknown samples are misclassified into other classes and the classification
accuracy rate of 81.65% is obtained unknown class. when the proportion between
known and unknown class sample is taken into consideration, this further proves that
the model works well in classifying unknown class samples. We got an average rate
of above 80% in both precision and recall (sensitivity) and overall average AUC of
84%, which further proves that open set domain adaptation can implemented in HSI
classification.
Table 5 gives the overall accuracy rates for the experiments discussed above. We
evaluated the performance of the model using classification accuracy for known
and unknown classes. we also calculated precision, recall and f1-score for further
studies. It is observed from the results that the model could accurately distinguish
the unknown class from known classes in hyperspectral data set. Since the number
of training samples chosen in both cases is low, it can be said that GAN plays an
important role in obtaining the desired results.

5 Conclusion

In this work, we applied open set domain adaptation to hyperspectral data set using
GAN. The obtained results shows that, open set domain adaptation works well with
HSI and it will increase the efficiency of HSI classification models. It is clear from the
experimental analysis that, various HSI data set can share a common training model
826 S. Nirmal et al.

Fig. 4 Confusion matrix obtained for the proposed method for PaviaU data set with 5 classes
(source with 1000 samples in each class) and 8 classes (target)

Table 5 Performance of the proposed open set domain adaptation method for hyperspectral image
data sets
Data set Known class accuracy Unknown class Overall accuracy (%)
(%) accuracy (%)
Salinas-A and Salinas 95.10 99.07 96.38
PaviaU with 5 and 9 85.97 81.65 82.69
classes

irrespective of the number of classes (excluding common classes) and samples with
the use of GAN network.
As the future scope of the present work, the hyperparameter tuning can be per-
formed to increase the efficiency of the proposed model and also it can be extended
to other HSI data set with different spectral ranges.
Open Set Domain Adaptation for Hyperspectral Image … 827

References

1. Pathinarupothi RK, Rangan ES, Gopalakrishnan EA, Vinaykumar R, Soman KP (2017) Sin-
gle sensor techniques for sleep Apnea diagnosis using deep learning. In: IEEE International
Conference on Healthcare Informatics (ICHI), pp 524–529
2. Charmisha S, Sowmya V, Soman KP (2018) Dimensionally reduced features for hyperspectral
image classification using deep learning. In: Proceedings of the international conference on
communications and cyber physical engineering, vol 500, pp 171–179
3. Srivatsa S, Sowmya V, Soman KP (2016) Empirical wavelet transform for improved hyper-
spectral image classification, intelligent systems technologies and applications, pp 393–401
4. Reshma R, Sowmya V, Soman KP (2018) Effect of Legendre-Fenchel denoising and SVD-based
dimensionality reduction algorithm on hyperspectral image classification. Neural Comput Appl
29(8):301–310
5. Lee H, Kwon H (2017) Going deeper with contextual CNN for hyperspectral image classifi-
cation. IEEE Trans Image Process 26(10)
6. Srivatsa S, Sowmya V, Soman KP (2018) Least square based fast denoising approach to hyper-
spectral imagery. AISC, vol 518
7. Bousmalis K, Silberman N, Dohan D, Erhan D, Krishnan D (2017) Unsupervised pixel-level
domain adaptation with generative adversarial networks. CVPR
8. Tzeng E, Hoffman J, Saenko K, Darrell T (2017) Adversarial discriminative domain adaptation.
CVPR
9. Salimans T, Goodfellow I, Zaremba W, Cheung V, Radford A, Chen X (2016) Improved
techniques for training gans. NIPS
10. Raina R, Battle A, Lee H, Packer B, Ng AY (2007) Self-taught learning: transfer learning from
unlabeled data. In: Conference on machine learning
11. Long M, Wang J, Jordan MI (2017) Deep transfer learning with joint adaptation networks.
ICML
12. Pan SJ, Tsang IW, Kwok JT, Yang Q (2011) Domain adaptation via transfer component analysis.
IEEE Trans Neural Netw 22(2)
13. Damodaran BB, Kellenberger B, Flamary R, Tuia D, Courty N (2018) DeepJDOT: Deep Joint
distribution optimal transport for unsupervised domain adaptation. [Link]
10081v1 arXiv:1803.10081v1
14. Yan H, Ding Y, Li P, Wang Q, Xu Y, Zuo W (2017) Mind the class weight bias: weighted
maximum mean discrepancy for unsupervised domain adaptation. CVPR
15. Panareda Busto P, Gall J (2017) Open set domain adptation. In: IEEE international conference
on computer vision. ArXiv1804.10427
16. Bendale A, Boult TE (2016) Towards open set deep networks, CVPR
17. Saito K, Yamamoto S, Ushiku Y, Harada T (2018) Open set domain adaptation by backpro-
pogation, ArXiv :1804.10427v2[[Link]]
18. Ganin Y, Lempitsky V (2015) Unsupervised domain adaptation by backpropagation, ICML
19. Hyperspectral image dataset available at [Link]
Hyperspectral_Remote_Sensing_Scenes
AHP-FVIKOR Based Access Network
Selection in Vehicular Communications

C. Suganthi Evangeline and Vinoth Babu Kumaravelu

Abstract Providing seamless connectivity is the major challenge in vehicular com-

munication. The vehicles need to perform vertical handover in order to select the best
target network. To identify the target network which satisfy the quality of service
(QoS) demands is the need of the hour. To provide solution for this issue, a two stage
fuzzy logic based target network selection is proposed where factor for handover
estimation is carried out in first stage and fuzzy VIKOR based target network selec-
tion is developed. Through simulations, the ranking method is compared with other
schemes.

Keywords Seamless connectivity · Fuzzy-VIKOR (FVIKOR) · Target network

selection (TNS) · Quality of service (QoS) · Vehicular network (VANET) ·
Handover estimation

1 Introduction

Vehicular Adhoc Networks provide the feature of communicating with other vehi-
cles and also with the infrastructure [1]. The process of communication includes
the exchange of messages, toll services, emergency messages, and also infotainment
services. To provide better quality of service the vehicle should be Always Best
Connected (ABC) with the network [2]. The network providing the services need
not be homogeneous but also heterogeneous involving Universal Mobile Telecom-
munications System (UMTS), Worldwide Interoperability for Microwave Access
(WiMAX), Wireless Fidelity (Wi-Fi). Several network selection approaches are avail-
able to select the appropriate network during handover process. Quality of service
(QoS) always accompanied with the network selection process, which follows three
phases like (i) Handoff Initiation, (ii) Network Selection and (iii) Handoff Execution

C. S. Evangeline
Karunya Institute of Technology and Sciences, Coimbatore, India
C. S. Evangeline (B) · V. B. Kumaravelu
Vellore Institute of Technology, Vellore, India
e-mail: evangelineme4@[Link]
© Springer Nature Singapore Pte Ltd. 2020 829
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
830 C. S. Evangeline and V. B. Kumaravelu

phase. As mentioned above, if the handover happens within the same network it is
referred as horizontal handover and when it happens between two different networks
it is referred as vertical handover.
To have seamless connectivity the heterogeneous network should include the
features of third-generation (3G) and fourth-generation (4G). 3G networks such as
UMTS and Code Division Multiple Access (CDMA) 2000 have wide coverage but
lack in terms of monetary costs and bandwidth meanwhile IEEE 802.16 WiMAX
has the feature of providing real-time data in Wireless Metropolitan Area Network
(WMAN). The main aim of this paper is to find the exact moment to initiate handover
and carry out the network selection process using fuzzy-VIKOR (FVIKOR), which
enables the user to select for better QoS in the heterogeneous environment.

2 Related Works

In [3], network selection algorithm based on Simple Additive Weighting (SAW)

method is presented in which normalized weight values for all networks and criteria
are simply added to produce the best utility factor. The paper [4] analyzed the impacts
of user velocity and traffic load on the node (mobile, vehicle) switching from one
network to another in heterogeneous wireless network. In [5] Fuzzy-Techniques for
Order Preference by Similarity to Ideal Solution (Fuzzy-TOPSIS) provide a QoS
aware vertical handover decision by considering bandwidth, jitter, delay, and bit
error rate. QoS enabled vertical handover decision is presented in [6], where weights
are assigned using Analytical Hierarchical Process (AHP). The comparative analy-
sis of various MADM approaches such as SAW, Grey Relational Analysis (GRA),
Multiplicative Exponent Weighting (MEW), VIseKriterijumska Optimizacija I Kom-
promisno Resenje (VIKOR), and TOPSIS has been presented in [7]. Fuzzy VIKOR
(FVIKOR) was presented in [6] which aims to use parallel fuzzy logic controllers
(FLC) with minimum number of rules to estimate necessity of handover and select
the target network.
Data is disseminated using clustering method [8] efficiently which improves the
QoS in VANET applications. The decision process for vertical handover in vehicular
adhoc network was proposed in [9], which gives insight about the probability of
unnecessary handover and handover failure probability rate with respect to vehicle
speed.

3 Proposed Method

The vehicular communication considered is of heterogeneous type as shown in Fig. 1,

and handover happening in such scenario is called as vertical handover. For the pro-
posed work, the network considered are UMTS, WiMAX, and Wi-Fi. The proposed
algorithm represented in Fig. 2, works in two stages. The stage 1, deals with the
AHP-FVIKOR Based Access Network Selection … 831

Fig. 1 Heterogeneous vehicular environment

Stage 1: Fuzzy Based Factor For Handover Estımatıon (FFHE)

Input: Received signal strength (RSS), velocity of the vehicle
Output: FHO
Steps:
1. RSS and velocity are given as input to fuzzy inference engine (FIE)
2. Fuzzification is done using membership function.
3. Decision is done based on the rules written to FIE
4. Defuzzification is done based on centre of centroid formula to get
the crisp value
Stage 2:FVIKOR Based TNS
Input:6 criteria (Cost,bandwidth, delay,jitter, packet loss,utilization of
network)
3 networks (WiMAX, Wi-Fi, UMTS)
Output: Handover to best target network
Steps:
1. Assigning weights to criteria using AHP
2. Rank the target network based on Fuzzy VIKOR
3. Handover to the network with high score value.

Fig. 2 Algorithm for the proposed work

factor for handover (FHO) which initiates the need for performing handover. The
vehicle with high FHO is allowed to execute next stage. In stage 2, it deals with the
target network selection (TNS) based on Fuzzy VIKOR ranking.

3.1 FFHE

Many handover initiation algorithm deals only with RSS [10]. The values are sam-
pled in each time instant to trigger the handover in heterogeneous environment. For
832 C. S. Evangeline and V. B. Kumaravelu

Fig. 3 FIE for proposed

work

RSS
FFHE

Factor for Handover

Velocity

Fig. 4 Surface plot RSS

versus velocity against FHO

vehicular communications, in addition to RSS the other main factor which influences
mobility is the speed of the vehicle, i.e., Velocity of the vehicle. FIE for proposed work
is represented in Fig. 3, surface plot RSS versus velocity against FHO is depicted in
Fig. 4 and the rules defined for FIE is mentioned in Fig. 5.

3.2 Weight Assignment Using AHP

AHP stands for Analytic Hierarchy Process. It is a method to support multi-criteria

decision-making and was originally developed by Saaty [11]. AHP derives ratio
scales from paired comparisons of criteria and allows for some small inconsistencies
in judgments. The steps involved in weight assignment in AHP are discussed below:
Step 1. Develop the weights for the criteria by
• developing a single pair-wise comparison matrix for the criteria;
• multiplying the values in each row together and calculating the root of said product;
• normalizing the aforementioned nth root of products to get the appropriate weights;
• calculating and checking the Consistency Ratio (CR).
AHP-FVIKOR Based Access Network Selection … 833

Fig. 5 Rules defined for FIE

Step 2. Develop the ratings for each decision alternative for each criterion by
• developing a pair-wise comparison matrix for each criterion, with each matrix
containing the pair-wise comparisons of the performance of decision alternatives
on each criterion;
• multiplying the values in each row together and calculating the nth root of said
product;
• normalizing the aforementioned nth root of product values to get the corresponding
ratings;
• calculating and checking the Consistency Ratio (CR).
Step 3. Calculate the weighted average rating for each decision alternative. Choose
the one with the highest score.

3.3 FVIKOR-Based Target Network Selection

Assume there exists m alternative such as A1 , A2 , A3 …Am and n criteria. The rating
of jth aspect is denoted by f ij, i.e., f ij is the value of jth criterion function for the
alternative Ai .
Step 1: Evaluation Matrix formation
834 C. S. Evangeline and V. B. Kumaravelu
⎡ ⎤
x̃11 · · · x̃1n
⎢ .. . . .. ⎥
D=⎣ . . . ⎦ (1)
x̃m1 · · · x̃mn

where 
xi j is the performance alternative j; j = 1, 2, 3…n in terms of criteria i; i = 1,
2, 3…m.
Step 2: Normalized fuzzy performance decision matrix construction
⎡ ⎤
f˜11 · · · f 1n
⎢ ⎥
F = ⎣ ... . . . ... ⎦ (2)
f˜m1 · · · f˜mn

x
where 
f i j = m i j xi j
j=1
Step 3: Determination of fuzzy best value and fuzzy worst value


f j∗ = max
xi j ∀i (3)


f J∗ = min
xi j ∀i (4)

Step 4: Computation of

i
Si , R

 j 
n w f j∗ − 
xi j

Si = (5)
j=1 f j∗ − 
 f j−

j 
w f j∗ − xi j

i = max
R ∀ j = 1, 2, . . . n (6)
f j∗ − 
 f j−

Step 5: Calculation of 
Q i value
   
 Si − S

− −

i − R
R
Qi = v + (1 − v) (7)

S − S−
+  +
R −R −

where

S
+ = max
+ = max R
Sι , R
ι
i i

S
− = min

i
− = min R
Sι , R ι
i i

from [12], v is the weight of VIKOR index value = 0.5

Step 6: Sorting
AHP-FVIKOR Based Access Network Selection … 835

The values of

i , 
Si , R Q i are sorting according to descending order. The maximum
value of Q i is based on merit points and it is denoted by A(1) . The second alternative
is denoted by A(2) .
Step 7: Ranking
The alternative of maximum value of  Q i is said to be the best alternative.

4 Results and Discussions

In order to validate the proposed method, the numerical calculations are done. The
weights for all criteria is assigned using AHP and followed by the ranking process
using FIKOR. By adopting the steps given in Sect. 3.2, Table 1 is constructed. Table 2
shows normalized fuzzy decision matrix.
Table 3 is constructed by performing calculations as per the Sect. 3.3 and the
ranking order in Table 4 is calculated by FVIKOR with maximum Q value. As per
the assumption the network A3 -UMTS is considered to the best alternative.
The aim to find appropriate target network with less number of handovers and
decision delay. By comparing FVIKOR, VIKOR, SAW, TOPSIS, and FSAW in terms
of executable handovers and decision delay, the proposed method with FVIKOR

Table 1 Calculated weights for criteria

C1 C2 C3 C4 C5 C6
weights 0.05 0.365 0.154 0.221 0.082 0.128

Table 2 Normalised fuzzy decision matrix

Criteria A1 A2 A3
C1 (0.33,1.33,3) 0.33,1.33,3) (7.66,9.33,10)
C2 (1,2.33,4.33) (0.33,1.33,3) (6.33,8,9.66)
C3 (8.33,9.66,10) (4.33,6.33,9.33) 0.33,1.33,3)
C4 (5.66,7.66,9.66) (0.33,1.66,3.66) (0.33,1.33,3)
C5 (6.33,8,9.66) (6.33,8,9.66) (4.33,6.33,8.66)
C6 (7,8.66,9.66) (0.33,1.33,3) (4.33,6.33,8)

Table 3 Computation of S, R and Q value

A1 A2 A3
Si [0.7827 1.2562 1.8293] [0.7284 0.9703 1.1869] [0.7554 1.0931 1.4733]
Ri [0.5904 0.7338 0.8181] [0.7000 0.8700 0.9700] [0.7000 0.8700 0.9700]
Qi [0.1969 0.4702 0.5000] [0.5000 0.5000 0.5000] [0.5978 0.7021 0.7229]
836 C. S. Evangeline and V. B. Kumaravelu

Table 4 Ranking of the

A1 A2 A3
network
Si 1.2894 0.961 1.107
Ri 0.714 0.846 0.846
Qi 0.389 0.5 0.6742
Rank 3 2 1

performs well in handover reduction as shown in Fig. 6 and gives less decision delay
when compared to TOPSIS as shown in Fig. 7.
The ranking methods combined with fuzzy adapts with the changing environment
and performs well when compared to the classical approach. The overhead in calcu-
lation provides slight higher delay when compared to SAW. But on considered the
criteria involved for deciding the target network, FVIKOR is the optimal solution
among various MADM schemes.

Fig. 6 Number of handover

Fig. 7 Decision delay for

different target networks
AHP-FVIKOR Based Access Network Selection … 837

References

1. Miller J (2008) Vehicle-to-vehicle-to-infrastructure (V2V2I) intelligent transportation system

architecture. In: IEEE intelligent vehicles symposium, IEEE, pp 715–720
2. Fodor G, Furuskar A, Lundsjo J (2004) On access selection techniques in always best connected
networks. In: ITC specialist seminar on performance evaluation of wireless and mobile systems,
pp 89–100
3. Afshari A, Mojahed M, Yusuff RM (2010) Simple additive weighting approach to personnel
selection problem. Int J Innov, Manag Technol 1(5):511
4. Hasswa A, Nasser N, Hassanein H (2005) Generic vertical handoff decision function for
heterogeneous wireless. In: Second IFIP international conference on wireless and optical
communications networks, IEEE, pp 239–243
5. Vasu K, Maheshwari S, Mahapatra S, Kumar CS (2011) QoS aware fuzzy rule based vertical
handoff decision algorithm for wireless heterogeneous networks. In: National conference on
communications, IEEE, pp 1–5
6. Zineb AB, Ayadi M, Tabbane S (2017) An enhanced vertical handover based on fuzzy inference
MADM approach for heterogeneous networks. Arabian J Sci Eng 42(8):3263–3274
7. Stevens-Navarro E, Martinez-Morales JD, Pineda-Rico U (2012) Evaluation of vertical handoff
decision algorithms based on madm methods for heterogeneous wireless networks. J Appl Res
Technol 10(4):534–548
8. Evangeline CS, Appu S (2017) An efficient data transmission in VANET using clustering
method. Int J Electron Telecommun 63(3):309–313
9. Evangeline CS, Kumaravelu VB(2017) Decision process for vertical handover in vehicular
adhoc networks. In: International conference on microelectronic devices, circuits and systems
(ICMDCS), IEEE, pp 1–5
10. Mohanty S, Akyildiz IF (2006) A cross-layer (layer 2 + 3) handoff management protocol for
next-generation wireless systems. IEEE Trans Mob Comput 5(10):1347–1360
11. Saaty TL (2005) Analytic hierarchy process. Encyclopedia of Biostatistics 1
12. Musani S, Jemain AA (2015) Ranking schools’ academic performance using a fuzzy VIKOR.
J Phys: Conf Ser 622(1):012036). IOP Publishing
Measuring Web Content Credibility
Using Predictive Models

R. Manjula and M. S. Vijaya

Abstract Web content credibility is a measure of believable and trustworthy of the

web content that is perceived. Content can turn out to be unreliable if it is not up-
to-date and it is not measured for quality or accuracy and therefore, web content
credibility is important for the individuals to access the content or information. The
analysis of content credibility is an important and challenging task as the content
credibility is expressed on essential factors. This paper focus on building predictive
models to discover and evaluate credibility of a web page content through machine
learning technique. A corpus of 300 web page contents have been developed and the
factors like Readability, Freshness, Duplicate Content are defined and captured to
model the credibility of web content. Two different labeling such as binary labeling
and numeric labeling are used for defining credibility. In case of binary labeling,
the high and low credibility of web content are represented by 1 and 0, respec-
tively, whereas in case of numeric labeling five-point scale rating is used to mark the
content credibility. Accordingly, two independent datasets have been developed. Dif-
ferent regression algorithms such as Linear Regression, Logistic Regression, Support
Vector Regression (SVR) are employed for building the predictive models. Various
experiments have been carried out using two different datasets and the performance
analysis shows that the Logistic Regression model outperforms well when compared
to other prediction algorithms.

Keywords Web content credibility evaluation · Machine learning · Prediction ·

Regression

R. Manjula (B) · M. S. Vijaya

PSGR Krishnammal College for Women, Coimbatore, India
e-mail: manjulacbe.r@[Link]
M. S. Vijaya
e-mail: msvijaya@[Link]

© Springer Nature Singapore Pte Ltd. 2020 839

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
840 R. Manjula and M. S. Vijaya

1 Introduction

Web content credibility is expressed as the believability of web content. Web content
credibility is created with two dimensions: trustworthiness and expertise. When a
web page conveys both these qualities, people will find, it is a credible content. When
it lacks one of these qualities, due to misrepresentation, inaccuracy, credibility will
suffer [1]. Content credibility is being utilized in different domain like economy,
healthy life-style, politics, personal finance, and entertainment. A credible website
can gather huge benefits on to the website and the corporate. People frequently choose
to respond to a significant message based on their perception of the communicator
[2].
Having a stylish, specialized looking webpage, offer credibility to the contents.
The single more crucial attribute of efficient content is credibility. There are four types
of web credibility such as Presumed credibility, Reputed credibility, Surface credibil-
ity, and Earned Credibility. Presumed credibility describes the general assumptions
about the product brand. Reputed credibility refers to the believed third party refer-
ence about the brand. Surface credibility expresses how much a perceiver believes
something based on simple inspection. Earned credibility states that the personal
experience of typographical text. Well-written web content keeps the customers
involved, and inspires them to explore their web content [3].
Web content credibility involves the abilities and competencies needed for read-
ing, writing and participating content on the web. Credibility is the single most
eminent attribute of great marketing content. Effective content must also be applica-
ble and precious. It’s also clear that lack of trust is weakening the impact of content.
In a modern survey of technology buyers by TrustRadius, survey participants are
asked to rate the helpfulness and trustworthiness of source of data used in buying
decisions. Credible content is authoritative [4].
To solve the problem of predicting web content credibility numerous computa-
tional techniques are accepted in the existing research. Machine learning utilizes
statistical techniques to realize and improve the performance of the predictive mod-
els. Hence, it is proposed in this work to develop an accurate predictive model by
learning various influencing parameters through supervised learning.

2 Literature Survey

Various research work had examined and explored on understanding the factors that
had an impact on credibility evaluations. In 2017, Michal Kakol et al., proposed “un-
derstanding and predicting web content credibility using content credibility corpus”.
The factors were based on empirical data. The content credibility corpus (C3) dataset
was used from a massive crowdsourced web credibility assessment. The factors such
as web media type, advertising, news source, official page, etc., were used to attain a
high level of quality. Random forest approach was used to indicate a comprehensive
set of credibility evaluation criteria [2].
Measuring Web Content Credibility Using … 841

In 2001, Fogg et al. concentrated on understanding the factors that had an impact
on credibility evaluations and spent two approaches for regulating credibility evalua-
tion factors. The first was a declarative approach, where respondents were requested
to evaluate credibility and precisely suggest which factor from a list was influencing
their decision. In 2003, Fogg et al. produced the second approach, the manual coding
of comments left by defendants evaluated credibility by two coders. Unsupervised
machine learning and NLP techniques was used from the content credibility corpus
(C3) dataset [5].
In 2013, Olteanu et al. proposed “web credibility: feature exploration and credi-
bility prediction”. The research work automatically assesses web credibility investi-
gated through various characteristic of webpages. The features were detected from
textual content, link structure, webpages design, as well as their social popularity
learned from popular social media sites. The random baseline approach for regres-
sion was concerned for the real dataset-based experiment under 75% accuracy for
classification and 53% of mean absolute error for regression [6].
In 2010, Joo Chung et al., proposed “An anatomy of the credibility of online
newspapers”. The research work exposed the primary components of credibility
of three types of online newspaper and the variance of credibility of news by that
type. The credibility scales were determined using seven-point likert-type scales and
mean was computed and the scale was explored for the similarities. The factors such
as expertise, trustworthiness, and attractiveness were used to show the significant
difference between online newspapers [7].
In 2014, Wawer et al. utilized natural language processing methods together with
machine learning to look for specific content terms that are predictive of credibility.
By this method, they found expected terms, such as energy, research, safety, security,
department, fed and gov. linguistic and social features was used based on machine
learning methods by the recognized trust level. Content-specific language features
were applied that greatly had enhanced the accuracy of credibility predictions. The
content organization factor was approximated by the analyzed CSS of the webpage.
The factor language quality was approximated using NLP techniques [8].
The factors defining the web content credibility are highly imperative in building
accurate prediction model through machine learning [9]. The existing research work
focused on prediction of credibility using factors like web media type, advertising,
news source, sales offer and objectivity of web page content. But these factors will not
contribute in controlling the content credibility [10]. It was detected that the factors
like readability, authority, understandability, accessibility, popularity, broken links,
freshness, page rank and duplicate content are also needed for evaluating the web
content credibility [11]. These factors can be taken and used as evaluation criteria
for web content credibility evaluation appropriate to ordinary web content for fine-
tuned credibility assessments [12]. Similarly, the quantitative predictive model was
built based on content credibility corpus dataset and it has the lower correlation
coefficient [13]. Hence, it is proposed in this work to build accurate web content
credibility prediction model by capturing the above efficient factors using regression
[14]. Several regression methods like linear regression, logistic regression have been
applied to boost up the correlation coefficient between the factors [15, 16].
842 R. Manjula and M. S. Vijaya

3 Methodology

The main focus of this work is to build an efficient predictive model to evaluate the
web content credibility based on leading factors. The bare bones of the proposed
methodology involve different phases such as data collection, feature extraction, and
building models using regression algorithms such as linear regression, logistic regres-
sion, and support vector regression. Various components of proposed methodology
are depicted in Fig. 1 and explained in the following section.

3.1 Data Collection

The corpus of 300 web pages have been selected from websites related to different
domain such as economy, healthy life-style, politics, personal finance, and entertain-
ment. The contents taken from Economy domain comprised of information about
the production and consumption of goods and services. Web page contents related
to short- and long-term health benefits, balanced diet is collected from Healthy Life-
Style domain. From webpages of politics domain, the contents are gathered based
on the activities related with the growth of a country or area. Web page contents
based on management of financial decisions are collected from personal finance

Fig. 1 Methodology for predicting and evaluating web content

Measuring Web Content Credibility Using … 843

domain. From entertainment domain, contents are collected based on information

about games, movie, etc., Likewise, details of author, publisher of webpages, web-
page updates, particulars of domain are collected from the particular pages. From
each of these five domains, 60 instances have been collected and stored in the form
of text files. Finally, a corpus with 300 instances have been developed.

3.2 Feature Extraction and Training Data

Feature extraction plays a prominent role for improving the performance of web con-
tent credibility. In this work, readability, authority, accessibility, understandability,
popularity, freshness, broken links, page rank, and duplicate content are considered
as vital factors for evaluating web content credibility. A set of nine factors is derived
from each content and are described below and the dataset has been developed.
Readability. The Readability depends on the contents and presentation of the
webpage. Readability is determined by the content visibility, reading speed and
legibility. Higher readability improves reading effort and speed for any reader. The
Flesch–Kincaid readability is used to compute the readability of contents. The value
for this readability factor is computed using R code.
Authority. Authority of a web page controls the publisher and organization of
the webpage and inspect whether it splits from the webmaster. Domain authority
helps to measure the authority of a website by comparing it with other websites and
it will predict the ranking of a website. A Domain Authority score ranges from one
to 100, with higher scores corresponding to a greater ability to rank. The value of
this authority factor is derived from nibbler tool.
Accessibility. The accessibility of the web page helps to remove barriers that
prevent interaction with or access to websites. Web accessibility depends on various
components working together, including web technologies, web browsers, authoring
tools, and other user agents, and websites. Accessibility value is obtained from the
nibbler tool.
Understandability. Content must be easy to follow and recognize for many users.
For most content, understandability simply avoiding overly complex sentences, ter-
minology and providing clear layout and design. The values from 0 to 100 provides
the range of content understandability. The understandability value is estimated using
R code.
Popularity. A webpage popularity is a tremendous way not only to build a website
but also to show others how the website is good. If the website or web page is
information-rich and attractive, the particular webpage will have high link popularity.
The more ratings a post gets, the more reliably the ratings tell the value. The value
of this factor is derived from nibbler tool.
Freshness. The freshness factor is an element of search algorithms that provides
greater weight to latest content over older content for some search queries. Search
844 R. Manjula and M. S. Vijaya

engines presented the freshness factor for searches related to trending topics, recur-
ring events (awards, sports scores, and so on) and breaking news. It provides the
updates of each web page. Updates of the web page are resulting from nibbler tool.
Broken links. A broken link is a hyperlink of a website which is related to an
empty or non-existent external webpage. A broken link or dead link is a link on a web
page that no longer works because the website is encountering an improper URL,
does not allow outside access. Broken links value is fetched from nibbler tool.
Page rank. PageRank refers to the system and the algorithmic method that Google
uses to rank pages as well as the numerical value is assigned to pages as a score.
PageRank is often considered to be a number between 0 and 10 (with 0 being the
lowest and 10 being the highest). The value of this factor is computed using R code.
Duplicate content. Duplicate content is content that seems on the web page in
more than one place. That one place is termed as a location with a unique website
address (URL)—so, if the same content appears at more than one web page, duplicate
content can be found. The value of this duplicate content factor is derived using R
code.
The above nine factors are used as values of predictor variables (Xi) of regression
algorithms. In this work, the credibility of web content is considered as the response
variable (Y ) for which the modeling is being done. In order to create the training
dataset enabling supervised learning, the value of credibility of web content is also
computed as below for each record and it is associated with the corresponding tuple.
Credibility. Credibility is said to be the quality or power of inspiring belief. The
credibility value is derived using crowdsourced environment like web of trust (WOT).
The binary labeling 0 is assigned to credibility if the content is not believable and 1
is assigned if the content is realistic. In case of numeric labeling 5-point scale rating
is used to spot the content credibility, where the rating from 1 to 3 specifies the low
credibility of content and the rating 4 and 5 represents high credibility.
Datasets. The factors derived from web page contents are used to train the pre-
diction models using supervised learning. In this work, two separate datasets have
been created, one with binary credibility labeling and another with numeric credi-
bility labeling. Both datasets consist of 300 instances with nine dimensions. Each
instance of independent variables is associated with binary credibility values and a
dataset Binary Credibility Dataset (BCD) is created. Similarly, second dataset named
Numeric Credibility Dataset (NCD) is developed by adding numeric credibility
values to all the 300 tuples.

4 Experiments and Results

The predictive models for web content credibility have been generated by imple-
menting the regression algorithms using python library in scikit-learn environment.
Scikit-learn perhaps the most appropriate library for machine learning in Python.
The real-time data has been collected, the content credibility factors are identified
and derived. Two separate datasets namely BCD and NCD have been created to show
Measuring Web Content Credibility Using … 845

the variations of binary and numeric credibility values of web page content. These
two datasets are employed for implementing the models.
In these experiments, prediction algorithms like Linear Regression, Logistic
Regression, Support Vector Regression have been employed to build the predictive
models. The performance of these models is evaluated using 10-fold cross-validation
based on correlation coefficient, mean squared error, root mean squared error and
mean absolute error. Correlation coefficient is utilized as a primary performance
measure for predicting content credibility values range from −1 to +1. A correlation
coefficient of 0 means that there is no relationship and +1 indicates a perfect positive
correlation. Mean squared error is said to be the average of the squared error and is
used as the loss function for least square regression. RMSE is an absolute measure
of fit. Mean absolute error is a measure of difference between continuous variables.
MAE is the model evaluation metric and is the mean of the absolute values of each
prediction error on all instance of the test dataset. The goodness of fit is evaluated
using these measures.
The web content credibility model based on NCD dataset is built using Linear
Regression and Support Vector Regression as the value of response variable is a
continuous-valued function in these models. The Logistic Regression algorithm is
implemented to create predictive model using BCD dataset since response variable is
binary in logistic regression. The performance of these regression-based credibility
prediction models is measured using metrics such as correlation coefficient, mean
squared error, root mean squared error and mean absolute error and are tabulated
(Tables 1, 2 and 3).
Comparative analysis. The performance results of the above three predictive
models are compared to their metrics. From the comparative analysis, it is detected
that the Logistic Regression attains highest result for content credibility prediction
model than Linear Regression and Support Vector Regression in the minimal learning
time. The Logistic Regression reaches the correlation coefficient of 0.896 and Linear
Regression accomplishes the correlation 0.875 and the correlation of 0.816 is gained

Table 1 Predictive
Measures Values
performance of linear
regression using NCD dataset Correlation coefficient 0.875
Mean squared error (MSE) 0.285
Root mean squared error (RMSE) 0.533
Mean absolute error (MAE) 0.314

Table 2 Predictive
Measures Values
performance of support vector
regression using NCD dataset Correlation coefficient 0.816
Mean squared error (MSE) 0.275
Root mean squared error (RMSE) 0.524
Mean absolute error (MAE) 0.307
846 R. Manjula and M. S. Vijaya

Table 3 Predictive performance of logistic regression using BCD dataset

Measures Values
Correlation coefficient 0.896
Mean squared error (MSE) 0.270
Root mean squared error (RMSE) 0.519
Mean absolute error (MAE) 0.301

by Support Vector Regression. The error rate is minimized in Logistic Regression

when compared to other two algorithms so the reliability of the system is improved.
Hence, it is concluded that Logistic Regression is appropriate than other algorithms
for evaluating the web content credibility and the comparative results are presented
in Table 4 and illustrated in Fig. 2.

Table 4 Comparative analysis of regression algorithms

Algorithms Correlation Mean squared Root mean Mean absolute
coefficient error squared error error
Logistic 0.896 0.270 0.519 0.301
regression
Linear 0.875 0.285 0.533 0.314
regression
Support vector 0.816 0.275 0.524 0.307
regression

2.5
2
1.5
1
0.5
0
Logistic Linear Support
Regression Regression Vector
Regression
Correlation Coefficient Mean Squared Error
Root Mean Squared Error Mean Absolute Error

Fig. 2 Comparison of prediction algorithms

Measuring Web Content Credibility Using … 847

5 Conclusion

This paper demonstrates the modeling of web content credibility using machine
learning techniques. The content credibility factors are recognized and derived by
collecting the real-time data. Two distinct datasets have been created for the variations
of binary and numeric credibility values of web page content. A predictive model
is designed and built for web content credibility evaluation based on a web content
dataset using supervised learning algorithms such as Linear Regression, Logistic
Regression, and Support Vector Regression. The performance of credibility predic-
tion model is measured using different metrics such as correlation coefficient, mean
squared error, root mean squared error, and mean absolute error. Through the exper-
iment, it is observed that the Logistic Regression model is best fitted for predicting
the web content credibility. This work can be further extended by adding more web
content instances and dimensions and repeating the experiment with other advanced
technique.

References

1. Wierzbicki A, Adamska P, Abramczuk K, Papaioannou T, Aberer K, Rejmund E (2014)

Studying web content credibility by social simulation, June 2014
2. Kakol M, Nielek R, Wierzbicki A (2017) Understanding and predicting web content credibility
using the content credibility corpus. Elsevier
3. Abdulla RA, Garrison B, Salwen M, Driscoll P, Casey D (2014) The credibility of newspapers,
television news, and online news
4. [Link]
5. Fogg BJ, Marshall J, Laraki O, Osipovich A, Varma C, Fang N, Paul J, Rangnekar A, Shon J,
Swani P, Treinen M (2001) What makes web sites credible? A report on a large quantitative
study. In: Proceedings of the SIGCHI conference on human factors in computing systems.
ACM
6. Olteanu A, Peshterliev S, Liu X, Aberer K (2013) Web credibility: features exploration and
credibility prediction. In: European conference on information retrieval. Springer, pp 557–568
7. Joo Chung C, Kim H, Hyun Kim J (2010) An anatomy of the credibility of online newspapers
8. Wawer A, Nielek R, Wierzbicki A (2014) Predicting webpage credibility using linguistic
features
9. Aladhadh S, Zhang X, Sanderson M (2014) Tweet author location impacts on tweet credibility
Nov 2014
10. Tumasjan A, Sprenger TO, Sandner PG, Welpe IM (2010) Predicting elections with Twitter:
what 140 characters reveal about political sentiment
11. Carlos C, Marcelo M, Barbara P (2013) Predicting information credibility in time-sensitive
social media 23(5):560–588
12. Korfiatis NT, Poulos M, Bokos G (2006) Evaluating authoritative sources using social networks:
an insight from Wikipedia. Online Inf Rev 30(3):252–262
13. Liviu L Predicting product performance with social media. Inf Educ 15(2):46–56
14. Sundar SS (2008) The main model: a heuristic approach to understanding technology effects
on credibility, pp 73–100
15. Sikdar S, Kang B, ODonovan J, Höllerer T, Adah S (2013) Understanding information
credibility on twitter. IEEE
16. Kak S (1999) Faster web search and prediction using instantaneously trained neural networks.
IEEE Intell Syst 14:79–82
Survey on Privacy-Preserving and Other
Security Issues in Data Mining

K. Karthika, R. Devi Priya and S. Sathishkumar

Abstract In present day, an ever-increasing number of researches in information

mining increases the seriousness about security issue. The security issues in infor-
mation mining can’t just be tended by limiting data integration or by reducing the
utilization of information technology. So as to keep up the security of the customer
in the process of data mining, different types of strategies have been proposed that
are dependent on the probabilistic perturbation of information records. Information
mining service requires precise information for their outcomes to be significant;
however, protection concern may impact clients to give fake data. Here, we present
a detailed survey on privacy and security issues on data mining by analyzing differ-
ent techniques from standard publishers of the year from 2010 to 2017. Based on
the techniques utilized and types of issues are analyzed and classified. Moreover, to
indicate the improvement and accuracy of all the research articles is also discussed.
Furthermore, the analysis is carried to find the importance of their approaches so that
we can develop a new technique to solve the security threats.

Keywords Data mining · Privacy and security issues · Security threats

1 Introduction

The present-day society achieved a lot of advancements from the advanced science
and its applications. It provides effective platform to convert the entire burdensome
task into easy tasks. Therefore, many industries and enterprises are focusing on
their service to be adopted based on advanced technologies [1]. The development
of Internet of Things (IoT) had paved the way to analyze the participatory sensing
(PS) where various individuals can gather and outline their information to an out-
sider information mining cloud administration [2]. The IoT based gadgets directs an

K. Karthika (B) · S. Sathishkumar

Department of Information Technology, Adhiyamaan College of Engineering, Hosur, India
e-mail: karthikainfoscience@[Link]
R. Devi Priya
Department of Information Technology, Kongu Engineering College, Erode, India
© Springer Nature Singapore Pte Ltd. 2020 849
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
850 K. Karthika et al.

outstanding measure of information in streams from different sensors by means of

everlasting communication [3]. IoT has reformed numerous fields including health
services, prosperity applications, public activity, environment observing, transporta-
tion, energy, and so on. The accessibility of minimal effort unavoidable detecting
gadgets has empowered IoT to develop in a regularly expanding way, thus IoT sen-
sors had turned into an indispensable source of big information. IoT incorporates
the application territories of WSN and RFID with extensional availability along the
web.
IoT has numerous applications related to intelligent healthcare services that are
controlled by the technical union of Wireless Body Area Network (WBAN), WSN,
and mobile crowdsensing (MCS) along with distributed computers [4].
The registered and specialized devices utilized by the cyber-physical IoT empow-
ered system create huge amounts of information. This information gives new accesses
yet, in addition, various difficulties arise particularly through their social ramifica-
tions [5]. Nowadays, Cloud storage is also very popular due to its service to preserve
the user data in the third-party platform. It has extraordinarily changed the manner
in which individuals store their information. Not at all like conventional storing, the
cloud gives gigantic capacity of storage which gives more benefits to the clients [6].
The outsourcing of users’ data into the cloud doesn’t concern about the storing of
information and maintenance [7]. Associations and people are slanted to utilize the
cloud service and takes advantages of the productive and practical nature of the cloud.
Cloud services, for example, Google Safe Browsing, PhishTank, and Malwr offers
blacklist of known vindictive URLs, domain, emails and so on [8]. The ceaselessly
expanding measures of information make their security a challenging and essen-
tial task especially when the information are exceptionally dimensional. Numerous
information holders need to distribute their miniaturized scale of information for
different purposes so that does not uncover the identity of individual [9].
In addition, some protection-sensitive system, for example, money-related and
medicinal services does not share their information freely because of the diverse
security strategies, yet joint information handling was unavoidable. Uncovering the
patients’ records is not just untrustworthy yet in addition unlawful as per the Health
Insurance Portability and Accountability Act [10]. The profound learning by clin-
ical and bio-medical analysts on their neighborhood data set overfit the learning
model and brings about wrong outcomes amid the deduction stage. For this situa-
tion, secrecy and security confinements fundamentally lessen utility [11]. Besides,
various new methodologies have been advanced so as to control quickly develop-
ing spam, malware, and DDoS assaults in the Internet to keep client’s information
and to accomplish full reliability from the customers [12, 13]. In this manner, these
distinctive kinds of trust and notoriety instrument were proposed to control various
undesirable traffic delays, for example, email spam. All in all, security protection con-
cerns are identified with verification, information getting to, information encryption
and information distributing [14, 15].
Survey on Privacy-Preserving and Other Security … 851

2 Literature Survey

2.1 Survey of Security Issues in Data Mining

Data mining is characterized as the way toward digging for certain unidentified and
conceivably fundamental data from dreadfully tremendous databases by effective
information disclosure systems. The huge open strategy tensions are the protection
and security of client data and there are getting increased enthusiasm by the gov-
ernment official and controller, protection advocates, and the media. In this article
we consider key online protection, security problems, and concern the job of self-
direction and client in protection and security assurances, information insurance
laws, administrative patterns and the perspective for security and secure enactment.
Normally such a procedure creates new suspicion measurements, identify new attack
designs and lift information security issues. Most latest advancements in data inno-
vations has empowered collection and preparing immense measure of user data such
as criminals database, web-based shopping, online managing of account, credit and
therapeutic history, driving records, and the legislature concerned information [16].
Information mining is depicted as a procedure of find or removing intriguing learn-
ing from immense measures of information put away in a few information sources, for
example, record frameworks, databases, information stockrooms, and so on. There
are a few information mining instruments used to anticipate and bring up patterns and
practices and enables companies to make proactive, learning-driven choices. At most
and mid-run organizations utilizes incredible frameworks for gathering information
and overseeing it in substantial databases. Anyway the bottleneck of transforming
this information into your triumph is the hardest assignment of removing learning
about the framework that you think about from the gathered information. Information
mining and its procedures can be very valuable in numerous regions, for example,
industry, business, government, instruction, and agribusiness, medicinal services,
etc. To break down huge databases utilizing information-digging apparatuses for
conveying answers to questions, for example, Which customers destined to react to
my next limited-time mailing, and why? Information mining has numerous favorable
circumstances yet at the same time it has disadvantages, for example, part of security
issues and hazards. The reason for this paper is to talk about Role of information
mining, its application and different difficulties and issues identified with it [17].
For each economy, industry, association, business work and individual, data has
turned into an essential part. Big Data is needed to distinguish the datasets and esti-
mate the capacity of database programming devices to store, overview and examine.
Thus, Big Data incorporates versatility, stockpiling bottleneck, commotion collec-
tion, fake connection, and estimation mistakes and furthermore presents remarkable
computational and factual difficulties. These issues are identified and new compu-
tational and measurable methods are adopted. This paper shows the survey about
the big data and mining with the issues and challenges. It also discuss about few
techniques to manage huge data [18].
852 K. Karthika et al.

Late development, fame, and improvement of data mining share the perilous risk
on the security of person’s data. Privacy protection and data mining (PPDM) is a
developing exploration theme in information mining and has been widely examined
in late years. PPDM manage the information to perform data mining estimation
properly without degrading the privacy of delicate information presented in the data.
Ongoing research in PPDM principally center around how to decrease the protection
issue caused by data mining, while maintaining the nature of undesirable exposure of
data may occur during the time utilized on data gathering, distributing and conveying.
In this article we chiefly center the security problem recognized during data mining
from a wide point of view and explore diverse methods that can avoid delicate data.
We recognize four kinds of users associated with the application of data mining.
They are data supplier, data gatherer, data excavator, and leader. For each client the
system examines the protection issue and strategies that are received to avert delicate
information. The rudiments of related research points are discussed to present best
survey and presented some fundamental abstraction on future research bearings. We
additionally proposed some survey based on diversion in hypothetical methodologies
for investigating the connections among different users in data mining which has
their own valuation on the touchy information. By classifying the duties of different
users concerning the security of sensitive data we need to give some valuable bits of
information in the investigation of PPDM [19].

2.2 Survey on K-Anonymization Techniques

PPDM techniques have been classified into various types such as Anonymization
based PPDM, perturbation based PPDM, and Cryptography based PPDM.
Exchange information record distinctive data about people, including their analy-
sis and buys, and are progressively distributed to help substantial scale and minimal
effort thinks about in different spaces, for example, drug and showcasing. Be that
as it may, the dispersal of exchange information may prompt security ruptures, as
it enables an assailant to connect a person’s record to their personality. As of late,
there are a few methodologies have been suggested that anonymous information by
killing certain qualities in a person’s record or by supplanting them with regular
qualities, yet they frequently produce information of restricted value due to embrace
esteem change procedures that don’t ensure information utility in planned applica-
tions and target estimates that may prompt over the top information mutilation. In
this review, we propose a novel methodology for anonymizing information to ful-
fill information distributors’ utility prerequisites and causes uninformed misfortune.
To accomplish this, we present a definite data misfortune measure and a successful
anonymization calculation that investigates a huge piece of the issue space. A broad
exploratory examination, utilizing clickstream, and restorative information, exhibits
that our methodology permits commonly more exact question replying than the best
in class techniques, while it is practically identical to them as far as proficiency [13].
Survey on Privacy-Preserving and Other Security … 853

Prior to the production of delicate information, the anonymization of inquiry logs

is a vital procedure that should be performed earlier. This guarantees the namelessness
of the clients in the record, an issue that has been as of now found in discharged logs
from surely understood organizations. This study utilizes small scale conglomeration
to presents the anonymization of question record. Our proposition guarantees the k-
namelessness of the clients in the question log while saving its utility. We demonstrate
the security and utility accomplished, and furthermore provide the analysis of our
proposition in real inquiry record, just as giving assessment to the utilization of data
in data mining forms based on grouping [20].
Anonymization of static information has been incredible research as of late. There
are two normal innovations for semi-identifier’s anonymizations are speculation and
concealment. Anonymization of information contrast from the anonymization of
static information dependent on the characteristics of information streams, for exam-
ple, potential interminability and high dynamicity. For anonymizing information
streams, static information anonymization strategies can’t be straightforwardly con-
nected. In this review, grouping based novel k-anonymization approach is proposed.
The new methodology examines a stream in one swing to perceive and reuse the
groups, so as to accelerate the anonymization procedure and decrease the data mis-
fortune and fulfills the k-obscurity guideline. It considers the time limitations on tuple
distribution and bunch reuse, which are explicit to information streams. Besides, the
methodology is improved to adjust the—decent variety standard. The analyses led on
the genuine datasets demonstrate that the proposed techniques are both productive
and powerful [21].
Information mining assumes an imperative job in examining the substantial mea-
sure of information gathered in this day and age. Protection Preserving Data Mining
(PPDM) systems have turned out to be fundamental because of the open’s rising
consciousness of security and absence of trust in association. A PPDM system guar-
antees singular security while permitting helpful information mining. In this study,
we present two novel information quality assessment procedures called EDUDS
and EDUSC, novel clamor expansion method called Forest Framework and a secu-
rity assessment system called SERS. Novel noise addition technique assembles a
forest from dataset and preserves every one of the examples (rationale rules) of
the backwoods while adding commotion to the dataset. The two frameworks, Forest
Framework to its Predecessor Framework have been thought about, and another setup
strategy, GADP then the correlation is finished utilizing our three assessment crite-
ria, just as Prediction Accuracy. The proposed work demonstrates the test results and
show the accomplishment of expansions to Framework and furthermore the support
of our assessment criteria [22].

2.3 Data Sanitization Issue

Moreover, Fosca Giannotti et al. [11] have developed affiliation rule mining assign-
ment inside a corporate security safeguarding structure. Here an assault display
854 K. Karthika et al.

dependent on foundation information and devise a plan for protection saving redis-
tributed mining. This plan guarantees that each changed thing is unclear as for the
assailant’s experience information, from in any event k − 1 other changed things.
The extensive test is gone up against a substantial and genuine exchange database
exhibit that this method is viable, versatile and ensures protection. Moreover, Jaya
krushna Sahoo et al.
Sahoo et al. [23] have built up a compacted portrayal for affiliation rules hav-
ing insignificant predecessor and maximal subsequent. This portrayal is produced
with the assistance of high utility shut thing sets (HUCI) and their generators. Here
calculations are created to produce the utility-based non-repetitive affiliation stan-
dards and strategies for reproducing all affiliation rules. Moreover, they depicted the
calculations which create high utility thing sets (HUI) and high utility shut thing
sets with their generators. These proposed calculations are executed utilizing both
manufactured and genuine datasets. What’s more, the test results show better quality
in the packed portrayal of the whole standard set under the thought about system.
Li et al. [24] have built up a cloud-helped regular thing set mining arrangement,
which is utilized to assemble an affiliation rule mining arrangement. These arrange-
ments are intended for re-appropriated databases that enable numerous information
proprietors to effectively share their information safely without settling on informa-
tion security. This arrangements release less data about the crude information than
most existing arrangements. In view of these analysis discoveries utilizing distinc-
tive parameters and datasets, they exhibit that the run time in every one of these
arrangements is just a single request higher than that in the best non-protection
saving information mining calculations. The asset utilization at the information pro-
prietor end is exceptionally low since the two information and processing work are
re-appropriated to the cloud servers. Moreover, Jordi Castro and Jose A. Gonzalez
[25] have introduced a substantially less expensive variation of Controlled forbidden
modification (CTA) that defines a multi-objective straight improvement (LO) issue,
where paired factors are pre-fixed, and the subsequent consistent issue is compre-
hended by lexicographic streamlining. Broad computational outcomes are accounted
for utilizing both business (CPLEX and XPRESS) and open source (Clp) solvers,
with either simplex or inside point techniques, on a lot of genuine examples. Most
cases were effectively tackled with the LO-CTA variation in under 60 min, while a
considerable lot of them are computationally pricey with the MILO-CTA detailing.
The inside point technique beat simplex in this specific application.

2.4 Other Issues Based on Cryptography

Database security is imperative as it contains sharpened data. Information digging

gives components to middle person portrayal of information. Protection conservation
is essential because of the security viewpoints. Protection demonstrates components
for enabling the information to be gotten to in verified way. Security is to shield
the delicate information from information excavators with the end goal that it isn’t
Survey on Privacy-Preserving and Other Security … 855

doable to get to the touchy information from database. The significant plans to change
the dataset from the first informational index are affiliation rule. An affiliation rule
with cryptographic procedures has been utilized. This calculation applies on gen-
uine informational collections to exhibit the materialness. An efficient protection
conservation information mining plan with information mining irritation blended
methodology and the affiliation rules with cryptography procedures, proposed in
this examination work. This paper exhibits how a neural system is being connected
for foreseeing the medicinal dataset and furthermore gives scope on how profound
convolution neural system can be connected for therapeutic investigation [5].
Dispersed information mining has assumed a critical job in different applica-
tion areas. In any case, in information mining, it is seen that there are number of
protection risk in person’s delicate data. We propose two conventions to address pro-
tection issue in circulate affiliation rule mining, which are safely creating worldwide
affiliation administers in evenly appropriated databases. Elliptic-bend based Paillier
cryptosystem is the primary convention, which helps in accomplishing the trust-
worthiness and validness of the messages traded among including destinations over
the uncertain correspondence channel. It gives protection of individual site’s data
against the including destinations and an outside foe and an assailant. Notwithstand-
ing, it influence the security of people because of the intrigue of two destinations.
We incorporate Shamir’s mystery sharing plan in the second convention to address
the issue. It gives security by averting conniving locales and outside enemy assault.
We investigate the two conventions to satisfy the protection safeguarding circulated
affiliation rule mining necessities [26].
Present-day mechanical advances have started the prominence and accomplish-
ment of cloud. This as of late created procedure is picking up a growing enthusi-
asm, since it gives cost-productive designs that help the transmission, stockpiling,
and serious figuring of information. Be that as it may, these promising stockpiling
administrations bring a few testing structure issues, for the most part because of both
loss of information control and theoretical nature of mists. The principle objective of
this review is to give a steady view about the two information security concerns and
protection issues that are looked by customers in distributed storage situations. In this
review, it investigates inquire about bearings and innovation patterns to address the
assurance of redistributed information in cloud framework and furthermore brings a
basic near examination of different cryptographic guard component [6].
In present-day protection saving information, mining has been a noteworthy
research. The fundamental objective of this field is to secure individual data and
anticipate revelation of this data amid the information mining process. There are
diverse methods proposed in the field of security safeguarding information mining.
Affiliation rules mining is one systems and the primary motivation behind affiliation
rules mining is to conceal delicate affiliation rules. Up until now, unique calculations
have been exhibited to this field so as to achieve the reason for delicate affiliation
rules stowing away. Every single calculation has its very own particular capacities
and techniques. This paper exhibits an electromagnetic field enhancement calcu-
lation (EFO4ARH) to conceal delicate affiliation rules. This calculation is utilized
856 K. Karthika et al.

Table 1 Overall analysis of

Security issues 2005–09 2010–14 2015–18
survey
Privacy issues [7, 16, 17, 19]
K-anonymization [13, 20, 21] [22]
Data sanitization [11] [23–25]
Other security [27] [5, 6, 26]
issues

to conceal the delicate affiliation administers by uses the information bending sys-
tem. Two wellness capacities are utilized to achieve the arrangement with the least
symptoms in this calculation and furthermore, the run time has been decreased. This
calculation comprises of a strategy for leaving from neighborhood optima point and
advancing toward worldwide ideal focuses. To assess the execution of the proposed
calculation by doing probes both genuine world and manufactured data set. As we
contrasted with the four reference calculations, the proposed calculation demon-
strates a decrease in the symptoms and better protection of information quality. The
execution of EFO4ARH is tried by standard deviation and means Friedman positions
of blunder for standard capacities (CEC benchmarks). Likewise, concealing analyses
demonstrate that our proposed calculation beats existing concealing algorithm [27].

3 Analysis and Discussion

In this section, we categorize the existing research work accordance with techniques.
This analyzes, very much helpful for which algorithm was mainly utilized in the olden
days and how to improve the algorithm further for our research. Here, each research
works are developed different years.
Here the Table 1 and Fig. 1 represents the issues of data mining process under
various years. Most of the researchers in 2015–18 concentrate on data sanitization
issue to protect highly sensitive data.

4 Conclusion and Future Work

Therefore, analyzing those diagnosis techniques can lead to new development in

this area. In the interim, straightforward information sharing constrained diverse
phishing assaults and malware helped security dangers. There are some protection
touchy applications like wellbeing administrations on cloud that are worked with
different financial and operational advantages require upgraded security. Therefore,
precise the internet security and alleviation against phishing rush ended up required
to avoid by and large information protection. In this paper, we have dissected different
issued of information mining procedure, for example, privacy saving, information
Survey on Privacy-Preserving and Other Security … 857

Fig. 1 Categorization based on techniques

cleansing, K-anonymization, data security on touchy information and so on. The

investigation portrays, to secure the protection data of information proprietor from
information clients and specialist organization, cleansing methodology is utilized in
this setup. The sterilized data can’t be recovered until the cleansing key is gotten.
Our proposition gives a productive answer for securing information proprietor’s data
in database.
So as to beat these security issues, we have built up an ideal method for verifying
touchy information.

References

1. Blanco-Justicia A, Domingo-Ferrer J (2018) Efficient privacy-preserving implicit authentica-

tion. Comput Commun 125:13–23
2. Lyu L, Bezdek JC, Law YW, He X, Palaniswami M (2018) Privacy-preserving collaborative
fuzzy clustering. Data Knowl Eng, corrected proof, Available online 12 May 2018 (in press)
3. Nakamura Y, Harada K, Nishi H (2018) A privacy-preserving sharing method of electricity
usage using self-organizing map. ICT Express 4(1):24–29
4. Ali I, Khan E, Sabir S (2018) Privacy-preserving data aggregation in resource-constrained
sensor nodes in Internet of Things: a review. Future Comput Inf J 3(1):41–50
5. Rajesh N, Arul Lawrence Selvakumar A (2018) Association rules and deep learning for
cryptographic algorithm in privacy preserving data mining. Clust Comput 1–13
6. Kaaniche N, Laurent M (2017) Data security and privacy preservation in cloud storage
environments based on cryptographic mechanism. Comput Commun 111:120–141
7. Jaseena KU, David JM (2014) Issues, challenges, and solutions: big data mining. Comput Sci
Inf Technol (CS & IT) 131–140
8. Dara S, Zargar ST, Muralidhara VN (2018) Towards privacy preserving threat intelligence. J
Inf Secur Appl 38(February):28–39
9. Abdelhameed SA, Moussa SM, Khalifa ME (2018) Privacy-preserving tabular data publishing:
a comprehensive evaluation from web to cloud. Comput Secur 72:74–95
10. Ma X, Zhang F, Chen X, Shen J (2018) Privacy preserving multi-party computation delegation
for deep learning in cloud computing. Inf Sci 459:103–116
858 K. Karthika et al.

11. Giannotti F, Lakshmanan LVS, Monreale A, Pedreschi D, Wang H (2013) Privacy-preserving

mining of association rules from outsourced transaction databases. IEEE Syst J 7(3):385–395
12. Zhang L, Yan Z, Kantola R (2017) Privacy-preserving trust management for unwanted traffic
control. Fut Gener Comput Syst 72:305–318
13. Loukides G, Gkoulalas-Divanis A (2012) Utility-preserving transaction data anonymization
with low information loss. Expert Syst Appl 39(10):9764–9777
14. Tudor V, Almgren M, Papatriantafilou M (2018) The influence of dataset characteristics on
privacy preserving methods in the advanced metering infrastructure. Comput Secur 76:178–196
15. Wang Y, Cai Z, Tong X, Gao Y, Yin G (2018) Truthful incentive mechanism with location
privacy- preserving for mobile crowdsourcing systems. Comput Netw 135:32–43, 22 April
2018
16. Singh DK, Swaroop V (2013) Data security and privacy in data mining: research issues &
preparation. Int J Comput Trends Technol 4(2):194–200
17. Sharma, BR, Kaur D, Manju A (2013) A review on data mining: its challenges issues and
applications. Int J Curr Eng Technol. ISSN 2277-4106
18. Ni L, Yuan Y, Wang X, Zhang M, Zhang J (2017) A location privacy preserving scheme
based on repartitioning anonymous region in mobile social. In: The proceedings of the 2017
international conference on identification, information and knowledge in the internet of things
Procedia computer science, vol 129, 2018, pp 368–371
19. Xu L, Jiang C, Wang J, Yuan J, Ren Y (2014) Information security in big data: privacy and
data mining. IEEE Access 2:1149–1176
20. Navarro-Arribas G, Torra V, Erola A, Castellà-Roca J (2012) User k-anonymity for privacy
preserving data mining of query logs. Inf Process Manage 48(3):476–487
21. Guo K, Zhang Q (2013) Fast clustering-based anonymization approaches with time constraints
for data streams. Knowl-Based Syst 46:95–108
22. Fletcher S, Zahidul Islam Md (2015) An anonymization technique using intersected decision
trees. J King Saud Univ-Comput Inf Sci 27(3):297–304
23. Sahoo J, Das AK, Goswami A (2015) An efficient approach for mining association rules from
high utility itemsets. Expert Syst Appl 42(13): 5754–5778
24. Li L, Lu R, Choo K-KR, Datta A, Shao J (2016) Privacy-preserving-outsourced association rule
mining on vertically partitioned databases. IEEE Trans Inf Forensics Secur 11(8):1847–1861
25. Castro, J, González JA (2017) A linear optimization-based method for data privacy in statistical
tabular data. Opt Methods Softw 1–25
26. Chahar H, Keshavamurthy BN, Modi C (2017) Privacy-preserving distributed mining of asso-
ciation rules using Elliptic-curve cryptosystem and Shamir’s secret sharing scheme. Sadhana
42(12):1997–2007
27. Teng Z, Du W (2009) A hybrid multi-group approach for privacy-preserving data mining.
Knowl Inf Syst 19(2):133–157
28. Upadhyay S, Sharma C, Sharma P, Bharadwaj P, Seeja KR (2016) Privacy preserving data
mining with 3-D rotation transformation. J King Saud Univ—Comput Inf Sci, corrected proof,
Available online 28 November 2016 (in press)
29. Jia J, Yu J, Hanumesh RS, Xia S, Jiang X (2018) Intelligent and privacy-preserving medication
adherence system. Smart Health, accepted manuscript, Available online 5 July 2018 (in press)
30. Demir S, Tugrul B (2018) Privacy-preserving trend surface analysis on partitioned data. Knowl-
Based Syst 144:16–20, 15 March 2018
A Study on Different Types of Robotics
Applications

B. U. Meghana and M. V. Prashanth

Abstract This paper’s main objective is to investigate the different types of robots,
their operation, and application. Different types of robots follow different methods
such as mobility, weight balance, distributed control system, biopsy, current 3G
technology, skid steering, robotic weeding, calculation of Godspeed, suction cups,
and network communication protocol. Here we discuss various types of robots, the
following methods, the parameters used and their use in different fields. Finally, we
can conclude that this paper briefly introduces robotic applicability in various areas.

Keywords Snake robot · Heavy duty robot · Medical robot · Car robot

1 Introduction

Robotics is an interdisciplinary branch of science and engineering. Robotics covers

the structure, development, activity, and use of robots and computer frameworks
for tracking, sensory feedback, and processing records. These advances are used
to build machines to update and imitate human activities. Robots can be used in
many situations and for many features, but in recent times many are used in risky
situations (bomb identification and deactivation), production techniques or wherein
humans cannot tolerate them (e.g., Space). Robots can also take any form, but some
appear to be human beings [1]. This is said to help and accept a robotic behavior that
people generally perform in certain replicative behavior. Such robots try to reflect
walking, lifting, speaking, cognizing and, in essence, something that can be done
by people. Many of the robots today are of course inspired and contribute to the
bio-inspired robotics enterprise.

B. U. Meghana (B) · M. V. Prashanth

Department of Information Science and Engineering, Vidyavardhaka College of Engineering,
Mysuru, Karnataka, India
e-mail: meghana1998megha@[Link]

© Springer Nature Singapore Pte Ltd. 2020 859

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
860 B. U. Meghana and M. V. Prashanth

1.1 Robotics and Artificial Intelligence

A robotic is a machine consisting of sensors, manipulators, power components and

software, all of which work together to meet a challenge. Motivation is the intrinsic
enthusiasm of an employee and the drive to carry out work-related activities. Artificial
Intelligence (AI) Artificial robots are artificial dealers in the real world. AI Robot aims
to manipulate objects through the perception, selection, movement, and destruction
of objects. Robotics is an AI department composed of different branches and robot
alertness. Power supply, actuators, electric motors (AC/DC), pneumatic air muscles,
muscle wires, piezo motors and ultrasonic motors, sensors are components of AI
Robots [2].

1.2 Aspects of Robotics and AI

AI robot has a mechanical design and shape to carry out a specific project. They have
electric components that make the machinery strong and manage it. They include a
few phases of a computer program that determines what a robotic does, when, and
how. AI Robot Locomotion: Locomotion is the mechanism that enables the AI robot
to move around it. Various types of locomotives are available: legged, wheeled, a
combination of the legged and wheeled locomotive, tracked slip/skid. There are five
most important components in the maximum fundamental stage [2].
A frame structure, a muscle gadget to move the frame structure, a sensory device
that gets facts approximately the frame and the encompassing environment, a power
supply to activate the muscle mass and sensors, a mind machine that strategies sen-
sory records and tells the muscles what to do. Applications of Artificial Intelligence
Robotics are Industrial, Medical Applications, Exploration, and Entertainment [2].
There are so many information about Robotics to be acknowledged. The Robots
are used in many fields like agriculture, industry, hotel, household works, hospital,
school, college, etc. Understanding their working is greater vital, in order that if any
problem is given, we can discover the answers by means of creating new varieties of
robots.

2 Literature Review

Chaudhry and Sharma [3] talked about snake robots mobility. In the future, snake
robots hold a lot and have a great scope in India. In future applications such as agricul-
ture, sanitation, and firefighting, serpent robots are used. The advanced technology
helped us to achieve “Beyond Human Capabilities”.
Chung et al. [4], introduced the heavy-duty industrial robot design process partic-
ularly for analytical technology to build the native model of KIMM. This model has
A Study on Different Types of Robotics Applications 861

a payload of 600 kg. For heavy-duty robot in static and dynamic payload more than
300 kg is essential for analysis to manipulate the design to select the basic compo-
nents. Already much-commercialized software is available in the market. Here the
author developed innovative software for inverse dynamic analysis called “Rodan”.
By using this software and its application the weight balancing component selection
is easily made for manipulation.
Wyeth et al. [5], designed an autonomous humanoid robot for playing soccer
with the help of image processing, artificial intelligence, and decision-making. He
controlled the Metal Fighter 2 humanoid robot with the communication system and
process the image with the Kalman filtering algorithm and proposed the 4 step model
based on 6 step reasoning model. Evaluated with the proposed system and got good
stability and competition ability towards humanoid robot soccer system. This paper
will help us to understand the mentality of human and how these robots behave like
human and communicate with a human is described. It works based on Mechanical
CAD model, Bipedal Walking Robots, Motor Choice, Electronics. It will illustrate
the design of a practical, affordable, autonomous, humanoid robot.
Azar and Samy [6] discussed the working of Medical Robots. These robots are
used to assist doctors and nurse. These robots help in telesurgery, brain biopsy,
telemedicine, surgical training, remote surgery, telemedicine, teleconsultation,
rehabilitation therapy, prosthetics.
Kaur and Kumar [7] worked on the wireless multifunctional robot to help the
defense and military applications with the help of 3G technology to monitor the
remote and border areas. This multifunctional robotic vehicle helps to examine the
border areas and it has to be communicated and controlled by an autonomous and
manual system. This robotic has a multifunctional feature like to detect human,
bombs, harmful gases and fire at the remote and border areas. For power consumption,
the robot uses renewable resources with the help of a solar panel. The limitation
behind the system is the frequency of the range. He also assessed the results of the
selection of the tilt angle of the solar panel and the energy consumption for the
autonomous and manual system. Users can control the movement of robots from any
part of the world by receiving live video feedback from the environment. It is used
as a security surveillance camera robot and as a rescue operation.
Shekhawat et al. [8], detailed study on Automatic car robot follows the method of
a wireless controller with the help of Bluetooth. It is used in our real-life like AGV,
used for dropping the parcels. It has the ability to sense the environment and decide
the navigation path without any human input.
Pedersen et al. [9], worked on Agricultural Robots, and developed autonomous
tractors and automatic steered systems for agricultural purpose. It uses robotic weed-
ing, micro-spraying, crop establishment, crop scouting, selective harvesting, eco-
nomic scenarios, field scouting, weed detection methods. It reduces labor costs and
restrictions on the number of daily working hours.
Hotel Service Robots were evaluated by Tussyadiah and Park [10]. Artificial
intelligence and robotics for tourism and hospitality, hotel service robots evaluation,
automatic emotional responses to hotel robots. Confirmatory factor analysis was per-
formed using Smart PLS 3.0, a multi-group analysis that compares path coefficients
862 B. U. Meghana and M. V. Prashanth

between respondents exposed to two different NAO robots, Relay. It is used in the
front desk and delivers items to the room.
Jagtap [11] discussed the use of robots for wall climbing. The study presents
an application of a rising glass and wall cleaning automaton. Automated operation
during movement, large remote control range. Cleaning robots structure, components
used in robots cleaning, robot mechanism, basic design calculations are explained.
It is used in a wide range of maintenance, building, inspection and process, and
construction safety applications.
The applications of ARP were explained by Clotet et al. [12]. The mobile teleop-
erated robot is designed as a live robot. Mechanical design, electronic components,
implementation of software, the architecture of video conferences, network commu-
nication protocol methods. Explains video communication, audio communication,
external network server, preliminary usability test, prevention of collisions. This
paper presents the design and application of a mobile teleoperated robot as a living
tool. It is used as a mobile video conference service, mobile telepresence service,
walking tool, scheduling tool, fall detection tool, mobile monitoring platform for the
environment.

3 Comparison of Different Kinds of Robotics

with Parameters

Table 1 shows the comparison of different kinds of robotics with parameters like
mobility, cost, performance, and flexibility. This comparison will help us to know
little more to achieve the better result in future.

Table 1 Comparison of different kinds of robotics with parameters

Robotics Mobility Cost Performance Flexibility
Snake robots Yes High High Yes
Heavy-duty handling robots Yes Low High Yes
Humanoid robots Yes High High Yes
Medical robots Yes High High Yes
Military robots Yes Low High Yes
Car robots Yes High High Yes
Agriculture robots Yes Low High Yes
Hotel service robots Yes Low High Yes
Wall climbing and glass cleaning robots Yes Low High Yes
Assisted personal robots Yes High High Yes
A Study on Different Types of Robotics Applications 863

4 Discussion

This paper presents the uses of different types of robots, such as snake robot that helps
in the field of agriculture, heavy-duty handling robots used for industrial purposes,
autonomous humanoid robot that helps people interact with machines, medical robots
that help in the field of medicine, military robot that is used as a security camera robot
and emergency rescue. Car robot that helps to drop parcels, an agricultural robot that
reduces labor costs and the number of working hours per day, hotel service robot that
helps to deliver items to the room, Wall Climbing which helps in construction, Tele-
Operated Assisted Living Robot which is used as mobile videoconference service.

5 Conclusion

Robotics is one of the emerging techniques and used across the world to help people.
It will help to a human being and make the work easier to achieve. In this paper, we
discussed the different applications of robotics, components, methods, parameters,
and their uses to create an innovative robot.

Acknowledgements The authors express gratitude towards the assistance provided by Accendere
Knowledge Management Services Pvt. Ltd. in preparing the manuscripts. We also thank our mentors
and faculty members who guided us throughout the research and helped us in achieving the desired
results.

References

1. Wikipedia—Robotics. [Link] Accessed 15 Sept 2018

2. Shailna Patidar.: AI Robot—Robotics and artificial intelligence. [Link]
robot-robotics-and-artificial-intelligence. Accessed 20 Sept 2018
3. Chaudhry N, Sharma S (2015) A review study on future applicability of snake robots in India.
IOSR J Comput Eng 17:2278–2661
4. Chung GJ, Kim DH, Park CH (2008) Analysis and design of heavy duty handling robot. 2008
IEEE Int Conf Robot Autom Mechatron RAM 2008. 00, 774–778
5. Wyeth G, Kee D, Wagstaff M, Brewer N, Stirzaker J, Cartwright T, Bebel B (2001) Design of
an autonomous humanoid robot. In: Proceedings of the Australian conference on robotics and
automation, pp 14–15
6. Azar AT, Samy E (2012) Medical robotics. In: Prototyping of robotic systems: applications of
design and implementation, p 35
7. Kaur T, Kumar D (2015) Wireless multifunctional robot for military applications. In: 2015
2nd international conference on recent advances in engineering and computational sciences,
RAECS 2015
8. Shekhawat RS, Sain A, Bhardwaj G (2016) Automatic intelligence car robot. Int J Sci Technol
Eng 2:504–508
9. Pedersen SM, Fountas S, Blackmore S (2008) Agricultural robots—applications and economic
perspectives. Service Robot Appl 369–382
864 B. U. Meghana and M. V. Prashanth

10. Tussyadiah I, Park S (2018) Consumer evaluation of hotel service robots. In: Information and
communication technologies in tourism, pp 308–320
11. Jagtap A (2013) Skyscraper’s glass cleaning automated robot. Int J Sci Eng Res 4:806–810
12. Clotet E, Martínez D, Moreno J, Tresanchez M, Palacín J (2016) Assistant personal robot
(APR): conception and application of a tele-operated assisted living robot. Sens (Switzerland)
16:1–23
UVRento—Online Product Rental
Application Using Cross-Platform
Service

Rahul Patil, Siddaram Sonnagi, Ashwini Dhummal, Subhas Kadam

and V. Sanju

Abstract According to the current situation of high demand for mobile applications,
combined with the mainstream and the enterprise activities (Yang and Zhang in
Computing technology, industrial information integration, [1]), the emergence of
E-commerce has been playing a vital role in each and every sector of our daily
lives. The number of new and different product developments increasing each year,
triggers probability of previous technology users, especially in electronic gadgets.
There are at least 40% of people who change their smartphones, laptops or any
other electronic devices in a minimum of 2–3 years (by social interactions). Hence
the cost of living is increasing steadily. For instance, cost of living in Bangalore in
the year 2015 was around 5000Rs–6000Rs, and now it has been increased around to
7000–10,000 per month ((B)Grady Booch Rational Santa Clara, California in object-
oriented analysis and design, [2]). Hence, Rental applications may reduce cost of
living without affecting the mandatory life choices and their preferring lifestyle. This
paper gives a detailed hands-on experience with the Rental Platform, UVRento, and
how it can be implemented using E-commerce model application services. UVRento
is an application developed using Ionic Framework with Firebase as its backend.
Ionic being a cross-platform framework, it makes easier to develop powerful and
efficient applications for every major platform running out there (iOS, Android,
Windows Phone, etc.). While Firebase along with Google’s security provides the

R. Patil · S. Sonnagi (B) · A. Dhummal · S. Kadam · V. Sanju

Department of Computer Science and Engineering, NMAMIT, Nitte, Karnataka, India
e-mail: [Link]@[Link]
R. Patil
e-mail: rahultpatil007@[Link]
A. Dhummal
e-mail: ashwinidhummal08@[Link]
S. Kadam
e-mail: subhaskadam840@[Link]
V. Sanju
e-mail: sanjuv21@[Link]

© Springer Nature Singapore Pte Ltd. 2020 865

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
866 R. Patil et al.

database facility which is more robust, more reliable and easy to integrate with Ionic
Framework (Ionic framework document, [3]; Learn firebase—tutorials, [4]).

Keywords UVRento · Mobile application · Hybrid · Cross platform · Rental

application service · Ionic · Firebase · Cost of living · Local marketing

1 Introduction

Most people want to rent out items to save their money. Hence they use the Online
Rental website system where they can get their items in lower prices and for a short
period of time. But in these systems, there is a communication gap between the real
vendors who give the items and the customers who actually use the items because,
in these systems, the Rental website organization buys the item from the vendor and
give it to the customers. Hence in the present market, there is a need of bridging the
gap between Customers and the Vendors for the benefit of the both.
• Our system UVRento aims to bridge the gap between the Customers and Vendors
[5, 6].
• Online platform where users can rent or buy items like electronic gadgets, books,
cycles, etc. based on their locality [7].
• This leads to efficient local marketing [8].

2 Proposed Business and Implementation Module

A new online rental service business model and Implementation modules in

e-commerce industry—UVRento application—is presented in this section.

2.1 Modules in UVRento Application

1. Admin:
• Register
• Login
• Verify the vendor details
• Verify the customer details
• Approve/Reject products
2. Vendor:
• Register
• Login
UVRento—Online Product Rental Application … 867

• Upload/Remove products
• Modify details
• Logout

3. Customer:

• Register
• Login
• Search product
• Place my order
• View my order
• Logout

2.2 Use Case Diagrams

Use Case Diagram (Fig. 1):

Use Case Diagram (Fig. 2):

2.3 Business Model with Use Case Diagrams

Vendor Side: Once vendor register their ID then about the product description their
product is not uploaded! But our admin verifies the product quality, quantity and

Fig. 1 Use case diagram for

profiles [2]
868 R. Patil et al.

Fig. 2 Use case diagram for

admin module [2]

the cost of the product. If our admin is satisfied and pass all the product check then
the product is given green signal and it’s uploaded to the UVRento platform! This
is done because to stop the fake product description uploaded on web by the direct
vendor…! And it this we don’t support the fake items.
Customer Side: Any customer who wants to take for Rent/Buy the products
will create an account and puts the selected products on cart, thus he places an
order. The payments of these products are always depends on customer and vendor
communication (Fig. 3).

2.4 Implementation Module

User Registration: User may be either Vendor or Customer, he need to register

by submitting required details such as mainly Name, Address, Aadhaar card, UID
number, PAN card, etc. Aadhaar card must be scanned using UVRento Scanner
which provided in application while user registration setup. Thus UVRento achieve
security over users.
Profile:
• Basically, UVRento has two profiles Vendor and Customer. Any user is allowed
to change his profile and he can act in both the profiles simultaneously as per his
requirement.
• Being vendor he can only perform sell/Put for Rent operations, being Customer
he can only perform buy/Take for Rent operations.
UVRento—Online Product Rental Application … 869

Fig. 3 Data flow diagram of business model

• By activating both the profiles he can perform Buy/Sell/Put for rent/Take for Rent
operations.
Vendor Side: Registered vendor is allowed to put any products for Rent by paying
the specified charges. Charges are different for each product; it depends on product
factors such as original cost, usage, damages, etc.
Admin (Specifically: Verification Team):
• Admin who is in the verification department will verify the product that vendor
has requested to put product on online UVRento platform for marketing.
• Admin can approve the product if it meets all the terms and conditions or Reject
the Vendor Product if it did not meet terms and conditions after verification.
UVRento Platform: After admin verification and approve procedures, vendor
product will be displayed on the marketing platform (UVRento) with all the required
details and specifications provided by the vendor.
870 R. Patil et al.

Customer Side:
• Registered customer can search and browse the details of available products such
as specifications, rate, date, ratings, etc.
• If he is interested in any product he can click Request item to notice vendor that
he is ready to take it for rent.
• If he gets positive action from vendor then UVRento displays vendor location,
phone number, and some other required details.
• By the provided details he can call/message and arrange a meet or courier the item
whatever, it’s completely depends on both of them.
• Thus, UVRento helps customer to get connected with vendor and to order required
product for rent.

2.5 Payments and Delivery

Payment and delivery method completely depends on Customer and vendor interac-
tions. They may go through doorstep delivery or courier process in delivery process
or online payment/cash on delivery in payment process.

2.6 Literature Survey

People nowadays don’t want to spend on products like Mobiles, Cameras and all
such electronic devices that lose up to 50% of its value the moment they purchase
it (Depreciation). They just want to use and experience the feel of such products
at lower prices. Hence Item Renting Websites such as [Link], Furlenco,
GrabOnRent, Rentickle, etc. have seen a rise in their usage. But all these item renting
websites systems have one thing in common; they all buy the items directly from the
vendors. This system doesn’t build up communication between the vendors and the
customers who actually use the products. And in some cases, the vendor may occur
losses due to selling of the items which may have potentiality of returns more than
its cost price. This benefit is taken by these existing systems.
UVRento—Online Product Rental Application … 871

3 Screenshots of User Interface (UI) Design

See Figs. 4, 5 and 6.

Fig. 4 Menu bar—menu bar is the button used to operate all the pages and activities of UVRento
app (References—UVRento application) [3, 4]

Fig. 5 Home—after user login home page acts as master node where user can get the nearby
available products for rent in UVRento application (References—UVRento application) [3, 4]
872 R. Patil et al.

Fig. 6 Add product—is a

menu node where user can
add his products for rent on
daily, weekly, monthly and
for sale basis
(References—UVRento
application) [3, 4]

4 Results and Discussion

• Users can get items without actually having to buy it, thus saving money.
• More Usage—Less Cost for Users.
• Vendors can collect their revenue easily with Renting system.
• Users can get items based on their requirements, for a specific period of time.
• UVRento can improve local marketing of Vendors.
• Rental Marketing is a Solid way and Scalable Marketplace.
• With the help of this project, people need not waste money for short term materials.
Helps to improve Local marketing. 24*7 Product showcase. Enables more usage
less cost for users, it may lead to customer-friendly platform.

References

1. Yang Y, Zhang Y (2017) Computing technology, industrial information integration. In: IEEE con-
ference 2017—mobile terminal development plan of cross-platform mobile application service
platform based on Ionic and Cordova
2. (B)Grady Booch Rational Santa Clara, California—Object oriented analysis and design. ISBN
0-8053-5340-2 15 1617181920 DOC 0 1 00 99 98 l5th Printing, Dec 1998
3. [Link] framework document
4. [Link] to learn firebase—tutorials
5. Mutiawani V, Rahmany S, Abidin TF (2018) Informatics department. In: IEEE conference, 2018
international conference on electrical engineering and informatics (ICELTICs), Sept 19–20, 2018
6. Bosnic S, Papp I, Novak S. The development of hybrid mobile applications with Apache Cordova.
Research and Development Institute RT-RK, Novi Sad, Serbia
7. Ravulavaru A. Learning ionic. Packt Publishing, Birmingham. ISBN:9781783552603
8. [Link] of living records
Autonomous Unmanned Ground Vehicle
for Enhancement of Defence Strategies

M. Someshwaran, Deepa Jose and P. Paul Jefferson

Abstract This paper explains about the Autonomous Unmanned Ground Vehi-
cle named Lakshman, which has the following features—carry payload (food and
medicines), follow the soldier like a companion, autonomous return to the base and
it can also locate an enemy, etc. The vehicle can be remotely controlled if required.
The technology been targeted are limited cross-county mobility, autonomous fol-
lowing techniques, localization, perception and path planning. It can carry stores
and ammunition for three days. The implementation setup is been divided into four
sections. They are mechanical, power management, core control, and autonomous
section. Lakshman can work continuously for three days. The frame structure is
made using mild steel to carry a payload around 200 kg. Power management unit
makes the device to sustain more working hours, making it reliable and flexible to
the requirement. The core control section controls the entire process and collects
data from various sensors. Autonomous section controlled by ARM cortex A53 to
process high data speeds. The most important future plan is to provide steganography
method of communication, which is a highly secured communication, so the vehicle
cannot be easily tracked by other people with enhanced technology. By implementing
a dedicated FPGA for encrypted steganography communication will provide greater
assist to the A-UGV. Thus Lakshman can be a very effective tool to enhance Indian
Defence Strategies.

Keywords SLAM—Simultaneous localization and mapping · ROS—Robot

operating system · AUGV—Autonomous unmanned ground vehicle

M. Someshwaran · D. Jose (B) · P. Paul Jefferson

Electronics and Communication, KCG College of Technology,
Karapakkam, Chennai 600097, India
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 873

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
874 M. Someshwaran et al.

1 Introduction

The growth of technology in the area of Unmanned Ground Vehicles has become
vast. The research in autonomous vehicles and artificial intelligence has infinite
applications. Using artificial intelligence in the field of autonomous vehicle would be
a great revolution, and dynamic in the field of defence. According to India’s defence
vehicles, this method of implementation of A-UGV provides robust and powerful
strategies in defence. This paper explains about the Autonomous Unmanned Ground
Vehicle which can do the following features: follow a solider, autonomously return
to the base station, transmit live video footage to the receivers end, and also able
to control the vehicle in a semi-autonomous way. The vehicle has four sections
in it mechanical, power management, core control and the autonomous sections,
respectively.

2 Implementation Setup

A. Mechanical Section:

For any vehicle, the frame of the vehicle is very crucial for it to withstand weights [1].
Lakshman is carefully designed to carry varying and flexible payloads up to 200 kg.
The frame structure is built using mild steel whose density is about 7.85 g/cm3 . Mild
steel is cheap and it is available in a large amount. Lakshman is strongly supported by
Caterpillar Tracks which provides high stability to vehicle and have better mobility
over rough terrains [1]. Track section consists of two sprockets on each end which
is permanently connected to High torque driving motor which finds better friction
during mobility [1]. The tension on the wheels is equally distributed through the
effective shock absorbers with hydraulic. This structure strongly supports the frame
from all sides. The basic version of Lakshman is concentrated to carry food and
medical supplies to soldiers. The way in which the vehicle has to be used can be
changed according to our need and purpose for various platforms like easy trans-
portable module, Articulated platform, Medevac, Supply transport, Reconnaissance,
UAV platform and mainly as Remote weapon station [2, 3].
B. Power Management Unit:
Power is the basic requirement for any vehicle. To make the system more efficient
UGV is been provided with both diesel as well as electrical power sources [1].
Lakshman consists of 1400 VA Diesel/petrol generator which acts as primary source
for the UGV. By using diesel generator instead of standalone Battery source enhances
the system efficiency higher. The reason for avoiding batteries as the main source
for the power consumption is that batteries require double the amount of time to
recharge again to be used. Whereas in the case of generators one can fill the diesel
and use it efficiently. Lakshman produces an AC output of 220 V @50 Hz and a
DC output of 12 V. The AC output is efficiently stepped down to 24 V to drive the
Autonomous Unmanned Ground Vehicle … 875

250 W high torque motors which is connected to the sprockets using a shaft. The
power management unit in the UGV is designed in such a way that it meets all the
constraint and it can work for longer hours more efficiently. Simultaneously two
12 V, 100 A which acts as secondary source. Batteries are charged and used to avoid
stall time. Water cooling system makes the vehicle to withstand from overheating.
C. Core Control and Communication Unit:
The key feature of this section is that it collects data from various sensors and controls
entire A-UGV. Lakshman is controlled by a 16 MHz microcontroller Arduino and
1.2 GHz microprocessor Raspberry pi v3. It comprises of different programmed func-
tions which makes it to work autonomously. The power consumption of this section
is very low compared to other sections. The system is made to work autonomous
with the help of the Kinect sensor which uses the LIDAR technology. It analyses
the depth of every object surrounding it and makes the vehicle to avoid disturbance.
Lakshman is controlled through multiple frequency bands 433 MHz and 5.8 GHz
[2, 3]. As 433 MHz frequency band has a high penetration through trees it is used
for the data control and 5.8 GHZ is used for video transmission and also to get a
clear quality of the data been transmitted. 5.8 GHZ is used for dedicated video trans-
mission approximately less than 5 km. The 433 MHz data transceiver unit is used
to wirelessly guide Lakshman from a long-distance [3]. Dedicated CMOS camera
is been provided to monitor the vehicle as a live feed and it is boosted by 5.8 GHz
video transmission unit.
D. Autonomous Unit:
Autonomous part of A-UGV has two features, following the group and return to
base. Lakshman is operated to a distance of 20 km from the controller side. The
vehicle can be controlled by the person at the control section if required. Autonomous
functionality is the most crucial system in this UGV where it processes numerous data
sets instantly by using 1 GB of flash that located within the Raspberry Pi. Autonomous
3d mapping technology is done through LIDAR system where it collects data through
Kinect sensor system and it provides visual data in a 3d architecture which makes the
UGV more precise in identifying the objects and location [4]. Green ant algorithm
is used to provide an optimal route planning which provides the shortest route to
the destination. It uses simultaneous localization and mapping technology. The G-
mapping package provides laser-based SLAM as a ROS node called slam G-mapping
[5, 6].
By using GPS positioning system that gets saved by the user will provide an
additional information to the UGV to return to base. By saving the initial base location
it can easily return to base with an optimal and shortest route possible [4].

3 Calculation of Remote Access

See Table 1 and Fig. 1.

876 M. Someshwaran et al.

Table 1 The calculation of

Required 433 MHz 5.8 GHz
remote access for 433 MHz
parameter transmission unit transmission unit
and 5.8 GHz
Transmit antenna 2 dB 2 dB
gain
Transmitting 17 dB 23 dB
power
Operating 433 * 106 5:8 * 109
frequency (Hz)
Receiver −70 dB −90 dB
sensitivity
Free space loss 107 dB −90 dB
Total estimated 1554.65 2315.75
range (meters)

Fig. 1 Sample image of the

UGV

4 Discussion of Results

A. Prototype
A small prototype of entire model is been made using thermocol material (Fig. 2).
B. Raspberry Pi Testing
Raspberry Pi is been interfaced with the libraries such as the Open NI and Open CV.
Programs which are required to make the motors run are been tested and their results
are successful. Motor runs with high amount of torque but at a slight slower speed.
C. Stability and Stress Analysis
A stabilized and more rigid UGV which can hold up to 500 kg has been tested. It is
found that the frame structure can withstand 500 kg without causing any degradation
in the performance of the vehicle. Below shown image is the frame structure of the
UGV (Fig. 3).
Autonomous Unmanned Ground Vehicle … 877

Fig. 2 Prototype model

Fig. 3 Analysis of frame structure

D. Object Detection
To make the vehicle detect a particular thing and to follow it is been tested. For this
purpose, a red-coloured object is been made to detect by the vehicle using the MAT
LAB. Results obtained are: the vehicle follows the red-coloured object (Fig. 4).
E. Outlined View of the UGV
The below-shown image displays an outlook of how the finished product of the UGV
will appear.
878 M. Someshwaran et al.

Fig. 4 Red colour object detection using MAT LAB

F. Suspension and Wheel Assembly

Types of suspension and the placement of the wheel assembly at the appropriate
places are been tested and their results are been obtained (Figs. 5, 6 and 7).
G. Dimensions
The below-shown image gives a complete view on how the dimensions of the cater-
pillar track structures are been designed. The measurements given are been given in
millimetres.
H. Skeletal Tracking
Human following is important function where UGV takes control of skeletal tracking
which is more precise and accurate. ROS system track each joints and follows back
(Figs. 8 and 9).

Fig. 5 The final outlined view of the UGV

Autonomous Unmanned Ground Vehicle … 879

Fig. 6 Image of the suspensions been designed and the placement of the wheel assembly

Fig. 7 Dimensions of the caterpillar tracks

I. Collision Avoidance
3D mapping is built using ROS system which is effectively used for collision avoid-
ance, path planning, and optimization, autonomous return to base. skeletal tracking
is very important for instructing the vehicle to follow the right solider who is been
assigned previously for this purpose of making the vehicle to follow him, respectively.
880 M. Someshwaran et al.

Fig. 8 Skeletal tracking for human following

Fig. 9 3D mapping for collision avoidance and path planning

References

1. Carlson J, Murphy RR (2005) How UGVs physically fail in the field. IEEE Trans Robot 21(3)
2. Noor MZH (2013) Design and development of remote-operated multi directional unmanned
ground vehicle (UGV). In: IEEE 3rd international conference on system engineering and
technology, 19–20 Aug 2013, Shah Alam, Malaysia
3. Murtaza Z (2014) Design and implementation of low cost remote-operated unmanned ground
vehicle (UGV). In: 2014 international conference on robotics and emerging allied technologies
in engineering (iCREATE), Islamabad, Pakistan, Apr 22–24, 2014
4. Seo DJ, Kim J (2013) Development of autonomous navigation system for an indoor service
robot application. In: 2013 13th international conference on control, automation and systems
(ICCAS 2013), Oct 20–23, 2013, Kimdaejung Convention Center, Gwangju, Korea
5. Jabbarpour MR, Jung JJ, Kim P (2017) A green ant-based method for path planning of unmanned
ground vehicles. Received Dec 30, 2016, accepted Jan 20, 2017, date of publication Jan 25, 2017,
date of current version Mar 13, 2017. [Link]
6. Ullah I, Ullah F, Ullah Q (2015) Real-time object following fuzzy controller for a mobile robot.
IEEE
A Parametric Study of Load Balancing
Techniques in Cloud Environment

Insha Naz, Sameena Naaz and Ranjit Biswas

Abstract Cloud Computing is one of the rapidly growing areas in the field of
computer science. A modern paradigm provides services through the Internet. An
Internet-based technology employs pay-as-you-go model (PAYG). Load balancing
is one of the important and vital issues in cloud computing. It (load balancing) is a
technique which improves the distribution of workloads across various nodes. Load
balancing distributes dynamic workload among various nodes so that no specific
node breaks down by heavy load. It is crucial to utilize the full resources of a parallel
and distributed system. Resource consumption and energy consumption both can be
limited by distributing the load evenly and properly using different load balancing
techniques. Lowering the use of resources increases the overall working performance
of the system thus cutting down the carbon emission rate and providing the greener
and safer environment. Different algorithms and techniques are employed to balance
the load on nodes. These techniques can be examined on different parameters such as
resource utilization, reliability of the system, system performance, related overhead
of the system, power saving feature, scalability and many more. This paper presents
the insight into the existing load balancing algorithms and their comparison on basis
of different parameters.

Keywords Load balancing · Cloud computing · Energy consumption · Green

computing

I. Naz (B) · S. Naaz · R. Biswas

Department of Computer Science and Engineering, Jamia Hamdard, New Delhi 110062, India
e-mail: inshanaz_sch@[Link]
S. Naaz
e-mail: snaaz@[Link]
R. Biswas
e-mail: rbiswas@[Link]

© Springer Nature Singapore Pte Ltd. 2020 881

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
882 I. Naz et al.

1 Introduction

Cloud computing as a service was first made known by McCarthy in 1961 and later it
was analysed by Licklider in 1963 [1]. Nowadays, cloud computing is the most pop-
ular term used in the field of computer science and IT. The frame ‘Cloud Computing’
refers to a kind of Internet-based technology which comprises of applications, stor-
age, on demand services, etc. These cloud computing services are available through
various organizations, like Google Cloud, Amazon Web Services and Microsoft
Azure, etc. and require a nominal amount of money and time for consumers to
access and process their data. There are three major components that describe the
cloud computing network and four cloud deployment models.
These components can be described as follows:

Components Models

IAAS (Infrastructure as a Public Cloud

service)
Private Cloud
PAAS (Platform as a service)
Hybrid Cloud
SAAS (Software as a service) Community Cloud

1.1 Software as a Service (SaaS)

This provides a complete or we can say a whole application as a service to the cloud
consumers in virtualized form typically on demand [2]. SaaS is very cost-effective
because of the use of the utility-based payment model as no infrastructure investment
is involved [3]. Common examples are Google apps like Google Mail and Salesforce,
spreadsheets, Zoho, etc.

1.2 Platform as a Service (PaaS)

For application development, PaaS platform provides a very convenient and suitable
environment. It provides the services to its consumers over the Internet on pay-per-
use basis. The user may be able to control the applications and configurations but does
not manage the infrastructure [4]. Examples are Microsoft Azure and [Link].
A Parametric Study of Load Balancing … 883

1.3 Infrastructure as a Service (IaaS)

It uses the virtualization technique. Iaas component provides the services like pro-
cessing, storage, data centre space, network equipments, etc. to its consumers on
demand [3]. In this environment, the consumer gets high degree of responsibility
and control. API is used for interaction. Amazon Web Services, GoGrid, 3 Tera are
some of the examples.

2 Cloud Deployment Models

2.1 Public Cloud

The public cloud user can easily access the systems and services that are provided by
the cloud service provider. In public cloud environment, there is a security risk since
applications from different users are mixed together on cloud servers but is cost-
effective [5]. Examples are Amazon Web Services (AWS), Microsoft and Goggle.

2.2 Private Cloud

A private cloud set-up is owned and managed by an organization singly. Resources are
managed by organization itself somewhat like Intranet [4]. This set-up is more secure
than public cloud set-up because the security is handled by the service providers
themselves. Eucalyptus system is one of the examples.

2.3 Hybrid Cloud

Hybrid cloud is the mixture of two or more clouds that remain unique entities, or in
other words, it can be stated as combination of private and public cloud [1]. Hybrid
cloud services are mostly used in healthcare and law.

3 Load Balancing

For the efficient use of resources and faster processing of tasks, we need to balance
the load and that is attained through load balancing [6]. Load balancing keeps a
check on the work/load on different nodes and makes sure that no particular node
or a server is heavily loaded thus preventing the breakdown of the system due to
884 I. Naz et al.

overloading. Domain Name Server (DNS) or a multilayer switch [7] is commonly

used for load balancing.

Key Features

1. The performance of the system is considerably improved.

2. To maintain the system stability.
3. To safeguard the machine consistency.
4. Deviations to the system are controlled.
5. Builds fault-tolerant systems by forming backups.

3.1 Load Balancing Algorithms

Load balancer allocates the tasks dynamically to servers/nodes to match up to

the increasing demand [8]. For smooth distribution of tasks/workload on different
servers, efficient load balancing algorithms and techniques are necessary to be imple-
mented. Depending on how the cloud is configured, load balancing can be dynamic
or static.

Static Approach
Non-pre-emptive technique. In the case of static load balancing, previous knowledge
of the system is required to balance the load and current state is not considered. It is
easily implemented and behaviour prediction is easy also [9].

Dynamic Approach
In case of dynamic approach, the work distribution takes place in runtime environ-
ment, and so the present status of the system is not needed to make any decision.
This approach is more efficient than static approach [10]. Usually, this algorithm is
composed of three strategies, namely the transfer strategy, the location strategy and
the information strategy [9].

4 Load Balancing in Cloud

Balancing the load in cloud environment is burning technologies in the current time.
As discussed earlier in this paper, load balancing improves the working performance
of the system by distributing the tasks among different processors in an efficient
manner so that the resources are utilized fully [11]. Due to the increasing number of
cloud users, balancing the load in cloud has become a critical issue. In the context of
cloud computing, the core intention of load balancing is to increase accessibility of
A Parametric Study of Load Balancing … 885

Fig. 1 Load balancing in cloud computing

resources and computing execution, present a backup plan, decrease response time,
scalability and reduce overall expenses (Fig. 1).
Some of the existing load balancing algorithms in cloud environment is discussed
below.

4.1 Ant Colony Optimization Technique

The ant colony optimization technique or simply ACO provides the finest resource
utilization; thus, the performance of the system is enhanced which in turn increases
the throughput of the system [12]. ACO is based on ant colony foraging behaviour,
the ants jointly hunt for food source then combatively make use of the offered food
sources to shift the food return to the shell. A pathway is made by these ants for
transferring from one node to another; using the following pathways, ants subse-
quently go back to the food. These techniques attain efficient resource consumption,
guarantee availability, a number of requests taken care by cloud are increased and
response time of requests decreased [13].
886 I. Naz et al.

4.2 Honeybee Foraging Behaviour

HFB is a technique of load balancing which is a direct implementation of natural

anomaly. It is a distributed load balancing technique [14]. A nature-inspired algo-
rithm is based on honeybee to gain private organization. It balances the load by
means of neighbouring server actions. Because of the increased infrastructure range,
the competence of the system is improved but throughput does not increase. This
approach is suitable in an environment where distinct sorts of service are needed.

4.3 HBB-LB

A load balancing technique which besides balancing the load also keeps the track
of preference of the jobs that were eliminated because of the overloaded VM’s [15].
The jobs that were eliminated are considered same as honey bees. The increase in
overall throughput reduces the response time of tasks. It is best suited for varied
cloud computing systems. Priority is the main key factor in this technique.

4.4 INS

In this approach, IP info and active stage index parameter are used to overcome the
network jamming [16]. In this algorithm, information replication and also repetition
are eliminated in cloud framework. The INS algorithm has following attributes to
compute the perfect fit: hash code of piece of info that is needed to formed, progress
state, victimized servers location and best net pace. This algorithm omits scanning
procedure of traditional backup thus decreasing the backup cost.

4.5 A2LB

The main contribution of this dynamic technique is proactive load calculations of

virtual machines. In this technique, whenever the load of the virtual machine arrives
at edge, load manager begins to search for the candidate vm’s using other DCs.
Anytime a virtual machine is burdened, the load is distributed in such a way that
the accessible resources get utilized proper manner such that no virtual machine is
overloaded and works properly [17].
A Parametric Study of Load Balancing … 887

4.6 Join-Idle-Queue

For the distribution of load in huge systems, JIQ approach is used. It can be dynam-
ically scaled since this algorithm is extendible in nature from both directions which
includes both varied processes and general arrivals. Job arrivals have no commu-
nication overheads in between the sender processes [18]. This approach has lower
response timings thus manages the load very efficiently.

4.7 PLBS

A centralized procedure-based strategy. The load disproportion on the nodes is

removed by boosting the response timings. Computational grid tactic is presented in
this approach so that the load unevenness is reduced and resource consumption is
increased. This approach works in this way, firstly it goes on hunting for best node for
computation purpose for those modules concerning the execution time and then the
next step is allotting the modules on specific node but keeping in view the capacity
index, and hence in this way, the finest job allocation is achieved [19].

4.8 Response Time-Based LB

Response time-based load balancing approach besides being dynamic in nature and
dropping the communication overhead also reduces the additional computation on
every server. This approach does not need to know the existing resource of the
system thus removing the requirement of unnecessary communication between the
load balancer and the virtual machine [20].

4.9 SBLB for Internet Distributed Services

For lessening the response time, this approach redirects the incoming requests to the
nearest server keeping in mind the server does not get congested. The time taken
by tasks to go from one node to another is quite less and also the time between
sending the requests and receiving their responses is smaller thus enhancing the
overall efficiency of the system [21]. It is a Unuique server-based load balancing
technique.
888 I. Naz et al.

Table 1 Summary
Approach Observation
Ant colony optimization Gives optimal resource utilization, resulting in
increase in throughput and performance
Honey bee foraging behaviour A distributed load balancing technique that achieves
load balancing using local server actions
HBB-LB Works well with heterogeneous cloud computing
systems. The increase in overall throughput reduces
the response time of tasks in this technique
INS Information replication and repetition is eliminated
A2LB Proactive load calculations of virtual machines
Join-idle-queue A distributed load balancing technique in large
systems. This technique efficiently reduces the load
without increasing the reaction time
SBLB for Internet distributed services A technique that redirects requests to nearest server
thus limiting service reply time
PLBS A centralized practice for load balancing multiple
tasks where each application is symbolized using
direct acyclic graph which has communication
requirements
Response time-based LB Dynamic and also cuts down the communication and
extra computation on all servers
Cloud server optimization A threshold-based contrast and balance approach

4.10 Cloud Server Optimization

This approach is basically a dynamic contrast and balance technique. It lays emphasis
on various host machines that need to be switched on so that the cost of cloud services
is reduced thus serving the purpose effective utilization of resources. A superior cloud
framework is used by this technique at host system level [22] (Tables 1 and 2).

5 Conclusion

After reviewing the existing load balancing techniques above, we came up with sev-
eral research concerns and breaches which can be treated for further contribution in
load balancing. Most of the techniques discussed above focuses on different param-
eters like scalability, throughput, overhead, etc. but very few techniques focused on
power saving and energy consumption issues. Thus, we need to present a load bal-
ancing approach that targets these two parameters and thus helps us to tackle high
energy consumption and carbon emission rates.
A Parametric Study of Load Balancing … 889

Table 2 Comparison table

Approach Performance Throughput Overhead Scalability Resource Fault
utilization tolerance
Ant colony Low Low ✓ × Low ×
optimization
Honey bee Low Low × × High ×
foraging
behaviour
HBB-LB Low High × × Low ×
INS High Low × × Low ✓
A2LB High Low × ✓ High ×
Join-idle-queue High Low ✓ × Low ×
SBLB for High Low × ✓ Low ×
Internet
distributed
services
PLBS High Low × ✓ High ×
Cloud server Low Low × × High ×
optimization

References

1. Alali FA, Yeh CL (2012) Cloud computing: overview and risk analysis. J Inf Syst 26(2):13–33
2. Harris T (2010) Cloud computing—an overview, Whitepaper, vol 462. Torry Harris Business
Solutions, pp 1–5
3. Shyam P, Dheeraj R, Jain P (2012) A survey paper on cloud computing. In: 2012 second
international conference on advanced computing and communication technologies. IEEE
4. Nazir Mohsin (2012) Cloud computing: overview and current research challenges. IOSR J
Comput Eng 8(1):14–22
5. Prasad MR, Naik RL, Bapuji V (2013) Cloud computing: research issues and implications.
Inter J Cloud Comput Serv Sci 2(2):134
6. Deepa T, Cheelu D (2017) A comparative study of static and dynamic load balancing algorithms
in cloud computing. In: 2017 international conference on energy, communication, data analytics
and soft computing (ICECDS). IEEE
7. Godha R, Prateek S (2014) Load balancing in a network. Inter J Sci Res Publ 4(10)
8. Katyal M, Mishra A (2014). A comparative study of load balancing algorithms in cloud
computing environment. arXiv preprint. arXiv:1403.6918
9. Hamadah S (2017) A survey: a comprehensive study of static, dynamic and hybrid load balanc-
ing algorithms. IRACST-Inter J Comput Sci Inform Technol Secur (IJCSITS) 7(2):2249–9555
10. Begum S, Prashanth CSR (2013) Investigational study of 7 effective schemes of load balancing
in cloud computing. Inter J Comput Sci Issues (IJCSI) 10(6):276
11. Kherani FF, Vania J (2014) Load balancing in cloud computing. Inter J Eng Dev Res 2(1)
12. Gupta E, Deshpande V (2014) A technique based on ant colony optimization for load balancing
in cloud data center. In: 2014 international conference on information technology. IEEE
13. Ahmad MO, Khan RZ (2018) Load balancing tools and techniques in cloud computing: a
systematic review. Advances in computer and computational sciences. Springer, Singapore, pp
181–195
14. Randles M, Lamb D, Taleb-Bendiab A (2010) A comparative study into distributed load balanc-
ing algorithms for cloud computing. In: 2010 IEEE 24th international conference on advanced
information networking and applications workshops. IEEE
890 I. Naz et al.

15. Krishna PV (2013) Honey bee behavior inspired load balancing of tasks in cloud computing
environments. Appl Soft Comput 13(5):2292–2303
16. Wu TY et al (2012) Dynamic load balancing mechanism based on cloud storage. In: 2012
computing, communications and applications conference. IEEE
17. Singh A, Juneja D, Malhotra M (2015) Autonomous agent based load balancing algorithm in
cloud computing. Procedia Comput Sci 45:832–841
18. Lu Y et al (2011) Join-Idle-Queue: a novel load balancing algorithm for dynamically scalable
web services. Perform Eval 68(11):1056–1071
19. Shahid M, Raza Z (2014) A precedence based load balancing strategy for batch of DAGs
for computational grid. In: 2014 international conference on contemporary computing and
informatics (IC3I). IEEE
20. Sharma A, Peddoju SK (2014) Response time based load balancing in cloud computing. In:
2014 international conference on control, instrumentation, communication and computational
technologies (ICCICCT). IEEE
21. Nakai AM, Madeira E, Buzato LE (2011) Load balancing for internet distributed services using
limited redirection rates. In: 2011 5th Latin-American symposium on dependable computing.
IEEE
22. Sahu Y, Pateriya RK, Gupta RK (2013) Cloud server optimization with load balancing and
green computing techniques using dynamic compare and balance algorithm. In: 2013 5th
international conference and computational intelligence and communication networks. IEEE
Logical Clustering of Similar Vertices
in Complex Real-World Networks

Md A. Rahman and Natarajan Meghanathan

Abstract We show that vertices part of a physical cluster (determined per the edges
that connect the vertices) in a complex real-world network need not be similar on the
basis of the values incurred for node-level metrics (say, centrality metrics). We adapt a
recently proposed approach (based on unit-disk graphs) to determine logical clusters
comprising of vertices of similar values for node-level metrics, but need not be
physically connected to each other. We use the Louvain algorithm to determine both
the physical and logical clusters on the respective graphs. We employ the Silhouette
Index measure to evaluate the similarity of the vertices in the physical and logical
clusters. When tested on a suite of 50 social and biological network graphs on the
basis of neighborhood and/or shortest path-driven centrality metrics, we observe the
Silhouette Index of the logical clusters to be significantly larger than that of the
physical clusters.

Keywords Logical clusters · Physical clusters · Centrality metrics · Silhouette

index · Complex network analysis

1 Introduction

Clustering (also referred to as community detection) is a critical component of com-

plex network analysis. In the traditional sense, a cluster (community) in a network
comprises a group of vertices that are more connected to each other than to ver-
tices outside the cluster [1]. We refer to such clusters as physical clusters. Several
clustering algorithms (like Girvan–Newman algorithm [2], Louvain algorithm [3],
and neighborhood-overlap-based greedy algorithm [4]) are available in the literature
to determine physical clusters of vertices of larger modularity. A physical cluster

M. A. Rahman · N. Meghanathan (B)

Computer Science, Jackson State University, Jackson, MS 39217, USA
e-mail: [Link]@[Link]; nmeghanathan@[Link]
M. A. Rahman
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 891

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
892 M. A. Rahman and N. Meghanathan

is considered to be more modular [1] if it has high intra-cluster density and low
inter-cluster density.
With the objective of maximizing edge-based intra-cluster density, it is difficult
to expect a clustering algorithm to group vertices that are similar to each other.
Similarity of vertices is typically assessed on the basis of node-level metrics that
could be topology or domain-driven. Centrality metrics are classical examples for
topology-based metrics that quantify the extent to which a node is important on
the basis of its location in the network [1]. The centrality metrics are used as the
node-level metrics for our analysis in this paper. We consider the following centrality
metrics: neighborhood-driven degree (DEG) [1] and eigenvector (EVC) [5] centrality
metrics and the shortest path-driven betweenness (BWC) [6, 7] and closeness (CLC)
centrality metrics [8, 9]. For more details on the centrality metrics, the interested
reader is referred to [10]. The analysis presented in this paper could be seamlessly
extended to any combination of domain-driven or topology-driven metrics as well.
In Fig. 1, we show a motivating example graph wherein the tuple next to a vertex
represents the degree (DEG) and eigenvector centralities (EVC) of the vertex. Any
well-known clustering algorithm in the literature would determine the two physical
clusters in sub Fig. 1a that have high modularity (larger intra-cluster density and
lower inter-cluster density). However, a closer look at the tuples for the vertices
within a physical cluster would indicate less similarity among the vertices on the
basis of their DEG and EVC values. On the other hand, in sub Fig. 1b, we show
two clusters each of which comprises vertices that are exactly similar on the basis of
their DEG and EVC values. The two clusters in sub Fig. 1b are not very modular (the
inter-cluster density is even larger than the intra-cluster density), but each of these
clusters would be more cohesive (i.e., comprised of vertices that are similar) on the
basis of their DEG and EVC values.
We propose the following approach (more details are in Sect. 2) to determine such
cohesive clusters of similar vertices in real-world network graphs. We distribute
the vertices in a coordinate system whose coordinates are the normalized values
of the node-level (centrality) metrics of the vertices. For such a logical topology,

(a) (b)

Fig. 1 Example graph: physical modular clusters versus logical clusters of similar vertices. a Phys-
ical clusters (larger intra-cluster density, but lower vertex similarity). b Logical clusters (lower
intra-cluster density, but larger vertex similarity)
Logical Clustering of Similar … 893

we iteratively attempt to connect the vertices together (an edge exists between two
vertices if the Euclidean distance between their normalized coordinate values is
within a threshold) in the form of a unit-disk graph. We use a binary search approach
to identify the minimum value for the threshold distance that would connect together
the vertices in the unit-disk graph. We run the Louvain clustering algorithm [3] on the
connected unit-disk graph to determine one or more (logical) clusters whose member
vertices are more similar compared to the vertices in the physical clusters obtained
by running the Louvain algorithm on the corresponding real-world network graph.
We use the Silhouette Index [11] measure to assess the similarity of the vertices in the
logical clusters vis-a-vis the physical clusters with respect to the centrality metrics
considered. In a recent work [12], we have successfully used the unit-disk graph and
binary search-based approach to quantify the similarity between any two vertices in
a network (in the form of a metric called the node similarity index). Our hypothesis in
this research is that for a set of centrality metrics, the Silhouette Index of the logical
clusters would be larger than the Silhouette Index of the physical clusters.
The rest of the paper is organized as follows: In Sect. 2, we explain the approach
to determine logical clusters of similar vertices on the basis of node-level (centrality)
metrics. In Sect. 3, we explain the formulation for the Silhouette Index measure. In
Sect. 4, we test our hypothesis on a suite of 25 biological networks and 25 social
networks on the basis of three sets of centrality metrics: (DEG, EVC); (BWC, CLC),
and (DEG, EVC, BWC, CLC). We present and analyze the Silhouette Index results
for the physical clusters and logical clusters obtained for the different combinations
of centrality metrics. Section 5 presents our conclusions and outline plans for future
work.

2 Logical Clusters of Similar Vertices

We describe the sequence of steps (see Fig. 2) involved in determining logical clusters
of similar vertices (for k centrality metrics) in a real-world network graph.
Step (i): Construction of a Logical Topology: Let the number of centrality metrics
considered for logical clustering of a real-world network graph be k. To get started,
for each of the k metrics, we determine their raw centrality values (see Fig. 2a) and
then independently normalize them (using the square root of the sum of the squares
of the raw values). Following this, we build a logical topology (k-dimensional coor-
dinate system) of the vertices wherein the coordinates of a vertex are its normalized
centrality values (ranging from 0 to 1). See Fig. 2b.
Step (ii): Binary Search Algorithm to Deduce a Connected Unit-Disk Graph:
We attempt to deduce (through a sequence of iterations) a unit-disk graph that would
connect all the vertices in the logical topology at a minimum threshold distance for
an edge√ to exist. For a set of k centrality
√ metrics, the threshold could range from
(0,…, k); if the threshold distance is k, we will have a unit-disk graph that is sure
to be connected (completely connected indeed!), but not connected at a threshold
894 M. A. Rahman and N. Meghanathan

Fig. 2 Example to illustrate the computation of the logical clusters and their evaluation
Logical Clustering of Similar … 895

distance of 0 (unless all the vertices are co-located). We use this observation as the
basis to run a binary search algorithm to determine the minimum possible value for
the threshold distance to obtain a connected unit-disk graph [12]. We start√the binary
search algorithm with the left index set to 0 and the right index set to k and go
through a sequence of iterations (see Fig. 2c).
Across all the iterations, the following invariant is maintained: If the threshold dis-
tance value corresponds to the left index, the unit-disk graph is not connected; if the
threshold distance value corresponds to the right index, the unit-disk graph is con-
nected. In each iteration, we first determine the middle index as the average of the
left index and right index, and seek to construct a unit-disk graph with the threshold
distance value corresponding to the middle index. If such a unit-disk graph is con-
nected, we set the right index to the value of the middle index; otherwise, we set the
left index to the value of the middle index. We continue the iterations until the right
index and left index differ not more than a cutoff parameter (∈). We use ∈ = 0.001
for all the analysis conducted in this paper. We set the minimum threshold distance
to correspond to the value of the right index in the last iteration of the algorithm.
Step (iii): Logical Clustering of the Connected Unit-Disk Graph: On the con-
nected unit-disk graph obtained for a minimum threshold distance, we run the Lou-
vain community detection algorithm [3] to determine (logical) clusters of vertices
that have a larger intra-cluster density (and a lower inter-cluster density) in the unit-
disk graph. The vertices within a logical cluster are expected to be more similar to
each other. The Louvain algorithm (a hierarchical community detection algorithm) is
designed to identify highly modular communities. To determine the logical clusters
using the Louvain algorithm, the weight of an edge in the connected unit-disk graph
is the Euclidean distance between their corresponding normalized coordinate values.
Note that the edge weights for the real-world network graphs are “1” when we run
the Louvain algorithm to determine the physical clusters.

3 Silhouette Index

We use the Silhouette Index [11] measure (ranges from −1 to 1) to evaluate the
extent of similarity among the vertices of the physical clusters and logical clusters
with respect to the normalized centrality values of the vertices. The larger the values
for the Silhouette Index for a cluster, the more similar are the vertices within the
cluster with respect to the centrality metrics in consideration. The Silhouette Index
for a cluster is the average of the Silhouette Index values of its member vertices.
The Silhouette Index for a network is the weighted average of the Silhouette Index
values of its clusters. The Silhouette Index for a vertex i in a cluster C k is calculated
as per formulation (1). Here, di,min is the minimum of the average of the Euclidean
distances for vertex i to vertices in the other clusters; di,Ck represents the average of
the Euclidean distances for vertex i to vertices in its own cluster. A negative Silhouette
Index for a vertex is an indication that the vertex is not in the appropriate cluster.
896 M. A. Rahman and N. Meghanathan

A negative Silhouette Index for a cluster is an indication that its member vertices
should have been in other cluster(s) for better cohesiveness.

di,min − di,Ck
Silhouette Index(i) =   (1)
max di,min , di,Ck

4 Evaluation on Real-World Networks

In this section, we test our hypothesis on a suite of 25 biological network graphs and
25 social network graphs of diverse degree distributions and present the results of
the Silhouette Index measure evaluated for the physical clusters and logical clusters
obtained by running the Louvain algorithm (per the approaches described in the ear-
lier sections). The biological networks analyzed are either gene–gene interaction net-
works, protein–protein interaction networks, interactions between different animal
species in a particular area, etc. The social networks analyzed comprise acquaintance
networks that capture the association between two users in a social network over a
certain time period and friendship networks for which no such time period is used
to capture association between two users.
In Fig. 3, we visually present some of the fundamental statistical information for
the biological networks and social networks (for more details, see [12]). The number
of nodes in the biological networks ranges from 62 to 2,640 with a median of 813.
The number of nodes in the social networks ranges from 22 to 1,882 with a median
of 75. The spectral radius ratio for node degree (λsp ≥ 1) [13] for a network graph
is the ratio of the principal eigenvalue [5] of the adjacency matrix for the graph and
the average node degree; the larger the λsp value, the larger the variation in node
degree. The edge density (0 ≤ ρ edge ≤ 1) is calculated as the ratio of the actual
number of edges in the network and the maximum possible number of edges in the
network (which is N(N − 1)/2 for a network of N nodes). The biological networks
are characteristic of having larger λsp values and lower ρ edge values; on the other
hand, social networks are characteristic of having smaller λsp values and larger ρ edge
values. For more details on the individual real-world networks analyzed in each of
these domains, the interested reader is referred to [12].

Fig. 3 Statistics for the biological and social networks

Logical Clustering of Similar … 897

In Figs. 4 and 5, we present and visually compare the Silhouette Index values for
the physical vs. logical clusters on the basis of the neighborhood-based (DEG, EVC);
the shortest path-based (BWC, CLC), and all the four centrality metrics together
(DEG, EVC, BWC, CLC). We observe all the data points to be above the dotted
diagonal line, indicating that the Silhouette Index values for the logical clusters for
all the real-world networks are appreciably larger than the Silhouette Index values
for the physical clusters. For each coordinate system and for each network category,
we measure (see Fig. 6a for a comparative bar chart) the average of the difference
in the Silhouette Index values for the logical clusters versus physical clusters. We
observe the biological networks to incur larger average difference in the Silhouette
Index values for all the three coordinate systems. The (DEG, EVC) coordinate sys-
tem incurs, respectively, the lowest (for social networks) and largest (for biological
networks) values for the average difference in the Silhouette Indexes for the logical
versus physical clusters. We also measure the median (see Fig. 6b for a compara-
tive bar chart) of the Silhouette Index values for the physical clusters versus logical
clusters. Through Figs. 4,5, and 6, we observe the Silhouette Index values for the
physical (logical) clusters in the biological networks to be relatively lower (larger)
than those in the social networks.

(DEG, EVC) (BWC, CLC) (DEG, EVC, BWC, CLC)

Fig. 4 Biological networks: Silhouette Index values for the physical versus logical clusters

(DEG, EVC) (BWC, CLC) (DEG, EVC, BWC, CLC)

Fig. 5 Social networks: Silhouette Index values for the physical versus logical clusters
898 M. A. Rahman and N. Meghanathan

(a) (b)

Fig. 6 Statistical comparison of the Silhouette Index values. a Avg. difference in the social networks
biological networks Silhouette Index values. b Median of the Silhouette Index values

5 Conclusions

We show that logical clusters of vertices could be more cohesive with respect to node-
level centrality metrics compared to physical clusters of vertices in complex real-
world network graphs. In this pursuit, we adapt a recently proposed unit-disk graph
approach (for node similarity assessment) [12] to determine such logical clusters
of similar vertices. We applied the approach on a suite of 25 biological networks
and 25 social networks and evaluated the extent of similarity of the vertices in the
logical clusters versus physical clusters using the Silhouette Index measure with
respect to three combinations of centrality metrics (DEG, EVC), (BWC, CLC), and
(DEG, EVC, BWC, CLC). For each combination, we observe all the 50 real-world
network graphs to incur significantly larger Silhouette Index values for the logical
clusters compared to the physical clusters. We observe the biological networks to
show a relatively larger difference in the Silhouette Index values between the logical
clusters and physical clusters. Likewise, the (BWC, CLC) coordinate system has
been observed to incur relatively larger Silhouette Index values for several real-
world networks. In future, we plan to extend the unit-disk graph approach for outlier
detection in complex networks and large datasets.

References

1. Newman MEJ (2010) Networks: an introduction, 1st edn. Oxford University Press, Oxford,
UK
2. Girvan M, Newman MEJ (2002) Community structure in social and biological networks. Proc
Natl Acad Sci USA 99(12):7821–7826
3. Blondel VD, Guillaume JL, Lambiotte R, Lefebvre E (2008) Fast unfolding of communities
in large networks. J Stat Mech Theor Exp P10008:1–11
4. Meghanathan N (2016) A greedy algorithm for neighborhood overlap-based community
detection. Algorithms 9(1, 8):1–26
5. Bonacich P (1987) Power and centrality: a Family of measures. Am J Sociol 92(5):1170–1182
6. Freeman L (1977) A set of measures of centrality based on betweenness. Sociometry 40(1):35–
41
7. Brandes U (2001) A faster algorithm for betweenness centrality. J Math Sociol 25(2):163–177
Logical Clustering of Similar … 899

8. Freeman L (1979) Centrality in social networks: conceptual clarification. Soc Netw 1(3):215–
239
9. Cormen TH, Leiserson CE, Rivest RL, Stein C (2009) Introduction to algorithms. MIT Press,
Cambridge
10. Meghanathan N (2016) Assortativity analysis of real-world network graphs based on centrality
metrics. Comput Inform Sci 9(3):7–25
11. Rousseeuw PJ (1987) Silhouettes: a graphical aid to the interpretation and validation of cluster
analysis. Comput Appl Math 20:53–65
12. Meghanathan N (2019) Unit disk graph-based node similarity index for complex network
analysis. Complexity. Article ID 6871874, p 22
13. Meghanathan N (2014) Spectral radius as a measure of variation in node degree for complex
network graphs. In: The 3rd international conference on digital contents and applications,
Hainan, pp 30–33
Capacity Enhancement Using
Delay-Sensitive Protocol in MANETs

V. Sivakumar, J. Kanimozhi, B. Keerthana and R. Muthulakshmi

Abstract In wireless modulation technologies, it has advanced frequency modula-

tion. It is capable of performing in Quality of Service (Qos) for utilizing the scope of
mobile ad hoc. It is measured only for calculating the ratio between two nodes. It will
form an implementation of single-hop delay. After that, to construct the multicast
tree for real-time in delay-sensitive tree, proposed model has been established in
MANET. While increasing a network in capacity, this proposed multicast protocol
can be minimized in the host and in a particular time the transmission node can block
the entire neighboring node so that there is no interrupt and wastage of data that can
be occurred in the data for properly adjusting data rates. Simulating results provides
lot of accuracy. It induces the expected result at the end of this paper.

Keywords Multicast protocol · Delay-sensitive · Qos · Multi-rate · Capacity

enhancement · One-hop delay

1 Introduction

In wireless network, the node moves randomly, and it can be able to communicate
with each other by multi-hop transmission. These transmissions would not have a
fixed infrastructure. And they do not have utilized centralized administration. Direct
transmission of packet is occurred in the initial node to its destination [1]. Multicast

V. Sivakumar · J. Kanimozhi (B) · B. Keerthana · R. Muthulakshmi

Department of IT, Adhiyamaan College of Engineering, Hosur, Tamilnadu, India
e-mail: kanijoh25@[Link]
V. Sivakumar
e-mail: sivcse45@[Link]
B. Keerthana
e-mail: bkeerthan1998@[Link]
R. Muthulakshmi
e-mail: muthukrishna9816@[Link]

© Springer Nature Singapore Pte Ltd. 2020 901

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
902 V. Sivakumar et al.

forms the cluster in communication; it transmits data packet in sharing the informa-
tion from source to destination. Multicast tree is efficient for sending same packets.
For a specified condition, multicast tree should request for delay requirements, where
multicast protocol is in delay-sensitive state. When compared to the predefine value,
these end-to-end delays are smaller [2]. For building a tree to find out the shortest
path in the given range, they form a large amount of transmission from one part to
another part, and delay is nothing but it transmits node from source to destination, it
means to its neighboring node.
A maximum cost can be transmitted by transmitter to send successful packets, and
it is dependent on the SINR and it is received. It is always directly proportional to the
transmission range. If a transmitter transmits, then the receiver has to perceive high in
SINR [3]. Most probably, the delay can be found out by the process of hopping data
from one node to another node. In order to have a complete transmission of packet, it
also requires short range in transmitter. Often in multi-rate MANETs, host may vary
in the neighborhood; later it is more difficult for the estimate of delay in MANET.
If a transmission occurs, then the neighborhood is blocked in MANETs. They share
channels’ ratio with their neighbors. If we design Quality of Service (QoS) for a
MANET, its neighboring data will be maintained for the estimation of resource,
and it is consumed by constructed QoS route to avoid violation in resource from
other neighbors. This Quality of Service is able to utilize only the limited wireless
resources efficiently.
Previously, multicasting protocols have been designed only for single-rate
MANETs or multi-rate. CEDAR and some other algorithm are neither guaranteed
by bandwidth, and if it begins to send its packet to nearest node, then the carrier
senses to block the neighbor [4]. HRP or HMRP is pointed, for the failing transmit-
ters while they admit the flow. It indicates that it is saturated in network traffic. They
can be avoided by two routing algorithms, namely hidden route problem (HRP) or
from the hidden multicast route problem (HMRP); it fails to collect information from
the neighboring resource [5]. L. Chen and W. Heinzelman have founded how to be
provided.
Bandwidth guarantee in QoS. C. C. Hu has been proposed about the delay-
sensitive multi-rate protocol. From IEEE system journals, we have read about the
single or multi-rate MANETs, when a host starts its transmission from its neighbor
to another if the former is within the transmission range of latter.
If the traffic occurs in network, then the bandwidth or requirements in delay
performance cannot be satisfied. Unlike other specifics, we use the neighboring
transmission to denote the neighbor [6]. Therefore, on-demand routing is resulted in
delay estimation method.
To calculate the shortest path and find out the implementation of multicast tree, we
have already studied about the existing system how to calculate the delay-sensitive
routing protocol in C. C. Hu. This proposed project is used to establish a lot of
hopping nodes in the Quality of Service. They would eliminate the hidden problem
and similarly they will eliminate the multicast neighbors [7]. There is no reply in
late simulation of packet. In existing system, it can be used to exit the unwanted data
and calculate multiple packets. It is greater and has large amount of valued data for
Capacity Enhancement Using Delay-Sensitive Protocol in MANETs 903

transmitting the packets. In packet forwarding, it senses and forwards more packets,
when it forwards neighbors should be blocked. If the forwarders increases, network
performance will go worse.
These systems will tend to minimizing the total transaction of information from
the particular node to its destinated node [8]. When choosing the neighboring data
packets in multicasting route, it uses single hopping system. Due to this process, the
capacity has been increased.

2 Related Work

The Quality of service (QoS) is mainly used in multicasting protocol in single-rate

MANETs. The transmission delay can be trying to reduce the protocol. In delay-
sensitive multicast tree, they are not delay sensitive, for failing the estimate from
the final late coming nodes [9]. If they do not eliminate the hidden multicast, it will
suffer from some missing protocol from HMRP. It cannot construct the simultaneous
routes.
The range of node can be varied due to its shortest path in algorithm [10]. From
this protocol, from given source which has been detected within a range, the capacity
in wireless network has been increased. According to delay in neighboring node, a
distributed routing protocol can be avoided in hidden problems, and it accepts the
requests from neighbor to process the delay control [11].
The simulation of packet can flow the request from every bandwidth in the liter-
ature. Markov-modulated Poisson process will be used in the estimation of average
delays. While analyzing the channel, most of the conditions were not assigned as
parameters in the traffic flows [11]. Estimates of one hop delay can be calculated by
using approximate model.
In one ratio analysis, the delay can be used to find out the shortest path within a
given range, and they can be able to modify the statement of data with or without
the permission of the author. But the irrelevant and incomplete data can be strictly
eliminated to make the transaction faster and easier for sender to receiver process
[12].
Most probably, all the nodes are sensitive toward the host, and there would not be
several attempts or increase in channel that will happen. Due to this existing system,
our proposed system gives lot of changes with good accuracy report which has been
shown in X-graph.

3 Protocol

The protocol plays a major role in this multi-rate MANET. This system is distributed
in the origin data. DSM has multiple attempts that can be used as algorithm to find
out the hidden problem with the definitions in MAC. For packet transmission, single
physical channel should be available, and every host perceived can be able to monitor
904 V. Sivakumar et al.

Fig. 1 Architecture diagram

the channel whether it is idle or busy. To sense the message from host with the base
rate, here we use MANET. For example, 1 Mb/s is the base rate. In the estimation of
delay, it senses the algorithm which uses default method in multicasting the data for
deriving all the busy ratio in threshold.
Figure 1 explains about the flow of packet and transmission of data from the
source to destination. When they started to share the data, host gets blocked by the
route.

3.1 Neighbor Host

In multi-rate MANETs has to identify in order to its r p —neighbors and from multi-
rate MANETs, every neighbors require to have a construction of two tables, named
with one-hop neighbors and the two-hop neighboring. pi,j gets transmitted from
source host of H j to H i . It allows several request flow and eliminates using HMRP
[13]. Later, it gets changed by the wireless network of signal has been indicated.
An auto-rate algorithm is used in receiver side, and rollback algorithm is used in
transmitter side. In Fig. 2, this proposed system explains about the generation of
node with the transmission, and HMRP and HRP have been blocked.
Capacity Enhancement Using Delay-Sensitive Protocol in MANETs 905

Fig. 2 Creation of nodes

3.2 Construction of Multicast Tree

In this existing system, delay can be reduced for low request when implementing
the single node. In order to take the proposed system, weight can be assigned with
its transmission of packets and its sum when blocking its overall transmission that
can be hosted. If the resources can be hosted in consumed multicast, they have more
flows in requesting the capacity of network enhancement.

3.3 Methodology of Implementing Trees

To construct trees, we use heuristic algorithm from proposed multi-rate MANETs.

Trade-off range has been used in data rate transmission range. To participate in
forwarding trees, higher data rates have been selected in packets. When carrier starts
to transmit the packet, the entire neighbor has been blocked. Due to this process, the
network degrades with its performance if it has large forwarders in number. In this
algorithm, their data rate can be changed.
In delay-sensitive multicast tree, the neighbors are used to sense the forwarders
into account. This proposed algorithm has been explained in the above section. A
multicast tree is used to invoke iteratively. To construct multicast tree, the basic
procedure is used. For specifying the destination of delay-sensitive to construction.
In order to consume the resource in reduced construction tree, the main feature is to
906 V. Sivakumar et al.

reduce the total transmission time and minimize the forwarders, and then it blocks
the host using blocking time.
In the above section, we explained in-depth and its illustrative example as provided
below. Implementation issues can be addressed. Implementation of tree construction
in variables can be used. In tree multicast current collect to the current multicast
tree, the destination was not connected in tree of multicast [14]. Here, F is used to
collect the forwarders. F is represented in forwarders, and R is used in F. D is set as
transmission.
There are four topologies, namely
1. Known Topology (singlecast)
2. Unknown Topology (singlecast)
3. Known Topology (multicast)
4. Unknown Topology (multicast)
It minimizes the weight route, in Dijkstras algorithm to find out the nearest path
(shortest). It is graphed from the weight graph. Dijkstras algorithm is used for con-
structing the shortest path which is equal to its weight from source vertex to its other
vertices. Multi-rate representation of directed graph in MANET is conveniently used
in vertex graph.
Is used as vertex and arch is used as unique element.
Delay-sensitive multicast is temporarily used in maintaining the multicast tree,
point to point is used from source to destination which transmits packets from point
to point without loss in data. It does not request the flow and delay have been violated.
Multicast tree is finally used to avoid the HMRP and HRP.
Especially, Delay_Sensitive_Route is being iteratively hosted in appending delay-
sensitive routes. If a host is violated in appending route, then it can be temporarily
violated delay in multicast for all the ongoing flows.
It may exceed the route delay requirement from one end to another end. In
following description, the above mentioned delay requirement can be hosted.
Implementation Issues: When recall the neighbors and carrier sense, it is men-
tioned with some denotations like r p —neighbors and r p —carrier sense neighbors. hi
can be estimated. When r p is used for the nd, later it is stored in neighbors in h data
rate. To identify the channel of busy/idle ratio, it is necessary to implement the hop
delay of one to one node. First three nodes are stored in hop delays, and the residual
delay happens due to the ongoing flow passes.

4 Performance Evaluation

Implementation can be done or perfomed by the ns2 tool. Every evaluation has
been required to transmit the channel in performance. The two-ray ground model
is adopted for predicting the signaling receiving power from the ground has been
considered as a reflection. Hence, in receiving the signal power and transmitting,
it in a distance of attenuated as 1/d 2 , where d is represented as the gap in center
Capacity Enhancement Using Delay-Sensitive Protocol in MANETs 907

transmitter and receiver. From CBR traffic, its flows can be injected from the sources
to its destination in the network, and every host is provided for a MAC FIFO queue
which is arranged in the 50th node [15].
In this existing system, we derived in delay sensitive. Choose delay in particular
node to simulation which can be conducted for preference operation using multiple
rates. Several operations cannot able to perform at a single delay-sensitive data rate:
In the existing system, we use the estimation that may vary in all trees. In evaluation
of packet, the data can be explained as follows.

Avoidance of HMRP
When avoiding this, point-to point information will be received in delay manner so
it delay and success ratio has been achieved by this two indicators can provide for
the effectiveness that can avoid HMRP. The success ratio may vary the flow of an
uncast flow, and it has been maintained the information with its successful.

Admission Ratio
To find out the admission ratio in this paper, we are going to implement the complete
graph. Therefore, all the node gets request message about the following information
which is going to transmit the data. This paper construction is based on few topologies
to different methods. These methods contain single and multicast format.
Maximum node can be avoided to get accurate result in the X-graph. At the initial
phase, these nodes can move from the start point to end point. Later, the flow would
not be affected (Fig. 3).
In Fig. 4, this paper demonstrates the transceiver nodes from source to destination
in the [Link] folder. Finally, the simulation is used to show the shortest path and
throughput for the following graph which can be represented in the form of graph.

Fig. 3 Simulation of packet

908 V. Sivakumar et al.

Fig. 4 Delay-sensitive protocol

In Fig. 3 simulation of packet, all the data gets broadcasted with its neighboring
node, and they can transmit packet without requesting the host in route. The proposed
multicast protocol is constructed by the trees which are shown in result of simulation.
If it was avoided HRP/HMRP, there are no interrupts. If it provided are multiple
data rates, same amount of data rates. So, the wireless resources in delay-sensitive
protocol can be utilized by its efficiently in crucial.
Figure 5 uses the representation for point to point delay in both existing system. We
have developed the proposed system and its rate in MANET for wireless resources in
this paper. For shared channel in ratio, our MANETs are an ongoing flow [1]. Mainly,
it is used for advanced technologies, and they may tend to satisfaction. The purpose
is used for requesting the flows. HRP and HMRP are avoided due to its failure in
estimation.

5 Discussion and Conclusion

By constructing and implementing the graph to find out the shortest path in multi-
rate. They can be minimized with the total sum of transmission time, and it can be
blocked by hosts, since the transmitting ranges are in host.
Capacity Enhancement Using Delay-Sensitive Protocol in MANETs 909

Fig. 5 X-graph

References

1. Sivakumar R, Sinha P, Bharghavan V (1999) CEDAR: a core-extraction distributed ad hoc

routing algorithm. IEEE J Sel Areas Commun 17(8):1454–1465
2. Xue Q, Ganz A (2003) Ad hoc quos on-demand routing (AQOR) in mobile ad hoc networks.
J Parallel Distrib Comput 63:154–165
3. Chen L, Heinzelman W (2005) Qos-aware routing based on bandwidth estimation for mobile
ad hoc networks. IEEE J Sel Areas Commun 23(3):561–572
4. Sinha P, Sivakumar R, Bhanghavan V (1999) MCEDAR: multicast core-extraction dis-
tributed ad hoc routing. In: Proceedings of the IEEE wireless communications and networking
conference, pp 1313–1317
5. Pagani E, Rossi G (2001) A framework for the admission control of Quos multicast traffic in
mobile ad hoc networks. In: Proceedings of the 4th ACM international workshop on wireless
mobile multimedia, pp 61–68
6. Darehshoorzadeh A, Dehghan M, Motlagh G (2007) Quality of service support for ODMRP
multicast routing in ad hoc networks. Lecture notes in computer science, vol 4686. Springer,
New York, pp 237–247
7. Hu CC, Wu EK, Chen GH (2008) Bandwidth-satisfied multicast trees in MANETs. IEEE Trans
Mobile Comput 7(6):712–723
8. Zhang B, Mouftah H (2005) Quos routing for wireless ad hoc networks: problems, algorithms,
and protocols. IEEE Commun Mag 43(10):110–117
9. Zhang B, Mouftah H (2004) Qos routing through alternate paths in wireless ad hoc networks.
Int J Commun Syst 17:233–252
10. Hsu JL, Rubin I (2007) Cross-layer multi-rate routing strategies in wireless multi-hop random
access networks. In: Proceedings of the IEEE global telecommunications conference, pp 609–
613
11. Venu S, Rahman AMJMZ (2018) FAODV, DSR, DSDV performance analysis for broadcasting
in MANET. TAGA J Graph Technol 14:1179–1187
910 V. Sivakumar et al.

12. Venu S, Rahman MdZ, Rahman AMJ (2014) Efficient routing for broadcasting in mobile ad
hoc networks. Inter J Appl Eng Res 9(27):9546–9552
13. Nafaa A, Ksentini A (2008) On sustained Qos guarantees in operated IEEE 802.11 wireless
LANs. IEEE Trans Parallel Distrib Syst 19(8):1020–1033
14. Zhai H, Fang Y (2006) Physical carrier sensing and spatial reuse in multi-rate and multihop
wireless ad hoc networks. In: Proceedings of the IEEE international conference on computer
communications, pp 1–12
15. Ma H, Vijayakumar R, Roy S, Zhu J (2009) Optimizing 802.11 wireless mesh networks based
on physical carrier sensing. IEEE/ACM Trans Netw 17(5):1550–1563
SCO-RNN: A Behavioral-Based
Intrusion Detection Approach for Cyber
Physical Attacks in SCADA Systems

N. Neha, S. Priyanga, Suresh Seshan, R. Senthilnathan

and V. S. Shankar Sriram

Abstract Supervisory control and data acquisition (SCADA) systems monitor and
control the critical infrastructures (CI) such as power generation, smart grids, oil–
gas pipelines, wastewater management, and nuclear power plant. Due to the drastic
increase in cyber attacks, maintaining SCADA systems has become a complex task.
Difficulty in securing the SCADA has gained the attention of researchers in design-
ing a robust intrusion detection system (IDS). However, existing machine-learning
and statistical approaches fail to detect the cyber physical attacks with high detection
rate. This paper presents a sine-cosine optimization based recurrent neural network
(SCO-RNN) to detect the cyber physical attacks against SCADA systems and the per-
formance of the proposed SCO-RNN was validated using the Secure Water Treatment
(SWaT) dataset in terms of accuracy and detection rate.

Keywords Intrusion detection system · Cyber physical system · Supervisory

control and data acquisition system (SCADA) · Recurrent neural network · Sine
cosine optimization

1 Introduction

Cyber-physical systems (CPS) are embedded systems which incorporate the network-
ing components, physical components, and computational algorithms that play a key
role in maintaining the CI [1, 2]. This paper focuses on one such CPS, SCADA.
SCADA systems monitor and control a wide range of Industrial Control Systems

N. Neha · S. Priyanga · S. Seshan · R. Senthilnathan · V. S. Shankar Sriram (B)

School of Computing, Centre for Information Super Highway (CISH), SASTRA Deemed to be
University, Thanjavur, Tamil Nadu 613401, India
e-mail: sriram@[Link]
N. Neha
e-mail: nehanageswaran@[Link]

© Springer Nature Singapore Pte Ltd. 2020 911

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
912 N. Neha et al.

(ICS) linked to networks, causing downtime of the system which in turn can cause
disastrous effects for national security [3]. Traditional SCADA systems were iso-
lated that they were not connected with the Internet thus shielded from all the remote
attacks [4]. Over the past decades, the SCADA systems were more susceptible to
cyber attacks which happened not only on the physical infrastructures but also on
the communication network.
There has been a significant increase in the number of attacks which have caused
catastrophic effects to the nation. According to the IBM X-Force Report, recent
security incidents like Qak-Bot (2017), Botnet-based CMDi LFI attacks (2017/18)
in which the attacker attempted to upload malicious files to servers and also attempted
to mine CryptoNote-based currencies such as Monero (XMR), IcedID emerges in
the USA and UK (2017), etc, accentuate the need to protect the digital ecosystem
from the security threats and intrusions [5].
The traditional security measures like firewall, authentication, etc., proved to be
less secure and made us understand the necessity for developing an efficient intru-
sion detection system (IDS) [6]. The main aim of IDS is to detect the anomalies
with less false-alarm rate. Based on the detection methodology, IDS can be catego-
rized into two—Signature-based IDS and Anomaly-based IDS [7]. Both approaches
have their own advantages and disadvantages which led the researchers to apply
several techniques such as statistical analysis, data mining, artificial intelligence,
and machine-learning approaches like k-nearest neighbor (KNN), support vector
machine (SVM), artificial neural network (ANN), and logistic regression, etc., are
used in the anomaly detection [8].
Artificial neural networks play an important role in intrusion detection. Recurrent
neural network (RNN) is one among ANN which overcomes the drawback of storage
space unlike other neural networks, [3, 9]. RNN plays an important role in speech
recognition, human action recognition, computer vision, etc. However, maintaining
high detection rate and low false-alarm rate helps in the design of potential IDS,
which significantly out-compete other models [10].
In order to achieve better performance, sine–cosine optimization—recurrent neu-
ral network (SCO-RNN)-based IDS is proposed in this paper which optimizes the
hidden layers [11]. This experiment was carried out using the standard Secure Water
Treatment dataset (SWaT) [12], and the performance of SCO-RNN is validated in
terms of classification accuracy and detection rate (Table 1).
SCO-RNN: A Behavioral-Based Intrusion Detection Approach … 913

Table 1 Related works

Authors Proposed method Objective Applications Advantages Performance metric
Turabieh Dynamic L-RNN Recovery of IoMT Recovery of • Accuracy
et al. missing data application missing data
[13]
Cinar Period-aware Missing Time series Can be • Attention weights
et al. content attention value forecasting performed on
[14] RNN imputation univariate and
multivariate
time series
Liu et al. Modeling Classification Medical Improved • Macro-F1
[15] asynchronous application performance • Precision
event sequences • Recall
Shitharth Intrusion-weighted Classification Attack Increased • Dice
and particle-based detection performance • Jaccard
Prince cuckoo search • precision
Winston optimization • Recall
[16] (IWP-CSP) and • Accuracy
hierarchical • Sensitivity
neuron • Detection rate
architecture-based • Specificity
neural network
(HNA-NN)
Zhao CNN-RNN based Classification Weather Practical • Precision
et al. [8] multi-label prediction implementation • Recall
classification is effective • F-score
Maglaras One class support Detection Information Increased • Accuracy
et al. [2] vector machine security accuracy • System performance
Maglaras IT-one class Detection Information Increased • False-alarm rate
and support vector security performance
Jiang [6] machine
Adepu State condition Analysis Water Easier and • Steady-state
et al. graphs (SCGs) treatment quicker response
[17]
Goh Recurrent neural Detection Time series Identifies • System state
et al. [3] network prediction attacked
sensors

2 SCO-RNN Proposed Methodology

This section discusses the working of proposed improved RNN based on sine–cosine
optimization. The workflow of the SCO-RNN as follows.
Pseudo Code—RNN-SCO

1. 51 Attributes of the SWaT dataset are fed as the input to the RNN
2. Now, 80% data of the SWaT data set are trained and 20% are tested
3. Now, set the upper bound and lower bound, the total number of iteration, and
initialize a constant ‘a’ as 2
4. Now, build an RNN model
5. While building RNN,
914 N. Neha et al.

(a) Initialize the number of units in the input layer to 51, which is the total
number of attributes in the SWaT dataset.
(b) Now, set the number of hidden units to 20 and grant it to the SCO
optimization algorithm as follows:
1. Initialize the set of search agents, upper bound, lower bound, and
constant ‘a’ = 2
2. Evaluate each search agents by the fitness function
3. Update the best solution obtained so far
4. Update the value of x1 , x2 , x3 and x4 as follows:
 
const
a.x1 = const-curriter ∗ (1)
tot_iter

where, const = 2 is a constant, curr_iter is the current iteration, tot_iter

is total number of iterations

b.x2 = 2π ∗ random_number (2)

c.x3 = 2 ∗ random_number (3)

d.x4 = random_number (4)

Here, x1 indicates the region between the destination and solution or

outside it, x2 indicates the distance from destination position, x3 indi-
cates a random weight for the destination, and finally, x4 here changes
depending on the Eqs. (5) and (6).
5. Now, update the position of each search agent as follows:
If x4 < 0.5 then,

X (i, j) = X (i, j) + (x4 ∗ sin(x2 ) ∗ |(x3 ∗ Destination_position ( j) − X (i, j))| (5)

else

X(i, i) = X (i, i) + x4 ∗ cos(x2 )∗ |(x3 ∗ Destination_position ( j) − X (i, j))| (6)

6. Repeat till it reaches the maximum number of iterations.

7. Obtain the best solution as the global optimum.
(c) Finally, set the number of units in the output layer as 10, which is a constant,
given for better performance.

6. Calculate the fitness value as given in Eq. (7)

7. Now, repeat the steps step 5.b.2 to step 5.b.7 and return the best solution as global
optimum
8. Print the accuracy, detection rate, and the fitness value.
SCO-RNN: A Behavioral-Based Intrusion Detection Approach … 915

2.1 Generation of Initial Population

The working of SCO-RNN starts with the initialization of parameters like search
agents, a number of iterations, and computation of parameters like x1 , x2 , x3 , x4 .
x1 depends on the current iteration, total number of iterations and a constant a. x1
parameter is initialized as given in the Eq. (2). x2 and x3 purely depend on random
numbers which is calculated as in Eqs. (3) and (4). The fitness function is chosen
such that it should give an optimized result compared to the existing technique as in
Eq. (7).

2.2 Training and Testing the RNN

The entire SWaT dataset considered for the experiment is divided into training and
testing samples. At first, training is done and then the trained samples and parameters
obtained from each population is used for training the RNN where its performance
is evaluated using fitness function along with the other testing samples.

2.3 Definition of the Fitness Function

The performance of SCO-RNN is assessed using the fitness function as in Eq. (7) in
order to bring out a better and optimized result. The performance of SSO-RNN is
evaluated in the following equation:

f Fit = W1 (DR) + W2 ∗ Acc + W3 (1 − FAR) (7)

where ‘W ’ for weights.

2.4 Termination Condition

The termination condition is satisfied when a maximum number of iteration or opti-

mal number of hidden layers has been reached. If the condition is true, then the
optimal number of hidden units is returned. Then, the population will be updated
with the best fitness value and with its position, using the Eqs. (5) and (6) accordingly
(Fig. 1).
916 N. Neha et al.

Fig. 1 Proposed flow diagram

3 Results and Discussion

3.1 Experimental Setup

SCO-based RNN was implemented on Python 3.6 with INTEL® Core™ i5-7200U
CPU processor @ 2.50 GHz architecture, 8.00 GB RAM and 64-bit OS in x64-based
processor running Windows 10, and WEKA tool is used for the validation process.

3.2 Dataset Description

Secure Water Treatment testbed (SWaT) dataset is a six-stage secure water treatment
which is a scaled-down version of the real-world industrial water treatment plant. This
dataset is collected from Singapore University of Technology and Design (SUTD)
[12]. SWaT is the most complex open-source dataset so far with a total of 51 properties
and 946,722 samples including six dual programmable logic control (PLCs) each for
one stage, along with a backup, 25 sensors, and 26 actuators altogether (Table 2).
Attacks are represented using a six-tuplet as follows: (M, G, D, P, S 0 and, S e )
where

M: Infinite set of procedures to launch the attack

G: Subset of a finite set of attacker events
D: Domain model for the attacks derived from CPS
SCO-RNN: A Behavioral-Based Intrusion Detection Approach … 917

Table 2 Attack types in

Attack category Number of attacks
SWaT
Single stage single point 26
Single stage multi point 4
Multi stage single point 2
Multi stage multi point 4

P: Finite set of attack points

S 0 , S e : Infinite set of attack points.

3.3 Data Pre-processing

Data pre-processing plays a key role in the simplification of obtained real-world

data, which are either incomplete or inconsistent. Here, data normalization is used
to improve the efficiency and accuracy of the proposed technique (SCO-RNN). Data
normalization generally modifies the data from the existing range to a particular
range. Here, normalization is performed for the values on which numerical encoding
is performed, which result in each value ranging between 0 and 1. There are many
methods in normalization out of which the method performed here is as follows:

x − xmin
X Norm = (8)
xmax − xmin

where x: Data point, xmin : Minimal among the data points, xmax : Maximal among
the data points, and X Norm : Data point normalized between 0 and 1.

3.4 Results and Discussions

Figure 2 shows a comparison of the accuracy of proposed SCO-RNN with other

existing techniques. From this, we can infer the accuracy of SCO-RNN is 98.05% is
higher than the existing approaches. The accuracy has been calculated using Eq. (9).

True Positive + True Negative

Acc = (9)
True Positive + True Negative + False Positive + False Negative

Table 3 shows comparison for the detection rate of the SCO-RNN technique
with other existing techniques. The proposed SCO-RNN has a higher detection rate
than the other existing approaches. As the proposed algorithm converges at a global
optimal solution, it achieves higher accuracy and detection rate in minimal time. The
detection rate has been calculated using Eq. (10).
918 N. Neha et al.

Accuracy (in %) Accuracy (in %)

99
98
97
Accuracy

96
95
94
93
Bayes Net Naive Bayes LogisƟc Decision One R SCO-RNN
Stump
Classifiers

Fig. 2 Accuracy

Table 3 Detection rate

S. No. Classifiers Detection rate (%)
1 Bayes net 95.98
2 Naive Bayes 95.1
3 Logistic 96.7
4 Decision stump 95
5 One R 92.1
6 SCO-RNN 97

True Positive
DR = (10)
True Positive + False Negative

4 Conclusions

In this paper, we have proposed an efficient intrusion detection technique that incor-
porates the sine–cosine optimization algorithm for optimizing the recurrent neural
network parameters. The SWaT, a benchmark intrusion dataset has been used for
validating the proposed SCO-RNN. Here, the hyperparameters of the RNN have
been optimized, and the outcome of the system shows a better result than other
machine-learning approaches in terms of accuracy and detection rate.

Acknowledgements This work was supported by Department of Science and Technology, India
through Fund for Improvement of S&T Infrastructure (FIST) programme (SR/FST/ETI-349/2013)
and Tata Consultancy Services. The authors would like to thank the DST-ICPS (T-615) for their
financial support, Department of Science and Technology, India and “iTrust, Centre for Research
in Cyber Security, Singapore University of Technology and Design” for providing the dataset.
SCO-RNN: A Behavioral-Based Intrusion Detection Approach … 919

References

1. Mahmoud MS, Hamdan MM, Baroudi UA (2019) Modeling and control of cyber-physical
systems subject to cyber attacks: a survey of recent advances and challenges. Neurocomput-
ing:1–15
2. Maglaras LA, Jiang J, Cruz TJ (2016) Combining ensemble methods and social network metrics
for improving accuracy of OCSVM on intrusion detection in SCADA systems. J Inf Secur Appl
30:15–26
3. Goh J, Adepu S, Tan M, Lee ZS (2017) Anomaly detection in cyber physical systems using
recurrent neural networks. In: Proceedings IEEE international symposium on high assurance
systems engineering, pp 140–145
4. Senthivel S, Ahmed I, Roussev V (2017) SCADA network forensics of the PCCC protocol.
Digital Invest 22:S57–S65
5. IBM (2018) IBM X-Force threat intelligence index 2018 notable security events of 2017, and
a look ahead. IBM Secur 43
6. Maglaras LA, Jiang J (2014) Intrusion detection in SCADA systems using machine learning
techniques. In: Proceedings of 2014 science and information conference (SAI), pp 626–631
7. Gauthama Raman MR, Somu N, Kirthivasan K, Liscano R, Shankar Sriram VS (2017) An
efficient intrusion detection system based on hypergraph—genetic algorithm for parameter
optimization and feature selection in support vector machine. Knowl Syst 134:1–12
8. Zhao B, Li X, Lu X, Wang Z (2018) A CNN–RNN architecture for multi-label weather
recognition. Neurocomputing 322:47–57
9. Liu H, Lang B, Liu M, Yan H (2019) CNN and RNN based payload classification methods for
attack detection. Knowl Syst 163:332–341
10. Kabir E, Hu J, Wang H, Zhuo G (2018) A novel statistical technique for intrusion detection
systems. Future Gener Comput Syst 79:303–318
11. Mirjalili S (2016) SCA: a sine cosine algorithm for solving optimization problems. Knowl Syst
96:120–133
12. Goh J, Adepu S, Junejo KN, Mathur A (2016) A dataset to support research in the design of
secure water treatment systems
13. Turabieh H, Abu Salem A, Abu-El-Rub N (2018) Dynamic L-RNN recovery of missing data
in IoMT applications. Future Gener Comput Syst 89:575–583
14. Cinar YG, Mirisaee H, Goswami P, Gaussier E, Aït-Bachir A (2018) Period-aware content
attention RNNs for time series forecasting with missing values. Neurocomputing 312:177–186
15. Liu S et al (2018) Modeling asynchronous event sequences with RNNs. J Biomed Inform
83(May):167–177
16. Shitharth S, Prince Winston D (2017) An enhanced optimization based algorithm for intrusion
detection in SCADA network. Comput Secur 70:16–26
17. Adepu S, Mathur A, Gunda J, Djokic S (2007) Algorithms and architectures for parallel
processing. Alg Archit Parallel Process 1:785–798
SSO-IF: An Outlier Detection Approach
for Intrusion Detection in SCADA
Systems

P. S. Chaithanya, S. Priyanga, S. Pravinraj and V. S. Shankar Sriram

Abstract Supervisory Control and Data Acquisition (SCADA) systems play a

prominent role in monitoring and controlling the Critical Infrastructures (CIs) such
as water distribution, nuclear plants, and chemical industries. On the other hand,
SCADA systems are highly exposed to new vulnerabilities as it highly relies on the
internet. Machine learning approaches have been employed to detect the cyberattacks
injected by the attackers in CIs. However, those approaches failed to protect the CIs
against the ever-advancing nature of cyberattacks. This work presents Salp Swarm
Optimization-based Isolation Forest (SSO-IF) to build an efficient SCADA intrusion
detection system, and the experiments were carried out using power system dataset
from Mississippi State University. The performance of SSO-IF was validated over
the state-of-the-art intrusion detection techniques in terms of classification accuracy
and detection rate.

Keywords SCADA · Intrusion detection system · Isolation forest · Salp swarm

optimization

1 Introduction

The Critical Infrastructure (CI) modernizes by the evolution of Cyber-Physical Sys-

tems (CPS) [1]. CPS integrates the physical system with computing and networking
technologies. SCADA systems are an essential part among the CPS that helps to
monitor and control the systems more efficiently and to communicate system issues
to mitigate downtime in CIs such as power plants, gas and oil distribution, electrical
power, water distribution, and chemical processing plants [2]. Primarily, the SCADA
systems are isolated devices, and then these are streamlined by interconnecting all the

P. S. Chaithanya · S. Priyanga · S. Pravinraj · V. S. Shankar Sriram (B)

School of Computing, Centre for Information Super Highway (CISH), SASTRA Deemed to be
University, Thanjavur, Tamil Nadu 613401, India
e-mail: sriram@[Link]
S. Priyanga
e-mail: priyangas54@[Link]

© Springer Nature Singapore Pte Ltd. 2020 921

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
922 P. S. Chaithanya et al.

embedded devices with the internet, which produces high reliability in the automa-
tion of the system. These sophisticated SCADA systems rely on cyberspace, which
makes the system vulnerable to cyberattacks. The cyber attackers focus to create
vulnerability on technical parts of the system in CIf.
Recently, the attacks in SCADA systems have increased tremendously which
results in compromising of basic security measures such as Confidentiality, Integrity,
and Authenticity of the system. This is conspicuous from the “Waterfall” report in
2017, Sophisticated Ukraine Attack causes physical damage to electric substations
and rotating equipment, Compromised Remote Site—A physical breach of remote
substation, ICS Insider which is a disgruntled insider, and it has the ability to par-
tially shut down the ICS-equipped plant by steal passcode of the equipment’s [3].
Therefore, the awareness about reducing these vulnerabilities leads to the evolution
of Intrusion Detection System (IDS).
IDS can be categorized into two, based on the detection techniques namely
Signature-based and Anomaly-based IDS [4]. The major focus of IDS is to achieve
high Detection Rate and reduce False Alarm Rate. Recent research works reveal the
importance of machine learning techniques (Decision Trees, KNN, Support Vector
Machines (SVM), Neural Networks, etc.) in the design of an efficient and adaptive
IDS [5].
This paper proposes the Isolation Forest (IF) technique which directly identifies
anomalies instead of identifying normal data points and constructed an ensemble of
isolation trees [6]. This technique segregates the records by selecting the features
randomly and sets a split value between the supreme and subordinate value. The
anomalies are identified by path length or anomaly score. It performs well for the
high-dimensional dataset, and the parameter used is the number of isolation trees and
sub-sampling size. Parameters are tuned to either increase the predictive power of the
model or to make it easier to train the model. Various optimization techniques like a
genetic algorithm, fruit-fly optimization, whale optimization, moth-flame optimiza-
tion, etc., have been used in the literature for tuning the parameters. Among these,
Salp Swarm Optimization technique [6] with Isolation Forest (SSO-IF) achieves high
detection rate, and it was validated with Mississippi’s power system dataset in terms
of accuracy and detection rate (Table 1).
SSO-IF: An Outlier Detection Approach for Intrusion Detection … 923

Table 1 Related works

Authors Proposed method Dataset Advantages Performance metric
Puggini Isolation forest OES data Better • EV
and performance • ENMSE
McLoone • EMRE
[7] • AUC score
Sun et al. Banzhaf random UCI machine Better • Banzhaf power index
[8] forest learning performance • Information gain
repository—12 ratio
datasets
Maglaras IT-one class SCADA traffic Increased • Accuracy
et al. [9] support vector dataset performance • System performance
machine
Alves and Random forest Six real-world Best • Gain metric
Morris datasets performance
[10]
Abellán Credal random 50 datasets Good • Accuracy
et al. [11] forest from the UCI accuracy,
repository of robust under
machine noise data
learning
Shirazi Random forest 201 High • Erosion ratio
et al. [12] regression experimental accuracy
data points
Nader One-class Mississippi High error • lp NORMS
et al. [13] classification State University detection,
SCADA low FAR
laboratory
Trombetta Critical state Private SCADA More feasible • State-state distance
et al. [14] analysis and state traffic data • State-critical states
proximity distance
• Distance evaluation
accuracy
Shitharth Intrusion Single-hop Increase in • Detection rate
et al. [15] weighted indoor real data performance • False positive rate
particle-based (SIRD), • Precision
cuckoo search single-hop • Sensitivity
optimization outdoor real • Specificity
(IWP-CSO) and data (SORD), • Accuracy
hierarchical multi-hop
neuron indoor real data
architecture-based (MIRD),
NN multi-hop
outdoor real
data (MORD)
924 P. S. Chaithanya et al.

2 SSO-IF Proposed Methodology

This section briefly discusses the working of enhanced Isolation Forest technique
based on Salp Swarm Optimization algorithm (SSO-IF) [16]. The following are
workflow of SSO-IF.
Pseudocode-SSO-IF:
1. Each among 15 datasets is fed as the input to the IF
2. Set the training and the testing ratio of the dataset
3. Now perform the IF with estimators being optimized by SSO technique

a. Initialize the lower and upper bound and total number of iterations
b. Calculate the swarm position and then
c. Update the leader and follower salp position using the parameter rnd1 , rnd2,
and rnd3
d. The leader salp is updated by the following equation
   
BF j + rnd1 ub j − lb j rnd2 + lb j  rnd3 ≥ 0
x 1j ← (1)
BF j − rnd1 ub j − lb j rnd2 + lb j rnd3 < 0

e. Here, the parameter rnd1 is calculated using the Eq. (3), and the rnd2 and
rnd3 are randomly generated
f. The follower salp positions are updated by

1 i 
x ij ← x j + x i−1
j (2)
2
g. Find the best optimum solution for the above
4. Set the best optimum solution as number of estimators
5. Predict the outlier present in the dataset
6. Build the confusion matrix
7. Compute the accuracy and detection rate using the Eqs. (6) and (7).

2.1 Generation of Initial Population

The working of SSO-IF begins with the parameters initialization such as the number
of populations, number of iterations, and computation of parameters like rnd1 , rnd2 ,
and rnd3 . The parameter rnd1 , rnd2 , and rnd3 are random numbers. Based on the
current and total number of iterations, rnd2 and rnd3 are defined and are evenly gen-
erated in the interval of [0, 1] [16]. In order to achieve better optimization result, the
better fitness function is chosen. The parameter number of trees in IF was generated
randomly in a specific range.
SSO-IF: An Outlier Detection Approach for Intrusion Detection … 925

rnd1 = 2e−( I )
4i 2
(3)

2.2 Training and Testing the Isolation Forest

The experiment splits the complete SCADA IDS dataset into training and testing
samples by random sampling. Here, the training and testing data ratio is chosen
randomly by the system in an appropriate ratio.

2.3 Definition of the Fitness Function

In order to compute the performance of SSO-IF, the weighted fitness function is used
and is defined with weights of the DR and FAR as follows:

Fitness_value = 0.8 ∗ DR + 0.1 ∗ (1.0 − FAR) + 0.1 ∗ (1 − (N /N f )) (4)

where N is the no. of selected features, Nf is the total number of features.

2.4 Termination Condition

Here, the termination condition verifies whether it reaches the maximum number of
iteration. If so, it returns the optimal parameter or else it would perform position
updation (Fig. 1).

3 Results and Discussions

3.1 Experimental Setup

SSO-based IF was implemented on Python 3.5 with INTEL® Core™ i3-5005U CPU
processor @ 2.00 GHz architecture running in Windows 10, and validation process
was carried out using Weka tool.
926 P. S. Chaithanya et al.

Fig. 1 Workflow of SSO-IF

3.2 Dataset Description

Power system attack dataset from Mississippi State University has been used for this
model. This dataset totally contains 128 features and is developed based on 37 event
scenarios which include eight Natural Events, one No Events, and 28 Attack Events
[17]. It is divided into three datasets binary dataset (CSV format), class dataset (CSV
format), and multiclass dataset (ARFF format), in which binary dataset consisting
of 15 sets has been used in this work.

3.3 Data Preprocessing

The preprocessing stage generally reduces the dataset size by resampling it or by

eliminating redundant records. It contains two main stages, in which initially, the
integers are bought between the range of 0–1 after the nominal features are mapped
with integer values. Further, to reduce the value range, logarithmic scaling (base
10) is enforced with the integer value ranges. Normalization is one of a kind in
preprocessing. The proposed technique SSO-IF uses min-max normalization which
finds the maximal and minimal feature value and then transforms them into linearly
scaled values ranging from 0 to 1. The formula for min-max normalization is as
follows:
SSO-IF: An Outlier Detection Approach for Intrusion Detection … 927

Fig. 2 Accuracy

x − xmin
X Norm = (5)
xmax − xmin

where x: Data point, xmin : Minimal among the data points, xmax : Maximal among
the data points, and X Norm : Data point normalized between 0 and 1.

3.4 Results and Discussions

Figure 2 shows the accuracy results for SSO-IF compared with other classifiers.
From this, it can be deduced that the accuracy of SSO-IF is higher than the other
techniques. The accuracy is calculated as in Eq. (6).

TP + TN
Acc = (6)
TP + TN + FP + FN

Table 2 shows the detection rate of SSO-IF compared with other classifiers in
which the proposed technique got higher DR when compared with other techniques.
DR is calculated as in Eq. (7)

TP
DR = (7)
TP + FN
928 P. S. Chaithanya et al.

Table 2 Detection rate

Dataset Random forest J48 Multi layer Logistic regression SSO-IF
perceptron
D1 95.4 81.4 76.1 55.5 98.9
D2 94.5 83.8 73.1 58 98.9
D3 96 88.3 65 59.1 98.9
D4 96.8 91.1 79.4 70.1 99
D5 97.7 92.4 85.4 73.5 98.8
D6 96.4 78.9 67.7 38.9 98.4
D7 98.4 85 77.1 62.9 98.5
D8 98.8 87.2 82.5 82.6 98.7
D9 93.7 78.7 75.2 64.7 98.8
D10 97.5 88.9 59.1 60.6 99.1
D11 98.6 82.6 67.9 51.7 98.3
D12 96 89.8 81.8 59.2 98.8
D13 98.6 85.8 70.8 53.4 99.3
D14 98.2 86.5 80.6 54.2 98.5
D15 96.2 84.9 82.4 62.4 99

4 Conclusions

In this work, we have proposed SSO-IF approach in which the parameters of isolation
forest have been optimized using Salp Swarm Optimization Algorithm to obtain
optimal number of trees. The proposed SSO-IF technique has been experimented
and validated using the benchmark power system dataset from Mississippi State
University. The performance of SSO-IF was evaluated using the metrics such as
accuracy and detection rate.

Acknowledgements This work was supported by Department of Science and Technology, India
through Fund for Improvement of S&T Infrastructure (FIST) programme (SR/FST/ETI-349/2013)
and Tata Consultancy Services. The authors would like to thank the MeitY (AAA-22/7/2018-CSRD-
MeitY) for their financial support.

References

1. Goh J, Adepu S, Tan M, Lee ZS (2017) Anomaly detection in cyber physical systems using
recurrent neural networks. In: Proceedings of IEEE international symposium high assurance
systems engineering, pp 140–145
2. Zhang J, Gan S, Liu X, Zhu P (2016) Intrusion detection in SCADA systems by traffic periodic-
ity and telemetry analysis. In: Proceedings—IEEE symposium on computers communications,
pp 318–325
SSO-IF: An Outlier Detection Approach for Intrusion Detection … 929

3. Ginter A (2017) The top 20 cyber attacks against industrial control systems, pp 2–4
4. Almalawi A, Yu X, Tari Z, Fahad A, Khalil I (2014) An unsupervised anomaly-based detection
approach for integrity attacks on SCADA systems. Comput Secur 46:94–110
5. Maglaras LA, Jiang J (2014) Intrusion detection in SCADA systems using machine learning
techniques. In: Proceedings of 2014 science and information conference, SAI 2014, pp 626–631
6. Liu FT, Ting KM (2018) Isolation forest. In: Eighth IEE international conference data mining,
2009
7. Puggini L, McLoone S (2018) An enhanced variable selection and isolation forest based
methodology for anomaly detection with OES data. Eng Appl Artif Intell 2017 67:126–135
8. Sun J, Zhong G, Huang K, Dong J (2018) Banzhaf random forests: cooperative game theory
based random forests with consistency. Neural Netw 106:20–29
9. Maglaras LA, Jiang J, Cruz TJ (2016) Combining ensemble methods and social network metrics
for improving accuracy of OCSVM on intrusion detection in SCADA systems. J Inf Secur Appl
30:15–26
10. Alves T, Morris T (2018) OpenPLC: an IEC 61,131–3 compliant open source industrial
controller for cyber security research. Comput Secur 78:364–379
11. Abellán J, Mantas CJ, Castellano JG (2017) A random forest approach using imprecise
probabilities. Knowl Syst 134:72–84
12. Shirazi SA, Parvandeh S, McKinney BA, Asgharpour A, McLaury BS, Zahedi P (2018) Random
forest regression prediction of solid particle erosion in elbows. Powder Technol 338:983–992
13. Nader P, Honeine P, Beauseroy P (2014) Lp-norms in one-class classification for intrusion
detection in SCADA systems. IEEE Trans Ind Inform 10(4):2308–2317
14. Trombetta A, Masera M, Nai Fovino I, Carcano A, Guglielmi M, Coletta A (2011) A multidi-
mensional critical state analysis for detecting intrusions in SCADA systems. IEEE Trans Ind
Inform 7(2):179–186
15. Shitharth S, Prince Winston D (2017) An enhanced optimization based algorithm for intrusion
detection in SCADA network. Comput Secur 70:16–26
16. Mirjalili S, Gandomi AH, Mirjalili SZ, Saremi S, Faris H, Mirjalili SM (2017) Salp swarm
algorithm: a bio-inspired optimizer for engineering design problems. Adv Eng Softw 114:163–
191
17. Borges Hink RC, Beaver JM, Buckner MA, Morris T, Adhikari U, Pan S (2014) Machine
learning for power system disturbance and cyber-attack discrimination. In: 7th international
symposium on resilient control systems, ISRCS
EDF-VD Scheduling-Based Mixed
Criticality Cyber-Physical Systems
in Smart City Paradigm

G. Naveen Balaji, M. Sethupathi, N. Sivaramakrishnan and S. Theeijitha

Abstract Dynamic data-driven simulation is introduced to improve the scheduling

performance. The customer will need a just in delivery of service scheduling. Mathe-
matical model of the scheduling problem is constructed, and a scheduling method is
proposed to improve the performance of scheduling. Four different optimizations for
the dynamic cloud manufacturing scheduling problems are presented in this paper,
namely average service utilization rate, average task delay time, weighted average
task delay time, proportion of delay tasks and constraints. The scheduling strategies
are constructed and simulated in the SIMSO software.

Keywords Embedded system · Worst-case execution time · Earliest deadline first ·

Earliest deadline first virtual deadline · Preemptive

1 Introduction

System is a manner of operating, organizing or doing one or greater duties in keeping

with the fixed plan, application or set of rules. A system is also an association in which
all its units gather and work collectively in step with the plan or software. An embed-
ded system is mixed operation of hardware and software or additional mechanical
or technical component to carry out desired characteristic. Any sort of tool which
incorporates programmable pc but itself is not always meant to be fashionable motive
laptop is said to be embedded system [1]. The lower layer of an embedded system
is printed circuit board that consists of busses and semiconductor devices. The top
layer is special application layer, and in between those two layers, there are another
vital layers referred to as device drivers and verbal exchange protocols. They need
electrical powered and digital interference. Embedded structures, all the time, are
limited assets to be had in phrases of memory, CPU, show display screen period,
key inputs, diskless, and those parameters play a important thing in the route of the
format, improvement and trying out of such structures.

G. Naveen Balaji (B) · M. Sethupathi · N. Sivaramakrishnan · S. Theeijitha

Department of ECE, SNS College of Technology, Coimbatore, India
e-mail: yoursgnb@[Link]

© Springer Nature Singapore Pte Ltd. 2020 931

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
932 G. Naveen Balaji et al.

1.1 Characteristics of Embedded Systems

In popular, embedded systems are designed to carry out any unique predefined project
that ought to meet any real-time constraint. The primary distinction among a laptop
and an embedded device is a computer is used to carry out a particular project, and
this is predefined by way of manner of the manufacturers.
Assembling all the actual-time tool constraints is a very vital feature of an embed-
ded machine. An actual-time constraint is split into two features. That is hard real-
time system and the upcoming one is soft real-time system [2]. Zero degree of
flexibility is for hard real-time system, and a determined amount of flexibility is
for soft real-time systems. The miss which happens while execution will collapse
the whole output of the system. The small amount of miss is accepted in the soft
real-time system which does not collapse the whole system. Devices which include
MP3s, cameras and TV remotes are the examples of standalone embedded system.

1.2 Architecture of Embedded System

An embedded system is constructed round a processor. The principal processing unit

does the important computation based on the enter it receives from diverse external
devices [3]. The functionality of the CPU is an embedded device is equal because
the capability of the CPU in a desktop, except that the CPU in an embedded system
is much less powerful.
The processor has restricted internal reminiscence, and if this internal remi-
niscence is not always enough for a given application, outside reminiscence gad-
gets are used [4]. The hardware additionally includes any additives that allow the
person-application interaction, including display units, keypads, etc.
A wide variety of 16 and 32-bit microprocessors are to be had from ARM, Atmel,
Intel, Motorola, National Semiconductors and so on. In order to increase an embedded
system with these processors, you first rate deal of peripheral circuitry. However,
microprocessors have better clock speeds and phrase-period, so they are able to
address higher reminiscence [5]. These type of processors are used for high-end
applications inclusive of hand-held computers, Internet get right of entry to devices
and so forth.
The reminiscence utilized in embedded system can be both internal and outside
[6]. The internal memory of a processor is very restricted. For small applications,
if this memory is enough, there is want to used outside reminiscence [7]. Figure 1
shows the system concept of embedded system.
EDF-VD Scheduling-Based Mixed Criticality Cyber-Physical … 933

Fig. 1 System concept of embedded architecture

2 Background

2.1 Cyber-Physical System

A cyber-physical system is the method wherein the computer primarily based set
of regulations is controlled and monitored and tightly included with the internet
and its patron [8]. In CPS, the bodily and software program application additives
are deeply intertwined, every taking walks on special spatial and temporal scales,
displaying more than one and excellent behavioral modalities and interacting with
every exceptional in quite some techniques [9]. Examples of CPS encompass clever
grid, impartial vehicle structures, scientific tracking, method manipulate functions,
robotics and automated pilot avionics.
CPS involves processes, merging concept of cybernetics, mechatronics, design
and manner technological know-how [10]. The system manage is known as embedded
devices. CPS is likewise similar to the (IoT), sharing the same primary structure,
despite the truth that CPS offers a higher mixture and coordination between bodily
and computational elements [9]. Figure 2 shows the architecture of Cyber-physical
system.

2.2 Real-Time Operating System

A real-time operating system (RTOS) imagined to serve real-time applications that

manner facts due to the fact it is far available in, normally without buffer delays.
RT systems require specific help from OS [11]. Tenths of seconds or the shorter
934 G. Naveen Balaji et al.

Fig. 2 Architecture of cyber-physical systems

increments of time are for measuring the working system processing time. RTOS is
a time positive device which has nicely described steady time constraints. Processing
should be carried out inside the described constraints or the failed devices [12]. They
both are event driven or time sharing. Event-pushed structures switch among respon-
sibilities primarily based on their priorities, while time-sharing structures switch the
challenge primarily based on interrupts. Most RTOS uses a preemptive scheduling
a set of rules.

2.3 Kernel

A kernel is the critical part of the working system that manages the operation of the
pc and the hardware most appreciably reminiscence and CPU unit. There are two
types of kernels. A microkernel, which only incorporates fundamental functionality.
A monolithic kernel, which contains many device drivers. It is capable of doing at a
simple level, speaking with hardware and handling assets, consisting of RAM and the
CPU. The kernel plays a machine check and acknowledges additives, which include
the processor, GPU and memory. It also tests for any related peripherals [7]. Kernel
is the software program chargeable for running packages and offering comfortable
access to the machine’s hardware. Since there are many packages, and resources are
constrained, the kernel also decides whilst and the way lengthy a software ought to
run [5]. Figure 3 shows the architecture of Kernel.
EDF-VD Scheduling-Based Mixed Criticality Cyber-Physical … 935

Fig. 3 Architecture of kernel

2.4 RTOS Scheduling Models

• Cooperative multitasking
• Preemptive multitasking
• Rate-monotonic scheduling
• Round robin scheduling.

2.4.1 Cooperative Multitasking

Non-preemptive multitasking is also known as cooperative multitasking is a style

of pc multitasking in which the strolling device in no manner initiates a context
switch from a system to each other technique. Strategies voluntarily yield and manage
periodically in order to permit multiple programs to be run concurrently [13, 14]. This
sort of multitasking is referred to as “cooperative” due to the fact all packages must
cooperate for the entire scheduling scheme to paintings. The technique scheduler of
an operating tool is known as a cooperative scheduler, having its function decreased
right down to beginning the approaches and permitting them to move lower back
control again to it voluntarily.
In comparison, preemptive multitasking interrupts programs and offers con-
trol to other methods out of doors the applications manipulate [15]. Cooperative
multitasking is used with look ahead to in languages with a single-threaded event.

2.4.2 Preemptive Multitasking

The time period preemptive multitasking is used to differentiate a multitasking oper-

ating tool, which allows preemption of obligations, from a cooperative multitasking
system wherein strategies or responsibilities should be explicitly programmed to
936 G. Naveen Balaji et al.

yield once they do no longer need gadget resources [16]. In easy terms: Preemptive
multitasking includes using an interrupt mechanism which suspends the presently
executing way and invokes a scheduler to determine which way needs to execute
next. Therefore, all techniques get a few amount of CPU time at any given time. In
preemptive multitasking, the working device kernel can also initiate a context trans-
fer to fulfill the scheduling insurance’s precedence constraint, therefore preempting
the lively undertaking. In general, preemption technique “earlier seizure of” is used
when the excessive priority assignment at that stage seizes the present undertaking
tasks, then it is referred to as preemptive scheduling.
Although multitasking strategies were in the beginning developed to allow mul-
tiple users to proportion a single system, it quickly has become apparent that mul-
titasking becomes beneficial no matter the number of users [17]. Many working
structures, from mainframes down to unmarried-person non-public computer sys-
tems and no-user manage structures (like those in robot spacecraft), have diagnosed
the usefulness of multitasking assist for a variety of reasons. Multitasking makes it
viable for a undetermined consumer to run multiple applications at the equal time,
or to run “historical past” procedures even as keeping manage of the laptop.

2.4.3 Rate-Monotonic Scheduling

In pc technological knowledge, rate-monotonic scheduling (RMS) is a set of rules

with priority utilized in real-time operating systems (RTOS) with a static-precedence
scheduling elegance [18]. The static-level priorities are assigned in line with the cycle
period of the pastime, so a shorter cycle length results in a higher venture precedence.
Rate-monotonic system is used alongside the ones systems to provide scheduling
ensures for a selected application.

2.4.4 Round Robin Scheduling

Round robin scheduling (RR) is one of the algorithms through device and community
schedulers [19]. As the time period is generally used, time slices (additionally called
time quanta) are assigned to system of every tasks in identical quantities and in round
order, handling the determined processes without precedence (additionally referred to
as cyclic government). Round robin scheduling is straightforward, smooth to position
into impact, and hunger-loose [9]. Round robin scheduling also can be completed
to different scheduling troubles, together with records packet scheduling in laptop
networks.
EDF-VD Scheduling-Based Mixed Criticality Cyber-Physical … 937

3 Existing System

3.1 Estimation of WCET

The worst-case execution time of ti is same to the smallest of t gratifying the

subsequent equality [20].

 t 
t = Cit + Cj. (1)
j<i
Tj

C it Worst-case execution time

Di Relative deadline parameter
T i Inter arrival separation time.

The worst-case workload of i with maximum precedence tasks over a determined

interval of period t,

 t 
Ci (2)
i
Tj

The worst-case workload of i with maximum precedence tasks over a determined

interval of period t,

Hi (t) = t − W (t) (3)

The pseudo-inverse function X i (c) of H i (t) is the smallest,

X i (c) = min{: Hi (t) ≥ c} (4)

t

The worst-case response time Ri of task is given by

Rit = max {X i − 1(kCi ) − (k − 1)Ti } . . . (5)

k=1,2,...

3.2 Task Response Time

The task response time has been estimated for the processor,
Kth job in task T i

Rit = max {X i − 1(kCi ) − (k − 1)Ti } (6)

k=1,...k
938 G. Naveen Balaji et al.

K* < +∞

Rik = X i − 1(kCi + I ) − (k − 1)Ti ≥ X i − 1(kCi ) − (k − 1)Ti (7)

It is non-decreasing function, [21]

Rit ≥ max{Rik } ≥ max {X i − 1(kCi ) − (k − 1)Ti } (8)

k>k∗ k>K ∗

We assume that

∀x :: f lb(x) ≤ f (x) ≤ f ub(x) (9)

Worst-case response time Rit ,

Wiub (t) ≥ Wi (t) (10)

Relationship for idle time,

Hilb (t) = t − Wiub (t) ≤ t − Wi (t) = Hi (t) (11)

The relationship between pseudo-inverse function,

 
X iub (c) = min t:Hilb (t) ≥ c ≥ min{t:Hi (t) ≥ c} = X i (c) (12)
t t

 
Riub = max ub
X i−1 (kCi ) − (k − 1)Ti ≥ Ri (13)
k=1,2,...

The maximum amount of time that the processor executes,


t
Wi (t) = w j (t) (14)
j=1

W 0j (t) is the maximum amount of time that the processor executes the task at any
time interval,

∀ j ∀t w0j (t) ≥ w j (t) (15)

The equation of linear bound is given by

 
w0j (t) ≤ U j t + C j 1 − U j (16)

The linear upper bound function,

EDF-VD Scheduling-Based Mixed Criticality Cyber-Physical … 939


i 
i
Wi (t) = w j (t) ≤ w0j (t)
j=1 j=1


i
  
≤ U j t + C j 1 − U j = Wiub (t) (18)
j=1

The worst-case response time for upper bound,

Ci+ j<i C j (1−U j )

Ri ≤ = Riub (19)
1 − j<i U j

i
  
Wiub (t) = Ujt + Cj 1 − Uj (20)
j=1
⎛ ⎞

t 
t
  
Hilb (t) = t ⎝1 − Ui j ⎠ − C j 1 − Ui j (21)
j=1 j=1

It is invertible,
i  
h+ C j 1 − Ui j
j=1
X iub (h) = (22)
1 − ij=1 Ui j

4 Proposed System

4.1 The Dynamic Cloud Manufacturing Scheduling

4.1.1 The Average Service Utilization Rate

⎛ ⎞  

N 
ni N
Ur (t1 , t2 ) = ⎝ u i, j (t1 , t2 ) ⎠ n ir (23)
i=1 j=1 i=1

The average carrier utilization price U is calculated via price ui,j within the cloud
manufacturing platform.
940 G. Naveen Balaji et al.

4.1.2 Average Task Delay Time

Reducing the common task put off time of all duties

 M(t ,t ) 
 1 2

Dr (t1 , t2 ) = di (Mr (t1 , t2 )) (24)

i=1

Dr (t1 , t2 ) is for average task delay time of the cloud manufacturing platform.
The delay time d i of task can be calculated

dir = Fir − air (25)

The completion time F i of T i is equal to the last subtask of T i

Fir = f ir,mir (26)

4.1.3 Weighted Average Task Delay Time

In cloud manufacturing environment, the different tasks will have different priority,
and the weighted average delay time W r (t 1 , t 2 ) of all tasks should be considered and
can be calculated
 M (t ,t )   M (t ,t ) 

ir 1 2 r1 2

Wr (t1 , t2 ) = pir dir pir (27)

i=1 i=1

4.1.4 Proportion of Delayed Task

The proportion of delayed task during time span [t 1 , t 2 ]

|B|
Rr (t1 , t2 ) = (28)
Mr (t1 , t2 )

B = {Ti /Ci < Fi , t1 ≤ ai , Fi ≤ t2 } (29)

4.1.5 Constraints

There are varieties of constrains for the scheduling solutions for the dynamic cloud
production scheduling problems
EDF-VD Scheduling-Based Mixed Criticality Cyber-Physical … 941

• Constrains of service selection

• Constrains of subtask start time
Constrains of Service Selection
For the constrains of service selection, the subtask can handiest pick provider which
can be able to execute the subtasks

vh,g = 1 (30)

where

h r = h x,y (31)

gr = gi, j (32)

Constrains of Subtask Start Time

Here, there are two tasks which are executing in a platform. The begin time of an
intermediate mission cannot be earlier than the completion time of the front subtask

air ≤ z i,1 (33)

f i, j−1 ≤ z i, j (34)

where 1 ≤ i ≤ Mr (t 1 , t 2 ) and 2 ≤ j ≤ mi .

5 Results and Discussion

Figures 4, 5, 6, 7, 8, 9 and 10.

6 Conclusion

The challenge contributions may be summarized as follows: the test for a WCET
estimation with EDF scheduling algorithm that adopts a worldwide method to task
allocation upon uniprocessors. That is, the behavior of algorithm EDF—one such
formerly described [10, 12] static-precedence global scheduling set of rules upon
uniprocessor systems. The simple sufficient conditions for determining whether any
given periodic undertaking system will be successfully scheduled by way of algo-
rithm EDF-VD upon a given uniprocessor platform. The situation of the reaction
time was calculated for further scheduling of obligations under EDF-VD algorithm.
942 G. Naveen Balaji et al.

Fig. 4 Gantt chart for uniprocessor scheduler under earliest deadline first

Fig. 5 Timing overhead for uniprocessor scheduling under earliest deadline first
EDF-VD Scheduling-Based Mixed Criticality Cyber-Physical … 943

Fig. 6 Task distribution under earliest deadline first scheduler

Fig. 7 CPU cycles (save and load count)

944 G. Naveen Balaji et al.

Fig. 8 Load by CPU under simulation

Fig. 9 Log file of the scheduler under earliest deadline first algorithm
EDF-VD Scheduling-Based Mixed Criticality Cyber-Physical … 945

Fig. 10 Log file of the scheduler under earliest deadline first algorithm

The outputs were visualized the use of the Gantt charts. The deadlines and miss had
been honestly seen within the output.
The circumstance of earliest deadline first together with virtual deadline by esti-
mating the shortest last time set of rules, the system model can be designed, and
outputs might also explicit sufficient schedulable project over uniprocessor plat-
forms. The uniprocessor set of rules can be better in addition to create a platform on
multiprocessor environments.

References

1. Nelissen EG, Nélis V, Tovar E (2015) How realistic is the mixed-criticality real-time system
model? In: Proceedings of the 23rd international conference on real time and networks systems
(RTNS ’15). New York, pp 139–148
2. Liu C, Layland J (1973) Scheduling algorithms for multiprogramming in a hard real-time
environment. J ACM 20(1):46–61
3. Davari S, Dhall SK (1986) An on-line algorithm for real-time tasks allocation. In: Proceedings
of real-time systems symposium, pp 194–200
4. Funk S, Goossens J, Baruah S (2001) On-line scheduling on uniform multiprocessors. In:
Proceedings of real-time systems symposium, pp 183–192
5. Burns, Davis R (2016) Mixed-criticality systems: a review, 7th edn. [Link]
[Link]/~burns/[Link]. Accessed 37 Mar/Apr 2018
6. Oh DI, Baker TP (1998) Utilization bounds for N-processor rate monotone scheduling with
static processor assignment. Real-Time Syst Int J Time-Crit Comput 15:183–192
946 G. Naveen Balaji et al.

7. Buxton JN, Randell B (eds) (1970) Software engineering techniques: report of a conference
sponsored by the NATO science committee, 27–31 October 1969. Scientific Affairs Division,
NATO, Rome, Brussels
8. Dertouzos M (1974) Control robotics: the procedural control of physical processors. In:
Proceedings of IFIP congress, pp 807–813
9. Vestal S (2007) Preemptive scheduling of multi-criticality systems with varying degrees of
execution time assurance. In: Proceedings in real-time systems symposium. Tucson, AZ, pp
239–243
10. Dhall SK, Liu CL (1978) On a real-time scheduling problem. Oper Res 26:127–140
11. Leung J, Whitehead J (1982) On the complexity of fixed-priority scheduling of periodic, real-
time tasks. Perform Eval 2:237–250
12. Burns, Davis R (2013) Mixed criticality on controller area network. In: Proceedings of 2013
25th Euromicro international conference on real-time systems (ECRTS ’13). Paris, pp 125–134
13. Baruah S, Burns A (2011) Implementing mixed criticality systems in Ada. In: Romanovsky
A, Vardanega T (eds) Reliable software technology—Ada Europe. Springer, Edinburgh, pp
174–188
14. Baruah S (2016) Schedulability analysis for a general model of mixed criticality recurrent
real-time tasks. In: Proceedings of 2016 IEEE real-time systems symposium
15. Fenn J, Raskino M (2008) Mastering the hype cycle: how to choose the right innovation at the
right time, Harvard Business School Press
16. Sha L, Rajkumar R, Lehoczky J, Ramamritham K (1988) Mode change protocols for priority-
driven preemptive scheduling. J Real-Time Syst 1(3):243–264
17. Sharma RK et al (2005) Balance of power: dynamic thermal management of internet data
centers
18. Ernst R, Di Natale M (2016) Mixed criticality systems—a history of misconceptions? IEEE
Des Test 33(5):65–74
19. Davari S, Dhall SK (1985) On a real-time task allocation problem. In: Proceedings 19th Hawaii
international conference on system science
20. Baruah S (2016) Schedulability analysis of mixed-criticality systems with multiple frequency
specifications. In: Proceedings of the 16th international conference on embedded software
(EMSOFT). Pittsburgh
21. Baruah S, Chattopadhyay B (2013) Response-time analysis of mixed criticality systems with
pessimistic frequency specification. In: Proceedings of IEEE international conference on
embedded real-time computing systems application (RTCSA). Taipei, Taiwan
Automatic Solid Waste Dumping Alert
System

K. Bhuvana, S. Deeksha, M. Deepthi and A. D. Radhika

Abstract One of the fundamental problems the world facing today is waste man-
agement. The main problem with waste management is that the dustbins placed by
the municipal company are filled earlier and overflow before the next cleaning pro-
cess (Mirchandani in 2017 International conference on big data, IoT and data science
(BID), pp 73–76, 2017 [1]). This poses environmental threats and causes health prob-
lems. This dangerous scenario can be avoided by installing an alarm system for the
dumping of solid waste. The main objective of this paper is to alert the municipal
web server to clearance of overloaded dustbins when solid waste flows over the pub-
lic dustbin. Here, we can build the dustbins using the MCU, RFID, GPS, ultrasonic
sensor and Wi-Fi module to prevent the overflow of the dustbins.

Keywords Node MCU · GPS · RFID · Ultrasonic sensor · Wi-Fi

1 Introduction

India’s main problem is waste management, which faces rapid population growth,
disorganization of the municipal government, lack of public awareness and public
involvement. Because of this, the overflow of waste in dustbins leads to an unhygienic
environment, which poses health problems, and there is also no clearance of the waste
on time. People should be responsible enough to use the dustbins properly and not
throw the waste out of the overloaded dustbins [2]. To prevent this, the dustbins can
be properly placed. Currently, we do not find advanced systems that use the method
of alerting municipal web servers to overfilled dustbins when the amount of garbage

K. Bhuvana · S. Deeksha · M. Deepthi · A. D. Radhika (B)

Department of Computer Science and Engineering, Vidyavardhaka College of Engineering,
Mysuru, Karnataka, India
e-mail: [Link]@[Link]
K. Bhuvana
e-mail: bhuvanak703@[Link]

© Springer Nature Singapore Pte Ltd. 2020 947

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
948 K. Bhuvana et al.

exceeds the capacity of the dumpster. This system is mainly focused on this. In order
to overcome these problems, we use dumpsters that include alert systems to alert the
municipal web server so that they can identify the locations of the dumpsters using
the GPS, and RFID is used for the unique ID verification process and helps to provide
an alert system for the overloaded dumpsters. When the dustbin is overloaded, the
Wi-Fi module sends the notice to the individual concerned. By using this technology,
we can achieve the cleanest and efficient use of intelligent dumpsters at a lower cost
[3].

2 Existing Methodology

The existing solid waste management system includes the conventional dumping
method in a dustbin, and then, the cleaning process performed by the municipality
concerned. Although this method is concerned with the management of solid waste,
it can lead to unhygienic environments, air pollution and some health problems if
there is no proper clearance of waste on time [4]. And there is also no proper way to
automatically alert the municipality concerned about clearing the filled dustbins. The
waste management system in different countries can be different, and it will also be
different in urban and rural areas [5]. It is, therefore, necessary to have an automatic
system that senses the waste level in dustbins and alerts the people concerned about
the cleaning process so that we can avoid overloading the dustbins. This paper is
about giving an alert message about the fully filled dustbins to the municipal web
server. Here, we use the efficient and cost-effective node MCU instead of Arduino,
using sensors, we feel the level of waste, and RFID tags and GPS are used to locate
the fully filled dustbin areas in an amortized manner. This paper, therefore, proposes
an automatic solid waste management system that is cost-effective and efficient.

3 System Architecture

We use components such as GPS, MCU node, wireless module, RFID and ultrasonic
sensors in system architecture. If the dustbins contain an MCU node, an ultrasonic
sensor is connected to the MCU node. This is essentially used to detect the level of
dumpsters, whether or not they are fully filled. Then, if the sensor senses the dump
level as fully filled, it uses the Wi-Fi module and sends a message to the person or
web server of the municipal corporation that the dumpsters are cleaned as they are
fully filled. The related people of the municipality will then know that the dustbin is
fully filled with the alert system provided. In this case, the GPS is used to locate the
location of the dustbins by sending the current position of the dustbin or the actual
tracking position of the dustbin, and the RFID is used to provide each dustbin with
a unique identification number to ensure which of the dustbin in a located area is
to be cleaned. This can be done using the RFID tag and the RFID reader where the
Automatic Solid Waste Dumping Alert System 949

Fig. 1 Architecture of components used in dustbin

RFID reader is used to obtain information from an RFID tag that is also used for
tracking objects. Radio waves are used here to transfer the data to a reader from
the tag. Therefore, we can locate the dustbin using the GPS and RFID tags, and the
cleaning process is performed by the individual concerned [6] as shown in Fig. 1.

4 Proposed Methodology

4.1 Node MCU

Node MCU is an open-source IoT platform, and it uses a Lua scripting language, a
lightweight multi-paradigm programming language designed primarily for embed-
ded systems and customers. MCU node is a development board with a popular Wi-Fi
chip ESP8266. Like a microcontroller, ESP8266 can be programmed. It can, there-
fore, be of great benefit to Arduino, who can connect to the Internet via Wi-Fi. The
breakout board ESP8266 has limited pins in the chip itself with many output ports
[7] as shown in Fig. 2.

Fig. 2 Node MCU

950 K. Bhuvana et al.

Fig. 3 Ultrasonic sensor

4.2 Ultrasonic Sensors

The HC-SR04 ultrasonic sensors are made of piezoelectric crystals, which use high-
frequency sounds to resonate the desired frequency and convert acoustic energy into
electric energy and vice versa. The exact distance between the object and the sensor
is measured, and the time intervals between transmission and reflection are measured
as shown in Fig. 3. The distance change is calculated repeatedly and produced using
the formal [8]

Test distance = high level time ∗ velcity of sound(340 M/S)/2.

This multiplies the value by 1/2 or 0.5 because ‘t’ is the time for the distance to go
and return. Initially, a pulse is generated at the base of the distance in the ultrasonic
sensor to send the data to the MCU node. The starting pulse is approximately 10 µs,
and the PWM signal at the base of the distance is 150–25 µs. If there is no obstacle,
the 38 µs pulse is generated for the MCU node, which determines that no objects
are present.

4.3 GPS

GPS is a device that interacts with the GPS satellites to receive the information
required from the satellites required to calculate the geographical position of the
device. In this article, we will interface the GPS module with the MCU node. A
simple server of the local web, i.e. using node MCU, the municipal server is created,
and the location details are updated on that server website. The GPS is used here to
send tracking position data in real time.
Automatic Solid Waste Dumping Alert System 951

4.4 Wi-Fi Module

Stations (STA) are those devices connected to the wireless network. An access point
(AP) provides a Wi-Fi connection. The other access point end is connected to the
cable network. Wi-Fi network accesses point to the Internet. SSID recognizes every
access point where SSID means identifier service set. This is the network name you
choose when you connect a station to the Wi-Fi (or call it a device). Each ESP8266
module can be connected to the Wi-Fi network as a station. It is thus able to connect
stations to these modules. The third ESP8266 can operate both in the station and soft
point mode at the same time as shown in Figs. 4, 5 and 6.

4.5 Radio-Frequency Identification

Radio-frequency identification (RFID) is essentially used for the purpose of verifi-

cation. Radio waves are used here for reading and capturing information stored on
a tag attached to an object. The main advantage over bar code is that, unlike a bar
code, it can be used to read the tag from a distance of up to several feet and does not

Fig. 4 ESP8266 operating in the soft access point mode

Fig. 5 ESP8266 operating in the station mode

952 K. Bhuvana et al.

Fig. 6 ESP8266 operating in the station + soft access point mode

need to be in the reader’s direct line-of-sight. RFID can, therefore, be used instead of
the bar code. An RFID reader is used here to read the information from the tag and
therefore to carry out the identification process. It has mainly two types of active and
passive RFID. Because the tags do not require batteries or maintenance in passive
RFID, we can use them here. The tags are also small enough to fit a label. And it has
three parts, an antenna, a semiconductor chip and a kind of encapsulation attached
to it. These are used to perform the actual identification operations in which the
antenna captures energy and transfers to the ID of the tag, and the chip coordinates
this process, whereas the encapsulation maintains the integrity of the tag and protects
it against external conditions [9, 10].

5 Conclusion

The aim of the solid waste dumping alert system is to promote a decent quality
of life, a clean and sustainable environment. The system proposed is an approach
for the planning and management of solid waste. This system prevents the irregular
cleaning of public bins by sending an alert to the person concerned at a regular time.
The technologies used here give visibility to the management of solid waste and help
to clear it regularly in public dustbins. The clearance here depends on whether the
dumpster is full, half or empty. It also proposes reducing costs and the efficient use of
intelligent dumpers. In general, it is intended to create a clean and green environment.

Acknowledgements The authors express gratitude towards the assistance provided by Accendere
Knowledge Management Services Pvt. Ltd. in preparing the manuscripts. We also thank our mentors
and faculty members who guided us throughout the research and helped us in achieving desired
results.
Automatic Solid Waste Dumping Alert System 953

References

1. Mirchandani S, Wadhwa S, Wadhwa P, Joseph R (2017) IoT enabled dustbins. In: 2017
International conference on big data, IoT and data science (BID), pp. 73–76
2. Vasagade TS, Tamboli SS, Shinde AD (2017) Dynamic solid waste collection and management
system based on sensors, elevator and GSM. In: 2017 International conference on inventive
communication and computational technologies (ICICCT), pp. 263–267
3. Shrivastava P, Mishra S and Katiyar SK (2015) A review of solid waste management tech-
niques using GIS and other technologies. In: 2015 International conference on computational
intelligence and communication networks (CICN), pp. 1456–1459
4. Riya AK, Yuvraj KK (2018) A review paper on IoT based smart garbage alert system. Int J Sci
Res 7(5):1810–1813
5. Pardini K, Rodrigues JJPC, Kozlov SA, Kumar N, Furtado V (2019) IoT-based solid waste
management solutions: a survey. J Sens Actuator Netw 8(5):1–25
6. Draz U, Ali T, Khan JA, Majid M, Yasin S (2017) A real-time smart dumpsters monitoring and
garbage collection system. In: 2017 Fifth international conference on aerospace science and
engineering (ICASE)
7. Mamatha D, Priyanka KE, Nidhi R, Pooja K (2017) Smart garbage monitoring system using
node MCU. In: international conference on signal, image processing communication and
automation, pp. 238–245
8. Kumar NS, Vuayalakshmi B, Prarthana RJ, Shankar A (2016) In: 2016 IEEE region
10 conference (TENCON) IOT based smart garbage alert system using Arduino UNO,
pp. 1028–1034
9. Manisha J, Sanket H, Lakshmi HR (2018) SmartBin-automatic waste segregation and col-
lection. In: 2018 Second international conference on advances in electronics, computers and
communications (ICAECC)
10. Muthukumaran P, Sarkar SB (2013) Solid waste disposal and water distribution system using the
mobile adhoc network. In: 2013 international conference on emerging trends in communication,
control, signal processing and computing applications (C2SPCA)
User Authentication for Mobile Phones
by Pseudo Signature Using CNN

Christy James Jose and M. S. Rajasree

Abstract Today, majority of the world population has one companion with them
always. This companion has all the private information of the person. That compan-
ion is the smart phone. Considering the private data its holding, the unauthorized
usage by friends, family members and strangers are to be avoided. The traditional
way of preventing the unauthorized usage is to lock the device with PIN numbers,
graphical patterns, fingerprint sensors and face unlock mechanisms [1]. The last two
are available with high-end phones only. Even these mechanisms rely on the first-
mentioned methods in case of a sensor failure. Primary authentication mechanisms
could be easily breached. In this paper, we are investigating the usage of handwrit-
ten signatures as an authentication method. We are suggesting a Pseudo signature, a
signature drawn on the touch screen of the phone using the finger tip. We have used
convolutional neural network classifiers to classify the genuine user and intruder.
Experimental results show our suggestion is promising and it could be used as an
easy-to-use user-friendly primary authentication for smart phones.

Keywords Authentication · Classification · Smart phones · Convolutional neural

networks

1 Introduction

Primary authentication on smart phones are based on PIN, GUI passwords, fingerprint
sensors and face unlock mechanisms. Most of the users are either satisfied with these
mechanisms or does not use it at all [2]. Protection of private data and prevention
of unauthorized usage of the device are the need of the hour. Problems with these

C. J. Jose (B)
Government Engineering College, Barton Hill, Thiruvananthapuram, Kerala, India
e-mail: jjchristy@[Link]
M. S. Rajasree
APJ Abdul Kalam Technological University, Thiruvananthapuram, Kerala, India
e-mail: rajasree40@[Link]

© Springer Nature Singapore Pte Ltd. 2020 955

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
956 C. J. Jose and M. S. Rajasree

methods lies in the fact that all these methods can be breached and ownership of
the device can be taken over by an intruder [3]. Our proposal suggests handwritten
signature, more precisely, pseudo signature drawn on the touchscreen of the mobile
phones for authentication. The motivation behind the suggestion is the fact that
graphical passwords or PIN numbers can be stolen by intruders by the methods
like shoulder surfing and smudge attack. Moreover, there is a chance of forgetting
the pattern or PIN number by the genuine user himself. Signatures have been used
by everyone for the purpose of identification and banking purposes. The advantage
of using signature is that usually nobody forgets his signature and we assume that
nobody can imitate others signature hundred percentage perfect. So, any person
imitating a genuine users signature to unlock a mobile device will be blocked instantly
and we expect this method will have highest rate of user acceptance since there is no
need to remember anything other than his own signature which is a very personal bio
metric. Here, what we suggest is that every user has to draw his signature with his
fingertip on the touch screen of his mobile phone. The device will verify the signature
with an already trained model and grants access based on the classifier result.

2 Related Work

Khuwaja and Laghari [4] proposed an offline signature recognition system to identify
the handwritten signatures using a neural network which is trained with the low-
resolution scanned images of the handwritten signatures. Eventhough they could
achieve an accuracy of 98%, there was no mention about the capability to reject
a forged signature. We could see so many works on signature verification, but the
difference with our work is that we are analyzing a signature drawn on the touch
screen using the finger, whereas most of the works have studied signature drawn
on paper using a pen or signature drawn using a stylus on touch screen. Chen et al.
[5] suggested pseudo signature as a bio metric, but the data they have created was
of shapes like circle, triangle, etc., drawn by different subjects. They could achieve
an FAR of 20%. Nowadays, the number of phones came bundled with a stylus are
few. Some of the works had used the well-known algorithms. Fischer et al. [6]
used dynamic time warping (DTW). In Wijesoma et al. [7] had used the root-mean-
square (RMS). Hidden Markov Model (HMM) was used by Fierrez et al. [12]. These
algorithms were used match the time-series data of the signature. The works by Feng
and Wah [9] were utilizing a statistical similarity evaluation by making use of the
descriptive features of the signature. Fierrez Aguilar et al. [8] has projected unique
features like the total duration taken to draw the signature, pen pressure and number
of times the pen lost its contact. Our work proposes deep learning as a solution to our
problem of classification. We had came through the work by Iranmanesh et al. [10]
that has analyzed the possibility of employing machine learning for the classification.
According to them, they have used it to improve the accuracy. Their work has shown
User Authentication for Mobile Phones by Pseudo Signature Using CNN 957

an accuracy of 82.42 but failed in recognizing an imitated signature. Auto encoder

was used in the work by Nam et al. [11] and they could achieve an accuracy of 92%
and were successful in recognizing the imitated signatures.

3 Proposed Method

The block diagram of the work is illustrated in Fig. 1. We would capture the signature
as an image using a suitable application. Then, use this image to recognize the user.
Since its been drawn casually on the screen, it may not be exactly matching a persons
signature drawn on a paper. So matching a pseudo signature with a signature drawn on
paper is out of scope. But we can be sure that the movement of your finger to draw the
signature on the screen would be showing away the same direction and almost same
speed in both [Link] Phase 1, Enrollment, Training and Validation, the signatures
of both genuine user and intruder are collected and labeled to form the training
and testing data. As we can see, the collected number of samples would be low in
numbers. In a real-time situation, this would be true. No user would be interested in
drawing a lot of his signature for enrollment. Keeping in mind this situation, we are
suggesting data augmentation so that the number of samples required for training the
CNN would be sufficient. Operations performed were rotation up to 10◦ , zoom, shear
and skew. Our assumption was that the pseudo signatures are time variant and it has
all the above-said deformations over time. For our test, we have used 260 samples
for training and 32 for validation. Finally, individual signature samples were tested
using the trained model. Finally, individual signature samples were tested using the
trained model. We can set threshold levels for the confidence levels of the classifier.

Signature Images of
Genuine User Labelled T
Signature Fully connected Trained
Signature Images of Intruder
Convolutional Layer - Model -
Labelled F
Neural Network Classification Classifier

Phase 1. Enrolment Training & Validation

Access Granted
GENUINE User

Signature Trained
Model -
Classifier
INTRUDER Access Denied

Phase 2. Testing Phase

Fig. 1 Block diagram

958 C. J. Jose and M. S. Rajasree

Fig. 2 Signature pad android app

We may grant access to non-critical applications if it is greater than 60% similarity

and for critical applications, it may be greater than 90%.

3.1 Data Set Creation

We have created our own data set for the experiment. We have chosen three subjects
for this experiment. All male candidates from our department in the age group 24–
43 years. Out of this three subjects, 1 is designated as genuine one and other two as
intruders. All data has been generated in a single session. Prior to data collection,
the designated genuine user was asked to display on the board how he is drawing
his signature. The details include the staring point, the direction followed and the
normal speed. This was repeated more than once. Genuine user was also directed
to sign on the touch screen in the presence of other subjects so that they can see
the signature details. For the data collection, we have used an App “Signature Pad”
from the Google Play Store. Screen shot of the app is given in Fig. 2. It essentially
stores whatever a user draws in the given window on the touch screen of the phone as
an image in the phone memory in JPEG format. The Signature Pad application has
a fixed size window. On this window, all subjects are supposed to draw the pseudo
signature using their finger tip. We have generated the data set in a single session. All
the signature images are stored in the internal memory of the phone. We have copied
these images and labeled them. Label T is given to the genuine user signature and
User Authentication for Mobile Phones by Pseudo Signature Using CNN 959

Fig. 3 Signature samples

label F to all other image. So we have taken more than 20 numbers of samples from
the genuine user and others are asked to imitate the signature (Fig. 3). Help from
the genuine user was there as to show how he signs. This is to ensure that whoever
tries to imitate the signature will have a close resemblance to the original one and
we would study performance of the classifier in a perfectly mimicked signature.
Data augmentation was performed on both training and test data, so as to increase
the number of samples. Then, these images were used for training and testing our
convolution neural network. We have used the Python 3.6 along with Keras deep-
learning library for all the coding and simulations.
In CNN-based classifier, convolution is performed on the image data set and it
extracts features in that image. Ultimately, it helps the machine to learn the image
characteristics. Pooling is the technique used to reduce the image size without losing
features. Flattening is used to transfer two-dimensional matrix of features into vector
features, so that it can be given as input to a neural network classifier (Fig. 4).

4 Results and Discussion

In this study we have used CNN for feature extraction and a fully connected layer
for classification of genuine user vs an intruder. The training and validation setup
are described as follows, Th epoch set to 5 after some trial and error, each epoch has
taken approximate 1–1 min and twenty seconds to complete. Both training loss and
960 C. J. Jose and M. S. Rajasree

Fig. 4 Model summary

Fig. 5 Training graph

validation loss were decreasing. We have got training accuracy of 99% (Fig. 5) and
validation accuracy of 97% (Fig. 6). This shows our proposal is good alternative to
traditional primary authentication methods.
Normally, the signature drawing on the touch screen will generate sensor data
especially from the accelerometer and the sensors associated with the touch screen,
like the coordinates the pressure finger size etc. These features can also be used for
recognizing the user, but in this work that is not taken into consideration.
The signature bio metric is known to vary with respect to time, moreover we are
drawing our signature using finger which is more likely to have deformations every
time. Another factor is the way an user holds the phone during his signing process.
Our data set was collected in a controlled environment. Every subject was sitting
and holding phone in their hands. In reality they may be lying, walking or the phone
may be on a desk. Situations mentioned above were not taken into consideration. It
has been observed that during our data collection, frequent errors are happening in
User Authentication for Mobile Phones by Pseudo Signature Using CNN 961

Fig. 6 Validation

signature drawing and if such data is there in the training set, it will affect the overall
accuracy.

5 Conclusion

Our study proposed pseudo signature, signature drawn using the finger on touch
screen as an authentication method in place of PIN and Graphical Pattern. Simula-
tion results shows that suggestion has a potential to replace the existing entry point
mechanism. From the users point of view, the signature bio metric is easy to remem-
ber and its resistant to forgery and stealing. Our study has achieved a remarkable
accuracy. The shortcomings of the study are as follows. The entire work carried out
on a limited data set. The time variance are not taken into consideration. As a future
work, we would consider the signature variance of the genuine user over time and
the model has to adapt the changes so as to make necessary modifications to the
learned model. Further, if we can add the accelerometer readings generated during
the signature drawing to the data set accuracy can be further increased.

Compliance with Ethical Standards All studies were conducted in accordance with principles
as defined in the Declaration of Helsinki Conference and approved by the relevant departmental
review board. Informed consent was obtained from all participants included in the study.
962 C. J. Jose and M. S. Rajasree

References

1. Bhagavatula R, Ur B, Iacovino K, Kywe SM, Cranor LF, Savvides M (2015) Biometric authen-
tication on iPhone and android: usability, perceptions, and influences on adoption. USEC ’15:
Workshop on Usable Security, 8 Feb 2015, San Diego, CA. Proceedings. 1–10. Research Collec-
tion School of Information Systems. Available at: [Link]
3967. [Link]
2. Bonneau J (2012) The science of guessing: analyzing an anonymized corpus of 70 million
passwords. In: 2012 IEEE symposium on security and privacy. San Francisco, CA, pp 538–
552
3. Harbach M, Von Zezschwitz E, Fichtner A, De Luca A, Smith M (2014) It’s a hard lock life:
a field study of smartphone (un) locking behavior and risk perception. In: 10th symposium on
usable privacy and security (SOUPS 2014), pp 213–230
4. Khuwaja G , Laghari M (2011) Offline handwritten signature recognition. World academy of
science, engineering and technology, open science index 59. Int J Comput Inf Eng 5(11):1304–
1307. [Link]
5. Chen J, Lopresti D, Ballard L, Monrose F (2013) Pseudo-signatures as a biometric. Int J IT
Eng Appl Sci Res (IJIEASR) 2(1), Jan 2013
6. Fischer A, Diaz M, Plamondon R, Ferrer MA (2015) Robust score normalization for DTW-
based on-line signature verification. In: Proceedings of the international conference on docu-
ment analysis and recognition (ICDAR). Nancy, France, 23–26 Aug 2015, pp 241–245
7. Wijesoma WS, Mingming M, Yue KW (2001) On-line signature verification using a compu-
tational intelligence approach. In: Proceedings of the fuzzy days. Dortmund, Germany, Oct
2001, pp 699–711
8. Fierrez Aguilar J, Nanni L, Lopez-Pealba J, Ortega Garcia J, Maltoni D (2005) An on-line
signature verification system based on fusion of local and global information. In: Proceedings
of audio- and video-based biometric person authentication (AVBPA). Rye Brook, NY, USA,
20–22 July 2005, pp 523–532
9. Feng H, Wah CC (2003) Online signature verification using a new extreme points warping
technique. Pattern Recognit Lett 24:2943–2951
10. Iranmanesh V, Ahmad SMS, Adnan WAW, Malallah FL, Yussof S (2014) Online signature
verification using neural network and Pearson correlation features. In: Proceedings of the
IEEE conference on open systems (ICOS). Subang, Selangor, Malaysia, 26–28 Oct 2014, pp
18–21
11. Nam S, Park H, Seo C, Choi D (2018) Forged signature distinction using convolutional neural
network for feature extraction. Appl Sci 8:153. [Link]
12. Fierrez J, Ortega Garcia J, Ramos D, Gonzalez-Rodriguez J (2007) HMM-based on-line sig-
nature verification: feature extraction and signature modeling. Pattern Recognit Lett 28:2325–
2334
Autonomous Mobile Robot Using RGB-D
SLAM and Robot Operating System

Jash Mota, Prutha Edwankar, Yash Shetye, Manisha Sampat

and Dattatray Sawant

Abstract Generally, the real-world environment is dynamic in nature. The transi-

tory and individuals items seem stationary for some time; however, they are later
moved, for example, chairs. A robot must somehow manoeuvre through the moving
objects for which it uses the SLAM algorithm. This paper explores Gmapping and
HectorSLAM on autonomous mobile robots for indoor applications using Microsoft
Kinect Robot Operating System (ROS).

Keywords Autonomous mobile robot · SLAM · RGB-D · Kinect · ROS ·

Navigation · Differential drive

1 Introduction

This paper explores autonomous navigation—one of the most challenging tasks in

robotics. Autonomous mobile robots are a category of robots capable of navigating
in a dynamic environment to the goal without any information from the user other
than the goal point. The autonomous navigation comprises of knowing where we
are in the dynamic environment, what does the environment look like at a particular
instant, and what is the transformation between the sensor and the wheelbase. A
two-wheeled differential drive with circular base has been chosen for the purpose of
this paper. Stepper motors provide the odometry data, while Kinect provides a laser
scan of the environment. All computation and communication happen through ROS.

J. Mota · P. Edwankar · Y. Shetye · M. Sampat · D. Sawant (B)

Department of Mechatronics Engineering, MPSTME, NMIMS University, Mumbai, India
e-mail: [Link]@[Link]
J. Mota
e-mail: jashmota720@[Link]

© Springer Nature Singapore Pte Ltd. 2020 963

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
964 J. Mota et al.

2 Slam

Simultaneous Localisation and Mapping is a problem to build an unknown envi-

ronment’s map and to determine its location in that environment after the mobile
robot is kept at an initially unknown position and orientation in the environment
[1]. The SLAM problem was first posed in 1986 at IEEE Robotics and Automation
Conference [2]. Eventually, it was realised that when the problem of mapping and
localisation is combined as a single estimation problem, the solution is more accu-
rate. SLAM estimates the location of the robot and the landmarks and plans a path
online, without any previous knowledge or a prebuilt map. There are many differ-
ent solutions to the problem of SLAM, with probabilistic SLAM, Extended Kalman
Filter SLAM (EKF SLAM), GraphSLAM and FastSLAM being the most prominent
ones [3]. The computation required to solve the problem increases quadratically with
an increase in the number of landmarks.

3 Localisation

The pose of the robot is necessary to plan a path to reach the navigation goal. Localisa-
tion can be achieved either through an inertial measurement unit (IMU) or by getting
feedback of the velocity of each motor [4]. An inertial measurement unit generally
comprises of an accelerometer, gyroscope and a magnetometer to calculate linear and
angular accelerations in three-dimensional space, which can be integrated to give the
value for orientation and position of the robot in space. Meanwhile, the rotation of
the wheels can also be measured using sensors like wheel encoders and can be used
to calculate linear velocities of each wheel [5]. Depending on the drive mechanism
of the robot, proper inverse kinematics can be computed to give the positional and
orientation values. For this paper, the robot has a differential drive with two wheels.
Also, stepper motors are used for driving the wheels. The advantage of using stepper
motors is the accurate open-loop control of the rotation of the motor. This allows to
precisely measure the rotation of the wheels, which will yield the angular velocity
over time, which would yield linear velocity for that particular wheel. Instantaneous
centre of curvature can be applied to find out the new pose of the robot relative to
previous pose (Fig. 1).
Mathematical model:

ω = θ × δt
 
L
vr = ω × R −
2
 
L
vl = ω × R +
2
Autonomous Mobile Robot Using RGB-D SLAM and Robot Operating … 965

Fig. 1 Computing pose for a differential drive

Which yields

L vl + vr
R= ×
2 vr − vl

And,

vr − vl
ω=
L
Therefore,

ICC = [x − R sin θ, y + R cos θ ]

       
x cos ωδt − sin ωδt x − ICCx ICCx
= × +
y sin ωδt cos ωδt y − ICC y ICC y

θ  = ωδt + θ

Robot’s position and orientation (pose) can be calculated using the above
equations.
966 J. Mota et al.

4 Mapping

In order to manoeuvre, the robot must know the map of its environment. In a dynamic
environment case, the map is constantly being updated and fed to the robot. There are
numerous ways to map any room. While LIDAR is a very accurate laser scanning
method, the sensors are usually expensive. Instead, for the purpose of this paper,
a Microsoft Kinect was used to get the landmark estimates. It incorporates several
advanced sensing hardware. Kinect contains a depth sensor, an RGB camera, micro-
phone array and an IR emitter. The IR emitter emits infrared light, the intensity of
which at different points is captured through an IR camera. The pattern captured is
correlated against a reference pattern. This image is matched with the image obtained
from the RGB camera image to obtain the distance of each point. The Kinect captures
307,200 depth points at a rate of 30 fps [6]. The data is stored as a point cloud, which
is a set of data points in a given coordinate system. Binding these point clouds from
various frames gives a 3D map of the environment.

5 System Setup

For our experiments, we built upon the TurtleBot 2 hardware, a set of sensors,
equipped with a computing system, microcontroller to communicate between the
wheel motors and the computer running Robot Operating System (ROS). ROS is
used for modelling, simulation and visualisation of output from laser scan topics,
along with providing support for distributed computing and providing packages for
navigation [7]. The computational platform is based on Intel Core i5, Radeon graph-
ics, which supports CUDA. Table 1 provides all the system specifications ROS phi-
losophy has the goal of being a peer to peer, tools based, multilingual, lightweight,
free and open-source middleware to provide a structured communication layer above
the host OS, Ubuntu 16.04 in this case, or a network of computers. Communication
happens between different nodes through topics where the data is published and
subscribed as messages. A node can publish a message on a topic to several other
nodes and subscribe to several other nodes too. This way, the whole procedure can
be broken down into modular nodes (Fig. 2).

Table 1 Robot’s system

Parameters Configurations
configuration
Processor Intel Core i5
GPU Radeon
RAM 12 GB
RGB-D Microsoft Kinect
OS Ubuntu 16.04 LTS
ROS Kinetic Kame
Autonomous Mobile Robot Using RGB-D SLAM and Robot Operating … 967

Fig. 2 System setup

A differential drive was used with stepper motors to provide the odometry data.
A circular base was chosen to allow the robot to pass through a given cross section
in any dimension in extreme cases, which is not possible in square bases.

6 Experimental Environment

The robot was initialised in a completely unknown environment with no prior knowl-
edge. Data from Kinect was published on laser scan topic on ROS. Value from IMU or
the wheel revolutions is published in the topic odom [8, 9]. Along with the informa-
tion about odometry and laser scan, a transforming message is necessary to transform
the point cloud value from the Kinect to the frame, in this case, the mobile base frame.
Gmapping [10] is a laser-based SLAM algorithm [11–13] and a widely used
SLAM package in the ROS community. Alternative to Gmapping is HectorSLAM,
which combines a 2D SLAM system [14] and 3D navigation [15], using an inertial
measurement system. Gmapping is recommended by Santos et al.
Gmapping takes input from /laserscan, /odom and /tf and localises the robot and
plans a path to reach the goal. Gmapping creates a local occupancy grid which is
a grid of estimates of the position of the robot. The estimate improves with more
landmarks and with an increase in runtime.
968 J. Mota et al.

7 Conclusion

This paper demonstrates how a robot can be built or configured to navigate

autonomously with the help of a laser scanner and an IMU. The paper also discusses
two prominent SLAM algorithms and how ROS helps incorporate communication
between various nodes, enforce distributed computing and working with navigation
stack of ROS. Further, visual tools like Gazebo and MoveIt! can also be explored
for simulation of the project. The experimental prototype setup demonstrated in
this paper can be used in homes to carry out routine tasks like cleaning homes or
delivering items from one room to another.

Acknowledgements The authors thankful to the NMIMS Deemed to be University’s Mukesh Patel
School of Technology Management and Engineering, Mumbai, India for providing the necessary
facilities for carrying out this work.

References

1. Santos JM, Portugal D, Rocha RP (2013) An evaluation of 2D SLAM techniques available

in robot operating system. In: IEEE international symposium on safety, security and rescue
robotics (SSRR), pp 1–6
2. Durrant-Whyte H, Bailey T (2006) Part I Simultaneous localization and mapping. IEEE Robot
Autom Mag 13(2):99–110
3. Bailey T, Durrant-Whyte H (2006) Part II Simultaneous localization and mapping (SLAM).
IEEE Robot Autom Mag 13(3):108–117
4. Quigley M, Conley K, Gerkey B, Faust J, Foote T, Leibs J, Wheeler R, Ng AY (2009) ROS: an
open-source robot operating system. In: ICRA workshop on open source software, vol 3, no 3,
p5
5. Ibragimov IZ, Afanasyev IM (2017) Comparison of ROS-based visual SLAM methods in
homogeneous indoor environment. In: 14th IEEE workshop on positioning, navigation and
communications (WPNC), pp 1–6
6. Thrun S, Montemerlo M (2006) The graph SLAM algorithm with applications to large-scale
mapping of urban structures. Int J Robot Res 403–429
7. Sturm J, Engelhard N, Endres F, Burgard W, Cremers D (2012) A benchmark for the evaluation
of RGB-D SLAM systems. In: IEEE international conference on intelligent robots and systems,
pp 573–580
8. Kerl C, Sturm J, Cremers D (2013) Robust odometry estimation for RGB-D cameras. In: IEEE
international conference on robotics and automation, pp 3748–3754
9. Dissanayake MG, Newman P, Clark S, Durrant-Whyte HF, Csorba M (2001) A solution to
the simultaneous localization and map building (SLAM) problem. IEEE Trans Robot Autom
17(3):229–241
10. [Link]
11. Endres F, Hess J, Engelhard N Sturm J, Cremers, D, Burgard W (2012) An evaluation of the
RGB-D SLAM system. In: ICRA, pp 1691–1696
12. Bailey T, Nieto J, Guivant J Stevens M, Nebot E (2006) Consistency of the EKF-SLAM
algorithm. In: IEEE international conference on intelligent robots and systems, pp 3562–3568
Autonomous Mobile Robot Using RGB-D SLAM and Robot Operating … 969

13. Leonard JJ, Durrant-Whyte HF, Cox IJ (1992) Dynamic map building for an autonomous
mobile robot. Int J Robot Res 11(4):286–298
14. [Link]
15. [Link]
Survey on Data Mining and Predictive
Analytics Techniques

S. Sathishkumar, R. Devi Priya and K. Karthika

Abstract Nowadays, predictive analytics is one of the most important big data
trends. Predictive analytics is the accumulation of extensive, mostly unstructured
data from various sources. The mixture of various information sources, for example,
online networking information, climate and traffic are improved by internal informa-
tion is especially basic. But both predictive analysis and data mining attempt to make
divination about possible events in the future with the help of data models. Predic-
tive analytics processes utilize various statistical strategies such as machine learning
or neural networks, regression and extrapolation to perceive in the information pat-
terns and infer algorithm. These algorithmic procedures are assessed depending on
test data and optimized data. It is to be noted that as data availability increases, the
accuracy of the algorithm also improved. By chance if the improvement procedure
is finished, the algorithm and the model can be connected to information whose
classification is obscure. Predictive analytics model captures connection between
various factors to assess chance with a specific set of conditions to distribute a score
or weightage. Successfully, on applying predictive examination, the organizations
can adequately explain huge information for their benefit. We present a detailed sur-
vey on data mining and predictive analytics here, by analyzing 15 techniques from
standard publishers (IEEE, Elsevier, Springer, etc.) of the year from 2008 to 2018.
Based on the algorithms and methods utilized which are inconvenient, the problems
are analyzed and classified. Moreover, to indicate the improvement and accuracy of
all the research articles is also discussed. Furthermore, the analysis is carried to find

S. Sathishkumar (B) · K. Karthika

Department of Information Technology, Adhiyamaan College of Engineering, Hosur, India
e-mail: ssathishkumaraeri@[Link]
R. Devi Priya
Department of Information Technology, Kongu Engineering College, Erode, India

© Springer Nature Singapore Pte Ltd. 2020 971

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
972 S. Sathishkumar et al.

the essential for their approaches so that we can develop a new technique to previse
the future data. Eventually, some of the research issues are also inscribed to precede
further research on the similar direction.

Keywords Data mining · Predictive analytics · Model modules · Parameter

estimation

1 Introduction

In present days, data managing becomes an essential task of many fields because
of the evolution of information technology leads to enormous amount of data. The
information technology provides a lot of information, but extracting only the essential
information becomes a difficult task; therefore, many techniques have been proposed
to deal with this issue. Among them data mining, data prediction, predictive analyt-
ics, etc. become an effective technique in current days. Information is expected to
expand as technology enhances the organization units inside the operations and it
gets connected in stable. Firms that recover and treat information as a vital resource
can make an incentive through predictive analysis [1]. This present work is focused
on predictive analytics which the brief introduction about the predictive analytics.
Analytics is an essential process of every field because it is utilized for discovering,
analyzing and understanding meaningful patterns from widespread data. The sig-
nificance of predictive analytics is inevitable in many fields because this technique
helps to predict the future about the particular field information [2]. With the help
of predictive analytics, the information about the future business performance can
be elucidated; therefore, the organizations utilized predictive analytics technique to
understand the complete growth, to improve the current scenario, to learn about the
development possibilities, etc. [3]. Besides, it is utilized for identifying trends, rela-
tionships and patterns within data that can be used to monitor the future event and
behavior [4, 5]. The general predictive methods often utilized by researchers are:
regression modeling, decision tree, Bayesian statistics, neural network, support vec-
tor machine and nearest neighbor algorithm [6]. In order to prove, we have studied
more about predictive analytics here, and the survey work is conducted based on
various papers. In addition, this work is categorized by three parts such as technique,
application and parameters measures.
Survey on Data Mining and Predictive Analytics Techniques 973

2 Survey on Data Mining and Predictive Analytics

Techniques

The survey is done by means of learning information from different research papers
from 2010 to 2018 in the standard journals such as IEEE, Elsevier, Springer and
miscellaneous international journals. Here, the survey is done by means of different
techniques and its categorizations.

2.1 KNN-Based Data Mining and Predictive Analytics

Bendre and Manthalkar [7] have proposed a novel technique for prediction of future
conditions on climatic station from big data by addressing the predictive approach
on the basis of time series and neural network utilized by MapReduce programming
model. They have included predictive analysis approach with a broad spectrum using
the models, such as examination and decay, arrangement and forecast. The time
arrangement-based decay approach was proposed to disintegrate and discover the
pattern, normal and modern components. The direct components were dealt with
time arrangement MapReduce-based autoregressive integrated moving average (M-
ARIMA) model and nonlinear segments were taken care of by MK-nearest neighbors
(M-KNN) display.
Additionally the MapReduce-based hybrid model (M-HM) is suggested to utilize
the benefits of time arrangement and neural network to enhance precision of divina-
tion. This analytics confirm the viability of developed model over the standard and
irregularity part of the information. The performance estimation and measurements
are performed to affirm the consistency of checked information. In addition, brilliant
accelerate, scale up and scrutinize are tried by varying the measure of information
collection. But when the information measure was expanded, the normal execu-
tion time was decreased by utilizing the MapReduce-based methodology over the
numerous node specialists.

2.2 Random Forest-Based Predictive Analytics

Wang et al. [8] have proposed a novel and effective predictive analytics. Business
Intelligence & Analytics (BI&A) had turned into an essential region for the spe-
cialists and experts. The ordinary business knowledge featured unmistakable and
symptomatic investigation to accomplish execution of estimation and the executives.
Besides, business analytics delayed for incorporating predictive and prescriptive
analysis to produce responsive activity plan. In the zone of BI&A, the accompanying
problems are basic yet hard to handle, particularly to legitimize the reliability of the
974 S. Sathishkumar et al.

proposed work, three sorts of personal computer firms included which are business
PC, modern PC and creation PC and were correspondingly utilized to describe differ-
ent plans of action in PC enterprises: Original Brand Manufacturing (OBM), Original
Design Manufacturing (ODM) and Electronic Manufacturing Services (EMS).
Appel et al. [9] have demonstrated the feasibility of utilizing methods from
machine learning and information mining to decide the upcoming opportunity and
dangers of individual properties and for neighborhoods utilizing an assortment of
basic, statistic, financial and city movement highlights with high accuracy. A larger
system of frameworks has been proposed which allows a city to move from decision-
making based on ‘educated anecdotes’. It also provided a systems solution that allows
the City of Syracuse to begin transforming to a proactive and preventative model using
analytics based on mathematical models and available data in the housing ecosystem.

2.3 ANN-Based Data Mining and Predictive Analytics

Lorenzo et al. [10] have suggested a novel technique based on ANN to investi-
gate the benefit of using a commercially accessible cloud-based machine learning
stage for examination of careful intercession in babies with pre-birth hydronephrosis
(HN). Probabilistic key component examination was utilized for information attri-
bution. Different clinical factors were incorporated into two-class choice wilderness
and neural system for model preparing, utilizing careful mediation as the essential
result. The predictive models were set out as a web administration in 35 s creating
an extraordinary API key for application and web page advancement. Individual
prediction dependent on the factors was conveyed as web based and batch execu-
tion record in 1 min. This cloud-based ML innovation permits simple construction,
arrangement and sharing of predictive investigation solution. Utilizing pre-birth HN,
for instance we propose a chance to address contemporary difficulties with informa-
tion investigation, detailing an inventive arrangement that moves beyond the cutting
edge standard.
Talaei-Khoei and Wilson [11] have proposed a novel technique to distinguish the
patients in danger of type 2 diabetes (T2D). There was a collection of writing that
utilized machine learning characterization calculations to anticipate advancement of
T2D among patients. The present investigation corresponds to the execution of the
characterization calculations to distinguish patients who were in danger of creating
T2D to put it plainly, medium and long terms. To create a balanced dataset syntactic
minority oversampling and random under sampling were used. The performance of
neural networks, support vector machine to identify the patients developing type 2
diabetes (T2D) in short, medium and long terms are compared. Through significant
investigation and data combination methods, the indicators of creating T2D were
recognized for short, medium and long haul hazard examination. The discoveries
showed that the execution of examination procedures relies on period and nature of
Survey on Data Mining and Predictive Analytics Techniques 975

expectation whether the prediction is to recognize individuals who will not advance
T2D or to decide in danger patients.
Lokhandwala and Nateghi [12] have proposed a new methodology where they
have given importance to the statistical and machine learning calculations to recog-
nize the key indicators of cooling request power, utilizing the EIA CBECS informa-
tion. They compared the utilization levels over the two years to seize how cooling
burden had developed. They saw that while the mean and middle estimations of
interest force have expanded sensibly since the mid-2000s, there had been a very
sharp expansion in greatest use power over that period. At an expansive scale level,
their information-driven investigation recognized atmosphere and building type as
the most essential indicators of interest in the years. Nourishment administrations
and in-persistent medicinal services units were famous as the most vitality serious
building types, with higher power.

2.4 Decision Tree-Based Predictive Analytics

Geological uncertainty indicates a natural risk for all mining ventures. Mining tasks
used asset square models as a noteworthy wellspring of information in arranging and
in basic leadership. Be that as it may, such operational choices were not free from
hazard and vulnerability. The idea moves on how vulnerability was dealt with and
an ability to enable in zones that make operational adaptability could direct potential
misfortunes. Information investigation was promoted as one of the significant intru-
sions in the twenty-first century and tasks that legitimately used information that
could make constant open doors in the time of a dubious future. Since associations
had bounty measures of unmistakable land information, a mix of information mining
and genuine alternatives could give an upper hand.
Ajak et al. [1] have suggested a new tool in this present work where predictive
data mining methods were connected to a real-time mining task to examine the likeli-
hood of experiencing dangerous metal in a mining program. The information mining
models yield is utilized to make conceivable genuine choices that the activities could
meet to manage dirt vulnerability. The most widely recognized information-digging
calculation picked for this errand was the characterization tree, which anticipated
the likelihood of dirt with 78.6% precision. Poisson dissemination and Monte Carlo
reproductions were connected to inspect different genuine alternatives. The ebb and
flow look into uncovered that activities could limit unprogrammed misfortunes in
the handling plant and could yield an upgraded esteem, if the prescient information
mining calculation was connected to make genuine alternatives.
976 S. Sathishkumar et al.

2.5 Naïve Bayesian Classifier-Based Predictive Analytics

Bellazzi and Zupan [13] have suggested a novel concept. The outstanding accessibil-
ity of new calculation strategies and tools for analyzing data and predictive models
requires restorative data scientists and professionals to methodically chosen the most
appropriate methodology to deal with clinical expectation issues. Especially, Pool
of methods are known as ‘Data Mining’ provides various solutions to deal with the
analysis of medical data and construction of various prediction models. A substan-
tial assortment of those strategies needs broad and basic rules that may help experts
in the reasonable determination of information mining apparatuses, creation and
approval of prescient models, alongside the engendering of prescient models inside
clinical situations. The objective of this friend audit was to talk about the degree
and job of the exploration region of prescient information mining and to develop a
structure to make do with the issues of making, evaluating and using information
mining model in clinical drug. It includes the analysis of clinical data warehouses
for clinical medicine, epidemiological studies and emerging studies in genomics and
proteomics.
Siryani et al. [14] have presented a framework for a decision-support system
(DSS) that operated within the IoT ecosystem. An ‘Internet of Things’ (IoT) that
creates a bridge between the society and frameworks speaks to an enormous idea
move. The DSS influence the advanced analytics of electric smart meter (ESM)
network communication in order to improve predictions of cost for smart meter
field operations and also provide recommendations regarding whether to send a
expert to a client location to deal with an ESM issue. Based on observation the
model was evaluated using data set from a profitable network. The framework of
decision support system is express with Bayesian Network prediction model and also
it is compared with three machine learning classifiers: Naïve Bayes, Random Forest
and Decision Tree. Results demonstrated that their methodology produced factually
amazing estimations and that the DSS would improve the cost-effectiveness of ESM
arrange tasks and support.

2.6 K-Means-Based Predictive Analytics

Ge et al. [15] have found a new technique where they discussed about data min-
ing. Data mining and analytics have assumed an essential job in learning revelation
and decision-making/bolstered in the process industry over the previous hundreds of
years. As a computational engine to data mining and investigation, machine learning
gave us essential instruments to data extraction, information design acknowledgment
and forecasts. From the information of machine learning, this exploration work gave
a survey on existing information mining and examination applications in the process
business over the previous hundreds of years. The best in class of information min-
ing and investigation was reassessed through eight unsupervised learning and ten
administered learning calculations, just as the application status of semi-regulated
learning algorithms.
Survey on Data Mining and Predictive Analytics Techniques 977

2.7 Summary of Survey

In this section, the data about the predictive analytics is grouped based on various
peer-reviewed papers from eminent journals includes IEEE, Elsevier Springer, etc. In
these papers, the efficiency of various techniques can be understood. In accordance
with that, the research has been conducted by different techniques that are utilized.
In the forthcoming section, the survey work based on different categorizations will
be discussed.

3 Categorization and Its Discussion

All the articles taken for the survey are categorized based on three criteria: technique-,
application- and parameter-based measures.

3.1 Categorization Based on the Novel Techniques

In this arena, we categorize the existing research work in accordance with the novel
techniques. This analysis meant for usage of algorithm in the olden days and realizes
how to improve the algorithm further for current research work. Each research work
is developed for different years. So we split the research work into four categories
such as 2008–2011, 2012–2013, 2014–2015 and 2016–2018. The number of research
work utilized in each year is explained in Table 1 for this survey, which relies totally
on 16 papers. The different algorithms used in various paper has been shown in Fig. 1.

Table 1 Overall analysis of survey 2008–2018

Technique 2008–2011 2012–2013 2014–2015 2016–2018
ANN – – – [7, 10, 11, 12]
KNN – – – [7]
SVM – – – [11]
K means – – – [15, 16]
Random forest – [9] – [8, 12]
Decision tree – – – [8]
Naïve Bayesian [13] – – [10]
PCA – – – [13, 15]
Nearest neighbor – – [14] –
Other techniques [15] – [14] –
978 S. Sathishkumar et al.

ANN
2016-2018
KNN
K Means

2014-2015 Random Forest

DecisƟon Tree
Naïve Bayesian
2012-2013
PCA
Nearest Neighbor

2008-2011 Other Techniques

Fig. 1 Summary of the survey

3.2 Categorization Based on Application

In this section, we tabulate the applications based on different papers such as EMS,
parallel and distributed processing, medical, etc. (Table 2).

3.3 Categorization Based on Parameters Measures

This section explains about the parameters measures utilized based on various papers
from 2008 to 2018 such as t-value, profit, R-square, etc. Table 3 shows the various
parameter measures applied in the literature survey.
This survey is executed for proposing an effective technique in future work with
advanced features. This work has shown about the techniques which have been
utilized in existing technique and gives us an idea about existing problems in it. With
the help of this survey, we can provide an effective predictive analytic technique in
our future work.
Table 2 Categorization based on application from 2008 to 2015
Article from 2008 to 2011
References EMS Parallel and Medical Energy Data Property vacancy Automated Computer
distributed information classification policies for cities library graphics
processing administration information
(EIA) exchanging
network
[7] ✓
[8] ✓
[9] ✓
[16] ✓
Survey on Data Mining and Predictive Analytics Techniques

[10] ✓ ✓
[11] ✓ ✓
[12] ✓
[13] ✓ ✓
[15] ✓
979
980 S. Sathishkumar et al.

Table 3 Categorization based on parameter measure

Measures 2008–2011 2012–2013 2014–2015 2016–2018
t-value – – – [7]
Profit – – – [8]
R-square – [9] – –
Average run time – – – [10]
Mean – – – [12]
Other [13] [1, 15]

4 Conclusion

Predictive analytics models help to create the connections among numerous compo-
nents to evaluate chance with a specific arrangement of conditions to allot a score or
weightage. By successfully applying predictive analytics, the businesses can effec-
tively interpret big data for their benefits. In this research work, we present a detailed
survey on data mining and predictive analytics by analyzing various techniques from
standard publishers (IEEE, Elsevier, Springer, etc.) during the year from 2010 to
2018. This survey work gives good vision about current techniques utilized for pre-
dictive analytics. This analysis helps to nurture future work in terms of overcom-
ing the current issues. Therefore, the effective predictive analytics can be proposed
significantly with effective features in future work.

Future Work
From the above survey work, we came to know the existing predictive analytic tech-
niques have some limitations such as big data handling, distributed computing, pre-
dictive analytics in cloud, etc. and therefore, the efficiency of the proposed technique
needs to be improved for future work. In addition, the effective advanced algorithm
should be utilized for improving the performance by proposing-effective-technique.

References

1. Ajak AD, Lilford E, Topal E (2018) Application of predictive data mining to create mine plan
flexibility in the face of geological uncertainty. Resours Policy 55:62–79
2. Alharthi H (2018) Healthcare predictive analytics: an overview with a focus on Saudi Arabia.
J Infect Public Health (in press, corrected proof). Available online 8 Mar 2018
3. Rajni J, Malaya DB (2015) Predictive analytics in a higher education context. IT Prof 17(4):24–
33
4. Bekiroglu K, Duru O, Gulay E, Su R, Lagoa C (2018) Predictive analytics of crude oil prices
by utilizing the intelligent model search engine. Appl Energy 228:2387–2397
5. Rousseaux F (2017) BIG DATA and data-driven intelligent predictive algorithms to support
creativity. Ind Eng Comput Ind Eng 112:459–465
Survey on Data Mining and Predictive Analytics Techniques 981

6. Dubey R, Gunasekaran A, Childe SJ, Papadopoulos T, Roubaud D (2017) Can big data and
predictive analytics improve social and environmental sustainability? Technol Forecast Soc
Change (in press, corrected proof). Available online 15 July 2017
7. Bendre M, Manthalkar R (2018) Time series decomposition and predictive analytics using
MapReduce framework. Expert Syst Appl (in press, accepted manuscript). Available online 8
Sept 2018
8. Wang C-H, Cheng H-Y, Deng Y-T (2018) Using Bayesian belief network and time-series
model to conduct prescriptive and predictive analytics for computer industries. Comput Ind
Eng 115:486–494
9. Appel SU, Botti D, Jamison J, Plant L, Varshney LR (2014) Predictive analytics can facilitate
proactive property vacancy policies for cities. Technol Forecast Soc Chang 89:161–173
10. Lorenzo AJ, Rickard M, Braga LH, Guo Y, Oliveria J-P (2018) Predictive analytics and mod-
eling employing machine learning technology: the next step in data sharing, analysis and indi-
vidualized counseling explored with a large, prospective prenatal hydronephrosis database.
Urology (in press, accepted manuscript). Available online 30 June 2018
11. Talaei-Khoei A, Wilson JM (2018) Identifying people at risk of developing type 2 diabetes: a
comparison of predictive analytics techniques and predictor variables. Int J Med Inf 119:22–38
12. Lokhandwala M, Nateghi R (2018) Leveraging advanced predictive analytics to assess
commercial cooling load in the U.S. Sustain Prod Consum 14:66–81
13. Bellazzi R, Zupan B (2008) Predictive data mining in clinical medicine: current issues and
guidelines. Int J Med Inf 77(2):81–97
14. Siryani J, Tanju B, Eveleigh TJ (2017) A machine learning decision-support system improves
the internet of things’ smart meter operations 4(4):1056–1066
15. Ge Z, Song Z, Ding SX, Huang B (2017) Data mining and analytics in the process industry:
the role of machine learning
16. Litsey R, Mauldin W (2018) Knowing what the patron wants: using predictive analytics to
transform library decision making. J Acad Librarianship 44(1):140–144
Implementation of In-flight
Entertainment System Using Light
Fidelity Technology

Viraj Savaliya, Kalp Shah, Dharmil Shah, Henish Shah and Poonam Kadam

Abstract We are often directed to switch off our mobile phones and other elec-
tronic devices while traveling in a plane due to the interference caused by the radio
transmitters installed in these devices. As a result, most customers do not enjoy the
benefit of data on-board except for some airlines who provide the paid data services.
Also, due to the recent outburst in the demand for data, there is soon going to be
a congestion of the radio frequency bands. Thus, developing a technology which
can entirely replace the current system of radio frequency-based data transmission
is beneficiary [1]. Light fidelity (Li-Fi) is a next-generation technology which uses
visible light as a medium for transmitting data [2]. Thus, it serves this increasing
demand for data due to its vast bandwidth and does not interfere with airplanes as
well. This paper aims on demonstrating the application of light fidelity (Li-Fi) in the
entertainment services provided inside the airplanes. All the fundamental principle
of visible light communication (VLC), the modulation techniques involved as well
as how a circuit can be realized for transmitting data in an airplane using those prin-
ciples have been discussed in this paper. Light fidelity (Li-Fi) when used to its full
potential can certainly prove that everyone can count on it when RF fails to serve the
purpose.

Keywords Light fidelity · Visible light communication · Wireless fidelity · Pulse

width modulation

V. Savaliya (B) · K. Shah · D. Shah · H. Shah · P. Kadam

Department of Electronics and Telecommunication, Dwarkadas J. Sanghvi College of
Engineering, Mumbai, India
e-mail: virajlee58@[Link]
K. Shah
e-mail: kshah.shah951@[Link]
D. Shah
e-mail: dharmil1997@[Link]
H. Shah
e-mail: henishshah18@[Link]
P. Kadam
e-mail: [Link]@[Link]
© Springer Nature Singapore Pte Ltd. 2020 983
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
984 V. Savaliya et al.

1 Introduction

The current in-flight entertainment system consists of an electronic tablet which

is fixed to backside of the seat and offers multiple sources of entertainment like
movies, TV series, music, etc. This tablet might be connected to a Wi-Fi or it may
have data stored inside it. Wi-Fi systems on-board are very expensive and the users
often have to pay an extra amount of money to avail such services. Also, installation
and maintenance cost of such electronic equipment on each seat leads to expensive
flight tickets. As the screen is attached on the backside of a seat, the user has to adjust
the viewing angle each and every time the person on the front seat lies down. Most of
the airplane companies use Wi-Fi services, and thus it further leads to the problem of
congestion of radio frequency bands [3]. To avoid such drawbacks and difficulties,
we can use an alternative wireless technology called Li-Fi.
Li-Fi technology has been proposed to overcome the limitations of finite band-
width of radio frequency. Li-Fi stands for ‘light fidelity’ which uses visible light for
data transmission. Visible light has a wider spectrum range of 375–775 nm [4]. Vis-
ible light communication (VLC) uses ordinary lamps or LEDs to transfer data from
one place to another. Light from these sources does not harm the human eye vision
and is immune to radio frequency interference [5]. In places, where Wi-Fi is not
allowed due to security purpose, Li-Fi makes it possible to have a wireless Internet
in specific environments. Also, as Li-Fi does not include the use of routers or optical
fiber cables, it is a cheaper alternative to Wi-Fi. The idea behind this project is to
use the ON–OFF activity of LEDs to transfer data [6]. LEDs have energy-efficient
illumination and high-modulation bandwidth. The devices flicker at a very high rate
which makes it possible to send a stream of bits at a very high speed. The trans-
mission is not visible to the naked eye because of the high-switching speeds. Binary
bits are used for communication, i.e., when the LED is on, a Binary 1 is transmitted
and when the LED is off, a binary 0 is transmitted [7]. This concept of Li-Fi was
introduced in a TED talk by Dr. Harald Haas who is a professor at University of
Edinburgh [8]. The data transfer between the transmitter and receiver is always done
in a line of sight manner. With the help of Li-Fi, it is possible to achieve speed higher
than 10 Gbps [9].
Since there is a light source above every seat on an airplane, we make use of that
to transmit data to the users (Fig. 1). Thus, we can save a huge amount of money
on the installation part of the system. The users can avail the services directly on
their mobile phones via a receptor module and there would be no need of any other
electronic equipment. With the help of Li-Fi modules, entertainment services can be
availed at a much cheaper rate and will not lead to expensive flight tickets. In this
paper, we present a methodology that makes use of light bulb to transmit data to the
airplane travelers (Fig. 2).
Implementation of In-flight Entertainment System … 985

Fig. 1 Li-Fi technology in aircraft using led bulbs

Fig. 2 Light spectrum for different frequencies [10]

2 Methodology

In the proposed Li-Fi data transmission system, we transmit data stream with the
help of Arduino board and LED(s), and receive the binary bits using a sensor (solar
panel) and an Arduino board which is at the receiver.
At the transmitter side, the Arduino’s code converts the desired data stream to be
transmitted into its corresponding binary values. This conversion to binary is done by
an in-built function in the Arduino’s library, and the binary data stream is modulated
using PWM and then sent to the transmitting device [11]. The bits are transmitted
using an LED or array of LEDs (LED bulb), and Logical 1 is transmitted as LED
‘ON’, whereas logical 0 is transmitted as LED ‘OFF’. Line of sight (LOS) is essential
for correct reception of bits at the receiver site [12].
986 V. Savaliya et al.

At the receiver side, initially a trigger detection procedure is established for further
bit synchronization during communication. The bit synchronization happens when
the receiver detects a constant stream of light for 20 ms (a predetermined threshold
is used for this). After bit synchronization, the sensor will detect the binary bits
transmitted by the transmitter and the Arduino board reads the sensor’s output value.
The output values of the sensor are then compared with a predefined threshold for
finding the corresponding binary values for each sensor reading. An additional noise
removal procedure is done, wherein the first bit which is received is discarded and
the array is updated to account for the noise in the first bit. Post noise removal data
represents binary for each frame of the video. These updated arrays of binary values
are then de-converted to form each frame of the video. Each frame is then displayed
with desired frame rate to avoid the flicker effect (Figs. 3 and 4).
Fig. 3 Trasmitter flowchart
Implementation of In-flight Entertainment System … 987

Fig. 4 Receiver flowchart

3 Implementation

The entire system consists of a remote, a transmitter circuit, a receiver circuit, array
of LED’s, a database, and a mobile-based application. The remote is installed on
each passenger seats in the aircraft. The transmitter circuit will be installed with the
LED lights above each passenger seat. Below the passenger seat, there will be a small
device consisting of receiver circuitry and pin provided for connection with a mobile.
The application is supposed to be downloaded on passenger’s mobile in order to run
the received data. The database acts as a central node for each aircraft privately and
provides the storage for available entertainment options and gives access to all the
passengers seating inside the airplane.
The application developed allows the user to choose from the available movies on
the in-flight entertainment system. Once the user selects the movie to be streamed
using the on-seat remote, the data transmission process using Li-Fi system begins.
An Arduino UNO board is used at both-transmitter and receiver side for doing the
desired processing. For transmission of data bits, the array of LED’s is connected to
the Arduino-based transmitter circuit. The data to be transmitted is initially converted
to its binary form using a function called bitRead(). Each frame of the video is
988 V. Savaliya et al.

initially converted to a set of segments. These segments are less than 64kB in size
so as to prevent clogging of Arduino’s serial buffer. This array of binary data is then
transmitted using the LED. Bit ‘1’ is transmitted with the output as high, whereas bit
‘0’ is transmitted with the output as low; this will result in corresponding flickering
of LED for each bit. At the transmitter side, it is essential to ensure that the array of
bits to be transmitted do not clog the buffer of the Arduino board. To avoid buffer
overload, a function called flush() is used repetitively after a fixed duration of time
to clear the buffer. After transmission of all the segment data for current frame, the
segment data for next frame is then transmitted. This process continues till data for
all the frames has been transmitted, i.e., till the complete video data is transmitted
(Fig. 5).
For data reception, a solar panel is used, which is interfaced with the Arduino -
based receiver circuit. To ensure proper bit detection, a bit synchronization procedure
is done. In this process, the LED at transmitter side is kept ‘ON’ for 20 ms. This is
detected by the receiver, which then proceeds to read the output value of the solar
panel. The output is read using a function called as analogRead(). After reading
all the data, the output of the sensor is in the form of voltage, hence, to convert
the data into binary it is compared with a predetermined threshold. The threshold
has to be updated every time the environment around the Li-Fi system changes.
The binary data array then undergoes a noise removal procedure to account for the

Fig. 5 Application interface

Implementation of In-flight Entertainment System … 989

noisy first bit of the received array. This updated array of bits then undergoes a de-
conversion/recombination procedure to make up each segment of each frame of the
video. After frame recombination, each frame is then displayed on the application’s
interface at the receiver site, which generally is a mobile phone or a tablet (Fig. 6).
The central database or the central node is accessed each time the passenger
presses the button on the remote to select a movie. The central node is basically
connected with all the nodes or passenger seats. The button pressed from a particular
remote sends a command to access that particular numbered movie to the central
database via a wired connection. Then that particular data is activated to be sent on
the node present from where the command was received. The node circuitry (above
passenger seat), on-seat remote, central database, and the receiver device connected
with the mobile works all in synchronization. The data on the central storage system
can be updated regularly (Figs. 7 and 8).

Fig. 6 Flowchart for transfer of data

Fig. 7 Central to other nodes connection

990 V. Savaliya et al.

Fig. 8 Transmitter circuit used

4 Results

The proposed system for transfer of data for an entertainment system in airplanes
using the visible light frequency spectrum is successfully tested. The system transmits
the video data using a transmitter circuit controlled by an Arduino. The transmitter
circuit is programmed using a software code via the Arduino. The software code
is designed so as to perform the necessary coding and modulations as required by
the model. Also the code as well the additional circuitry of the transmitter sections
take care of any other modulation or scaling required. The data is then successfully
transmitted using an array of LED’s and hence using the visible light frequency
spectrum. The LED’s here blink in a prescribed manner according to the light signal.
The receiver solar panel accepts the lights rays falling on it and converts it into
electrical signal. Then the software code at the receiver end performs the decoding
and carries out demodulation to generate the original signal sent by the user. The
transmission of data is successfully done after taking into consideration all the error
signal present in the vicinity of the model. Also, the important observations made in
functioning of the proposed system are that works to its complete efficiency only for
a limited range and also the speed is limited. As the distance between the transmitter
and receiver side increases, there starts occurring more errors and false transmission
of bits. The bit error rate keeps on increasing with increase in the distance making the
system difficult to revive the original signal and hence making the model inefficient.
But with better and improved modulations, we can tackle that limitation.
Implementation of In-flight Entertainment System … 991

5 Conclusion and Future Scope

The proposed system to transfer video data successfully fulfills its objective and is
efficient to work in real-world environment. The system also takes care of the noises
present in the vicinity and produces the original signal accordingly. The system has
limitations on the range, speed, and capacity of data to be transmitted. All these
limitations can be worked upon by performing better modulations and using higher-
end processors having more power. But, in comparison with the currently in use
entertainment system in flights, the proposed systems outweigh it on many factors.
The cost of the systems reduces drastically as it uses the already installed lights
for transfer of data and also uses passenger’s personal electronic device to run the
data. Also, the proposed system is more comfortable for the passengers as they can
sit in a position which is relaxing and preferable by them to watch a movie rather
than sticking to a fixed installed device. Also, the angle at which the user wants
to watch can be adjusted. The maintenance cost of the system also decreases as
there is not much circuitry involved and it replaces the currently installed screens in
airplanes. Hence, the cost of travel for passengers also reduces along with charges.
Also, only those passengers who want to access the entertainment system can access
rather than providing it mandatorily to all. The screen size for running the data
will also be as preferred by the passenger as they are running it on their personal
devices. Hence, passengers will not have to stick to already provided device at their
seat. The mobile-based application can also be designed in a way to make it more
user-friendly and attractive by the aircraft travel companies which can also give
them a chance to enhance their marketing strategies. For extension of the proposed
system, more amounts of data can be provided by using a bigger and better database
and its management systems. Also, the transmitter and the receiver circuitry can
be made more advanced by making necessary modifications for higher speed and
data-carrying capacity.

References

1. Sonnad AM (2013) Led recent advancements in Li-Fi technology. Int J Electr Electron Data
Commun 1(10). ISSN: 2320-2084
2. Vinnarasi A, Aarthy ST (2017) Transmission of data, audio signal and text using LI-FI. Int J
Pure Appl Math 117
3. Proakis JG (2001) Digital communications, 4th edn. McGraw-Hill, New York
4. Abdallah W, Boudriga N (2016) Enabling 5G wireless access using Li-Fi technology: an OFDM
based approach. 978-1-5090-1467-5/16/ ©2016 IEEE, pp 1–6
5. Adwani A, Nagtode S (2016) LI-FI: information transferring through LED’S. In: ICEEOT
6. Tanwar K, Gupta S (2014) Smart class using Li-Fi technology. In: IEEE transactions on
engineering and science, pp 336–338
7. Mahendran R (2016) Integrated Lifi (light fidelity) for smart communication through illu-
mination. In: International conference on advanced communication control and computing
technologies
992 V. Savaliya et al.

8. Komine T, Nakagawa M (2003) Integrated system of white LED visible-light communication

and power-line communication. IEEE Trans Consum Electron 49:71–79
9. Rani J, Chauhan P, Tripathi R (2012) Li-Fi (light fidelity)-the future technology in wireless
communication. ISSN 0973-4562
10. IEEE standard for local and metropolitan area networks—part 15.7: short-range wireless optical
communication using visible light, pp 1–309
11. Jovicic A, Li J, Richardson T (2013) Visible light communication: opportunities, challenges
and the path to market. IEEE Commun Mag 51:26–32
12. Damodaran S, Shaikh T, Taylor NK (2016) Using mobile phone based camera to read
information from a Li-Fi source. 978-1-5090-5527-2/16© 2016 IEEE, pp 165–170
A Survey of Pre-processing Techniques
Using Wavelets and Empirical-Mode
Decomposition on Biomedical Signals

Prasanth M. Warrier, B. R. Manju and Rajkumar P. Sreedharan

Abstract Recorded biomedical statistics are utilized for predicting various syn-
dromes in humans. Recorded electrical activity of heart can be used for predicting
cardiovascular ailment likelihood. Several steps are involved to process biomedical
signals, among which the first step related to pre-processing, in which a noisy sig-
nal is processed for generating noise-free signal, which can be utilized for further
operations. This work gives a detailed understanding of de-noising techniques those
have been used for the last decade, for cardiac signals. These techniques utilize the
benefits of discrete wavelet transforms (DWT), Bayesian approach, singular value
decomposition (SVD), artificial neural networks (ANN), empirical-mode decompo-
sition (EMD), adaptive filtering, and finite impulse response (FIR) filtering. These
techniques have been implemented for de-noising of biosignals, individually as well
as combining with other techniques, for better results.

Keywords ECG · Empirical-mode decomposition · Filtering · De-noising

techniques · DWT

1 Introduction

Biomedical signals can be processed and studied carefully for envisaging anoma-
lies in human body. For example, electroencephalogram (EEG) signals are used
for predicting epileptic nature of human brain and electrocardiogram (ECG) can
be used for predicting cardiovascular diseases. These biomedical signals are very
well applied for applications like biometric identification, emotion recognition, etc.

P. M. Warrier (B) · R. P. Sreedharan

Department of Electronics and Communication, Amrita Vishwa Vidyapeetham, Amritapuri, India
e-mail: prasanthmw@[Link]
R. P. Sreedharan
e-mail: rajsreedharan@[Link]
B. R. Manju
Department of Mathematics, Amrita Vishwa Vidyapeetham, Amritapuri 690525, India
e-mail: manjubr@[Link]
© Springer Nature Singapore Pte Ltd. 2020 993
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
994 P. M. Warrier et al.

Depending on the purpose, the analysis of biomedical signals may contain different
phases in which the first phase could be the pre-processing [1]. In this work, several
de-noising techniques have been discussed that can eliminate or deteriorate effect
of noises embedded in data acquired through electrodes, from subject’s body. In the
case of ECG, noises can be from different sources like (1) power-line interference (50
or 60 Hz), (2) baseline wander (0.5–0.3 Hz), (3) muscle artifacts (electromyography:
EMG), (4) electrode contact noise, (5) electrode motion artifact, (6) electrosurgical
noise, and (7) instrumentation noise. To explore, significant studies in this literature,
works were taken from IEEEXplore, and ScienceDirect. This paper is prearranged
as, Sect. 2 contributes details of pre-processing techniques, developed in the last
decade, Sect. 3 gives data base details, Sect. 4 gives details about various parame-
ters for success measures, Sect. 5 discusses inference from the work and finally, the
Sect. 6 gives conclusion.

2 Pre-processing Techniques

The eminent techniques implemented in various environments are briefed here.

2.1. Savitzky–Golay filtering: This method helps to smooth the noisy ECG sig-
nal. Since sequential operations of polynomial fitting and reevaluation are
identically same as discrete convolution with finite impulse response, SG fil-
tering can be termed as a simple finite impulse response (FIR) filter [2]. If
x(n) represent the data and h(n) is the impulse response, then the convolution
result is given by [2],

N −1

y[n] = h(k) · x(n − k) (1)
k=0

2.2. Adaptive filter: For removing power-line interface (50/60 Hz), an adaptive
filter, which is tuned with ECG signal morphology, can also be used. If the
width of morphological filter is larger than width of noise, useful signal can
be separated from noisy signal. A disadvantage found is, QRS wave, along
with pits and peaks of nearby regions will too be removed [3].
2.3. Wavelets transform: Different wavelets transform (WT) techniques used for
de-noising purpose has been discussed here (Fig. 1).
2.3.1. Adaptive filter with WT and neural network: An adaptive filter
method, which combines discrete wavelets transform (DWT) and arti-
ficial neural network (ANN), can be used for de-noising ECG [4].
Instead of using an inverse DWT, ANN utilized for finding inverse
and a nonlinear adaptive filtering to remove noise [4] (Fig. 2).
2.3.2. DWT with hybridizing β-hill climbing: Beta-hill climbing algo-
rithm has been utilized for choosing the finest wavelet parameters and
A Survey of Pre-processing Techniques Using Wavelets and … 995

Fig. 1 Combinations of WT and other different techniques

Fig. 2 Combination of adaptive filter, DWT, and ANN

used with DWT techniques, so that mean squared error is minimum

[5].
2.3.3. Discrete wavelet transform (DWT) and cauchy distribution at
sub-bands: A cauchy probability function can be used for model-
ing wavelet coefficients in each sub-band, where thresholds are being
calculated at each level. Individual lead noisy ECG records from
cardiovascular Centre of Glasgow University are being used [6].
2.3.4. To de-noise ECG signal by using non-local estimation with DWT:
Though DWT technique is highly effective with larger decomposition
levels, in removing HF noise, it can be also used to eradicate low
frequency [7]. A method has been proposed, in which a combination
of DWT with non-local means estimation is applied. Denoising using
NLM would give a minimized mean error and falsehood improvement
but with longer computational time [7].
2.3.5. Discrete wavelet transform (DWT) and S-median threshold: ECG
signal is converted into wavelets and then the variance of noise sig-
nal in each sub-band is estimated using mean absolute deviation
(MAD) [8]. Threshold is evaluated using S-median (sub-band level-
dependent median threshold) and a soft thresholding is applied with
sub-band level-dependent threshold, after which signal reconstruction
is implemented using inverse wavelet transform (IWT) [8].
996 P. M. Warrier et al.
    
tl,k = σk · 2 · log(n) / Sl,k + b where, k = 1, 2, . . . , l. (2)

k
Sl,k = 2 L− L =−band level dependent parameter. (3)

L deepest level of decomposition,

σ k variance of noise,
k level at which thresholding is done.

median(|x|)
variance of noise, σk = , k = 1, 2, 3, 4, . . . , l. (4)
0.6745
S-median DM utilizes mean-value differences of both sets and found
that S-median DM is better [8]. In modified S-median thresholding
technique, an optimal Symlet function has been used along with a
tuning factor [9].
2.4. Empirical-mode decomposition (EMD): Empirical-Mode decomposition
characterizes random signal as sum of intrinsic mode functions (IMFs) which
are chosen and processed for de-noising [10]. Figure 3 represents various
combinations of EMD, for de-noising purpose.
2.4.1. Adaptive switching mean filter in EMD: Three types of noises have
been considered here namely, EMG, WGN, power-line interference
[11] (Fig. 4).

Fig. 3 Combination of EMD with other techniques

Fig. 4 De-noising method using EMD and adaptive switching mean filter
A Survey of Pre-processing Techniques Using Wavelets and … 997

2.4.2. Ensemble EMD (EEMD) with genetic algorithm-based threshold-

ing for adaptive de-noising of ECG: Kullback-Leibler divergence
method and a probability density function is used for segregating
IMFs into signal dominant group and noise dominant group. Adaptive
de-noising is performed with the help of genetic algorithm, and then
signal dominant IMFs and de-noised noise dominant IMF are selected
for reconstructing pure signal [12].
2.4.3. Combination of ensemble empirical-mode decomposition
(EEMD) and block LMS adaptive algorithms as well as dis-
crete wavelet transform (DWT) and neural network (NN): Those
combinations have been developed for de-noising and they have
been analyzed in [13]. Their performance was later compared with
conventional EMD (CEMD), CEEMD, EEMD + LMS and DWT
thresholding methods. Among the above mentioned techniques,
based on performance, EEMD + LMS stands best.
2.4.4. Noise exclusion from ECG using Eigenvalue decomposition (using
Hankel matrix EMD with DWT): Eigenvalue breakdown of Henkel
matrix is utilized for de-noising process of ECG [14]. Notch filters can
be used for removing PLI, but with ringing issue. For decomposition,
basis functions which are Eigen vectors are obtained by decomposition
of Hankel matrix and for reconstruction, Eigenvalue pairs are used
[14].
2.4.5. EMD with NLM: Non-local means (NLM) technique has been com-
bined with empirical-mode decomposition (EMD) to preserve the
edges in ECG signal [15] (Fig. 5).
2.4.6. EMD with adaptive filtering: Morphological structure of ECG sig-
nals can get fluctuated by power-line interference of 50 Hz. That is, a
noise cancelation should work within 48–51 Hz [16]. In this proposed
method, 50 Hz noise is removed using adaptive filter in first IMF,
since only first IMF contain PLI and rest are free from PLI, and those
coefficients are regulated according to LMS [16].
2.4.7. Combination of EMD and wavelet techniques for de-noising ECG:
In this work, a combination of EMD and DWT has been tried, in which
no IMFs have been omitted and thresholding for wavelet transform
has been done adaptively [17] (Fig. 6).

Fig. 5 De-noising process

using EMD and NLM
998 P. M. Warrier et al.

Fig. 6 Noise reduction using EMD and DWT

2.5. Extended Kalman filter: 2-d EKF structure has been modified into 17-d EKF
structure, for de-noising, compression, and estimation of 15 model parame-
ters, by utilizing benefits of synthetic ECG signal proposed by McSharry and
co. An SNR improvement of 1.8 dB has been observed and found it better
than multi-adaptive bionic wavelet transform (MABWT) and EKF2 though
it suffers from filtering performance [18].
2.6. Periodicity-based NLM de-noising technique for ECG in small SNR con-
ditions: De-noising of ECG signals is tried along with NLM, by making use
of periodic nature of ECG [19]. Using FIR and IIR filters, it is challenging to
remove EMG noises or other noises that overlay, in 20–200 Hz [19].
2.7. Mean-shift algorithm: Discrete-time ECG samples are synthesized to contin-
uous signal, keeping signal morphology, using embedded Gaussian smoother
and the extreme of continuous-time signal can be located, which corresponds
to peak values in original ECG signals [20]. For modeling, mainly Gaussian
functions were used, while wavelet synthesis and neural network are poor in
adjusting morphology of typical output [20].
2.8. Adaptive Fourier decomposition: Signal is represented as sum of mono-
components and standard remainder [21].

∞
 
N
G(t) = Ck · e jkt ∈ H 2 space = Sn (t) + Rn (t);
k=0 n=1
∞
where (|Ck |)2 < ∞. (5)
k=0

AFD found better than Butterworth LPF, WT, EMD, EEMD, for muscle and
motion artifacts. Results obtained were better than EMD/EEMD algorithms,
in which, it shows, later algorithms do not hold strong mathematical basis
[21].
2.9. Bayesian approach with phonocardiogram (PCG): Phonocardiogram is
a graphical representation of acoustic vibrations of heart, which may con-
tain heart sounds and murmurs [22]. Based on PCG morphology, a dynami-
cal model for PCG is constructed for synthetic PCG signals and a Bayesian
framework has been opted for de-noising [22].
2.10. Approximation of EMG in ECG signals: Translation invariant approx-
imation is done using stationary WT. It is very difficult to identify and
remove EMG noise and patient electrode motion artifacts. Karhunen-Loever
transform (KLT) is the one, which is used for noise removal in [23].
A Survey of Pre-processing Techniques Using Wavelets and … 999

2.11. Wearable ECG monitoring devices and ECG de-noising: Using multi-
stage decision tree algorithm, once small and large frequency components are
detected, then it classifies, small frequency into BW/ABC using local dynamic
amplitude feature range and classifies high frequency into PLI/EMG, using
local autocorrelation function extreme peak feature. Parameters like average
sensitivity was observed to be 97.88%, positive productivity = 91.18%, and
accuracy = 89.06% [24].
2.12. Compression and noise reduction of biomedical signals by singular value
decomposition: In this method, Bijective mapping of signal vector into matrix
is done for de-noising any biomedical signal [25]. This matrix is decomposed
to get rank-1 matrices, which are summed up to give singular values. SVD
de-noising is accomplished by attenuating all small singular values [25].

3 Data Bases

Research and analysis of arrhythmia and related things have been supported by
MIT-BIH. Data base provides data from 47 subjects as given in Table 1 [1, 26, 27].

4 Improvement Measures

4.1. SNR improvement measure



(|x n (i) − x(i)|) 2

SNRimprovement [dB] = SNRoutput − SNRinput = 10 · log i ,
i (|x d (i) − x(i)|)
2

(6)

x pure ECG,
x d de-noised ECG signal, and
x n noisy ECG.

Table 1 MIT-BIH data base details used often by researchers

Data bases Records Subjects Duration Sampling Bits per Channels
(min) frequency sample (leads)
MIT-BIH 48 47 30 360 11 2
arrhythmia
MIT-BIH 18 18 24 h 128 – 2
normal
sinus
rhythm
1000 P. M. Warrier et al.

Table 2 Four possibilities of

True condition Positive assessment Negative assessment
dimension and class
Real-positive Positive-true Negative-false
Real-negative Positive-false Negative-true

4.2. Error measurements


 N
1 
Rootmean Square Error, RMSE = [w(i) − z(i)]2 , (7)
N i=1

w Original-signal and
z Filtered-signal.

4.3. Success measures

In problem of statistical classification, error matrix or confusion matrix
allows performance visualization of an algorithm (Table 2).
While classifying signals, constraints such as average sensitivity, precision
or positive predictivity and accuracy have been utilized.
 
TP
Average Sensitivity, Se = ∗ 100 (8)
TP + FN
 
TP
precision or positive predictivity, +P = ∗ 100 (9)
TP + FP
 
TP
accuracy = ∗ 100 (10)
TP + FN + FP

5 Discussion

In this work, 22 different algorithms have been discussed, which are used for de-
noising biomedical signals, especially ECG. From this survey, it became apparent
that wavelet transform techniques are more mathematically stable and whose com-
bination along with neural networks can give faster results, thereby saving lot of
computation time. On the other hand, de-noising techniques, in real time, still needs
lot of improvement.
A Survey of Pre-processing Techniques Using Wavelets and … 1001

6 Conclusions

For pre-processing, various techniques are being followed and this work discusses
about various de-noising techniques. Methods like discrete wavelet transforms
and various thresholding techniques, Bayesian approach, artificial neural networks,
empirical-mode decomposition, singular value decomposition, adaptive filtering, and
finite impulse response filtering, have been studied. Parameters of interest, for suc-
cess measures, utilizes error matrix/confusion matrix, for SNR improvement, in most
existing works. Above-mentioned methods as well as their combination also being
tried to improve overall result. This paper gives an insight about significant de-noising
techniques that were developed, in the past one decade.

Acknowledgements We would like to thank our beloved Chancellor Sri (Dr). Mata Amritanan-
damayi Devi, for giving a superior inspiration to implement this research work.

References

1. Berkaya SK, Uysal AK, Gunal ES, Ergin S, Gunal S, Gulmezoglu MB (2018) A survey on
ECG analysis. Biomed Signal Process Control 43:216–235
2. Savitzky A, Golay MJ (1964) Smoothing and differentiation of data by simplified least squares
procedures. Anal Chem 36(8):1627–1639
3. An-dong W, Lan L, Qin W (2012) An adaptive morphologic filter applied to ECG de-noising
and extraction of R-peak at real time. AASRI Procedia 1:474–479 (Elsevier, ScienceDirect)
4. Poungponsri S, Yu X-H (2013) An adaptive filtering approach for ECG signal noise reduction
using neural network. NeuroComputing 117:206–213 (ScienceDirect)
5. Alyasseri ZAA, Khader AT, Al-Betar MA, Awadallah MA (2018) Hybridizing β-hill climbing
with wavelet transform for de-noising ECG signals. Inf Sci 429:229–246
6. Manthalkar R, Ardhapurkar S, Gajre S (2010) Wavelet based ECG de-noising by employing
Cauchy Distribution at sub bands. In: ICSP2010 proceedings
7. Singh P, Pradhan G, Shahnawazuddin S (2017) De-noising of ECG signal by non-local
estimation of approximation coefficient in DWT. Biocybern Biomed Eng 37:599–610
8. Poornachandra S (2008) Wavelet based de-noising using sub-band dependent threshold for
ECG signals. Digit Signal Process 18:49–55 (ScienceDirect)
9. Awal MA, Mostafa SS, Ahmad M, Rashid MA (2014) An adaptive level dependent wavelet
thresholding for ECG de-noising. Biocybern Biomed Eng 34:238–249
10. Blanco-Belasco M, Wengb B, Barnerc KE (2008) ECG signal de-noising and baseline wander
correction based on empirical-mode decomposition. Comput Biol Med 38:1–13
11. Rakshit M, Das S (2018) An efficient ECG de-noising methodology using EMD and adaptive
switching mean filter. Biomed Signal Process Control 40:140–148
12. Nguyen P, Kim J-M (2016) Adaptive ECG de-noising using genetic algorithm based
thresholding and EEMD. Inf Sci 373:499–511
13. Kærgaard K, Jensen SH, Puthusserypady S (2016) A comprehensive performance analysis of
EEMD-BLMS and DWT-NN hybrid algorithms for ECG De-noising. Biomed Signal Process
Control 25:178–187
14. Sharma RR, Pachori RB (2018) Baseline wander and power line interference removal from
ECG signals using Eigen value decomposition. Biomed Signal Process Control 45:33–49
1002 P. M. Warrier et al.

15. Kumar S, Panigrahy D, Sahu PK (2018) De-noising of electrocardiogram (ECG) by using

empirical-mode decomposition (EMD) with non-local mean (NLM) technique. Biocybern
Biomed Eng 38:297–312
16. Suchetha M, Kumaravel N, Jagannath M, Jaganathan SK (2017) A comparative analysis of
EMD based filtering method for 50 Hz noise cancellation in ECG signal. Inform Med Unlocked
8:54–59
17. Kabir MA, Shahnaz C (2012) De-noising of ECG signals based on noise-reduction algorithms
in EMD and Wavelet domains. Biomed Signal Process Control 7:481–489
18. Sayadi O, student member, IEEE, Shamsollahi MB, member, IEEE (2008) ECG de-noising
and compression using a modified extended Kalman filter structure. IEEE Trans Biomed Eng
55(9)
19. Lee Y, Hwang D (2018) Periodicity based NLM de-noising method for ECG in low SNR
non-white noisy conditions. Biomed Signal Process Control 39:284–293
20. Yan J, Lu Y, Liu J, Wub X, Xu Y (2010) Self-adaptive model based ECG de-noising using
features extracted by mean shift algorithm. Bio-med Signal Process Control 5(2):103-113.
[Link]
21. Wang Z, Wan F, Wong CM, Zhang L (2016) Adaptive Fourier decomposition based ECG
de-noising. Comput Biol Med 77:195–205
22. Almasi A, Shamsollahi MB, Senhadji L (2013) Bayesian de-noising framework of phonocar-
diogram based on a new dynamical model. IRBM 34:214–225 (Elsevier, ScienceDirect)
23. Marouf M, Saranovac L, Vukomanovic G (2017) Algorithm for EMG noise level approximation
in ECG signals. Biomed Signal Process Control 34:158–165
24. Satija U, Ramkumar B, Manikandan MS, School of Electrical Sciences, Indian Institute of Tech-
nology Bhubaneswar (2015) A simple method for detection and classification of ECG noises
for wearable ECG monitoring. In: 2015 2nd international conference on signal processing and
integrated networks (SPIN)
25. Schanze T (2018) Compression and noise reduction of biomedical signals by singular value
decomposition. IFAC-PapersOnLine 51(2):361–366 (Elsevier)
26. MIT-BIH Arrhythmia database. Available Online [Link]
mitdb/,accessed on 25-8-2018
27. Goldberger AL, Amaral LAN, Glass L, Hausdorff JM, Ivanov PCh, Mark RG, Mietus JE,
Moody GB, Peng C-K, Stanley HE (2000) PhysioBank, PhysioToolkit, and PhysioNet: com-
ponents of a new research resource for complex physiologic signals. Circulation 101(23):e215–
e220. Circulation Electronic Pages; [Link] 13
June 2000
Analysis of Process Scheduling Using
Neural Network in Operating System

Harshit Agarwal and Gaurav Jariwala

Abstract Process scheduling plays a vital role in multitasking for any operating
system. There are many factors involved during process scheduling like priorities,
free memory, user demand and processor which if not handled properly can be very
complex and time consuming. Neural network has adaptive nature which can be used
to handle the complex part easily. The main aim of this paper is to review different
types of scheduling algorithms working on the principle of neural network and offer
constructive criticism to improve their efficiency.

Keywords Backpropagation · Bayesian system · Decision tree learning · Genetic

algorithm · Neural network · Process scheduling · Rule-based system

1 Introduction

Since different users use their computer systems in a different ways, neural network
in process scheduling will help to optimize the work. Neural network has the ability
to learn and solve complex problems while finding the most optimal solutions.
Efficient process scheduling and context switching is the most important step of
multitasking. This could be the stepping stone for the AI-based programs and neural
network. Their applications are endless. Knowing what job is to be scheduled by
studying the user behavior and doing that in an optimal way is a crucial step. It is
necessary for the operating system to understand which processes are important and
which are not. Researchers have focused on developing algorithms for learning the
user behavior on system by recording the application’s use and generating a predictive
loop. The drawback of this system is that it requires a lot of time to process and is
unpredictable.

H. Agarwal (B) · G. Jariwala

Sarvajanik College of Engineering and Technology, Surat, India
e-mail: 9arshit@[Link]
G. Jariwala
e-mail: gjariwala9@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1003

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1004 H. Agarwal and G. Jariwala

For process scheduling, there are many algorithms used. All techniques have some
merits and demerits to them which we will discuss in this paper. Methods that will be
discussed are job shop scheduling, preemptive scheduling, multithreading, genetic
algorithm, decision tree learning, Bayesian system, rule-based system and neural
network.
The remainder of this paper is organized as follows. Section 2 reviews related
work. Section 3 describes process scheduling based algorithm in detail. Section 4
describes AI techniques based algorithm in detail and existing research. Section 5
shows comparison table between different algorithms. Finally, Sect. 6 draws
conclusions and discusses future work.

2 Related Work

Ajmani and Sethi [1] proposed a fuzzy logic-based CPU scheduling algorithm to
overcome drawbacks of conventional algorithms for efficient utilization of CPU.
This paper has compared the proposed fuzzy CPU scheduling algorithm (PFCS)
with priority algorithm; they found that the priority algorithm has more average
waiting time and average turnaround time than PFCS. Rehaiem et al. [2] presented
an approach for real-time scheduling of reconfigurable embedded systems using
neural networks. Sharma et al. [3] have given a optimize credit based scheduling
algorithm based on load balancing to improve the performance in cloud computing.

3 Process Scheduling Algorithms

Process scheduling is a procedure that arranges the order and time period of different
processes during which they can use the CPU. If not for processing algorithms, a lot
of CPU cycles would have been wasted ideally when the process is waiting for I/O
or memory. Algorithms in use make our system efficient and fair.

3.1 Job Shop Scheduling

Job shop scheduling is also called linear workshop process scheduling [4]. It can be
considered the simplest form of process scheduling. There are different job shops
presently capable of handling the jobs, see [5]. They accept the contract and complete
the work. In this case, the processes are jobs that are to be finished. The problem is
to schedule these jobs in such a way that the jobs have to wait for minimum amount
of time before they are processed, and the entire system works efficiently.
Analysis of Process Scheduling Using Neural Network … 1005

For example, before data analysis process is done, it is necessary to obtain data.
Hence, I/O process cannot be scheduled after analysis process. Job shop scheduling
requires details for the process, time of completion beforehand. We cannot add a
new process at runtime which is unsuited for OS. Hence, this method is too static for
process scheduling in OS [4].

3.2 Preemptive Scheduling

A scheduling discipline is categorized as preemptive if once the process has been

given the CPU, it can be taken away. In preemptive scheduling, a process is given a
quantum time during which it is allowed to run. After the quantum time is expired for
a process, the CPU control is returned to kernel. This is done using a clock interrupts.
The kernel then saves the state, and then, it decides which process will now be given
control of the CPU.
If the process has completed before its quantum time expires, the process can
forfeit its ownership of the CPU voluntarily to the kernel. This system is unlike non-
preemptive scheduling where only when the process itself gives up the CPU kernel
cannot access it. If the process is not completed, it goes back to the ready queue and
waits for its turn again. The priority of the process in that queue is then depended on
the algorithm the processor is running on.
Algorithm that is based on preemptive scheduling is Round Robin. The shortest
job first (SJF) and priority scheduling under specific circumstances can be classified
as preemptive scheduling. This kind of algorithm is better suited for background
process than user process as shutting down the user process against their need will
defeat the purpose of user-friendly OS.

3.3 Multithreading

Threads are called lightweight processes (LWP). Threads within a process share
address space and other resources. The term multithreading is used to describe
multiple threads in same process. When a multithreaded process is running on
single-CPU system, the threads run one at a time just like processes in multipro-
gramming on single-CPU system. The CPU switches between threads so quickly
that it gives an illusion of threads running in parallel. True parallelism of running
threads can be achieved using multithreaded processors. The minimum requirement
for multithreaded processor is the ability to pursue two or more threads in parallel
within the processor pipeline—i.e., it must provide two or more independent program
counters—and a mechanism that triggers a thread switch [6].
The best way to understand the usefulness of threads is through example. Word
processors use threads to increase productivity and improve interactivity. When a
user types a character at the keyboard, one thread is used to read that character. Word
1006 H. Agarwal and G. Jariwala

processor can execute other threads between keyboard interrupts. Like, storing the
document in the disk to prevent data loss and checking for spelling mistakes. Each
feature is implemented using different threads so even if a thread is blocked due to an
I/O operation (storing the document), the word processor can response to keyboard
interrupt.
Researchers have attempted to combine advantages of both user threads and ker-
nel threads called scheduler activation (M:N model) [7]. The goals of the scheduler
activation work are to mimic the functionality of kernel threads, but with the better
performance and greater flexibility usually associated with threads packages imple-
mented in user space [8]. A neural network scheduler can be designed for threads
and process in similar way so in this paper we will concentrate on process scheduling
rather than thread scheduling.

4 AI Technique Based Algorithms

Algorithms discussed above may not always work. They might be efficient, but they
are too static. In order to create a perfect neural system-based operating system, it
is necessary to have components that learn from the user over a period of time. AI
techniques can be used for process scheduling. It learns user behavior and reprograms
the system to predict user’s need.

4.1 Genetic Algorithm

Genetic algorithm was inspired by the Darwin theory of natural evolution. This
algorithm reflects the process of natural selection where the fittest individuals are
selected for reproduction in order to produce offspring of the next generation [9].
The solutions are selected on the bases of their fitness and then promoted further.
They are grouped in pairs and combined using genetic operator like mutation and
crossover. The child solution created from the parent using these methods then acts
like a parent in further iterations.
Figure 1 shows the flowchart of genetic algorithm where initial set of individual
solutions are called population. An individual is characterized by a set of parameters
(variables) known as Genes. Genes are joined into a string to form a chromosome
(solution) as presented in Fig. 2. This process is repeated and evaluated until there is
a solution in the set that exceeds a minimum fitness. Crossover operator is applied to
the selected chromosomes during which better string is created, while the information
in the parent string is optimally preserved. The crossover is done in hope of creating
a better string, but the result obtain may or may not be the desired one depending on
the use. After crossover, the strings are subjected to mutation [10]. Mutation causes
bit-wise reversal. The bit’s position is chosen with probability of Pm. The process
of mutation is done to maintain diversity in population since selection process may
Analysis of Process Scheduling Using Neural Network … 1007

Fig. 1 Flowchart for genetic

algorithm. Source [9]

Fig. 2 Elements of genetic

algorithm. Source [9]

promote same breed of strings which does not optimize search algorithm and helps
to retrieve information that may be lost during crossover operator.
Limited research projects have been on implementing genetic algorithm-based
schedulers [11, 12]. The process scheduler based on genetic algorithm shows similar
results like shortest job first algorithm [13] in certain cases. But shortest job first
algorithm cannot be implemented at the level of short-term CPU scheduling. There is
no way to know the length of the next CPU burst. Here, performance can be improved
by using genetic algorithm. The other studies involved did not use algorithm in the
scheduler but rather in the application. The study showed that genetic algorithm had
same result as the priority scheduler and the algorithm was time consuming. But the
time consumption was mostly due to hardware. With even more simplified algorithm,
this problem can be easily solved. The genetic algorithm evolving technique does
provide more flexible mechanism than FIFO [8] scheduling and adapts itself to
changing environment.

4.2 Decision Tree Learning

As the name says it is a tree-like model of decision. A decision tree contains non-
leaf (internal nodes) nodes which represent input features and the leaf nodes are
the possible solutions. The splitting at non-leaf nodes is done according to certain
discrete function of input attribute values (as in our case) is called classification tree.
1008 H. Agarwal and G. Jariwala

Fig. 3 Decision tree to play

football

As shown in Fig. 3, an example decides whether to play football or not (in this
case Yes or No). At the root node, an instance is classified by testing the attribute
specified by this node, and then moving down tree branch according to the value of
the attribute. This process is repeated at each node until leaf node is reached.
This algorithm is recursive in nature as same approach is used at each node to
split the groups formed. As this algorithm never backtracks to reconsider earlier
choices, it is also called greedy algorithm. The algorithm is very simple and easy
to implement. The main disadvantage of using this algorithm is its very sensitive to
noise meaning small variations in data might result in completely different tree being
generated. Furthermore, due to the greedy characteristic of decision tree it cannot
guarantee the globally optimal solution. See Doh et al. [14] for decision tree base
scheduling for flexible job shops.

4.3 Bayesian System

Bayesian networks also known as belief networks are a graphical model that shows
probabilistic relationships among set of variables [15]. It is a graph called a directed
acyclic graph (DAG) which contains no directed cycles. Bayesian network requires
probability distribution P(Y|X) for each node Y. If node Y has no parents, then it is
just P(Y ). P(X|Y ) is known as conditional probability. For example, to calculate the
probability of computer failure, see [16].
Bayesian network can be used, even in the case of missing data, to learn the
causal relationships and gain an understanding of the various problem domains and
to predict future events [17]. As Bayesian network is a DAG, it cannot be used when
the graphical model contains cycle. This approach is computationally expensive. In
our case, there is no real structure that can easily be imposed on the data of the
system.
Analysis of Process Scheduling Using Neural Network … 1009

4.4 Rule-Based System

Rule-based system uses a set of IF-THEN rules for classification. If the condition
mention on the “if” side is satisfied, then the rule is said to be triggered, and the
action corresponding in the “then” part executes. This system is the most accurate
in comparison with other systems; this is because other system works on probability
functions that may not give us always the desired or the most appropriate result. The
rule-based system is only as accurate as the rules programmed in the system, but this
turn creates problems such as, what if more than one rule is fired or none of them
does. This issue can be solved with help of size ordering, rule ordering or class-
based ordering [18], the programmer needs to cover the entire basis which makes
this impractical for complex systems.
An example of a rule-based system is the domain-specific expert system that uses
rules to make deductions or choices. For example, an expert system might help a
doctor choose the correct diagnosis based on a cluster of symptoms or select tactical
moves to play a game [19].
However, by creating a hybrid system using decision tree and rule-based system
can have vast applications with the least amount of drawbacks. We can extract rule
from the decision tree system where rules are created from the path of root to leaf
node. Using this system helps to convert “IF-THEN” static system to dynamic system
which increases its application arena.

4.5 Neural Network

Neural networks or artificial neural networks (ANNs) are analogy from biological
neural networks in which neurons are used to transmit signals, while in ANNs nodes
are used for the same. These nodes form layers, and each node is connected with
every node of next layer with some weights.
There are three types of layers in ANNs, input layer, hidden layers and output
layer. Any number of hidden layers can be there in a neural network. The input layer
is used to feed the data to the neural network, while the output layer gives the result
of the provided input. Each node has its activation upon which the output is decided.
The activation of each node is calculated by summing the products of activation of
every incoming node with their associated weight and adding it with the bias. Then,
the resultant value is passed to a function called sigmoid to get the activation value
of the node between 0 and 1.
Backpropagation algorithm is the most commonly used technique in ANNs. In
this, each weight is adjusted from the output layer to input layer so that the network
error can be reduced. There are other techniques that are used in ANNs which are
out of scope for this paper.
Neural network is the most fitting technique to be used in process scheduling as
it can adapt to different situations. Since the data of the process are continuous in
1010 H. Agarwal and G. Jariwala

Fig. 4 Comparision of scheduling algorithm. Source [20]

nature so are input and output. The demerit of the neural networks is that it is hard
to interpret the model as they are trained.
Many studies were done in which stimulations were run for using different algo-
rithms on kernel level. Results obtained from these studies showed that neural
network-based algorithms were faster than conventional system used currently.
As shown in Fig. 4, genetic algorithm had the least waiting time among the
different algorithm used for job scheduling. Its use also provides more flexibility
than other algorithms [20]. Results obtained from different studies [4, 14] were also
in favor of neural network-based kernel. During initial phase of the use, system
was slow which caused the running applications’ performance in percentage of FPS
drop and loading speed of the multimedia. But with time, the system was trained
and performance showed significant improvement in waiting time for the jobs and
percentage of FPS drop. The initial stage problem was reduced with help of pre-
training but was not eliminated. Use of decision tree-based approach reduced flow
time and increased tardiness. Stable performance was achieved with improvement
in the method.

5 Comparison of Algorithms

See Table 1.
Table 1 Comparison between different scheduling algorithms
Name Method Merits Demerits Improvements
Job shop scheduling – Queues the job in linear – Simplest form of – Creates a static system – Using multilayer function
manner scheduling – Limits multitasking for scheduling to increase
– Compatible with all the productivity
systems
Preemptive scheduling – Assigns quantum time to – Single process cannot – It is not suitable for – User-based processes
process during which monopolize the CPU user-based background should have programming
process takes the CPU. – Priority-based quantum processes to override quantum time
After which it is put back time can be assigned
into queue
Multithreading – More than one thread work – Improves system’s – Debugging becomes – Banker’s algorithm can be
concurrently within a performance complex used to avoid deadlock but
process – Better use of CPU – Increase occurrence of for using this algorithm
resources deadlock maximum numbers of
resources needed by each
process should be known
Genetic algorithm – Selection, crossover and – It overcomes the limitations – Crossover may cause – Creation of simpler
Analysis of Process Scheduling Using Neural Network …

mutation functions are used of shortest job first unwanted results to reoccur program to avoid
to select the most suitable algorithm – Results are based on complexity
process probability function’s – Using adaptive crossover
accuracy function
Decision tree learning – Non-leaf nodes are split, – It is easy to implement – Overfitting is caused due to – Using pruning size of the
and the output is generated – Nonlinear parameters can noise in the dataset decision tree can be
by the series of decisions be handled – Does not produce globally reduced to tackle overfitting
optimal solution
(continued)
1011
Table 1 (continued)
1012

Name Method Merits Demerits Improvements

Bayesian system – It works on the bases of – Can be used even in the – It cannot be used when the – Its accuracy is very good on
probabilities case of missing data graph contains cycle small datasets but it may
– It is computationally asymptote to a high error
expensive rate, making it less useful
as a classifier for very large
databases [15]
Rule-based system – Uses set of IF-THEN rules – Most accurate system – All the rules need to be – Hybrid system of rule
to trigger the rule – Can be integrated with any mentioned manually based and decision tree can
system have more applications
Neural network – It works on layered – Creates a user-specific – Specialized hardware needs – Creating better
structure and output is system to be used to process data compression system for fast
generated using the – Adapt to different situations – Hard to interpret the computing
activation structure of neural network
H. Agarwal and G. Jariwala
Analysis of Process Scheduling Using Neural Network … 1013

6 Conclusion and Future Work

In this paper, process scheduling algorithms and AI-based techniques were analyzed
for their implementation in operating system. While in comparison study of some of
these algorithms gave similar results, the more advance approach of neural network-
based algorithms had the advantage of being more flexible. With recent increase in
research in field of artificial intelligence and machine learning, these systems are
more likely to use in the future. Job shop scheduling is obsolete and static to handle
multitasking used in the current system, while Bayesian system implementation
proves to be only theoretical in our case. Though initial uses of neural network
systems were slow in comparison with regular process scheduling algorithms, with
respect to time the training program and prediction loop provided more efficiency.
Until training program reaches to optimal level use of generic algorithms will be
more beneficial and then switching of algorithms could take place.

References

1. Ajmani P, Sethi M (2013) Proposed fuzzy CPU scheduling algorithm (PFCS) for real time
operating systems. BIJIT - BVICAM’s Int J Inf Technol 5(2) (New Delhi, India)
2. Rehaiem G, Gharsellaoui H, Ahmed SB (2016) A neural networks based approach for the real-
time scheduling of reconfigurable embedded systems with minimization of power consumption.
In: IEEE/ACIS 15th international conference on computer and information science (ICIS),
Okayama, 2015, pp 1–6
3. Sharma A, Gupta AK, Goyal D (2018) An optimized task scheduling in cloud computing
using priority. In: Proceedings 3rd international conference on internet of things and connected
technologies (ICIoTCT), Jaipur, India, 26–27 Mar 2018
4. Bex P (2008) Implementing a process scheduler using neural network technology. Radbound
University Nijmegen
5. Yoo BY (1977) Methods and techniques used for job shop scheduling. University of Central
Florida
6. Ungerer T, Robic B, Silc J (2002) Multithreaded processors. Comput J 45(3):321–348
7. Anderson TE, Berchad BN, Lazowska ED, Levy HM (1992) Scheduler activations: effec-
tive kernel support for the user-level management of parallelism. ACM Trans Comput Syst
10(1):53–79
8. Tanenbuam AS (2009) Modern operating system (3rd edn). Upper Saddle River, NJ, 07458
9. Mallawaarachchi V, Introduction to genetic algorithms. [Link]
10. Pai GAV, Rajasekaran S (2011) Neural networks, fuzzy logic and genetic algorithms - synthesis
and applications, India, July 2011
11. Sharma M, Sindhwani P, Maheswari V (2013) Genetic algorithm optimal approach for
scheduling processes in operating system. Int J Eng Res Technol 2. ISSN: 2278-0181
12. Elrad T, Jinlong L, Cork DJ (1998) Evolutionary computation for scheduling controls in
concurrent object-oriented systems. Int J Comput Their Appl 5(3):11–20
13. Arpaci-Dusseau RH, Arpaci-Dusseau AC (2014) Operating systems: three easy pieces. Arpaci-
Dusseau Books
14. Doh HH, Yu JM, Kwon YJ, Shin JH, Kim HW, Nam SH, Lee DH (2014) Decision tree based
scheduling for flexible job shops with multiple process plan. World Acad Sci Eng Technol
8(3):621–627
1014 H. Agarwal and G. Jariwala

15. Kohavi R, Becker B, Sommerfield D (2001) Improving Simple Bayes. In: Proceedings of the
European conference on machine learning
16. Horný M (2014) Bayesian networks. Technical report no. 5, Department of Health Policy &
Management, Boston University School of Public Health, Apr 2014
17. Ben-Gal I (2007) Bayesian networks. In: Ruggeri F, Faltin F, Kenett R (eds) Encyclopedia of
statistics in quality & reliability. Wiley, London
18. Jiawei H, Kamber M, Jian P (2012) Data mining concepts and techniques (3rd edn)
19. Gupta A, Newell A, Wedig R, Forgy C (2005) Parallel algorithms and architectures for rule-
based systems. IEEE Computer Society Press, Los Alamitos, CA. Tokyo, Japan
20. Dr. Kumar R, Er. Kumar R, Er. Kaushik A, Er. Gill S (2010) Genetic algorithm approach to
operating system process scheduling problem. Int J Eng Sci Technol 2:4247–4252
An Enhanced Trust Based Fuzzy Implicit
Cross-Layer Protocol for Wireless Sensor
Networks

Kompalli Anusha and Ambidi Naveena

Abstract Cross-layer procedure integrates functionalities from first layer of OSI

model (physical layer) to transport layer. It enables flexibility, trustworthy and effec-
tiveness in communication process. In this approach, it collects system parameters
from multiple layers to enhance the capability of the network. The standard level
is decreased by enabling flexibility through inter-layer information exchange. The
node selection mechanism is done through fuzzy logic system to provide an effi-
cient communication. Among these benefits, the cross-layering approach faces a
problem with security threats in a network. To mitigate these attacks in a network,
a trust based cross-layering framework (T-XLM) initiates a trust estimation mech-
anism using fuzzy logic system to articulate approximate experimental knowledge
which is used in reputation building in nodes to avoid defaults in future actions. The
TRUFIX is a T-XLM based protocol which is used to permit and hold inter-layer
data exchange to accommodate traffic awareness and improve system version. The
extension of TRUFIX is E-TRUFIX in which the node if it identifies a malicious node
it takes an alternate neighbor route and sends the packet toward the destination. By
taking into account with simulation results, E-TRUFIX was compared with FUGEF
and TRUFIX which shows an increment in the packet delivery ratio and delay due
to the alternate neighbor route.

Keywords Cross-layer approach · Resource bound security solutions · Fuzzy logic

system · Wireless sensor networks · Black hole · Sybil · Malicious node

K. Anusha (B)
Branch of WMC, G. Narayanamma Institute of Technology and Science, JNTUH, Hyderabad,
India
e-mail: anusha.kompalli22@[Link]
A. Naveena
Department of ETM, G. Narayanamma Institute of Technology and Science, JNTUH, Hyderabad,
India

© Springer Nature Singapore Pte Ltd. 2020 1015

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1016 K. Anusha and A. Naveena

1 Introduction

Wireless sensor network comprises massive amount of nodes that gather and sense
physical variables from the sensing limit and transfers the data toward the destination.
As sensor node has definite energy resources, less energy must be consumed which
in turn improves the effectiveness of the system. The dynamic transport procedures,
such as Greedy Perimeter Routing protocol [1], Contention-Based Forwarding [2],
Implicit Geographic Forwarding protocol [3] and cross-layer routing protocol [4], are
designed using cross-layer method. An improvement in energy and QOS is achieved
by cross-layer protocols than traditional layered protocols. In traditional layered
protocols, the information is not exchanged between layers compared to cross-layer
protocols. In cross-layer approach, most of the approaches failed to consider the
concept of entire security phenomenon to contempt its value in the present system and
communication methods. In order to provide security in cross-layer protocols, they
failed in bringing security at less than three layers which in turn consume resources
to implement security mechanisms such as encryption and decryption mechanisms
which include keys, respectively, which tend to exert substantially on resources. The
memory, bandwidth and energy are tended to be consumed more in key management
mechanism which employs cryptography when affirmed to multi-hop network at
each sender node required to perform encryption and decryption while preserving
original sender cipher. Due to these circumstances in a network, it faces the problem
of raised delay, minimum lifetime and null delivery due to drain nodes. To overcome
these problems, the trustworthy system is introduced to provide secure data delivery.
Trust is assuredness of honesty between two entities which are involved in com-
munication and it is executed in a field of network security to protect and handle
interactions between nodes. This process is accomplished by accommodating gen-
erated evidence from previous events and restored to give a report to manage future
nodes interactions. This trust theory arises from secured loop feedback rule method
joining subjectivity, uncertainty. The superimposed layout of node is not able to cre-
ate a code which is capable of mitigating all threats in a network. In this paper, our
proposed framework has extraordinary adaptable feature which builds on cross-layer
module. The framework which is implemented on the basis of XLM is trust based
cross-layer module which employs the concept of trust in it to secure a sensor based
network. The protocol which is implemented based on these two frameworks is trust
based fuzzy implicit cross-layer protocol, and the enhancement of this protocol is
E-TRUFIX that holds a modified IEEE 802.11 Distributed Coordination Function
Media access Control and uses FLS to design a report mechanism to deliver easy
transmission mechanism. Among the simulation observations held, the assured per-
formance of E-TRUFIX and TRUFIX, FUGEF was compared in the presence of
malicious node and obtained better performance.
An Enhanced Trust Based Fuzzy Implicit Cross-Layer … 1017

2 Literature Survey

The protocols inspired by this XLM framework include cross-layer routing protocol
[4] which drags parameters from the layers of OSI model which involves physical,
MAC and transport layer to give a decision which determines that nodes are ready
to participate in a communication process. Energy-efficient beaconless geographic
routing protocol [5] provides a guaranteed loop-free delivery from source to sink
until the network is connected. It utilizes functionalities related to physical, MAC
and routing. EBGR is further extended to lossy sensor networks to deal with dynamic
topology. It is accessible only to attack free environments and suffers from energy
insufficiency problem. MACRO [6] integrates data layer and routing layer operations
in order to transfer packet to destination.
SIGF [7], a resource bound security solution, is based on IGF non-
deterministic/MAC transmission protocol which has no state and which permits to
manage dynamic topologies. It consists of three protocols and implements protection
by modifying their routing semantics. It keeps no state and routing information, but
it accomplishes an increase in PDR probabilistically. DWSIGF [8] routing proto-
col enhances performance on selection process in SIGF in order to select malicious
nodes by implementing collection window period which increases its window period
dynamically to create shift in time in protocol interpretations. This protocol uses data
link layer and routing layer functionalities. FUGEF [9] is implemented to select a
forwarding candidate node which eliminates substantial packet losses in network
and provides better security in network acquisition. It has low packet delivery ratio,
and spatio-temporal predictions are not possible. The FUGEF outdoes DWSIGF in
terms quality of service performance, energy consumption and overall performance
of security provision when subjected to black hole attacks. When subjected to Sybil
attacks, no results were shown as with DWSIGF and FUGEF as these are tuned only
to black hole attacks. It has low packet delivery ratio.

3 Proposed Method

The proposed protocol trust based fuzzy implicit cross-layer protocol is implemented
based on two frameworks:

1. XLM framework.
2. T-XLM framework.
1018 K. Anusha and A. Naveena

3.1 The XLM Framework

The XLM framework plots its parameters to the sensor protocol store to acquire its
combination. It pulls the minimum accessible resources to fend successful commu-
nication. The above process was acquired by the combination of the most built in
layer capabilities to a single component to fend necessities for efficient communi-
cation. In this XLM framework, it gives whole information to a node about when to
take part in the transmission mechanism. The communication method starts with a
initialization phase which involves a variable initiative denoted as Id is assigned to
1 if the neighboring node satisfies all the four conditions and 0 if elsewise as shown
in Eq. (1). The following conditions are identified by parameters which illustrate
the inherent capabilities of the protocol stack and which consists: relay packet level
λrelay , left over buffer capacity β, availability and survivability of node E rem and
signal-to-noise ratio ξrts .

(1)

The method of node selection in transmitting data is known as initiative determi-

nation. This method is initiated whenever a sender node broadcasts the RTS messages
to the neighboring nodes in available broadcast range. On receiving this messages,
the nodes are ready to take part in the communication process by sending a CTS
messages which include values defining the beginning constants like εRTS which
defines connection reliableness among the neighboring node and sender node, λrelay
the control of traffic which avoids congestion, β avoids barrier redundancy due to
uncontrollable passage and E rem the remaining energy of a node through span of
transmission. The only node which satisfies the above conditions is chosen as for-
warding candidate. This framework
 acquires less energy absorbed using congestion
control measures λrelay , β to avoid packet loss which causes to retransmission. This
subject is to have a reliable, disseminative, adjustable communication model.
An Enhanced Trust Based Fuzzy Implicit Cross-Layer … 1019

3.2 Trust Estimation Processes

Trust is defined as a credit which is generated from control theory suggested in the
field of E-commerce to choose dependable trade objects. Investigation progressed
by applying the notion into various realms using qualified policies to effectively
estimate trust through the transacting objects in Fig. 1.
The trust is evaluated from preceding originated proofs combined with reputation
transferred from participating objects within a network. The trust estimation process
is divided into four classes as shown in Fig. 2.
Probability based approach: This approach was based on two theories which
are Dempster–Shafer evidence theory and Bayesian probability theory [10–14]. This
process was furthermore divided into:
i. Objective estimation: It hangs on information break out of entities over
immediate attention.
ii. Subjective estimation: It reasonably switches on record of the proof supplied
directly or indirectly as gathered from entities.

Fig. 1 Typical trust model

Fig. 2 Trust estimation processes

1020 K. Anusha and A. Naveena

Outputs acquired after the evaluation stick exactly to a dual outcome.

Fuzzy based estimation approach: Evaluation of trust estimates is pertained
through the act of system so that the act can be graded perfect or faulty up to certain
extent.
Priority depended evaluation: This method is published by weighting the
dealings of sharing nodes through certain duration.
Miscellaneous approach: These are implemented using adopted process or
embolden concepts from scientifically and non-scientifically proven thesis.

3.3 The T-XLM Framework

The trust based XLM scheme is a prolonged edition of cross-layer module frame-
work. It exploits trust in choosing a data transferring node to forward packet toward
destination. The trust based cross-layer module theory TI is explained in Eq. 2, a
connection among the original resolution (I) and report (R).

TI = I ⊗ R (2)

The modified initiative resolution (I) is shown in Eq. 3

Th
relay relay
Th
positive if op op
Th
Th
rem rem
Th
rts rts
I= min Th
rts rts rts
min Th (3)
neutral if relay relay relay
Th max
op op op
min Th

min max
rem rem rem

Unsuitable if elsewise

I is stated in a congenital language that is easily derived. The parameters are

outlined as εRTS is the derived SNR value of the Request To Send telecast resolved
from derived SNR, ωrelay is defined as relay packet rate of a node deducted by the
holding interval of packets gathering the RTS telecast and βop is the barrier residency
interval, E rem is known as remaining energy of node.
An Enhanced Trust Based Fuzzy Implicit Cross-Layer … 1021

(4)

R determine value of the node report that is the modernize trust T value. Addi-
tionally, sr expresses strike rate to illustrate a nodes capability in transmission of
packet, fr is integrity rate to assure way modification
  and τ amounts the duration of
data transmission. The trust value of node n Tmn by node m inside its transmission
domain as detail of the transit stats (∅) and transit capacity (WÄ ) as shown in Eq. 5:

= f( ,ῼ) (5)

The traffic statistics and traffic volume are variables observed for a single leap
acquaintance, it is defined in Eqs. 6 and 7:

∅ = g(α1 , α2 , α3 , α4 , α5 , α6 ) (6)

Ω = h(∂1 , ∂2 ) (7)

α1 packets forwarded through m to n that are dripped by m

α2 entire no. of packs dripped over m
α3 packs dripped through m due to obstruction
α4 packs dripped through m overdue to undetermined details
α5 n’s imposition of m’s priority to m’s self-packet versus all other nodes Packets
α6 packets transferring waiting period by m
∂1 packs diverted through m
∂2 packs maliciously inserted through m.
1022 K. Anusha and A. Naveena

3.3.1 Routing Process in T-XLM

The routing process initializes with channel reservation phase in which the forward-
ing entity sends a request to send packet to the neighboring points to begin the
contending process. At this instant, invader resolves coherently to forward such a
broadcast message that helps the forwarding entities in packet transmission and also
affects the energy consumption among hosts aspiring to compete. Evaluating span
of the acquired request to send a telecast beacon and the total count of forwarded
packets aids to determine that the signal was reproduced through malignant system
or none.
The reception of RTS among nodes initiates to take part by replying off CTS signal
by attaching the variables εRTS , ωrelay , βop , T and E rem as a result of certain span of
time. At this instance of time, an invader fight toward election by speeding up in order
to make visible a similar response rates like less halting period, connection level and
anything as its reply. Suitable election in such times is done once clear to send replies
which are resolved utilizing border constraints, such as interval compared to level is
lesser than neglected or connection level upon some level is selected. The variables
implemented to weight a neighbor node are graded as positive, neutral, unsuitable
derived on rated responded through nodes.
A chosen sending host is permitted in order to continue to the subsequent stage,
wherein information is forwarded to it. On completion of transmission process, the
forwarding entity is dissected down the variables Sr, fr, τ and graded entrusted,
unclear and mistrusted. Assaults here involve wherein fewer malignant points deter-
mine to grip data drip each and every and few implicit volume of the information
afore transmitting to rear system accomplishment. This suggested framework must
be able of discovering and mollifying the result of strikes like blocking, black hole,
gray hole and Sybil. Due to the new threat involving more kinds of threats to a
network, our framework is not vulnerable to all security threats in a network.

3.4 The Proposed TRUFIX Protocol

The TRUFIX protocol excluded persuasions that hold trust and distrust as contradic-
tory offsets of scale. Here nodes congested behavior is marked for malicious behavior
and marked distrusted. The TRUFIX protocol that implements routing process in two
phases:
1. Channel allotment phase
2. Packet substitution phase
The FLS is used in both phases to route packets toward destination.
An Enhanced Trust Based Fuzzy Implicit Cross-Layer … 1023

3.4.1 The Channel Allotment Phase

In this process, the sender node S sets the NAV timer and on its expiry it perceives a
vacant channel for distributed inter-frame space-time duration and transmits an ORTS
signal containing the sender position and the destination position to neighboring hosts
inside its transmission extent. The transmission orbit contains transmitting contestant
nodes and halting contestant nodes. On taking message, the halting nodes abort their
replies by fixing their network allocation vector timer. On the other hand, setting
their CTS response time, and upon its termination, it reacts back with CTS replies.
The gathered response variables consist of forward range value (d) representing
the gap among the transmitting point and the sender node S, clear to send reply interval
(ω), that is a behavior of link excellence derived from node allocation interaction
to range and added holding interval because of inter-frame spacing and the original
trust rate that is fixed to 0.5. Each and every host reply is operated by the first fuzzy
logic system to choose the subsequent transmitting contestant node and rated them
as positive, neutral or unsuitable.

3.4.2 The Packet Substitution Phase

In this process, information is transmitted out of transmitting node S to the chosen

delivering contestant. After the finalization of data substitution, the sending node is
analyzed based on three parameters which contains; strike rate Sr that is a measure of
trust ability in packet transmission, data transfer period τ that measures the amount of
time held out of initiation of data swap interval to its accomplishment corresponding
to count of packs transferred inside that time duration and fairness ratio (fr) which
raises the dispersal of subsequent stage broadcast options between alike functioning
acquaintances and reached to FLS and rated as (trusted, distrusted, unsuited) and
returned as feedback to update the trust value by reputation R as shown in Fig. 3.
The mandatory fairness method permits every node to maintain a schedule of the
nodes that take part in the pathfinding method. A node selected in order to take part
within transferring method, if identified to have performed in the method before is
punished by decreasing its fairness ratio and data transfer period rates to guarantee
that the node is not chosen in the preceding steps. Identically malignant nodes that

Fig. 3 Proposed fuzzy logic design

1024 K. Anusha and A. Naveena

Fig. 4 Flowchart for routing process

have manipulated to work out in performing greater than once are also punished. The
whole transmission procedure is outlined in Fig. 4.

The Fuzzy Logic System (FLS)

The FLS is a computationally intelligent system which is used to perform human like
decisions that are simple and easy to understand. From Fig. 3, fuzzy logic system
manages three variables using 27 rules. In the first fuzzy logic system, the distance
calculated between the sender and chosen candidate, reply time and trust to select a
forwarding candidate node and rated as (positive, neutral and unsuitable). The FLS2
process the variables sr, fr and τ and graded as entrusted mistrusted and unclear.
The result generated is restored back as trust input to the first logic system which is
generated from the second fuzzy logic system, thus updating the preceding misprision
value 0.5 as original trust value for that detail node. The greatest chance value is
selected as the best transmitting candidate and the trusted sender node from Fig. 5.

3.5 The Extended E-TRUFIX Protocol

In this protocol, after the processing of the two fuzzy logic systems the packet trans-
mission takes place. In the packet transmission process, if a malicious node occurs in
the process, it takes an alternate route and delivers packet toward destination which
shows an improvement in the packet transfer rate. The total delay in the transmission
process is increased as it takes an alternate route to choose a candidate node in order
to forward packet toward destination.
An Enhanced Trust Based Fuzzy Implicit Cross-Layer … 1025

Fig. 5 Input membership functions for TRUFIX

4 Performance Evaluation

The performance of the RBSS based protocols existing protocol (FUGEF), TRU-
FIX and E-TRUFIX is implemented using NS2 simulation. The variables which are
evaluated in these protocols are packet transfer rate (PDR) and end-to-end delay.
From Fig. 6 among existing (FUGEF), TRUFIX and E-TRUFIX, the proposed pro-
tocol has achieved high packet delivery ratio and less possibility of attacker selection.
Thus, our proposed protocol outperforms all the other three protocols by mitigating
the security threats in a network.
From Fig. 6, the enhanced protocol E-TRUFIX achieved high packet delivery
ratio due to the node selection criteria by using forced fairness approach and FLS.
From Fig. 7, the E-TRUFIX has high delay due to the processing in fuzzy logic
system and to choose an alternate neighbor route which increases delay than other
protocols TRUFIX and existing protocol (FUGEF).
1026 K. Anusha and A. Naveena

Fig. 6 Packet delivery ratio

Fig. 7 End-to-end delay

An Enhanced Trust Based Fuzzy Implicit Cross-Layer … 1027

5 Conclusion

The proposed T-XLM based protocol lacks. So, in order to enhance security the
proposed protocol TRUFIX is implemented. The FUGEF protocol provides secu-
rity by changing their routing metrics to nodes by applying fuzzy logic for node
election. The proposed TRUFIX involves node election process using fuzzy logic
based trust evaluation, and mandatory fairness approach attains optimally suggestive
commutation between security and quality. The extension of this proposed protocol
is E-TRUFIX has shown an improvement in packet delivery ratio by providing an
alternate route and high possibility of attacker selection. In this process, the delay is
increased due to alternate route selection.

References

1. Karp B, Kung HT (2000) GPSR: greedy perimeter stateless routing for wireless networks.
In: Proceedings of the ACM 6th annual international conference on mobile computing and
networking 2000, pp 243–254
2. Füÿler H, Widmer J, Käsemann M, Mauve M, Hartenstein H (2003) Contention-based
forwarding for mobile ad hoc networks. Ad Hoc Netw 1(4):351–369
3. Son S, Blum B, He T, Stankovic J (2003) IGF: a state-free robust communication protocol for
wireless sensor networks. Tech. Rep., 2003, Department of Computer Science, University of
Virginia, Charlottesville, VA, USA
4. Vuran MC, Akyildiz IF (2010) XLP: a cross-layer protocol for efficient communication in
wireless sensor networks. IEEE Trans Mobile Comput 9(11):1578–1591
5. Zhang H, Shen H (2010) Energy efficient beaconless geographic routing in wireless sensor
networks. IEEE Trans Parallel Distrib Syst 21(6):881–896
6. Galluccio L, Leonardi A, Morabito G, Palazzo S (2007) A MAC/routing cross-layer approach
to geographic forwarding in wireless sensor networks. Ad Hoc Netw 5(6):872–884
7. Wood AD, Fang L, Stankovic JA, He T (2006) SIGF: a family of configurable, secure routing
protocols for wireless sensor networks. In: Proceedings of the 4th ACM workshop on security
of ad hoc sensor networks 2006, pp 35–48
8. Hanapi ZM, Ismail M, Jumari K, Mahdavi M (2009) Dynamic window secured implicit
geographic forwarding routing for wireless sensor network. In: Proceedings of the interna-
tional conference on wireless communication sensor networks. World Academy of Science,
Engineering and Technology 2009, pp 173–179
9. Umar IA, Hanapi ZM, Sali A, Zulkarnain ZA (2016) FuGeF: a resource bound secure
forwarding protocol for wireless sensor networks. Sensors 16(6):943
10. Quercia D, Hailes S, Capra L (2006) B-trust: Bayesian trust framework for pervasive computing.
In: Proceedings of the international conference on trust management 2006, pp 298–312
11. Zouridaki C, Mark BL, Hejmo M, Thomas RK (2009) E-hermes: a robust cooperative trust
establishment scheme for mobile ad hoc networks. Ad Hoc Netw 7(6):1156–1168
12. Zouridaki C, Mark BL, Hejmo M, Thomas RK (2005) A quantitative trust establishment frame-
work for reliable data packet delivery in MANETs. In: Proceedings of the 3rd ACM workshop
on security ad hoc and sensor networks 2005, pp 1–10
13. Shaikh RA, Jameel H, d’Auriol BH, Lee H, Lee S, Song Y-J (2009) Group-based trust man-
agement scheme for clustered wireless sensor networks. IEEE Trans Parallel Distrib Syst
20(11):1698–1712
14. Yao Z, Kim D, Doh Y (2006) Plus: parameterized and localized trust management scheme
for sensor networks security. In: Proceedings of the IEEE international conference on mobile
adhoc sensor systems (MASS), Oct 2006, pp 437–446
1028 K. Anusha and A. Naveena

Kompalli Anusha Pursuing [Link] in the department of WMC, G. Narayanamma Institute of

Technology and Sciences, under JNTUH, Hyderabad, Telangana, India.

Ambidi Naveena at present working as Assistant Professor in ETM Department , G.

Narayanamma Institute of Technology and Sciences, Hyderabad, She completed [Link] from G.
Narayanamma Institute of Technology and Sciences, Hyderabad. M.E from Osmania University,
Hyderabad. At present pursuing Ph.D from JNTUH. She has 12 years of teaching experience.
Implementation of Public-Key
Infrastructure for Smart Parking System
Using MQTT Protocol

Rajilal Manathala Vijayan, R. Ezhilarasie and A. Umamakeswari

Abstract Applications of IoT are endless, and it has been into many fields like
wearable devices, home appliances, etc. This paper investigates the smart parking
management application domain, in which the focus is on the users (user requesting
for the parking) privacy. Since the number of devices connected to the IoT network
grows exponentially, security of the user connected to IoT network is of paramount
importance. A scenario like hacking the confidentiality of high-profile users can
happen. Thus, this system proposes a solution to protect the identity of the user’s by
averting the exchange of private information by adapting the zero-knowledge proto-
col (ZKP) with elliptic curve cryptography (ECC) implementation. ECC, compared
with other public-key crypto algorithms, is the best choice for cryptographic imple-
mentation on resource-constrained devices. In this framework, the MQTT protocol
has been used to establish efficient communication between the user and parking
system.

Keywords IoT · MQTT · Security · ZKP · ECC

1 Introduction

IoT applications are spread into many fields like military, medical [1], health mon-
itoring, industrial, intelligent and smart transport system, smart homes, and smart
grids [2]. Smart city concept is now becoming possible with the advent of the Inter-
net of Things. One of the critical issues that needed to be taken care in smart cities
is car parking facilities [3, 4]. Car parking facility system issues include searching
an available parking spot and security issues related to user privacy. Many types of
research have been done on developing algorithms for web-based parking manage-
ment. But, little research has been done on the security issues related to unveiling the
identity of users during data transmission between the user and parking system. The
hiding of user’s location privacy is of much importance as discussed in [5]. Sensitive

R. M. Vijayan (B) · R. Ezhilarasie · A. Umamakeswari

School of Computing, SASTRA Deemed University, Thanjavur 613401, India
e-mail: rajilalmv@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1029

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1030 R. M. Vijayan et al.

information like the privacy of high-profile users, for example, where does the person
spend the majority of the time, an address of residence, user with disabilities, etc.,
needs to be secured. The proposed system provides a novel solution that is smart
enough to preserve the privacy of the user.
The proposed smart parking application provides security and privacy to the users
by averting the interchange of vital information. The user’s privacy is protected by
adopting ZKP over ECC.
ECC, a public-key cryptography (PKC), provides an adequate level of security
compared with other PKC algorithms.
ZKP is a technique where the prover proves the apprehension of a secret, without
revealing the confidential information to verifier [6]. Schnorr’s interactive protocol,
a type of ZKP, is used in the proposed system for implementing zero-knowledge
algorithm.
While considering the data transmission between systems, the established com-
munication protocols which are used on the internet cannot be directly used in
IoT network. Because certain things like resource-constrained environments of con-
trollers, unreliable networks, security, etc., need to be considered. The proposed smart
parking system application used the message queuing telemetry transport (MQTT),
an IoT communication protocol, between the user and parking system. MQTT, an
IoT application protocol, is a lightweight, open source, simple, and easy to imple-
ment publish/subscribe protocol suitable for resource-constrained environments and
applicable in the Internet of Things (IoT) context.
The leftover sections in the paper are structured as Sect. 2 talks about the literature
survey done on smart parking systems. Section 3 describes preliminaries required
for implementation. Section 4 presents the formulated work. Section 5 explains
experimentation and validation done on the proposed system.

2 Related Works

While investigating the smart parking management system, the papers listed mainly
concentrate on either directing the users to the free or unparked slots or its automation
[7, 8]. In [9], a smart parking system is proposed, in which users can view, choose,
and reserve an available parking slot before reaching the parking area. But, a little
research has been done on the user’s privacy. In [10], user security is discussed, in
which security research is limited to network security. But, no study has been done
on how security protocols can be implemented on resource-constrained devices.
In [11], constrained application protocol (CoAP) protocol is used as a commu-
nication protocol for resource-constrained devices. User datagram protocol (UDP)-
based CoAP has less overhead than the transmission control protocol (TCP)-based
MQTT. However, due to the lack of TCP retransmission mechanisms, the chance of
packet loss is more when using CoAP. Hence, MQTT protocol is preferred for com-
munication between systems over CoAP, as CoAP needs to re-transmit the whole
Implementation of Public-Key Infrastructure for Smart … 1031

message more often than MQTT does, with a high packet loss, which leads to better
performance under these conditions for MQTT.

3 Preliminaries of the Work

3.1 ECC

Security is one of the main concerns in the IoT paradigm. Even though MQTT
provides the network layer security, the proposed smart parking system application
describes the security at application level. The idea behind the smart parking system
application is to protect the identity of the users by not revealing any information
during the communication between the user and parking system. The advantages
and disadvantages of existing internet protocol (IP)-based security, in the context
of the IoT paradigm, are examined in [12]. The research portrays that PKC is more
appropriate for IoT networks.
ECC replaces RSA on environments, having memory and computation limita-
tions [13]. A 160-bit key in an ECC cryptosystem provides an identical amount of
security to that of a 1024-bit key in a conventional cryptosystem as proposed by
National Institute of Standards and Technology (NIST). These reasons make ECC
suitable for constrained environment implementations as it saves memory space and
computational time, and consequently reduces energy requirements.

3.2 ZKP

Implementation of ZKP demonstrates how the privacy of the user can be protected
by adapting zero-knowledge proofs based on ECC. ZKP consists of two operators, a
prover and a verifier. This proof helps the prover to prove the knowledge of a secret
without revealing any vital information [6]. Completeness and soundness are the
two properties of zero-knowledge proofs. Complete property describes that a proof
is complete if the protocol triumphs with overwhelming probability and if an honest
prover and verifier are given. Sound property describes that a proof is sound, if the
dishonest prover has negligible probability to complete the proof [14].

4 Proposed System

Little research has been done in the area of maintaining privacy and identity of the
users in the smart parking system application. User privacy can be secured either by
securing the data that is meant for communication or by maintaining the anonymity
1032 R. M. Vijayan et al.

of the user itself. The first method of maintaining user privacy is vulnerable to
attack from third parties as the data transmission happens between systems. This
paper approached the second method, where a novel solution is proposed to maintain
the anonymity of the user by totally averting the exchange of private information.
The proposed solution uses ZKP over ECC for establishing the protection of user’s
privacy. MQTT protocol is used for establishing communication between the devices
as this protocol is the best application protocol in the IoT paradigm.

4.1 Algorithm

Schnorr’s interactive protocol depends upon the elliptic curve discrete logarithm
problem (ECDLP), is used to implement the ZKP algorithm.
The following are the Schnorr’s protocol steps based on an elliptic curve: Prover
and verifier consent on an elliptic curve E, over a prime field F n , and a generator
point G ∈ E/F n . Prover and verifier know M ∈ E/F n and former needs to prove that
he/she knows y such that M = y · G to later without revealing y.

4.2 Schnorr’s Protocol Steps

1. User calculates the point N = r · G, where r is a random value r ∈ F n.

2. Point N is sent to parking system by the user.
3. Parking system computes random d = HASH (G, M, N) and sends d to user.
4. User computes x = r + d · y (mod n) and sends m to parking system.
5. Parking system checks that Z = x · G − d · M = (r + d · y) · G − d · M = r · G
+ d · y · G − d · y · G = r · G = N.

The implementation of ECC in the work follows the library in [15]. A NIST-
approved elliptic curve [16] over prime fields having equation y 2 = x 3 + a · x + b
(mod p) is used in this work. Order of the curve (r) and the generator point G(x, y)
are given in Table 1. Secure hash algorithm-2 (SHA-2) is used [17] for computing
the random value “c.”

Table 1 Elliptic curve implementation variables

Variable Value
r FFFFFFFF FFFFFFFF FFFFFFFF FFFFFFFE BAAEDCE6 AF48A03B
BFD25E8C D0364141
x 79be667ef9dcbbac55a06295ce870b07029bfcdb2dce28d959f2815b16f81798
y 483ada7726a3c4655da4fbfc0e1108a8fd17b448a68554199c47d08ffb10d4b8
Implementation of Public-Key Infrastructure for Smart … 1033

5 Experimentation and Validation

The proposed smart parking management system consists of two systems. The
approaching vehicles can be termed as prover, and parking system can be termed
as the verifier. The terms prover and verifier are with respect to the ZKP. In this
work, only one verifier and one prover are implemented. To validate the work, a
system is designed (Fig. 1) using Raspberry Pi 3 model B, along with force sensor
(FSR402) and nodeMCU (ESP8266). The force sensor detects the presence of the
vehicle and passes the information to the parking system. After this, parking sys-
tem starts processing the requests from the vehicle. Data flow representation of the
system is shown in Fig. 2.
The prover and verifier communication is established using the MQTT protocol.
Raspberry Pi controller is used for programming in both sides. The Raspberry Pi
at the verifier side serves as the server with MQTT broker. Since the interactive
ZKP requires multiple communications between prover and verifier, both prover and
verifier serve as both MQTT publisher and MQTT subscriber. After this step, prover
subscribes a computed random value from the verifier. And based on the subscribed
random value, prover computes a value and published the value to the verifier.
In the experiment, free open-source Mosquito broker and Paho Python MQTT
client are used for doing programming in Raspberry Pi.

Fig. 1 The system designed using Raspberry Pi along with nodeMCU and force sensor

Process Description
1 Force sensor detects the presence of vehicle
2 Force sensor passes the vehicle presence information to Node MCU
3 NODE MCU publishes the vehicle topic to the parking system(Raspberry Pi)
4 Parking system starts communication to user(Raspberry PI)
5 User responds to parking system

Fig. 2 Data flow representation of the proposed system

1034 R. M. Vijayan et al.

5.1 Prover System

The prover wants to prove that he/she is the authorized user approaching
for the parking slot. Prover can be connected to the MQTT server by using
[Link]().connect(mqtt_server_ip, Port_no,keepalive). The parameter keepalive
describes the maximum portion of time in seconds that is allowed between commu-
nications with the broker. publish(“Topicname”, data), can be used to publish data,
to which topic name and data are passed. [Link]().subscribe(“Topic_name”) is
used to subscribe data from a topic on the broker.

5.2 Verifier System

The verifier verifies whether the approaching user is a valid user or not. Verifier is
programmed to subscribe data from only one topic. If the verifier is programmed to
set separate topics for the receiver, the computation process and code size will be
more. So, a single topic is maintained.

5.3 Cheater System

A cheater can cheat or act unfairly toward the system if he/she could find the value
of Z = x · G − d · M, before the arrival of verifier’s hash value, d. Otherwise, the
cheater will not be able to act upon the system.
Below are the simulation snippets.
Step 1: Prover (user) sends the computed value to the verifier, as shown in Fig. 3.
Step 2: Verifier (parking system) receives the key from prover and computes the hash
value and sends back to prover, as shown in Fig. 4.

Fig. 3 Prover sends the

computed value
Implementation of Public-Key Infrastructure for Smart … 1035

Fig. 4 Verifier sends the

computed hash value

Step 3: Prover receives the hash value and computes the hash value and sends it to
the verifier, as shown in Fig. 5.
Step 4: Verifier receives the value and verifies it, as shown in Fig. 6.
Hence, the verifier verifies the proof from the prover and validates the prover.

Fig. 5 Prover sends the

computed value

Fig. 6 Verifier validated the

prover
1036 R. M. Vijayan et al.

6 Conclusions and Future Work

Maintaining the privacy of the user in the IoT paradigm is of uppermost importance.
Preserving the anonymity of the high-profile users who are approaching the parking
system is of high importance. An algorithm, based on ZKP over ECC, is proposed
to protect the anonymity of the user identity. ECC has the advantage of providing
equivalent security with lesser key size when compared with other public-key cryp-
tographic algorithms (e.g., RSA). The future works include the establishment of an
algorithm which offers less proof size than the interactive ZKP used which will be
helpful in the resource-constrained environments.

Acknowledgements The authors are grateful to the Department of Science and Technology,
New Delhi, India (SR/FST/ETI-371/2014). They also wish to acknowledge SASTRA University,
Thanjavur for extending infrastructural support to carry out the work.

References

1. Khan R, Khan SU, Zaheer R, Khan S (2012) Future internet: the internet of things architecture,
possible applications and key challenges. In: 10th international conference on frontiers of
information technology
2. Bhat SM, Ahmed S, Internet of things: an insight into emerging applications and architectures,
national conference on recent advances in computer science and IT (NCRACIT). Int J Sci Res
Comput Sci Eng Inf Technol 4(1). ISSN: 2456-3307
3. Zhang Z-K, Cho MCY, Wang C-W, Hsu C-W, Chen C-K, Shieh S (2014) IoT security: ongoing
challenges and research opportunities. In: 2014 IEEE 7th international conference on service-
oriented computing and applications
4. Zhou F, Li Q, Parking guidance system based on ZigBee and geomagnetic sensor technology.
In: 2014 13th international symposium distributed computing and applications to business,
engineering and science (DCABES), pp 268–271
5. Krumm J (2009) A survey of computational location privacy. Pers Ubiquitous Comput
13(6):391–399
6. Menezes AJ, Vanstone SA, Oorschot PCV (1996) Handbook of applied cryptography. CRC
Press Inc, Boca Raton, FL
7. Wang H, He W (2011) A reservation-based smart parking system. In: 2011 IEEE conference
on computer communications workshops (INFOCOM WKSHPS). IEEE, pp 690–695
8. Pala Z, Inanc N (2007) Smart parking applications using RFID technology. In: RFID Eurasia,
2007 1st annual, pp 1–3. [Link]
9. Mutiara GA, Agung AAG, Handayani R (2018) Implementation of smart parking system with
real time monitoring. Far East J Electron Commun 8(2):277–290
10. Alqazzaz A, Alrashdi I, Aloufi1 E, Zohdy M, Ming H (2018) SecSPS: a secure and privacy-
preserving framework for smart parking systems. J Inf Secur 9:299–314
11. Li W, Jung C, Park J (2018) IoT healthcare communication system for IEEE 11073 PHD and
IHE PCD-01 integration using CoAP. KSII Trans Internet Inf Syst 12(4)
12. Nguyen KT, Laurent M, Oualha N (2015) Survey on secure communication protocols for the
internet of things. Ad Hoc Netw 32(C):17–31. [Link]
13. Gura N, Patel A, Wander A, Eberle H, Shantz SC (2004) Comparing elliptic curve cryptography
and RSA on 8-bit cpus. In: Cryptographic hardware and embedded systems-CHES 2004.
Springer, pp 119–132
Implementation of Public-Key Infrastructure for Smart … 1037

14. Almuhammadi S, Sui NT, McLeod D (2004) Better privacy and security in e-commerce: using
elliptic curve-based zero-knowledge proofs. In: CEC, pp 299–302
15. Malan DJ, Welsh M, Smith MD (2008) Implementing public-key infrastructure for sensor
networks. TOSN 4(4)
16. Certicom research (2010) Sec 2—recommended elliptic curve domain parameters. [Link]
[Link]/[Link]
17. Eastlake D, Hansen T, US secure hash algorithms (SHA and HMAC-SHA). [Link]
org/html/rfc4634
Non-bipolar Evaluation
and Visualization of Online Text Reviews

Keerthana Chigateri and Rekha Bhandarkar

Abstract Research in the field of Sentiment analysis is blowing day by day expo-
nentially in the recent past and currently it is the much-acknowledged discipline.
This is due to the outpouring users of internet and thus generated effusion of data in
the form of reviews, comments, blogs, communications, etc. These are treasure trove
of information needed to comprehend the fact-based opinions of diverse consumers.
Elucidating those contents is propitious to various stakeholders as right sentiments
can be gathered through it. But trading with such ginormous unstructured inputs
embodies diverse challenges. These challenges trigger the raise in desideratum and
thus summon the need to instigate pioneering logics to meet the same and to opti-
mize the existing approaches. Sentiment analysis and opinion mining terms are used
equivalently. This existed from ancient days. But the modes employed to perform
this were different like in the form of surveys, elections, etc. It was carried out either
by individuals or a group or an organization. Individuals used to consult friends or
family before changing themselves from strangers to customers of any product or
audience of any events. Groups or Organizations used to conduct surveys. But gradu-
ally most of the decisions were to be data-driven and thus appropriate decisions were
to be made in short-run to bridge the gap between businesses and consumers and to
gain the competitive advantage in market. So now this process was to be automated
for mutual benefits by exploiting the escalation in technology. Various approaches
are emerging every day to strengthen and ease the process. In the same direction, this
paper addresses how sentiment analysis of online reviews can serve as suggestions
to strangers and help them better in gaining additional insights to make well-judged
decisions or choice by visualizing the analysis in the graph form.

Keywords Reviews · Sentiment · Opinion · Suggestive · Non-bipolar

K. Chigateri (B)
Department of CSE, N.M.A.M.I.T, Nitte, India
e-mail: keerthanabc@[Link]
R. Bhandarkar
Department of ECE, N.M.A.M.I.T, Nitte, India

© Springer Nature Singapore Pte Ltd. 2020 1039

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1040 K. Chigateri and R. Bhandarkar

1 Introduction

In ancient times the amount of data available was very less. This was due to various
reasons like the absence of smart gadgets, or access to internet by very less people.
Initially many people were associated with one basic communicative device with
no other option. It was many-to-one association. So the amount, quality and kind of
data generated was very less. This data was productive only for telephone industry.
Gradually phones with more options other than just communications like camera
surfaced the markets. Along with this, association also increased from many-to-one
to one-to-one. Thus the data and the kinds of data generated was more than before.
Analysis of these data was not very challenging as data was of less variety and infre-
quently produced compared to present-day environment. Later smartphones slowly
flooded the market which supported various features. Along with smartphones, smart
devices also floated into the market. So the scenario again advanced from one-to-one
association to one-to-many association. Increase in smart devices and their associa-
tions with humans not only eased the lifestyle but also led to increase frequency and
the kinds of data generated enormously [1]. Now life is almost irrevocable without
these smart devices. There is almost no field untouched by this smart technology.
Every domain generates its own kind of data different from other domains [2]. So the
approach of collecting, processing, extracting, interpretation of data differs to certain
degree. These steps need to be carried out regularly, swiftly and meticulously. Thus
lot of data is made available on everyday basis which is highly concentrated with
valuable information, analysis of which would mitigate the gap between public and
service providers effectively by targeting right style of customers at the right time
for the right product based on the demographic factors. This yields tangible benefits
wholly. To accomplish this, analysis system demands augmentation in techniques
continuously. Either the good tool or good data should be available. Having both is
almost impossible. So first is recommended solution as latter is less attainable. The
analysis needs to get more precise and relevant eliminating the redundancies. More
the relevance less is the data entered, collected. Thus processing, storage required
also is less and speed increases thereby improving efficiency. The current conven-
tional review system allows the user to review the entire product. This is very abstract
as the information provided by most of such reviews is insufficient to make decisions
as they do not spot fundamental criteria as of the product. Researchers are focusing
and exploring this area to address these limitations. This paper also contributes to the
same by analyzing only those reviews targeting the duration of working of hard-disk
in non-bipolar way in terms of days, months and year unlike only positive, negative
and neutral analysis of the whole reviews.
Non-bipolar Evaluation and Visualization … 1041

2 State of the Art

Technology and so our lives are flourishing parallelly. Ergo presently, internet and
Smart gadgets are the indispensable part of today’s lifestyle of any person through
various surveys as they are available within the financial means of people. This
has not only led to very effortless communication but also equally in verbalizing
of individual’s panorama. Research needs to go beyond the current binary (positive
and negative) or ternary (positive, negative and neutral) analysis of these panoramas.
Reviews along with being positive, negative, neutral or sarcastic, they can also serve
as suggestions to others [3]. Such suggestive reviews need to be identified, extracted
and analyzed [4]. The largeness of data does not matter but the richness of data matters
very much. So analysis of this data is much acknowledged across the globe as it yields
many tangible benefits. Initially, this analysis was carried out at the document level
usually for blogs to categorize blogs into binary or ternary classes. Gradually this
analysis was narrowed down to Sentence level like reviews [5]. Later it was further
narrowed down to entity-level like a particular product is positive/negative/neutral.
Finally, it’s now attenuated to aspect level where a particular feature of a particular
entity is positive/negative/neutral [6]. Figure 1 shows this hierarchy of sentiment
analysis. Figure 2 represents these five elements of an opinion. Figure 3 shows
the block diagram of the approach. The outcomes of the present Bipolar and tri-
polar assessment of data is two labels positive or negative and three labels positive,
negative or neutral, respectively. But a review can be positive for some aspects,
negative for certain features and neutral for few factors [7, 8]. Thus a single review
can be positive, negative and neutral at the same time. So concluding it with only one
polarity is incorrect. But non-bipolar assessment of data determines multiple labels
which connects more to the strangers [9]. It also helps the trade to understand the

Fig. 1 Hierarchy of
sentiment analysis
1042 K. Chigateri and R. Bhandarkar

demographics. It is observed through research that changing the colour of the button
increases the conversion by 15% [10]. Comprehending the sentiments of consumers
to a greater depth and more precisely is very important to ameliorate the customer
satisfaction through which business can lead in the competitive market [11]. Such
analysis refrains vague services and enables grasping and focusing on the prime
preferences of consumers. Thus the resources and time are used effectively and
wastage of the same is reduced and productivity is boosted. This paper addresses the
analysis of text reviews which is an extension to binary analysis. The emphasis of
binary analysis is on only one dimension, i.e., direction-positive/negative. Besides
this, there are other four more dimensions of an opinion which needs to be analyzed
[12] viz.,
– Intensity: It unfolds the severity or extremity of buyer’s viewpoint. It is usually
expressed using certain adjectives. For example, worst is more intense or severe
form of negativity. Excellent is the extreme form of positivity.
– Stability: based on the context, it conveys various factors like the firmness in
services, durability of the product, reliability on a service provider, etc.,
– Informational: it contributes additional details about prime aspect based on which
the direction of the opinion can alter.
– Social Support: It enhances the quality of business by giving broader focus on
one’s particular service/brand/product, etc. It by-product is positive image and
well-being in the marketplace.
These other dimensions serve as suggestive to strangers who can choose easily to
change into customers or not on viewing this diagrammatic analysis [13].

3 System Design

The main objectives of this research are to design an algorithm to extract only
informative data from large repository and to develop optimized techniques for
pre-processing the data extracted from the data warehouse. An algorithm to effi-
ciently identify different sentiments if present and to identify and classify the dif-
ferent dimensions of the sentiment needs to be established. This paper works on the
dimension, i.e., Stability. Similarly, the other four dimensions viz, Direction, Inten-
sity, Social Support, Information, needs to be worked upon. The meaning of stability
differs from context to context. Here, stability means the length of duration or the
time period during which its working or functioning is considered to be normal or
as expected. Rest of the cases are considered to be abnormal.
Non-bipolar Evaluation and Visualization … 1043

4 System Implementation

Here dataset containing 443 reviews on hard disk is considered. In this set, only those
reviews are targeted which are expressing about its functioning. It is very laborious
and time-consuming to identify only relevant reviews when the number of reviews
is very huge. Figures 4 and 5 represents the frequency of working and not working
of hard disk, respectively. The Y-axis represents duration in terms of days, months
and years and the X-axis represents the number of people speaking for each case
in both figures. This representation is very comprehensive, effortless and less time-
consuming in making decision and benefits mutually for both service providers to
improve services or to identify loopholes and thus bridge the gap between them,
connect effectively and customers to decide quickly. This analysis is beyond binary
or ternary analysis. Hence the name non-bipolar. Following is the algorithm for
analyzing the reviews speaking about the working and not working of the hard disk
in terms of days, months and year. The frequency of reviews about its working is
considered as positive and the frequency of reviews about it not working is considered
as negative. The keywords found for were 2 days, 2 months, 8 days, 15 days, 10 days,
1 week, 3 years, 1 month, 1 day, 1 year, 20 days, 2 weeks, 8 years, 3 months,
7 months, 5 months, 2 years, 3 days, 3 weeks, 12 days. If the term “working” or any
other positive terms (in dictionary) appeared within the window size less than five
terms after days/months/years, it was listed under working. Similarly, the term “not
working” or any other negative terms (in dictionary) appeared within the window
size less than five terms after days/months/years, it was listed under not working.
This window size can be varied. For example, in the review “I got these product
8 days before but the hard-disk is unable to detect”. Here the stop words like the, but,
etc., were remove while pre-processing. The term “unable” appeared within window
size <=5 and it is in the dictionary of negative terms. Thus, it is classified under
frequency of non-working reviews. The same explanation holds good for positive
reviews classification.
Every opinion can contain five main elements:

{E 2 , Fab , Oabcd , Hc , Td }

where
Ea Entity a
F ab Feature b of Entity a
Oabcd Opinion of entity a for feature b by opinion holder H at the time d.
Hc Opinion holder c
Td at the Time d.
Along with these five elements [5], every opinion should contain a target. An
opinion without a target is not substantial for analysis and thus its lifespan is very
short. Figure 2 represents these five elements of an opinion.
Step 1: Read the excel file and store it in the form of a DataFrame object
1044 K. Chigateri and R. Bhandarkar

Fig. 2 Five elements of an

opinion

Step 2: Calculate the shape of the DataFrame

Step 3: Read the reviews, break it down into individual reviews and store it as an
individual element in a list
Step 4: Make lists of factors to review the comments and two lists for all the positive
and negative keywords for the keyword search
Step 5: Set the counters for all the review factors as zero
Step 6: Create a dictionary to make a key-value pair of the keywords and its occurrence
counters
Step 7: Search for the key in the list and match it with the required reviews and store
it in the dictionary along with its count
Step 8: For plotting the graph use the matplotlib library
Step 9: Append all the required labels in one list, calculate the required rows and
columns and plot the bar graph
Step 10: Repeat the steps for more sections of the analysis (Figs. 3, 4 and 5).

5 Conclusion and Future Enhancements

In this paper, non-bipolar assessment of opinions of text input processing. The same
idea can be made domain-independent in future. It turns out to be the prerequisites
for the companies in the current and future scenarios for the business to prosper.
By adopting this, they can enjoy the tangible benefits like grasping the issues and
acknowledge the trends and competitors influencing the society, embrace fresh cus-
tomer visions, optimize the strategies, and meet the warnings in the preliminary
stages to build an enduring system and services.
Non-bipolar Evaluation and Visualization … 1045

Fig. 3 Block diagram of the

methodology

Fig. 4 Graph of reviews on working

1046 K. Chigateri and R. Bhandarkar

Fig. 5 Graph of reviews on not working

References

1. Shayaa S et al (2018) Sentiment analysis of big data: methods, applications, and open
challenges, vol 6. IEEE
2. Mäntylä MV, Graziotin D, Kuutila M (2018) The evolution of sentiment analysis—a review of
research topics, venues, and top cited papers. Comput Sci Rev 27:16–32. ISSN 1574-0137
3. Qazi A, Raj RG, Hardaker G, Standing C (2016) A systematic literature review on opinion
types and sentiment analysis techniques: tasks and challenges. Internet Res 27(3)
4. Shayaa S, Jaafar NI, Bahri S, Sulaiman A, Wai PS, Chung YW, Piprani AZ, Al-Garadi MA
(2018) Sentiment analysis of big data: methods. Appl Open Challenges. IEEE
5. Vo A-D, Nguyen Q-P, Ock C-Y (2018) Opinion–aspect relations in cognizing customer feelings
via reviews, vol 6. IEEE Access
6. [Link]
7. Songpan W (2017) The analysis and prediction of customer review rating using opinion mining.
IEEE SERA 2017(June):7–9
8. Al-Smadi M, Al-Ayyoub M, Jararweh Y, Qawasmeh O (2019) Enhancing aspect-based senti-
ment analysis of Arabic Hotels’ reviews using morphological, syntactic and semantic features.
Inf Process Manag, vol 56(2):308–319
9. Bouazizi M, Ohtsuki T (2017) A pattern-based approach for multi-class sentiment analysis in
Twitter, vol 5. IEEE Access
10. [Link]
11. Qazi A, Raj RG, Tahir M, Cambria E, Syed KBS, Enhancing business intelligence by means
of suggestive reviews. Sci World J 2014
12. [Link]
13. [Link]
2015. Available [Accessed 1 Dec 2018]
Status Monitoring System-Based Defense
Mechanism (SMS-BDM) for Preventing
Co-resident DoS Attacks in Cloud
Environment

S. Rethishkumar and R. Vijayakumar

Abstract Co-occupant DoS assaults are discovered as most defenseless dangers in

relation to distributed computing which is an asset and generally is obligatory in
nature. Co-inhabitant DoS assaults would deplete out the cloud assets which impair
the certifiable cloud clients from executing cloud utilization. Consequently, aversion
and conclusion of the event related to co-inhabitant assaults result in fundamental
assignment of cloud form. In our past research strategy, two-player game approach
(TPGA) is acquainted which points that show evasion of co-inhabitant DoS assaults
by learning and arranging the virtual machine requested from clients on account
of the low, medium, and high hazard demands. Anyway, this strategy decreased
in its execution regarding discovery of hazard dimension of client ask for VMs.
There is no predefined method for recognizing hazard estimation of VM assets. And
furthermore, it would be troublesome to keep up and refresh the hazard status data
of the cloud assets. This is settled in the proposed research strategy by presenting
the system called status monitoring system-based defense mechanism (SMS-BDM)
or state observation-based co-resident DoS attack detection (SO-CRDoS-AD). In
the proposed research technique, at first, hazard estimation of each asset dependent
on client demands is assessed by utilizing hazard recognition metric. In light of
this hazard metric esteem, state estimation of each client asks for as far as VM is
refreshed. The states that are considered in this work are security state, vulnerability
state, attacked state, positive state, negative state, degenerate state, and failure state.
These state estimations of VM assets are refreshed occasionally with the assistance
of Markov chain display. This examination technique is actualized in the CloudSim
condition from which it is demonstrated that the proposed research strategy can
guarantee the exact recognition of assault status of VM asks for; therefore, the security
level can be upgraded.

S. Rethishkumar (B) · R. Vijayakumar

School of Computer Sciences, Mahatma Gandhi University, Kottayam, Kerala, India
e-mail: rethishsnair3@[Link]
R. Vijayakumar
e-mail: vijayakumar@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1047

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1048 S. Rethishkumar and R. Vijayakumar

Keywords Cloud security · Defense mechanism · Status monitoring or state

observation · Attack state · Risk level of requests

1 Introduction

Virtualization has turned into an appealing study in the present distributed computing
condition [1]. The capacity to share the procedure are forms of assets of a solitary
physical machine server between a few detached virtual machines (VM) empowering
a more streamlined equipment use, and in addition, the less demanding administra-
tion and movement of a virtual framework contrasted with its physical partner have
offered to ascend to new security policies and VM portion designs [2]. Specifically,
virtualization procedures are a key component in distributed computing [3].
In cloud platform, the security issues are related to virtual machine allocation,
while distributed computing gives numerous points of interest in availability, versa-
tility, and cost proficiency, and it additionally presents various new security dangers
[4]. This paper focuses on the co-resident hit, where malicious users are targeted to
genuine VMs through co-locate VM from the same server and then provide side-
channel attacks to the targeted VMs and taken private data from genuine VMs [5].
Majority of the previous works are connected with avoiding of the side channels [6].
Be that as it may, sometimes are unrealistic for the current business cloud platform
[7].
We approach the issue from an alternate point of view, and concentrate on how
to limit the assailant’s probability of co-finding their VMs with the objectives, while
keeping up a palatable remaining task at hand at par and one with low power uti-
lization for the framework [8]. Our answer does not require any progressions to
the fundamental foundation. Subsequently, it tends to be effortlessly actualized in
existing distributed computing stages [9].
VM allocation policies against co-resistance attacks in cloud infrastructure are
based on the elimination of malignant users from VMs for reducing side-channel
attacks and lead to secure the VMs from co-resident attacks [10]. Previous works
focused that malignant user targeted a VM called victim VM who is utilizing the
side channels [11].
The primary commitment of this exploration work is to present the anchored
system for keeping the co-occupant DoS assaults occurring on the cloud assets;
subsequently, the ideal asset dealing with can be guaranteed. The principal objective
of this examination strategy is to arrange the required asset for the clients ask for
according to their necessities and avert pointless use of cloud assets from wastage.
Status Monitoring System-Based Defense Mechanism (SMS-BDM) … 1049

2 Related Works

VM level assurance is pivotal in a virtualized or distributed computing condition. By

making a security edge around each VM along these lines, the undertaking can co-find
applications with various trust levels on a similar host and can shield VMs in a com-
mon, multi-inhabitant condition. In this area, distinctive VM security calculations
have been talked about in detail.
Yinqian Zhang et al. depicted about “Home Alone: Co-Residency Detection in
the Cloud by means of Side-Channel Analysis” [12]. The abode itself is a framework
that gives an occupant a chance to confirm its VMs’ select utilization of a current
machine.
Adam Bates et al. clarified about “Identifying Co-Residency with Active Traffic
Analysis Techniques” [13]. Co-occupant watermarking, a traffic investigation assault
that permits a malevolent co-inhabitant VM to infuse a watermark signature into the
system stream of an objective example. Thus, our methodology is hard to protect
without expensive underutilization of the physical machine.
Yi Han et al. depicted about “Security Games for Virtual Machine Allocation
in Cloud Computing” [14]. It focuses on the co-occupant assault, where pernicious
clients expect to co-find their virtual machines (VMs) with target VMs on the equiv-
alent physical server, and afterward, abuse side channels to remove private data from
the person in question.
Rodrigo N. Calheiros et al. clarified about “CloudSim: a toolbox for displaying
and recreation of distributed computing situations and assessment of asset provi-
sioning calculations” [15]. The ongoing endeavors to plan and create cloud advance-
ments center around characterizing novel strategies, arrangements, and instruments
for effectively overseeing cloud foundations.

3 Status Monitoring System-Based Defense Mechanism

(SMS-BDM) or State Observation-Based Co-resident
DoS attack detection Model

The proposed research strategy is presenting the procedure called status monitor-
ing system-based defense mechanism (SMS-BDM) or state observation-based co-
resident DoS attack detection (SO-CRDoS-AD). In the proposed research technique,
at first, hazard estimation of each asset dependent on client demands is assessed by
utilizing hazard discovery metric. In light of this hazard metric esteem, state estima-
tion of each client asks for as far as VM is refreshed. The states that are considered in
this work are security state, vulnerability state, attacked state, positive state, negative
state, degenerate state, and failure state. These state estimations of VM assets are
refreshed intermittently with the assistance of Markov chain demonstrated work.
1050 S. Rethishkumar and R. Vijayakumar

3.1 Problem Definition

Think about the accompanying situation: in a distributed computing arrangement of:

K servers S = {s1, s2, …, sK},
M clients U = {u1, u2, …, uM} & begin
N virtual machines V = {v1, v2, …, vN}.
A mapping X: U × V → S, apportions each VM from every client to an explicit
server, XV × S×U = {xv, s, u | xv, s, u = 1 iff VM v of client u is allotted to server
s}. An assailant A 
means to co-find their VMs with the VMs of legitimate client L,
i.e., Target (A) = t VM (L, t). Amid, one assault began at time t, A dispatches
|VM (A, t)| VMs, and the objective is to expand the proficiency and additionally
inclusion rate. Given this assault situation, our new arrangement ought to fulfill the
accompanying destinations:
1. Security—Under the new strategy, the aggressor needs to begin countless to
accomplish a nonzero productivity or inclusion rate; i.e., VMmin is high. Likewise,
the inclusion rate does not increment or increments gradually with |VM (A, t)|.
So as to accomplish these two, one outstandingly prominent case is that VMs of
various clients are never allotted to a similar server. In view of such a thought,
we limit the normal number of clients per server, i.e.,
S: min 〖1/K − (s ∈ S)[|u|u ∈ U, x_(v, s, u) = 1, ∀v ∈ V |]〗
2. Remaining burden balance—As we referenced in the presentation, the signifi-
cance of outstanding task at hand balance is twofold. For cloud suppliers, equally
disseminating VMs helps decline the likelihood of servers being over-used. For
straightforwardness, in our new arrangement, we utilize the quantity of running
VMs per server as the rule to spread the outstanding task at hand (equivalent to
the least VM strategy). What is more, clients would likewise lean toward if their
VMs are not all distributed together on a similar server. As it were, overall, the
quantity of servers that have a client’s VMs ought to be amplified, i.e.,
W: max 〖1/M − (u ∈ U )[|s|s ∈ S, x_(v, s, u) = 1, ∀v ∈ V |]〗
3. Power consumption—How to successfully decrease the power utilization is a sig-
nificant issue for cloud suppliers. So as to rearrange the issue, here, we just think
about the most direct methodology—limiting the quantity of running servers,
i.e.,
P : min|{s|s ∈ S, ∃u ∈ U, v ∈ V, x_(v, s, u) = 1 }|
What is more, we make the accompanying presumptions:
1. The limit of a server is not expressly included. Be that as it may, when another
VM asks for are being prepared and just the servers with adequate assets left are
considered—we allude to these as authentic servers. As it were, we center around
structuring a calculation to sort these genuine servers and dispense the new VM
to the best-positioned server, with the goal that the over three destinations can
be fulfilled.
Status Monitoring System-Based Defense Mechanism (SMS-BDM) … 1051

2. The multi-target improvement is improved the situation each approaching VM

ask for when it arrives, with the end goal that just the present framework state
and the VM asks for are mulled over; i.e., no look-forward system is utilized;
3. VM live relocation is not mulled over, which implies once a VM is dispensed to
a server, it will keep running on that server until the point when it is halted or
ended by the client.
4. Cloud suppliers do not have any earlier information of the aggressors or
unfortunate casualties, and all VM asks for are dealt similarly.

3.2 Risk Value Measurement

The risk value of every user request is calculated by using RISK equation, and the
state value of every VM is also updated by using a Markov chain model. Finally, a
risk detection metric is used with state transition model so as to optimize the VM
allocation in cloud infrastructure. The connection motor uses a tree of consistent
conditions, i.e., rules, or AND/OR tree, and it is actualized by OSSIM, an open-
source framework. The motor uses Eq. 1 to characterize the hazard an incentive for
each gathering of cautions and at whatever point the hazard winds up bigger than or
equivalent to one, an alert will be let go:

RISK = (AssetValue ∗ AlertPriority ∗ DetectionReliability)/NF (1)

where
AssetValue → value of the attacked resource.
AlertPriority → how dangerous the alert is. This value is set by the firing IDS and
is adapted in the normalization step.
DetectionReliability → the probability that the attack defined in a correlation level
is real.
NF → fixed normalized factor defined by the administrator in the IDS configuration
phase.

3.3 State Value Observation and Updation Using Markov

Chain Model

The risk factor of every user request is calculated by using risk detection metric,
and after that, the state value of every resources is updated by using a Markov chain
Model because the security situation of a cloud has been changed in each point of time.
The HMM comprises a state change likelihood network P, a perception likelihood
framework Q, and an underlying state circulation vector π and is signified by a tuple
1052 S. Rethishkumar and R. Vijayakumar

Sp Ss Security State
(xi)
Sv Vulnerable State
(v)
(iv) Sa Attacked State
(viii) Sp Positive State
(i) (iii) (vii)
Sn Negative State
Ss Sv Sa Sd Sf
Sd Degenerate State
Sf Failure State
(ii) (vi)
(ix)
Without manual intervention
Sn (x) With manual intervention

(i) accident or pre-attack action (ii) detect before attack (iii) begin scanning
(iv) eliminate all negative effect (v) transparent recovery (vi) undetected
(vii) graceful degradation (viii) fail with alarm (ix) restoration/reconfiguration/evolution
(x) restoration/reconfiguration/evolution (xi) restoration/reconfiguration/evolution

Fig. 1 State transition model

(P, Q, π ). In this work, 7 states are considered. The states that are considered in
this work are security state, vulnerability state, attacked state, positive state, negative
state, degenerate state, and failure state. The state change model of this exploration
technique is shown in the accompanying Fig. 1.
Figure 1 shows the state transition (ST) demonstration where the state set I =
{S s , S v , S a , S p , S n , S d , S f } can be commonly traveled as a system of depicting the
dynamic practices of security in cloud. Because of the dynamic use of cloud assets,
the security is without a doubt impacted by the accomplished setting. The state
change of ST model can be portrayed as pursues.
(i) If the protective techniques, e.g., verification, get to control, encryption, and
fix for known vulnerabilities lose adequacy, the framework will enter the
helpless state S v from the security state S s and effectively experiences being
infiltrated and investigated by an aggressor. Be that as it may, if catching the
assaults previous interruption, the framework could in any case remain in the
state S s .
(ii) If recognizing the assaults amid being entered and investigated, the framework
could come back to the state S s from the defenseless state S v . The framework
will as often as possible sweep the present state and could quickly dispatch
the correct countermeasure after distinguishing assaults.
(iii) If the powerlessness is misused effectively, the framework will be under the
assaulted state S a and the potential harm may happen. In the assaulted state S a ,
the framework needs to lead different conceivable ways to deal with recouping
from harm and to relieving the misfortune. In spite of the fact that there may
be more than one reaction accessible, the framework essentially means to
reestablish the security state S s .
Status Monitoring System-Based Defense Mechanism (SMS-BDM) … 1053

(iv) If the framework dispenses with every single ruinous impact caused by an
assault and effectively reestablishes the security state S s , and afterward, the
framework will go into the positive state S p .
(v) By straightforward recuperation, the framework will recoup from the positive
state S p to the security state S s .
(vi) If the framework neglects to perceive the dynamic assaults and along these
lines takes no reaction, the framework will enter the negative state S n .
(vii) Having distinguished assaults however unfit to dispense with every one of the
impacts, the framework needs to contract the degree of harm while keeping
up the basic administration, which is known as the savage state S d . The fun-
damental administrations allude to the functionalities that ought to be kept up
to fulfill the foundational necessities regardless of whether confronting the
unfriendly, disappointments, and mishaps.
(viii) If the above countermeasures do not work by any stretch of the imagination,
the framework needs to enter the disappointment state S f at long last.

(ix), (x), and (xi) By manual mediation, the framework recoups the full administra-
tions after an assault and comes back to the security state S s from
the negative state S n , the savage state S d , and the disappointment
state S f , individually.

After calculating the risk value of every user request, then the alert risk of each
stage has been verified by using the below-mentioned equation.

     
ARs = ACs ∗ AP ∗ DRs /NFs = ACs ∗ Cseverity ∗ Noccurance /Afrequency ∗ DRs /NFs (2)

where
ARs → alert risk at a specific state s.
ACs → asset cost at a specific state s. AC is computed using the C vector, and it
represents the potential consequences of the state s on the asset in question.
AP—<alert priority. It is computed based on C severity , N occurance , and Afrequency as
shown in Eq. 2.
C severity → current alert severity defined by the firing IDS.
N occurance → number of occurrences of current alert in a specified correlation time
slot defined in the correlation process.
Afrequency → acceptable frequency of this alert per day based on the training data
computed from the attack dataset.
DRs → detection reliability at a specific state s. It is computed according to the alert
position corresponding to s in Matrix Å.
NFs → A fixed normalization factor that is computed according to the maximum
values appeared during training phase for ACs , AP, DRs , and maximum alert risk
(MR) where ARs belongs to the range (0–MRs ). All these values are computed for
each state independently. Thus, NFs = (Max(ACs ) * Max(AP) * Max(DRs ))/MRs .
1054 S. Rethishkumar and R. Vijayakumar

3.4 Two-Player Security Game

As expressed [5], when the client accounts are characterized, the diversion approach
is utilized for upgrading the barrier system. The assailant’s conduct, as found prior,
when the aggressor begins their first VM (there is no motivating force for aggressors
to begin more than one VM at first, as none of them will co-situate with the objectives),
they are marked as medium hazard. So as to be renamed as okay, the assailant needs
to keep the first VM running before beginning more VMs. This is known as the
underlying expense. Subsequent to being named as generally safe, the aggressor can
make the same number of VMs as they need. In any case, they need to deliberately
control the pace, with the goal that they won’t be renamed as medium or even high
hazard. When it turns out to be progressively costly to keep the present record being
considered as okay than to make another record (i.e., pay the underlying expense
once more), the assailant will dispose of the present record.

4 Results and Discussion

The execution of the proposed status monitoring system-based defense mecha-

nism (SMS-BDM) or state observation-based co-resident DoS attack detection (SO-
CRDoS-AD) is assessed in CloudSim. The outcomes are contrasted, and the co-
area safe (CLR) calculation proposed in [24], PSSF-based amusement hypothetical
methodology (alluded as PSSF in charts) [10], and out past work two-player diversion
approach-based protection system (alluded as two-player approach in diagrams) as
far as record order exactness, accuracy, review, and assailants by and large expense.

Accuracy: Characterization exactness

Exactness is the rate relating to the effectively done order of client accounts as lawful
and vindictive clients.
(TP + TN)
Accuracy = ∗ 100
(TP + FN + FP + TN)

Figure 2 shows the comparison of the defense mechanisms in terms of accuracy.

From the graph, it can be found that the proposed SO-CRDoS-AD provides highly
accurate classification. The proposed research method SO-CRDoS-AD shows 3.2%
increased accuracy than the two-player approach, 7.9% increased accuracy than PSSF
approach, and 14.4% increased accuracy than the CLR approach.

Precision

Precision is the correctness of the classification of the user accounts.

TN
Precision = ∗ 100
(TP + FP)
Status Monitoring System-Based Defense Mechanism (SMS-BDM) … 1055

CLR PSSF Two player approach SMS-BDM

96
94
Accuracy (%) 92
90
88
86
84
82
80
78
76
Methods

Fig. 2 Accuracy comparison

Figure 3 shows the comparison of the defense mechanisms in terms of precision.

From the graph, it can be found that the proposed SO-CRDoS-AD provides précised
classification, thanks to the optimal solutions for the large-size data center. It is
evident that the proposed defense mechanism can avoid the possibility of attacks
with greater probability. The proposed research method SO-CRDoS-AD seems to
provide 3.3% increased precision rate than two-player approach, 6.8% improved
precision outcome than the PSSF method, and 17.7% improved precision outcome
than the CLR method.

Recall

Recall is the completeness of the classification done in the cloud user accounts.

TN
Recall = ∗ 100
(TP + FN)

Fig. 3 Precision comparison CLR PSSF Two player approach SMS-BDM

0.95

0.9
Precision

0.85

0.8

0.75

0.7
Methods
1056 S. Rethishkumar and R. Vijayakumar

Fig. 4 Recall comparison CLR PSSF Two player approach SMS-BDM

1

0.95

0.9

Recall
0.85

0.8

0.75

0.7
Methods

Figure 4 shows the comparison of the defense mechanisms in terms of recall.

From the graph, it can be found that the proposed SO-CRDoS-AD provides classi-
fication with high recall. The proposed approach increases the defense against co-
resident DoS attacks with higher accuracy. Recall of the proposed research method is
improved 3.2% better than the two-player approach, 12.9% better than PSSF method,
and 18.5% better than the CLR method.

Attacker’s Overall Cost

This parameter helps in assessing the expense caused for starting an assault, i.e.,
making another record for starting another VM. The expense is spoken to in US
dollars ($) for normal cost assessment.
Figure 4 shows the comparison of the attacker’s overall cost for initiating a new
VM for beginning a co-resident DoS attack in the presence of evaluated defense
mechanisms. From the graph, the cost incurred by the attacker to initiate a co-resident
DoS attack is higher in proposed SO-CRDoS-AD than the other methods. It is evident
that the proposed defense mechanism makes it difficult for an attacker by making
an attack process highly expensive. Thus, it forces the attacker to reduce risks and
behave as a normal user. The proposed research method shows 346% improved
performance than two-player approach, 90% improved performance than the PSSF
method, and 171% improved performance than the CLR method (Fig. 5).

Fig. 5 Attacker’s overall CLR PSSF Two player approach SMS-BDM

cost comparison 40
35
30
25
Cost ($)

20
15
10
5
0
Methods
Status Monitoring System-Based Defense Mechanism (SMS-BDM) … 1057

5 Conclusion

In this work, we focus on a specific threat—the co-resident attack that targets the
virtualization level and the bottom of the software stack. In this type of attack, the
attacker has a clear set of target virtual machines (VMs), and they intend to extract
private information from these victims, by co-locating their own attack VMs with
the target VMs on the same physical server and then building different kinds of
side channels. This is resolved in the proposed research method by introducing the
technique called status monitoring system-based defense mechanism (SMS-BDM)
or state observation-based co-resident DoS attack detection (SO-CRDoS-AD). In
the proposed research method, initially, risk value of every resources based on user
requests is evaluated by using risk detection metric. Based on this risk metric value,
state value of every user request in terms of VM is updated. This research method is
implemented in the CloudSim environment from which it is proved that the proposed
research method can ensure the accurate detection of attack status of VM requests;
thus, the security level can be optimized.

References

1. Yan Q, Yu FR, Gong Q, Li J (2016) Software-defined networking (SDN) and distributed denial
of service (DDoS) attacks in cloud computing environments: a survey, some research issues,
and challenges. IEEE Commun Surv Tutorials 18(1):602–622
2. Cardosa MD, Gopisetty S, Korupolu MR, Singh A (2016) U.S. Patent No. 9424094. U.S. Patent
and Trademark Office, Washington, DC
3. Mijumbi R, Serrat J, Gorricho JL, Bouten N, De Turck F, Boutaba R (2016) Network func-
tion virtualization: state-of-the-art and research challenges. IEEE Commun Surv Tutorials
18(1):236–262
4. Han Y, Chan J, Alpcan T, Leckie C (2017) Using virtual machine allocation policies to
defend against co-resident attacks in cloud computing. IEEE Trans Dependable Secure Comput
14(1):95–108
5. Han Y, Alpcan T, Chan J, Leckie C, Rubinstein BI (2016) A game theoretical approach to defend
against co-resident attacks in cloud computing: preventing co-residence using semi-supervised
learning. IEEE Trans Inf Forensics Secur 11(3):556–570
6. Zuo P, Hua Y, Wang C, Xia W, Cao S, Zhou Y, Sun Y (2017) Bandwidth-efficient storage
services for mitigating side channel attack. arXiv preprint arXiv:1703.05126
7. Beaumont O, Eyraud-Dubois L, Lorenzo-del-Castillo JA (2016) Analyzing real cluster data
for formulating allocation algorithms in cloud platforms. Parallel Comput 54:83–96
8. Nalinipriya G, Varalakshmi PJ, Maheswari KG, Anita R (2016) An extensive survey on co-
resident attack in dynamic cloud computing environment. Int J Appl Eng Res 11(5):3019–3023
9. Levitin G, Xing L, Dai Y (2017) Co-residence based data vulnerability versus security in cloud
computing system with random server assignment. Eur J Oper Res
10. Qiu Y, Shen Q, Luo Y, Li C, Wu Z (2017, August) A secure virtual machine deployment strategy
to reduce co-residency in cloud. In: Trustcom/BigDataSE/ICESS, 2017. IEEE, pp 347–354
11. Ezhilchelvan PD, Mitrani I (2017) Evaluating the probability of malicious co-residency in
public clouds. IEEE Trans Cloud Comput 5(3):420–427
12. Zhang Y, Juels A, Oprea A (2011) Home alone: co-residency detection in the cloud via side-
channel analysis. In: 2011 IEEE symposium on security and privacy
1058 S. Rethishkumar and R. Vijayakumar

13. Bates A, Mood B, Pletcher J, Pruse H, Valafar M (2010) Detecting co-residency with active
traffic analysis techniques
14. Han Y, Alpcan T, Chan J, Leckie C (2011) Security games for virtual machine allocation in
cloud computing
15. Calheiros RN, Ranjan R, Beloglazov A, De Rose CAF (2010) CloudSim: a toolkit for modeling
and simulation of cloud computing environments and evaluation of resource provisioning
algorithms
Resource Scheduling Algorithms
for Cloud Computing Environment:
A Literature Survey

V. Arulkumar and N. Bhalaji

Abstract Nowadays, resource scheduling in cloud environment is a challenging task

as the number of customers increases for utilizing the cloud services. In this cloud
environment, allocation of suitable resources to the corresponding VM depends on
the QoS requirement of the specified applications. Researchers have developed so
many resource scheduling algorithms. However, the service providers in cloud envi-
ronment still find it difficult to choose the appropriate algorithm for their applications.
This is due to the heterogeneity of resource types, interdependencies, uncertainty and
dispersion of assets in the cloud environment. This paper reviews all the available
load balancing algorithms in a nutshell.

Keywords Load balancing · Cloud computing · Resource scheduling

1 Introduction

Cloud computing recently emerged as a promising technology for providing unusual

services through Internet to satisfy customer requirements based on their demand
with least spending. As a matter of fact, cloud is a broadly shared figuring which
is determined by economies of scale, where a pool of preoccupied, virtualized, pro-
gressively versatile, oversaw registering power, stockpiling, stages, and any adminis-
trations could be conveyed to outer clients given by devoting Internet. The client can
profit or anticipate all kind of administrations or assets without building, observing
and keeping up their interior resources. It delivers infrastructure, platform and soft-
ware as a service to the customers. The cloud computing provides added services,
which are system infrastructure dependent even if the clients do not have infras-
tructure. Since customers using the cloud providers’ infrastructure on pay as used
and on demand basis, all of us can accumulate resources and operational venture.

V. Arulkumar (B) · N. Bhalaji

Department of Information Technology, SSN College of Engineering, Chennai, India
e-mail: arulkumarv@[Link]
N. Bhalaji
e-mail: bhalajin@[Link]
© Springer Nature Singapore Pte Ltd. 2020 1059
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1060 V. Arulkumar and N. Bhalaji

Due to this, customers can upload their data on the providers’ platform instead of
on their own environment. Their task could run on the cloud and share the server’s
space within the cloud to do dealing out and manipulation of data, etc. A bundle of
specialist co-ops are putting forth distributed computing administrations by convey-
ing server farms at different areas around the globe. In flow situation, the specialist
co-ops evaluated that server farms expend 0.5% of the world’s all out power use. Sup-
plier server farms are requesting more vitality, and it continues multiplying every
year. This prompts harm of condition and furthermore builds consumption of the
specialist organization. Eventually, situation demands for the researchers to find rea-
sonable scheduling or resource management plan to deploy on the service providers’
data centers to ensure their customers to serve speedy manner. Advanced scheduling
scheme supports to cloud service providers in effective way of deploying their data
centers. Resource scheduling can be achievable through any of the following ways:
schedule VMs to conserve data center utilization, effective way of managing the VMs
and primary infrastructure, decrease working inefficiencies for non-essential tasks
and optimize data center load. Existing work investigates the compelling method for
VM executives for the most part fits on any of the followings: watch cloud utilization
and machine load in this circumstance when load diminishes: Live relocate VMs to
increasingly used hubs and shutdown unused hubs. The potential outcomes of oppo-
site side for example load builds: Use hold up states to fire up holding up hubs and
timetable new VMs to new hubs. Burden adjusting is an imperative undertaking in
cloud executives to accomplish most extreme use of assets. Large burden adjusting
task has been investigating either static or dynamic techniques. Static strategy is easy
to reproduce yet does not suit heterogeneous condition. The dynamic technique is
hard to recreate; however, it suits for heterogeneous condition. Here, primary center
dimension is at which hub the static or dynamic calculation has been actualized.
This assumes a noteworthy job in choosing the adequacy of calculation. Successful
method for actualizing load adjusting calculations on cloud foundation basically cen-
ters around proficient usage of assets and finishes the assignment in least time with
least use charges. Figure 1 gives an overview of LB algorithms in cloud computing
environment.
In this paper, we present an audit of current LB algorithms in cloud computing
situation. Examination of various algorithms is done bringing out notable parts of
every one of the categories.
The benefits of various algorithms are viewed as dependent on numerous parts of
the execution. Some of the aspects are shown in Table 1.

2 Literature Review

Numerous specialists have worked in the region of LB in cloud computing condi-

tion. These are gone for decreasing the overutilization or under usage of VMs in the
cloud. There are static and dynamic LB plans. Opportunistic load balance (OLB) is
one of the generally utilized LB plans, in which the undertakings are additionally
Resource Scheduling Algorithms for Cloud Computing Environment … 1061

Pre-Emptive
Based on
Interruption

Non Pre-emptive

Immediate Processing
LB Based on
Algorithms in Processing
cloud
computing Batch Processing

Sender Initiator
Based on
Initiator Receiver Initiator

Symmetric Initiator
Semi Distributed

Centralized
Centralized
Dynamic
Based on state
of system Non-Cooperative

Distributed
Static
Co-Operative

Fig. 1 Load balancing algorithms in cloud computing

Table 1 Performance metrics considered for load balancing

S. No Parameter Remarks
1. Throughput (TP) In light of the quantity of assignments finished effectively
2 Migration time(MT) Time elapsed when the resources move between nodes
3 Response time (RT) Time taken to respond by the resources to respond for a
request
4 Fault tolerance (FT) The ability of a calculation to complete uniform burden
adjusting notwithstanding when interface falls flat
5 Resource utilization (RU) Measured utilization of resources
6 Scalability (Sc) Capacity of a calculation to execute as arranged when the
quantity of hubs increments

partitioned into sub-errands and are doled out for completing the remaining task at
hand at all conceivable time. The OLB algorithm offers better efficiency with the
hosts being kept busy to the maximum extent. However, makespan suffers in typical
cloud environment. Another dynamic algorithm is known as Minimum Execution
Time (MET) algorithm. This approach tries for better makespan in the system by
1062 V. Arulkumar and N. Bhalaji

appropriate balancing of cloud resources. This methodology offers fluctuated con-

trasts in the heap over the framework. A Minimum Compilation Time (MCT) is a
calculation, in which prepared to-execute time and the normal execution time are
viewed as together for adjusting. This methodology can be considered under static
and dynamic techniques. The Simulated Annealing calculation looks to abstain from
getting kept to neighborhood minima and scans for all inclusive ideal arrangement.
The First Come First Served (FCFS) calculation is one of the most straightforward
calculations. In this, the primary occupation in the line is dispensed to the main free
VM, with no inclinations. This calculation is non preemptive. The turnaround and the
reaction time in this calculation are extremely high when contrasted with numerous
calculations. min–min scheduling depends on the idea of allocating an undertaking
having least fulfillment time first for execution on the asset, which has the base con-
summation time (quickest asset). In any case, this algorithm is disadvantageous if
the quantity of bigger measured errand is more than the quantity of shorter assign-
ments. Max–min scheduling depends on the idea of doling out an assignment having
greatest finishing time first for execution on the quickest asset. Be that as it may, this
approach is required to expand the all out reaction time of the framework. Special-
ists have dealt with the above general plans and have displayed better exhibitions in
regard to certain aspects. The service provider is relied upon to pick whichever suits
the prerequisites close by.
Rajput and kushwah [1] surveyed distinctive static and dynamic LB calculations
in-task planning established on load balance, opportunistic load balancing algorithm
(OLB), round-robin load balancer, max–min calculation, min–min calculation, ran-
domized, FCFS serve calculation, most brief reaction time first, similarly spread cur-
rent execution and asset mindfulness scheduling algorithm regarding their relative
favorable circumstances and weaknesses.
Aslan and shah [2] reviewed diverse intuitive LB algorithms which are tending
to various issues from various perspectives and give distinctive arrangements. The
authors classified the focal points and weaknesses of different static calculations like
static LB, round-robin, min–min, max–min and dynamic calculations like honey
bee, ant province, container and throttled LB calculations. The exhibitions of every
one of these calculations are classified for various parameters, for example, fairness,
response time, throughput, overhead, adaptation to internal failure, execution, asset
use, speed and multifaceted nature.
Nitaka [3] proposed another strategy for appropriating the heap in cloud condi-
tion by actualizing round robin, similarly spread current execution calculation and
throttled LB algorithm successively so as to improve the general framework reaction
time and furthermore to get diminished postponement with the server farm preparing
time. The proposed work is executed in three unique dimensions. In level 1, round-
robin calculation is executed which disperse the heap on irregular premise, second
dimension similarly spreads the outstanding burden and considers less VM as per
the circulation procedure and third dimension is executed by throttled calculation.
Samal and Mishra [4] made a near examination of variations of various round-
robin calculations, to be specific, round-robin (RR), modified round robin (MRR),
time slice priority-based round robin (TSPBRR). The investigation is accomplished
Resource Scheduling Algorithms for Cloud Computing Environment … 1063

for different parameters like setting switch, throughput, CPU usage, turnaround time,
holding up time and reaction time. They opined that the TSPBRR calculation gives
better throughput least turnaround time, holding up time and reaction time contrast
with RR and MRR.
Zhang et al. [5] proposed a double tree-like structure comprising of leaf hubs, child
hubs, parent hubs and so forth which partition the simulation region into subdomains.
The outstanding task at hand will be isolated into sub-areas dependent on quick
adjusting calculation. Outstanding burden is determined dependent on the calcula-
tion which gives high productivity, quicker adjusting rate and less correspondence
overhead. Be that as it may, the topology is not kept up.
Dhinesh Babu and VenkataKrishna [6] proposed a method dependent on char-
acteristic conduct of honey bee for finding and harvesting sustenance to adjust the
remaining task at hand between the virtual machines for amplifying the throughput.
It plainly distinguishes the VM with work of lesser need. At the point when high need
remaining task at hand comes, it will be coordinated to VMs with lesser need work.
It is exceptionally intended to give most extreme throughput, minimum waiting time
and with less correspondence overhead. In any case, the principle downside of this
calculation is that in the event that more works with higher need come, at that point
the lower need work will sit tight in the line for long-lasting.
Chandrakanta and Piyush [7] have likewise proposed a novel honey bee enlivened
LB calculation dependent on scavenging conduct of honey bees in a cloud situation.
The undertakings are doled out to VMs which are running in parallel to create the
yield with less execution time. It utilizes the pare to predominance idea for both
choosing ideal VM, and diverse QoS parameters are considered for setting needs to
the assignment. It is expected that the assignments are free, and the VMs utilizes
preemptive errands scheduling.
Dong et al. [8] proposed a LB concept dependent on a distributed architecture
to conquer the issues of dynamic file movement, versatility and accessibility of the
parallel record framework. In this method, an appropriated leader will be utilized, in
which every I/O server can take their very own choice to allocate remaining tasks at
hand to VMs contingent on the present circumstance of the framework to give better
adaptability and accessibility.
Deng and Lau [9] proposed two warmth dispersion algorithms global and local
dissemination for circulated virtual conditions where the load has gotten to by simul-
taneous clients may make issues when number of clients increments. In this propose,
the virtual condition is part into huge number of square cells and each square cell hav-
ing a hub. Global diffusion algorithm comprises of two phases (i) global scheduling
stage and (ii) local load migration stage. In local diffusion algorithm, local decision
making and effective cell choice plans are utilized which basically contrast the neigh-
boring hub loads with the contiguous hub loads. In the event that heap is little, at that
point, the exchange of load ends up conceivable. Here, global dissemination works
better and give less correspondence overhead, fast and require just less measure of
figuring, however, needs in system delay.
Esch and Tobias [10] presented the idea of overlay systems to interconnect each
one of the machines that makes the foundation of an online situation. LB algorithm
1064 V. Arulkumar and N. Bhalaji

encourages the whole system to be separated into littler cells, and every cell is
furnished with different hotspots which are utilized to figure without a doubt the mass
of the item. Every single cell is checked and overseen by an open server. Hotspot
precision expanded when the system load increases. It gives dependable, versatility,
proficient directing and adaptation to internal failure and furthermore gives number
of connections doled out to every hub when the system over-burden. In any case, it
sets aside some effort to adjust the heap as the open servers are put haphazardly.
Godfrey et al. [11] have presented the idea of virtual server to improve the exe-
cution of the system in a dynamic load balancing system. In this, load information
of the peer nodes is stored in different directories which are used to schedule the
workload to different virtual servers for better balance. It is more qualified for shift-
ing remaining tasks at hand and number of hubs with high hub use and improved
versatility. Be that as it may, the reassignment work of virtual server is troublesome.
Kokilavani [12] proposed a min–min method. It is a basic and quick executing
approach. In this method, the allotment happens in two stages. In the main stage,
every single autonomous undertaking without any information reliance on different
assignments is gathered, and their individual least execution time is determined. The
second stage comprises of designation of undertaking with least execution time to
the reasonable assets. The fulfillment time of the offset assets is refreshed with the
expansion of the culmination time of as of now running project. When an undertaking
is finished, it will be expelled from pending rundown until all errands are allotted.
Despite the fact that this calculation offers better throughput, reaction time and asset
use, it creates high correspondence overhead. This calculation cannot adjust the
outstanding task at hand legitimately since it focuses on the little undertakings first.
Chen et al. [13] presented another priority-based improved min–min load balanc-
ing calculation standard to diminish the fulfillment time of assignments and increment
the asset uses. It depends on the standard min–min algorithm. The authors executed
standard min–min, improved min–min and priority-awarded load balancing min–min
algorithm using MATLAB. The outcomes for makespan, average resource utiliza-
tion range, average VIP tasks completion time and average ordinary tasks completion
time are contrasted by shifting the assets and diverse speed and errand with various
size. The outcome demonstrates that this calculation decreases 20% of the normal
finish time of VIP errands.
Anandharajan and Bhagyaveni [14] proposed a load balancing algorithm based
on random sampling of the system. This method, which is appropriate for extensive
framework, works in decentralized manner accomplishing self-association in adjust-
ing the outstanding task at hand. This is conceivable by developing a virtual chart
with the availability of every server. The graph is drawn depicting the degree directed
to the free resources of the server. But, it is found that the performance is degraded
when there is an increase in population diversity.
Chauhan and Joshi [15] have developed max–min algorithm by modifying only
in the second phase of min–min algorithm. In this, the task with maximum expected
completion time is taken first and assigned to suitable resources. It gives importance
to long tasks first and suitable for some environments.
Resource Scheduling Algorithms for Cloud Computing Environment … 1065

Bhoi and Ramanuj [16] built up an upgraded max–min planning calculation

by completing a little change in improved max–min calculation. This calculation
depends on expected execution time as a determination premise rather than culmi-
nation time. Be that as it may, it allocates assignments with normal execution time
to asset that produces least consummation time in improved max–min. On the off
chance that biggest undertakings are excessive since quite a while ago contrasted
with different errands, it expands the general make-length since it is executed just by
slowest asset first. Different errands are executed by quicker assets. This downside
is overwhelmed by upgraded max–min calculation. The calculation is simulated by
CloudSim for different assets. The outcomes demonstrate that it diminishes by large
makespan and can adjust the heap crosswise over assets.
Li et al. [17] proposed an improved max–min task scheduling algorithm. Their
algorithm provides elasticity in cloud computing by maintaining a status table for
all tasks and calculates the existing workload for each machine along with expected
time of completion for each task. This table helps in allocating the remaining work-
load among VMs. The algorithm is assessed by simulation by using CloudSim.
The parameters considered are task pending time, task reaction time, hypothetical
simultaneousness and asset use. This Elastic cloud Max–Min (ECMM) calculation
gives 21.4–30% improvement in results for normal undertaking pending time as con-
trasted and max–min and round-robin calculations. Be that as it may, ECMM needs
in assignment reaction time proportion contrasted with max–min calculation.
Konjaang et al. [18] introduced a new optimized scheduling algorithm based
on max–min algorithm. It identifies the tasks with maximum completion time and
minimum completion time. The algorithm assigns anyone of the tasks to the available
resource to increase the throughput. This helps to overcome the increased makespan
time of standard max–min algorithm. Comparison between the results with standard
max–min and round-robin algorithm concludes that the max–min offers better results
in terms of makespan time and load balancing.
Elzeki et al. [19] presented an improved max–min calculation making a remark-
able change in the standard max–min calculation. It considers the effect of RASA
calculation in appointing errands alongside max–min methodology. It allocates the
undertaking with greatest execution time to asset with least fruition time. RASA is
a hybrid algorithm of max–min and min–min algorithms. The calculation ascertains
the normal consummation time for each assignment on the accessible assets. After the
computation, max–min and min–min calculations are connected again to exploit the
two calculations wiping out their disadvantages. The calculations of max–min, RASA
and improved max–min are reproduced utilizing CloudSim to check for makespan.
The reproduced outcomes demonstrate that improved max–min dependably offers
less makespan contrasted with standard max–min and same or littler makespan as
RASA calculation.
Tawfeek et al. [20] proposed another closed task scheduling algorithm dependent
on ant colony optimization to minimize the makespan time for the given errand
administration. This algorithm takes care of NP-hard problem, for example, voyaging
sales rep issue, graph shading issue and vehicle routing algorithm. This calculation is
reenacted utilizing CloudSim, and the makespan is evaluated for various errands with
1066 V. Arulkumar and N. Bhalaji

round-robin and FCFS calculation. The outcomes are acquired with the presumption
that the errands are commonly free and not preemptive. The outcomes demonstrated
that the ant colony algorithm works better as far as makespan time as contrasted and
round-robin and FCFS.
Gupta and Garg [21] proposed a meta-heuristic methodology of ant colony opti-
mization calculation to take care of the LB issue in distributed computing condition.
It is intended to accomplish least makespan or execution time and better LB. The
calculation is mimicked in CloudSim for ascertaining makespan time and burden
adjusting level by shifting the quantity of undertakings and processors. Results con-
trast well and most famous calculations, for example, NSGA-II with the presumption
that the errands are autonomous. The reproduction results demonstrate that the ACO
calculation delivers less makespan time, and LB level contrasted with NSGA-II.
Shah et al. [22] proposed a algorithm that combines the concepts of opportunistic
load balancing (OLB) and Load balance Min-Min (LBMM). This load schedul-
ing algorithm achieves the advantages of both algorithms, namely, better executing
efficiency, better utilization of resources and load balancing of the system.
Ru and Keung [23] proposed a scheduling calculation which is created with mix of
undertaking gathering, prioritization of data transfer capacity mindfulness and short-
est job first (SJF) calculation so as to lessen the processing time, waiting time, finish
time (makespan) and overhead. In this calculation, the outstanding burdens are made
utilizing Gaussian distribution, and resources are generated using random distribu-
tion. This work is mimicked in CloudSim condition, and the outcomes demonstrate
that the calculation produces 30% improved outcomes for holding up time, preparing
time and makespan alongside better use of assets with minimum overhead.
Agarwal and Jain [24] proposed a new generated priority algorithm working based
on the priority of tasks. The priority of the tasks is defined based on the cloudlet size,
memory, bandwidth scheduling policy, etc. Here, the authors compared time shared,
space shared and a generalized priority algorithm. The above work simulated using
CloudSim for various data centers, host VMs, scheduling and provisioning policies,
and the results are compared with FCFS and round-robin.
Zheng and Zhang [25] proposed a streamlined water wave optimization algo-
rithm dependent on three wave propelled administrators to be specific, propagation,
refraction and breaking. The proposed improved WWO calculation disentangles the
computation by overlooking the refraction administrator. It acquaints a populace
estimate decrease with equalization the load better. They reason that the simplified
WWO works better when contrasted with WWO as far as computations; however,
they do not know whether simplified WWO can supplant WWO totally.
Ghanbari and Othman [26] proposed a priority-based job scheduling algorithm
based on the theory of Analytical Hierarchy Process (AHP). T. Saaty has developed
the AHP model based on a mathematical model combining multi-criteria decision
making (MCDM) and multi-attribute decision making (MADM). The algorithm con-
siders three levels of priorities such as Scheduling level (Objective level), Resources
level (attribute level) and Job Level (alternative level). But this algorithm suffers
from complexity, consistency and finish time.
Resource Scheduling Algorithms for Cloud Computing Environment … 1067

Wang et al. [27] proposed a new job spanning time and load balancing genetic
algorithm (JLGA) based on the original adaptive genetic algorithm. It considers the
new characteristic of cloud computing to improve the performance. This algorithm
initializes the population and describes the load existing among nodes and weight
multi-fitness function. The simulation is done for comparing the performance of
AGA and JLGA with no priorities among jobs. The results prove that JLGA works
better in both total tasks and average task consuming of jobs, and it balances the load
among the systems.
Dubey et al. [28] proposed a new modified Heterogeneous Earliest Finish Time
(HEFT) algorithm to overcome the drawback of inefficient task distribution by HEFT
algorithm. Their algorithm distributes the workload between different resources in an
effective way to reduce the makespan. The algorithm works in two different phases.
In the first phase, the nodes are assigned with a weight, and communication costs
between the nodes are setup. Then rank is calculated for every task in DAG traveling
from the last node toward the root node. In second phase, the task will be effectively
assigned to available resources based on results obtained from the first phase. The
results of the simulations show that the modified HEFT produces less makespan time
as compared to HEFT and CPOP algorithm.

3 Conclusion

There are different alternatives accessible for giving resource scheduling in cloud
computing condition. The researchers, as talked above, have not executed or mim-
icked their calculations for comparative situations, in the sense, the quantity of servers
and their capacities and furthermore the quantity of clients utilizing the cloud and the
assignments that they are performing are not indistinguishable. Each methodology
might be reasonable for one circumstance. In this way, there cannot be a straightfor-
ward measuring stick to quantify the exhibitions of various mists on various angles.
The specialist co-op must think about all angles and introduce a heap adjusting cal-
culations that may best suit the individual prerequisites as far as CPU use, memory
utilization and every other parameter, for example, makespan, throughput and so
forth.

References

1. Rajput SS, kushwah VS (2016) A review on various load balancing algorithms in cloud
computing. Int J Adv Res Comput Sci Soft Eng 6(4). ISSN: 2277 128x
2. Aslam S, Shah MA (2015) Load balancing algorithms in cloud computing: a survey of modern
techniques. NSEC, IEEE
3. Nitika M, Shaveta M, Raj MG (2012) Comparative analysis of load balancing algorithms in
cloud computing. Int J Eng Sci 1(1)
1068 V. Arulkumar and N. Bhalaji

4. Samal P, Mishra P (2013) Analysis of variants in round robin algorithms for load balancing in
cloud computing. Int J Comput Sci Inf Technol 4(3):416–419. ISSN: 0975-9646
5. Zhang D, Jiang C, Li S (2009) A fast adaptive load balancing method for parallel particle-based
simulations. Simul Model Pract Theory 17:1032–1042
6. Dhinesh Babu LD, VenkataKrishna P (2013) Honey bee behavior inspired load balancing of
tasks in cloud computing environments. Appl Soft Comput 13:2292–2303
7. Chandrakanta K, Piyush G (2015) A novel honey bee inspired algorithm for dynamic load
balancing in cloud environment. Int J Adv Res Electr Electron Instrum Eng 4(8)
8. Dong B, Li X, Qimeng W, Xiao L, Ruan L (2012) A dynamic and adaptive load balancing
strategy for parallel file system with large-scale I/O servers. J Parallel Distrib Comput 72:1254–
1268
9. Deng Y, Lau RWH (2010) Heat diffusion based dynamic load balancing for distributed virtual
environments. In: Proceedings of the 17th ACM symposium on virtual reality software and
technology, ACM, pp 203–210
10. Esch M, Tobias E (2010) Decentralized scale-free network construction and load balancing in
massive multiuser virtual environments. In: Collaborative computing networking, applications
and work sharing, collaborate com, 6th international conference on IEEE, pp 1–10
11. Godfrey B, Lakshminarayanan K, Surana S, Karp R, Stoica I (2004) Load balancing in dynamic
structured P2P systems. In: INFOCOM 2004, twenty-third annual joint conference of the IEEE
computer and communications societies, vol 4. IEEE, pp 2253–2262
12. Kokilavani T (2011) Load balanced min-min algorithm for static meta-task scheduling in grid
computing. Int J Comput Appl (0975–8887) 20(2)
13. Chen H, Wang F, Helian N, Akanmu G (2013) User-priority guided min-min scheduling algo-
rithm for load balancing in cloud computing. In: National conference on parallel computing
technologies, pp 1–8
14. Anandharajan TRV, Bhagyaveni MA (2011) Co-operative scheduled energy aware load-
balancing technique for an efficient computational cloud. Int J Comput Sci Issues (IJCSI)
8(2)
15. Chauhan SS, Joshi RC, A weighted mean time min-min max-min selective scheduling strategy
for independent tasks on grid. Electronics and Computer Engineering Department, Indian
Institute of Technology Roorkee-247667, India
16. Bhoi U, Ramanuj PN (2013) Enhanced max-min task scheduling algorithm in cloud computing.
Int J Appl Innov Eng Manag (IJAIEM) 2(4). ISSN No: 2319-4847
17. Li X, Mao Y, Xiao X, Zhuang Y (2014) An improved max-min task-scheduling algorithm for
elastic cloud. In: International symposium on computer, consumer and control
18. Konjaang JK, Ayob FH, Muhammed A (2017) An optimized max-min scheduling algorithm
in cloud computing. J Theor Appl Inf Technol 95(9). ISSN: 1992-8645
19. Elzeki OM, Reshad MZ, Elsoud MA (2012) Improved max-min algorithm in cloud computing.
Int J Comput Appl (0975–8887) 50(12)
20. Tawfeek MA, El-Sisi A, Keshk AE, Torkey FA (2015) Cloud task scheduling based on ant
colony optimization. Int Arab J Inf Technol 12(2)
21. Gupta A, Garg R (2017) Load balancing based task scheduling with ACO in cloud computing.
In: International conference on computer and applications IEEE transactions
22. Shah R, Veeravalli B, Misra M (2007) On the design of adaptive and decentralized load-
balancing algorithms with load estimation for computational grid environments. IEEE Trans
Parallel Distrib Sys 18(12)
23. Ru J, Keung J (2013) An empirical investigation on the simulation of priority and shortest job
first scheduling for cloud-based software systems. In: 22nd Australian conference on software
engineering
24. Agarwal A, Jain S (2014) Efficient optimal algorithm of task scheduling in cloud computing
environment. Int J Comput Trends Technol 9
25. Zheng Y-J, Zhang B (2015) A simplified water wave optimization algorithm. Evolutionary
computation (CEC), [Link]
Resource Scheduling Algorithms for Cloud Computing Environment … 1069

26. Ghanbari S, Othman M (2012) A priority based job scheduling algorithm in cloud computing.
In: International conference on advances science contemporary engineering, pp 778–785
27. Wang T, Liu Z, Chen Y, Xu Y (2014) Load balancing task scheduling based on genetic algo-
rithm in cloud computing. In: International conference on dependable, autonomic and secure
computing 978-1-4799-5079-9/14, IEEE transactions
28. Dubey K, kumar M, Sharma SC (2017) Modified HEFT algorithm for task scheduling in cloud
environment. In: 6th international conference on smart computing and communications ICSCC,
Elsevier
Cloud-Based Healthcare Portal
in Virtual Private Cloud

R. Mahaveerakannan, C. Suresh Gnana Dhas and R. Rama Devi

Abstract Healthcare system providing cloud-based storage makes possible to store

the patient’s therapeutic records to the remote server than keeping the files and radi-
ological images on a hard drive or local storage device which enables the patient
to access their medical records at any place from the Internet through a web-based
application. The structure of cloud application guarantees the privacy and security
of health-related data to preserve the sensitive health information. The architectural
design of cloud computing is to alleviate the privacy concern and to fulfill the con-
fidence and trust of the cloud-based healthcare organization. This work proposes a
structure for a cloud-based healthcare system to allow patients get to their medi-
cal images and reports from the cloud, ensuring that the data are available when it
meets the prerequisite of a particular contract that is authorized. The requirements,
architecture design, software components, and validation methods of cloud-based
healthcare system are introduced.

Keywords Healthcare system · Public cloud · VPC · VPN · NAT · Client system

1 Introduction

The organization requires different types of software and hardware to run. They
need a different expert to maintain the software and hardware [1]. These are the
challenging task before cloud computing. After cloud computing, an organization
can get rid of those problems. If the organization has not had enough resources

R. Mahaveerakannan (B)
Information Technology St. Peter’s University, Chennai, India
e-mail: mahaveerakannan10@[Link]
C. Suresh Gnana Dhas
Vel Tech High Tech Dr. Rangarajan Dr. Sakunthala Engineering College, Chennai, India
e-mail: [Link]@[Link]
R. Rama Devi
Janson Institute of Technology, Coimbatore, India
e-mail: ramarengaraju@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1071

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1072 R. Mahaveerakannan et al.

to invest in infrastructure and platforms to deploy their applications, they can take
advantage of the cloud services to suit their specific needs. Since the cloud provides
a scalable infrastructure to handle the load effectively, the organization can pay for
what they used based on a pay-as-you-go model. Since they can deploy and run
the different application in the cloud, they do not have to buy and maintain the
infrastructure like hardware, software, networking, security, firewall, etc. The cloud
service provider should product their data from the negative impact on business.
So, the cloud service provider provides a high level of services as the expectation
of the organization. Google and Microsoft provide web-based email service. The
emails are stored in a Google server and Microsoft server instead of storing in a
client computer. Microsoft also offers a web-based office online app, web-based apps,
social networking sites, and media services. Healthcare sensitive data must be treated
with high-level security of current legislation [2]. This work encompasses to preserve
the privacy and technical requirement of the healthcare data to allow ubiquitous and
secure access in the cloud environment and highlight the threats to the health data
in the cloud and present the requirements to be fulfilled to mitigate the threats [3].
Public cloud computing provides the virtual private cloud (VPC) which allows the
organization to isolate their cloud instance from other organizations in order to meet
the security issues and provides great control over the cloud environment, and also,
it creates a hardware virtual private network (VPN) to connect the own data centers
with public cloud.

2 Requirements

Cloud-based healthcare system should maintain the security and privacy of data
within the organization in order to protect the sensitive data.
The cryptographic and non-cryptographic methods are used to preserve the privacy
and security concerns [4]. The taxonomy of security and privacy requirements [5] is
shown in Fig. 1.

2.1 Confidentiality

The sensitive healthcare data are not only protected from cloud service provider but
also protected from unauthorized insiders.

2.2 Integrity

The health care data should not be modified through any illegal action of authorized or
unauthorized persons. Identifying attributes such as name, date of birth, and address
Cloud-Based Healthcare Portal in Virtual Private Cloud 1073

1. Confidentiality
2. Integrity
Security Requirements 3. Anonymity
4. Authenticity
5. Unlink-ability

1. Patient-Centric Access Control

2. Access Revocation
Privacy Requirements
3. Auditing

Fig. 1 Security and privacy requirements

are encrypted with cryptographic method and access control mechanism enforced
on the attribute of healthcare data file stored in the cloud.

2.3 Collision Resistance

Collision resistance mechanisms provide resistance among unauthorized users or

authorized users to prevent the illegitimate actions among authorized or unauthorized
users. It ensures the privacy of health data not only from the unauthorized users, but
also from the authorized users.

2.4 Anonymity

The methodology of pseudonyms is identifiers that are used to identify the data
owner instead of real name. The methodology is used to prevent the revealing of the
data owners’ identities such as name, security number form CSP, authorized, and
unauthorized users.

2.5 Authenticity

Develop the robust public-key infrastructure through message encryption and digital
signature and provide the valid authentication code and keys that should only grant
the users access their health-related data.
1074 R. Mahaveerakannan et al.

2.6 Unlink-Ability

Maintain the unlink-ability between patient identifiers and their health-related data to
protect the patient’s privacy. Develop the policy to request query to classify whether
the query is malicious or legitimate.

2.7 Patient-Centric Access control

Patients can encrypt their personal health data before storing at the cloud. Users
are permitted to access the patient’s data based on role, such as doctor, nurse, and
insurance broker and based on access control such as read, write, and print.

2.8 Access Revocation

Since the patients can able to revoke the access rights to different entities over the
health information, other users should not be able to access the health information.

2.9 Auditing

Auditing of healthcare data ensures that manipulation activities of patients’ record

are monitored by either themselves or another role nominated by a patient.

2.10 Architecture Design

2.10.1 Healthcare System (HCS) in the Public Cloud Service

The following architecture is proposed in the figure. It comprises the three systems:
1. The health care system (server)
2. Amazon web service (public cloud)
3. Client system (authorized clients).
The public cloud supports sensitive data of different departments in HCS. The
client application is running under the HTTP and interacts with HCS. The architecture
of HCS in public cloud is shown in Fig. 2.
HCS can create the different application, web services, and database storage in
elastic cloud computing (EC2) instance in the public cloud. Moreover, other organi-
zation can create the instance in the EC2 server in the public cloud, and the instances
Cloud-Based Healthcare Portal in Virtual Private Cloud 1075

Client System 1
Web Server

Health Care Application Server

System Other
Organization
Database Server

Public Cloud
Client System 2

Fig. 2 Healthcare system in flat network

are shared by flat network. Since no organization has own network or own IP address,
it is not possible to prevent the others to try to get accessing own instance in the cloud.
Anyone can directly launch EC2 anywhere in the cloud. Each instance can be directly
accessed by public IP address and based on access control mechanism, so each entity
should be individually managed. Since the hackers are easier to intercept data on the
flat network, it is creating network security issues on the network [5].

2.11 Proposed Architecture

2.11.1 Healthcare System (HCS) in Virtual Private Cloud (VPC)

The work introduces the virtual private cloud (VPC) in the public cloud to allocate the
part in the public cloud. So, the existing network and VPC become as a single network.
The own defined virtual network allows the organization to control the network
environment. When the existing customer gateway (CG) in HCS is connected to
virtual private gateway (VPG) in VPC using VPN or direct connect to establish the
VPN connection, it allows internal system (IS) in HCS to connect to the VPC through
the private IP address without the Internet or public IP address. It can also allow VPC
peering in different region in public cloud that can connect the multiple VPC within
public cloud to talk with each other without the Internet or direct connect which is
shown in Fig. 3.
1076 R. Mahaveerakannan et al.

VPN Connection

Health Care System

Virtual Private
Cloud
Virtual Private
Cloud

Direct Connection

Virtual Private Web Server

Customer
Gateway
Gateway

Application Server
Other
Organization
Private IP Address

Database Server
Internal System

Public Cloud

Virtual Private Cloud

Fig. 3 Virtual private cloud for security

3 Establish Flexibility of Access in VPC

HCS has different server, such as application server, web server, and database server.
It can improve the security in the EC2 instance and in public cloud by providing
appropriate access control to a different server, and each server can be individually
controlled by providing appropriate subnet, for example, connect the Internet gateway
(IGW) to the web server to establish the external access to the web server [6]. The
database server and application server are protected from external server. The large
network is segmented by subnet to provide public and private subnet to increase
flexibility of access. Customer gateway in HCS can connect privately with private IP
address into the VPC in public cloud through virtual private gateway (VPG) without
using the Internet gateway. It can protect the sensitive data by providing private IP
address if it properly secures the network, which is shown in Fig. 4.

Health Care System

Web Server
Application Server
Customer Virtual Private
Gateway Gateway

Database Server Internet Gateway

Server
VPC in Public Cloud

Fig. 4 Create private cloud for healthcare system

Cloud-Based Healthcare Portal in Virtual Private Cloud 1077

VPC [Link] / 16

Public Routing Table

S D
Web
[Link] / 16 Local
Server
[Link] IGW
Application Database
Server Server ENI
[Link] / 24

EC2 Public Subnet

Virtual Private Router

Gateway

Public Routing Table

Web
Server S D

Database [Link] / 16 Local

Application NI
Server Server [Link] / 24

EC2 Public Subnet

Fig. 5 Public and private subnet for servers

3.1 Create the Public and Private Subnet for VPC IP Address
[Link]/16

RFC 1918 standard of IP Range: [Link]/16

Availability Zone1
Public subnet1A: [Link]/24 for web servers
Private subnet1B: [Link]/24 for database servers.

Access control lists (ACL) act as a firewall for controlling the traffic into and out
of the subnet [7]. All web servers are managed in public subnet to allow inbound and
outbound access. The public subnet [Link]/24 wants to communicate to the private
subnet [Link]/24 which sends the request to the router to search for the routing
information. It checks the destination IP address is local to the VPC that communicate
locally; otherwise, it communicates through the Internet gateway [8]. Since public
routing table is associated with the public subnet, it can possibly communicate both
locally and outside. All the database servers have managed in the private subnet to
prevent the inbound and outbound access [9]. Since private routing table is associated
with private subnet, it has the only possibility to communicate local subnet which is
shown in Fig. 5.

3.2 Network Address Translation (NAT)

Proposed works introduced network address translation (NAT) gateway which is

used instead of developing software to perform IP filtering and intrusion detection
1078 R. Mahaveerakannan et al.

VPC [Link] / 16

Public Routing Table

S D
Web
[Link] / 16 Local
Server
[Link] IGW
Application Database
Server Server ENI
[Link] / 24

5
4
EC2 Public Subnet

Router

3
2
1
Public Routing Table
Web
Server S D

Database [Link] / 16 Local

Application NI
Server Server [Link] / 24

EC2 Public Subnet

Fig. 6 Secured database server with NAT gateway

and prevention (IDP) to enable connection in the private subnet to the Internet or
other services in public cloud and also block the inbound public access through the
Internet [10]. For example, private subnet can communicate with the external IP
address [Link] through a NAT gateway which is shown in Fig. 6.
1. Database server wants to communicate with external IP address [Link], and
it sends the request to router.
2. Router search in the private routing table.
3. Get the NAT gateway.
4. NAT gateway connects the router.
5. Communicate with external IP address through IGW.
The healthcare system’s web server can handle the patients’ request and delivers
the medical information anywhere through the NAT gateway with the corresponding
security policy of authorization and authentication [11]. Then, it is responsible for
information is wrapped with their authentication keys for converting the data into an
encrypted object under the security and privacy requirements [12].
Cloud-Based Healthcare Portal in Virtual Private Cloud 1079

Table 1 Evaluations
Goal Metric
parameter
Patient can access medical Patients name, identification
images and reports number, and number of tries
Patient’s privacy protected VPC, access control list
Patient’s data protected Private subnet

4 Evaluations Framework

The goal of evaluations parameter of patient’s authentication and privacy require-

ments is given in Table 1.

5 Conclusion and Proposed Work

The healthcare system keeps up and protects the health-related data in VPC architec-
ture. They can keep the advantages of public cloud with respect to flexibility, scal-
ability, elasticity, performance, availability, and the pay-as-you-use pricing model
for both small- and large-scale health organizations. The healthcare sensitive data
in VPC are intended to meet the prerequisite of security, privacy, and confidential-
ity. The future work is to extend to the analysis of cryptographic solutions along
with a digital signature and build up the robust methodologies of SSH-based lay-
ered encryption approach to ensure the integrity for transferring data from one point
to another. The cloud-based healthcare system can gain the overall acceptance by
providing strong security mechanisms to protect the patients’ records.

References

1. Abbas, Khan SU (2014) A review on the state-of-the-art privacy preserving approaches in

e-health clouds. IEEE J Biomed Health Inf 18(4):1431–1441
2. Jacob A (2009) Infrastructure in the cloud era. In: Proceedings at international O’Reilly
conference velocity
3. Zhang R, Liu L (2010) Security models and requirements for healthcare application clouds.
In: 3rd IEEE international conference on cloud computing, Miami, FL, pp 268–275
4. Li J (2013) Electronic personal health records and the question of privacy. Computers. https://
[Link]/10.1109/MC.2013.225
5. Leng C, Yu H, Wang J, Huang J (2013) Securing personal health records in the cloud by
enforcing sticky policies. TELKOMNIKA Indones J Electr Eng 11(4):2200–2208
6. Subashini S, Kavitha V (2011) A survey on security issues in service delivery models of cloud
computing. J Netw Comput Appl 34(1):1–11
7. Metri P, Sarote G (2011) Privacy issues and challenges in cloud computing. Int J Adv Eng Sci
Technol 5(1):001–006
1080 R. Mahaveerakannan et al.

8. Li ZR, Chang EC, Huang KH, Lai F (2011) A secure electronic medical record sharing mech-
anism in the cloud computing platform. In: 15th IEEE international symposium on consumer
electronics (ISCE 2011), pp 98–103
9. Ramgovind S, Eloff MM, Smith E (2010) The management of security in cloud computing.
In: Information security for South Africa (ISSA), IEEE, pp 1–7
10. Fabian B, Ermakova T, Junghanns P (2015) Collaborative and secure sharing of healthcare data
in multi-clouds. Inf Sys 48:132–150
11. Google Online Documentation (2009) Google’s approach to IT security. Google White Paper.
Winkler V (2011) Securing the cloud computer security techniques and tactics. Elsevier. ISBN:
978-1-59749-592-9
12. Balaram A, Pushpa S (2018) Sybil attack resistant location privacy in VANET. Int J Inf Commun
Technol 13(4)
A Hybrid Approach for Video
Steganography by Stretching the Secret
Data

B. Karthikeyan, M. M. Anishin Raj, D. Yuvaraj

and K. Joseph Abraham Sundar

Abstract In this Internet era, the digital communication is unavoidable, and its
contribution to the development is significant. Due to the enormous data commu-
nication, there is always a question about the security of the data. In this paper, a
hybrid reversible data hiding method is proposed to transmit the data in secure man-
ner through video files. Secret data like one-time password (OTP) can be embedded
into one of the frames in the video. This achieves more security than the regular
ways. The proposed method has compared with mean square error (MSE) and peak
signal-noise ratio (PSNR).

Keywords Steganography · Video · Stretching · Least significant bit (LSB) ·

Video frames

1 Introduction

Steganography is a technique used to hide the data, and in this technique, a stegano-
graphic model is represented which uses cover video files to obscure the existence
of other subtle data without considering their format [1–4]. In this technique, the
secret message is divided into blocks before being embedded into the cover image.
In this paper [5], the four models of secret data which are quantitatively evaluated are
proposed. This scheme can be used in both spatial and frequency domain streams.
In this technique, the technique the encoding capacity is high as compared to the

B. Karthikeyan (B) · K. Joseph Abraham Sundar

Computer Vision & Soft Computing Lab, School of Computing, SASTRA Deemed University,
Thanjavur, India
e-mail: mbalakarthi@[Link]
M. M. A. Raj
Department of CSE, Viswajyothi College of Engineering & Technology, Vazhakulam, Kerala,
India
D. Yuvaraj
Department of CSE, Chian University, Cihan University, Duhok, Kurdistan Region, Iraq

© Springer Nature Singapore Pte Ltd. 2020 1081

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1082 B. Karthikeyan et al.

other techniques. The key data, transform domain and the number LSB bits encoded
must be known for the hackers to extract a data from the cover file in this proposed
scheme.

2 Related Works

Steganography is a process which involves hiding the data inside a cover data in
order to protect it from illegal access. The basic discrete cosine transform (DCT)
and discrete wavelet transform (DWT) are applied on the multiple object tracking
(MOT) algorithm, and error-correcting codes are implemented for a secure and robust
method for video steganography. In this proposed system, first the secret message
is encrypted and then encoded using Hamming code and BCH codes, and later, it is
embedded into the DCT coefficients of the video frames. In this, proposed algorithm
is three-fold, specified as the option-based MOT algorithm, data embedding and the
data extraction. In this technique, a better way for the visual quality is achieved
compared to the other steganographic techniques used for video steganography. This
method improves the security in multimedia field and makes it undetectable. The
proposed algorithm provides privacy and protects it from various illegal accesses.
By applying both MOT and error-correcting codes (ECC), it improves the privacy of
the data [6].
This is a paper which presents a method of motion vector-based video steganog-
raphy. In this improved technique of video steganography, the modification is made
on the LSB. The motion estimation method searches the least MV value. Experi-
mental techniques are performed on the illegally accessed videos, and by different
steganographic methods, the values are converted into bit rates and video codes [7].
This paper discusses about illegal attacks against the governmental agencies are
focused. Video compression on standard H.264/AVC, an algorithm is implemented
on secret sharing and the error-correcting codes. Discrete cosine transform (DCT)
blocks are chosen by grey relational analysis along with a partition mode in video
compression standard H.264/AVC. First and foremost, the secret information is pro-
cessed by the technique called secret sharing, and later on, ECC technique is used
to process the obtained information. We choose the DCT blocks using GRA, and
later rules are used to hide the pretreated information to DCT coefficients form of
the video frames. This implemented algorithm provides robustness and invisibility
as compared to other steganographic techniques. This reduces the attacks and the
PSNR ratios. Videos have a very great capacity of hiding information, and in this
technique, symmetric blocks have been used which increases the capacity to hide
the information secretly and in a large amount [8].
In this paper [9], hiding of data in an efficient method is being discussed. Here,
the data is hidden in such a way that the data is hidden in a tricky way. Patch-Wise
Code Formation (PCF) is used for secure video transformation. The main purpose of
this technique is to transfer the data efficiently and to maintain the secrecy of the data
which is to be transmitted. This method reduces the time complexity and increases
A Hybrid Approach for Video … 1083

the security as compared to other methods. In this paper, a technique is proposed to

hide the existence of the message so that it becomes hard for the attacker to notice it.
In Patch-Wise Code Formation method, the entire video is divided into patches. In
this method, the numbers of bits are reduced and increase the security as compared
to other methods.
Steganography is the method of hiding secret data. In this, proposed system of
video steganography method using spiral LSB and factorization methods is used.
Factorization of the cover frames of cover video files is done. Later, the secret file
is embedded into the scramble of the cover frames. This is done using spiral LSB
techniques. As compared to the other methods, this method is quite efficient and is
undetectable to the attackers. By applying prime factorization method on the cover
frames, scrambling and descrambling are done, respectively. This method is much
more efficient and provides more security and robustness, and PSNR technique is
used to measure the quality of steganography [10].
A 2D chaotic map is used for shuffling the pixel positions, and in this technique,
it is based on nonlinear feedback shift register and Tinkerbell chaotic maps. Using
the technique of chaotic map and NLFSR, a video steganography mechanism is
developed in which the data is hidden into segregated frames by embedding them.
Unlike other hiding techniques, in this technique, the hiding limit is exponential.
In Internet and the mobile system video-based steganography, the files are in video
formats. It is hard to get the encoding position if the initial value of the chaotic map
is not known. For the security factors of the steganographic data, it is aimed that the
improved LSB is good for steganography [11].
In paper [12], the author discusses about how data is hidden in compressed
videos. This improves the security of the motion vector-based steganography with
the presently available steganographic techniques. A syndrome-trellis code is used to
minimize the overall embedding impact. In this proposed scheme, it provides higher
security and the smaller level of the impact on coding as compared to the other
video steganographic methods. The entire embedding effect is reduced by using the
syndrome-trellis code, thereby increasing the security and privacy [12].

3 Design Methodology

Sample video file is converted into number of frames. Randomly, two frames have
been chosen out of the n-1 frames, i.e. except the first frame. The first frame is used
to store the metadata like frame number and length of the secret data. The metadata
can be stored in some other frames also. Figure 1 shows the overall procedure for
the entire scheme.
After selection of two frames, the secret data is expanded [13] which introduces
more complexity for the intruder. The secret data can be expanded in many ways, but
it should be well suited for decoding procedure to get back the secret data. The first
character of secret data is taken, and it is divided by 256. The remainder is stored in
the first pixel of cover image 1, and the remainder is stored in the first pixel of cover
1084 B. Karthikeyan et al.

Sample Video

Convert into frames

Random frame
selection
(Cover Image)

Embed
Secret data Stego
data using image
expand
method

Fig. 1 Data embedding procedure

image 2. The secret data is embedded in the cover images by using least significant
bit (LSB) substitution method. This repeated for all the characters in the secret data.
Since all the steps are reversible, the secret data is extracted from the cover images
with the help of first frame, i.e. metadata. The decoding procedure is shown in Fig. 2.

Frame Apply
selection inverse
Extract Secret
(Stego expand
image) data data
method

Fig. 2 Decoding procedure

A Hybrid Approach for Video … 1085

4 Results and Discussion

Proposed method has been tested with number of images with various size of secret
text. The sample results are shown in Fig. 3.

Size of
Secret Cover Image Stego Image
Text

50

50

100

100

Fig. 3 Sample cover and stego images taken from video clip
1086 B. Karthikeyan et al.

Table 1 Performance
Size of text MSE PSNR
comparison for sample
images 50 0.0018 173.8668
50 0.0008 181.7260
100 0.0083 158.7158
100 0.0013 177.2854

The proposed work achieves not only high-level complexity in embedding the
text but also better results in mean square error (MSE) and peak signal- to- noise
ratio (PSNR). The results are tabulated in Table 1.

5 Conclusion

The stretching concept proposed in this paper increases the security of the data to be
embedded. If the intruder wants to know the secret data which is embedded in the
cover image, then they are supposed to know the stretching procedure and the key
frames. Because of this, the proposed work on video-based steganography achieves
better security than other techniques. Experimental result shows a better accuracy.

References

1. Sriram S, Karthikeyan B, Vaithiyanathan V, Raj MMA (2015) An approach of cryptography

and steganography using rotor cipher for secure transmission. In: 2015 IEEE international
conference on computational intelligence and computing research, ICCIC 2015, 17 Mar 2016,
Article number 7435669
2. Karthikeyan B, Ramakrishnan S, Vaithiyanathan V, Sruti S, Gomathymeenakshi M (2014) An
improved steganographic technique using LSB replacement on a scanned path image. Inter J
Netw Secur 16(1):14–18
3. Charan GS, Kumar SSVN, Vaithiyanathan V, Lakshmi D, Karthikeyan B (2015) A novel
LSB based image steganography with multi-level encryption. In: ICIIECS 2014–2015, IEEE
international conference on innovations in information, embedded and communication systems,
12 Aug 2015, Article number 7192867
4. SSVN Kumar, Charan GS, Karthikeyan B, Vaithiyanathan V, Reddy MR (2016) A hybrid
approach for data hiding through chaos theory and reversible integer mapping. In: International
conference on computational intelligence, cyber security and computational models, ICC3
2015, vol 412. Coimbatore, pp 483–492
5. Hanafy AA, Salama GI, Mohasseb YZ (2008) A secure covert communication model based
on video steganography. In: IEEE military communications conference, San Diego
6. Mstafa RJ, Elleithy KM, Abdelfattah E (2017) A robust and secure video steganography method
in DWT-DCT domains based on multiple object tracking and ECC. IEEE Access 5:5354–5365
7. Wang K, Zhao H, Wang H (2014) Video steganalysis against motion vector-based steganog-
raphy by adding or subtracting one motion vector value. IEEE Trans Inf Forensics Secur
9(5):741–751
A Hybrid Approach for Video … 1087

8. Zhang Y, Zhang M, Yang X, Liu L, Guo D (2017) Novel video steganography algorithm based
on secret sharing and error-correcting code for H.264/AVC. Tsinghua Sci Technol 22(2):198–
209
9. Rajalakshmi K, Mahesh K (2017) Video steganography based on embedding the video using
PCF technique. In: International conference on information, communication and embedded
systems
10. Jha VK, Roy S, Mukherjee S, Sanyal G (2017) Video Steganography technique using factor-
ization and spiral LSB methods. In: International conference on computer, communications
and electronics
11. Kar N, Aman MAA, Mandal K, Bhattacharya B (2017) Chaos-based video steganography. In:
8th international conference on information technology
12. Cao Y, Zhang H, Zhao X, Haibo Y (2015) Covert communication by compressed videos
exploiting the uncertainty of motion estimation. IEEE Commun Lett 19(2):203–206
13. Karthikeyan B, Sameer SZ, Srinath P, Raj MMA, Vaithiyanathan V (2017) A novel stretching
approach for multiple image steganography using bit stuffing. In: International conference on
communication and Security
An Acknowledgment-Based Approach
for Implementing Trusted Routing
Protocol in MANET

K. Dhanya, C. Jeyalakshmi and A. Balakumar

Abstract Security and dependable transmission is testing errand in portable ad

hoc system along the portability of system gadget bargained with assault and loss
of information. For the aversion of assault and dependable transmission, different
creators proposed a technique for verified directing convention, for example, SAODV
and SBRP (secured backup routing protocol). The procedure of these strategies works
alongside route disclosure and route maintenance, inventing, and route prolong kept
up required more power utilization for that procedure. For the verification of gathering
hub, gathering mark procedure is utilized, and rest mode edge idea is utilized for
power minimization. Our proposed strategy is reenacted in ns-2 and contrast with
other directing convention gives a superior act in contrast with vitality utilization
and throughput of system. There are various approaches to compute trust for a node
such as fuzzy trust approach, trust administration approach, and hybrid approach.
Adaptive information dissemination (AID) is a mechanism which ensures the packets
in a specific transmission, and it analyzes if there are any attacks by hackers. It
encompasses of ensuring the packet count and route detection between source and
destination with trusted path. Trust calculation based on the particular condition or
context of a node, by sharing the context data onto the other nodes in the system
would give a better solution to this problem. Here, we present a review on different
trust association approaches in MANETs. We bring out immediate response from the
methodologies for building up trust of the taking part hubs in a dynamic and unsure
MANET atmosphere.

Keywords Internet of things · Mobile ad hoc network · Trust management

approach · Adaptive information dissemination · Route discovery · Attacker · Data
transmission

K. Dhanya (B) · C. Jeyalakshmi · A. Balakumar

K. Ramakrishnan College of Engineering, Trichy, India
e-mail: dhanyaeece2506@[Link]
C. Jeyalakshmi
e-mail: lakshmikrce.2016@[Link]
A. Balakumar
e-mail: balakumar2712@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1089

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1090 K. Dhanya et al.

1 Introduction

Remote sensor networks (WSNs) and wireless sensor and actuator networks
(WSANs) both assume a key job in the work of Internet of things (IoT) frameworks
since they portray the association between the present computational frameworks and
the physical world. In a WSAN, the two sensors and actuator control the physical area
with amounts, for example, temperature, weight, sound dimension, and light force
being continuously estimated by gadgets associated with either a WSN or a WSAN.
Estimated information are sent by remote transportations to handling gadgets for
investigation, and the essential control information are passed on back to WSAN-
associated actuators. Both WSNs and WSANs are utilized in a countless of utilization
spaces including environmental checking and area/following, basic assembling appli-
cations, shrewd networks, and medicinal services. WSN and WSAN transportations
for the most part include multi-jump which commands directing usefulness of all
related gadgets that are not end framework hubs.
To fabricate, trust association relies upon some factor setting, conduct, and under-
standings. It is all the more difficult to figure precisely. So, enhancement can be
capable by considering those setting mindful measurements which measure MANET
execution. Setting mindful measurements could incorporate versatility mindfulness,
vitality mindfulness, control mindfulness, accessibility, difference mindfulness, and
clog cognizance. Counting such measurements in the created conventions should
propel MANET execution.

2 Existing System

So as to exploit the trust in neighbour hubs, keen enemies may profess to be generous
hubs by encouraging bundles preceding propelling parcel dropping assaults. Be that
as it may, course of action number assaults propelled by parcel lessening foes may
create explicit sort of examples in manufacturing some field esteems (e.g., grouping
number and jump tally), in the control bundles amid course disclosure process. A
trust-based plan (TRS-PD) embraces an example revelation instrument [1] to break
down the set down system esteems to caught/got the command bundles. The system
principles are set down in two gliding windows: (i) the main gliding window (GL1)
set down end hub’s character, recent time, jump tally, and finally, goal plan count
and (ii) the second gliding window (GL2) set down goal hub’s personality and recent
time, then alteration among the end arrangement quantities around the got answer
parcel and accommodating solicitation bundle.
A calculation tolerating the [2] representation technique of same distinction inves-
tigates [3] the set down information and then yields in case the connecting hub pursues
the assault design [4–6]. Course revelation procedure can upheld through the instru-
ment insolating the boycotted opponent who can dispatch bundle falling on assaults
(Fig. 1).
An Acknowledgment-Based Approach for Implementing … 1091

Fig. 1 Block diagram for routing proposal

2.1 Drawbacks

• While enemy’s consequent certain assault examples may get saw byte theme dis-
closure instrument, the other bundle dropping foes (which do not pursue any
example) are recognized by the trust display.
• Nonetheless, such hub repossession its trust is expelled from the boycott later on
the off chance that it doesn’t pursue assault designs and believed neighbors have
additionally not recommended it as a questioned hub.

3 Proposed System

A proposal filtering component is used to adequately screen out false honors even
in amazingly threatening surroundings in which the larger part recommenders are
noxious. It utilizes hub to-undertaking task MANET accommodation with multi-
target improvement (MOO) necessities. Versatile information dissemination (AID)
is a system which guarantees the compartments in a specific communicate, and it
investigates if there are any assaults by programmers. It involves shielding the bundle
check and course revelation among source and endpoint with confided in way, with
the goal that the parcel plunging proportion is decreased and conveyance rate of
bundles is expanded. It likewise gives the safe information communicate through
IOT with MANET condition.

3.1 Trust Models

The accompanying tributes for secure IOT steering convention configuration have
been proposed in [7]:
1092 K. Dhanya et al.

• Secure course development [8].

• Self-adjustment [9].
• Effective pernicious hub ID framework.
• Lightweight computations [9].
• Location protection.

4 Proposed Algorithm for Information Routing

On making trust perceptive course assurance plans, we acknowledge the WSN can
be an outline with vertices specialist indicator center points along outskirts address-
ing affirmation interfaces along pinnacle. Charts are sensible portrayal to delineate
composite frameworks, for instance, WSN. The hugeness on a vertex implies extra
essentialness of that center point, and the weight on an edge exhibits the measure of
imperativeness that a center requires to transmit a unit of information along the edge
[10]. The lingering vitality in a course is characterized as the most minimal vitality
dimension of any hub on the course.

4.1 Distributed Trust Evaluations

Conveyed trust appraisal can be delegated: neighbor detecting (direct trust),

proposals-based trust, and half-and-half technique. In appropriated impromptu sys-
tems, trust levels are contrived from the examination of gathered information from
perceptions for explicit activities. It could log that a particular node forwards some
packages as normal and then drops other packets. It could receive this through direct
neighbor sensing and calculate trust from direct knowledge.
• The node state is initialized.
• The value is set 0 for initial selection. 
• Ascertain the intensity of vitality of chose hub for demand as P = , + 1. Here,
the gathering of hub is M−1, and hub determination is 0 to M. In the event that
power of hub is least Pi, at that point, gathering of enactment is chosen.
• The activation phase group should be created.
• GAi[t] ← 0, t = 0 . . . . . . GA − 1
• ti ← 0
• Presently, determination of single
 hub in gathering hub figure completes intensity
of Transceiving power as = (, + 1 +) for choice of dynamic hub and for
ascertaining a neighbor edge as
• Tval = −
• In the event that estimation of Tval is not
exactly chosen hub control esteem, at that point, lower control hub is chosen as
ace.
An Acknowledgment-Based Approach for Implementing … 1093

(1) Secured course revelation over the hub.

(2) Backup hub setup.
(3) Route upkeep over the hub.

5 Results and Discussion

The proposed work has obtained a healthy network by considering the distinctive
features like mobility, security, and excellence of service (Figs. 2, 3, 4, 5).
The above results explained that secure routing path can be established in
MANET. Throughput measurement analysis, efficiency analysis, and data bandwidth
measurement analysis (Table 1 ).

5.1 Performance Parameter

Throughput. It is the part of the channel limit utilized for helpful transmission (Data
bundles accurately conveyed to the goal) and is characterized as the complete number
of parcels gotten by the goal. It is actually a proportion of the viability of a directing
convention.

Fig. 2 Result analysis of secure routing in MANET

1094 K. Dhanya et al.

Fig. 3 Result for throughput with routing proposal

Fig. 4 Comparison result for efficiency

Average delay. This incorporates all conceivable deferrals brought about by

buffering amid course disclosure inertness, lining at the interface line, retransmission
delays at the MAC, and spread and exchange times.
Packet delivery proportion. The proportion of the information parcels conveyed
to the goals to those created by the traffic sources.

6 Conclusion

In this paper, we adjusted the verified stateless convention for verified directing and
limited the use of intensity amid way finding and foundation. For the validation of
gathering hub, gathering mark method is utilized, and rest mode edge idea is utilized
An Acknowledgment-Based Approach for Implementing … 1095

Fig. 5 Result analysis for data bandwidth

Table 1 Parameters
Parameters Value
simulated in our network
Duration 98 s
Area 950*950
Mobile node count 50
Packet rate 4 packet/ s

for power minimization. The proposed calculation partition hub in two states is rests
mode and dynamic mode. Transmission from hub rest to to dynamic mode figures
need of all rest hub and contrast and number-crunching mean of edge. The estimation
of rest mode is more noteworthy, and the equivalent to limit along these lines goes
about as ace hub in gathering. In this style, the usage of intensity limited the on time of
gathering correspondence. Our trial result indicates most extreme lifetime arrange in
contrast with AID steering convention. Trust is a multifaceted, complex, and setting
subordinate idea. The trust establishing and the executives in MANET face difficul-
ties from the serious hold requirements, the open idea of the remote medium, the
mind boggling reliance between the correspondences framework, the interpersonal
organization, and the application arrange, and consequently the perplexing reliance
of any trust metric to highlights, parameters, and interchanges inside and among
these systems.
1096 K. Dhanya et al.

References

1. Argyroudis P, Mahoney D (2005) Secure routing for mobile ad hoc networks. In: IEEE
communication
2. Zhang Z, Nait-Abdesselam F, Ho PH, Lin X (2008) RADAR a reputation based scheme for
detecting anomalous nodes in wireless mesh networks. In: IEEE communications society
3. Djahel S, Nait-Abdesselam F, Khokhar A (2008) An acknowledgment-based scheme to defend
against cooperative black hole attacks in optimized link state routing protocol. In: IEEE
communications society
4. Nakayama H, Kurosawa S, Jamalipour A, Nemoto Y, Kato N (2009) A dynamic anomaly
detection scheme for AODV-based mobile ad hoc networks. In: IEEE transactions on vehicular
technology
5. Li Z, Chigan C, Wong D (2008) AWF-NA: a complete solution for tampered packet detection
in VANETs. In: IEEE communications society
6. Djahel S, Nait-Abdesselam F, Zhang S (2011) Mitigating packet dropping problem in mobile
ad hoc networks: proposals and challenges. In: IEEE communications surveys
7. Chang JM, Tsou PC, Chao HC, Chen JL (2011) CBDS: a cooperative bait detection scheme to
prevent malicious node for MANET based on hybrid defense architecture. In: IEEE transaction
8. Akyildiz IF, Wang X, Wang W (2004) Wireless mesh networks: a survey in Science Direct
9. Sunderasan K, Papagiannaki K (2008) The need for cross layer information in access point
selection algorithms. In: ACM IMC
10. Liu K, Deng J, Varshney PK, Balakrishnan K (2007) An acknowledgment-based approach for
the detection of routing misbehavior in MANETs. In: IEEE transaction
11. Cho JH, Swami A, Chen IR (2011) A survey on trust management for mobile ad hoc networks.
In: IEEE communications surveys
12. Bali S, Machiraju S, Frost V (2007) On the performance of ad-hoc networks in scheduling
techniques. In: WPN
On-line Frequency Reallocation for Call
Requests in Linear Wireless Cellular
Networks

Narayan Patra

Abstract On-line frequency allocation problem (FAP) for wireless linear cellular
networks reusing the frequency of drop calls is studied. In this paper, FAP is investi-
gated on ring networks. The coverage area of highway that surrounds a large mountain
is divided into number of regular hexagonal regions called cells. Each of the cells is
aligned with exactly two adjacent cells, thus forming a ring topology of the network.
Base station (BS) which heads a cell is the transceiver of new or drop call requests
from users of same or neighbouring cells. Utilization of spectrum must be properly
managed so that no calls generated from same or adjacent cells should be left without
getting services from the BSs. In the proposed on-line algorithm for FAP, the drop
call releases frequency, and it is reallocated to an ongoing call using greedy strat-
egy. The algorithm is also implemented on 2- or 3-colourable graph model of ring
depending on even or odd number of cells that constitute the network, respectively.
The performance of the algorithm is analysed on 2-colourable ring network, and it is
found that the competitive ratio of the algorithm is χ /2, where the chromatic number
of ring network is χ . It has been also shown that there is no on-line algorithm of
less than 1 and 3/2-competitive ratios for FAP in 2-colourable and 3-colourable ring
networks, respectively.

Keywords Competitive ratio · Frequency reallocation · On-line algorithm ·

Off-line algorithm · Ring topology

1 Introduction

The problem of frequency interference and its optimal use in wireless cellular net-
works is highly affected today due to the exponential growth in cell phones. The
commonly used frequency division multiplexing (FDM) technology in wireless net-
works separates the local area into disjoint cells known as regular hexagons. As per

N. Patra (B)
Department of Computer Science and Engineering, Siksha O Anusandhan Deemed to be
University, Bhubaneswar, India
e-mail: narayanpatra@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1097

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1098 N. Patra

[1], the centrally located base station (BS) of a cell provides service to the mobile
users from inside and outside periphery of the cell. To receive signals from each
parts of the coverage area, it divides the local area into small regions (cells). Mobile
users send requests to BSs, whenever they require frequency to transmit message to
other mobile users. Depending on the way of receipt of requests and get served by
base station, the frequency allocation problem is classified as on-line and off-line. In
off-line procedure, the entire requests to be responded must be known in advance to
the system, whereas in on-line procedure, requests feed to the system partially over
time. Therefore, the frequency allocation problem is of two types known as off-line
FAP and on-line FAP. Distinguished researchers have been investigated on aforesaid
varieties of FAPs for wireless cellular networks over the last decade [1–8]. Due to
the recent technological innovations in software as well as in hardware for wide use
of digitization with limited availability of spectrum, the researchers are encouraged
to adapt on-line version of FAP.
The wireless traffic on a highway that surrounds a large mountain area in the heart
of a busy metropolitan city forms a ring topology known as ring highway network.
Thus, frequency allocation problem for this type of networks is called FAR. Since
ring topology is a model of cycle graph, it is considered as a special case of linear
network. From Figs. 1 and 2, it is observed that the small geographical regions known
as cells in the network are arranged in such a way that each of the cells has exactly
two neighbouring cells. As an illustration, Fig. 1 shows the topology of general ring

Fig. 1 Ring cellular

wireless network C2 C3 C4

C1 °
Ci
CN

°
CN-1
Cj
°

Fig. 2 Ring cellular

wireless network N = 9 C1 C2

C9 C3

C8 C4

C7 C6 C5
On-line Frequency Reallocation for Call Requests in Linear … 1099

wireless cellular network with N number of cells, and Fig. 2 shows the network with
N = 9 cells.
Let S = {C1r , C2r , . . . Cir , . . . C rj , . . . Cnr } be the sequence of call requests received
by the base station of a cell from n-mobile users in a given time period, where
1 ≤ i ≤ j ≤ n. Each of Cir ∈ S identifies the cell from which ith call initiates. The
Boolean values of r in Cir classify the call requests either a new call or a drop call
depending on r = 1 or r = 0, respectively. Whenever the base station receives a new
call request from its own or from two adjacent cells, it instantly serves the request
allocating available frequency if and only if the request is a new call request. On the
other hand, if the request is drop call request, it reallocates the frequency of dropped
call to an active call in the same cell or free this frequency. Let Z + = {2, 3, 4, . . .} be
the set of positive integers such that frequency f i ∈ Z + is allocated to the ith—call
request in Cir for r = 1. Then, for some integer j > i and frequency f j ∈ Z + ,
f j = f i is allocated to C rj , where either Cir = C rj or Cir is adjacent to C rj . Depending
on the duration of calls, FAP can view as finite or infinite scenario. In infinite scenario,
call never terminates, and the frequency once allocated remains unaltered until end
of call. On the other hand, if a call terminates after a finite time interval, the frequency
released from this terminated (dropped) call may be reused for other existing calls.
However, the basic objective of FAP for wireless cellular network (WCN) is to serve
all the call requests using optimum span of frequency.
The competitive analysis cited in [9] is a metric for measuring the performance
of an on-line algorithm over an off-line algorithm. It is defined as follows.
Let cost (ALGO) and cost (OPT ) be the span of frequencies of an on-line algo-
rithm ALGO and an optimal span of an off-line algorithm OPT, respectively. An
on-line algorithm is α-competitive if and only if there is a constant α ≥ 1 such that
cos t (ALG O) ≤ α ∗ cos t (O P T ) + β, for some constant β. The competitive ratio
becomes absolute for β = 0.

1.1 Overview of Graph Colouring

Since last many years, it has been observed that using graph colouring techniques,
various kinds of optimization problems have been solved by the researchers. In this
paper, the theory of chromatic number of a graph is used in order to increase the
efficiency of FAP. Let the graph G = (V, E), where V is non-empty set of vertices
called cells, and E is set of edges known as links among the base stations and the
mobile phones. Using vertex colouring strategies, the vertices are coloured such that
two adjacent vertices should be assigned with distinct colours. So, colouring is a
function c: V (G) → Z + such that if (i, j) ∈ E, then c(i) = c( j), where c(i) and
c( j) are colours of vertices i and j, respectively. Graph colouring is an optimization
problem in order to minimize the maximum number of colours required to colour
all the vertices of G, i.e. maxi∈V (c(i)) is minimized. The least number of colours
required to colour G is called chromatic number of G denoted by χ (G).
1100 N. Patra

Since regular hexagonal graph is a suitable model for ring highway network, each
of the nodes in this graph is treated as a cell. These cells are topologically regular
hexagonal and are aligned in such way that every cell has exactly two neighbouring
cells. The number of cells required to form a ring highway network which surrounds a
large mountain area depends on the geographical region to be covered by the network.
Let N be a positive integer, the number of cells required to design the network whose
value depends on the length of the highway to be covered by the traffic. Therefore,
the chromatic number of the corresponding graph varies depending on the values of
N. Hence, there is a theorem on the chromatic number of ring graph.
Theorem 1 The chromatic number (χ ) of a ring graph depends on the number of
vertices N of the graph G.
Proof Assuming N ≥ 3, we will apply proper colouring to the vertices of the graph G.
Since G is a ring, every vertex has exactly two neighbours. Let us suppose each vertex
is indexed with 1, 2 and 3…N in a sequence. Using one colour for even indexed vertex
and another colour for odd indexed, it is observed that when N is even minimum
number of different colours required to colour the graph is 2. Similarly, when N is
odd, it needs minimum number of distinct colours to colour the graph is 3. Thus,
χ (G) = 2, when N = even number and χ (G) = 3, when N = odd number.
Finally, the contributions to this paper are listed below.
• The proposed on-line algorithm HYBRID_RING is for solving the frequency
allocation problem using the special case of linear network which is known as ring
network.
• This algorithm is investigated on two models of WCN with ring topology that are
of 2-colourable and 3-colourable ring networks.
• For 2-colourable ring highway, the performance of the proposed on-line algorithm
demands a competitive ratio 1 subject to call duration is limited, and the existing
calls may be reallocated with the frequency of drop call.
• By similar assumption, in case of 3-colourable ring highway, the proposed
algorithm proves a competitive ratio 3/2.
The rest of this paper is organized as follows. In Sect. 2, related work explores the
frequency allocation algorithms using on-line strategies with different constraints for
WCN. Section 3 presents the proposed work in which HYBRID_RING, an on-line
algorithm for solving FAP for ring network with frequency reallocation (FAR_RE),
is discussed. The theoretical results are proved using the bounds for the competitive
ratios of the proposed algorithm in Sect. 4. And the paper is concluded with future
scopes in Sect. 5.

2 Related Work

The on-line algorithm HYBRID_RE in [10] solves the frequency allocation problem
using graph colouring technique. In this case, the networks are modelled as χ −
On-line Frequency Reallocation for Call Requests in Linear … 1101

colourable graphs, where χ is called chromatic number (least number of colours

needed to satisfy the property of perfect colouring) of the graph. Fixed assignment
(FA) [1] and greedy algorithm [11] are jointly used in HYBRID_RE.
Using FA, each of the cells in a cellular network is treated as independent set
depending on frequency distributions among the cells. In general, 3-colourable graph
is a suitable model of WCN. Authors [11] claim that the competitive ratio of their
on-line algorithm meant for frequency allocation to the network is 3.
On the other hand, greedy strategy allocates frequency to the new call request
by selecting lowest available frequency from the frequency spectrum. However, fre-
quency allocated to call requests in this way should not be interfered with the fre-
quency allocated to calls from the same or adjacent cells. The performance of on-line
algorithm [11] for FAP in WCN proves competitive ratio is at least 17/7 and at most
2.5. It finds a tight bond of competitive ratio 17/7 in [2] using greedy strategy. An
on-line version of HYBRID algorithm [12] states that there is a competitive ratio
of (χ + 1)/2 for χ -colourable network. In particular, it has been shown by them
that the competitive ratio for solving FAP in WCN is 2, which is also optimal. For
χ -colourable network, [10] claims HYBRID_RE is (χ + 1)/3-competitive.
The algorithm cited in each of [1, 11–13] deals with infinite call duration. There-
fore, as a consequence, frequency allocated initially to a call will be remained unal-
tered for the entire call duration. In this scenario, it results in a poor usage of frequency
spectrum.
In general, the initiation and termination of a call take place after a finite interval
of time, and also, the availability of frequency spectrum is limited. It is also observed
that the hybrid algorithm in [12] does not consider the drop call for limited time
period. To serve all the new and drop call requests, [10] resolves all the pitfalls of
FAP in WCN considering the call duration is finite instead of infinite.

3 Proposed Work

The frequency allocation problem for the ring (highway) cellular network (FAR)
is investigated in this paper. Since ring graph is a model of aforesaid network, χ -
colourable graph is considered for the solution of FAR. A new on-line algorithm
for FAR is proposed using χ-colourable network called HYBRID_RING, whose
competitive ratio is at most χ /2.
Let us consider a χ -colourable ring network in which χ = 2 or χ = 3. Whenever
the base station receives a call request, it instantly allocates the available frequency or
reallocates the frequency to existing call in the same cell. In order to avoid frequency
interference, the given integer frequency set Z + = {1, 2, 3, . . .} is partitioned into
two sets using equivalence classes FR = [0]χ and FG = [1]χ when χ = 2.
Thus, for χ = 2,

FR = {2, 4, 6, 8 . . .} = {2χ , 4χ , 6χ , . . .}
(1)
FG = {3, 5, 7, 9 . . .} = {χ + 1, χ + 3, χ + 5, . . .}
1102 N. Patra

When χ = 3, the frequency set Z + = {1, 2, 3, . . .} is divided into three partitions

FR = [0]χ , FG = [1]χ and FB = [2]χ , where
⎫
FR = {3, 6, 9, 12, . . .} = {χ , 2χ , 3χ , . . .} ⎬
FG = {4, 7, 10, 13 . . .} = {χ + 1, χ + 4, χ + 7, . . .}and (2)
⎭
FB = {5, 8, 11, 14, . . .} = {χ + 2, χ + 5, χ + 8, . . .}.

3.1 HYBRID_RING Algorithm and Its Principles

Depending on types of call request received by a cell, it serves each of the requests
allocating available frequency to a new call using greedy approach or reallocating
the released frequency of the dropped call to another call with the highest frequency.
It follows two schemes for serving the each call request.
Allocation Scheme: Let us suppose a new call request is generated in some node
(cell) say v with colour red or green. Then, it allocates frequency either from FR or
from FG based on the following rules:
I. Let fr = min(FR ) such that no call from v or from its adjacent uses frequency
fr .
II. Let f g = min(FG ) such that no call from v uses frequency f g .
III. Let f m = min( fr , f g ). Then, allocate frequency f m to the new call.
The reallocation of released frequency from drop call is done as per the given
scheme.
Reallocation scheme: For each drop call request Cir =0 , supposing that it originates
from a node v with colour red or green. The highest frequency f h to a call Cir =1 is
reallocated with the frequency f d of dropped call Cir =0 , if f d < f h and both requests
Cir =0 and Cir =1 belonging to same cell, otherwise free f d .
Input: Request sequence (S = {C1r , C2r , . . . Cir , . . . C rj , . . . Cnr }), where r = 0 or
1.
Output: Allocating the optimal span of frequencies.
Let us suppose that node (cell) v with colour red or green received a new call
request Cir =1 or a drop call request Cir =0 in time period t.
On-line Frequency Reallocation for Call Requests in Linear … 1103

4 Result Analysis

The proposed on-line algorithm HYBRID_RING is implemented on χ − colourable

ring network, and it determines the upper and lower bounds for its competitive ratio
in theorem 2 and theorem 3, respectively. As 2- or 3-colourable graph is the suitable
model for a ring network, it has been shown the counter results in corollary 4.1 and
corollary 4.2 with respect to the results that were previously found for FAP.

Theorem 2 The on-line algorithm HYBRID_RING for χ − colourable networks is

χ /2 is competitive.
1104 N. Patra

Proof Suppose Ci cell with colour red or green receives the sequence of call requests
S in an instance of time period. If each of the requests is a new call and the frequency
once allocated to it survives until call ends, authors in [12] prove that the competitive
ratio does not exceed χ + 1/2. Assuming call duration is finite, and possibly, the call
may be terminated due to various reasons. In this case, frequency allocated earlier
to a call may be altered later. In view of getting better competitive ratio and also to
minimize the frequency span, it is wise to reuse the released frequency from drop call
allocating to some active calls in the same cell of the same network. Let us suppose
that the highest frequency f h is currently allocated to a call in a particular cell with
one of the colours say red or green. If the next request in the same cell is a drop call
request, the proposed algorithm HYBRID_RING proves the competitive ratio is at
most χ /2.

Case 1. Let f h = χl, for some integer l ≥ 1 be the highest frequency belonging
to the frequency set FR being allocated to the call originates from the cell Ci with
colour red or its two adjacent cells with colour green. So l-calls are being in active
through the allocated frequencies in the cell Ci or its adjacent cells Ci−1 or Ci+1 .
Suppose f d = χl  , be the frequency of dropped call belonging to the frequency
set FR such that f d < f h for some integers l and l  with l  < l. Then, f d is allocated
to a call holding the highest frequency f h provided both the calls originate from the
same cell say Ci . On the other hand, if the existing call with highest frequency f h
is in Ci and the dropped call with frequency f d is either in Ci−1 or in Ci+1 , then
f h cannot be reallocated with f d in order to avoid interference of frequencies in the
neighbouring cells. Again, similar arguments can also be made for the case when
frequency of dropped call is f d = χl  + 1,
Case 2. Let f h = χ m + 1, for some integer m ≥ 1 be the highest frequency
belonging to the frequency set FG being allocated to the call originates from the cell
C j with colour red or its two adjacent cells with colour green. So m-calls are being
in active through the allocated frequencies in the cell C j or its adjacent cells C j−1
or C j+1 .
Suppose f d = χ m  + 1, be the frequency of dropped call belonging to the fre-
quency set FG such that f d < f h for some integers m and m  with m  < m . Then,
the call with highest frequency f h is reallocated with f d provided both the calls
originate from the same cell say C j . On the other hand, if the existing call with
highest frequency f h is in C j and the dropped call with frequency f d is either in
C j−1 or in C j+1 , then f h cannot be reallocated with f d in order to avoid interference
of frequencies in the neighbouring cells. Now, similar arguments can also be made
for the case when frequency of the dropped call is f d = χ m  ,
From case 1 and case 2, it is observed that the total number of calls including the
recent most call request is l + m + 1, which are allocated with frequencies from set
FR ∪ FG .
If f h < f h , this implies χl < χ m + 1, for some integers m and l, m ≥ l.
Therefore, the net highest frequency becomes f h = χ m+1. By using the proposed
algorithm, the new highest frequency is now denoted by f hALG = χl. However, the
On-line Frequency Reallocation for Call Requests in Linear … 1105

optimal span of frequencies required for allocation to avoid frequency interference

using off-line algorithm is not less than 2l + 1.
Hence, upper bound of competitive ratio is f hALG /2l + 1 = χl/2l + 1 ≤ χ /2,
for m ≥ l.
If f h < f h , this implies χ m + 1 < χl, for some integers m and l, for l ≥ m + 1.
Therefore, the highest frequency becomes f h = χl. The proposed
HYBRID_RING algorithm updates the highest frequency to f hALG = χ m + 1.
Again, the well-known off-line algorithm uses minimum of 2(l + 1) frequencies to
avoid interference among frequencies.
Hence, the upper bound of competitive ratio be f hALG /2(l + 1) = χ m + 1/2(l +
1) ≤ χ /2 because l ≥ m + 1.
In both cases, the competitive ratio is at most χ /2.
Thus, theorem 2 proves the upper bound of the competitive ratio using the pro-
posed on-line algorithm is χ /2 which dominates the competitive ratio χ + 1/2 cited
in [12] using infinite call duration. Moreover, χ /2-competitive is logically more
efficient than χ + 1/3-competitive due to [10].
Since the ring networks in FAR are a special case of linear networks and also a
model of ring topology, the number of cells used in these networks is either even
or odd. Therefore, the chromatic number of the proposed ring graph is either 2 or 3
depending on the even or odd number of cells, respectively.

Corollary 4.1 On-line algorithm HYBRID_RING for 2-colourable networks

reduces the competitive ratio to 1 which exploits the competitive ratio 5/3 for FAL
due to [14].
In [14], authors proposed an algorithm BORROW with competitive ratio is at most
5/3 in 2-colourable ring networks. Algorithm suggested in [5] with chromatic number
2 proves its competitive ratio 1. Using 2-colourable network and applying theorem
2, HYBRID_RING algorithm finds at most 1-competitive ratio which opposes the
less tight upper bound 5/3 due to the algorithm BORROW. Therefore, corollary 3.2
states the following statement.

Corollary 4.2 Algorithm HYBRID_RING for 2-colourable networks obtains a

tighter upper bound than the upper bound of competitive ratio of BORROW.

Thus, HYBRID_RING algorithm generalizes the works of [14] using a special

case of linear networks.
Lower Bound

Theorem 3 Frequency allocation problem for 2-colourable ring cellular network

contributes minimum of 1-competitive ratio.

Proof Consider a ring cellular network with cells C1 , C2 , C3 , C4 , C5 , C6 , C7 and C8

with colour red and green alternatively in Fig. 3.

Suppose there is an on-line algorithm ALGO for solving FAR. Let us consider
three calls in cell C1 and two calls in C2 are in active with frequency sets {2, 3, 4}
1106 N. Patra

Fig. 3 2-colourable ring

C8 C1 C2
cellular network
2,3,4 5, 6

C7 2 C3

C6 C5 C4

and {5, 6}, respectively. If the highest frequency among the five calls in both the
cells is given by the integer frequency f h = 6 and the current drop call request with
frequency f d , then three situations arise. First, if f d = f h and both frequencies are
used in same cell, then ALGO supports maximum span of frequency is not more
than 5, and the adversary makes one call in cell C2 which needs optimally at least
five different frequencies. Thus, at least 1-competitive ratio results due to on-line
algorithm ALGO. Secondly, if a new call is made in C3 , then the algorithm allocates
integer frequency 3 without conflicting with other frequencies. In this case, it is
observed also the maximum span of frequency be 5 which proves the same bound
for competitive ratio. Third, if f d = 2 such that f d < f h and both being allocated
in the same cell either in C1 or in C2 , where adversary generates a new call in C3 ,
then the highest integer frequency be 5. These arguments prove that proposed on-line
ALGO is at least 1-competitive, as there must exists an off-line algorithm which takes
an optimal span of frequency 5.
Corollary 4.3 Frequency allocation problem for 3-colourable ring cellular net-
works contributes minimum of 3/2-competitive ratio.
Proof Let there is odd number of cells that constitutes the proposed ring cellular
network. According to theorem 1, this network is 3-colourable, and the chromatic
number is 3. Suppose, as an illustration, Fig. 4 is a 3-colourable network which
contains cells C1 , C2 , C 3 , C4 and C5 using colours red, green and blue such that no
two adjacent cells have same colour.
From this network, it is concluded that f h = 7 being the highest frequency, and
it is allocated to a call in cell C 5 . If the call request is a drop and it is also generated
in cell C5 , ALGO finds new highest frequency f h = 6. Since it requires at least four
distinct frequencies optimally using an off-line algorithm, thus the proposed on-line
algorithm proves at least 3/2-competitive ratio.

Fig. 4 3-colourable ring

cellular network {3} {4} {5,6} {3,7}
{4}

C1 C2 C3 C4 C5
On-line Frequency Reallocation for Call Requests in Linear … 1107

5 Conclusion

The proposed algorithm divides the frequency set into 2 or 3 partitions depending on
whether the ring network is 2-colourable or 3-colourable, respectively. Considering
the duration of calls is finite and reusing the frequency free from drop calls for some
existing calls, it has been proved that the competitive ratio (χ +1)/3 is reduced to χ /2
for χ -colourable wireless ring cellular networks. The lower bounds of the algorithm
for 2-colourable as well as 3-colourable ring networks have been derived. It is shown
that for 2-colourable ring, the lower bound is at least 1, and for 3-colourable ring,
it is at least 3/2. The future study will focus on the randomized version of on-line
HYBRID_RING algorithm for both linear and nonlinear networks.

References

1. MacDonald VH (1979) Advanced mobile phone service: the cellular concept. Bell Syst Tech
J, 15–41
2. Chan JWT, Chin FYL, Zhang D, Ye Y, Zhu H (2007) Greedy on-line frequency assignment in
cellular networks. Inf Process Lett 102(2–3):55–61
3. Aardal KI, van Hoesel SPM, Koster AMCA, Mannino C, Sassano A (2007) Models and solution
techniques for frequency assignment problems. Ann Oper Res 153(1):79–129
4. Hale W (1980) Frequency assignment: theory and applications. Proc IEEE 68(12):1497–1514
5. Christ MG, Favrholdt LM, Larsen KS (2013) On-line multi-colouring on the path revisited.
Acta Informatica 50:343–357
6. Pantziou GE, Pentaris GP, Spirakis PG (2002) Competitive call control in mobile networks.
Theory Comput Syst 35(6):625–639
7. Katzela I, Naghshineh M (1996) Channel assignment schemes for cellular mobile telecommu-
nication systems: a comprehensive survey. IEEE Pers Commun 3(3):10–31
8. Chrobak M, Sgall J (2009) Three results on frequency assignment in linear cellular networks.
In: 5th international conference on algorithmic aspects in information and management, vol
5564, LNCS, Springer, pp 129–139
9. Borodin A, El-Yaniv R (1998) On-line computation and competitive analysis. Cambridge
University Press
10. Patra N, Ray BNB, Mohanty SP (2014) On-line frequency reassignment for new and drop calls
in wireless cellular networks. In: 13th international conference on information technology,
proceedings, pp 137–141
11. Caragiannis I, Kaklamanis C, Papaioannou E (2002) Efficient on-line frequency assignment
and call control in cellular networks. Theory Comput Syst 35(5):521–543
12. Chan JWT, Chin FYL, Ye D, Zhang Y (2007) On-line frequency assignment in cellular
networks. In Proceedings of 19th Symposium on Parallel Algorithms and Architectures (SPAA)
13. Chin FYL, Zhang Y, Zhu H (2007) A 1-local 13/9-competitive algorithm for multi-
colouring hexagonal graphs. In: Proceedings of the 13th annual international Computing and
Combinatory Conference (COCOON)
14. Chan JWT, Chin FYL, Ye D, Zhang Y, Zhu H (2006) Frequency assignment problems for linear
cellular networks. In: Proceedings of the 17th annual International Symposium on Algorithms
and Computation (ISAAC), pp. 61–70
A Novel Secure IoT Based Optimizing
Sensor Network for Automatic Medicine
Composition Prescribe System

M. Bowya and V. Karthikeyan

Abstract Growing population and continuous increase in health issues increases the
significance on the healthcare centers which are demanding for an efficient healthcare
system. 24/7 clinical service is not available to all common people and many life
losses occurs due to unavailability of medicine on time and many researches were
made to make the medicine globalized. We propose an IoT based module where the
health-related parameters of patients are recorded using sensors, symptoms computed
are uploaded to the server, by comparing the computed symptoms to the database
the problem is detected and the medicine and dosage level for the detected health
problem is prescribed automatically. Database must be created by group of authorized
specialist doctors and the database management can be done with the help of SQL
server.

Keywords IoT (Internet of Things) · Database management · Medicine

prescription

1 Introduction

Healthcare monitoring is an essential one to maintain the well being of the people.
Due to increased health-related issues, there has been continuous increase in pressure
on the hospitals and the clinical services are not available to all rural areas. The
existing healthcare methodology is hospital-based which is time-consuming and
cost-intensive. This adds to long queues in hospitals and there are several cases that
life loses occurred since the timely medical solution is not obtained. This fast-moving
world needs technology-based advanced healthcare monitoring especially for elderly
people [1]. IoT is creating a great impact on most of the fields today and medical field
is not an exception [2]. Remote patient monitoring can be achieved with the help of

M. Bowya (B) · V. Karthikeyan

Department of ECE, Vivekanandha College of Engineering for Women, Namakkal, India
e-mail: bowya217@[Link]
V. Karthikeyan
e-mail: karthick77keyan@[Link]
© Springer Nature Singapore Pte Ltd. 2020 1109
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1110 M. Bowya and V. Karthikeyan

biomedical sensors and the parameters recorded can be sent to healthcare providers
with the help of IoT, thus the health parameters can be recorded continuously and an
alert intimation may be given in case of emergency. In addition it serves the purpose
of maintaining the medical history of patients but does not provide any solution to
the patients. In this paper, we propose a module where with the combined technology
of Embedded Systems and IoT the health parameters of the patients are sensed and
the health issue is detected by analyzing the symptoms and an immediate medicinal
solution is suggested, i.e., the prescription for the detected problem is generated
automatically with the help of database.

2 Existing System

The existing healthcare system is hospital-based which requires presence of doc-

tors manually. It leads to more transportation time, waiting time in queue, pre-
appointment, and anytime consulting service. Continuous hospitalization is required
for patients with chronic diseases [3] where in such cases the patients may feel iso-
lated from the external world, thus distance monitoring is quite important in this
fast-moving world especially for the people with chronic disorders. Remote moni-
toring of patients can be achieved with the help of IoT technology where patients
at home can be monitored continuously using sensors [4]. The at most work done
in existing methodology is to send the sensed health parameters of the patients as
real-time data to the health care centers [5]. This existing method does not focus on
providing solution to the patients, it only focuses on maintaining the perfect medical
history of the patients so that the further decision regarding the type of treatment
that the patient can be proceeded with is determined effectively. Only in some emer-
gency cases, alert is created that the patient is to be hospitalized immediately. The
Architecture of existing system is shown in Fig. 1.

Data mining
Wi-Fi Cloud
Sensors Microcontroller and
Module Server
Learning

Fig. 1 Architecture of Existing system

A Novel Secure IoT Based Optimizing Sensor Network … 1111

2.1 Sensors

The health parameters are collected by sensing using biomedical sensors and various
wearable sensing devices [6]. Certain important signs including body temperature,
heartbeat rate, respiratory rate and value of blood pressure are monitored and recorded
[7]. Other parameters like blood glucose level need to be monitored in case of diabetic
patient. Thus the parameters to be sensed depends upon the medical status of the
patients, e.g., the factors such as ECG, rate of heartbeat, saturation level of oxygen
and weight are sensed for heart failure patients. For people with disabilities and
elderly people, continuous activity monitoring is required [8].

2.2 Microcontroller

It is mandatory to process the sensed parameters into transmittable format using

smartphones [9], FPGA (Field Programmable Gate Array), hardware platforms,
microcontrollers or microprocessors. This combines the data from sensors, transfers
the data to server for storing and then finally processes the data. It is the compu-
tational block of the architecture and is an essential one for decision making and
further analysis in future.

2.3 Wi-Fi Module

The sensed and recorded data are to be sent to the microcontroller. This involves
several protocols by which the connection and data transmission can be done. The
technologies like Wi-Fi [10], Bluetooth, ZigBee, LTE (Long-Term Evolution) or any
other high-level communication protocols can be implemented. Both Bluetooth and
Wi-Fi provide wireless communication but the difference resides in their efficiency.
Various devices can be connected without the usage of cable by using Bluetooth
while the access to internet can be achieved using Wi-Fi.

2.4 Cloud Server

The collected and processed data need to be stored for further analysis. The data is
stored without any loss using cloud servers. There are several platforms available for
data storage including ThingWorx, Open IoT, Google Cloud, Amazon, GENI [11].
Cloud computing is used to manage the large database thus the efficiency of data
storage is improved. Here the data is stored in distributed server so that the data is
available on demand without any sort of traffic.
1112 M. Bowya and V. Karthikeyan

2.5 Data Mining and Learning

This layer focus on converting the collected and stored data into knowledge in deci-
sion making. Data mining is an algorithm-based process of generating new infor-
mation by analyzing and examining huge database. Here the frequency of similar
data of patient is registered. Machine Learning provides the system the ability to
automatically learn and improve from experience.

3 Proposed Work

The overall architecture of proposed system shown in Fig. 2a consists of several

client devices (i.e., the device the patients are provided with) connected to a server
via IoT. Patient’s health parameters are sensed, processed and finally uploaded to
the authorized server using the client device. The server is connected to the database
where the symptoms and basic details of patients are analyzed, the health issue of
the patient is detected and the standard medicine for the problem and dosage level
according to the age factor is suggested to the patient instantly.

Client
Device 1

Wireless
Client IoT Server
Device 2 Network

.
.
.
.

Database
Client
Device n

Fig. 2 a Overall Proposed Architecture

A Novel Secure IoT Based Optimizing Sensor Network … 1113

3.1 Client Device

Each client device consists of n number of biomedical sensors which are used to
sense the vital signs of patients [12]. The recorded data are fed as input to the signal
conditioning circuit which is used to manipulate the recorded values and covert
them from analog to digital signal since the sensed parameters can’t be directly
given as input to the Digital Signal Controller. Both Digital Signal Controller and
signal conditioning unit are provided with the power supply unit [13]. DSCs can
be used in variety of applications including power conversion, motor control, and
sensor processing applications. Here the Digital Signal Controller uploads the health
parameters of patients to the server. And after analyzing the recorded symptoms of
patients the detected problem and the medicine along with dosage level is sent to the
client device. Client module is shown in the Fig. 3.

3.1.1 Power Supply Unit

Every electronic system is to be provided with DC power supply. Different circuits

operate under different power supplies and the rating depends upon the load current
and voltage. The load current depends on the load resistance, i.e., the load current
is inversely proportional to load resistance. So it is quite mandatory to provide the
electronic circuit with matched designation of power supply. In our proposed module
we use two power supplies, DC 5v with GND and DC 12v with GND. A LM341
is a three-terminal positive voltage regulator is used to maintain the output voltage
constant irrespective of its input voltage.

3.1.2 Sensors

The biomedical sensors are implemented in the client device to sense the health
parameters of the patients. In our proposed module Temperature sensor (LM35)

Power
Supply
Unit IoT

Signal Digital
Sensors Conditioning Signal GSM/GPRS
Circuit Controller Module

Fig. 3 Client device

1114 M. Bowya and V. Karthikeyan

and Heartbeat sensor are used to record the body temperature and heartbeat rate.
The output of LM35 series shown in Fig. 5 is a voltage which is comparative to
temperature in Celsius. The light source (LED) is made to be incident onto any
muscular region and the amount of light reflected back is estimated using light
detector. The amount of light reflected changes with respect to the change in the
volume of blood flow, thus the rate of heartbeat can be estimated. Ultrasonic sensors
with low power oscillators [14] can be implemented for better efficiency.

3.1.3 Signal Conditioning Circuit

The sensed analog value cannot be directly given as input to the next stage thus
the signal conditioning circuit is implemented so that the measured analog health
parameter of the patient is processed into a form such that the input requirements
of the next stage is met. The signal conditioning circuit accepts any type of sensor
inputs and the outputs can be voltage, current, frequency, timer or counter.
Signal conditioning processes: Signal Conditioning circuit involves two main
process including Filtering and Amplifying. The analog parameter sensed by sensors
is subjected to filtering so that any high-frequency noise present can be eliminated,
then the resultant signal is amplified. Amplified analog signal is converted into digital
signal using ADC converter. Thus, its function is to convert the measured analog
signal into processed digital signal which is the input to next stage (Digital Signal
Controller).

3.1.4 Digital Signal Controller

A Digital Signal Controller (DSC) is considered to have advanced features than

Digital Signal Processors (DSPs) and Microcontrollers. The digital output from the
Signal-Conditioning circuit need to converted into proper symptoms so that it can
be uploaded to server and compared with the database in the authorized server. DSC
performs the major role of processing the data and uploading the processed data to
the server using GSM/GPRS module and DSC can be generally programmed using
the C programming language.

3.1.5 GSM/GPRS Module

GPRS module is used to efficiently transfer the data between client and server by
using circuit-switched mechanism. It is the responsibility of the GSM/GPRS module
to find the best path with minimal traffic to transfer the processed data from the DSC
to the authorized server. Both uploading of parameters to server and receiving the
feedback response given by the server, i.e., prescribed medicine and dosage level, at
high speed can be achieved using this module.
A Novel Secure IoT Based Optimizing Sensor Network … 1115

3.1.6 IoT

The Internet of Things (IoT) plays a major role in our work since it allows the com-
munication between the client and server module. It is the network of devices which
allows connection, interaction and exchanging of data between various things. IoT is
the technology involving extending of internet connection beyond standard devices
to the devices that are used in day to day life. With the IoT technology, the devices
may be able to communicate among one another and controlling of a device from is
remote location can also be achieved.

3.2 Server and Database Management

The collected and processed health parameters of the patients are uploaded to the
authorized server which is connected to the database. The database management is
done with the help of Microsoft SQL server and the programming is done using php.
Php is an easiest web programming language that is used to create a database. The
basic details of patients would be registered at the hospital server. Once registered the
further login can be done using the provided username and password. The database
created consists of the health issues which is selected based on the recorded param-
eters of patients and the dosage level is selected according to the age factor of the
patients.

4 Results and Discussion

Each client device consists of n number of biomedical sensors which are used to
sense the vital signs of patients. The recorded data are fed as input to the signal
conditioning circuit which is used to manipulate the recorded values and covert them
from analog to digital signal since the sensed parameters can’t be directly given as
input to the DSC. Both DSC and signal conditioning unit are provided with the power
supply unit. Here the DSC uploads the health parameters of patients to the server.
And after analyzing the symptoms of patients the detected problem and the medicine
with dosage level is sent to the client device (Fig. 4).
Server and Database Management:
The basic details of the patients would be registered to a authorized hospital server
and once registered the patients can be provided with the user name and password
with which the patient can log in (Figs. 5, 6 and 7).
Once the login is done the patient’s health parameters are uploaded to the server
via GPRS module and the symptoms of the patients are compared to the database
created and the detected problem, medicine for the problem detected along with the
dosage level is displayed to the patients instantly (Fig. 8).
1116 M. Bowya and V. Karthikeyan

Fig. 4 Client Device

Fig. 5 Login page of automatic medicine prescribe system

Fig. 6 Login can be done using corresponding username and password

A Novel Secure IoT Based Optimizing Sensor Network … 1117

Fig. 7 No invalid login can be done

Fig. 8 Medicine and dosage level prescription

5 Conclusion and Future Work

In this paper, remote monitoring and diagnosing using sensors and automatic
medicine prescription generation using the database is achieved. In our work mini-
mum parameters like body temperature and heartbeat rate are analyzed and different
health status along with the corresponding medicine and dosage level to be taken
is given as response to the patients. Initial dosage level suggestion would be based
on the age factor and further dosage level is based on response of patient’s health
response to the medicine (with the help of medical history maintained in the server).
It decreases pressure on hospitals and reduces healthcare cost. In future furthermore
biomedical sensors can be added and by analyzing further more symptoms many
health issues can be detected and the prescription for many problems can be gen-
erated. The database of medicine can be created by super-specialist doctors, and it
1118 M. Bowya and V. Karthikeyan

can act as an life-saving equipment by providing immediate first aid medicine in

emergency situations like heart attack.

References

1. Jimenez F, Torres R (2015) Building an IoT-aware healthcare monitoring system for elderly
people. In: 34th international conference of the chilean computer science society (SCCC), pp
1–4
2. Karthikeyan V, Bowya M (2018) A review on IoT based healthcare monitoring. Indo-Iran J
Sci Res 2:24–30
3. Dierckx R, Pellicori P, Cleland JGF, Clark AL (2015) Tele monitoring in heart failure: big
Brother watching over you. Heart Fail Rev 20:107–116
4. Kelly SDT, Suryadevara NK, Mukhopadhyay SC (2013) Towards the Implementation of IoT
for environmental condition monitoring in homes. J IEEE Sens 13:3846–3853
5. Nguyen HH, Mirza F, Naeem MA, Nguyen M (2017) A review on IoT healthcare monitoring
applications and a vision for transforming sensor data into real-time clinical feedback. In:
International conference on computer supported cooperative work in design, pp 257–62
6. Karthikeyan V (2016) A novel design of low energy oscillator based ultrasonic sensor edge for
intravascular applications with CMUT technology. J Middle-East J Sci Res 24:3716–3720
7. Karthikeyan V, Christina J (2016) Review of ultrasonic transducers for medical applications
with CMUT implementation. J Int J Adv Eng Glob Tech 7
8. Bjorkman M, Causevic A, Fotouhi H, Ahmed MU, Linden M (2015) An overview on the
Internet of Things for health monitoring systems. In: 2nd EAI international conference on IoT
technologies for healthcare healthyIoT 169:429–36
9. Bisio I, Lavagetto F, Marchese M, Sciarrone A (2015) A smartphone centric platform for
remote health monitoring of heart failure. Int J Commun Syst 28:1753–1771
10. Narendra Swaroop K, Chandu K, Gorrepotu R, Deb S (2019) A healthcare monitoring system
for vital signs using IoT. Internet Things 5:116–129
11. Al-Fuqaha A, Guizani M, Mohammadi M, Aledhari M, Ayyash M (2015) Internet of Things:
a survey on enabling technologies, protocols, and applications. IEEE Commun Surv Tutorials
17:2347–2376
12. Fortino G, Parisi D, Pirrone V, Di Fatta G (2014) Body Cloud: a SaaS approach for community
body sensor networks. Future Generation Comput Syst 35:62–79
13. Karthikeyan V, Christina Jasmine (2017) A squat energy oscillator based readout edge with
CMUT technique for medical applications. J Adv Nat Appl Sci 11:364–641
14. Karthileyen V, Jasmine Christina J (2017) Design of low power oscillator for medical ultrasonic
sensors with CMUT implementation. Asian J Appl Sci Technol 1:68–72
High-Speed Polar Decoder Architecture
for Next Generation 5G Applications
Using Radix-k Processing Engine

R. Kavipriya and M. Maheswari

Abstract Among the various coding techniques, polar codes are very useful, since it
achieves the ultimate channel coding characteristics. In this paper, a decoding archi-
tecture is proposed which reduces the memory consumption and delay. To overcome
the problem of concurrency, a parallel processing is done. The further research is in
processing on the polar codes to be applied for next-generation applications. Polar
code is preferred since it increases the speed of the operation for large number of
bits (i.e.) for large code length. This implementation process is further extended by
combining the encoding and decoding process by using radix-k based design.

Keywords Polar decoder · Polar codes · High speed

1 Introduction

The main aim of coding theory is to approach a new code in order to satisfy the
Shannon limits as compared with the other codes [1]. Turbo codes and LDPC codes is
found to satisfy the Shannon limit. Apart from these codes, Polar codes is found to be
an ultimate capacity-achieving code so it is used for next-generation 5G applications.
It is also utilized in order to perform reliable data transmission.
Channel coding is employed to mitigate the occurrence of error. Generally encod-
ing is an operation which transforms the original code into another code at the time of
transmission. To recover the original information which is transmitted a decoding is
required. The channel at which the information is transmitted can be any base station
[2]. In polar codes, the information is transmitted at bit positions ‘0’ and ‘1’. If the
information is sent at ‘1’ position then it guarantees the transmission of information
and it is found to be unreliable if it is sent at ‘0’ position [3].

R. Kavipriya (B) · M. Maheswari

K. Ramakrishnan College of Engineering, Tiruchirappalli, India
e-mail: kavipriya1096@[Link]
M. Maheswari
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1119

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1120 R. Kavipriya and M. Maheswari

Many efforts are done in order to implement polar code for encoding and decoding
operation without increasing its complexity and robustness during transmission [4].
In the next chapter, we outline the survey of the various decoders. In the chapter III,
the proposed polar decoding is implemented and its operation is explained in brief.
In the last chapter, the simulated results are shown for the above-mentioned code.

2 Literature Survey

To reduce the complexity in the Successive Cancellation decoding algorithm the

abandoned frozen bits are used so that it reduces 20% of its complexity without
increasing the loss [5]. By reducing the redundancy, the complexity is also reduced.
The pipelined architecture method is employed by performing the FFT operation so
that the efficiency is high.
In the SC decoder, Merged Processing Element is employed since it reduces
many conversions such as magnitude and sign conversions, thereby increasing the
throughput [6]. The latency increases as the complexity is increased. In order to
avoid that problem, the original decoding graph is diving into many small graphs
to overcome various problems [7]. This problem can be further reduced by using
multiple decision approach [8, 9].
To balance the latency of the adjacent stages, the processing elements are separated
from each other and then combined together in Fast SSC decoding algorithm to
optimize its critical path [10]. The frequency and the power consumption is also
found to be reduced. To avoid the problem of reordering of the information bit
butterfly diagram is preferred [11]. Since the decoder is directly placed at the router
it can correct the error after decoding the bits [12].
In Successive cancellation List Decoding, a pruning technique is employed to
overcome the time complexities and space complexities by using the Maximum
Likelihood method of decoding [13].

3 Proposed System

The output of the encoder is applied as the input to the decoder and the resulting
output will be the original information bit which undergoes various steps to avoid
the leakage of the information. Figure 1 indicates the basic channel diagram of the
polar code. The basic polar code channel is shown in Fig. 2.
The whole operation is subdivided into many small stages. The encoded output
x0, x8 is applied as the input to A0 and x1, x9 is applied as the input to A1. Similarly
x2 and x10 is applied as input to A2, x3 and x11 is applied as input to A3, x4 and
x12 is applied as input to A4, x5 and x13 is applied as input to A5, x6 and x14 is
applied as input to A6, x7 and x15 is applied as input to A7. The outputs is obtained
by performing XOR operation and the process goes on.
High-Speed Polar Decoder Architecture for Next Generation 5G … 1121

Fig. 1 Various coding techniques

Fig. 2 Basic polar code

channel

The output of A0, A1, A2, A3, A4, A5, A6, A7 is applied as the input to stage 1
and it is denoted by wij where the edge is denoted by ‘j’ and the stage is denoted by
‘i’. By performing the addition operation with the inputs A0 and A4, the output B0
is obtained. Similarly, by performing the subtraction operation of the above inputs,
the output B4 is obtained. The above operation is repeated with the input values A1
and A5 from the previous value obtained to generate B1 and B5 which is obtained by
performing addition operation and subtraction operation. Likewise, with the inputs
A2, A3, A6, A7, the corresponding outputs B2, B3, B6, and B7 are obtained. Stage
1 output is applied as the input to stage 2 for processing.
The outputs of stage1 is multiplied with the complex coefficients W NK before
passing as the input to stage 2. The complex coefficients W NK is also known as phase
factors or twiddle factors which can be calculated by using the formula as shown
below,

W N = e− j2π/N (1)

The values of phase factors W80 is found to be 1, W81 is found to be 0.707 +

j0.707, W82 is found to be j. Similarly W84 is found to be −0.707 + j0.707. The
output A4 is multiplied with W08 and the corresponding value is applied as the input
to stage 2. Similarly A5, A6, and A7 is multiplied with W81 , W82 , W83 to generate the
corresponding outputs. Figure 3 represents the data flow graph of polar decoding.
In the second stage, the values which are obtained after performing multiplication
operation with the multiplier coefficients is applied as the input, i.e., the output C0
1122 R. Kavipriya and M. Maheswari

<----stage1----><----stage2-----><----stage3--->

Fig. 3 Data flow graph of polar decoding

is obtained by performing addition operation with the inputs B0 and B2, the output
C2 is obtained in a similar manner by performing subtraction operation instead of
addition operation. The outputs C1 and C3 is obtained by performing addition and
subtraction operation with the inputs B1 and B3. The operation which is performed
with the first four information bits are repeated to generate C4, C5, C6, C7 but with
the different input values B4, B5, B6, B7.
The outputs obtained in the second stage is then multiplied with the complex
coefficients before passing to next stage for processing. The output C2 and C3 is
multiplied with the twiddle factors W80 and W18 and then passed to next stage for
processing. Similarly, C6 and C7 is multiplied with W80 and W81 . The phase factors
comprises of complex values and so the output will be the combination of both the
real and the imaginary values.
The output of stage 2 is then applied as the input to stage 3. D0 is obtained
by performing addition operation with the inputs C0 and C1, D1 is obtained by
performing the subtraction operation with the same inputs shown above. The above
process is repeated with different inputs such as C2, C3, C4, C5, C6 and C7 to
generate the outputs D2, D3, D4, D5, D6, and D7.
The entire process is divided into various stages and the operation of next stage
depends upon the previous stage values, therefore a register is needed in order to
store the values. The final decoded outputs are generated in the normal order which
is then divided by N to generate the decoded values. The above process is said to be
decoding process which increases the security during the transmission through the
channel.
High-Speed Polar Decoder Architecture for Next Generation 5G … 1123

4 Results and Discussion

In this chapter, the proposed system simulation results are outlined. The process is
simulated with various inputs and the related outputs are obtained. The inputs is of
7 bits for out0 and out4 and 12 bits for out1, out2, out3, out5, out6, out7. The input
information bits are shown in the Fig. 4,
The decoder accepts the inputs in the bit reversed order and generates the asso-
ciated outputs in natural order (i.e.) the order is found to be the same as the input
values. The 12-bit information is decoded by using the polar decoding scheme and
the corresponding outputs are generated.
The proposed decoding scheme is implemented in Model Sim version 10.4a soft-
ware. To generate the synthesis report it is also implemented in Xilinx ISE (Integrated
Synthesis Environment) Software. It provides various information regarding the uti-
lization of logic gates and the number of slice LUTs and bonded IOBs. The outputs
of the proposed system, i.e., decoding operation are shown in Fig. 5.
The synthesis report of the proposed system which comprises of the number of
slice LUTs and number of bonded IOBs is shown in Table 1.
The Maximum combinational path delay for the proposed decoder system is
48.646 ns. The memory consumption of the proposed decoding process is 265,712
kilobytes.

Fig. 4 Inputs generated before decoding

1124 R. Kavipriya and M. Maheswari

Fig. 5 Proposed decoder output

Table 1 Synthesis report

Logic utilization Used Utilization (%)
Number of slice LUTS 1836 20
Number of bonded IOBs 230 123

5 Conclusion

The decoding complexity should be considered while decoding the information bits
since it can change based on various parameters. The complexity varies based on
the encoding outputs and the code length. In our proposed system, to overcome
these kind of problems the encoded outputs are checked before decoding and if there
is no occurrence of imaginary part then there is no need for usage of multiplier
coefficients. Therefore, the memory requirement is reduces which in turn speeds
up the performance. The proposed system is implemented using Model Sim 10.4a
Software and its synthesis is done using Xilinx ISE Suite 14.7.

References

1. Srinath P, Samuel John J (2016) Implementation of parameter based partially parallel encoder
architecture for long polar codes. IJARECE 5(10), October
2. Robert G, Maunder CTO (2018) The implementation challenges of polar codes. AccelerComm,
February
High-Speed Polar Decoder Architecture for Next Generation 5G … 1125

3. Korada SB, Sasoglu E, Urbanke R (2010) Polar codes: characterization of exponent bounds,
constructions. IEEE Trans Inf Theor 56(12):6253–6264
4. Condo C, Land I, Biogloi V (2018) Design of polar codes in 5G new radio.
arXiv:1804.04389v2[[Link]]
5. Liu A, Zhang Q, Yi X, Liang X (2017) A reduced complexity successive cancellation decoder
of polar codes. In: 3rd international conference on computer and communications
6. Babu GS, Madala LR, Gopalakrishnan L, Sellathurai M Low complex processing element
architecture for successive cancellation decoder. Integr, VLSI J
7. Viterbo E, Vangala H, Hong Y (2014) A new multiple folded successive cancellation for polar
codes. In: IEEE information theory workshop (ITW)
8. Yuan B, Parhi KK Low-latency successive-cancellation list decoders for polar codes with
multibit decision. IEEE Trans Very Large Scale Integr (VLSI) Syst
9. Gross WJ, Leroux C, Raymond AJ, Sarkis G (2001) A semi-parallel successive- cancellation
decoder for polar codes. IEEE Trans Sig Process 61(2):289–299
10. Cui J, Zeng Q, Higgs R, Yan X, Zhang X (2018) High throughput fast SSC polar decoder for
wireless communication. Wireless Mobile Computing
11. Duhamel P (1986) Implementation of split-radix FFT algorithms for complex, real and real-
symmetric data. IEEE Trans Acoust Speech Sig Process (ASSP) 34(2)
12. Maheswari M, Murugeshwari B (2018) Random and triple burst error correction code
with low redundancy for network-on-chip link. In: International Conference on Computer
Communication and Informatics (ICCCI), Coimbatore, 4–6 Jan 2018
13. Lin J, Chen K, Niu K (2013) Improved successive cancellation decoding of polar codes. In:
IEEE transactions on communications, 61(8)
PAPR Reduction in F-OFDM
Modulation Scheme for 5G Cellular
Networks Using Precoding Technique

Sasidharan Jiji and M. Ponmani Raja

Abstract The emerging 5G system has promising advances in the near future. The
5G system will be offering many features that are not being possessed in the past
generations. To meet the several requirements, orthogonal frequency division mul-
tiplexing (OFDM) is a best choice. In the existing OFDM technique, model and
framework of 4G LTE, chosen mainly for mobile broadband (MBB) service, are not
that sensitive to recession or authenticity. Despite the fact that OFDM provides high
spectrum efficiency through orthogonal frequency multiplexing, the OOBE (out-of-
band emission) of OFDM is still not very satisfactory, and also, OFDM requires
global synchronization which comes at the price of extra signaling. Indeed, when the
user is perfectly synchronized both in time and frequency domain with the base sta-
tion (BS), in terms of bit error rate (BER), the performance offered by OFDM is very
good and resistance to the carrier frequency offset (CFO). These circumstances are
energy costly as the user needs to exchange messages with the BS to ensure this syn-
chronization. Therefore, if these conditions are not satisfied, the OFDM BER may
be high. Nevertheless, the OFDM modulation suffers from high side lobes which
decrease the spectral efficiency and create adjacent channel interferences. For these
reasons, several MCM schemes have been developed these recent years as candidates
for 5G systems such as filtered OFDM (F-OFDM). Filtered OFDM (F-OFDM) is an
alternative to the OFDM modulation in 5G system. It offers all the advantages main-
tained by OFDM such as efficient performance and flexible frequency multiplexing
that meets the needs of future generation. It also meets OOBE requirements and thus
helps in efficient spectrum utilization.

Keywords Adjacent channel interference (ACI) · Bit error rate (BER) ·

Complementary cumulative distribution function (CCDF) · Cyclic prefix (CP) ·
Fast Fourier transform (FFT) · Filtered orthogonal frequency division multiplexing
(F-OFDM) · Inter-channel interference (ICI) · Inter-symbol interference (ISI) ·

S. Jiji (B) · M. Ponmani Raja

Electronics and Communication Engineering, Jyothi Engineering College, Thrissur, India
e-mail: jijisasidharan95@[Link]
M. Ponmani Raja
e-mail: ponmani@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1127

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1128 S. Jiji and M. Ponmani Raja

Inverse fast Fourier transform (IFFT) · Orthogonal frequency division multiplexing

(OFDM) · Out-of-band emission (OOBE) · Peak-to-average power ratio (PAPR) ·
Power amplifiers (PA) · Power spectral density (PSD)

1 Introduction

Nowadays, the field of multimedia wireless communication is growing rapidly which

influences the life of people creating unstoppable demands and needs that increase
day by day. People are in search of a technology that is faster than the current gen-
eration, which has very high-speed transmission rates, and supports mobility and
efficient utilization of the network resources and available spectrum. The current
modulation scheme used in 4G system is orthogonal frequency division multiplex-
ing (OFDM), and it is one of the best solutions to achieve these goals. OFDM is
having high spectral utilization. The basic method is encoding digital data on multi-
ple carrier frequencies. It improves the bandwidth efficiency and effectively combats
the multipath fading channel. At the same time, to provide a reliable transmission, it
increases the system capacity. In OFDM, it first splits the high-rate data stream into
a number of lower rate data streams which are then transmitted simultaneously over
a number of subcarriers. These subcarriers are overlapped with each other and are
orthogonal, where each frequency channel is modulated with simpler modulation
scheme. However, it introduces inter-symbol interferences (ISI) and inter-channel
interferences (ICI). ISI is the effect of adjacent symbols, whereas ICI is the effect
exerted by the subcarriers. In order to reduce the effect of ISI and ICI, we introduce
a guard interval in between the OFDM symbols.
One of the major drawbacks in OFDM is that a very high peak-to-average power
ratio (PAPR) is being exhibited by the composite transmit signal when the input
sequences are highly correlated. Corresponding to different phase values, the signal
will be having different values with respect to each other. The output symbol will be
having peak points in the overall envelope, when all the points achieve the maximum
value simultaneously. The uneven peaks in the OFDM system occur due to the signal
distortions. The presence of large number of subcarriers makes the peak values very
high. Peak-to-average power ratio is defined as the ratio of the peak power to average
power value of the system. PAPR and bandwidth efficiency are directly proportional,
i.e., when the number of subcarriers increases PAPR, it also increases the bandwidth
efficiency. Thus, a reduction of PAPR affects the efficiency of the system being very
crucial situation.
We are aiming for a new generation (5G) in mobile communication which will
be providing us with the new features that were not employed in 4G generations.
The main application will be provided by it are divided into three major categories:
enhanced mobile broadband (eMBB), ultra-reliable and low-latency communication
(URLLC), massive and machine-type communication (mMTC). [1] Figure 1 depicts
the applications of 5G [1].
The properties that an ideal waveform should have are the following [1]:
PAPR Reduction in F-OFDM Modulation Scheme for 5G Cellular … 1129

Fig. 1 Future applications of 5G [1]

• Low complexity—low-cost transceivers are needed for machine-type communi-

cation (MTC); thus, low complexity should be maintained.
• Good spectral containment—the use of guard bands in between the OFDM
symbols may affect the spectrum utilization; thus, spectrum for each service must
be efficiently utilized.
• Carrier frequency offset (CFO)—CFO at the receiver can be amplified by the
use of poor oscillators; thus, it will be robust against CFO.
• Support multiple input multiple output (MIMO)—MIMO is the best technol-
ogy for the future since spatial multiplexing is a multiple antenna technique; as
compared to single antenna techniques, it increases the data rate.
• Time localization—a well-localized waveform in time domain is needed for
fulfilling the demand of low latency.
• Flexible numerology—in order to satisfy different traffic types, a flexible
waveform is needed.
From these benefits, it is clear why OFDM can be considered for future generation.
Even though there are several benefits of OFDM, there also exist several limitations
which enable the generation of a new modulation scheme that can efficiently satisfy
the requirements for an efficient future generation (5G). Several limitations of OFDM
systems are as follows [1]:
• Sensitivity to frequency and time offset: Due to CFO and Doppler effect, several
impairments occur. Similarly, if the signals are not synchronized well, then it will
result in fading and irregularities of signals; hence, synchronization is an important
factor.
• Cyclic prefix overhead: The addition of cyclic prefix to reduce the interferences
in the symbol results in overhead in the transmission and performance degradation.
1130 S. Jiji and M. Ponmani Raja

• Large peak-to-average power ratio (PAPR): Larger PAPR occurs due to alter-
ations or signal distortions which will result in unwanted peaks in the resulting
signal; this will degrade the performance and efficiency of the system.
• Large out-of-band emissions (OOBE): In time domain, each OFDM symbol
has a shape of rectangular pulse which seems as a sinc function in the frequency
domain. Therefore, it results in a bad spectral behavior as the OOBE falls very
slowly and imposes the need for large guard.
From the limitations listed above, we can understand that the OFDM cannot
provide an efficient performance in the future 5G systems. However, we can think
of making use of the advantages offered by OFDM to produce a new modulation
scheme. The main goal of this paper is to provide a new waveform that imposes all
the advantages of OFDM and at the same time introduces an efficient technique to
reduce the demerit of high PAPR in it. In [2–9], we can see the several techniques
used in modulation techniques in several channels. It will provide us with a detailed
study of related works in the field of OFDM.
In Section I, we have given a brief introduction of the paper. In Section II, the liter-
ature survey related to the topic is being done. Section III mainly focuses on the pro-
posed methodology, and finally, section IV concludes the paper with corresponding
results.

2 Related Work

In [10], Vipul D. Sahni, Nitesh Kumar, and Vishal Narain Saxena had discussed
about several hybrid techniques for reducing the high PAPR in OFDM. Four main
techniques like

(1) Precodings with repeated clipping and frequency domain filtering (RCF).
(2) Companding techniques along with RCF.
(3) Repeated frequency domain filtering and clipping (RFC) along with
companding.
(4) Precodings along with companding.

are being introduced in their work. From that, they are concluding that precoding has
been considered better among all these techniques, because it improves the PAPR
without increasing more complexity and without hindering with the orthogonality
between subcarriers. It also increases the BER performance in comparison with other
stated hybrid methods because of diversity gain.
In [11], Dan Wu, Xi Zhang, Jing Qiu, Liang Gu, Yuya Saito, and Anass Benjebbour
discuss that to enable flexible waveform for 5G and improve the spectrum utilization,
a new modulation scheme, named filtered OFDM (F-OFDM), is being introduced. In
that work, they have mainly discussed about the filter design and its implementation.
In this work, they have recommended the use of sinc filter design for F-OFDM.
According to their results, they are concluding that the F-OFDM is having more
PAPR Reduction in F-OFDM Modulation Scheme for 5G Cellular … 1131

efficiency than OFDM which is currently being used. F-OFDM enhances spectrum
efficiency over the conventional OFDM by providing a reduced spectrum leakage.
In [12], Qiwei Zheng, Fanggang Wang, Xia Chen, Yinsheng Liu, Deshan Miao,
and Zhuyan Zhao focus on performance evaluation of waveforms in the fifth gener-
ation, i.e., universal-filtered multicarrier technique (UFMC), resource block-filtered
orthogonal frequency division multiplexing (RB-F-OFDM), filtered OFDM (F-
OFDM), and filter bank multicarrier (FBMC) in high-speed scenarios. With respect
to signal-to-noise ratio (SNR), terminal speed, and carrier frequency offset (CFO),
the block error rates (BLERs) of the channel model of tapped-delay-line-D (TDL-D)
defined in 3GPP TR 38.900 are evaluated. Simulation results indicate that RB-F-
OFDM is recommended for high-speed scenario due to the robustness in terms of
high mobility and CFO. Main disadvantage is the high peak-to-average power ratio.
A single-input and single-output (SISO) system is considered; here, the principle
of filtering-based waveforms UFMC, RB-F-OFDM, F-OFDM, and FBMC are pro-
posed for future 5G and compared their performance in high-speed scenario, using
the TDL-D channel model in 3GPP TR 38.900. Simulation results show that RB-
F-OFDM outperforms other waveforms in terms of high mobility and robustness
against ISI and CFO.
In [13], Lei Zhang, Ayesha Ijaz, Pei Xiao, Mehdi Molu, and Rahim Tafazolli
propose that F-OFDM system is an efficient modulation scheme for 5G system. In
their work, they are concentrating on the mathematical model of F-OFDM, and then,
they have discussed about how to reduce the interferences in the system, thus leading
to equalization. Then, they aimed on reducing the computational complexity of the
system. The efficient obtained results reduce the computational complexity of the
subcarriers.
From the above discussed works, we can conclude that F-OFDM is an efficient
modulation scheme for efficient communication of the future generations. There still
persists the demerit of high peak-to-average power ratio. From [10], it is clear that
the efficient technique for reducing the high PAPR in OFDM is precoding technique;
thus, we can implement the precoding technique for reducing the PAPR in F-OFDM.

3 Proposed Work

In order to fulfill the need for 5G cellular communication, we are hereby implement-
ing a new modulation scheme F-OFDM. Basic difference between the OFDM and
F-OFDM is that there is filtering at both the transmitter and the receiver sections.
The block diagram of F-OFDM is given in Fig. 2.
In filtered OFDM (F-OFDM), a filtering algorithm is used after the IFFT/cyclic
prefix of OFDM system; the entire assigned bandwidth is first divided into several
sub-bands. In each sub-bands, the different services are being accommodated as
per the requirements. Compared to conventional OFDM technique, the F-OFDM
provides throughput gain. Significant reductions on usage of guard band are also
being offered in F-OFDM which leads to more efficient spectrum utilization. Even
1132 S. Jiji and M. Ponmani Raja

Fig. 2 Block diagram of F-OFDM

though it is having several advantages, still a demerit that is high PAPR which will
affect the performance of the system persists. Thus, as per [10], we are proposing a
precoding-based technique to reduce the PAPR in F-OFDM modulation scheme.

3.1 Filter Design

F-OFDM should satisfy certain criteria for achieving appropriate filtering. It includes
flat passband, sharp transition band, and sufficient stop-band attenuation. Thus, these
criteria can be fulfilled by using a filter with a rectangular frequency response, i.e.,
a sinc impulse response; hence, a soft truncated filter with specific window is rec-
ommended to achieve the appropriate filtering in F-OFDM. This in turn will create
a trade-off between time and frequency localization (i.e., ISI and ICI). It is an easy
implementation for flexible sub-band configuration. From Fig. 2, we can see that
in F-OFDM, the input signal is first sent to the N-point IFFT which is a serial-to-
parallel converter, and then, it is added with cyclic prefix to reduce the interferences
in the symbol. After that, we are applying designed filtering to the system. This is
the transmitter section; output from filtering is being sent through the channel to the
receiver and a similar filtering is being applied at the receiver part. Then, the reverse
operation of transmitter part is being employed at the receiver such as removal of
cyclic prefix followed by N-point FFT; then finally, equalization is being employed.
The filtering algorithm employed enables the spectrum efficiency, improves latency,
and reduces the out-of-band emission. Even though F-OFDM has all the advantages
of OFDM, still there persists the high PAPR ratio. Thus, to reduce the PAPR, we can
use the precoding technique [10] which is very efficient. From the papers [10, 14–
20], we can get a detailed comparison of several techniques for reduction of PAPR
in OFDM, and from that, we can understand that a very efficient technique for PAPR
reduction is precoding technique.
PAPR Reduction in F-OFDM Modulation Scheme for 5G Cellular … 1133

Fig. 3 Basic block diagram of precoding technique

3.2 Precoding Technique

In precoding technique, we will multiply the modulated symbols of OFDM and

F-OFDM with the precode matrix before IFFT block, and this precode matrix P
is of a dimension of L × N where L = N + N p , N is the number of baseband
modulated signals, and overhead subcarriers as 0 ≤ N p < N. When an input matrix
‘x’ is multiplied to a precoding matrix P, then the output Y will be Y = P ⊗ X. Figure 3
implies the block diagram of precoding technique. It depicts that the precoder is being
used after the serial-to-parallel converter and before the IFFT block. The transpose of
the precoding matrix is being multiplied to the receiver part in precoding technique.
From [20] & [21] we can find the precoding technique beings depicted on ofdm
modulation scheme along with companding technique.

4 Result Analysis

In Fig. 4, we can see the impulse response of the rectangular pulse sinc filter used
in F-OFDM. Figure 5 depicts the power spectral density of OFDM and F-OFDM.
Table 1 depicts the values of parameters being used in this modulation technique. In

Fig. 4 Impulse response of

sinc filter
1134 S. Jiji and M. Ponmani Raja

Fig. 5 PSD of OFDM and F-OFDM

Table 1 Parameters required

Parameters Values
in the modulation schemes
Number of FFT points 1024
Number of resource blocks 50
Number of subcarriers per resource block 12
Cyclic prefix length 72
Modulation 64 QAM
Channel model AWGN
Filter length 513
Tone offset or excess bandwidth (in subcarriers) 2.5

Fig. 6, we can see the CCDF plot of PAPR for F-OFDM and OFDM before applying
precoding technique, whereas Fig. 7 depicts the CCDF plot of PAPR for F-OFDM
and OFDM after applying precoding technique. Table 2 shows the results of PAPR in
both OFDM and F-OFDM before applying precoding technique and after applying
precoding technique.
Comparing the plots of the spectral densities for OFDM and F-OFDM schemes,
we can see that F-OFDM has lower side lobes that allow a higher utilization of the
allocated spectrum, hence leading to increased spectral efficiency. It also depicts that
the OOBE is reduced in F-OFDM as compared to the OFDM.
From Table 2, it is very clear that the PAPR for OFDM is very large compared
to F-OFDM even after applying precoding technique; thus, F-OFDM is an efficient
mechanism for 5G systems.
PAPR Reduction in F-OFDM Modulation Scheme for 5G Cellular … 1135

Fig. 6 CCDF plot of PAPR for OFDM and F-OFDM without precoding technique

Fig. 7 CCDF plot of PAPR for OFDM and F-OFDM with precoding technique
1136 S. Jiji and M. Ponmani Raja

Table 2 Final values of

Parameter Value in dB
PAPR in dB
PAPR of OFDM without precoding 11.0173
PAPR of F-OFDM without precoding 8.3904
PAPR of OFDM with precoding 8.8571
PAPR of F-OFDM with precoding 5.2942

Figures 8, 9, 10, and 11 depict the BER versus SNR in dB plots. BER values range
in between 10−2 and 10 which is an optimum value for efficient transmission rates.

Fig. 8 BER V/S SNR plot

of OFDM without precoding
technique

Fig. 9 BER V/S SNR plot

of OFDM with precoding
technique
PAPR Reduction in F-OFDM Modulation Scheme for 5G Cellular … 1137

Fig. 10 BER V/S SNR plot

of F-OFDM without
precoding technique

Fig. 11 BER V/S SNR plot

of F-OFDM with precoding
technique

References

1. Kodheli O OFDM-based schemes for next generation wireless systems. PhD diss.
2. Larsen Y, Leus G, Giannakis GB (2004) Constant modulus and reduced PAPR block differential
encoding for frequency-selective channels. IEEE Trans Commun 52(4):622–631
3. Kim Y-J, Kwon U-K, Seol D-Y, Im G-H (2009) An effective PAPR reduction of SFBC-OFDM
for multinode cooperative transmission. IEEE Sig Process Lett 16(11):925–928
4. Wang C-L, Ku S-J, Yang C-J (2010) A low-complexity PAPR estimation scheme for OFDM
signals and its application to SLM-based PAPR reduction. IEEE J Sel Top Sig Process 4(3):637–
645
5. Sohn I (2014) A low complexity PAPR reduction scheme for OFDM systems via neural
networks. IEEE Commun Lett 18(2):225–228
6. Lasya PR, Kumar MS (2015) PAPR and out-of-band power reduction in OFDM-based cog-
nitive radios. In: 2015 International conference on Signal Processing and Communication
1138 S. Jiji and M. Ponmani Raja

Engineering Systems (SPACES). IEEE

7. Ogunkoya FB, Popoola WO, Shahrabi A, Sinanovi’c S (2015) Performance evaluation of pilot-
assisted PAPR reduction technique in optical OFDM systems. IEEE Photonics Technol Lett
27(10):1088–1091
8. Khan MA, Rao RK (2016) Low-complexity PAPR reduction technique for OFDM systems
using biased subcarriers. Can J Electr Comput Eng 39(1):19–25
9. Hossain MS, Shimamura T (2016) Low-complexity null subcarrier-assisted OFDM PAPR
reduction with improved BER. IEEE Commun Lett 20(11):2249–2252
10. Sahni VD, Kumar N, Saxena VN (2016) Novel hybrid technique (s) for PAPR reduction in
OFDM systems. In: 2016 International Conference on Signal Processing and Communication
(ICSC). IEEE
11. Wu D, Zhang X, Qiu J, Gu L, Saito Y, Benjebbour A, Kishiyama Y (2016) A field trial of
f-OFDM toward 5G. In: 2016 IEEE Globecom Workshops (GC Wkshps), pp 1–6
12. Zheng Q, Wang F, Chen X, Liu Y, Miao D, Zhao Z (2017) Comparison of 5G waveform candi-
dates in high speed scenario. In: General assembly and scientific symposium of the international
Union of Radio Science (URSI GASS), 2017 XXXIInd. IEEE
13. Zhang L, Ijaz A, Xiao P, Molu MM, Tafazolli R (2018) Filtered OFDM systems, algorithms,
and performance analysis for 5G and beyond. IEEE Trans Commun 66(3):1205–1218
14. Ermolova N (2002) A comparison of two schemes for peak-to-average power ratio reduction in
a multicarrier transmission. In: Proceedings of ICCSC’02. 1st IEEE international conference
on circuits and systems for communications, 2002. IEEE
15. Xin Y, Fair IJ (2005) Low complexity PTS approaches for PAPR reduction of OFDM signals.
In: 2005 IEEE International Conference on Communications, 2005. ICC 2005, vol 3. IEEE
16. Yang L, Chen RS, Siu YM, Soo KK (2006) PAPR reduction of an OFDM signal by use of PTS
with low computational complexity. IEEE Trans Broadcast 52(1):83–86
17. Jiang T, Ni C, Guan L (2013) A novel phase offset SLM scheme for PAPR reduction in Alamouti
MIMO-OFDM systems without side information. IEEE Sig Process Lett 20(4):383–386
18. Badran EF, El-Helw AM (2011) A novel semi-blind selected mapping technique for PAPR
reduction in OFDM. IEEE Sig Process Lett 18(9):493–496
19. Sohn I, Kim SC (2015) Neural network based simplified clipping and filtering technique for
PAPR reduction of OFDM signals. IEEE Commun Lett 19(8):1438–1441
20. Agarwal D, Sharan N, Raja MP, Agarwal A (2015) PAPR reduction using precoding and
companding techniques for OFDM systems. In: 2015 International Conference on Advances
in Computer Engineering and Applications (ICACEA). IEEE
21. Lakamana AS, Prasad AM (2016) An effective composite PAPR reduction technique of OFDM
by using DFT precoding with piecewise linear companding. In: International Conference on
Communication and Electronics Systems (ICCES). IEEE
Privacy Preservation Using Top K
Multi-Keyword Synonym Fuzzy Search
Scheme

Manjubala Sekar and Kamalanathan Kandasamy

Abstract Owing to the numerous advantages arising out of storing data in the cloud
such as flexible costs, improved mobility, many users are outsourcing their data to the
cloud. However, it poses security challenges and rigidity issues. The cloud is often
honest but curious, and hence, the need to encrypt the data arises. However, it renders
the basic processes like searching difficult. Hence, there is a need to implement search
algorithms on encrypted data. Plaintext fuzzy search and semantic search techniques
cannot be implemented on encrypted data. To save its resources, the cloud may return
partially correct results, and thus, there is a need for verification of the results returned.
Access control mechanisms for multiple users should be implemented ensuring the
confidentiality of unauthorized data. This paper deals with the design and analysis
of a privacy-preserving top k multi-keyword synonym/similarity fuzzy search.

Keywords Fuzzy search · Cloud computing · Privacy preserving · Synonym ·

Multi-keyword

1 Introduction

Advancements in cloud computing have made storing data in cloud a popular choice.
Numerous benefits include reduced costs in provisioning whenever storage is needed
and de-provisioning when it is not needed, reduction in data center footprint, avail-
ability of data anytime from anywhere and more. Cloud is of three types: public
cloud, private cloud and hybrid cloud. Private cloud stores sensitive information,
and hybrid cloud contains both public and private cloud services. Services provided
by the cloud are Infrastructure as a Service (IaaS), Platform as a Service (PaaS),
Software as a Service (SaaS) and Recovery as a Service. IaaS provides pre-installed

M. Sekar (B) · K. Kandasamy

Amrita Center for Cyber Security Systems & Networks, Amrita Vishwa Vidyapeetham, Amrita
University, Kollam, Kerala, India
e-mail: manjubalasekar@[Link]
K. Kandasamy
e-mail: kamalanathan@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1139

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1140 M. Sekar and K. Kandasamy

and configured hardware through a virtualized interface. PaaS provides computing

platform and solution stack in addition to IaaS services. SaaS and RaaS provide
web-based applications and backup service, respectively.
However, storing data in a remote server is not fully secure. The cloud is honest
but curious. Sensitive information must remain integral and confidential. Hence,
the need to encrypt the data but processes like searching becomes difficult. The
existing solutions are homomorphic encryption of the data, oblivious RAM and
search encryption to search through encrypted data. Search encryption is so far the
best approach. The traditional and secure method of searching was to extract all the
files from the cloud, decrypt the files and conduct plaintext search on the decrypted
files. This will be a waste of computation power, decryption of all the encrypted files.
Thus, we must employ techniques to search through encrypted data. There are
three techniques, (i) search in the encrypted data (ii) secret sharing and (iii) index-
based approach as discussed by Brinkman [1]. In the first approach, all the files and
the keywords are encrypted, and the encrypted keyword is searched through every
line of the encrypted files. In secret sharing method, the cryptographic secret is split
into m secrets such that any n out of m secrets can be used to construct the original
secret. In this context, for example, if 7 is the data and it is split among three owners
as 32,5,11 and 7 is obtained by adding them and applying modulo 41 over them. In
the index-based approach, keywords are extracted from the files before encryption,
and secure index is constructed. Both the index and files are encrypted, and when a
user sends a query, the keywords are encrypted, and a match is searched for in the
index. This reduces search time in searching over the entire file documents.
The basic entities are data owner, data user and the cloud server. The data owner
uses the key extraction algorithm to extract the relevant keywords. Then, he/she
encrypts the keywords to form the secure index along with the file ids and send it to
the cloud server. When the data user sends a search request, the query is encrypted,
and if it matches with the keywords in the secure index, then the encrypted files
corresponding to the file ids are returned to the user (Fig. 1).
The rest of the paper is organized as follows. Section 2 deals with the related
work, Sect. 3 explains the methodology followed in this paper, and Sect. 4 provides
the implementation details. The results of our experiments are given in Sect. 5 and
followed by conclusion in Sect. 6.

Fig. 1 Index-based
approach general procedure
Privacy Preservation Using Top K Multi-Keyword Synonym Fuzzy … 1141

2 Related Work

Fuzzy Search: Since the data is encrypted along with keywords, a spelling mistake
in the keyword will be encrypted to a different value and the search fails, i.e., there
is no tolerance for typing errors. Words which originate from the same stem are
also encrypted to different values, and the search fails unless the exact keyword
is specified. In wild card-based approach [2], the possible fuzzy words are formed
with a pre-defined edit distance. For example, the wild card fuzzy set of ‘cloud’
with edit distance 1 will be {*cloud, *loud, c*loud, c*oud, cl*oud, cl*ud, clo*ud,
clo*d, clou*d, clou*, cloud*, cloud}. The possible combinations are n *2 +2 where
n is the length of the keyword. In the K-gram-based approach [3–5], the keywords
are split into k-grams and the encrypted individually in the index. A search query
containing the same or almost similar keyword will have more k-grams in common.
For example, the bi-gram representation of the word ‘secure’ is {se, ec, cu, ur, re}.
Symbol-based trie-traverse search scheme [6] has trapdoors of the keywords mapped
to a pre-defined symbol set, and it is inserted into trie-based tree where the root node
is null. The keywords with the common prefixes are grouped together.
Another technique involving bloom filters and LSH is described as follows [5,
7]. This eliminates the need for a dictionary, and the index for each file is a bloom
filter. Every keyword is converted into a bi-gram set. There are 26 *26 possible bi-
grams, and the bit is set to one if a bi-gram is present. If there is a single spelling
mistake, only two bi-grams will vary. The vector is then hashed using LSH from p-
stable LSH family. This hashing ensures that two keywords with one letter difference
are mapped to a hash value of almost similar value. This similarity is measured by
Euclidean distance. The vector is then inserted into the bloom filter. The query is
also generated in the same way. When the user sends a query, the inner product
of the query and the index will give a high value as the vectors will be set to one
in similar positions. This scheme also supports dynamic update, new files can be
inserted, and files can be deleted as each file is indexed separately. The disadvantage
of this scheme is false positives introduced by bloom filter which can be mitigated
by setting a minimum threshold for the inner product and at the same time ensuring
that the relevant files are not missed out.
In another fuzzy method, it is assumed that the user may omit, swap, include up
to two characters maximum, and hence, the positioning of letters in the keyword can
be moved up or down by a maximum of two [8]. The example for positioning in case
of one letter addition, omission or swapping is as follows for the word ‘secure’ {s0,
s1, e1, e2, e3, c2, c3, c4, u3, u4, u5, r4, r5, r6, e5, e6, e7}. When the position of ‘c’
comes in either second, third or fourth position, the letter is mapped like ‘seecure.’
Semantic Search: In a query, certain keywords should be given higher impor-
tance, certain words are general terms, and some are specific. For example, if the
query is ‘blue pants,’ it means pants are more generic to blue. First, the results con-
taining pants should be found and narrowed down to blue pants. To understand and
implement this, semantic relations must be studied, and search made on that basis.
Synonym search is essential as users may search for files with different keywords
1142 M. Sekar and K. Kandasamy

but with the same meaning. For example, university/college both refer to a institu-
tion, and hence, both the words should form the keyword set for the files containing
any one of the keyword. Improving fuzzy search [9, 10] instead of discarding the
remaining elements or the fuzzy set of the keyword when the exact match succeeds,
the trapdoors of the fuzzy set of the keyword are also formed to search for similar
words.
Porter’s stemming algorithm [5] is used to extract the root word from which the
derivatives can be constructed, thus enabling more matches. For example, the root
word of ‘detected’ will be ‘detect,’ and other words derived from the root word
are ‘detectable,’ ‘detects,’ ‘detecting,’ ‘detective.’ The above words are then used
to construct the keyword set with the root word as the head node. The assigning of
keyword weights [11] will ensure the importance of a keyword in the search query,
and thus, a higher score for the document containing keyword will be calculated.
Conceptual hierarchy [12] is a tree-like structure where the leaf nodes will have a
more specific meaning, and the nodes at the top will have a general meaning. E.g.,
‘college,’ ‘institution,’ ‘university’ will occupy the higher level of nodes, and ‘Amrita
school of engineering’ and ‘Kollam’ will occupy the lower level of nodes. Another
method is applying sentence scoring followed by conceptual graphs [13]. First, the
most relevant sentence is extracted from the document known as sentence scoring.
A conceptual graph consists of objects and relations, and every object is connected
to another object by a relation.
Preserving Privacy: Apart from encrypting the data in the files, it is also necessary
to protect the privacy of the keywords and other statistical data associated. Else, the
cloud will be able to deduce which set of files are accessed more frequently and
which files are sensitive. The keyword is split [14, 15] into two and then encrypted
such that even when the attacker obtains the trapdoor, he/she will not be able to
determine the keyword. Smaller keywords will be subjected to keyword guessing
attack, and hence, randomizing [16] and increasing the keyword length will secure
it against keyword guessing attacks, and similarly, dummy keywords [17] are also
introduced into the secure index. To protect against the cloud from determining the
words in the encrypted files, the same keywords in various files are encrypted to
different cyphertexts [18].
Verifiable Results: Since the cloud is not a trusted entity, the results returned
by the server should be authenticated. Sometimes, to save its resources, the server
may not complete its search process and may return partial results. Check mecha-
nisms should be employed to ensure the authenticity of the data. In this challenge-
based verification technique [19], the user composes a challenge from the returned
results and sends it to the cloud. If the cloud can solve the challenge, thus verifying
the authenticity of the ciphertext, the ciphertext is accepted by the user else, it is
discarded.
In another random challenge technique [20], similarity scores are asked for any
random documents. If the cloud has returned the top k results, the similarity scores
of the random documents requested by the user should be less than the similarity
scores of the top k documents. In this technique [21], a verifiability code is generated
for each file based on the file and the similarity score. The cloud will return the set
Privacy Preservation Using Top K Multi-Keyword Synonym Fuzzy … 1143

of files along with the VC code when a request has been made. The data user will
reconstruct the VC from the file and the shared secret key and verify if it matches
with the returned VC code.
Ranking of Results: When a user sends a query, certain results may be more
relevant than some, and hence, the results should be ordered with the most relevant
results at the top. Thus, the need for similarity score arises. The scores are usually
calculated based on how unique yet frequent the word appears in the document
relative to other documents. Some common algorithms are TF-IDF, Cosine similarity
[22] and Co-ordinate matching.
Dynamic Update: The search index should facilitate dynamic update [7, 23] as
new files might be added, and old files might be replaced and modified. This will
lead to a change in keywords and similarity scores. Hence, the secure indices must
support dynamic update.
Multi-user Access Control: There are two types KBAC and ABAC. KBAC
assigns each file’s decryption key directly to authorized users. ABAC attaches a set of
attribute values to a user and designs access policy for a file. A file can be accessed if
and only if the attribute values satisfy the access policy. Predicate encryption [24] is
a very powerful technique for enforcing fine-grained access control where the owner
of the ciphertext can provide partial keys to the users for decrypting partial data.
In this paper, LCS algorithm is applied initially followed by calculation of distance
between the words returned to obtain the most relevant word (Fig. 2).

Fig. 2 Architecture
1144 M. Sekar and K. Kandasamy

3 Methodology

3.1 Model Description

The components are data owner, data user, search server and storage server. The data
owner extracts all the keywords from the files and forms synonym/similar keywords.
He/She forms a linked list of the keywords with the keyword found in the file at
the top of each linked lists (dictionary). It will have m distinct keywords, and n
is the highest number of elements in the synonym and fuzzy set of each keyword.
The data owner extracts the frequency of the keywords from the data files as well.
He/She encrypts the keyword and the frequency to form the secure index (trie-based
structure) and sends it to the search server along with file ids. The file ids, encrypted
files and the access control of each file are sent to the storage server. The data user
encrypts his query using public key of the data owner and sends the query to the
data owner. The data owner decrypts the query using his private key and tries to find
a match in the dictionary. For fuzzy search, the LCS algorithm is used to find the
edit distance between two words. If there is a match with any keywords, i.e., exact
match or fuzzy match within the specified edit distance, trapdoors of the keyword,
synonyms and similar words are computed and sent to the search server. The search
server compares the trapdoors are compared with the index table, and the file ids are
returned.
The top k results are obtained by calculating the frequency of the keywords using
term frequency * inverted document frequency (TF*IDF). The files which have all
the keywords of the query are at the top even if its individual frequencies are com-
paratively lesser than other files having higher frequency but not all the keywords
of the query. The data user receives the file ids and sends the file ids to the storage
server to obtain the encrypted files. The storage server sends the encrypted files that
match with the file ids to the user after verifying the access controls of the user. The
data user will decrypt the received file.

3.2 Assumptions

This model is applicable where there is a single owner like a corporate company
where the company-related projects are to be made available to only its employees.
Here, the encryption of keywords does not vary, i.e., the same keyword will be
encrypted to the same value, and hence, the cloud may reduce the frequency of the
encrypted keywords queried. Similarly, the cloud might deduce the files which are
accessed frequently and form a relationship between the encrypted keywords and the
encrypted files.
Privacy Preservation Using Top K Multi-Keyword Synonym Fuzzy … 1145

3.3 Algorithm Description

Hirschberg’s Algorithm: This algorithm finds LCS of two strings in O(mn) time com-
plexity and O(m + n) space by applying divide and conquer method. This follows the
divide and conquer method where the strings are split recursively, and string compari-
son is done. In the above example, the strings are ‘TATGC’ and ‘AGTACGCA’ where
the second string is split into two, and the following method is done to determine
where to split the first string to obtain LCS sequence.
This algorithm saves space as it does not store the entire table but just two rows at
any time. LCS algorithm is applied on the first half of the string as it is and applied
on the reversed half of the second string as shown in the figure above. Here, it is
assumed that if the characters match, a value if 2 is added, and if does not match, −2
is added in case of insertion and deletion and −1 in case of substitution. Whether a
character must be inserted, deleted or substituted is done by comparing the diagonal,
top and left values, and the minimum is taken for the operation, i.e., −2 is added if
the top or left value is minimum, and −1 is added if the diagonal value is minimum.
Finally, the last two rows are aligned, and the subsequent columns are added. The
column which yields the minimum values is taken as the divider. Therefore, the LCS
of the above two strings is TAGC (Figs. 3 and 4).

Fig. 3 Hirschberg’s algorithm

Fig. 4 Divide and conquer

LCS algorithm output
1146 M. Sekar and K. Kandasamy

4 Implementation Details

‘Reuters-21578, Distribution 1.0’ comprising over 20,000 documents (appeared on

the Reuters newswire in 1987) was encrypted and used as a dataset and used Python.
For the formation of the index, the keywords formed are nouns and noun-phrases.
Stemming algorithm is applied to increase the match frequency. The ranking of the
results is done based on TF-ID where

TF − IDF = TF ∗ IDF;

TF = Frequency of words in the text

IDF= 1/log (1 + p); p = Total documents/Number of documents containing the
keywords.
A split-keyword matrix structure as described in [6] is constructed with key-
words, TF-IDF value and, the file ids. When a keyword is encrypted, the attacker can
reconstruct the plaintext of the encrypted keyword by brute force. If it is split and
encrypted, the attacker may not be able to reconstruct the keyword. For example,
unive, 5, (0.2215879871480701, ‘1_49’), (0.4238681120052753, ‘1_49’), rsity, 5.
The index is then encrypted using AES-CBC ad stored in the search server. The
synonym set is formed using WorldNet, and it is the matrix containing the keyword
present in the file in the first column followed by the synonyms. For the search pro-
cess, the query to be sent to the owner is encrypted using the RSA algorithm. The
received query is decrypted at the owner side, and pre-processing is done where fuzzy
search, construction of synonym sets and stemming process. The trapdoor is formed
and sent to the search server. The file ids of the matched trapdoor are returned from
the search server, and the corresponding files are obtained from the storage server.

5 Results

The time taken for generating the search index was studied, and it is found to increase
linearly as shown in the graph below. Using the TF-IDF algorithm is inefficient when
the number of documents increases for every keyword, the whole document set must
be checked.
Using an intermediate server to correct spelling mistakes and the formation of
synonym set increases the accuracy rate. However, using the LCS algorithm has its
disadvantages as explained below.
The algorithm described is inefficient in differentiating two words which are
similar but semantically different. In the table below, it is shown that the distance
between the actual word and the word that forms after applying LCS is the same.
However, it is evident that the user meant to search for ‘cloud.’ It observed that the
edit distance of the first entry is four compared to five in the second entry. Therefore,
Privacy Preservation Using Top K Multi-Keyword Synonym Fuzzy … 1147

Table 1 Disadvantage of
Actual word Entered word Applying LCS
LCS algorithm
Cloud Cluod Clud
Loud Cluod lud

Fig. 5 Index generation

time versus number of files

the results should be returned such that results related to ‘cloud’ to be returned first
followed by results related to ‘loud’ (Table 1).
This can be improved by studying the edit distance between the words formed after
applying the LCS algorithm and the entered word. It is observed that the difference
is one in the first case and two in the second case. Hence, it can be concluded that
the entered word was supposed to be ‘cloud’ and not ‘loud’ (Fig. 5).

6 Conclusion and Future Work

This system model improves fuzzy search and preserves privacy by the combination
of split-keyword technique, and the query is transmitted after encrypting with RSA.
The usage of two separate servers will prevent the cloud from forming a relationship
between the search query and the files, and usage of the LCS algorithm followed
by calculation of edit distance between the actual word and the returned words will
return the most relevant result. The runtime of LCS algorithm can be improved by
using Kuo-Cross algorithm whose runtime is O((r + n) log(n)). In the future, this
system will be modified to see if the cloud returns the correct results and enables
multi-user access.
1148 M. Sekar and K. Kandasamy

References

1. Brinkman R (2007) Different search strategies on encrypted data compared. Data-centric sys-
tems and applications. Security, privacy, and trust in modern data management, Part 3, pp
183–196
2. Li J, Wang Q, Wang C, Cao N, Ren K, Lou W (2010) Fuzzy keyword search over encrypted
data in cloud computing. IEEE INFOCOM. San Diego, CA
3. Zhou W, Liu L, Jing H, Zhang C, Yao S, Wang S (2013) K-gram based fuzzy keyword search
over encrypted cloud computing. J Softw Eng Appl 6(1):29–32
4. Wang D, Fu S, Xu M (2013) A privacy-preserving fuzzy keyword search scheme over encrypted
cloud data. In: 2013 IEEE 5th international conference on cloud computing technology and
science, Bristol, pp 663–670
5. Wang B, Yu S, Lou W, Hou YT (2014) Privacy-preserving multi-keyword fuzzy search over
encrypted data in the cloud. In: IEEE INFOCOM 2014—IEEE conference on computer
communications, Toronto, ON, pp 2112–2120
6. Raghavendra S, Girish S, Geeta CM, Buyya R, Venugopal KR, Iyengar SS, Patnaik LM (2015)
IGSK: index generation on split keyword for search over cloud data. In: 2015 International
Conference on Computing and Network Communications (CoCoNet), Trivandrum, pp 374–380
7. Fu Z, Wu X, Guan C, Sun X, Ren K (2016) toward efficient multi-keyword fuzzy search
over encrypted outsourced data with accuracy improvement. IEEE Trans Inf Forensics Secur
11:2706–2716
8. Ahsan MAM, Chowdhury FZ, Sabilah M, Wahab AWBA, Idris MYIB (2017) An efficient fuzzy
keyword matching technique for searching through encrypted cloud data. 2017 International
Conference on Research and Innovation in Information Systems (ICRIIS), Langkawi
9. Wang C, Wang Q, Ren K (2011) Towards secure and effective utilization over encrypted
cloud data. In: 31st international conference on distributed computing systems workshops, pp
282–286
10. Wang C, Ren K, Yu S, Urs KMR (2012) Achieving usable and privacy-assured similarity search
over outsourced cloud data. IEEE INFOCOM, Orlando, FL, pp 451–459
11. Shen Z, Shu J, Xue W (2018) Preferred search over encrypted data. Front Comput Sci
12(3):593–607
12. Fu Z, Xia L, Sun X, Liu AX, Xie G (2018) Semantic-aware searching over encrypted data for
cloud computing. IEEE Trans Inf Forensics Secur 13(9):2359–2371
13. Fu Z, Huang F, Sun X, Vasilakos A, Yang C (2016) Enabling semantic search based on
conceptual graphs over encrypted outsourced data. In: IEEE transactions on services computing
14. Raghavendra S, Girish S, Geeta CM, Buyya R (2018) Split keyword fuzzy and synonym search
over encrypted cloud data. Multimedia Tools Appl 77:10135
15. Li J, Wang Q, Wang C, Cao N, Ren K, Lou W (2009) Enabling efficient fuzzy keyword search
over encrypted data in cloud computing. IACR Cryptology ePrint Arch 593
16. Ahsan MAM, Idna Bin Idris MY, Bin Abdul Wahab AW, Ali I, Khan N, Al-Garwi MA, Rahman
AU (2018) Searching on encrypted e-data using Random Searchable Encryption (RanSCrypt)
Scheme. Symmetry 10:161
17. Shen Z, Shu J, Xue W (2017) Keyword search with access control over encrypted cloud data.
IEEE Sens J 17(3):858–868
18. Khan NS, Krishna CR, Khurana A (2014) Secure ranked fuzzy multi-keyword search
over outsourced encrypted cloud data. In: 2014 International Conference on Computer and
Communication Technology (ICCCT), Allahabad, pp 241–249
19. Miao Y, Ma J, Wei F, Liu Z, Wang XA, Lu C (2016) VCSE: verifiable conjunctive keywords
search over encrypted data without secure-channel. In: Springer Science + Business Media
New York
20. Wan Z, Deng RH (2018) VPSearch: achieving verifiability for privacy-preserving multi-
keyword search over encrypted cloud data. In: IEEE transactions on dependable and secure
computing, vol 15, issue no 6, pp 1083–1095, 1 Nov–Dec 2018
Privacy Preservation Using Top K Multi-Keyword Synonym Fuzzy … 1149

21. Dai H, Zhu X, Yang G, Yi X (2017) A verifiable single keyword top-k search scheme against
insider attacks over cloud data. In: 2017 3rd international conference on Big Data Computing
and Communications (BIGCOM), Chengdu
22. Fu Z, Sun X, Xia Z, Zhou L, Shu J (2013) Multi-keyword ranked search supporting synonym
query over encrypted data in cloud computing. In: 2013 IEEE 32nd International Performance
Computing and Communications Conference (IPCCC), San Diego, CA, pp 1–8
23. Xia Z, Wang X, Sun X, Wang Q (2016) A secure and dynamic multi-keyword ranked search
scheme over encrypted cloud data. IEEE Trans Parallel Distrib Syst 27(2):340–352
24. Katz J, Sahai A, Waters B (2008) Predicate encryption supporting disjunctions, polynomial
equations, and inner products. In: Smart N (eds) Advances in cryptology—EUROCRYPT 2008,
EUROCRYPT
High-Speed Variable-Length Decoder
Using Segmented Code book

Sujata Bhavikatti and R. M. Banakar

Abstract Various communication applications from Internet video streaming to

video broadcasting require high-speed throughput variable-length code decoder. The
fundamental techniques (parallelism and pipelining) used to improve the speed of
algorithm cannot be applied to VLC decoder, due to variable length of symbol codes
and due to no boundary defined between symbols. Considering the basic algorithm of
VLC decoder, the speed of decoder is improved by reducing the processing time while
identifying the code word and corresponding code length from lookup table. The
rearrangement of code words in code book is proposed in this paper to minimize the
time. For variable-length code of 17 bits, the proposed design depicts an improvement
of 46% in speed compared to traditional single length code book.

Keywords VLC decoder · Code book · Lookup table

1 Introduction

The video from Internet, personnel devices, television, movies, etc., use compressed
form of digital video. Video codec is the basic building blocks of various com-
munication applications including health monitoring systems, defense services, and
multimedia services. These applications crave for high resolution, high frame rate
which necessitates larger bandwidth. Video compression is essential to reduce the
size of video, so that storage requirements are brought down to a manageable level.
Even the video data can be brought to channel capacity level. Basically, there exist
two types of video compression techniques: (a) Lossy type of compression is used
in Internet-connected devices and telecommunication applications. The data that is
produced after decompression does not exactly match the original form; instead, data

S. Bhavikatti (B)
Tontadarya College of Engineering, Gadag, India
e-mail: sujatabhavikatti@[Link]
R. M. Banakar
BVB College of Engineering & Technology, Hubli, India
e-mail: banakar@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1151

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1152 S. Bhavikatti and R. M. Banakar

is reconstructed in a useful form. (b) Lossless compression technique is based on the

removal of redundant data so that decompression gives exactly original data.
With an intention of making the compressed video data observable by all devices
across the globe, compression techniques are defined as IEEE standards by ISO/IEC
and ITU-T, for example, MPEG standards and H.264 standards. MPEG-2 standard
was defined in 1993 by ISO. The encoder and decoder architecture requirements are
defined in the form of syntax and semantic rules. Codecs contain lossy and lossless
compression tools. Variable-length coding is a lossless compression tool.
The appropriate use of source residual redundancy is the key point of decoding
technique. VLC decoder is the first tool of MPEG-2 decoder. The VLC generates
code word for each symbol with different length. Frequently occurring symbols are
represented using less number of bits, whereas less common symbols are represented
using long length code word. During the design of MPEG-2 decoder, various opti-
mized algorithms are developed to implement VLC decoder. VLC decoder archi-
tecture is a bitwise operator. Hence, VLC decoder contributes major computation
time for MPEG-2 decoder. VLC produces code word for each symbol. Each input
symbol is mapped to a code word, and code words may have varying length, but
contain integral number of bits. Frequently occurring symbols are represented using
less number of bits, whereas less common symbols are represented using long VLCs.
Over a sufficiently large number of encoded symbols, data is compressed.
Following are the main steps of VLC decoder:
• Identify the code word length and code word: Code words are identified from the
series of encoded input bit stream. Using code book in the form of lookup table,
the (run, level) values are identified.
• Shift the input bit stream by the length of the previous decoded symbol: After
identifying the length of the previous decoded symbol, the bits of input sequence
are discarded from input bit stream.
• Arrange the decoded symbol: The decoded symbol from code book is the output
of the decoder.

1.1 VLC Code book

The VLC decoder requires knowledge of source symbols, code word, and corre-
sponding code length. These details are stored in the form of lookup table called
code book. The size of code book is decided by the number of symbols and the prob-
ability of symbols. ISO 13818-2 defines 15 VLC tables for the video data, motion
vectors, and syntax elements necessary for decoder. The statistical analysis of 15
tables shows that Tables B-14 and B-15 defined as “VLC for DCT coefficients Table
zero” and “VLC for DCT coefficients Table one,” respectively, are larger in size. So
VLC decoder design is illustrated by designing a code book for these two tables,
which are available in ISO 13818-2 document.
High-Speed Variable-Length Decoder Using Segmented Code book 1153

The further arrangement of paper is as follows. Various algorithms of VLD are

designed to reduce the complexity, minimize power, and increase the speed and are
listed in Sect. 2. Section 3 explains the parallel decoder architecture using table
lookup method and partitioning of lookup table to improve index speed. The simu-
lation results are given in Sect. 4. The proposed work focuses on the simulation of
VLC decoder design referring to lookup table using the system-level simulation and
hardware implementation. Finally, the paper ends with the conclusion.

2 Observations

The VLC encoder and decoders have different implementation requirements. The
encoder finds the VLC value from the standard tables and generates bit stream. The
bit stream is the sequence of binary data without any field identification of code
“begin” or “end.” The symbols are encoded using table lookup defined as per the
IEEE standard specifications. The run, length combinations, and respective code
words are defined in the tables. The run and length combinations are generated
during encoding. In encoding process, after DCT and quantization, zigzag scanning
generates the array of values. This array is represented in the form of run-length
combination. Each of these run-length values is assigned with code word. As each
symbol is encoded independently, parallelism technique can be used to speed up the
coding process.
The input to decoding algorithm is the bit stream, and decoding procedure uses
standard tables. Since bit stream does not contain any information of start or end of
the code, the decoder architecture is complex. Basically, VLD is implemented using
binary tree approach or parallel decoding approach. The tree-based approach uses
the concept of Huffman tree. The direct implementation of VLC decoder is bit-by-bit
sequential decoding. Thus, VLD algorithm implementation has a tradeoff between
decoding speed and size of memory. For example, a simple and efficient represen-
tation of VLCs for a frame is using binary tree. An associated decoding algorithm
starts at the root node, extracts a bit from the input, and follows the corresponding
branch to the next node. The process repeats until a leaf is reached, which in turn
indicates the symbol. The complexity of this algorithm depends on the depth of the
tree. Thus, the main constraint of the approach is it decodes one bit at a time.
In [1] the use of concurrent method to improve the throughput is explained. The
algorithm is based on pipelined tree-based architecture. The fully pipelined design is
implemented using concurrent decoding of independent bit stream. Due to pipelin-
ing method, high-level optimization is achieved in terms of speed and bit length.
Since algorithm is based on VLSI architecture, the main limitation is reprogram-
ming for change in source statistics, hence less flexible. The PLA-based algorithm is
improved by using plane separation. The performance of this architecture is improved
by parallel processing in feedback path.
The soft input and soft output VLC decoder algorithm [2] uses the source redun-
dancy appropriately at decoder and improves the error correction capability. This
1154 S. Bhavikatti and R. M. Banakar

gives better results for concatenated schemes with VLC codes. The algorithm is less
complex. There is small loss due to insufficient quality of the soft outputs generated.
Since the delay is inserted during code book indexing during decoding process, it
counts for overall delay of VLC decoder.
The author in [3] proposes an algorithm to reduce the complexity of VLC decod-
ing by using SOSA algorithm. The reliable information of bits is collected from VLC
encoder. This information is used during decoding to minimize the errors. The reliable
information is about tree structure, probability of symbols. The iterative decoding
algorithm is less complex and also more robust to errors. The advancement in tech-
nology introduced many international video codec standards, which contain VLC
decoder as one of the tools. Lamy and Perros-Meilhac [4] propose multi-standard
VLC decoder. The hardware complexity is reduced by efficient representation of code
word table. The reduced hardware complexity gives better results for MPEG-2 and
H.264 video decoders. In [5], a high-speed Huffman decoder algorithm is proposed.
By using typical nature of Huffman coding which is based on single-side growing,
the search time is reduced. The memory efficiency is improved. Simulation results
are tabulated for MP3 type video.
In [6], a method to reduce the complexity of VLC decoder is explained. The
method is commonly defined for both MPEG-2 standard and H.264 standard. VLD
is implemented using sub-tree (ST) methods. Sub-tree refers to the prefix part and
tree part of VLCs. Since prefix part is redundant, using ST type of classification, the
redundancy appearing between different VLC trees is removed. This minimizes the
storage requirements.
In [7], VLCs are compared to fixed-length codes (FLCs). Since number of bits
used by VLC is less compared to FLC, VLCs are used in IEEE standards. It is
observed that VLCs are more prone to errors. By using trellis-based representation
for VLCs and BCJR algorithms, the error effect in VLCs is minimized. The author
Jae proposes a plane separation method to improve the speed of VLC decoder. The
delay involved in the feedback path of data flow is minimized. Shifting operations and
decision process of correct code word identification are carried out simultaneously.
Due to parallelism in process, the delay is reduced. Tsai and Chen in [8] propose
architecture based on parallel VLD. The architecture consists of code detector and
lookup table. The arrangement of code book is done such that VLC itself acts as
address. The size of lookup table and code detector implementation is considered
to design low-power algorithm. Subtable concept is used in code book arrangement.
The code search process is executed sequentially to find match in each subtable. The
table partitioning is done to minimize power consumption. The amount of power
consumed in code detection is explained with examples.

3 Parallel Decoding Approach

The parallel decoding technique of VLD is a search algorithm for finding match for
multiple bits from input bit stream to the code book. The algorithm implementation
includes the code book arrangement and comparing unit as the main parts. Code
High-Speed Variable-Length Decoder Using Segmented Code book 1155

book arrangement is considered to speed up the decoding process. The code book is
arranged in an increasing order of its code length. The lowest length code is the most
frequently occurring code. During search time, the number of memory references
can be minimized by arranging the code book in the increasing order of code length.
The code book reference and comparison are sequential processes performed till the
bits of input sequence are defined with symbol. Thus, decoding procedure is defined
using sequential finite state machine modeling as shown in Fig. 1. The FSM input
is VLC-coded bit stream, and outputs are decoded source symbols. The main hurdle
for designing of high-speed decoder is recursive dependency.
The decoder is loaded with bit stream indicated by flag “load.” Then comparison
is done for search with code book contents. During search, load = “0.” Search is
an iterative process done by comparing input bit stream with every code in code
book. Once the match is found, the load is reset. The input bit streams are shifted by
length of code word identified. The search process continues till all bits of a block
are decoded.
Figure 2 is the architecture of variable-length decoder. Input MPEG-2 bit stream
is serial input data to VL decoder. The variable length of the code words makes
the decoding to a sequential process. In order to decode a code word, the previous
code words must be decoded in sequence in order to determine where in the input
buffer the code word starts. This data dependency is limiting the throughput of the
VLC decoders. The variable-length decoder consists of a barrel shifter, comparator,
and code book. The comparator is designed to match the input bit stream with the
code word of code book. The code book is the main functional block, in the form
of a table. It contains information about number of code words, code length, and
code word. The major drawback here is the relatively large critical timing path in the
feedback loop that includes barrel shifter, code word table (ROM), and accumulator.
This feedback path cannot be pipelined and will therefore limit the performance of
this architecture. During decoding, the input bits are serially read. The length of code
word is unknown. It requires serial search in the code book. In order to speed up the
decoding process, search algorithm [9] is to be framed. By changing the arrangement
of code book [6], the search time can be minimized.

Fig. 1 Conventional FSM of load=1

VLC decoder

S0

Read bits
load=1

S2 S1
Shift the
bit stream by Read bits
codelen
1156 S. Bhavikatti and R. M. Banakar

Fig. 2 Parallel VLC MPEG-2

decoder architecture bit stream

Reg2 Reg1 Load

Reg3 Barrel shiŌer

Reg4 Comparator

Code Code Symbol

Symbol
length Word

The decoding logic is based on the code book access which in turn depends on
code book arrangement and overhead involved in code book access. The partitioning
of VLC tables into two parts is considered to minimize the power. This method
is considered to partition the code book into segments. These segments are defined
based on the code length. The address length required to refer to code book location is
reduced by two bits. ISO/IEC defines the total of 15 tables for variable-length code
assigned as per the MPEG-2 standard. The tables include the VLC of run-length
pair, motion vectors, and DCT coefficients. Among these tables, T14 and T15 are the
maximum length tables. The VLC decoder implementation is illustrated with respect
to Table T14. The length of Table T14 is 114. Since the arrangement of 114 VLCs
into one code book requires 7 bits address length. The address length is reduced due
to dividing the code book into segments.
Figure 2 shows the parallel decoding architecture of VLC decoder. The input is
MPEG-2 bistream. VLD architecture is composed of code book and decoding logic.
Further decoding logic consists of reading the code book, comparing logic to identify
the code match and find the symbol. The low-power design approach considers the
code book arrangement and decoding logic. Tsai and Chen [8] say that the VLC
decoding logic and size of VLC table decide the power and throughput. Thus to
minimize the power, the code book arrangement is considered. With the normal code
book arrangement, the complete code book is selected for the decoding operation.
During decoding, the data of selected row is transferred to register. The code book
remains in active state until the code is decoded and symbol is identified. The amount
of power consumed to keep the code book as a whole is more compared to memory
segment form. Using memory segmentation approach, the memory segment which
is referred for code access is activated keeping other memory segments in idle state.
Thus, the power consumption is minimized due to activating the part of code book.
High-Speed Variable-Length Decoder Using Segmented Code book 1157

Bit stream contains the frame data coded, as per MPEG-2 standard and syntactic
and semantic rules, necessary information to decode the bit stream. “maxlen” rep-
resents the highest length of code word in code book. In synchronous with clock,
number of bits = “maxlen” is read from MPEG-2 bit stream into reg1 and buffered
in reg2. The code word from code book and corresponding code length are read into
reg3 and reg4. The reg1 content is shifted by the number of bit positions equal to
the length of code word read from the code book. The inputs to the comparators are
shifted bits of reg1 and code word from code book. If comparator output is one, the
corresponding symbol is output. If match is not found, then comparator gives output
as zero. The comparator iteration continues for next code word from code book.
After finding the symbol, the contents of reg2 are shifted by that code word length
and rest of bits in reg2 are concatenated with input MPEG-2 bit stream. The process
repeats till all bits of block are decoded.

4 Simulation Results

In this paper, VLC decoder algorithm is verified using FPGA hardware implementa-
tion. The algorithm is tested using system-level simulation. For testing of algorithm,
bit stream is generated using system-level simulation and text file “[Link]” is gen-
erated. Using [Link] bit stream is generated. The serial input to VLC decoder is
given using buffer. The lookup tables or code book is in the form of ROM memory.
Discrete ROM memory separately for code of particular length eliminates the need
of length storage in code book. Since 17 is maximum length to read maximum of 17
values, 5 bits address length is assigned. Thus, the total memory length required is
5 * 114 = 570. Total memory length = 570 + 1921 = 2491. Thus, it is concluded
that memory segmentation minimizes the memory size. For memory segmentation
method, the number of bits is identified by referring to bits values which are referred.
It counts to 1879. The amount of memory saving is (1879/2491) * 100 = 75%. The
address length is reduced by 2 bits.

Table 1 Search time of

Variable length Single code book (ns) Segmented code
single code book and
book (ns)
segmented code book
4 8.3 8.2
6 10 8.3
10 11.8 8.1
14 15.9 8.1
17 18 8.2
1158 S. Bhavikatti and R. M. Banakar

The bit stream of a block contains VLC of differential DC and run-length pair.
The decoder module is simulated using Spartan-6, XC6SLX45T device. The simu-
lation results are tabulated comparing the search time of VLC for different lengths
considering a single code book and segmented code book in Table 1. It is observed
that the search time remains almost the same for segmented code book arrangement
as compared to the single code book arrangements. The search time increases with
increase in length for single code book arrangement, whereas the search time remains
almost the same for segmented code book arrangement. The average clock rate is
120 MHz in this design.

5 Conclusion

A new hardware-based VLC decoding algorithm is implemented using code book.

VLC decoder is the first block in MPEG video decoders. The modified VLC decoder
algorithm is simulated using system-level simulation, and power minimization is
shown with hardware implementation. The algorithm uses sequential decoding of
symbols. Since we assume that the number of symbols and code length is known, the
exact number of symbols can be decoded. The code book arrangement according to
the code length improves the memory indexing and reduces the memory access time.
The functioning of algorithm is tested on “[Link]” file. ISO/IEC 13818 data
is used to design the VLC decoder.

References

1. ISO/IEC JTC1/SC29 (1994) Generic coding of moving pictures and associated audio, ISO/IEC
13818-2. Draft International Standard
2. Jeon JH, Park YS, Park HW (2000) A fast variable length decoder using plane separation. IEEE
Trans Circuits Syst Video Technol 10(5):806–811
3. Aspar Z, Yusof ZM, Suleiman I (2000) Parallel Huffman decoder with an optimized look up
table option on FPGA. In: TENCON, Proceedings, vol 1, pp 73–76
4. Lamy C, Perros-Meilhac L (2003) Low complexity iterative decoding of variable length codes.
In:THALES communications, pp 1–6
5. Fryza T, Prokopec J (2009) Experimental application of channel coding for wireless transmission
of compressed video data. In: Ultra modern telecommunications & workshops, 2009. ICUMT
‘09. International conference, pp 1–4
6. Venugopal N, Ramachandran S (2009) Design and FPGA implementation of fast variable length
coder for a video encoder. IJCSNS Int J Comput Sci Netw Secur 9(7):178–184
7. Lo C-C, Hsu CW, Shieh M-D (2013) Low complexity multistandard variable length coding
decoder using tree based partition and classification. IET Image Process 7:185–190
High-Speed Variable-Length Decoder Using Segmented Code book 1159

8. Tsai T-H, Chen W-C (2003) A low power VLSI implementation for variable length decoder in
MPEG-1 layer 3. IEEE Tran
9. Zhu K, Liu WD, Du J (2012) Hardware JPEG decoder and efficient post-processing functions
for embedded application. In: Computer and information technology (CIT), 2012 IEEE 12th
international conference, pp 814–817
Efficient Cruise Control of Electric
Motorcycle Using PID

Rohan Sawant and Andrew Kidd

Abstract The automotive industry is progressing expeditiously. The popularity of

electric vehicles (EV) has shot up over the last few years. As a matter of fact, electric
drive systems such as DC motor drives, one of the electrical drives, are rapidly gaining
prominence because of their high efficiency, better dynamic response and low upkeep.
The following work deals with the drivetrain of the pre-existing electric motorcycle
comprised of a brushed DC motor. In this work, a system is developed to maintain the
speed of the electric motorcycle at a set speed, under various disturbances using PI
control. A detailed mathematical model, transfer function and simulation of the same
are obtained using the software package MATLAB, SIMULINK. The controlled DC
motor is made to track a variable speed set point with zero steady-state error and
desirable disturbance rejection capabilities.

Keywords DC motor · PID · Matlab

1 Introduction

DC motor drives play a significant role in mechanical and other applications such
as steel moving plants, electric trains and robotics. For the most part, an elite engine
drive framework must have great powerful speed order and load controlling reaction
to perform undertaking. DC drives, due to their straightforwardness, simplicity of
utilization, dependability and favourable cost have for quite some time been a spine
of mechanical and robotic applications where speed and position control of motor
are required. DC drives are less complex, as when it comes to motorcycles, the power
conversion factor from DC to AC is eliminated. Also, the speed-torque attributes of
a DC motor are robust [1].

R. Sawant (B)
Vidyalankar Institute of Technology, Mumbai, India
e-mail: rohansawant96@[Link]
A. Kidd
Teesside University, Middlesbrough, UK
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1161

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1162 R. Sawant and A. Kidd

Although three-phase induction motors are the most widely recognized sort of
motors utilized in industrial applications, DC motors are expanding in prevalence
because of their execution points of interest over AC motors for applications ranging
from high-speed automation to electric motorbikes. Applications where speed should
be varied, or torque should be controlled with high exactness, brushed DC motors are
used. Today, locomotives are being offered with powered gadgets, in addition design-
ers are seeking to improve the performance of the equipment they design, which often
entails upgrading from AC motor to a DC motor, specifically, places where increasing
accelerations are required. DC drives are additionally being fabricated in regularly
expanding volumes for much Bijou applications, such as solar-based gadgets, toys
and cell phone handsets [2].
In this work, a corresponding DC motor similar to the one used in an electric
motorcycle is considered. A transfer function model of the motor is obtained using
the reaction curve method and analysed in MATLAB, compared to physical data
from the test motor to check for goodness-of-fit. Further using this model and the
direct design synthesis (DDS) control design approach, the speed of the motor will
be made to track a variable set point and then be maintained under the influence of
disturbances, such as a change in the inclination of the surface the bike is travelling
on (i.e. up/down a hill) (Fig. 1).

2 Mathematical Modelling

2.1 Transfer Function Derivation of a DC Motor (Bottom-Up

Approach)

A separately excited DC motor system is considered. To obtain the mathematical

model of the same, consider, supply voltage ‘v’ applied to the motor circuit (refer
Fig. 2), and the motor generates torque (T ) that is proportional to the product of the

Fig. 1 DC motor equivalent

circuit
Efficient Cruise Control of Electric Motorcycle Using PID 1163

magnetic field flux ‘φ’ and the rotor armature current ‘i’,

T = KT i φ

The Back EMF is a counter voltage that is proportional to the product of magnetic
field flux (φ) and the rotor angular velocity (ω),

e = KEφ ω

Making use of KVL voltage law in the above motor circuit,

di
v = Ri + L +e
dt
Considering the above fundamental equations into Laplace transform, respec-
tively, and further simplifying it [3, 4], we get the first-order transfer function model
of the DC motor as,
1
(s) KE
= (1)
v(s) 1 + τm s

where τ m = Electromechanical time constant.

2.2 System Identification—Reaction Curve Method

The first-order plus time-delay (FOPTD) model has been widely used to design and
implement the process controllers of the system. The terms related to the low-order
plus time-delay processes are very useful for describing the dynamic characteristics
of given process. The FOPTD process derived for DC motor system is given in
Eq. (1).
From Eq. (1), it can be realized that K and τ are called the static gain and the
time constant, respectively. The step input applied to the motor is u, and the maxi-
mum rated voltage under which the motor operates is represented by y. Using the
formulae, the terms K, τ and d are computed.

y
K = (System Gain)
u

τ = 1.5(t2 − t1 ) (Time Constant)

 
t2
d = 1.5 t1 − (Time delay)
3
1164 R. Sawant and A. Kidd

Fig. 2 Sample DC motor behaviour plot

The computed values are fed in the FOPTD model of the motor and from the
schematic obtained in the SIMULINK (refer Fig. 3), and a transfer function graph
of the DC motor is acquired [5].
In the proposed system, from Eq. (1), the term d, i.e. time delay is eliminated since
the value derived from the transfer function graph is insignificant (refer Fig. 4). Also,
the transfer function acquired from the reaction curve method is in time domain, i.e.
continuous transfer function. It is further converted to discrete form so as to get the
same sampling time, and the discrete and continuous elements do not mix within the
SIMULINK model to be used in the digital circuit. The sampling frequency used is
kept same as that in the input.
A comparison between the actual DC motor response and the transfer function
model developed using the above-derived terms is plotted. It can be observed that
the transfer function model developed is a decent fit.

Fig. 3 SIMULINK schematic to acquire TF

Efficient Cruise Control of Electric Motorcycle Using PID 1165

Fig. 4 Transfer function response

3 Control System Design

A proportional controller makes the control signal directly proportional to the mag-
nitude of the error signal. One of the main drawbacks of proportional—only control
is its inability to leave zero steady-state error. Increasing the gain reduces the off-
set but can lead to noise problems and instability. Alternatively, by introducing the
Integral action in the controller, which perhaps act as an automatic bias adjuster can
eliminate the offset without introducing much of instability. But, the downside is that
the integral action can actually introduce oscillatory behaviour, and this is eliminated
only if the parameter is set correctly. The introduction of an integrator now exhibits
dynamic behaviour of its own. Now, the system will have two adjustable parameters
K P and K I , the proportional and the integral gain.
Using Pade approximation [6, 7], slight variations in the approach of direct design
synthesis (DDS), which has achieved widespread industrial acceptance, are used to
compute the proportional and integral parameters of the system.
For a FOPTD model with a PI controller, the parameters are derived as follows,

2τ + d
KP =
2K λ
2
KI =
2τ + d
1166 R. Sawant and A. Kidd

Derivative control is highly susceptible to noise. The noise sensitivity is due

to high rate of change within the signal when there are noise spikes. Also, in the
proposed system, the output of the DC motor consists of fair share of noise. Hence,
by incorporating derivative control in the system, entire system can turn unstable. As
a matter of fact, the given system attains stability by incorporating PI control only.

4 Implementation

4.1 PI Control Observed Under Disturbances

The output (speed) of the DC motor is measured and fed back to the PI controller
via a closed-loop system. The feedback signal is first compared with the reference
signal. An error signal is generated at the summing point by calculating the difference
between the actual output and the reference signal by simple subtraction. This error
signal is passed through to the PI controller for the required modification set by
specification of controller parameters. The control outputs an appropriate control
signal such that the error is then eliminated (Fig. 5).
A mechanical disturbance in the form of brake can be applied to the motor. This
is analogous/equivalent to an electric bike travelling on a now steeper gradient (i.e.
uphill). Due to the incorporation of PI control, it now maintains the speed of the
motor at the certain set speed.

4.2 PI Control Observed Under Varying Pre-defined Speeds

PI control is also observed whilst setting different set point speeds. The proposed
system consists of manual/auto switch. If the switch is on manual mode, the driver
will have to control the speed of the motorcycle manually using the throttle. If the
switch is on auto mode, a set point speed is decided by the driver, and the motorcycle

Fig. 5 PI control under disturbances, schematic

Efficient Cruise Control of Electric Motorcycle Using PID 1167

Fig. 6 PI control under varying pre-defined speeds, schematic

will drive at the set speed even under most disturbances. For the purpose of simulation
only, the varying set point speed block can be represented by signal builder in the
SIMULINK. The varying signals are the step signals of different magnitudes. The
PI controller will try to eliminate the error signal as mentioned and match with the
new set speeds at earliest (Fig. 6).

5 Results and Discussions

5.1 Disturbance Rejection Capability

Mechanical disturbance is added to the system in the form of braking. It can be seen
that the disturbance is added in the system at t = 5.5 s, and it takes around 4 s to settle
at pre-defined value since λ = 4. The PI controller recovers from this disturbance
and adjusts itself to the pre-defined speed. This is the disturbance rejection capability
of the PI controller in the system.

5.2 Set Point Tracking Capability

Set point tracking capability allows the user to set the set point speeds of the motor-
cycle. In the proposed system, signal builder allows to set varying amplitude for the
step input for the motor (Figs. 7 and 8).
This varying step input does act like different set point speeds for the motorcycle,
and PI controller tries to adapt to these speeds.
1168 R. Sawant and A. Kidd

Fig. 7 Disturbance rejection behaviour using PI

Fig. 8 Set point tracking capability using PI

6 Conclusions and Future Enhancements

The transfer function model derived from the reaction curve method finds a decent
fit with the motor output. This transfer function can be used as a basis for tuning. PI
control approach is incorporated to trace this transfer function model. The proposed
Efficient Cruise Control of Electric Motorcycle Using PID 1169

system can efficiently work in maintaining the speed of the motor and perform the
function of cruise control in electric motorcycles.
Induction motors are now used widespread. One such notable example would be
four-wheel based electric vehicles. An equivalent mathematical model of an induction
motor can be computed using the process defined in the proposed system. Further
analysis can be resolved, and a similar cruise control system for four-wheeled system
based on induction motors can be devised.

References

1. Wang JB (2001) Control of electric machinery. Gau Lih Book co., Ltd, Taipei, Taiwan
2. DC motor drives. [Link]
3. Huang G, Lee S (2008) PC-based PID speed control in DC motor. In: International conference
on audio, language and image processing, 2008. ICALIP 2008, pp 400–407. IEEE
4. Hanselman DC (2003) Brushless permanent magnet motor design, The Writers’ Collective
5. Sung SW, Lee J, Lee IB (2009). Process identification and PID control. Wiley
6. Rivera DE, Morari M, Skogestad S (1986) Internal model control: PID controller design. Ind
Eng Chem Process Des Dev 25(1):252–265
7. Al-Mashakbeh ASO (2009) Proportional integral and derivative control of brushless dc motor.
Eur J Sci Res 35(2):198–203
Transmission of Watermarked Image
in WSN Using ELSM Algorithm

A. Umapriya and P. Nagarajan

Abstract To improve the copyright protection of the digital image, the watermarking
technique is used. By changing the entropy, identify the location where the secret data
gets embedded into the original image. It provides less distortion and high robustness.
The watermarked image is send to the receiver through the wireless sensor network.
In WSN, clustering is the best technique to save the energy. In cluster-based WSN,
cluster head requires more energy to receive the data from the sensor nodes and trans-
mitting it to the base station. To maintain the lifetime of WSN, the proper selection
of CH is essential. In this paper, we propose the energy-based least squares multiple
algorithms. Simulation results show that the ELSM algorithm is more efficient com-
pared to the WLSM algorithm to improve the range of bandwidth, cluster overhead,
skew rate, offset rate, carrier signal, reference signal and control output signal.

Keywords Wireless sensor network · Cluster head · Energy-based least squares

multiple · Weight-based least squares multiple

1 Introduction

Data security is necessary for every field to transfer the secret data. For this purpose,
we use the digital watermarking technique. It provides high robustness to the host
images. To improve the quality of image and strength and to remove the noises, many
processes are available.
[1] The host image provides self-recovery, tampered localization and ownership
verification. Greyscale watermark is proposed to the watermark insertion. The last
two LSBs get changed after the insertion of watermarking. While using LSB, it is
easy to identify the host image.

A. Umapriya (B) · P. Nagarajan

Department of ECE, Vivekanandha College of Engineering for Women, Namakkal, India
e-mail: umapriyakaviarun@[Link]
P. Nagarajan
e-mail: greennagarajan@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1171

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1172 A. Umapriya and P. Nagarajan

[2] The survey of digital watermarking techniques provides discrete cosine trans-
form, discrete Fourier transform, discrete wavelet transform and also the performance
analysis metrics. All these techniques are compared.
[3] Clock synchronization using least common multiple algorithm is used to avoid
the clock skew and clock offset. Two levels of synchronizing are used. One is nodes
within the cluster synchronize and another level is the cluster head synchronize to
remove the cluster overhead.
[4] Using particle swarm optimization algorithm in wireless sensor network, the
energy-efficient head cluster selection is proposed.
[5] Using this algorithm, it provides a better performance of the total energy
consumption, lifespan of the network and also the base station that received many
packets.
[6] For the security purpose, we use the watermarking technique to embed the
data in the image [7]. The image gets watermarked. In wireless sensor network [8], it
contains many sensor nodes that are available to transfer the data from the sender to
the receiver through the base station [9]. To save the energy, network gets divided into
groups called as cluster using the energy-based least squares multiple algorithms.
Each cluster has its own cluster head used to find the path among the sensor nodes
[10].
Cluster heads are used to find the shortest path to the bath station. The base
station communicates with the public networks [11]. Then the watermarked image
received at the receiver end. The secret data can extract from the host image [12].
The ELSM algorithm is used to improve various parameters such as the range of
bandwidth, cluster overhead, skew rate, offset rate, carrier signal, reference signal
and control output signal [13].

2 Methods and Materials

2.1 Watermark Insertion

In any field, we hide some information for security purpose. When we transfer the
secret data, it gets stolen by someone. To avoid this problem, we use the watermarking
technique. The watermark insertion is the process of embedding the secret data in
the original image. After the insertion of secret data in the image, the secret key will
generate. The image gets watermarked. Figure 1 shows the watermark insertion.

2.2 Clock Synchronization in WSN

Wireless sensor network contains more number of sensor nodes. When we transmit
the image from sender to receiver, it consumes huge power because of the presence
Transmission of Watermarked Image in WSN Using ELSM Algorithm 1173

key

Watermark Watermarked
Image insertion image

Watermark

Fig. 1 Watermark insertion

of more sensor nodes. For saving the energy in wireless network, we propose the
energy-based least squares multiple algorithms. In this technique, divide the sensor
nodes into groups. Each group contains the superior called as cluster head which is
used to select the shortest path of the transmitting image.
Figure 2 shows the clock synchronization in WSN using ELSM algorithm. The
cluster head is automatically selected in the network using ELSM algorithm. Two
levels of synchronizing are used. One is nodes within the cluster synchronize and
another level is the cluster head synchronize to remove the cluster overhead. Cluster
head transmits the image to the base station. And also the ELSM algorithm is used to
improve the range of the bandwidth, cluster overhead, skew rate, offset rate, carrier

Fig. 2 Clock synchronization using ELSM algorithm

1174 A. Umapriya and P. Nagarajan

signal, reference signal and control output signal. Then the base station sends the
image to the public networks. Finally, the receiver receives the image by giving the
correct password.
By saving the energy consumption, the bandwidth gets improved and reducing
the overhead. And also provides the long distance to transmit the image.

2.3 Transmitting and Receiving Process of WSN

The transmitting and receiving processes of the wireless sensor network using ELSM
algorithm are given below. We need to set the password. If we give the password
correct, the secret data can be embedded in the original image by using watermark
technique. Then the secret key will generate automatically.
The host image is sent from the source to the destination through the wireless
sensor network. In this network, there are more number of sensor nodes which are
available. When we transmit the image through this network, the energy loss will
occur for more sensor nodes. The sensor nodes are grouped by the cluster. Each
cluster has its own cluster head. For choosing the proper cluster head, we use the
ELSM algorithm. The cluster head is used to select the shortest path via sensor nodes.
The cluster head selects the sensor nodes which have the highest residual energy.
When we select the proper cluster head, we can save the energy. By saving the energy,
we can transmit the image for long distance. Then the transmitted image reaches the
base station (Fig. 3).
The receiving section using ELSM represents Fig. 4. We need to give the password.
If the receiver gives the wrong password, the secret data does not generate instead they
receive the ordinary image. The same process will continue for selecting the cluster
head in transmitter side. Finally, the data can be retrieved by using the watermark
extraction. By using this technique, we can save the energy of all sensor nodes. Each
iteration will continue by using this algorithm.

2.4 Watermark Extraction

The watermark extraction is used to extract the secret data from the watermarked
image. Figure 5 represents the watermark extraction process. When we provide the
correct password, we can receive the secret data. Otherwise, we can receive the
ordinary image.
Transmission of Watermarked Image in WSN Using ELSM Algorithm 1175

Fig. 3 Transmitting section

using ELSM START

CHECK NO
PASSWORD

YES

SELECT IMAGE

ENTER SECRET DATA

WATERMARKING

SECRET KEY GENERATION

SELECT HEAD CLUSTER

ELSM

SEND TO THE BASE STATION

3 Result and Discussion

The simulation result is taken by MATLAB using C program. For the security pur-
pose, we use the watermarked image. Simulation results show that the ELSM algo-
rithm is more efficient compared to the WLSM algorithm to improve the range of
bandwidth, cluster overhead, skew rate, offset rate, carrier signal, reference signal
and control output signal.
1176 A. Umapriya and P. Nagarajan

Fig. 4 Receiving section

using ELSM

Secret keySecret

Watermark Extracted image

Watermarked
Image extraction

Fig. 5 Watermark extraction

Figure 6 represents the graphical representation of error rate, carrier signal and
control signal using ELSM algorithm. Figure 7 represents the graphical representa-
tion of skew rate, offset rate and cluster overhead. By saving the energy, we improve
the lifespan of the network and also provide the long transmission distance. Finally,
we extract the secret data using the watermark extraction.
Transmission of Watermarked Image in WSN Using ELSM Algorithm 1177

Fig. 6 Error rate, carrier signal, control signal

Fig. 7 Skew rate, offset rate, cluster overhead

4 Conclusion

In this research, we proposed the ELSM algorithm for the proper selection of the
cluster head to save the energy. This algorithm is to extend the lifespan of the network
by saving the energy among the nodes. Two levels of synchronizing are used. One is
nodes within the cluster synchronize and another level is the cluster head synchronize
to remove the cluster overhead. The watermarked image was received by receiver.
By saving the energy, the transmission range is for a large distance. The experimental
1178 A. Umapriya and P. Nagarajan

results of ELSM algorithm performs better than the existing algorithm in terms of
bandwidth, cluster overhead, skew rate, offset rate, carrier signal, reference signal
and control output signal.

References

1. Ansari IA, Pant M (2017) Multipurpose image watermarking in the domain of DWT based on
SVD and ABC. Pattern Recogn Lett
2. Hemani, Singh S (2017) A survey of digital watermarking techniques and performance
evaluation metrics. IJETT
3. Tabassum N, Geetha D (2017) Clock synchronization in wireless sensor networks using least
common multiple. AEU
4. Srinivasa Rao PC, Jana PK (2016) A particle swarm optimization based energy efficient cluster
head selection algorithm for wireless sensor networks, Springer Science Business Media
5. Ernawan F, Kabir MN (2018) A blind watermarking technique using redundant wavelet
transform for copyright protection. In: 14th IEEE colloquium on signal processing and its
applications
6. Panda J, Mouriya S, Dang R (2016) Analysis of robustness of image watermarking algorithm
using the dual tree complex wavelet transform and just noticeable difference. In: International
conference on signal processing and communication
7. Tyagi S, Singh HV (2016) Digital watermarking techniques for security applications. In:
International conference on emerging trends in electrical, electronics and sustainable energy
systems
8. Ernawan F (2016) Robust image watermarking based on psychovisual threshold. J ICT Res
Appl
9. Fazli S, Moeini M (2016) A robust image watermarking method based on DWT, DCT, and
SVD using a new technique for correction of main geometric attacks. Optik Int J Light Electron
Opt
10. Kakkirala KR, Chalamala SR (2014) DWT-SVD based blind audio watermarking scheme for
copyright protection. In: IEEE conference
11. Makbol NM, Khoo BE (2013) Robust blind image watermarking scheme based on redundant
discrete wavelet transform and singular value decomposition. Int J Electron Commun (AEÜ)
12. Ling H-C, Phan RC-W, Heng S-H (2013) Comment on robust blind image watermarking
scheme based on redundant discrete wavelet transform and singular value decomposition. Int
J Electron Commun (AEU)
13. Abu NA, Ernawan F, Suryana N, Sahib S (2013) Image watermarking using psychovisual
threshold over the edge. In: Information and communication technology, ICT-EurAsia
International Students’ Gender Impact
to Use Mobile Learning in Tertiary
Education

Sujit Kumar Basak, Marguerite Wotto and Paul Bélanger

Abstract This study aims to design a model on mobile learning used by international
students in tertiary education. It has used a survey questionnaire distributed through
the LimeSurvey. A total of 12 international students from a tertiary education (6
female students and 6 male students) participated in Canada. The results revealed
that male students got higher impact to the performance expectancy (PE), social
factors (SF), and facilitating condition (FC) but on the other hand, female students
got an impact on PE and FC. Furthermore, female students got a significant impact
on PE and SF on behavioral intention (BI). Gender is having an impact on EE
and FC. Finally, PE and SF have an impact on the BI to use mobile learning in
tertiary education. These findings will help practitioners, educators, policymakers to
implement mobile learning in tertiary education for international students in Canada
and abroad.

Keywords Mobile learning (ML) · International students · UTAUT · SEM

1 Introduction

In tertiary education, mobile learning (ML) has significantly increased and is playing
a crucial role for international students. Therefore, use of mobile learning in tertiary
education helps for international students to be connected anywhere and anyplace.
Berking et al. [1] stated that m-learning has a significant impact on the teaching
and learning environments. In this context, the tertiary education may encourage
international students to use m-learning for learning purposes. According to [2], a
mobile device is used by many users because it is affordable and easy to carry than

S. K. Basak (B) · M. Wotto · P. Bélanger

Université du Québec à Montréal (UQÀM), Montreal, Canada
e-mail: sujitbasakmca@[Link]
M. Wotto
e-mail: [Link]@[Link]
P. Bélanger
e-mail: [Link]@[Link]
© Springer Nature Singapore Pte Ltd. 2020 1179
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1180 S. K. Basak et al.

personal computers. The development of mobile and technology has opened a new
window to learn. Students believe that mobile learning has positive influence on the
learning environment, although students are well known with barriers [3]. Levy’s
study indicated that learners of mobile devices are significantly higher motivated
as compared to who did not use mobile device [2]. Using mobile devices, various
materials can be downloaded. However, most of the smart mobile devices can easily
play video features and also record where learners use their mobile devices to collect
scientific data and visual materials [4]. The implementation of mobile learning can
add a significant impact on the existing formal learning and assessment [5].
Gender issue in m-learning has been identified in the past decades on the com-
puter and education research [6]. Male or female does not use the technology in a
similar way, but as a result, some differences exist [7]. On the other hand, female
students got significantly higher attitudes as compared to their male counterparts
[8]. Moreover, a study by Muhanna and Abu-Al-Sha’r [9] indicated males got more
positive attitudes than females. Similarly, Kwon and Chidambaram [10] argued that
there is no significant impact on the gender attitude.

2 Problem Statement

Wang et al. [11] stated that although mobile technology use is increasing, using
ML still in the infancy stages. The limitations of ML are categorized into four
categories, namely physical attributes, network connection, physical environmen-
tal (using mobile devices outdoors), and limitations of the software design (difficult
to install and also built in function) [12]. In this context, implementing ML for stu-
dents could be a challenge because students are already used to what they are using
in the past. Several researchers indicated that gender differences exist among college
students or adult users about Internet use, attitudes toward the Internet, frequency
to use Internet, and to measure self-skills of the Internet [13, 14]. Using a small
keyboard and touch screen need more time to search for some information and to
read them [15]. Existing research shows that female got anxiety than males on the
computers and Internet use [8, 16]. Above, evidences show that the use of mobile
learning by international students is increasing day by day. “The comparative analy-
sis of the variables shows that social factors have a significantly higher impact on the
behavioral intention to use for the m-learning by the first-year international students
followed by performance expectancy, effort expectancy, and facilitating condition”
[17]. In addition, evidences also show that students (male and female) are meeting
more challenges to use mobile learning. This study is very important because it will
help to analyze the international students’ gender impact on the mobile learning very
broadly in Canada.
International Students’ Gender Impact to Use Mobile Learning … 1181

3 Literature Review

The existing study shows the impact of gender plays a crucial role in tertiary educa-
tion. A study with the undergraduate university learners in using a digital library [18]
indicated male students got a higher impact on the perceptions than female students.
Another study was conducted by Enoch and Soker [19] with students who use the
Web-based instruction method. The results found that male students and female stu-
dents both are continuously increasing the use of the Internet and the difference is still
significant. Recent development on the Internet technology, namely online “gender
difference, is believed to be less significant” [20]. A study on the differences of male
and female students indicated that male students have more positive impacts on a
computer use. In addition, male students use more frequently computer as compared
to female counterparts [21].
Similarly, Hilao and Wichadee [22] also conducted a study on 122 students (65
females and 57 males) on gender differences to use a mobile phone. Results revealed
there is no difference in the usage and attitudes of male and female students in
terms of language learning and performance. Most of the research findings show
that male students were well experienced and very positive toward m-learning as
compared to female students [23, 24]. In Taiwan, conducted a study by Ong and Lai
[25] on 156 employees where 67 of them female employees and 89 of them male
employees. These employees participated from six multinational companies. Results
revealed male’s usages decisions are having a significant impact by their perceptions
of the usefulness of mobile learning. Morahan-Martin [26] study also argued female
students used the Internet very less frequently, and they also spend less time than
their male counterparts.
A study was conducted by Li and Kirkup [27] with 465 students where 220
students were Chinese and 245 students were British. Their study indicated that there
is a difference with the gender in Chinese and British students. A study conducted by
Kay [28] with 659 secondary school students on gender differences, and the results
found that there is a positive attitude with the male students than female students.
Tsai and Tsai [29] “investigated and compared the gender gaps of college students’
Internet use and Internet attitudes between 1995 and 2002 with a conclusion that both
gaps were lessoning or disappearing altogether in 2002.” A study was conducted by
Wang et al. [11] in Taiwan and their respondents were 330. Results from [11] indicated
that gender has a positive influence on the social as well as on self-learning.

3.1 Theoretical Framework

This study will use the UTAUT model. The UTAUT model was proposed by
Venkatesh et al. [30], and it was developed by merging eight models of technol-
ogy acceptance to information technology or information system research study. In
1182 S. K. Basak et al.

this research, “performance expectancy” means the belief of use ML by the interna-
tional students; “effort expectancy” means international students are associated with
the ML; “social factors” means international student usages with the ML; “facilitat-
ing conditions” means that the technical infrastructure, as well as the organizational
infrastructure, are available to use ML; and finally, the “behavioral intention” means
performance of international students’ interaction.

4 Aim

This paper’s aim is to design a model for the international students’ gender impact
to use mobile learning in tertiary education.

4.1 Research Questions

(a) Analyze the first-year international students’ gender impact on “performance

expectancy” to the “behavioral intention” in tertiary education;
(b) Analyze the first-year international students’ gender impact on “effort expectan-
cy” to the “behavioral intention” in tertiary education;
(c) Analyze the first-year international students’ gender impact on “social factors”
to “behavioral intention” in tertiary education;
(d) Analyze the first-year international students’ gender impact on “facilitating
condition” to “behavioral intention” in tertiary education.

To design a model for the international students’ gender impact to use mobile learning
in tertiary education.

4.2 Research Hypotheses

The following research hypotheses were taken into consideration:

H0 1 Gender influence positively on “performance expectancy” to “behavioral
intention” m-learning by international students in tertiary education;
H0 2 Gender influence positively on “effort expectancy” to “behavioral intention”
m-learning by international students in tertiary education;
H0 3 Gender influence positively on “social factors” to “behavioral intention” m-
learning by international students in tertiary education;
H0 4 Gender influence positively on “facilitating condition” to “facilitating condi-
tion” m-learning by international students in tertiary education.
International Students’ Gender Impact to Use Mobile Learning … 1183

5 Methodology

This pilot study began by collecting data anonymously from first-year international
students, and these students participated voluntarily from a tertiary education in
Canada. The questionnaires of this study were distributed between November 27 and
December 23, 2017. International students (first year) participated anonymously from
various departments. Before distributing questionnaires, a permit was granted from
the Comité institutionnel d’éthique de la recherche avec des êtres humains (CIEREH)
of the institution where the research took place. In this study, 17 international students
participated, but 12 international students were considered for this study due to the
equal number of participants were considered (6 female international students and 6
male international students). All the data were captured on Microsoft Excel. Having
captured data on the Microsoft Excel, data were analyzed using the WarpPLS 5.0
version. Furthermore, a UTAUT model was used to validate the questionnaire and
these variables were “performance expectancy (PE)” (section A), “effort expectancy
(EE)” (section B), “social factors (SF)” (section C), and the “facilitating conditions
(FC)” (section D) that predict the “behavioral intention (BI)” (section E) [30].

6 Results

6.1 Cronbach’s Alpha Values

According to [31], if the value of α is 0.70 or even greater 0.70, then the alpha value
(α) is considered acceptable. On the other hand, Cronbach [31] indicated that when
variables are less than 10 items, in that case, the alpha value (α) may be 0.5 or less.
This study considered five variables and Table 1 shows that alpha (α) coefficients
are less than 0.5 only with the facilitating conditions, but with all other variables
Cronbach’s alpha coefficients are greater than 0.7.

Table 1 Cronbach’s alpha (α) values, composite reliability values, and total items used in each
variable
Variables Alpha (α) values for Cronbach Composite reliability Total item used
coefficients
PE 0.870 0.907 5
EE 0.884 0.916 5
SF 0.876 0.914 5
FC 0.498 0.666 5
BI 0.983 0.986 5
1184 S. K. Basak et al.

6.2 A Model for Female International Student’s to Use

M-Learning

Figure 1 shows that female students have a significant impact on the performance
expectance (PE) with a value of β = −0.42 (when P < 0.01) and facilitating condition
(FC) with a value of β = −0.54 (when P < 0.01) but female students do not have
significant impact on the effort expectancy with a value of β = −0.12 (when P =
0.13) and social factors with value of β = −0.14 (when P = 0.08). Figure 1 also shows
that female international students having impact on the performance expectancy and
social factors to behavioral intention to use the mobile learning in tertiary education
when R2 = 0.89. However, female students do not have much impact on the effort
expectancy and facilitating condition to behavioral intention using mobile learning
by international students in tertiary education.
Figure 2 shows that male students have a significant impact on the performance
expectance (PE) with a value of β = −0.51 when P < 0.01, social factors (SF) with
a value of β = −0.38 when P < 0.01 01, and finally, the facilitating condition (FC)
with a value of β = −0.27 when P < 0.01. Nevertheless, male student does not
influence significantly to effort expectancy (β = −0.15 when P = 0.07). Figure 2
also shows that male students got a significant impact on the PE, EE, to behavioral
intention to use mobile learning in tertiary education when R2 = 0.98. Furthermore,
male international students do not have much impact on the SF to BI the mobile
learning in tertiary institutions.
Having analyzed Fig. 1 as well as Fig. 2 show that male international students got a
significant higher impact on the PE, SF, and FC; but on the other hand, female student
having an impact on the PE and FC. Female international students got a significant

Fig. 1 Female usages of m-learning in tertiary education

International Students’ Gender Impact to Use Mobile Learning … 1185

Fig. 2 Male usages of m-learning in tertiary education

impact on the PE and SF to BI to use mobile learning in tertiary education. Finally,

PE, EE, and FC also significantly influence to BI to use mobile learning in tertiary
education by male students.
Figure 3 shows that gender has a significant impact on the effort expectance (PE)
with a value of β = −0.51 when P < 0.01, facilitating condition (FC) (when β =
−0.46 and P < 0.01). On the other hand, gender does not have a significant impact
on the performance expectancy and social factors. Finally, the PE (when β = 0.26
and P < 0.01) and social factors with a value of β = 0.74 when P < 0.01 got a

Fig. 3 A model of the gender (male and female) impact to use the m-learning in tertiary education
1186 S. K. Basak et al.

Table 2 Fit index, model,

Fit index Model Recommendation
and recommendation
APC 0.303 P = 0.002
ARS 0.294 P = 0.003
AARS 0.214 P = 0.012
AVIF 2.530 Acceptable if ≤5, ideally ≤3.3
AFVIF 6.121 Acceptable if ≤5, ideally ≤3.3
GoF 0.461 Small ≥0.1, medium, ≥0.25, large
≥0.36
SPR 0.875 Acceptable if ≥0.7, ideally = 1
RSCR 0.999 Acceptable if ≥0.9, ideally = 1
SSR 0.875 Acceptable if ≥0.7
NLBCDR 1.000 Acceptable if ≥0.7
APC Average path coefficients, GoF Tenenhaus GoF, ARS
Average R-squared, SPR Simpson’s paradox ratio, AARS Average
adjusted R-squared, RSCR R-squared contribution ratio, AVIF
Average block VIF, SSR Statistical suppression ratio, AFVIF
Average full collinearity VIF, NLBCDR Nonlinear bivariate
causality direction ratio

significantly positive impact to BI to use mobile learning by international students

in tertiary education.

6.3 Model Fit and Quality Indices

A model can be measured two ways are namely convergent and the discriminant
validity. In the case of convergent, it is analyzed with different categories. These
categories are the questions, the reliability, constructions of the reliability, and the
variance extracted with the help of constructs. On the other hand, discriminant validity
can be measured based on the correlation [32]. Table 2 shows how good the model
fit is.

6.4 Mean and Standard Deviation for PE, EE, SF, FC,
and BI

Table 3 shows that PE significantly influence on BI to use the mobile learning by

international students in tertiary education when the mean of PE is 2.50 and standard
deviation of PE is 1.00 for the x-axis. Similarly, in the case of y-axis, the mean of
BI is 2.17 and standard deviation of BI is 1.27. EE got a significant influence to BI
in order to use mobile learning by international students in tertiary education when
International Students’ Gender Impact to Use Mobile Learning … 1187

Table 3 Mean and standard

Variables Mean SD Axis
deviation of x-axis and y-axis
for PE, EE, SF, FC, and BI PE 2.50 1.00 X
BI 2.17 1.27 Y
EE 2.08 0.51 X
BI 2.17 1.27 Y
SF 2.42 1.08 X
BI 2.17 1.27 Y
FC 2.42 1.16 X
BI 2.17 1.27 Y
SD Standard deviation, X x-axis, Y y-axis

the mean of EE is 2.08 and the standard deviation is 0.51 for the x-axis. Similarly, in
the case of y-axis, the mean of BI is 2.17 and the standard deviation is 1.27. Social
factors (SF) also influence BI to use the mobile learning by international students in
tertiary education when the mean of SF is 2.42 and the standard deviation is 1.08 for
the x-axis. Similarly, in the case of y-axis, the mean of BI is 2.17 and the standard
deviation is 1.27. FC also influence to BI in order to use the mobile learning by
international students in tertiary education when the mean of FC is 2.42 and the
standard deviation is 1.16 for the x-axis. Similarly, in the case of the y-axis, the mean
of BI is 2.17 and the standard deviation is 1.27.

6.5 Correlation

Table 4 indicates p-values correlations when PE is 1.000, EE is 1.000, SF is 1.000,

FC is 1.000, BI is 1.000, and finally GN (gender) is 1.000 for the use of mobile
learning in tertiary education.

Table 4 Values of correlations for PE, EE, SF, FC, BI, GN

PE EE SF FC BI GN
PE 1.000 0.271 0.005 0.455 <0.001 0.457
EE 0.271 1.000 0.467 0.445 0.369 0.060
SF 0.005 0.467 1.000 0.507 <0.001 0.887
FC 0.455 0.445 0.507 1.000 0.948 0.135
BI <0.001 0.369 <0.001 0.948 1.000 0.358
GN 0.457 0.060 0.887 0.135 0.358 1.000
1188 S. K. Basak et al.

7 Discussion and Conclusion

The m-learning aspects are very important in tertiary education in Canada because
it is in the early stage. Twelve participants participated in this pilot study, and all
the participants were from first-year international students. A questionnaire was
distributed through the line survey among first-year international students. This study
has used the UTAUT model, and all the variables in the questionnaire were validated
by the UTAUT model. Firstly, data were captured in the Microsoft Excel. Secondly,
having captured the data, it was analyzed using the WarpPLS 5.0 version. Three
structural equation modelings have designed, and these are for the male students,
female students, and for the gender (male and female). Model for male students’
shows that male students having an impact on the PE, SF, and on FC. Furthermore,
PE, EE, and FC described 98% of the difference in the BI to use mobile learning.
Structural equation model (SEM) for female students’ shows that female students
having an impact on the PE and on the FC. Moreover, PE and SF described 89%
of the difference to BI in terms of using mobile learning. Structural equation model
(SEM) for gender shows that gender has an impact on the EE and on the FC. In
addition, PE and SF described 89% of the difference to the BI in order to use mobile
learning and these results are in line with the findings [33].
The findings of this research in line with research question and the first research
question answer indicated that PE (β = −0.24 when P = 0.02) does not have a
significant impact on BI to use the mobile learning in the tertiary education. The
second research question answer indicated that EE (β = −0.50 when P < 0.01)
has a significant impact to BI to use the mobile learning in tertiary education. Third
research question indicated that SF (β = −0.05 when P = 0.32) does not have a
significant impact to BI in terms of using the mobile learning in tertiary education.
Fourth, research questions answered FC (β = −0.46 when P ≤ 0.01) got a significant
impact to BI in order to use the mobile learning in tertiary education. Second and
fourth hypotheses got accepted; however, first and the third hypotheses got rejected
since PE and SF do not significantly get the impact to BI in order to use the mobile
learning by international students. Having comparatively analyzed, it shows that
male international students got a significantly higher impact on the PE, SF, and FC;
nevertheless, female students got a significant impact on the PE and SF to BI in
order to use the mobile learning in tertiary education. Finally, PE, EE, and FC have a
significant impact on BI to use mobile learning by male students in tertiary education.

Acknowledgements Consent: All the author state that there is no conflict of interest. We take the
responsibility involved in the publication of our paper.
International Students’ Gender Impact to Use Mobile Learning … 1189

References

1. Berking P, Birtwhistle M, Gallagher S, Haag J (2013) Mobile learning survey report. Advanced
Distributed Learning Initiative
2. Taleb Z, Sohrabi A (2012) Learning on the move: the use of mobile technology to support
learning for university students. In: International conference on education and educational
psychology (ICEEPSY 2012). Proc Soc Behav Sci 69:1102–1109
3. Mansouri S, Kaghazi B, Kharmali N (2010) Survey of students attitude about M-learning in
Gonbad Payam-noor University. In: The first conference of Iran’s mobile value-added. IRIB
Conference Center, pp 1–9
4. Kafyulilo A (2012) Access, use and perceptions of teachers and students towards mobile phones
as a tool for teaching and learning in Tanzania. Educ Inf Technol 19(1):115–127
5. Ozuorcun NC, Tabak F (2012) Is m-learning versus e-learning or are they supporting each
other. Proc Soc Sci Behav 46:294–305
6. Turkle S (1997) Life of the screen: identity in the age of the internet. Touchstone, New York
7. Mitra A, Willyard J, Platt C, Parsons M (2005) Exploring web usage and selection criteria
among male and female students. J Comput Mediat Commun 10(3)
8. Zhang YX (2005) Age, gender, and Internet attitudes among employees in the business world.
Comput Hum Behav 21:1–10
9. Muhanna W, Abu-Al-Sha’r A (2009) University students’ attitudes towards cell phone learning
environment. Int J Interact Mob Technol 3(4)
10. Kwon HS, Chidambaram L (2000) A test of the technology acceptance model the case of
cellular telephone adoption. In: Proceedings of the 33rd Hawaii international conference on
system sciences, 2000. IEEE, Hawaii, pp 1–10
11. Wang Y, Wu M, Wang H (2009) Investigating the determinants and age and gender in the
acceptance of mobile learning. Br J Edu Technol 40(1):92–118
12. Kukulska-Hulme A (2009) Will mobile learning change language learning? Recall 21(2):157–
165
13. Hargittai E, Shafer S (2006) Differences in actual and perceived online skills: the role of gender.
Soc Sci Quart 87(2):432–448
14. Li N, Kirkup G (2007) Gender and cultural differences in Internet use: a study of China and
the UK. Comput Educ 48(2):301–317
15. Motiwalla LF (2007) Mobile learning: a framework and evaluation. Comput Educ 49(3):581–
596
16. Ong C, Lai J (2006) Gender differences in perceptions and relationships among dominants of
e-learning acceptance. Comput Hum Behav 22(5):816–829
17. Basak SK, Wotto M, Bélanger P (2018) University students’ m-learning adaption behavioral
factors: a pilot study. In: 2018 IEEE 9th annual information technology, electronics and mobile
communication conference (IEMCON), pp 68–73
18. Koohang A (2004) Students’ perceptions toward the use of the digital library in weekly web-
based distance learning assignments portion of a hybrid program. Br J Edu Technol 35(5):617–
626
19. Enoch Y, Soker Z (2006) Age, gender, ethnicity and the digital divide: University students’ use
of web-based instruction. Open Learn 21(2):99–110
20. Pew Internet and American Life Project (2008). [Link]
21. Kay RH (1992) An analysis of methods used to examine gender differences in computer-related
behaviour. J Educ Comput Res 8(3):323–336
22. Hilao MP, Wichadee S (2017) Gender differences in mobile phone usage for language learning,
attitude, and performance. Turk Online J Distance Educ TOJDE 18(2):68–79
23. Durndell A, Thomson K (1997) Gender and computing: a decade of change. Comput Educ
28(1):1–9
24. Whitely BE Jr (1997) Gender differences in computer related attitudes and behavior: a meta
analysis. Comput Hum Behav 13(1):1–22
1190 S. K. Basak et al.

25. Ong CS, Lai JY (2006) Gender differences in perceptions and relationships among dominants
of e-learning acceptance. Comput Hum Behav 22:816–829
26. Morahan-Martin J (1999) Women and internet: promise and perils. Cyber Psychol Behav
3(5):683–696
27. Li N, Kirkup G (2007) Gender and cultural differences in Internet use: a study of China and
the UK. Comput Educ 48:301–317
28. Kay RH (2009) Examining gender differences in attitude toward interactive classroom
communications systems (ICCS). Comput Educ 52:730–740
29. Tsai MJ, Tsai CC (2010) Junior high school students’ Internet usage and self- efficacy: a
re-examination of the gender gap. Comput Educ 54:1182–1192
30. Venkatesh V, Morris M, Davis G, Davis F (2003) User acceptance of information technology:
towards a unified view. Manage Inf Syst Quart 27(3):425–478
31. Cronbach LJ (1951) Coefficient alpha and the internal structure of tests. Psychometrika
16(3):297–334
32. Hair JF, Black WC, Babin BJ, Anderson RE (2010) Multivariate data analysis, 7th edn. Prentice
Hall, Englewood Vliffs
33. Alharbi S, Drew S (2014) Mobile learning system usage: scale development and empirical
tests. Int J Adv Res Artif Intell (IJARAI) 3(11):31–47
Industrial Automation of Process
for Transformer Monitoring System
Using IoT Analytics

Vaishali Khairnar, Likhesh Kolhe, Sudhanshu Bhagat, Ronak Sahu,

Ankit Kumar and Sohail Shaikh

Abstract Multiple devices interconnected with each other via the Internet are the
key concept behind IoT. It allows autonomous devices with the possibility to use the
Internet for communication and exchange of data. This paper focuses on monitoring
the transformer in real-time fault detection and records distinct operating parame-
ters of the transformer like voltage imbalance, load current, transformer oil levels,
temperature, vibration. Based on these parameters, the transformers fail state (i.e. a
state where transformer malfunctions or completely stops working) and health (i.e.
voltage, current, oil levels, temperature and vibration) are predicted by making use
of an artificial neural network (ANN) algorithm. Use of this technology can mini-
mize working efforts, thereby improving accuracy, stability, efficiency. Thus, remote
monitoring and machine control are achieved, as well as ANNs help to determine
the performance and yield appropriate measures accordingly. In this case, sensors
are used to sense the important parameters of equipment such as current, voltage
oil level in any operating transformer. By analyzing relevant data using ANNs, this
system will be beneficial in many industries. Likewise, this system is generalized to
be used in a wide array of industrial automated machines.

V. Khairnar · L. Kolhe · S. Bhagat (B) · R. Sahu · A. Kumar · S. Shaikh

Information Technology, Terna Engineering College, Navi Mumbai, India
e-mail: sudhanshubhagat101@[Link]
V. Khairnar
e-mail: vaishalikhairnar@[Link]
L. Kolhe
e-mail: likhesh8@[Link]
R. Sahu
e-mail: ronaksahu0204@[Link]
A. Kumar
e-mail: akumar.kumar043@[Link]
S. Shaikh
e-mail: sohailks118@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1191

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1192 V. Khairnar et al.

Keywords Real-time monitoring · Artificial neural networks · Prediction ·

Communication module

1 Introduction

In our daily life, electricity has become very essential and life without electricity is
hard to imagine. Similarly, electricity is also very crucial for different industries, and
an industry comprises a variety of different machines and appliances which need a
constant supply of electricity for their smooth functioning. These machines are also
responsible for effective production and manufacturing services which determine the
growth of any industry. Malfunctioning of these machines due to voltage instabilities
and currency fluctuations will lead to losses on large scale also due to the heavy
current applications of the industry, and the workers working on the field can be
subjected to dangerous shocks.
In order to avoid any damage to the machine as well as glitches like voltage
and current, distortions can be resolved and fixed with the help of a transformer.
A custom current transformer constantly measures the electric current, and even if
the current level exceeds due to electric fluctuations, the transformer automatically
eliminates excess current flow, thereby reducing the damage to a bare minimum. In
other words, transformers are designed to use electromagnetic induction to modify an
alternating current voltage that runs from one electric circuit to another. Proper care
and maintenance of the transformer are often neglected largely due to its reliability
leading to a cutback in the life span of the transformer, which will eventually result
in downturn and loss. Keeping this in mind, perception, veracity, and analysis of data
are important, which is obtained by monitoring of the transformers. To improve the
transformer’s functioning and reliability, evaluation of the maintenance data gathered
during monitoring is important. Furthermore, by making use of powerful (ANNs)
artificial neural networks, relevant information on existing issues can be obtained that
may result in predicting the need for the replacement or repair of the transformer.

2 Literature Survey

Transformers are important to power system components whose health condition

needs to be ensured for safe and reliable operation of any power system network [1].
The main aim of this system is the distribution transformer monitoring and controlling
through IOT. If there are any deviations from the normal or an emergency situation
occurs, the system sends SMS messages to the mobile phones containing information
about the deviation according to some predefined instructions programmed in the
microcontroller [2].
Industrial Automation of Process for Transformer Monitoring … 1193

3 Methodology

In this automated transformer monitoring system, we make use of technologies like

IoT and analytics which help to increase in the performance and throughput of the
system.
In order to perform analytics on any system using neural networks, we require
a large amount of dataset. This dataset is collected in real-time using different sen-
sors like temperature sensor (LM35), vibration sensor, humidity sensor (DHT11),
and ultrasonic sensor. These sensors are controlled and monitored with the help of
Arduino UNO microcontroller which is an open-source microcontroller board based
on the Microchip ATmega328p, developed by [Link]
The data collected by these sensors is then stored in SQL database with regular
time intervals. These sensors also notify the real-time state of the transformer. If a
condition arises with the transformer where it is overheating or damaged, it will be
quickly detected by the sensors attached to it and the microcontroller will inform the
server. A web portal is developed to interface this complex system so that the user
can have a high benefit of the system. This web portal has admin and user access.
The admin will have the privilege to monitor the entire system in real time, whereas
the user will focus on the output of the system.
Furthermore, once a large amount of data related to different factors of the trans-
former that influence its performance are collected by different sensors attached to
the transformer, the next step is to feed this data to the neural network algorithm.
The end goal of this system is to analyze the different sensor data and to determine
the fail state of the transformer. With the help of neural networks algorithm, we can
predict the health and the life of a transformer, and we can repair any damage to the
transformer prior to the further loss in the system (Fig. 1).
The functional block is divided into four fundamental blocks of the system. The
first block is being the power supply module which powers the entire system and
provides adequate voltage to the entire sensor network.
The second block includes the sensor section which includes all the different
sensors which help to monitor the different influential factors of the transformer.
The DHT11 sensor is used to measure the external temperature and humidity level
around the transformer. The ultrasonic sensor monitors the oil level in the oil tank
of the transformer and helps to calculate the oil consumption of the transformer.
Vibration sensors are sensors used for measuring and analyzing linear velocity and
displacement. It is used to identify and determine the condition and state of the
transformer. The LM35 temperature sensor is used to detect the precise temperature
in centigrade. It is used to measure the internal temperature of the transformer.
1194 V. Khairnar et al.

Fig. 1 Functional block

The next and very important block is the microcontroller where the sensors are
connected and controlled by the Arduino UNO microcontroller. These sensors are
programmed to calculate and measure different parameters of the transformer and
record them at regular time intervals on the server. These parameters determine
the present state and health of the transformer, and also in any conditions where
the sensors parameters indicate an irregular behavior of the transformer leading to
malfunction of any kind; in such cases, the microcontroller is programmed to take
appropriate measures suitable for the transformer in that present state which includes
powering down the transformer, thereby reducing the risk of high damage to the
transformer.
Industrial Automation of Process for Transformer Monitoring … 1195

The final and most crucial block of this system is the IoT block where the web
portal plays a major role in interfacing the entire system to the user, thereby making it
simpler to manage the system. Furthermore, the powerful neural network algorithm
is used for the prediction of the fail state of the transformer and also to determine
the health so that we can take necessary actions before the damage to the machine
increases. The ANN algorithm provides better accuracy and more precise results.

3.1 Artificial Neural Networks (ANNs)

The artificial neural network is the key element of this system, which takes in the
sensor values, stores in the database as an input parameter, processes this parameter
to produce a predictive analysis of the transformer, and determines its fail state. The
ANN algorithm undergoes data preprocessing which prepares the raw values stored
in a database for further processing. Relevant data might not get recorded due to
misunderstandings like equipment or sensor malfunction, the recorded data history
or modifications to the data can be overlooked the missing data, and particularly, some
attributes of tuples with missing values may need to be inferred. Data preprocessing
is a form of data mining approach where raw data is converted into a logical format.
It is a proven method for resolving such issues. The traditional data preprocessing
method starts with data which is assumed ready for analysis without any feedback.
The main difficulty for data preprocessing is an inconsistency between datasets. The
following are the major tasks involved in data preprocessing (Figs. 2 and 3).
• Data Cleaning: It is the process of filling the missing values, identifies, removes
outliers, smoothes the noisy data, and resolves inconsistencies.
• Data Integration: It is the process of integrating multiple files, data cubes, and
databases.
• Data Transformation: It involves the task of data normalization and data
aggregation, e.g., to transform V [min, max] to V [0, 1].
• Data Reduction: It is the process of reducing representation in terms of volume
but producing similar analytical results.
• Data Discretization: It is a part of data reduction along with specific importance
for numeric data.
We make use of ‘Keras’ library which gives a high level of abstraction and is
built on top of TensorFlow library, and we use TensorBoard to analyze our model’s
efficiency.
1196 V. Khairnar et al.

Fig. 2 Steps involved in

data preprocessing

4 Results

Predicting the fail state of a transformer is the most crucial part of this system. The
ANN will consider the transformer parameters for prediction and notify the user
accordingly. The developed system is designed in such a way that it can activate the
emergency alert. Figure 4 shows the confusion matrix of the ANN model.
Figure 5 is the accuracy graph which shows the comparison of accuracy of training
set of the data with accuracy of testing set of the data. Figure 6 is the loss graph
which compares the losses occurred in the training set of the data with the testing
set of the data. Real-time data is collected and processed by ANN, and the result is
predicted; taking into consideration, the testing set of the data is displayed on the
web application.
Industrial Automation of Process for Transformer Monitoring … 1197

Fig. 3 Process of
preprocessing

5 Conclusion

This paper discusses the complete solution for monitoring and controlling the trans-
former by using sensors, relay module, communication module, and web application.
The designed system also can predict the fail state of the transformer and alert the
user about it.
1198 V. Khairnar et al.

Fig. 4 Confusion matrix

All the transformer sensors are continuously monitored by ANN. The user is
notified when the transformer experiences an abnormal condition, appropriate actions
can be taken to prevent any disastrous failure of a power transformer.
By using web application, user can view historical data and monitor real-time
parameters of the transformer easily.
Industrial Automation of Process for Transformer Monitoring … 1199

Fig. 5 Accuracy graph

Fig. 6 Loss graph

1200 V. Khairnar et al.

References

1. Balaji Kamlaesan K, Kumar A, Alex David S Analysis of transformer faults using IOT. In:
2017 IEEE international conference on smart technologies and management for computing,
communication, controls, energy and materials. Veltech Dr. RR & Dr. SR University, Chennai
2. Sivaranjani S, Lokesh S, Vignesh M, Vijayaragavan N, Goutham S IOT based distribution trans-
former monitoring system. Int J Current Trends Eng Res (IJCTER) (Department of Electrical
and Electronics Engineering, V.S.B Engineering College, Karur)
A Comparative Study on Various Search
Techniques for Gaming Applications

B. Sunil, M. R. Naveen Kumar, B. N. Gowrishankar and N. S. Prema

Abstract The main objective of this research work is to deduce best among the
existing search techniques. The various methods or the algorithms that we are going to
discuss here are breadth-first search, depth-first search, and A* algorithm. Comparing
various search techniques that are used in artificial intelligence problems for gaming
applications.

Keywords Search techniques · DFS · BFS · A*

1 Introduction

Searching is a process of determining whether an element is a part of the data set or

not. Most of the search belongs to either sequential or binary. A sequential search is
conducted on small data sets where the list is unsorted. Binary search is performed
on large data sets of the sorted list. The motivation behind the searching technique
is to solve artificial intelligence problems; here, we are considering many gaming
applications and how the different search algorithms are used to solve the game
problems and how the searching algorithms are modified to get accurate results
depending on the type of problems. Performance of artificial intelligence problems
mainly depends on the complexity of the searching techniques.

2 Review of Literature

In [1], the author has proposed a better algorithm to solve the Japanese puzzle which
is a combination of logical rules and DFS where the performance of the algorithm is
better than the DFS. The results which they have got are for 5 * 5, 5 * 6 and 10 * 10;

B. Sunil (B) · M. R. Naveen Kumar · B. N. Gowrishankar · N. S. Prema

Department of Information Science and Engineering, Vidyavardhaka College of Engineering,
Mysuru, Karnataka, India
e-mail: sunilbayanaboyana12@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1201

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1202 B. Sunil et al.

the time taken is less than 0.1 s. The author has concluded that the Japanese puzzle
problem is solved by two phases; first, with the logical rules to solve cells as many
as possible and in the second phase, the unknown cells are solved using DFS with
branch and bound scheme based on column information. The method which they
have used gives correct answers for black patterns quickly and also fastens the DFS
algorithm.
In [2], the author has proposed efficiency, comparison rapidity, and intelligence
about seven pathfinding an algorithm for 3D gaming applications and also proposed
depth direction A* method which uses linear graph theory to find shortest paths in
maps and to avoid barriers in the multilayer environment. The result which they
have got for one layer of ground that is outdoor is as follows for different search
algorithms considering various maps. The author has concluded that the iterative
depth D A* algorithm uses the least memory with time complexity O(b * d) and
space complexity O(d). The depth direction A* expands more node than A* and has
a very smooth pathway.
In [3], the author has proposed an algorithm for Babylon tower game that uses
DFS column by column to reach the goal state. The game goal is to sort each ball of
the same color on the same column and sort each ball in each column based on the
brightness of the color of the ball. The average number of steps taken to reach the
goal state is 151 steps for easy, 295 steps for medium difficulty, and 431 steps for
hard difficulty. And, the time taken for easy is 9.17680325 ms, for medium difficulty
13.659335 ms, and for hard difficulty 16.8398884.
In a study in [4], the author has proposed the combination of A* algorithm and
dynamic pathfinding A* algorithm for a game of racing car which states about the
obstacles in the dynamic environment of the game and how dynamic pathfinding
A* addresses this problem. In the result, the author has proposed the comparison
between the combined algorithm A* and DPA with DPA for different states of car
bending at 60°, 90°, spiral trajectory, lane letter S, empty trajectory and trajectory
with obstacles based on the results of all the abovementioned states combination of
A* and DPA has given the car reaching the final state and in the case of obstacles,
the finish line is reached three times out of five experiments, and finally, the author
has concluded that by combining A* and dynamic pathfinding algorithm can be
implemented in car racing game; the grid representation method used has an effect
on the route real obtained by A* algorithm. The DPA has affected by the obstacle
position and trajectory shape.
In [5], the author used the concept of artificial intelligence where goat must escape
the enemies, so the player playing in goat’s place must win the game. A* algorithm
is used to find the shortest distance between the goat and the plants in the form of a
labyrinth and should be able to mislead the enemy. Pathfinding is the application of
the A* algorithm to the game.
A* is an algorithm used to solve many problems. In pathfinding, this algorithm
checks the most promising node location it has ever seen. If the node visited is the
promising or the target node, then the algorithm will stop the execution. If not, the
algorithm will store the node’s neighboring data for further nodes. It is the famous
algorithm used to find the path in the artificial intelligence field [6].
A Comparative Study on Various Search Techniques … 1203

3 Search Techniques Methodology

3.1 Breadth-First Search

If we consider the BFS searching technique, a guaranteed solution will be given if the
searching element is in the list. The complexity in BFS is that it visits each and every
node in the list or tree and those nodes will be stored. If memory is not a constraint
in your problem, then BFS technique gives a very good result.
BFS is most similar to Dijkstra’s Algorithm for the shortest path. Here, the search
is exhaustive, entire graph or tree at each level is expanded, and the search operation
is performed [7]. The search will go up to the deepest level of the tree and memory
to store each and every node is more.
Algorithm: BFS
a. Create a Node_List variable and set it to start.
b. Until there is a goal state or Node_List is empty:

a. Remove from Node List the first element and call it E. If Node_List has been
empty, quit.
b. For each instance where each rule corresponds to the state described in E do:
i. Use the rule to create a new state.
ii. If the goal state is the new state, quit this state and return it.
iii. If not, add the new state to Node_List’s end.

3.2 Depth-First Search

It works on the technique of uniform search by generating the decedents of most

recently expanded node until the goal node cutoff reached and then backtracked one
of the most recently expanded nodes. In Depth-first search, only the nodes from
initial to goal node are stored, so less memory is required [8]. One of the limitations
of DFS is that if the goal node is not present in the graph or tree, then the search will
be exhaustive and it searches only one side of the tree and reach deadlock situation.
Algorithm: DFS
a. If the starting state is the goal state, quit success and return it.
b. Otherwise, do the following until you find or fail the goal state.

a. Generate the initial state’s successor, E. Signal failure when there are no more
successors.
b. Call DFS as the starting state with E.
c. Signal success if success is returned. Continue in this loop otherwise.
1204 B. Sunil et al.

3.3 A* Algorithm

A* is the most popular algorithm used for finding a path between two locations in
the map area [9]. The A* algorithm has two sets one is an open set and another one
is the closed set. The open set is used to record the areas adjacent to those already
evaluated and the closed set is used to record the areas already evaluated [10]. The
heuristics which is used in A* algorithm are

F(n) = g(n) + h(n)

where g(n) represents the cost of the path and h(n) represents the heuristic value that
is estimated cost form vertex n to goal.
a. Two sets are OPEN and CLOSED. The OPEN node contains the candidate nodes
for examination. There is only one element in the OPEN list which is the starting
position.
b. The CLOSED set contains the nodes that were already examined and will be
empty initially closed.
c. There is a loop that repeatedly pulls out and examines the best node n in the
OPEN list; if n is the goal node, then we are done. Otherwise, n for the OPEN
list will be removed and moved to the CLOSED list.
d. A neighbor who is CLOSED has been seen before, so we do not need to look at
it, so if his f value becomes the lowest in OPEN, a neighbor who is in OPEN
will be examined. Otherwise, we will add it to OPEN, the parent being set to n.
The cost of the path is n , g(n) is g(n) + momentcost (n, n ).

4 Comparison of Various Search Techniques

Table 1 gives the idea of how the various search techniques are performing under
different parameters for gaming applications.

5 Conclusion

By comparing the different search techniques for gaming applications to identify the
shortest path, A* algorithm is better than the DFS and BFS. A* algorithm is well
suited for gaming applications in a dynamic environment with obstacles to play. In
future planning to propose a new algorithm for the shortest path from the initial state
to a final state of gaming applications.
A Comparative Study on Various Search Techniques … 1205

Table 1 Comparison of various search techniques

Parameters Breadth-first search Depth-first search A* algorithm
Memory High Low Low
Structure of Wide and short Narrow and long Straight line
constructed tree
Optimality Optimal, not in cost Not optimal Optimal
Data structures to Queue Stack Graph
store nodes
Performance Low High High
Space complexity More critical Less critical Less critical
Time complexity Less High Less
Basic Vertex-based Edge-based algorithm Abstract data type
algorithm
Traversing fashion Oldest unvisited Vertices along the Predecessor
vertices are explored edge are explored in
at first the beginning

References

1. Jing MQ, Yu CH, Lee HL, Chen LH (2009) Solving Japanese puzzles with logical rules and
depth first search algorithm. Proc Int Conf Mach Learn Cybern 5:2962–2967
2. Khantanapoka K, Chinnasarn K (2009) Pathfinding of 2D & 3D game real-time strategy with
depth direction A* algorithm for multi-layer. In: 2009 8th international symposium national
language processing SNLP’09, pp 184–188
3. Rahmat RF, Harry Syahputra MF, Sitompul OS, Nababan EB (2018) The depth-first search
column by column approach on the game of Babylon Tower. In: Proceedings of the 2nd
international conference on informatics and computing, ICIC 2017, pp 1–6
4. Sazaki Y, Satria H, Syahroyni M (2017) Comparison of A∗ and dynamic pathfinding algorithm
with dynamic pathfinding algorithm for NPC on car racing game. In: 2017 11th International
conference telecommunication systems services and applications, pp 1–6
5. Harsani P, Mulyana I, Zakaria D (2018) Fuzzy logic and A* algorithm implementation on goat
foraging games. IORA-ICOR 2017. IOP Conf Ser Mat Sci Eng 332:1–7
6. Xiao C, Hao S (2011) A*-based pathfinding in modern computer games. Int J Comp Sci Netw
Secur 11(1):125–130
7. Breadth First Search. [Link]
8. Depth First search. [Link]
9. Introduction to the A* Algorithm. [Link]
10. A* Search Algorithm. [Link]
Crowdsourcing Data Analysis for Crowd
Systems

Vignesh Loganathan, Gopinath Subramani and N. Bhaskar

Abstract Crowdsourcing permits huge scale and adaptable summon of human con-
tribution for information social affair and examination, which presents another world
view of information mining process. Customary information mining techniques fre-
quently require specialists in investigative areas to comment on the information. In
any case, it is costly and ordinarily takes a long time. Crowdsourcing empowers the
utilization of heterogeneous foundation information from volunteers and circulates
the explanation procedure to little segments of endeavors from various commitments.
The data analytics process is done via logistic regression.

Keywords Crowdsourcing · Data analytics

1 Introduction

Crowdsourcing is a type of human calculation, where human calculation is a tech-

nique for having individuals do things that we may somehow think about doling
out to a processing gadget to compute consequently, e.g., a dialect interpretation
errand. A crowdsourcing framework encourages a crowdsourcing procedure to fin-
ish an indicated errand. To do an undertaking, a crowdsourcing framework enrolls a
“swarm” of human specialists to help take care of a characterized issue. Data ana-
lytics is the study of breaking down data to change over data to helpful information.
This information could enable us to comprehend our reality better and in numerous
settings empower us to settle on better choices. While this is the wide and fabulous

V. Loganathan (B) · G. Subramani · N. Bhaskar

Department of Information Technology, KCG College of Technology, Chennai, India
e-mail: vigneshloganathan0598@[Link]
G. Subramani
e-mail: [Link]@[Link]
N. Bhaskar
e-mail: bhaskar@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1207

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1208 V. Loganathan et al.

target, the most recent 20 years have seen steeply diminishing expenses to accumu-
late, store, and process data, making a considerably more grounded inspiration for
the utilization of experimental ways to deal with critical thinking.

2 Overview

2.1 Automated

With the data analytics necessities and data as the information sources, it can natu-
rally achieve end-to-end data analytics assignments with human exertion. It under-
stands the struggle between the requirement for critical data examination in little and
medium assembling undertakings also, existing data investigation stages being time
[1].

2.2 Well-Disposed to Non-master Clients

An easy to understand interface is not excessively intricate, however rather is direct,

giving brisk access to normal highlights or directions. This approach is highly reliable
to non-expert users [1].

2.3 Human Intelligence

A Human Intelligence Task, or HIT, is an inquiry that needs an answer. A HIT speaks
to a solitary, independent errand that a worker can chip away at, present an answer,
and gather a reward for finishing [2].

2.4 Applications

There are several examples can be given for the process of crowdsourcing and data
analytics. Among that, the most important and popular example of crowdsourcing
was Amazon Mechanical Turk. Amazon Mechanical Turk (MTurk) is a commercial
center for work that requires human insight [2]. The crowdsourcing has been utilized
in number of research and business applications. The Wikipedia is a precedent in
which set of data is being distributed and later on number of individuals can include
Crowdsourcing Data Analysis for Crowd Systems 1209

and upgraded the data at long last the best of the best-upgraded information is acces-
sible for the client which is impossible by utilizing a solitary PC alone and could be
costly to accomplish through expert domain process [3].

3 Related Work

3.1 Data Analysis Framework [1]

Data analysis depends on R language. Every one of the information investigation

tasks in Data analysis are converted into R proclamations. Data analysis resem-
bles an interlayer between information investigation prerequisites and information
investigation undertakings portrayed by R language [4]. The general system struc-
ture of GDMA is appeared in Fig. 1. To be exceptionally mechanized and conven-
tional, GDMA mostly depends on three noteworthy framework parts, an information
processor, a calculation selector, and a parameter enhancer.

System workflow
An entire information investigation process starts with a client’s order. GDMA gets

Fig. 1 GDMA framework

1210 V. Loganathan et al.

the errand depiction and information processer naturally chooses the properties type
and shows it to the client. The client can modify the traits sort utilizing directions
on the off chance that they wish. At that point, calculation selector chooses the most
suitable calculation as indicated by the errand type and information attributes point
by point. On the off chance that fundamental information processor will process the
information to ensure that it is appropriate for the chosen calculation. For instance,
multilayer perceptron does not endure missing or discrete qualities and these must be
evacuated. Next, if the calculation has parameters to be resolved, parameter stream-
lining agent will decide the ideal parameter values, as point by point. Following this,
GDMA utilizes the chose calculation and decided parameter settings to dissect the
handled information and build up an investigation demonstrate. At last, the examina-
tion results are acquired and clients can affirm on the off chance that they are happy
with it, if not, they can modify the parameters or even change the calculation until
the point that they are.

Advantages
In this framework, the major advantage was to analyze the data given for query
processing and publish the satisfied result to the user.

Challenges
In this concept, the greatest challenges were to optimize the parameter [5] and
selection of algorithm for query processing.

3.2 Data-Less Big Data Analytics [6]

Scalable, efficient, and accurate (SEA) examination is getting to be the holly ves-
sel of information science research in the enormous information time. All research
networks associated with enormous information science have perceived and grasped
the intrinsic difficulties. Since the turn of the century, new frameworks, encourag-
ing adaptability with enormous capacity, the board and parallel/conveyed preparing
have been created. Also, the requirement for additional productivity, notwithstand-
ing adaptability, was getting to be clear and new frameworks for quick investigation
developed to fulfill such needs [7, 8].
Imagine an information framework, which can process examination questions
without getting to any base information, while giving precise answers! This would
be a definitive in versatility (as question preparing times move toward becoming
accepted obtuse to information sizes), what is more, a definitive in effectiveness,
as tedious, asset hungry co-operations inside appropriated and complex enormous
information examination stacks are totally stayed away from. Figure 2 given below
will explain the process of data-less big data analysis. The upper half portrays tradi-
tional processing, while the base demonstrates the data-less processing worldview.
Inquiries are submitted to the framework as previously. Utilizing known and gen-
erally acknowledged outstanding burden qualities, (i.e., that questions commonly
Crowdsourcing Data Analysis for Crowd Systems 1211

Fig. 2 Data-less analytics processing

characterize overlapping data subspaces), our methodology basically infuses an “in-

telligence” which screens and gains from the investigator framework collaborations
and creates novel ML models which are utilized to precisely foresee answers to future
inconspicuous diagnostic questions [6].
The endeavors here spotlight on comprehension and utilizing the conduct of inves-
tigators (clients), explicitly, to determine novel calculations, structures, and models,
which can gain from what investigators are doing. What are their questions? Which
data subspaces would they say they are focusing on? At the point, when do their inter-
ests move? Up until now, we have favored a demonstrating of investigative inquiries
as vectors in multidimensional vector spaces. Checking what is more, gaining from
client conduct at that point adds up to, in pith, inquiry space quantization. This
comes connected at the hip with portion works that characterize likeness between
said inquiry quanta and new approaching inquiries.
In the meantime, the emphasis is on creating novel models, calculations, and
structures that gain from what the framework does in light of explicit examiner
questions. What are the appropriate responses to examiner inquiries? How were
they registered? Utilizing which data subspaces? The end objective here is to learn
and display the attributes of questioned data subspaces. For instance, what are the
1212 V. Loganathan et al.

circulations of measurements on enthusiasm for the data things inside questioned

subspaces.

Advantages
Analytical query preparing then continues by utilizing a fitting closeness separate
capacity for deciding the nearest query quanta(s) to the query point. [9, 10, 11].

Disadvantages

1. Maintain said models within the sight of query space updates and information
space refreshes.
2. Infer the most suitable models for the question and data spaces.

3.3 Crowdsourcing Predictors [2]

The fundamental idea of this framework is to make accessible a spot where farmers
can make inquiries and answer the inquiries asked by their friends. Figure 3 demon-
strates the stream of the framework. The new client needs to enlist her/him to the
framework. After an effective login, the existing client will choose the classification
of organic products, blooms, vegetables. As indicated by the class, she/he will post
an inquiry and can give a response to an unanswered inquiry. The arrangement of
posting the pictures of surrendered and harmed items is given. The job of domain
expert is predominant. On the off chance that client is not happy with the appropriate
responses given by different clients, at that point, she/he can tap on not fulfilled.
After that, the inquiry will send to the domain expert. Domain expert will give the
response to that question and that answer will be given back to the clients. A client

Fig. 3 System flow diagram

Crowdsourcing Data Analysis for Crowd Systems 1213

can give a rating to the appropriate response of different clients. The beneath Fig. 3,
clarifies the stream of the framework and will clarify the working of framework.

Advantages
The facility of giving pictures as opposed to utilizing words to clarify makes the
framework not quite the same as others. The framework is totally free of expense
and only a piece of administration to serve the country [2].

Disadvantages
System does not show the predicted result since this has no analysis of data and
shows all results regarding the queries [2].

4 Proposed System

The proposed system comprises the process of crowdsourcing and data analytics.
Our approach overcomes the limitations that are occurred in the previous system.
Crowdsourcing system is designed in a way to help the farmers to gather the sugges-
tions in the new/alternate techniques of the agriculture. The system helps to share
information between the farmers. The farmer can able to post queries as a requester
and farmer can able to share their suggestions as a worker. The suggestions are getting
shared throughout the system. This suggestion is getting monitored by the expert in
the system. The expert has the authority to remove the unwanted suggestions given by
the workers/requestors. The analysis of queries is done using the Logistic regression
algorithm for categorical data.

Logistic Regression
The idea of Logistic Regression is to find a relationship between features and proba-
bility of particular outcome. E.g.… This type of a problem is referred to as Binomial
Logistic Regression, where the response variable has two values 0 and 1 or pass and
fail or true and false.
Prediction from categorical data using Logistic Regression Algorithm (sample)
(Table 1):

Table 1 Analysis of the process

GDMA Data-less Expert system Proposed system
framework analytics
Data analysis ✓ ✓ ✘ ✓
User friendly ✘ ✘ ✓ ✓
Analysis of data ✓ ✓ ✘ ✓
(by Logistic
Regression)
1214 V. Loganathan et al.

#Farm dataset and test

> farm < - alldata[!([Link](Survey))]
> farm [,Survey : = [Link](Survey)]
> test < - alldata[[Link](Survey)]
> test [,Survey : = NULL]
#Logistic Regression
> model < - glm(Survey ~ ., family = binomial(link = ‘logit’),
data = farm[,-c(“FarmerId”,“Name”,“Location”)])
> summary(model)

5 Conclusion

Crowdsourcing approach audits ongoing work on crowdsourcing-based data mining

strategies. Crowdsourcing can do data mining and concentrate expansion data from
the datasets more effectively and keenly than conventional techniques. It needs to
manage bunches of difficulties like the low nature of answers from the groups to
apply crowdsourcing to data mining. In our approach, we bring up research process
to overcome these challenges and present the general systems of an incorporated
data mining undertaking in crowdsourcing.

References

1. Zhang H, Wang H, Li J, Gao H (2018) A generic data analytics system for manufacturing
production. Big Data Min Anal 1(2): 160–171. ISSN: 2096-0654 06/06
2. Pawar CV, Patni SS, Wale SK, Chemate HB, Dasgupta A (2015) India exploring agriculture
sector using Crowdsourcing predictors (IJRASET). Department of Information Technology,
Amrutvahini College of Engineering, Sangamner, Maharashtra, Apr 2015
3. USAID (2013) From the American people, briefing paper Crowdsourcing application for
agricultural development in Africa, pp 1–6
4. Racine JS (2012) RStudio: a platform-independent IDE for R and sweave. J Appl Econom
27(1):167–172
5. Keser SB, Yayan U (2016) A case study of optimal decision tree construction for RFKON
database. In: Proceedings of 2016 international symposium innovations in intelligent systems
and applications (INISTA), Sinaia, Romania, pp 1–6
6. Peter Triantafillou Department of Computer Science (2018) UK data-less big data analytics
(towards intelligent data analytics systems). In: 2018 IEEE 34th international conference on
data engineering. University of Warwick
7. Melnik S, Gubarev A, Long JJ, Romer G, Shivakumar S, Tolton M, Vassilakis T (2011) Dremel:
interactive analysis of web-scale datasets. Commun ACM
8. Engle C, Lupher A, Xin R, Zaharia M et al (2012) Shark: fast data analysis using coarse-grained
distributed memory. In: Proceeding of ACM SIGMOD
Crowdsourcing Data Analysis for Crowd Systems 1215

9. Anagnostopoulos C, Triantafillou P (2015) Learning set cardinality in distance nearest

neighbours. In: Proceeding of IEEE international conference on data mining (ICDM15)
10. Anagnostopoulos C, Triantafillou P (2015) Learning to accurately count with query-driven
predictive analytics In: Proceeding of IEEE international conference on big data
11. Anagnostopoulos C, Triantafillou P (2015) Efficient scalable accurate regression queries in in-
dbms analytics. In: Proceeding of IEEE international conference on data engineering (ICDE17)
Optimized Database Using
Crowdsourcing

Vignesh Balakrishnan and N. Bhaskar

Abstract RDBMS is the most normal and antiquated way to deal with database
arrangements. The data is kept in an extremely organized methodology in style of
tables or relations. With appearance of Big Data in any case, the organized method-
ology misses the mark to serve the needs of monstrous data frameworks that are
basically unstructured in nature. Expanding ability of SQL, however, allows huge
amount of data to be overseen, and it does not generally make a difference as a
response to Big Data frameworks that anticipates brisk reaction and quick speedy
adaptability. To comprehend this disadvantage, a sensibly new database framework
suggested to as NoSQL was presented. NoSQL framework is acquainted which give
the fast adaptability and unstructured stage for Big Data application.

Keywords Crowdsourcing · MySQL · NoSQL · SSIS

1 Introduction

As the technology is being evolved, human too evolved with the technology. They
always want things to be done in a most efficient and simple manner. The main
objective of this approach is to showcase the throughput time variation between
MySQL and NoSQL databases in form of retrieving same data from both databases.
In the existing system, uploading large amount of data is slow and time consuming.
In proposed system, uploading large amount of data is fast and time reducing. The
proposed system is to show an optimized database by the performance comparison
using throughput time variation between the MySQL and NoSQL databases using
crowd sourcing application. Deduplication, which can save storage cost by enabling
us to store only one copy of identical data, becomes unprecedentedly significant with
the dramatic increase in data stored in the databases.

V. Balakrishnan (B) · N. Bhaskar

Department of Information Technology, KCG College of Technology, Chennai, India
e-mail: bvignesh253@[Link]
N. Bhaskar
e-mail: bhaskar@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1217

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1218 V. Balakrishnan and N. Bhaskar

2 Overview

2.1 SQL Server Integration Services

Performance task of uploading large amount of data in database is increased. If error

occurred in the insertion of data or the column left null, SSIS indicates about these
things to the user to rectify it. It is a programmable object model that allows the
developer to write their own piece of code. Deduplication operation is adopted for
eliminating duplicate copies of redundant data [1].

2.2 Service-Level Agreement

A service-level agreement is a formal contract concerning a given service between

two parties: the admin and the user. The SLA explicitly defines the concerned service
using measurable metrics such as the service availability and performance. If the user
violates the agreement, the SLA indicates to the admin regarding the violation. And
also used to guarantee the satisfaction of the service quality expected by the customer.
An administration-level ascension (SLA) is an agreement between a specialist co-op
and its inward or outside clients that archives what benefits the supplier will outfit
and characterizes the administration models the supplier is committed to meet [2].

3 Related Work

Recent crowdsourcing systems, such as CrowdDB and Qurk and Deco, provide an
SQL-like query language as a declarative interface to the crowd. The query that works
as indicated by our necessity is not constantly valuable for us, for example, on the
off chance that information is refreshing on non-endurable time. We have to advance
question for better execution. Each inquiry gives same outcome yet the question
with best time normal is the thing that we need and use expertly. The database query
optimizer to choose efficient implementations using indices to evaluate predicates or
selecting join algorithms. The user has to submit an SQL query and the system takes
the responsible for compiling the query, generating the execution plan.
Optimized Database Using Crowdsourcing 1219

Advantages
In paper [3], develops an efficient algorithm for optimizing selection queries, join
queries, and complex selection-join queries and the second stage validates our
approach.

Challenges
In crowdsourcing concept, there is a slight difficulty in dealing with a large amount
of data processing. Lack of a quantitative decision support process to identify better
concepts [3].

4 Proposed System

Crowdsourcing allows large-scale and flexible invocation of human input for data
gathering and analysis. With the help of this, admin gathers information from the
employee and manager and produces an output to the user as per their request.
Admin designed those menus with some restrictions involved in it. The menus are
provided as an instant template when they login their page within a second. And
then, the optimized tables are stored as a report in a dynamic query table. Using key
deduplication, duplicate data are avoided while storing in the databases.
The crowdsourcing Web application has been designed in a way for the user to
show the performance comparison between the two different databases. So, to check
the comparison, the graphical image is designed to see the variation throughput time
from the given input. User can check the comparison efficiency between MySQL
and NoSQL databases. As we are introducing service-level agreement algorithm in
this crowdsourcing application, privacy is also maintained between each user and
the admin. SSIS method, is a platform to creates efficient data integration solutions.
Thus, the following table will show the different process of execution in different
databases (Table 1).
Efficiency: The capacity, calculation and correspondence overheads related with
the enormous information deduplication ought to be as little as could be allowed,
and the expense of seeking for copied information ought to likewise be limited [5].
Confidentiality of data: Security analysis demonstrates that key duplication
ensures data confidentiality and convergent key security, and well protects the
ownership privacy simultaneously [2].
Result: The final result will be performance variation between two different
databases is shown in Fig. 1.

5 Conclusion

Hence, by our approach, performance variation between two different databases are
shown in the form of graph. It helps the user to choose right database regarding the
1220 V. Balakrishnan and N. Bhaskar

Table 1 Basic queries in two different databases [4]

Query MySQL database NoSQL database
Create query CREATE TABLE table_name No need for defining schema
(column_name1 datatype,
column_name2 datatype)
Insert query INSERT INTO table_name db.collection_name. insert ({name1:
(column_name1, column_name2) value1,name2: value2})
VALUES(value1,value2)
Delete query DELETE FROM table_name db.collection_name.remove
WHERE (condition) ({condition})
Import query BULK INSERT table_name FROM mongoimport–db database_name
file_name WITH -collection collection_name–type
{FIELDTERMINATO R =’,’, csv–file “file_name”
ROWTERMINATOR =’\n’} GO
Select query SELECT column_name FROM db.collection_name({}, {condition})
table_name

Fig. 1 Performance comparison graph between MySQL and NoSQL

process. The future research should include many databases with proposed system
and brings out the throughput time variation among those databases.

References

1. Chauhan D Using the advantages of NOSQL: a case study on MongoDB. Department of

Computer Science, Himachal Pradesh University, Summerhill, Shimla, India
2. ETL Function Realization of Data Warehouse System Based on SSIS Platform Tong Wu College
of Computer Science and Information Technology, Guizhou University Guiyang, China 550025.
wtxx@[Link]
Optimized Database Using Crowdsourcing 1221

3. Forbes HL, Schaefer D (2018) Crowdsourcing in product development: current state and future
research directions. In: International design conference, design 2018
4. Chandra AK, Merlin PM (1977) Optimal implementation of conjunctive queries in relational data
bases. In: Published in the proceedings of ninth annual ACM symposium theory of computing
(STOC)
5. Liu L, Zhang Y, Li X (2018) KeyD: secure key-deduplication with identity-based broadcast
encryption. IEEE Trans Parallel Distrib Syst 26(5):1206–1216
Anti-Counterfeit on Medicine Detection
Using Blockchain Technology

R. Anand, Khadheeja Niyas, Sorjeeta Gupta and S. Revathy

Abstract The goal of this project is to find whether a given medicine is fake or
original using blockchain technology. Blockchain is an incorruptible digital ledger
of economic transactions that can be programmed to record not just financial trans-
actions but also has the potential to disrupt other markets. It removes the need for
trusted intermediaries, can facilitate faster transactions and add more transparency.
A medical product is counterfeit when there is false representation in relation to
its identity or source. In the case of medicines, the implications are more severe.
This technology stops the entry of fake drugs into the supply chain, mainly the
part between the manufacturer and consumer. The technology uses digital signature
where each block gets a unique crypto id. This digital signature is provided for each
block and it gives a strong control of ownership.

Keywords Blockchain · Supply chain · Healthcare

1 Introduction

Nowadays, for any transaction to take place between two parties, there is always a
trusted intermediary in between. But using blockchain, there is no need of a mid-
dle party between the consumer and supplier. Blockchain technology is a new and
rapidly growing technology. This technology was first introduced in 1991 and they
considered assigning a time period to digital documents and so, it is not possible

R. Anand (B) · K. Niyas · S. Gupta · S. Revathy

Department of Information Technology, KCG College of Technology, Chennai, India
e-mail: nowhereanand@[Link]
K. Niyas
e-mail: khadheejaniyaz123@[Link]
S. Gupta
e-mail: get2sorjeeta@[Link]
S. Revathy
e-mail: rey.revathy16@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1223

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1224 R. Anand et al.

to make any changes just like an agreement. Using this technology, we can handle
the data, manage the exchange or transaction between any two parties, and store the
details about any new product. Implementation of this technology in many indus-
tries leads to its massive success. Blockchain is a distributed network and it is a
write-once-read-only record of details. The data are stored in peer-to-peer network
where information about any product can be viewed by anyone that is in the network.
The chain stores information about any transaction or exchanges that occur in the
network. It is similar to all other network but what makes blockchain technology dif-
ferent is that no data can be deleted or altered by anyone in the network. No changes
can be made in the network until it is validated by all the authorized users of the
network.

Blockchain is a chain of blocks and each block contains information. A block is

a set of data and is identified with a cryptographic hash and timestamp. Each block
contains the newly entered data, unique hash value, and the previous hash value.
When a new block is added, it contains the previous hash values so that there will
be an order from the first block that was first ever generated in the chain. This first
block is called genesis block. The same process is repeated again and again to grow
and maintain the block. The blocks of data are stored in a linear chain fashion. It
is a continuously growing list of digital records which are linked and secured using
cryptography.
A block is just a collection of information that is needed by the organization for
the purpose of keeping track of the processes or to identify the addition or removal
of products from the chain (Fig. 1).
Two people wish to interact with each other for transaction purpose or for exchange
of data. Suppose A wants to send money to B in a network, the transaction takes
places through the blocks that are interconnected in the chain. This transaction takes
a block and that block is broadcasted to every other party in the network. Those in
the network check whether the block is valid or not. If it is valid, then they approve
for the transaction to occur. The block can then be added to the chain to be used for
future references. The money moves from A to B.
Anti-Counterfeit on Medicine Detection … 1225

Fig. 1 Block working

process

In blockchain technology, there are two keys: private key and public key. The main
purpose of these keys is to create a secure unique identity. Identity is based on the
private and the public cryptographic keys. The combination of a private and a public
key generates a highly secure digital signature. This digital signature is provided for
each block, and it gives a strong control of ownership.
In the future, many industries will use blockchain technology. Some industries
already using blockchain technology are:

Banking and payments

Banks access financial services and it connects billion of people from different coun-
tries in the world, like bitcoins allow anyone to transfer money, banks also increas-
ingly allow anyone to do the payments. Banks are now investing in blockchain; the
BARCLAYS is the bank which is already using the technology.

• Cyber security

Allowing the blockchain ledger public, the data is encrypted using cryptography and
this may lead the data to be less hacked and it cannot be changed. It also eliminates
the need of a middleman system and makes it more efficient, this allow only the
authorized user to access the data.

• Supply chain management

With blockchain technology transaction, the supply chain can be documented in a

permanent decentralized record, and it also monitored securely and transparently.
This helps to reduce the time delay, loss of data, and human mistakes. It also verifies
1226 R. Anand et al.

the authenticity of a product by tracking them from the origin, and it also understands
the environmental impact of the products.
• Healthcare

One of the challenges healthcare faces is the lack of a secure platform to store and
share data and they are often victims of hacking because of outdated infrastructure.
This technology allows hospitals to safely store data like medical records and share
it with authorized professionals or patients. Gem and Tierion are the two companies
that are working on disrupting the current healthcare data space.

• Insurance

The global insurance market is based on trust management. Blockchain is a new way
of managing trust, and it is used to verify many types of data in insurance contract,
like the insured person’s identity.

2 Proposed System

This project helps to track the manufacturing process of the medicines, right from
the raw materials to the retailers and consumers, using blockchain technology. This
ensures that there is no tampering in important information about the medicines. The
proposed system uses blockchain technology because by this technology, the network
can be open to everyone. All users of the network are aware of the transactions that
take place. If a new user has to enter the network, then it has to validate with all
the other users in the network. The network is open to everyone; it is a peer-to-peer
network. Using a simple web application, every end user can easily find out the
correct details of the medicine and find out whether the medicine is original or not.

3 Problem Formulation

The use of counterfeit medicine can be harmful to health. Their use may result in
treatment failure or death. They also undermine confidence in health service. Solving
the counterfeit drugs problem is important to ensure that the patients do not lose faith
in the benefits of pharmaceuticals and become non-adherent with their treatments.
Controlling the availability of counterfeit drugs is not easy, but it is necessary, given
the tremendous public health issues concerning counterfeit drugs, which can harm
or kill people.
Anti-Counterfeit on Medicine Detection … 1227

4 Existing System

Detecting a substandard drug is not an easy task. Some are entirely fake while some
are potentially dead (does not produce a desired result or ineffective). Consumption of
fake medicine can lead to consequences ranging from delayed treatment to dangerous
side effects. Consumers do not have testing equipment at home, but there are ways
by which counterfeit medicines can be identified and avoided such as checking the
packaging, checking the tablets/dosage form, physical attributes of tablets, verifying
medicines by online or SMS but by this way, information is not open to everyone,
only the user is aware of the transactions that take place and is not secure; so, we are
making use of blockchain technology which is the network that is open to everyone
and is secure.

5 Architectural Design

Counterfeit medicines are difficult to track and test. To detect the counterfeit drugs
before they reach the customers through the global supply chain, there must be enough
systems that can detect counterfeit medicines. To find out whether the medicine is
counterfeit or not, we need regulators to properly check its pharmaceutical ingredi-
ents. Regulators also analyze the packaging, the medicines serial number, and also
the inserts in order to determine the originality. Therefore, the process of finding
out the counterfeit medicines is a time-consuming and difficult task. Counterfeit
1228 R. Anand et al.

medicine results in major financial issue. In the case of medicines, the implications
are more severe because the products might not contain the right active ingredients
and therefore be useless to harmful. Using blockchain technology, we can reduce
the possibilities of counterfeit. This technology stops the entry of fake drugs into the
supply chain, mainly the part between the manufacturer and consumer.
The diagram illustrates the algorithm used by the counterfeit drug detection net-
work. When the manufacturing company manufactures a new medicine, they assign
a unique serial number to each new medicine. For that, a unique protocol asset called
hash is used. Then these medicines are added into the supply chain. Each medicine
is added as blocks to the chain. The samples are split into subset. To find out whether
the medicine is original or fake, we have to carry out package inspection where the
medicines ingredients are checked. From the added ingredients first, we identify
each ingredient. Then we check the amount of each ingredient added. From those
added ingredients, we see if there are any toxic substances such as chalk or flour, or
even toxic, made from rat poison or antifreeze. Or sometimes, a counterfeit medicine
does not contain the expected dose of whatever active ingredient they are supposed
to deliver, or they are expired. If any of these situations are found, then that medicine
is said to a fake one. If it is proved to be not fake, then the medicines are distributed
to the wholesale distributor and then to the users. Solving the counterfeit medicine
problem using blockchain technology is important to ensure that patients do not
lose faith in the benefits of pharmaceuticals and become non-adherent with their
treatments. The expansion of the internet has made the controlling of counterfeit
medicines very difficult.

6 Algorithm

• User enters the medicine details in the system.

• Admin responses to the user and sends the details or request to the network and
ensures authentication.
• In the blockchain, the details of particular medicine are entered from the
manufacturer while creating blocks.
• Only one block which has the same ingredient will response at a time.
• Each time the user’s crypto will generate for each transaction that is private key.
• The detail shared by the user is compared with the datasets already present.
• Admin will verify the details.
• Sends the notification to the user whether the medicine is fake or not.
Anti-Counterfeit on Medicine Detection … 1229

7 Implementation

7.1 Tablets

7.1.1 Tablet Name: Paracetamol

Ingredient: The ingredients are maize, starch, potassium sorbate, purified talc,
povidone, and soluble starch.
Manufactures.

7.1.2 Tablet Name: Vic Vicks

Ingredient:
Each uncoated tablet contains ingredients:
Paracetamol I.P. 500 mg,
Diphenhydramine HCl I.P. 25 mg,
Phenylephrine HCl I.P. 5 mg,
Caffeine (anhydrous) I.P. 30 mg.
Excipients q.s. Color: Lake Brilliant Blue FCF.
Dosage:
One tablet every 4–6 h does not exceed more than four doses in 24 h.
Caution:
Do not use with other Paracetamol products. Taking more daily dose of Parac-
etamol may cause serious liver damage or allergic reactions (e.g., swelling of face,
mouth, difficulty in breathing, itching, or rash). In the case of overdose, consult a
doctor eventually even if no symptoms appear.
Manufactures:
Product Description:
Vicks Vapo Rub medicated vapor begin to work fast to relieve our cough. Use on
chest and throat temporarily relieves cough. Also, great use as a topical analgesic to
relief minor aches and pains in our muscles and joints.
Manufacturer:
Soothing vapor prevents cough. Vapo Rub is applied on the chest or throat so
it does not interact with other medication the way pills are and does not to cause
drowsiness or side effects. Vapo Rub may be used for oral decongestant for people
who suffer from tension and may be suited for diabetics who are looking for a sugar-
free cough suppressant. Vicks Vapo Rub can also be rubbed on sore muscles to relieve
aches and pains.
1230 R. Anand et al.

7.2 Blockchain Client

Blockchain:
A system in which a record of transactions made in bitcoin maintained across
several computers that are linked in a peer-to-peer network.
The blockchain network has no central authority. It is an independent system. It
is a shared and immutable ledger so information inside it is open for everyone to see.
Hence, anything that is built on the blockchain is visible to everyone involved and is
accountable for their actions.

• Multiple nodes can be added to the blockchain.

• Proof of work (PoW).
• Simple conflict resolution between nodes.
• Transactions with RSA encryption.
• Blockchain client can be the user as well as manufactures.
• According to network, the user also receives the request and response as well as
manufactures.
• Both of them share their requirement in the network if they having the public key
of them means they can able to access the transaction and can respond to the user.
• It can be managed by the admin side.

7.3 Blockchain Admin

• History immutability.
• Un-hackability of the system.
• Data persistence.
• No single point of failure.

– Transactions can be managed using the admin or block miner.

– The admin will manage the transaction between the manufactures as well as users.
– User would not know the manufactures public key as well as manufacture also do
not know the users public key.
– The transaction can be done by the admin so that It will be secure one.
– Every “post” can also be called a “transaction” (it is all simply data at the end of
the day).
– The transactions are packed into blocks.
– A block can contain one or more transactions.
– The blocks containing the transactions are generated frequently and added into the
blockchain.
– Each block should have a unique id (because every block should be uniquely
identifiable).
Anti-Counterfeit on Medicine Detection … 1231

8 Implementation

9 Conclusion

In blockchain-based applications and experiments, faith on the longevity of

blockchain technology is increasing. Scalability and consensus algorithms are areas
of growing research in order to make blockchain more adaptable for businesses of
larger scale. Areas like taxation, education, and insurance are yet to see a major over-
haul via blockchain adoption and these can be the focus areas of future research in
blockchain. Acceptance of cryptocurrency by governments and establishment of reg-
ulations governing them are very important to ensure ethical use of cryptocurrency.
The public blockchains also provide an opportunity of mining interesting patterns of
cryptocurrency usage, user behavior, and monetary networks across the globe.

Bibliography

1. Barbora Hornáčková BC (2018) Using Blockchain smart contracts in the DEMO Methodology.
Czech Technical University in Prague Faculty of Information Technology, Jan 2018
2. Nakasumi M (2017) Information sharing for supply chain management based on block chain
technology. In: 2017 IEEE 19th conference on business informatics (CBI), Thessaloniki, pp
140–149
1232 R. Anand et al.

3. Cui G, Shi K, Qin Y, Liu L, Qi B, Li B (2017) Application of block chain in multi-level demand
response reliable mechanism. In: 2017 3rd international conference on information management
(ICIM), Chengdu, pp 337–341
4. Xu Y, Wu M, Lv Y, Zhai S (2017) Research on application of block chain in distributed energy
transaction. In: 2017 IEEE 3rd information technology and mechatronics engineering conference
(ITOEC), Chongqing, pp 957–960
5. Judmayer A, Stifter N, Krombholz K, Weippl E, Bertino E, Sandhu R (2017) Blocks and chains:
introduction to bitcoin, cryptocurrencies, and their consensus mechanisms. Morgan & Claypool
6. Ewen H, Mönch L, Ehm H, Ponsignon T, Fowler JW, Forstner L (2017) A Testbed for simulating
Semiconductor Supply Chains. IEEE Trans Semicond Manufact 30(3):293–305
7. Ro YK, Liker JK, Fixson SK (2007) Modularity as a strategy for supply chain coordination: the
case of U.S. Auto. IEEE Trans Eng Manage 54(1):172–189
Fault Detection in an Indoor Wireless
Sensor Network Using RSSI-Based
Machine Learning Technique

R. Pradheepa, M. Bhuvaneshwar, S. Ajay Kumar, B. Ajay Raj

and K. S. Anusha

Abstract WSN plays a significant role in various fields like in communication,

military, etc., and as it is being applied in various fields, the faults in it have to be
taken seriously. There were different approaches taken to detect these faults. But,
there were very less number of approaches of detection of faults through RSSI. There
can be many reasons for change in the measured value of a node. But, change in RSSI
can only be limited to distance, disturbance or due to the health of the node. Due to this
reason, we are motivated that detection of fault in WSN using RSSI can be reliable
than other methods in suitable scenarios. Also, a suitable machine learning technique
has been implemented to reduce human involvement in the detection process.

Keywords Wireless sensor networks (WSN) · RSSI · Neural networks

1 Introduction

Wireless sensor network (WSN) is defined as the network of self-governing sensors,

which are dispersed spatially across a region that is to be monitored, and that records
physical or environmental conditions and communicates the gathered information
through wireless links with each other are called wireless sensor networks. Wireless

R. Pradheepa · M. Bhuvaneshwar (B) · S. Ajay Kumar · B. Ajay Raj · K. S. Anusha

Department of Electronics and Communication Engineering, Amrita School of Engineering,
Amrita Vishwa Vidyapeetham, Coimbatore, India
e-mail: bhuvander@[Link]
R. Pradheepa
e-mail: pradheeparamasundaram@[Link]
S. Ajay Kumar
e-mail: ajayk0077@[Link]
B. Ajay Raj
e-mail: ajayraj98b@[Link]
K. S. Anusha
e-mail: ks_anusha@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1233

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1234 R. Pradheepa et al.

sensor networks (WSNs) have its application on various fields such as healthcare (less
invasive patient monitoring), environmental monitoring such as continuous monitor-
ing of air, water, soil, structural monitoring for buildings and bridges, animal farm
conditioning, forest fire detection, air pollution or radiation level detectors, traffic
congestion control, adaptive street light control and various military and wildlife
monitoring.
Since WSNs are mostly applied in hostile and harsh conditions, they are suscep-
tible to node failures and damage which can impact the efficiency of monitoring.
Some of the faults can be offset faults, battery faults, hardware faults, etc. [1].
RSSI is chosen as our fault detection parameter as it is directly related to the
battery level of the nodes.
Objective of paper:
• To detect faults in a wireless sensor network
• To apply machine learning technique to detect the faults
Reminder of the paper will be presented as follows: Sect. 2 will be related works
where summary of various papers is discussed. In Sect. 3, problem statement and
project methodology have been discussed where the algorithm has been explained
briefly. In Sect. 4, the final results have been discussed.

2 Related Works

In this paper [1], various kinds of faults and its error reduction techniques have
been discussed. Author is trying to emphasize the importance of error reduction
because increase in demand of WSN has made the system very accurate that even
error in small quantity is considered to be intolerable. Errors were classified on the
basis of its locality as data-centric and system-centric. The fault reduction has three
main approaches as follows: centralized approach, distributed approach and hybrid
approach.
In previous works on finding faults in WSN was done using a machine learning
tool called SVM [2]. Support vector machine (SVM) is considered to be the most
powerful machine learning tools used in many of practical applications. SVM method
is used to detect the sensor received data and to identify the faults. Kernel function has
excellent adaptation threshold for nonlinear classification in case of fault detection.
Initially, faults in WSN were discussed, and as fault detection kernel function was
used. SVM is a statistical approach where a nonlinear model was introduced in order
to reduce the faults. Here, humidity is taken as the parameter for detecting faults in
WSN.
RSSI is another type of parameter which can be used for detecting faults in WSN.
Through RSSI, battery faults can be detected as RSSI, and the module’s battery level
is interrelated [3]. This paper is based on the performance, design and implemen-
tation parameters in autonomous energy monitoring and cultivating irrigation sys-
tem through Internet. This prototype, due to its least cost and user-efficient interface
Fault Detection in an Indoor Wireless Sensor Network 1235

through Internet application, could be deployed by the farmers into practice very eas-
ily and effectively, and this is an introductory-level application without investing in
new technologies. This system is developed using tools like programming languages,
software and hardware due to which a cost efficient and user-friendly practices could
be made possible. This prototype by exploring through wireless sensor network may
give different benefits in terms of crop yield. This checking subsystem empowers
precise observation of each and every condition within the development, and this
web application would provide us a convenient model for crop protection and thus
by helping the farmers to face adverse situation.
Machine learning can be used to reduce human involvement in the fault detection
process [4]. The overall idea behind most of the machine learning techniques is that
a computer learns from training datasets to perform a task. Machine learning has
two methodologies that are supervised machine learning and unsupervised machine
learning. In supervised machine learning, the training dataset consists of desired
output of the task along with that data. Supervised machine learning includes a set
of classification algorithms/methods, which takes input in the form of dataset and
the classifications of each data set so that the computer can efficiently learn how
to identify and classify new data. In unsupervised machine learning, the training
dataset contains data but with no solutions, so the system creates a solution of its own.
Unsupervised machine learning includes clustering algorithm, which takes input as
a dataset that covers various part and converts it into clusters in order to process the
task in a effectively manner.
A statistical way of approach towards the detection of faults in wireless sensor
networks [5].
The proposed method in this paper learns the possible outcome effectively without
any training data set. So, this method can be used in the identification and classifi-
cation of data and system faults by considering the structural relationship between
two kinds of hidden Markov model (HMM). The main focus at present lies on the
effective calibration on data and system faults.
HMM is a statistical model in which the system being modelled is assumed to be
Markov processed with unknown parameter, and determining the hidden parameters
from the overall observable parameters. In an HMM, the hidden state is not directly
identified, but the state variables used in the detection are directly observable. Each
state has a probability distribution function over all the possible output observation.

3 Problem Statement and Methodology

WSN plays a significant role in various fields like in communication, military, etc.,
and as it is being applied in various fields, the faults in it have to be taken seriously.
There were different approaches taken to detect these faults. But, there were very
less number of approaches of detection of faults through RSSI. There can be many
reasons for change in the measured value of a node. But, change in RSSI can only
1236 R. Pradheepa et al.

be limited to distance, disturbance or due to the health of the node. RSSI values are
directly related to battery level of the node which makes detection of battery fault in
WSN simple.
As mentioned above, the process of detecting faulty nodes is found monitoring
the RSSI values of each node. Since there is a direct relationship between RSSI and
distance, we have designed a model with the help of this relationship. The topology
considered here is a mesh network as all the nodes are interconnected to each other
which helps us to detect the faulty nodes effectively.
The coordinates of each node are obtained from the data of Intel Berkeley WSN
[1]. It contains data for about 54 sensors that were deployed in Intel Berkeley Research
Lab. It contains the X and Y coordinates of 54 sensors places in an experimental setup
in the lab and also measured parameters like moisture, temperature, humidity and
light. Here, only the XY coordinate is used.
The XY coordinates from the excel sheet are imported into MATLAB, and they
are plotted in a graph. With the sensor coordinates, a distance matrix was developed
as follows (Fig. 1):
Distance between each node with each other node is obtained with this matrix.
Distance between two nodes is found by the equation:

d= (x2 − x1 )2 + (y2 − y1 )2 (1)

Friis equation gives us the direct relation between distance and RSSI,

Pt G t G r λ2
Pr = (2)
4π R 2
Since the obtained received power is in mW, we converted those mW values to
dBm using the following equation:
 
Pr
RSSI = 10 ∗ log (3)
1 mW

Distance Node 1 Node 2 Node 3

Node 1 0 Distance between Distance between

nodes 1 and 2 nodes 1 and 3

Node 2 Distance between 0 Distance between

nodes 2 and 1 nodes 2 and 3

Node 3 Distance between Distance between 0

nodes 3 and 1 nodes 3 and 2

Fig. 1 Matrix for node in mesh network

Fault Detection in an Indoor Wireless Sensor Network 1237

RSSI Node 1 Node 2 Node 3 Node 4

Node 1 inf

Node 2 inf

Node 3 inf

Fig. 2 Matrix for diagonal elements in mesh network

In our project, we considered the characteristics of an XBee module. From the

datasheet, it can be found that the maximum range of the XBee modules can range
up to 40 m. Hence, we found out the RSSI value for the maximum range to be −
90 dBm and hence can be said as the maximum RSSI value that can be available
in this sensor. Any value beyond which has to be faulty. Using this −90 dBm as
threshold for good functioning, the RSSI matrix has been analysed.
If anyone node (here node 3) loses its both transmission and receiving capability,
the RSSI matrix will be affected as follows (Fig. 2):
Here, in this table, rows are considered for reception, and column is for
transmission.
When a sensor node goes faulty, practically all the values in the highlighted area
might not exceed the threshold value. Similarly, in a good functioning node, not all
values in the highlighted area need to be within threshold due to some unpredictable
events. Hence, a node is said to be faulty only if all the node’s RSSI values exceed
the threshold value. A value of either 0 or 1 is assigned to each node depending on
if the corresponding node is faulty or not. ‘0’ is assigned to healthy nodes, and ‘1’
will be assigned to the faulty ones.
A supervised neural network is trained to detect the faulty nodes in a wireless
network. To train a neural network, one would need an input and output training
dataset. We have trained the machine learning model with an input dataset of 40,000
samples and an output dataset of 2000 samples (i.e. for every 2000 input samples,
one output sample is mapped to it). After training the dataset, we get the following
error percentage which will be displayed in the results section.
1238 R. Pradheepa et al.

Fig. 3 Error percentage after training the model using training dataset from neural networks

Fig. 4 Final output (faulty nodes are encircled with red circles)

4 Results and Conclusion

Even though this model gives us an accuracy of 98.8%, testing with hardware will
give us the concrete proof which can be done in future. Here, free space is considered,
and for future scope, obstruction can be considered (see Figs. 3 and 4).

References

1. Muhammed T, Shaikh RA (2017) An analysis of fault detection strategies in wireless sensor

networks. J Netw Comput Appl 78:267–287
2. Zidi S, Moulahi T, Alaya B (2018) Fault detection in wireless sensor networks through SVM
classifier. IEEE Sens J 18(1):340–347
3. Georgios M, Christos G (2015) Web based monitoring and irrigation system with energy
autonomous wireless sensor network for precision agriculture. In De Ruyter B, Kameas A,
Chatzimisios P, Mavrommati I (eds) Ambient intelligence. AmI 2015. Lecture notes in computer
science, vol 9425. Springer, Cham
4. Louridas P, Ebert C (2016) Machine learning. IEEE Softw
5. Warriach EU, Tei K (2013) Fault detection in wireless sensor networks: a machine learn-
ing approach. In: 2013 IEEE 16th international conference on computational science and
engineering, Sydney, NSW, pp 758–765
Image Description Generation Using
Deep Learning

Neha Supe, Deepti Patil, Revathi Mahadevan, Tanvi Pandhre

and Bharti Joshi

Abstract Over the years, various attempts have been made to make social media
and devices more disabled friendly. Screen readers and other assisting technologies
have made it easier for people with visual disabilities like low vision and blindness
to interact with social media. However, images on the social network are still inac-
cessible to them. Developing the description of an image in natural language is an
upcoming problem which is at the intersection of disciplines of artificial intelligence,
computer vision, and natural language processing. The image captioning task, also
known as visual captioning, makes the technical chassis of many important applica-
tions of semantic visual search and visual questioning, photo and video sharing in
social media, visual intelligence in chatting robots, and aid for visually impaired peo-
ple to perceive surrounding objects and happenings. To describe an image effectively,
it involves the detection of objects in images, identifying scenes and attributes of the
objects. Then, these labels are used to construct semantically meaningful sentences
to generate paragraphs which describe the images. Our research makes an attempt
to carry out the task to produce a paragraph like a description of images.

Keywords YOLO · CNN · LSTM · Deep learning · Image description

N. Supe · D. Patil · R. Mahadevan (B) · T. Pandhre · B. Joshi

Ramrao Adik Institute of Technology, Navi Mumbai, Maharashtra 400706, India
e-mail: revathim15@[Link]
N. Supe
e-mail: neha_supe@[Link]
D. Patil
e-mail: deepti4597@[Link]
T. Pandhre
e-mail: tanvippandhre@[Link]
B. Joshi
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1239

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1240 N. Supe et al.

1 Introduction

The task of describing the contents of the image is a complex one. Screen readers
which are available now can only read the text on the screen of the device to the
user. A visually disabled user must rely on text read out aloud by the screen reader to
understand what the image is about. People with cognitive disabilities have difficulty
in interpreting images, but they can understand the text. Therefore, the proposed
system aims to tackle these challenges by generating a description of the image in
natural language.
Describing an image is an easy problem for a human who has a perfect vision, but
it is a difficult task for a machine as it involves both understanding the content of an
image and then translating the information that is generated into natural language.
The task of image description generation involves generating a multi-line paragraph
of text which will describe the objects and actions in the image. A great deal of
research is already available in this domain.
The image description framework can be divided into two modules logically, one
is an ‘image-based model’ [1] which extracts the features and nuances out of our
image, and the other is a ‘language-based model’ [1] which translates the features
and objects given by our image-based model to natural sentences. The steps involved
are as follows:
1. Scene Detection
2. Object Detection
3. Attribute Classification
4. Sentence Generation
The usual approach is to use CNN and RNN [2, 3]. The object detection and
attribute classification are done by CNN [2–4], whereas the RNN generated sentences
[5–7].
The proposed framework will be used in social media to describe the contents
of the images in the user’s feed. Our approach would enable the relationships and
attributes between the objects that are detected by the object detector and classifier be
described in detail. A multi-lined description is generated to describe the happenings
in the image in a detailed understandable manner.

2 Related Work

Johnson et al. [2] proposed model generates detailed captions by creating and local-
izing regions in images. They used fully convolutional localization network (FCLN),
recurrent neural network (RNN), and long short-term memory (LSTM). The archi-
tecture processes each image with a single, efficient forward pass and does not
need external regions proposals. This image can be trained with only one round of
optimization.
Image Description Generation Using Deep Learning 1241

In [3], a technique is used to adjust weights of various samples, and it uses a

two-stage optimization approach for missing concepts mining. This method detects
added semantic concepts, and high accuracy is obtained. The methodology employed
by them is RNN, CNN-RNN, Online Positive Recall and Missing Concepts Mining
(OPR-MCM). Semantic concepts are identified, and high accuracy is achieved.
In the research [4], InstaPIC-1.1M method tries to solve personalized image cap-
tioning problem using YFCC100M data sets. It produces a describing sentence for a
user image, computing for prior knowledge such as users active vocabulary or writing
pattern in users previous documents. The methodologies used are natural language
processing with convolutional neural networks (CNN). They have performed two
tasks: the hashtag prediction helps in predicting a series of hashtags for an image,
and the post-generation generates a natural sentence of normal words, emoji, and
hashtags.
In another paper [8], scene graphs are generated using the objects detected in an
image with their pairwise relationship predicted, while region captioning creates a
natural language description of the objects, their attributes, relations, and other con-
text information. The methodology used is Multi-level Scene Description Network
(MSDN) which depends on the convolutional layers of VGG-16 and an LSTM-based
language model to prepare natural sentences to interpret the region. In MSDN, given
an input image, a graph is dynamically built up to form the relationship among
regions with different linguistic meaning. The graph gives an innovative approach to
allineate features from different tasks.
In [5], VGG-16 is used for object detection and scene classification, and a hier-
archial recurrent neural network of two levels is used for sentence generation. The
output produced by this model is in a paragraph and is detailed compared to the
previous CNN models.
When we compare the existing projects with the proposed system, it is clear that
there has been extensive research done towards the improvement in efficiency of
generating image captions by combining different techniques and applying them on
different types of data sets.
The system proposed is inclined towards applying object detection, scene recogni-
tion, attribute extraction, and sentence generation using CNN and LSTM. Therefore,
we use YOLO [9] as an object detection algorithm as it is faster, accurate, and more
efficient than CNN. It is also real-time. Detailed paragraphs are generated using
two-level hierarchical RNN as it has shown to generate more descriptive sentences.

3 Proposed Methodology

3.1 Object Detection

You only look once (YOLO) model processes the image and predicts the class of
object and bounding boxes in a single evaluation. Initially, we give the image as
1242 N. Supe et al.

input to this model, it divides the image in S × S grid. Each grid predicts B bounding
boxes with the confidence score. The probability that the object detected belongs to
a particular class (Dog, cat, banana, car, etc.) is its confidence score. Each bounding
box being evaluated takes five elements under consideration: x, y, w, h, and a box
confidence score. We consider x and y as offset values, the height and width of the
bounding box are normalized with respect to those of the image, so the values for x,
y, w, and h are between 0 and 1.
The ideal size of images for YOLO is 416 × 416, but as it is made up of only
convolutional and pooling layers, it can resize images on the fly. YOLO divides the
image into the grid size like S × S, and each cell of the grid predicts five bounding
boxes with different aspect ratios and size. YOLO generates a tensor for every bound-
ing box, which tensor will give its location within the grid, the height and width of
the grid, with respect to that of the image, and the confidence score. It is possible
that every bounding box may not contain an object. If centre of an object falls in
the grid cell, then that cell is responsible for predicting that object. Bounding boxes
give confidence score, which reflects the confidence of model that the bounding box
contains an object and accuracy of its prediction.
Formula for the confidence score is Pr(Object) × IOUtruth pred . Confidence score is
zero for boxes containing no objects, else it is equal to Intersection over Union (IOU)
between prediction and ground truth.
Grid cells also detect the C class of the object using conditional class object prob-
abilities, Pr (Classi|Object). Only one class per grid cell is predicted irrespective
of the number of bounding boxes. At the time of testing, conditional class proba-
bilities are multiplied by individual box confidence prediction, Pr(Classi|Object) ×
pred = Pr(Classi) × IOUpred this value gives class-specific predic-
Pr(Object) × IOUtruth truth

tion for each bounding box. For finding the final bounding boxes, we need to do the
following two steps:
Remove the bounding boxes which have no object and also those that predict a
confidence score less than a threshold of 0.24.
From the bounding boxes which claim to have an object, using Non-Max Suppression
and Intersection over Union remove the redundancy of identifying the same object.

3.2 Scene Detection

Modern deep learning methods focus on using semantic context for smoothing or
regularizing the predicted label maps while ignoring another rich semantic context
available in the data sets. The data set uses class names of scenes in the images, and
image patches’ label map statistics for supervision signals for learning deep feature
representations. The data set images are categorized according to classes.
We create a two-level label hierarchy for each of the original classes by exploit-
ing semantic context, and CNN is fine-tuned with the proposed label hierarchies.
Image Description Generation Using Deep Learning 1243

The representations of features learned by the model are more descriptive and dis-
tinguishable and provide accurate classification into classes. On the other hand, the
label map statistics of image patches also provide crucial prior information on the
appearance of patches since they specify the spatial layout of the semantics in the
surrounding region.
During the training period, we adopt stochastic gradient descent (SGD) with mini-
batch sizes of 64, 64, and 10 to optimize the three CNN models, respectively The
learning rate is initialized as 10−3 and decreased by a factor of 10, and step size is
2 × 104 . During testing, the efficient pixel-wise forward-propagation algorithm was
adopted for the Clarifai and the OverFeat models.

3.3 Sentence Generation

It is a complex task to generate paragraphs for images which require the model to
understand fine-grained images and also develop long-term language reasoning. The
generated paragraphs of multiple sentences to describe images should be detailed and
also form coherent stories. The paragraph result for an image has to be informative
and complex linguistically compared to single sentence description.
For our method, we will use the data set which comprises paragraph annotations,
region-wise annotations, corresponding images. We will use data from MS COCO
[10] and Visual Genome [11] which have images annotated with a multi-line sentence
description.
In this paper, we are using long short-term memory (LSTM) network. It produces
one word at each time step based on previously generated word, hidden state, and
context vector.
The first level of LSTM: It is a single-layer LSTM. It will be given the pooled
region vector as input, and the result produced will be a sequence of vectors where
each vector will be used to generate one sentence each to form a paragraph. Each
vector will give the topic the sentence is about.
The second level of LSTM: The second level of LSTM is a two-layer LSTM. The
output of the first level of LSTM which is topic vectors, one for each sentence will
be the input for the second level of LSTM. These vectors are combined to form one
pooled vector and will be used to generate words of the sentences. The hidden state
of the previous LSTM layer will be used to predict the words of the sentence at every
time step. After an “end” token has been encountered, all the topics will be used to
predict words to form sentences and will form a paragraph.

4 Results

We evaluated our generated image descriptions with METEOR [5]. METEOR is

used for automatic metric evaluation which takes human judgement into account.
It is used to evaluate a machine generated translation by calculating a score based
1244 N. Supe et al.

on word to word matches comparison between the translation from model and a
reference translation by human. It produces good correlation with human judgement
at the sentence and segment level. The METEOR score evaluated for our model is
12.35.

5 Conclusion

The visually impaired is not able to access and interpret images. The existing tech-
nologies enable screen readers which do not interpret images itself. There is a lack
of systems which would generate such descriptions in a comprehensive way. There-
fore, the proposed system aims to tackle these challenges by generating a descriptive
caption of the image. The system would generate a natural language description of
an image in real time. Above research shows that the descriptions generated are
accurate and semantically correct. This output would be in the form of speech so
as to enable the visually impaired to comprehend the image. Some modifications or
possible upgradation like describing GIFs and videos can be pursued in future.

References

1. Kinghorn P, Zhang L, Shao L (2017) Deep learning based image description generation. In:
2017 IEEE
2. Johnson J, Karpathy A, Fei-Fei L (2016) DenseCap: fully convolutional localization networks
for dense captioning. In: IEEE conference on computer vision and pattern recognition
3. Zhang M, Yang Y, Zhang H, Ji Y, Shen HT, Chua T-S (2018) More is better: precise and detailed
image captioning using online positive recall and missing concepts mining. IEEE Trans Image
Process
4. Li Y, Ouyang W, Zhou B, Wang K, Wang X (2017) Scene graph generation from objects,
phrases and region captions. In: IEEE international conference on computer vision
5. Krause J, Johnson J, Krishna R, Fei-Fei L (2017) A hierarchical approach for generating
descriptive image paragraphs. In: IEEE conference on computer vision and pattern recognition
6. Gu J, Wang G, Cai J, Chen T (2017) An empirical study of language CNN for image captioning.
In: IEEE international conference on computer vision
7. Feng Y, Lapata M (2013) Automatic caption generation for news images. In: IEEE transactions
on pattern analysis and machine intelligence
8. Mao J, Huang J, Toshev A, Camburu O, Yuille A, Murphy K (2016) Generation and com-
prehension of unambiguous object descriptions. In: IEEE conference on computer vision and
pattern recognition
9. YOLO9000 Better, faster, stronger [Online]. Available [Link]
[Link]
10. Lin T-Y, Maire M, Belongie S, Hays J, Perona P, Ramanan D, Dollar P, Zitnick CL (2014)
Microsoft coco: common objects in context. In: ECCV
11. Krishna R, Zhu Y, Groth O, Johnson J, Hata K, Kravitz J, Chen S, Kalantidis Y, Li L-J,
Shamma DA, Bernstein M, Fei-Fei L (2016) Visual genome: connecting language and vision
using crowdsourced dense image annotations. arXiv preprint: arXiv:1602.07332
Multi-agent-Based Interference
Mitigation Technique in Wireless
Communication System

P. Aruna and A. George

Abstract The popularity of agents in artificial intelligence (AI) can be seen almost
everywhere in the form of effective decision making, predictive analysis, expert sys-
tems, and so on. In parallel, wireless communication systems in the form of wireless
local area networks (WLANs) and wireless personal area networks (WPANs) are
also providing tremendous support for data communication. Exploring and utilizing
the benefits of artificial intelligence and wireless communication systems together
in an environment are still in the nascent stage due to the divergent characteris-
tics of the respective domain. One of the major drawbacks in the existing wireless
communication system is interference in 2.4 GHz band. This work proposes a multi-
agent-based interference mitigation technique to mitigate the interference issues in
wireless communication system and aims to enhance the communication process in
terms of packet delivery ratio. This work is implemented as a pilot project in an edu-
cational institution for teaching students in wireless communication environment,
the result shows that technology has added new dimensions to teaching–learning
process, and performance of the wireless communication system is also increased in
terms of packet delivery ratio.

Keywords Artificial intelligence · Agents · Wireless local area network · Wireless

personal area network · Interference · 2.4 GHz

P. Aruna (B)
Department of Software Engineering, Periyar Maniammai Institute of Science and Technology,
Thanjavur, India
e-mail: [Link]@[Link]
A. George
Department of Mathematics, Periyar Maniammai Institute of Science and Technology, Thanjavur,
India
e-mail: amalanathangeorge@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1245

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1246 P. Aruna and A. George

1 Introduction

The emergence and advancement of communication technologies for flexibility

among applications, virtual communities, and Web services are increasing day by
day. This technological development accelerated a lot in the wireless communi-
cation devices, and these communication devices can be broadly classified into
wireless personal area network devices and wireless local area network devices.
Each device in WPAN and WLAN is having its own characteristics and restrictions.
When both WPAN and WLAN mobile communication devices coexist in a particular
environment, there arises interference due to the usage of unlicensed ISM band of
2.4 GHz.
In wireless communication, the frequency-hopping spread spectrum (FHSS) and
direct-sequence spread spectrum (DSSS) are used by WLAN and WLAN. Trans-
mission can be initiated [1] either by FHSS or DSSS. When the receiver detects
the unwanted signal in the received signal along with original signal or it receives
corrupted signal, then it is termed as interference or noise. When there is the best
performance between data transmission among Bluetooth and Wi-Fi devices, then
there is less interference [2]. If there is interference among the devices, then there
might be a need to resend data among the communications which leads to worst or
poor performance. Following are the foremost negative effects of interference.
• A decrease in the wireless range between devices;
• A decrease in data throughput;
• Intermittent or complete loss of the connection;
• Difficulty pairing between device’s discovery phase.
Some of the positive aspects of wireless communication such as distribution,
cooperation, decentralization, and mobility can be interrelated to the characteristics
of agents. In addition to these common attributes, agents offer more intelligence and
autonomy to act on its own in the environment. The main contributions of this work
are as follows: (1) use of agent technology in the domain of wireless communication,
(2) agents are used to mitigate the interference issues among the WPAN and WLAN
devices, and (3) use of agents to increase the throughput and packet delivery ratio.
This work is designed using Java Agent Development Framework [3].
The remainder of this paper is organized as follows. Section 2 provides the under-
lying motivations for this proposed work. Section 3 discusses related works on agent-
based technologies for wireless communication systems. Section 4 represents the
proposed work in detail. Section 5 represents the implementation of proposed work
with a case study to measure the performance of wireless communication systems,
and finally, Sect. 6 concludes the paper with a brief summary of the work and also
gives an outline for the future work.
Multi-agent-Based Interference Mitigation … 1247

2 Motivations

A multi-agent-based approach is presented in this proposed work to deal with inter-

ference in wireless communication systems. An agent technology solution is pro-
posed since agent characteristics have the ability to perceive the environment and act
accordingly. This will be used to enhance the performance of the wireless commu-
nication systems in terms of resource optimization. Concerning the choice and the
usage of agent technology to come up with new interference mitigation technique
for wireless communication system, it is simply admired by the dynamic character-
istics of agents such as mobility, interactivity, autonomy, and intelligence. Indeed,
the proposed work can be used to significantly enhance the performance of wireless
communication system and particularly the elimination of interferences, by moving
the data collected by the proposed agents in the wireless communication devices.

3 Related Work

It can be observed that researchers [4–6] have already proposed agent-based tech-
nologies for the wireless communication systems, but serves different purposes. Our
foremost objective is to design and develop a multi-agent-based system to support
wireless communication system in the 2.4 GHz band with low interference among
the devices.

4 Proposed Approach

This approach is based on the usage of multi-agents in the wireless communica-

tion environment. This work mainly used two kinds of dynamic topologies namely
network topology and agent topology. Network topology describes physical layout
and the relationship among the devices in the environment. Agent topology describes
logical organization of various agents in the environment. To generate network topol-
ogy and agent topology, this system utilizes various agents. The role of agents in this
approach is to detect the list of devices in the environment, to classify the devices
based on their characteristics, to perform the routing activities among the devices,
and to mitigate the interference among the devices in the environment.

4.1 Architecture

The multi-agent architecture is illustrated by Fig. 1. This work reveals that users along
with WPAN and WLAN devices are working in specific locations and it is monitored
by the environment agent. WLAN and WPAN agents, hold details information about
1248 P. Aruna and A. George

Fig. 1 Multi-agent
architecture for wireless
communication

the WLAN and WPAN devices in the environment. Routing agent: this agent holds
the routing information between the transmitter and receiver devices. This routing
agent supports both the unipath and multipath routing protocols [7] to disseminate
information to all the devices in the environment. When a source node has to send data
packets to a destination node, it searches its route table to find if the route already
exists. If no route exists, then the source node initiates a route discovery process
by broadcasting route request (RREQ) packet to all its neighbors. When a RREQ
reaches a mobile node and if it is not the destination node, it adds its ID to the next
node and forwards/multicasts to its neighbors and broadcasts a route reply (RREP)
packet to the sender along the reverse path in network. This continues until the RREQ
packet reaches the destination node or the TTL reaches a threshold value. If a path
is discovered, then the path is stored in the routing table along with a time stamp
value T. Frequent agent, spatial agent, and temporal agent play a vital role mitigating
interference among the WLAN and WPAN devices to ensure enhanced performance
of the wireless communication system. The role of frequent agent is to monitor the
frequency of communication between the transmitter and receiver, and priority is
given to the most frequent communication devices based on ranking mechanism.
The role of spatial agent is to keep Wi-Fi and Bluetooth devices logically separately
with certain distance along with their respective antennas. Insulating material can
also be used to separate the devices. Bluetooth/Wi-Fi spatial isolation is impossible
when both radios share a common transmission medium which may be an antenna,
transmitter, and receiver.
The role of temporal agent in the proposed architecture is to offer the Bluetooth
and Wi-Fi radios that are embedded in the same device but takes time slice between
them in transmitting or receiving information. It is one of the coexistence methods.
This kind of isolation can be implemented between Wi-Fi and Bluetooth with a
support of printed circuit board which informs all the connected devices other than
transmitting device to refrain from sending data during this time.
Multi-agent-Based Interference Mitigation … 1249

4.2 Proposed Interference Mitigation Algorithm

To overcome interference in the wireless communication system, this work proposes

the following FST (frequent spatial, temporal) algorithm mechanism in the agents.
Proposed Algorithm
1. Detecting list of devices accessing the system and its neighborhood with the
support of environment agent;
2. Monitoring each device performance using WPAN and WLAN agent;
3. Routing agent is deployed for each and every category of WPAN and WLAN
agent to route data among the devices and its uses frequent spatial and temporal
agent to overcome the interference with the devices and sharing of data.
4. Classifying the devices based on its category/signal strength using frequent,
space, and/or time.
5. Each agent is interrelated to overcome the isolation error, and cooperation is
achieved with the support of agent communication language.
The above proposed algorithm mitigates interference issues in the heterogeneous
wireless communication network.

5 Performance Measure

To evaluate the efficiency of the proposed multi-agent-based system for wireless

communication, the proposed work is implemented in JADE environment and the
data communication and transmission are selected from the learning management
system (LMS) of an educational institution. Various types of data from the LMS
such as audio, video, text, and animation are taken into account as packets, and the
performance of these packets is measured in terms of packet delivery ratio.
Packet delivery ratio is defined as the ratio of data delivered packets to the corre-
sponding destinations to those generated in terms of constant bit rate. Illustration of
the experiment setup is shown in Fig. 2.
The experiment was conducted with two cases with the support of same set of
twenty-five users and data packets from different locations: the first one with the
support of multi-agent-based wireless communication system exclusively for WPAN
devices, WLAN devices and with the coexistence of WLAN and WPAN devices and
the second one was without the support of multi-agents for WPAN devices, WLAN
devices and along with the coexistence of WLAN and WPAN devices. Tables 1,
2, 3 and Fig. 3 give the packet delivery ratio for the experiment setup with UDP
transmission. Tables 4, 5, 6 and Fig. 4 give the packet delivery ratio for the experiment
setup with TCP transmission. It can be observed that the multi-agent-based wireless
communication systems perform better in terms of packet delivery ratio for all the
WPAN communication, WLAN communication and for the coexisted WLAN and
WPAN communication in both UDP and TCP transmission.
1250 P. Aruna and A. George

Fig. 2 Illustration of the experimental setup

Table 1 Packet delivery ratio

Wireless personal area
for the experiment setup
network (WPAN)
WPAN devices for UDP
Wireless communication s:4306
without multi-agents r:2139
r/s Ratio:0.4967
f:0
Wireless communication with s:4336
multi-agents r:2639
r/s Ratio:0.6086
f:0
Multi-agent-Based Interference Mitigation … 1251

Table 2 Packet delivery ratio

Wireless local area network
for the experiment setup
(WLAN)
WLAN devices for UDP
Wireless communication s:4287
without multi-agents r:4282
r/s Ratio:0.9988
f:1737
Wireless communication with s:4387
multi-agents r:4385
r/s Ratio:0.9995
f:1737

Table 3 Packet delivery ratio

WPAN + WLAN
for the experiment setup
WPAN and WLAN devices Wireless communication without s:4311
for UDP multi-agents r:2445
r/s Ratio:0.5672
f:0
Wireless communication with s:4341
multi-agents r:2885
r/s Ratio:0.6672
f:0

Fig. 3 Packet delivery 1.2

ratio—UDP for the 1
experiment setup
0.8
S/R raƟo

0.6
Without MulƟagent
0.4
With MulƟagent
0.2

0
WPAN WLAN WPAN + WLAN
Wireless CommunicaƟon

Table 4 Packet delivery ratio

Wireless personal area
for the experiment setup
network (WPAN)
WPAN devices for TCP
Wireless communication s:20
without multi-agents r:16
r/s Ratio:0.8000
f:2
Wireless communication with s:20
multi-agents r:18
r/s Ratio:0.9000
f:4
1252 P. Aruna and A. George

Table 5 Packet delivery ratio

Wireless Local area network
for the experiment setup
(WLAN)
WLAN devices for TCP
Wireless communication s:20
without multi-agents r:19
r/s Ratio:0.9500
f:2
Wireless communication with s:20
multi-agents r:20
r/s Ratio:1.0000
f:2

Table 6 Packet delivery ratio

WPAN + WLAN
for the experiment setup
WPAN + WLAN devices for Wireless communication without s:20
TCP multi-agents r:14
r/s Ratio:0.7000
f:0
Wireless communication with s:20
multi-agents r:16
r/s Ratio:0.8000
f:2

Fig. 4 Packet delivery 1.2

ratio—TCP for the
1
experiment setup
0.8
S/R ratio

0.6 Without Multiagent

With Multiagent
0.4

0.2

0
WPAN WLAN WPAN + WLAN
Wireless Communication
Multi-agent-Based Interference Mitigation … 1253

6 Conclusion

The underlying interference issues between WPAN and WLAN devices in the wire-
less communication systems must adapt themselves to a dynamically changing envi-
ronment and devices; also, the performance should be increased in the wireless com-
munication in terms of throughput and packet delivery ratio. With these objectives,
this paper proposed multi-agent-based interference mitigation technique in wireless
communication system, and this study deploys the agent technology for the commu-
nication between devices and to mitigate the interference issues among the WPAN
and WLAN.
From the case study implementation, it is noted that the proposed model is having
technology agnostic, friendly to legacy technologies, extensible, scalable (over a
broad range of dimensions), dynamic and adaptive, resilient and trustworthy.

References

1. Qureshi A, Guttag J (2005) Horde: separating network striping policy from mechanism. In:
MobiSys
2. Barford P, Crovella M (1998) Generating representative web workloads for network and server
performance evaluation. In: SIGMETRICS
3. Bellifemine FL, Caire G, Greenwood D (2007) Developing multi-agent systems with JADE, vol
7. Wiley, Hoboken
4. Harrabi S, Chainbi W, Ghedira K (2013) A multi-agent approach for routing on vehicular ad-hoc
networks. Procedia Comput Sci 19:578–585
5. Cicirello V, Peysakhov M, Anderson G, Naik G, Tsang K, Regli W, Kam M (2004) Designing
dependable agent systems for mobile wireless networks. IEEE Intell Syst 19(5):39–45
6. RoyChoudhuri R, Bandyopadhyay S, Paul K (2000) Topology discovery in ad hoc wireless
networks using mobile agents. In: Mobile agents for telecommunication applications, pp 531–
542
7. Choudhury RR, Paul K, Bandyopadhyay S (2004) MARP: a multi-agent routing protocol for
mobile wireless ad hoc networks. Auton Agent Multi-Agent Syst 8(1):47–68
Implementing Machine Learning
on Edge Devices with Limited Working
Memory

A. Harish, Saksham Jhawar, B. S. Anisha and P. Ramakanth Kumar

Abstract The architecture is aimed at pushing the computing towards the edge.
When most of the computation occurs towards the edge device where the data is
generated, the processing becomes faster and more efficient. This improves the user’s
wait time and delivers results faster to the user. Machine learning techniques are
implemented in the edge devices. While one can process data at the sensor, what one
can do is limited by the processing power available on each IoT device. Data is at the
heart of an IoT architecture, and one needs to choose between immediacy and depth
of insight when processing that data. The more immediate the need for information,
the closer to the end devices your processing needs to be. We propose an architecture
to use machine learning algorithms in the limited memory of the edge device.

Keywords Edge device · Fog layer · Working memory · Thin client

1 Introduction

Edge computing is a technique that introduces a new computation layer between the
cloud and end-users. The motivation behind developing this layer is to overcome
the flaws present in cloud computing. Cloud computing suffers various problems
ranging from processing of huge amounts of data (network congestion) to security
issues arising from physical distance between the cloud and its end-user. As the
physical distance increases, latency in data communication increases. To counter

A. Harish (B) · S. Jhawar · B. S. Anisha

Information Science and Engineering, R.V. College of Engineering, Bengaluru, India
e-mail: [Link].a@[Link]
S. Jhawar
e-mail: [Link]@[Link]
B. S. Anisha
e-mail: anisha@[Link]
P. Ramakanth Kumar
Department of Computer Science and Engineering, R.V. College of Engineering, Bengaluru, India
e-mail: ramakanthkp@[Link]
© Springer Nature Singapore Pte Ltd. 2020 1255
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1256 A. Harish et al.

these problems, edge computing devices store and process some amount of data
in the newly introduced layer. This layer is present physically closer to end-users,
lowering the latency problems.
Traditional edge computing machine learning models are based on the concept of
training classifiers present on a single edge device. However, this results in excessive
load on edge devices. To increase the efficiency and reliability of these devices,
another system that acts as a server can be used to train classifiers based on existing
dataset. The model parameters generated on the server can be sent to edge device to
assist with the decision-making process. This would significantly decrease the load
on edge devices and would enable them to take quick actions.

2 Related Work

A team comprising of individuals from the city of Antwerp, Belgium and the Flemish
region collaborated to develop large-scale living laboratory to test smart city appli-
cations [1]. The team worked with local government to put their anomaly detection
approach to test. The process included mounting air quality sensors on the vehicles
used by Bpost (postal service of Belgium). These vehicles covered almost every cor-
ner of the city while delivering goods and at the same time provided the team with
real-time air quality data. This in turn eliminated the need of using static sensors for
data collection. The objective of the test was to alert civilians about the air pollution
and observe huge amounts of organic compounds present in the atmosphere.
Conventionally, data from sensor is sent to the centralised cloud layer, where it
is processed. This approach suffers from latency flaws and network congestion. To
counter this, the team designed their system in such a way that most of the computa-
tion takes place at computation nodes, also called as fog resources. These nodes are
positioned in close proximity of the sensor to provide enhanced data transmission.
The team used BIRCH algorithm to cluster data samples. The parameters related to
air quality aberration were identified using robust covariance algorithm. As a result,
in the case of anomaly detection, computation nodes notified the centralised server
and IoT sensors.
To conclude, introducing fog resources enabled faster response time [2–6] and
showcased wide view of network behaviour [7–10].

3 Architecture and Implementation

To overcome the latency in transferring the data to server for adequate action and
the error rate in taking actions based on simple constraints, we propose a model that
inculcates the swiftness of edge computing and accuracy of machine learning models
trained by the resource-rich server.
Implementing Machine Learning on Edge Devices … 1257

The proposed model has two stages of operation that happens in different physical
locations on different devices. The division of the system into stages further eases
the process and implementation. The two stages are:

3.1 Training the Model in the Server/Cloud

We deal with a heavy server system in this model. The server system has resources
like RAM size and features like multithreading that compensate for the lighter client
edge node. The majority of the training occurs on the server where high processing
speed is available.
The data collected for the specific use case is used to train the machine learning
model. Different models such as neural networks, linear logistic regression and poly-
morphic logistic regression are trained and tested over the cross-validation dataset.
The model that provides the most accuracy and is the most suitable for the given use
case is chosen for the application.
The model is now trained using the dataset collected from various sources and is
used for further processing.

3.2 Computation/Decision-making on the Edge Device

The edge device is a very thin client. This paper concentrates on getting the edge
to work using a very small working memory. This is suitable for devices that are
embedded circuits and have limited working memory to function on.
Noting that the training of the machine learning model is already complete on
the server, the edge device inherits the parameters of the model from the server. The
edge device now collects the required and agreed upon data from the environment.
Following this, the edge device is ready to be put to use. The parameters obtained
from the server and the newly obtained environmental values are used to compute
the appropriate result for the given situation.
In a dynamic system with real-time data, the cycle of this process can be used to
obtain relevant results from time to time.
For the purpose of simplicity, we have used the theory logistic regression [11] with
a linear graph to discuss the model. However, we can use other more complicated
models such as neural networks, support vector machines, etc., and obtain results
accordingly without much variation in the model or the latency incurred.
Further, we provide details about the implementation of a fire detection system that
we implemented. The system uses logistic regression model to detect the presence
of fire with the usage of physical parameters such as temperature, relative humidity,
etc. (Figs. 1 and 2).
The machine learning model to be implemented is chosen considering the merits
and demerits of various models. The model is trained using the data previously
collected. Once the model achieves its required efficiency, it is ready to undertake
connection requests from edge devices.
1258 A. Harish et al.

Fig. 1 Training the machine

learning model in the server

Fig. 2 Computation using

machine learning at the edge

The edge device connects with the server and obtains the parameters of the
machine learning model that was trained. Now, the device is ready to function to
its needs. The edge device senses the external environment and computes the result
according to the machine learning system, thus giving better and faster results and
moving the analytics towards the edge device (Fig. 3).
The traffic towards the server is now reduced. The edge device performs a part of
the computing and thus reduces the load on the server.
Further, the model can be modified to send feedback to the server to improve the
performance of the machine learning model developed.

4 Experimentation and Results

The proposed model was implemented, and the application was put to test. We
implemented the server using a HP laptop with 8 GB of RAM. We chose a device
with light computational capabilities because it is easier and cheap to access and
program. The added advantage of cost was on our side.
Implementing Machine Learning on Edge Devices … 1259

Fig. 3 Architecture of the

complete system

The application we chose to implement was that of a fire detection model in a

normal household. We chose this application for two reasons.

4.1 Ease of Obtaining Data

The input data needed for the training of the machine learning model at the server
was easily available for different sources such as the [Link] site and our own data
sources from various locations across the city of Bengaluru.
Temperature and relative humidity were used in the initial trial runs. Other factors
such as atmospheric pressure, SO2 and NO2 levels of the air were used to experiment
with the model. The model gave similar results as that of the results specified here.

4.2 Ease of Implementation and Testing

The proposed model could be easily implemented for the application. We imple-
mented logistic regression [11] with the available data to classify the situations of
existence of fire and the non-existence of fire for the input read from the sensors.
We noticed that the proposed architecture includes the goodness of server training
with that of edge device decision-making. We compared the results of this hybrid
model with that of a simple constraint-based decision system. In this system, the
edge device made decisions based solely on the fixed constraints of the physical
environment. For example, say temperature >35 °C and relative humidity greater
than 50%.
1260 A. Harish et al.

Table 1 The test results

Type of model Type of input
Positive (112 test Negative (112 test
samples) samples)
Simple constraints 101 105
Proposed 111 112
architecture

The results were in favour of the proposed architecture.

The positive samples represent the test scenarios where there was an existence of
fire, whilst the negative samples represent the absence of fire.
From Table 1, it is clear that the proposed model outperforms the simple constraint
model that has been used historically [12] for edge computing.

5 Conclusion

The concept of edge computing is still young, and the research to explore this domain
is evolving exponentially. In essence to the points discussed in this paper, edge com-
puting proposes a solution to solve problems related to network congestion and
latency by reducing the physical distance between centralised cloud and end-user
devices. As per the results obtained, introducing edge nodes between centralised
cloud and end-user devices has proven to be more efficient than the traditional sys-
tems. The proposed method which incorporates a machine learning model not only
improves the scalability but also increases adaptability and the ease of maintenance.
Processing raw data in edge nodes and passing on the relevant information to cloud
expedite the overall communication process.

References

1. Santos J, Leroux P, Wauters T, Volckaert B, De Turck F (2018) Anomaly detection for smart
city applications over 5G low power wide area networks. April, 2018
2. Shi W, Cao J, Zhang Q, Li Y, Xu L (2016) Edge computing: vision and challenges. October,
2016
3. Villari M, Fazio M, Dustdar S, Rana O, Rajan R (2016) Osmotic computing: a new paradigm
for edge/cloud integration. December, 2016
4. Yi S, Li C, Li Q (2015) A survey of fog computing: concepts, applications and issues. June,
2015
5. Ai Y, Peng M, Zhang K (2017) Edge computing technologies for internet of things: a primer.
July, 2017
6. Gao Y, Hu W, Ha K, Amos B et al (2015) Are cloudlets necessary? October, 2015
7. Mach P, Becvar Z (2017) Mobile edge computing: a survey on architecture and computation
offloading. September, 2017
Implementing Machine Learning on Edge Devices … 1261

8. Dinh HT, Lee C, Niyato D, Wang P (2011) A survey of mobile cloud computing: architecture,
applications, and approaches. October, 2011
9. Luan TH, Gao L, Li Z, Xiang Y, Wei G, Sun L (2015) Fog computing: focusing on mobile
users at the edge. February, 2015
10. Tong L, Li Y, Gao W (2016) A hierarchical edge cloud architecture for mobile computing.
April, 2016
11. Peng J, Lee KL, Ingersoll GM (2002) An introduction to logistic regression analysis and
reporting. September, 2002
12. Sen K, Sarkar J, Saha S, Roy A, Dey D, Baitalik S, Nandi CS (2015) Automatic fire detection
and controlling system. May, 2015
Smart Transportation: An Edge-Cloud
Hybrid Computing Perspective

Aashish Jaisimha, Salman Khan, B. S. Anisha and P. Ramakanth Kumar

Abstract Internet of things is enhancing various industries by connecting all devices

to the Internet to solve key challenges. Transportation is a huge sector where exten-
sive research and development is undergoing. The current developments by various
independent corporate research companies use the power of cloud computing for
storage and processing of data. With the need for quicker decision-making in critical
services, a new paradigm, called edge computing is under research, where latency is
tackled. Since, each has its own limitations, a new approach is given that involves both
these paradigms—called hybrid computing. We would like to discuss the merits and
demerits of each of these paradigms and propose two variants of hybrid computing
with respect to smart transportation and smart tires in specific.

Keywords Internet of things · Cloud computing · Edge computing · Edge

analytics · Hybrid computing · Centralized architecture · Distributed architecture ·
Embedded systems-on-a-chip

1 Introduction

Transportation, being an ever-challenged sector due to the rapid increase in its cus-
tomer base poses lots of technological challenges. The need to design smarter solu-
tions is now more than ever. Of the total sector, the major percentage is accounted
by road transport. Tires are the only points of contact between the vehicle surface

A. Jaisimha (B) · S. Khan · B. S. Anisha

Information Science and Engineering, R.V. College of Engineering, Bengaluru, India
e-mail: aashishcool99@[Link]
S. Khan
e-mail: salmank123417@[Link]
B. S. Anisha
e-mail: anisha@[Link]
P. Ramakanth Kumar
Department of Computer Science and Engineering, R.V. College of Engineering, Bengaluru, India
e-mail: ramakanthkp@[Link]
© Springer Nature Singapore Pte Ltd. 2020 1263
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1264 A. Jaisimha et al.

and the road. With every vehicle, there are multiple tires associated which need to
function in a synchronized manner for the vehicle to run smoothly. Thus, the success
of road transport is largely dependent on the tires and the lack of maintenance is a
primary reason for failure of the system, and hence contributes to major accidents.
Yet, it is one of the least explored domains under transportation. Tire bursts are com-
monly caused due to improper inflation, tire-aging, overloading, low quality of tires,
and over-speeding.
Internet of things (IoT) connectivity is redefining today’s world in a vast spectrum
of applications such as smart-watches, smartphones and every possible appliance to
build smart-homes and smart-cities. Thus, by incorporating IoT connectivity and
using the power of sensors and data, smart tires can be developed to improve road
safety, tire life, and fuel efficiency.
Multiple paradigms can be approached for the storage and processing of data.
The most common in use is the cloud computing paradigm, which is the centralized
architecture. The newer and the fast advancing paradigm is the edge computing
paradigm, which is the distributed architecture. Most of the current development in
smart tires are using cloud computing. It has not been explored much in the direction
of edge computing. We discuss how the mixture of both of these paradigms can be
useful. The goal is to bring the right balance between performance and speed.
Organization: The remaining research paper is structured as follows: In Sect. 2,
we discuss the existing technology and the developments made in the field using IoT
connectivity. In Sect. 3, we discuss and differentiate the various data computation
paradigms. In Sect. 4, we propose two variants of hybrid architecture derived from
both edge and cloud computing. Section 5 poses the key challenges in implementing
our proposed architecture. We conclude the paper in the final section.

2 Existing Technology

2.1 Conventional Identification Technique

In the conventional days, there used to be a ‘retread’ mark on the tires. The wearing
up of that particular mark used to be an indication to either change the tire or subject
it to retreading process. This was a naïve approach to observe the normal aging of
the tire and did not consider the other aspects leading to tire bursts (Fig. 1).
Another measure that helps to determine the aging of the tire is minimum legal
tread depth. Tread wear indicators in the groove must not be let to align with the tire
[1, 2].
Smart Transportation: An Edge-Cloud Hybrid Computing Perspective 1265

Fig. 1 Conventional identification of tire wearability

2.2 IoT Connectivity

Based on the requirements, various kinds of sensors have been deployed into the
tires, most of which are to serve the common purpose of safety. Inflation pressure is
an important quantity that is a good warning for most abnormalities in the tire. The
pressure sensor is responsible for indicating when the internal pressure decreases
or increases than a certain threshold. This data is useful to notify the owner to get
the tires inflated. There are multiple contributing factors to changes in pressure, viz.
temperature, speed, load, etc. Thus, the monitoring of tire parameters can be done
by collecting the data. Another common observation that can be made on highways
is the burning of tires. This can be detected either by the temperature sensor or the
odor sensor (since tire burns cause a typical rubber burning odor). A transportation
technology company, continental has developed its product called ContiConnect for
digital tire monitoring [3, 4].
In the present scenario, the data that is collected is transferred onto a server or
a cloud for processing. The traditional cloud architecture for processing of data
has its own limitations, such as high network traffic, data management cost, and
most importantly latency. In life-critical scenarios, latency plays a huge role and
near-instantaneous decisions are of utmost importance. For example, tire design and
purpose would be a lot different in the case of autonomous vehicles. The decision
that needs to be taken would be based on more complex scenarios, such as driver
behavior, pothole detection, and automatic breaking. Hence, pushing all data onto
the cloud for processing will consume a lot of time and does not serve the service
concerned.
1266 A. Jaisimha et al.

3 Computation Approaches

3.1 Cloud Computing

In a cloud computing paradigm [5, 6], the data is sent from the tires, which is our
edge node to the server for processing. The need for cloud is justified by the low
processing power and the storage capability of the tire as the edge node. However, it
causes issues such as network congestion and latency when there are continuous data
and low response time required. In the process, availability and decision-taking time
are compromised. Thus, in real-time delay-sensitive applications, where a decision
has to be taken extremely quickly, it is not feasible to send the data to the server,
provided the data can be processed locally (Fig. 2).

3.2 Edge Computing

When cloud computing was introduced, the end-devices had limited computing
capacity and the devices were only capable of data collection. But with the rapid
advancements in embedded systems-on-a-chip, it is now possible to perform com-
plex computations on small end-devices. The edge computing paradigm proposes
the processing of the data to be done locally. Therefore, it solves the issue of latency
posed in the cloud architecture, where the processing can be done in the tire, which
is the edge node itself. By this, the decision can be taken quickly. Another impor-
tant advantage of edge computing over cloud computing in the transportation space

Fig. 2 Cloud interaction with tire and vehicle. Source Pirelli

Smart Transportation: An Edge-Cloud Hybrid Computing Perspective 1267

would be the lack of need of network connection to work. In many situations, such
as mountainous terrain, forest area, or rural area for that matter, the network con-
nectivity cannot be expected to be flawless. Thus, in low bandwidth conditions, the
edge computing is the appropriate approach.
Further, distributed clouds in the local server called a cloudlet can be used for
processing of data with each such cloudlet being a part of the vehicle software [6,
7]. The research has also been in the direction of introducing more layers in between
the edge and the cloud where computation can be distributed, giving rise to fog
computing [8].
Edge Analytics
The primary reason for data to be sent to the cloud in IoT is to gather insights from
the data; take a decision and strike an action. With rapid advancements in embedded
systems-on-a-chip (SoC) has resulted in powerful and resourceful devices that are
capable enough to run a full-fledged operating system or a complex algorithm. Edge
analytics [9, 10] can be used for the computational analysis of the data gathered
locally on the edge device, i.e., the tire. The time taken is drastically improved for
taking data-driven decisions at real time. Edge analytics involves developing the
analytics model, deploying and executing it on the edge. The complexity of analytics
model is largely dependent on the processing power and the storage capacity of the
device in the tire.

4 Hybrid Computing

With cloud computing and edge computing paradigms, both having its own advan-
tages and disadvantages, it is best to bring about the right balance. Thus, both the
paradigms can be incorporated based on the type of data and the service required
with the data. We would like to propose two variations of hybrid architecture [6].

4.1 Non-interleaving Edge-Cloud Architecture

In this variation, the services direct the computation approach which is dependent
on the criticality of the said tire services. A highly critical service would require low
latency and high response time and would therefore process the data at the edge.
The criticality of the service should be decided by the developer or manufacturer.
Therefore, for every service, the analytical model must be designed at its target
computation facility. For example, to determine the wearing of the tires, the cloud
computing approach is more suitable, whereas, for auto-braking system, edge com-
puting is more suitable. The primary goal would be to classify what data needs to be
processed in the cloud and what data must be processed in situ (Table 1).
1268 A. Jaisimha et al.

Table 1 Service and their

Service Criticality Paradigm
appropriate computation
paradigm based on criticality Inflation detection Medium Cloud
Automatic braking High Edge
Life-cycle assessment Low Cloud
Tire burst detection High Edge
Hump alert High Edge
RFID vehicle identification Medium Cloud

Figure 3 shows hybrid non-interleaving edge-cloud architecture. It is important

to know when the data needs to be processed locally, which is dependent on the type
of service being demanded.

Fig. 3 Decision flowchart for non-interleaving edge-cloud architecture

Smart Transportation: An Edge-Cloud Hybrid Computing Perspective 1269

4.2 Interleaving Edge-Cloud Architecture

The first variation discussed above affiliates one paradigm approach to a service of
an application and the computation is carried out. But, for a service, to improve the
performance, the edge-cloud paradigms can both be applied, provided the commu-
nication between the edge and the cloud is minimized efficiently. For improving the
data-driven decisions taken by the edge device, machine learning could be applied
which improves the accuracy drastically. But, most machine learning algorithms are
computationally intensive and cannot be done at the edge without the aid of external
modules. This problem can be solved by the interleaving edge-cloud architecture
that we propose.
In this architecture, edge and cloud act as two different modules interacting with
each other. Training being the most computationally intensive task of machine learn-
ing, the model can be trained at the server with predefined data for the service. The
trained parameters could be passed onto the edge device. The edge device would
be responsible for data collection and taking decisions using the gathered data and
the parameters obtained from the server. In order to make the model more robust
and the service dynamic, the edge device would also have the role of calculating the
variance of incoming real-time data. If there is considerable fluctuation in the data
detected, some historical data can be passed onto the server for retraining the model.
This provides for updating of the parameters that can be sent back to the edge.
This architecture distributes the tasks between the edge and the server, with
computation-intensive tasks done at the server end and lighter tasks done at the
edge, thereby reducing the workload on the server. The interactions between edge
and server are not completely cut out, but are minimized by a large extent. Thus,
network traffic is reduced and hence latency is reduced to the maximum extent. This
architecture could be expected to work really well for classification problems (Fig. 4).
Table 2 summarizes the difference between the two-hybrid models on various
distinguishing aspects.

Fig. 4 Interleaving edge-cloud architecture

1270 A. Jaisimha et al.

Table 2 Distinguishing aspects of the two-hybrid models proposed

Hybrid models Non-interleaving Interleaving
1. Data processing point Depends on the service Computation intensive—At
server
Lighter task—At the edge
2. Key features • Processing happens at a • Interaction between edge
single facility and cloud
• Depends on criticality of • Task distribution based on
service computation intensity
3. Low bandwidth Cloud: Availability is The system should
performance compromised completely rely on edge until
Edge: Works effectively network is restored

5 Key Challenges

The hybrid computing approach, though takes the aid of both the paradigms, has
issues related to integrating both the paradigms. This is because the technological
advancements in the two approaches are not proportionate. Cloud computing has
been researched for a long time and a lot of progress has been made, whereas edge
computing is relatively new and still under exploration. For the best results from
the hybrid approach, substantial advancements must be done in the areas of edge
computing.
Further, in the interleaving edge-cloud hybrid model, since there is communica-
tion between the edge and the cloud, network connectivity affects the system. The
communication will be disrupted under low bandwidth conditions and the processing
can take place only on the edge.

6 Conclusion

The transportation industry is largely dependent on the tire manufacturing sectors.

The progress toward intelligent tires gives a total new dimension for smart trans-
portation. With autonomous transportation under huge research and development,
the prime focus should be on to reduce the latency in decision-making and commu-
nication. A pure cloud computing or a pure edge computing approach, both have its
own merits and demerits. Thus, the hybrid approach discussed in the paper throws
light on exploiting the better of the existing two approaches and working toward the
synchronization of working between the two approaches.
Smart Transportation: An Edge-Cloud Hybrid Computing Perspective 1271

References

1. Tyre industry going IoT, [Link]

Partanen_RAIN_Tire_Industry_Going_IoT_PUBLIC_2017.pdf
2. Johnson S, Schmeitz A (2014) Study on some safety-related aspects of tyre use, 10–22, May
27, 2014. [Link]
discussion_document.pdf
3. ContiConnect. Digital tyre monitoring, anytime, anywhere, [Link]
com/transport/tyre-monitoring/conticonnect
4. Reinventing the wheel: why smart tyres give the IoT real grip, [Link]
reinventing-the-wheel-smart-tyres-and-the-iot
5. Atlam HF, Alenezi A, Alharthi A, Walters RJ, Wills GB (2017) Integration of cloud computing
with internet of things: challenges and open issues. In: 2017 IEEE international conference
on internet of things (iThings) and IEEE green computing and communications (GreenCom)
and IEEE cyber, physical and social computing (CPSCom) and IEEE smart data (SmartData),
Exeter, pp 670–675
6. Cloud, edge or hybrid IoT solutions: which computing model is best suited for your IoT appli-
cation?, [Link]
which-computing-model-is-best-suited-for-your-iot-application
7. Hassan N, Gillani S, Ahmed E, Yaqoob I, Imran M (2018) The role of edge computing in
internet of things. In: IEEE communications magazine, vol 56, no 11, pp 110–115, Nov 2018
8. Bermbach D et al (2018) A research perspective on fog computing. In: Braubach L et al (eds)
Service-oriented computing—ICSOC 2017 workshops. Lecture notes in computer science,
vol 10797. Springer, Cham
9. A guide to edge IoT analytics, [Link]
analytics/
10. Moor insights and strategy: IOT analytics at the edge, Aug 2016. [Link]
[Link]/wp-content/uploads/2016/08/IoT-Analytics-at-the-Edge-by-Moor-
[Link]
A Low-Cost Ideal Fish Farm Using IoT:
In the Context of Bangladesh
Aquaculture System

Md. Kalim Amzad Chy, Abdul Kadar Muhammad Masum,

Mohammad Emdad Hossain, Md. Golam Rabiul Alam,
Shahidul Islam Khan and Mohammed Shamsul Alam

Abstract This paper presents an IoT-based low-cost aquaculture system for auto-
matic monitoring of fish firms to increase the production of fishes in order to meet
the protein demand. A microcontroller is employed in measuring several essential
parameters of the water like pH, temperature, water level, oil layer, and to monitor
the fish behavior to understand their hunger level that is directly affect in the growth
of firm fishes. A mobile application and an interactive Web interface are designed
to notify the measured parameters value and necessary recommendation. Bluetooth
and ESP8266 Wi-Fi module are integrated with the system to deliver the data to the
mobile app and Web interface. In case of any abnormality, the system informs the
concerned authority to take the immediate steps. In addition, the forecasting of the
several water parameters in the following day to take an earlier preventive action
makes the proposed framework more exceptional. Therefore, the system will assist
the fish farmer to enhance the production of quality fishes, which may help to meet
the protein challenges for the large population.

Md. Kalim Amzad Chy · A. K. M. Masum (B) · S. I. Khan · M. S. Alam

Department of Computer Science and Engineering, International Islamic University Chittagong,
Chittagong, Bangladesh
e-mail: akmmasum@[Link]
Md. Kalim Amzad Chy
e-mail: [Link]@[Link]
S. I. Khan
e-mail: nayeemkh@[Link]
M. S. Alam
e-mail: alam_cse@[Link]
M. E. Hossain
Department of Business Administration, International Islamic University Chittagong, Chittagong,
Bangladesh
e-mail: mehapstat@[Link]
Md. Golam Rabiul Alam
Department of Computer Science and Engineering, BRAC University, Dhaka, Bangladesh
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1273

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1274 Md. Kalim Amzad Chy et al.

Keywords Smart fish farm · pH sensor · Oil layer detection · Support vector
regression (SVR) · Random forest regression

1 Introduction

Ocean is a great source of protein. So, we are increasing stress on the ocean that
creates natural imbalance. To meet the protein demand, alternate sources should
be explored. Aquaculture-based fish farming is one of the solutions to this crucial
problem. Fish farming is very cheap as only 1.5 lb feed is needed to produce 1 lb
fish while in case of meat from cow its 8–9 lb feed and 8000 L water [1]. In respect
to financial gain, a significant difference between Internet of things (IoT)-based fish
farm and non IoT-based fish farm has been found in earlier research [2].
The primary objective of this work is to develop a low-cost smart fish monitoring
and controlling system that can measure different important water characteristics
like pH, temperature, oil level, water level, and fish hungry level by observing the
fish behavior. Several low-cost sensors are equipped with the system to accomplish
the assigned task. To develop the system only $50–$55 is required. This amount is
such a low cost that a fish farmer can easily adapt.

2 Related Work

Nowadays, IoT makes a revolutionary change in diverse fields [3]. Consequently, IoT
improves the fish farming system that resolves a large portion of the nutrition problem
for more growing population. Conventional fish farming faces a lot of difficulties
in a change of the climate like heavy rainfall, extreme weather, limited water in
a farm, and high temperature. Food and Agriculture Organization (FAO) highly
recommends that modern technologies should be integrated with traditional farming
to overcome the challenges. SK telecom developed an IoT-based fish farm system
that observes pH, temperature, dissolved oxygen, and notifies the fish farmer in real
time through their smartphones [4]. To test the water quality, the conventional method
takes several samples from various locations and process in the laboratory. To smart
this method, Shareef et al. [5] presented an IoT-based decisions support system using
wireless sensor network (WSN) technologies that determine various physio-chemical
characteristics of water and inform the end user through smart mobile. Simbeye and
Yang [6] also developed a fish farm monitoring system to examine water quality based
on WSN technology. Their equipped sensors collect temperature, oxygen level, pH,
the water level of the fish farm, and transfer the data to the mother server via ZigBee
protocol. If any water parameter exceeds its threshold level than an auto-generated
message will be sent to the user. They utilized LabView software to show the gathered
information. But the high cost is the main drawback of this system.
A Low-Cost Ideal Fish Farm Using IoT … 1275

Besides, most of the work does not identify which sensors they have employed
or equipped high-cost sensors that make the system tough to cope up in the fish
farm [7]. Moreover, most of the system supply the fish feed after a certain period
ignoring feeding when they required. Prediction of the amount of next week fish feed
is missing in most of the system.
In the context of Bangladesh aquaculture system, lots of fisheries follow the
traditional methods to cultivate the fish. As a result, the cost becomes high, but the
output is not the expected level. To the best of our insight, this is the first attempt to
develop a low-cost fish farm monitoring and understanding when fish require feed
by analyzing their movement. Also, the forecasting features of the next week fish
farm parameters will help the fish farmer to take a preventive step in advance.

3 Methodology

The IoT-based fisheries monitoring and controlling system are presented in Fig. 1.
The figure depicts how the fisheries information will be gathered and transferred to
the end users of the system. The system estimates several parameters of the water
like pH, temperature, oil layer, water level, and the movement of the fish that directly
affect on the quick growth of a fish. Firstly, to measure the targeted water parameters
and fish movement, the system takes a few moments to initialize the equipped sensor.
As pH sensor does not give a proper reading at a static situation, it needs to shake
slightly to measure the pH of the water appropriately. The system uses a servo motor
to perform this task that rotates the pH sensor 30° in one direction and 30° in reverse
direction. For more accuracy and lower the amount of power consumption of the
system, it takes an average of five pH value where each value is read in every 3 min.
By using Eq. 1, this average value is calculated where P is the value of pH in every
3 min and pH is the desired result.

Fig. 1 Block diagram of IoT-based aquaculture system

1276 Md. Kalim Amzad Chy et al.
 5 

pH = Pi /5 (1)
i=1

IR optical sensor measures the density of the oil layer and stores the data in
memory. For pH value measuring, the value of the oil layer is calculated using Eq. 2,
where D is the value of the oil layer in every 3 min and Oil_layer is the ultimate
result.
 5 

Oil_layer = Di /5 (2)
i=1

Usually in a fishery, after a certain period, fish food is supplied to the fishes. The
system can determine either the fishes are extreme hungry or medium hungry or not
so that enough fish food is supplied. The abnormal movement, i.e., not a frequent
movement of the fish can help us to understand the fish hunger level. Three ultrasonic
sensors (a, b, c) are deployed in this purpose where each ultrasonic sensor takes six
values after every 10 s. For each sensor, the average value is calculated using Eqs. 3,
4, and 5 where a is a location of a fish measured by sensor a and b and c are like
a. Then, RMS value is calculated for each sensor using Eqs. 6, 7, and 8. a_rms is
measured RMS value by sensor a and b_rms and c_rms are like a_rms. From these
three RMS value of the sensor, we can determine the hungry level easily. If three
RMS value exceeds the threshold value, then the hungry level is extreme, if two
values exceed then medium hungry level, if one value exceeds then normal hungry
level.

a = [a1 , a2 , . . . a6 ] b = [b1 , b2 , . . . b6 ] c = [c1 , c2 , . . . c6 ]

 

6
a_avg = ai /6 (3)
i=1
 

6
b_avg = bi /6 (4)
i=1
 

6
c_avg = ci /6 (5)
i=1

a_rms = (a1 − a_avg)2 + · · · + (a6 − a_avg)2 (6)

b_rms = (b1 − b_avg)2 + · · · + (b6 − b_avg)2 (7)

c_rms = (c1 − c_avg)2 + · · · + (c6 − c_avg)2 (8)
A Low-Cost Ideal Fish Farm Using IoT … 1277

Too hot and too cold temperature is a great reason of the slow growth of a fish.
The system gets the current temperature of fisheries through temperature sensor. In
similar to pH and oil_layer value measuring, the system measures the temperature
of the fishery using Eq. 9 where T is the temperature in every 3 min and temperature
is the estimated value.
 5 

Temperature = Ti /5 (9)
i=1

Similarly, too high and low water level is not suitable for the fishes. A liquid level
sensor can measure the water level but costly. To make the system more cost-effective,
a low-cost ultrasonic sensor is equipped to measure the water level. In Eq. 10, W is
the water level in every 3 min and water_level is the outcome of the right-hand side
calculations
 5 

Water_level = Wi /5 (10)
i=1

Through two communication medium, the microcontroller sends the gathered data
to the end user. A mobile application and Web interface are integrated with the system
to operate it easily. Bluetooth module sends the data in a mobile app so that local
fishery supervisor can easily gain the necessary information. ESP8266 Wi-Fi module
sends the data to the Web interface so that one can monitor the fishery remotely. In
the Web interface, the data is plotted graphically and stored in the database that will
help the supervisor of the fishery to check the previous data when necessary.

3.1 Circuit Design and Softwares

The proposed system is developed by following the designed circuit. The ground and
VCC pin of every sensor are connected to the ground and 5 V pin of Arduino Uno.
The pH sensor relates to the analog pin of Arduino. From the analog output of the
pH sensor, the value of pH is calculated using the formula of pH calculation. Servo
motor, buzzer, temperature sensor, IR optical sensor, ultrasonic sensor, Bluetooth
module, and Wi-Fi module is connected to the Arduino UNO microcontroller board
according to the developed circuit diagram. Arduino IDE and language is used to
write the code and burnt into the system that is easy to understand and reliable.
1278 Md. Kalim Amzad Chy et al.

3.2 Algorithm

The entire flowchart of an IoT-based ideal fishery system appears in Fig. 2. Primarily,
all sensor values are initialized. To measure the pH value, servo motor rotates 30°
left and 30° right as pH sensor needs to shake for proper value. By using pH formula,
pH is determined 5 times in every 3 min. Average of the pH value is calculated
for 15 min duration. As optimal pH value for speedy growth of fishes is 7–8.5, the
system turns on the alarm if it exceeds this range [8]. IR optical sensor engages to
measure the oil layer of the water. By removing the heat and noise value form the
entire value received from IR optical sensor, the system gets the actual value. If the
oil layer value less than 500 than pH sensor notifies the user and turns on the alarm.
To estimate the fish moving rate, every 10 s is taken in 1 min duration. 4.5 cm RMS
value of the sensor act as a threshold value to determine the hungry level of the fishes.
Average value of the water level and temperature within 15 min is calculated like the
pH sensor. [9] says that 22–27 °C is the optimum temperature for the rapid growth

Fig. 2 Working process of the proposed system

A Low-Cost Ideal Fish Farm Using IoT … 1279

of the fish. So, in case of exceeding this range, the system informs the owner of the
fishery through mobile app and Web interface. The flow of the data from the device
to the mobile app and Web is performed through the Bluetooth and Wi-Fi module.

3.3 Predicting the Following Day Fish Feed Consumption

The system can forecast the various essential parameters and notifies the user. For
simplicity, here, we have considered just forecasting of the probably required fish
feed in next day. The system can predict other parameters. We have recorded about
200-day data of fish feed and used the data to train our model. Dataset is split into
70% and 30% where 70% data is for training and 30% data is reserved for testing
the model. Through different kinds of machine learning regression algorithms such
as linear regression (LR), decision tree regression (DTR), polynomial regression
(PR), support vector regression (SVR), and random forest regression (RFR), we fit
the training data to prepare the model to forecast the amount of the fish feed in the
following day. After finishing the model training part with training data, the model
is examined by testing data. Then, the mean absolute error and accuracy level are
calculated to evaluate the model. From these results, we have found RFR gives better
outcomes than other regression models.

4 Experimental Results and Discussion

A prototype of the proposed system is shown in Fig. 3. The pH, temperature, oil
layer, water level, and fish hungry level are the outcomes of the system. The essential
parameters of the water are sent to the mobile app through Bluetooth module for
local user, and data is sent to the Web server through the ESP8266 Wi-Fi module for
the remote user. Also, data is saved to the cloud storage for further analytics.

Fig. 3 Photograph of the

prototype
1280 Md. Kalim Amzad Chy et al.

Fig. 4 a A sample of everything ok, b an example of fish hungry notification, c a sample of pH

and hungry level notification, d an example of oil layer notification

In Fig. 4, some sample outcomes of the system taken at different time and different
situation. In Fig. 4a, all parameter is ok while at the time of medium hungry level
the system notifies it successfully (Fig. 4b) and inform the user. In case of pH level
decreasing the system informs the user, shown in Fig. 4c while in the oil layer
exceeding the threshold value case the system successfully determines the situation
(Fig. 4d).
In the Web interface, the system sent the measured data and visualized it graph-
ically to make it more interactive with the user. In Fig. 5, the system visualizes the
pH data over time and store it into the server for further analytics. Like the pH data
visualization, the system also visualizes the temperature data (Fig. 6), oil layer data
(Fig. 7), and water level data (Fig. 8) graphically into the Web interface for the remote
user.
Some forecasted data are displayed in Table 1 where the first column “Date Test”
refers to the forecast of the power consumption of that date. “MonthDate” format
is used to represent a date where the last two digits refer to data and remainder
value to month. For example, “218” implies second-month 18th day, “717” implies
seventh–month 17th day, and so forth.
Accuracy level and mean absolute error are considered to judge the model that is
shown in Tables 2 and 3, respectively. Shifting the degree of PR analysis provides

Fig. 5 pH observation
A Low-Cost Ideal Fish Farm Using IoT … 1281

Fig. 6 Temperature observation

Fig. 7 Oil layer observation

Fig. 8 Water level observation

Table 1 Testing data

Date test Fish feed (g) LR PR SVR DTR RFR
113 10096.59 10101.15 10101.31 10294.05 10602.16 10524.65
218 10173.3 10176.01 10178.49 10294.79 10693.29 10418.38
303 10235.84 10236.61 10237.68 10294.44 10732.53 10593.94
515 10394.56 10387.76 10383.78 10295.23 10561.82 10582.49
717 10572.98 10531.77 10539.52 10295.58 10732.19 10689.75
1282 Md. Kalim Amzad Chy et al.

Table 2 Accuracy comparison

Accuracy LR PR SVR DTR RFR
Training set 85.45 85.49 87.49 92.09 95.75
Testing set 84.76 84.85 86.05 90.94 94.24

Table 3 Output observations

Evolution of the linear model LR PR SVR DTR RFR
Mean absolute error Training set 44.2489 44.2063 29.2189 26.3687 23.0747
Testing set 43.6042 43.2575 35.5301 30.2692 23.8518

a good result than LR. RFR gives a better result than DTR and other models. So,
it can be decided that the RFR-based model is more suitable to forecast the water
parameters.

5 Conclusion

The traditional fish cultivation method fails to meet the demand of protein because
of the lacking in real-time monitoring the adverse aqua environment of fishes. The
proposed framework is more cost-efficient and smarter than existing traditional and
smart systems. The design and development of low-cost autonomous and smart fish-
ery monitoring system while integrating state-of-the-art sensors and technologies to
determine ideal environment for aquaculture is our future research direction. The
integration of the proposed IoT-based smart fishery monitoring system assists fish
farming to increase the fish production.

Acknowledgements All author states that there is no conflict of interest. No animals (fishes) and
humans are not involved in this work.

References

1. Ruan J, Wang Y, Chan FTS, Hu X, Zhao M, Zhu F et al (2019) A life cycle framework of green
IoT-based agriculture and its finance, operation, and management issues. IEEE Commun Mag
57:90–96
2. Zhang Y, Hua J, Wang YB (2013) Application effect of aquaculture IOT system. In: Applied
mechanics and materials, pp 1395–1401
3. Vernandhes W, Salahuddin NS, Kowanda A, Sari SP (2017) Smart aquaponic with monitoring
and control system based on IoT. In: 2017 second international conference on informatics and
computing (ICIC), pp 1–6
4. Sharma R (2014) SK telecom starts pilot operation of IoT-based fish farm management system.
Available: [Link]
A Low-Cost Ideal Fish Farm Using IoT … 1283

5. Shareef Z, Reddy SRN, Delhi IGDTUW (2016) Design and development of IOT based
framework for aquaculture. In: National conference on product design (NCPD 2016)
6. Simbeye DS, Yang SF (2014) Water quality monitoring and control for aquaculture based on
wireless sensor networks. J Netw 9:840
7. Beardmore JA, Porte J (2003) Genetically modified organisms in aquaculture
8. Y. E. Corporation (2016) pH in fish farming. Available: [Link]
resources/application-notes/ph-in-fish-farming/
9. Hobbyist TF (2019) Temperature control. Available: [Link]
basics/[Link]
IoT-Based Smart Monitoring System
to Ensure Worksite Safety—A Context
of Garment Industry in Bangladesh

Abdul Kadar Muhammad Masum, Ahmed Shan-A-Alahi,

Abdullah Al Noman, Mohammad Nazim Uddin, Khairul Islam Azam
and Mohammed Golam Sarwar Rakib

Abstract In this paper, we have proposed and designed a hybrid system that is
able to detect fire breakout, gas leakage and noise pollution as well as providing
location of the affected area and opening fire extinguish system. Raspberry Pi is
integrated with MQ-5 sensor, humidity sensor, flame sensor, sound sensor and camera
module. A 360° servo motor is accumulated with camera module to capture affected
location even at any angle. To increase the reliability of this system, an authorized
person is assigned to assess the real situation. If fire is detected, camera module
takes a snapshot of affected region and sends it to admin’s email through 802.11n
LAN wireless module. Different sensors’ value also transmits to server through that
module in one-minute interval. Moreover, a buzzer is activated in control room when
data and picture is sent to admin. If admin confirms the incident, the system will
raise the alarm in whole workplace, uncover the water valve of affected region and
send message to the owner and nearby fire brigade. Thus, a garment can secure the
workplace for its workers.

Keywords Raspberry pi · Sound intensity · Fire detection · Garment industry

A. K. M. Masum (B) · A. Shan-A-Alahi · A. Al Noman · M. N. Uddin · K. I. Azam ·

M. G. S. Rakib
Department of Computer Science and Engineering, International Islamic University Chittagong,
Chittagong, Bangladesh
e-mail: akmmasum@[Link]
A. Al Noman
e-mail: alnomancse143@[Link]
M. N. Uddin
e-mail: nazimuddinasifiiuc@[Link]
K. I. Azam
e-mail: [Link]@[Link]
M. G. S. Rakib
e-mail: rakibsarwar3@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1285

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1286 A. K. M. Masum et al.

1 Introduction

In Bangladesh economy, ready-made garments (RMG) industry has acquired a sub-

stantial facet in comparison with any other sectors for foreign exchange earnings
and growth. According to the world trade organization (WTO), Bangladesh is the
second-largest apparel supplier that holds 6.5% market share in the world in 2017
[1]. In every year, nearly about $5 billion earned from this sector which is around
two-third of total exports. Around 2,500 garment factories are in Bangladesh, on
which ten million livelihoods are indirectly or directly depend on it [2]. This sector
also participates significant contribution in exporting which is almost 80% of total
export of country [3]. Unfortunately, nowadays, this sector is under threat because
of security risk of garment factories.
The best precedent is Tazreen Garment’s fire, which is the deadliest fire incident in
the country’s history. It was November 24, 2012, a tragic fire incident was experienced
by Tazreen Garment’s Factory, where 117 people were dead [4]. On December 14,
2010, in Ashulia, Dhaka, “That’s It Sportswear Ltd” garment factory was affected
by fire where at least 28 more people had died, and also huge number of workers
was grimly injured. Again at the “Garib and Garib” factory, situated at Dhaka, 21
peoples lost their lives in the deadly fire broke out [5]. In garment industry, the
pollution level is increased along with the advent of the technology. There are many
reasons for pollution: intense sound, temperature and humidity. It can cause hearing
problem, heat illness, fatigue, heatstroke and cold-related medical conditions. So, we
need to maintain a reasonable range of sound, temperature and humidity to control
the pollution level. Consequently, worker feels less comfort to work. Therefore, these
terrible incidents demonstrate that an early warning system is crying need to handle
the risky situation in the garment factory.
Our proposed system is assuring the early warning in a factory’s workplace. In this
system, we used an intelligent algorithm to decide when to warn for fire breakout and
gas leakage. If fire breakout and gas leakage are detected via flame and gas sensor,
then their value sends to the controller office with snapshot of affected region and
activates the buzzer in the office. The camera module is placed in the middle of the
workplace and positioned at zero degree. If an incident is occurred, the system will
start firing suppression system using uncovered water valve and also stop electricity
and gas supplies. At the same time, GSM module sends the SMS to the owner and
nearby fire service station for informing this incident. The system also collects the
noise level using sound sensor and temperature and humidity level using humidity
sensor to make the workplace comfortable for the workers. The main aim of this paper
is to develop IoT-based smart early warning system in workplace. To the best of our
knowledge, smart warning system in workplace is not implemented in Bangladesh
yet. Moreover, operating the proposed system is more comfortable than other existing
related systems.
IoT-Based Smart Monitoring System to Ensure Worksite Safety … 1287

2 Literature Review

The garment industry of Bangladesh is the largest manufacture sector which has
a great impact on the socio-economic development of Bangladesh. Unfortunately,
different types of sufferings in the workplace of factories are incredibly increasing
nowadays that badly effects on this sector. At least 1,601 workers died in garment
factories in last 2005–2016 where 280 workers in fire breakout, 1,221 in building
collapse and 100 in gas leakage [6]. So, a smart warning system is essential to save
valuable lives and growth of this sector.
Imteaj et al. [7] implemented a system for fire detection in workhouse using
Raspberry Pi 3. They used gas and light intensity sensor to sense any indicators of
fire and gas leakage. The system can extinguish fire breakout using water valves
and can notify by SMS with location. But, they did not work on noise level also did
not explain how to extinguish fire breakout. Moreover, Fuzi et al. [8] used Zigbee
wireless module to sense signs of fire. The system composed of temperature sensor,
Arduino Uno microcontroller, buzzer and operating software. To detect fire, they used
only temperature sensor. Similarly, Islam et al. [9] also used Zigbee and localization
technique to find the distance and position of fire which is cost expensive. Again,
Sowah et al. [10] developed a system to sense fire using fuzzy logic in vehicle. They
can put out fire breakout using air-conditioning system of vehicle.
For alerting an fire occurrence in any industrial premises, Sathishkumar [11]
worked on automated fire voice alert system. They used automatic voice recorder
to extinguish fire and GSM inside the system to send up-to-date information of
surrounding area to the company’s IP. The Yu et al. [12] proposed video processing
based on fire alarming system. To detect the possible fire breakout, they adopted
smoke color and its spreading features. But, detection of fire via video processing is
time consuming.
So, our motive is to design an intelligent and early warning system to detect the
fire breakout, gas leakage and noise pollution in the worksite which has to overcome
the defect observed in the earlier system.

3 System Description

3.1 System Architecture

The system we designed consists of IoT and wireless technology that equipped in a
room to monitor symptom of fire, gas leakage and noise pollution. We use Raspberry
Pi 3 as our main device and use a series of sensors and module which are flame
sensors, gas sensors, sound sensors, humidity sensors, servo motor, GSM module,
camera module, two-channel relay module and water valve. The flame sensor can
detect the fire in particular place. The existence of gas is detected by gas sensor. The
camera module is used for taking the snap of affected area with the help of servo
1288 A. K. M. Masum et al.

motor. The sound sensor is used to detect the noise level of industry. The output of
the flame sensor, gas sensor and sound sensor is a digital value. We use a converter
to get the analog value from flame, gas and sound sensor.
Data transmits from sensors to cloud server and picture sends to email through
802.11n LAN wireless module which is built in Raspberry Pi 3. When the certain
condition is true, then the relay module activates the alarm and uncovers the water
valve. At the same time, GSM module notifies owner and nearby fire station by
sending SMS after admin confirmation. To measure the level of noise pollution, we
use sound sensor. Humidity sensor is used for measuring temperature and humidity
of workplace. By these sensors, we can see the situation of pollution level by cloud.
In the flow chart, flame sensor is denoted as F, and Gas sensor is denoted as G.
To detect the fire breakout and gas leakage, we used four pairs of flame and gas
sensors and also used a camera module which is embedded with servo motor for
rotating 360°. Camera module is placed in the middle of the room. Initially, we set
the direction of the camera module at 0°. If the value of F1 and G1 is true, then the
camera rotates 45° angle with the help of servo motor and takes a snap of the first
quadrant. Similarly, it rotates 135°, 225° and 315° angles for second, third and fourth
quadrant, respectively, for taking a snap of these quadrant.

3.2 Blog Diagram and Flow Chart

After giving power supply, the Raspberry Pi 3 gets power, and the input–output
port gets ready as shown in Fig. 1. Flame sensor, gas sensor, sound sensor and
humidity sensor transmit data to Raspberry Pi 3, and camera module also sends snap
through it. Raspberry Pi 3 used it’s built in 802.11n LAN wireless module for sending
data and snapshot to the admin. When certain condition is satisfied, then buzzer is
activated, and GSM module is used for sending SMS to owner and fire service. The
implementation of our proposed system is followed by this flow chart which is shown
in Fig. 2, and circuit diagram is shown in Fig. 3.

3.3 System Implementation

The code for total device is written in Python. Python program checked every part
individually and set threshold value for flame sensor and MQ-5 sensor. If the threshold
exceeded, then Raspberry Pi Infrared Camera Module takes snap of this. A sim is
placed in GSM to communicate with administration and send location of specific
affected area.
Each sensor is connected to Raspberry Pi 3. Where, Flame and MQ-5 sensors
have four pins. We used ADS 1115 16-Bit I2C ADC module to get the analog value
from sensors. The signal pin of flame is connected with ADC module pin A1, and
MQ-5 is connected with ADC module pin A2. SCL and SDA pin of ADC module
IoT-Based Smart Monitoring System to Ensure Worksite Safety … 1289

Fig. 1 Block diagram of the proposed systems

Fig. 2 Flowchart of the designed system

is connected to GPIO2 and GPIO3 of Raspberry Pi 3. DHT11 and RKI-3103 sound

sensor has three pins. Where, signal pin of RKI-3103 sound sensor is connected
with ADC module pin A0. Signal pin of DHT11 is connected with GPIO18 pin of
Raspberry Pi 3. VCC and GND pins of all sensors are connected with 5 V and GND
of Raspberry Pi 3. The servo motor and relay module have digital pin. Signal pin
1290 A. K. M. Masum et al.

Fig. 3 Circuit diagram

of relay is directly connected with pin GPIO22 and GPIO10 of Raspberry Pi 3. The
signal pin of servo motor is connected with GPIO25 pin of Raspberry Pi 3. VCC and
GND pin is connected with 5 V and GND of Raspberry Pi 3 in both sensors. The
ribbon of camera module is connected to Raspberry Pi 3’s CSI camera port.
Then, we enable the camera module by raspi-config. A Python script is written
to take snap when flame and gas are detected. A servo motor is used to rotate the
camera. The servo motor and relay module have digital pin. Signal pin of relay is
directly connected with pin GPIO22 and GPIO10 of Raspberry Pi 3. The signal pin
of servo motor is connected with GPIO25 pin of Raspberry Pi 3. VCC and GND pin
is connected with 5 V and GND of Raspberry Pi 3 in both sensors. We use SIM800
GSM/GPRS to send SMS. Where RX pin of SIM800 is connected to GPIO14 TX
pin, and TX pin of SIM800 is connected to GPIO15 RX pin of Raspberry Pi 3. The
ground pin of SIM800 and Raspberry Pi 3 is connected to each other. In our system,
the Raspberry Pi 3 is connected with an external 5000 mAh power bank. A 2 A power
supply adapter is used to operate SIM800 GSM/GPRS.

4 Experiments and Results

Flame sensor, gas sensor, camera module, sound and humidity sensor based on a
smart warning system can measure symptom of fire breakout, gas leakage, pollution
of noise, humidity and temperature level of workplace. The camera module takes
the snap after satisfying required value of flame and gas sensor. A prototype of our
proposed system is shown in Fig. 4. Humidity, sound, flame and gas sensor send data
to the server through 802.11n LAN wireless module which has built in Raspberry
Pi. At the server, the gas and flame sensors’ data are presented graphically in Figs. 5
and 6. Representation of noise level and temperature level is shown in Figs. 7 and 8.
Camera module sends the snapshot of affected area through this module in admin’s
IoT-Based Smart Monitoring System to Ensure Worksite Safety … 1291

Fig. 4 Prototype of our

proposed system

Fig. 5 Gas sensor’s data

representation

Fig. 6 Flame sensor’s data

representation

Fig. 7 Representation of
noise level
1292 A. K. M. Masum et al.

Fig. 8 Temperature
representation

email as shown in Fig. 9. Controller office checked it and activated the buzzer alarm,
uncovered the water valve and shut off the power circuit if data is exceeded. The
system also sends SMS to owner and nearest fire service station as shown in Fig. 10.

Fig. 9 Email from

Raspberry Pi

Fig. 10 SMS alert

IoT-Based Smart Monitoring System to Ensure Worksite Safety … 1293

5 Conclusion

To provide the safety of workstation, traditional system is not sufficient and also has
many limitations. This paper designed a smart warning system to monitor worksta-
tion efficiently and effectively. By investigating sensor data, our system gives the
authentication of incident and warns at real time. Cloud server is used for presenting
sensor data graphically and storing purpose. This is a remarkable idea in context of
developing nations, particularly Bangladesh. The entire framework is cost effective
than existing system. If our designed system can be successfully implemented in
every factory, then it is expected that the damage of life and wealth because of the
fire and gas accidents and noise pollution will minimize remarkably, and the nation’s
economy won’t be faltered by such heartbreaking accidents. When data gets massive,
some features like Hadoop HDFS, MapReduce are initiated to handle big data. So,
our proposed system should integrate into every factory to change the current terrible
situation of the workstation of Bangladesh.

References

1. Bhuiyan MSA (2019) Sustainable economic growth: a challenge to embrace for Bangladesh.
Glob J Hum-Soc Sci Res 4(2):34–39
2. Islam S, Roman RI (2019) Assessment of fire hazard on the readymade garment industry in
Chittagong City, Bangladesh. Indonesian J Environ Manage Sustain 3(4):20–28
3. Rahman MA (2015) Impact of garments industry in Bangladesh economy, Aug 20, 2015.
Available: [Link]
4. Dhaka Bangladesh clothes factory fire kills more than 100, Nov 25, 2012. Available: https://
[Link]/news/world-asia-20482273
5. At least 28 more garment workers die in Bangladeshi factory fire, Apr 4, 2013.
Available: [Link]
in-bangladeshi-factory-fire
6. Alamgir F, Banerjee SB (2019) Contested compliance regimes in global production networks:
insights from the Bangladesh garment industry. Hum Relat 72(2):272–297
7. Imteaj A, Rahman T, Hossain MK, Alam MS, Rahat SA (2017) An IoT based fire alarming and
authentication system for workhouse using Raspberry Pi 3. In: 2017 international conference
on electrical, computer and communication engineering (ECCE), pp 899–904
8. Fuzi MFM, Ibrahim AF, Ismail MH, Ab Halim NS (2014) HOME FADS: a dedicated fire
alert detection system using ZigBee wireless network. In: 2014 IEEE 5th control and system
graduate research colloquium, pp 53–58
9. Islam T, Rahman HA, Syrus MA (2015) Fire detection system with indoor localization using
ZigBee based wireless sensor network. In: 2015 international conference on informatics,
electronics & vision (ICIEV), pp 1–6
10. Sowah R, Ampadu KO, Ofoli A, Koumadi K, Mills GA, Nortey J (2016) Design and imple-
mentation of a fire detection and control system for automobiles using fuzzy logic. In: 2016
IEEE industry applications society annual meeting, pp 1–8
11. Sathishkumar R (2016) Design and development of automatic fire detection using sms and
voice alert system. Int J Sci Eng Res 7:114–117
12. Yu L, Wang N, Meng X (2005) Real-time forest fire detection with wireless sensor networks.
In: Proceedings 2005 international conference on wireless communications, networking and
mobile computing, pp 1214–1217
Evading Gratuitous Energy
Consumption Due to Activation
of Superfluous Nodes in WSN

Alok Misra and Divakar Singh Yadav

Abstract In the recent period, wireless sensor networks are mostly used in diverse
applications of sensing that includes medical, armed forces, civil, adversity manage-
ment, environmental and commercial applications. Wireless sensor networks typi-
cally comprise a hefty amount of sensors. Sensors are device that produces a measur-
able response in changing the environmental conditions like temperature, humidity,
pressure, etc. As the sensor has the limited energy, to boost the duration of network
and maintaining coverage preservation, we necessitate an approach that involves
least sensors in communication of sensed data to base station. In this research work,
we amalgamate the conception of genetics and extended search to evade gratuitous
energy consumption which is due to activation of superfluous nodes. We aim to pre-
pare a schedule in which least number of sensors are stimulated and cover each point
of concerns.

Keywords Energy-efficiency · Full coverage preservation · Network life span

addition · Sensor scheduling

1 Introduction

Wireless sensor network consists of a assortment of sensor nodes and running on

limited magnitude of battery power. Each sensor node of WSN can sense, process
and transmit data to base station (BS). If all the packets are passed to BS straightfor-
wardly by sensor nodes, the nodes which are far away from BS will depart untimely.
Alternatively, among sensor nodes transmitting packets via a couple of hops, sensors
nodes which are in close proximity to the BS are inclined to depart untimely. Thus,
some areas of network become totally un-examined, and network partitions are cre-
ated. Lifetime of sensor nodes is required to be extended by minimum consumption

A. Misra (B)
Dr. A.P.J. Abdul Kalam Technical University, Lucknow, India
e-mail: alokalokmmm@[Link]
D. S. Yadav
Institute of Engineering and Technology, Lucknow, UP, India
© Springer Nature Singapore Pte Ltd. 2020 1295
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1296 A. Misra and D. S. Yadav

of power in transmission [1]. Thus, to curtail the power consumption, we aim to cover
each point of concern (POC) at least by one sensor. Typically, the superfluous sen-
sor nodes, which are chosen by scheduling policies [2–4], should be kept in sleeping
mode for energy preservation. As soon as the lively node loses its entire energy, there
is urgent need to wake up one or more sleeping nodes to reinstate that dying node. As
a consequence, the coverage control is assured, and original coverage is preserved
after switching off superfluous nodes. To discover the most optimal schedule, in
which we decide which sensor to stimulate and which sensor to be deactivated, we
exploit the conception of genetic algorithm.

1.1 Genetic Algorithm

In genetic algorithms, we have a set or a populace of feasible solutions to the given

hitch. These solutions are then subjected to recombination and mutation (as in natural
genetics). Fitness value is determined for each candidate solution, and the greater
suit persons are given a larger likelihood to breed superior “vigorous” individuals.
This is driven by theory of Darwinian that is “Survival of the fittest”. In this way, we
proceed to “evolve” with higher individuals or solutions over generations, until we
attain a criterion of detention. In the genetic algorithm, we repeat selection, crossover
and mutation procedure unless a pre-defined criterion is fulfilled.

1.2 Sensing Coverage Representation

Consider target area A in which set of sensor nodes is delineated as S = {s1 , s2 , s3 ,

…, sM }, where si is located at coordinates {ai , bi }, where i varies from 1 to M and M
be the total nodes which are deployed in area A. Each sensor has sensing radius Ra.
Let P be set of POCs distributed over the area A. If N is number of POCs, then P
= {p1 , p2 , …, pN }, where POC pj is situated at {aj , bj }, where j varies from 1 to N.
A binary coverage variable C i,j which signifies whether sensor si covers the POC pj
is delineated as follows:
  2  2
1 if ai − aj + bi − bj < Ra2
Ci,j =
0 Otherwise
Evading Gratuitous Energy Consumption Due to Activation … 1297

2 Literature Review

Currently, many academicians have investigated the optimization algorithm for the
coverage and location of nodes in wireless sensor networks. Cardei’s TianD algorithm
[5], Wang’s CCP algorithm [6] and Liang’s Huang algorithm [7] are some effective
approaches in this context.
Chen et al. [8] proposed a fusion memetic scaffold (Hy-MFCO) for optimizing
coverage. From real-world experimentations and computer simulations, they pro-
duced the outcomes that specify that Hy-MFCO is proficient in maximizing detection
coverage and, at the same moment, attaining energy effectiveness.
Liang et. al. [7] proposed a method that uses the adaptive group head communica-
tion method to guarantee that power burning is unprejudiced throughout the network
and can perk up the network time phase. Cardei et al. [5] designed a scheme to
ascertain a “maximum separation set coverage problem”. In this scheme, the nodes
that are incorporated in the utmost separation coverage area are in operation, while
residual nodes remain in dormant state. Thus, much energy is saved, and life cycle
of the network is enhanced. Jia et al. [9] presented a weighted genetic algorithm
and optimization coverage mechanism based on the genetic restriction algorithm.
According to the fitness function generated to perform the operation of the genetic
algorithm, a complete coverage province is needed for the guessing the finest set of
nodes and the absolute assortment of the work node, thus prolonging the endurance
of the network. Han [10] proposed a scheme anchored in a genetic algorithm cov-
erage programming and evolutionary inclusive search techniques to monitor all the
targets and that can discover the optimum coverage set, extending the life span of
the network.

3 Proposed Approach

3.1 Problem Formulation

In this study, the focal point of our work is to manage the POC coverage with least
energy efficiently. Consequently, our endeavor is to locate such specific set of sensor
nodes such that each POC is covered by at least one node. This hitch can be devised
mathematically as follows:
Optimization Model:

Min costi .ai i = 1, 2, 3, 4 . . . M
Subject to :

Ci,j ai ≥ 1, j = 1, 2, 3, 4 . . . N
xi = 1 or 0, i ∈ [1, M]
1298 A. Misra and D. S. Yadav

where Costi is the outlay of stimulating ith sensor node; ai ’s are the key assess-
ment variables which are determined by proposed approach. The proposed approach
decides the value of a zero if it should be inactive otherwise a is set 1. The objective
function diminishes the total number of nodes required to be activated such that each
POC in sensing vicinity is covered.

3.2 Absolute Sensing-Area Coverage Using Extended

Genetic Algorithm

The proposed “Absolute sensing-Area Coverage Using extended Genetic Algorithm”

(ASCeGA) comprises two optimization strategies: an extended GA approach for
schedule determination for sensor nodes and a stir-up proposal. The foremost scheme
deactivates superfluous nodes in clustered WSN in accordance with the proposed
schedule for nodes. The working of proposed approach is exemplified in Fig. 2.
The subsequent stir-up proposal handles the energy-proficient coverage optimiza-
tion every time. The initial populace is usually produced in random manner. Selec-
tion, crossover and mutation are genetic operations, which are used in evolutionary
process.
In the ASCeGA, we use the fitness function to evaluate the rectitude of every
individual solution (gene). After completion of genetic operations, we apply the
extended search to further improve the rectitude of the solutions. After extended
search, a novel populace of superior genes is produced. Individuals in the novel
populace are h in the vicinity of to the overall finest solution. Additionally, when
termination criterion is satisfied, it will be assumed that finest solution has been
found, and thus the evolutionary process will be ended.
Figure 1 depicts the genetic representation for the power-proficient coverage opti-
mization. There is X number of genes, and allele li,j denotes whether sensor node sj
in gene i is active or not. As there are M nodes, the length of each gene is M. Here,
we take fixed size of populace size for every populace generated in each generation.
With the aim to preserve power and attain the finest coverage ratio, a best schedule

Fig. 1 Genetic
representation used in the
ASCeGA
Evading Gratuitous Energy Consumption Due to Activation … 1299

for sensor nodes is included into the ASCeGA to dormant these superfluous nodes
(Fig. 2).
We also propose a coverage vector CV to characterize the coverage of each POCs.
By using the sensing coverage model (Sect. 3.1), we delineate the coverage vector
of si as CVi = [C i, 1, C i, 2, …, C i, N ], where si ∈ S. In the similar way, for another

Fig. 2 Flowchart of the ASCeGA

1300 A. Misra and D. S. Yadav

sensor node s j ∈ S, the coverage vector is would be CVj = [C j, 1 , C j, 2 , …, C j, N ],

where i = j. By using binary model, following synthetic coverage vector (SCV) can
be delineated for si and sj to symbolize whether a specified POC is covered by these
sensors.
   
SCV si , sj = CVi ∨ CVj = Ci,1 ∨ Cj,1 , Ci,2 ∨ Cj,2 , . . . , Ci,N ∨ Cj,N

where SCV designates a synthetic coverage vector, which determines whether si and
sj cover every POC in combined manner or not. Consequently, the SCV for a gene k
is delineated as follows:
   
SCV(k) = lk, 1 .CV1 ∨ lk, 2 .CV2 . . . ∨ (lkM , CVM )

As the coverage assessment procedure is made simpler into binary operations, the
working of ASCeGA can be enhanced significantly. Additionally, we can establish
the coverage ratio (CoR) for the gene k by:

SCV(k)
CoRk =
N

where SCV(k) represents the number of POC covered by gene k. The utility ratio
(UR) of nodes for gene k is computed by:


N
lk, p
p=1
URk =
M

where Np=1 lk, p represents total nodes that have been chosen to be stimulated. We
delineate fitness function f k as the goodness of gene k and devise it as:

fk = CoRk − URk

Substituting CoRk and URk :


N
lk, p
SCV(k) p=1
fk = −
N M
The constrained boundaries are 0 ≤CoRk ≤1, 0 ≤URk ≤ 1 and 1≤ f k ≤1. Thus,
the solution encoded in gene k is considered better if it has higher value of f k.
Genetic Operations
Selection, crossover and mutation are the operations generally used in genetic
algorithms. For selection, we have numerous strategies but proportional fitness,
roulette wheel and fixture selections are mostly used. The fixture assortment strategy
Evading Gratuitous Energy Consumption Due to Activation … 1301

has been exploited in proposed ASCeGA. We have opted for this just because it
gives the best fitness of each generation more powerfully. In the fixture assortment
approach, a competition is organized among arbitrarily selected individuals and then
chooses the conqueror for crossover. In proposed ASCeGA, every allele signifies
the state of a sensor which is kept 1 for active and 0 for dormant. The gene with
elevated fitness value implies that the clustered WSN has an improved schedule for
sensor nodes while applying ASCeGA. The result of selection procedure is used for
crossover. A single-point crossover is used in crossover task.
After the crossover, we apply the mutation operation which changes one or more
values in any probable gene allele. It supports the whole GA to thwart the populace
from being ensnared in a local best solution. Therefore, the newly constructed indi-
vidual (gene) is added to the original gene pool. The concluding offspring have an
elevated fitness because of iterative operations of crossover as well as mutation.
Extended Search Scheme
In this research work, an extended exploration plan is derived so that proposed
ASCeGA converges rapidly thus further perk up the righteousness of the populace
computed by genetic operations. In this approach, we alter the value of every allele
from one to zero and keep the updated gene if the fitness value of new gene is greater
than the fitness value of original gene.
In this way, gene can be polished to an improved one if superfluous nodes are
found. Contrasting the traditional genetic approaches, the proposed ASCeGA gives
superior outcome by performing the extended search process. Using the developed
extended exploration scheme, the ASCeGA converges swiftly.
The ASCeGA keeps sprouting until a extinction condition is fulfilled. Here, the
evolutionary process is finished when its optimal result is unaffected for η subsequent
generations.
Stir-up
Our stir-up mechanism comes into picture when active node looses its entire
energy. When such situation occurs, some POC becomes uncovered. Thus, some
dormant nodes are required to be stimulated to recover the coverage of uncovered
POCs.
After each transmission, each active sensor of gene is checked whether it is still
active or exhausted its whole energy. If one sensor si drains its entire energy,
 the BS

Si
will re-examine the coverage of network and find the uncovered POCs CVuncovered
by taking exclusive OR between the original SCV of all sensor nodes including
sensor si , and the SCV without sensor si .
Let node si ∈ S, where S is the set of all sensor nodes, and i =1, 2, 3 … M.
The ui signifies the set of all adjacent sensors of si , where i =1, 2, 3 … M. If SCV
(synthetic coverage vectors) of some sensors which belong to ui, have overlapped
Si
with CVuncovered , let OPi be the set of such sensors. Additionally, let SOPi, k characterize
the k-th probable subset of OPi , and w is total number of subsets in OPi.
Process of the stir-up proposal
best_comb, best_value = 0;
map=[0, 0, …,0], res = [0, 0, …,0], length = N;
1302 A. Misra and D. S. Yadav

If (sensor si has exhausted its energy and cannot work acceptably) then

{ Loop k =1 to w

{ Loop every element d of S OPk

{ map =CVd map ;}
S
res=map CVunicovered
1
if a > best_value then
a re s size ( S OPk ) 1
1
best_value= b
b re s size ( S OPk ) 1

S OPk
best_comb= ;
}
output = best_comb;
}

Stir-up proposal spawns a best schedule to stimulate and makes some sensors in
a dormant mode. As mentioned above, some dormant nodes will be activated by
BS according to the finest schedule of nodes generated by the stir-up method at the
commencement of the next round. If any sensor loses its complete energy yet again,
the sit-up method will re-examine the coverage and resolve to stimulate some other
sensor nodes to convalesce the uncovered POCs. In this way, we succeed to preserve
the coverage of POCs in effective manner.

4 Experimental Evaluation

We simulated our proposed approach in a field with dimensions 100 m×100 m where
10 POCs and 100 nodes disseminated randomly in a sensing field. All the POCs are
also deployed randomly. MATLAB is used to implement the simulations. The sensing
range Ra is assumed as 12.61 m considering 20% active nodes. Nodes with space
less than 12.61 m are delineated as neighbors. Number of generations to which the
optimization carried out is 30. Crossover rate (Rc) is 0.8, and mutation rate (Rm) is
0.006. Transmit/receive energy (Eelec) is 50 nJ/bit, amplification energy (Eamp) is
100 pJ/bit/m2 , and data aggregation energy EDA is 5 nJ/bit/report. Location of base
station is (50, 200). Data packet size is 400 bits. Initial energy of each node is 0.5 J.
Figure 3 depicts the deployment of nodes and coverage for POCs after applying
the proposed optimal schedule.
Evading Gratuitous Energy Consumption Due to Activation … 1303

100
100 63
16 36 68 45
44 416842 45 11 66
69
80
80 8 77 15
1280 5 52 3095 24
34
97
74 49 17
84 37 27
100
60 3547 8562 61 7572 18 96 60 62
92 13
99 31 83 6
87
94 32 90 239 81
40 7 40
9865 23 19 40
88 19
5370 59 8226 10
55
28
20 64784858
43
86
9320
25 54 79 20 78
89 14 21 50 4 1 3876
7339227133 91
51 46
57 60
0 29 67 56 51 57
0
0 50 100 0 20 40 60 80

Fig. 3 Unvarying deployments of sensors and POCs

We evaluate our proposed approach with VP-NL [11] and FDA [12] and K cov-
erage GA [13]. We conducted simulations using the identical network mentioned
previously where nodes and POCs are deployed randomly.
Sensing coverage ratio is exemplified in Fig. 4 against number of rounds. This
figure noticeably designates that VP-NL and FDA provide meager capabilities in
maintaining coverage ratio (CoR). While the ASCeGA reaches to 0% sensing cov-
erage ratio at round number approx 4000 which is greater than all other compared
approaches.
Figure 5 shows the comparison of life span of propose and other approaches. We
detect that the sensors loose entire energy rapidly using VP-NL and FDA methods
because of lack of efficiency in scheduling strategy. Conversely, the K coverage

1
ASCeGA
0.9 K Coverage GA
VP-NL
0.8 FDA

0.7
Sensing coverage ratio

0.6

0.5

0.4

0.3

0.2

0.1

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500
Number of rounds

Fig. 4 Rounds verses sensing coverage ratio versus

1304 A. Misra and D. S. Yadav

4500
ASCeGA
4000 K Coverage GA
VP-NL
FDA
3500
Number of rounds

3000

2500

2000

1500

1000

500
0 10 20 30 40 50 60 70 80 90 100
Percentage of node death (%)

Fig. 5 Number of rounds versus percentage of dead nodes

GA and ASCeGA can deactivate the superfluous sensor nodes via node-scheduling
strategy that accumulate a large amount of energy, so the network life span can be
enhanced.

5 Conclusion

As sensor partakes in the network operations only for the time they have energy, then
energy competence in the blueprint of every aspect of such nodes is required. Energy
burning in sensors occurs chiefly due to computational processing. In this paper, we
designed an energy-proficient approach by exploiting the genetic algorithm so that
only minimum nodes should be active at any time and participate in communication.
If any active node dies, any sleeping node to recover the network coverage is woken.
Our experimental evaluation shows that by using the extensive genetic algorithm,
we attained the better network life span and coverage.

References

1. Biswas S, Das R, Chatterjee P (2018) Energy-efficient connected target coverage in multi-

hop wireless sensor networks. Industry interactive innovations in science, engineering and
technology, Springer, Singapore, pp 411–421
2. Liu Y, Pu J, Zhang S, Liu Y, Xiong Z (2009) A localized coverage preserving protocol for
wireless sensor networks. Sensors 9:281–302
3. Bellur U, Jaiswal N (2006) Power aware duty scheduling in wireless sensor networks. Lect
Note comput Sci 4308:546–551
Evading Gratuitous Energy Consumption Due to Activation … 1305

4. Boukerche A, Fei X, Araujo RB (2007) An optimal coverage-preserving scheme for wireless

sensor networks based on local information exchange. Comput Commun 30:2708–2720
5. Cardei M, Du DZ (2005) Improving wireless sensor network lifetime through power aware
organization. Wirel Netw 11:333–340
6. Wang XR, Xing GL, Zhang YF, Lu CY, Pless R, Gill CD (2003) Integrated coverage and con-
nectivity configuration in wireless sensor networks. In: Proceedings of the 1st ACM conference
on embedded networked sensor systems (SenSys’03), Los Angeles, CA
7. Liang Y, Zeng P, Yu H (2006) Energy adaptive cluster-head selection for wireless sensor
networks. Inf Control 35:141–146
8. Chen CP, Mukhopadhyay SC, Chuang CL, Lin TS, Liao MS, Wang YC, Jiang JA (2015)
A hybrid memetic framework for coverage optimization in wireless sensor networks. IEEE
transactions on cybernetics 45(10):2309–2322
9. Jia J, Chen J, Chang J, Zhao L, Wang G (2007) Optimal coverage scheme based on genetic
algorithm in wireless sensor networks. Control Decis 1(22):1289–1292
10. Gil JM, Han YH (2011) A target coverage scheduling scheme based on genetic algorithms in
directional sensor networks. Sensors 11:1888–1906
11. Yang Q et al. (2015) Variable-power scheduling for perpetual target coverage in energy harvest-
ing wireless sensor networks. In: International symposium on wireless communication systems
(ISWCS), IEEE, pp 281–285
12. Wan X, Wu J, Shen X (2015) Maximal lifetime scheduling for roadside sensor networks with
survivability. IEEE Trans Veh Technol 64(11):5300–5313
13. Elhoseny M et al. (2017) K-coverage model based on genetic algorithm to extend WSN
lifetime.” IEEE Sens Lett 1.4:1–4
Detection of Phishing Websites Using
Machine Learning

Ahmed Raad Abbas, Sukhvir Singh and Mandeep Kau

Abstract Phishing is defined as imitating a creditable company’s website aiming to

take private information of a user. These phishing websites are to obtain confiden-
tial information such as usernames, passwords, banking credentials and some other
personal information. Website phishing is the act of attracting unsuspecting online
users into revealing private and confidential information which can be used by the
phisher in fraud, blackmail or other ways to negatively affect the users involved. In
this research, an approach had been proposed to detect phishing websites by apply-
ing a different kind of algorithms and filters to achieve a reliable and accurate result.
The experiments were performed on four machine learning algorithms, e.g., SMO,
logistic regression and Naïve Bayes. Logistic regression classifiers were found to be
the best classifier for the phishing website detection. In addition, the accuracy was
enhanced when the filter had been applied to logistic regression algorithm.

1 Introduction

Online phishing is considered a criminal way to deceive users to reveal their important
information, like username, password, as it mentioned in Fig. 1, credit card detail,
etc., by pretending to be a trustworthy entity in the online communication. They get
the users trust by claiming that they are from a legitimate party, like well-known
services providers or financial institution such as bank or PayPal, etc., and then they
misguide the users to a fraudulent website to grab their credentials. Phishing was
considered a crime for the industrial economy, due to the massive loss which they
suffered [1].
The number of the phishing attempts according to APWG is getting higher and
expanding.

A. R. Abbas (B) · S. Singh · M. Kau

Department of Information Technology Development (IT), University Institute of Engineering
and Technology (UIET), Punjab University, Chandigarh 160015, India
e-mail: [Link]@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1307

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1308 A. R. Abbas et al.

Fig. 1 An example of phishing

The statistic reports have displayed that the entire phishing attack was enlarged
significantly from 180.577 to 263.538 cases from the last quarter of 2017 until the
first quarter of the year 2018. As for the monetary losses, it was massive [2].
A lot of users without realizing click on phishing domain daily. The hackers did
not just target the clients yet the companies as well. In the 3rd Microsoft Safer Index
Report, which was out on 2014, the phishing attacks can influence very high a 5
billion dollar. The lack of awareness of the users is the main reason behind this
huge lost [3]. So, security protectors must take an action for clients not to face any
harmful sites. The protectors mostly attempt to rise to awareness of the company
against phishing to prevent any damage and build immune security system which
can confront phishing and prevent it.

2 Literature Review

A new methodology [4] was proposed to detect phishing websites, by utilizing arbi-
trary forests, for example, the sorting algorithm with the aid of RStudio. Therefore,
they were able to extract eight features which they have found it to be the best and
work it out to achieve an overall accuracy as they claimed 95%.
A new system was introduced by Sahingoz [5] which involves using seven dif-
ferent algorithms, for example, the Decision Tree, Ad boost, K-star, KNN (n = 3),
Random Forest, SMO and Naïve Bayes. And 3 kinds of properties, and present them
as NLP based features, word vectors, and hybrid features. Meanwhile, they have
constructed their individual dataset with 73,575 URLs. This set comprises 36,400
legitimate URLs and 37,175 phishing URLs. Moreover, they have found that the
NLP-biased features have performed the best among the other features’ types. More-
over, the utilization of NLP grounded properties and word vectors jointly similarly
will increase the functioning of the phishing detection structure. Authors use [6] a
Detection of Phishing Websites Using Machine Learning 1309

nonlinear regression approach to detect if the website is phishing or not. They have
used harmony search and support vector machine metaheuristic algorithms to train
the system. And as they have claimed, harmony search preforms an improved exact-
ness percentage of 94.13% and 92.80% for training process, respectively, by utilizing
about 11,000 web pages has proposed [7] an anti-phishing method, which utilizes
machine learning by selecting nineteen different properties in the user’s part to dif-
ferentiate between phishing website and legitimate. They have used 2141 phishing
pages from Phish Tank and Open phish, and 1918 genuine web pages from Alexa
popular websites, some online payment gateways, and some top banking websites. By
means of the machine learning, their anticipated method achieved 99.39% accurate
percentage.
In other studies, such as [8], the researcher uses a hybrid technique by merg-
ing machine learning methods with image checking. A significant drawback of the
image/visual grounded phishing discovery was around the requirement of a prelim-
inary image database or previous familiarity (web history) of that web page; never-
theless, the given method is unrestricted of these barriers. They have implemented
three featured groups: hyperlink-based features, third-party grounded features, and
URL obfuscation features. Despite the have used third-party services, they manage
to achieve high accuracy with 99.55%.
In [9], a classifier model was adopted which is able to detect phishing sites in
a smart and computerized method by utilizing a dataset freely available. The used
random forest algorithm has achieved 97.36% accuracy, and as they claimed that
their outcomes exhibited that RF is quicker, strong and more precise than the rest of
the classifiers.
The authors of [10] studied specific features which can increase the accuracy
efficiency to detect the phishing websites. As they claimed in their study that (URL-
grounded features, domain-grounded features, page-grounded features and content-
grounded features) are the best features for machine learning-grounded discovery of
phishing sites.
The authors of [11] in their study have recommended three new features which
were measured with the current features. The investigational findings show that the
new features are very powerful in contrast with the current features. The new features
should be recognizable among the phishing and non-phishing websites.

3 Detection of Phishing Websites

The proposed approach architecture is shown in Fig. 2. This proposed work starts
with investigating the accuracy the phishing detection system with complete set of
features. Then, applying different algorithms on the divided or grouped features is
investigated.
1310 A. R. Abbas et al.

Fig. 2 Proposed approach architecture

3.1 Pre-processing

In the pre-processing stage, phishing website data was gathered from UCI repository
website. Features have then been taken out from each website, all the features for
all websites are given by rows, and each row presents the possibility of the website
whether being legitimate, suspicious or phishing. By presenting different kind of
features which in return they were chosen very carefully to fulfill the gaps in the
phishing websites. Approximately, 30 features are presented, they were divided into
subgroups to cover all area of website phishing, and each group can deal with specific
and different area. The set of features were categorized into four collections: address
bar-based features (contains 12 features), abnormal-based features (contains 6 fea-
tures), HTML, JavaScript-based features (contains 5 features) and domain-based
features (consists of 7 features).

3.2 Selection of Features

As it was discussed before, the making of four groups is grounded on the structure
and the content of the website. This creates the collections as shown in Table 1.
Merged Groups Feature Selection:
Detection of Phishing Websites Using Machine Learning 1311

Table 1 Groups of the

S. no The features group Number of features
features for phishing website
detection 1 Address bar-based features 12
2 Abnormal-based features 6
3 HTML, JavaScript-based 5
features
4 Domain-based features 7
5 All the features 30

Table 2 Merged groups

S. no The features group Number of features
1 Group (1 + 2) 18
2 Group (1 + 3) 17
3 Group (1 + 4) 19
4 Group (2 + 3) 11
5 Group (2 + 4) 13
6 Group (3 + 4) 12

All these groups had been combined in Table 2, where each group presents a
combination of two groups from the four features groups. The aim is finding the best
group of features to apply in area of phishing detection for more accuracy.

3.3 The Used Algorithms

Four supervised ML algorithms were used, to prepare and examine the overall exact-
ness with the grouped features for phishing websites detection. They were preferred
and selected because of their different training strategy they can be used to figure out
the mechanism and the rules of testing and learning, and the selected algorithms are
well-known and listed below:
NAÏVE BAYES
SMO
Logistic Regression
1312 A. R. Abbas et al.

3.4 Resample Filter

Resample filter adds instances to a class. This is understood by merely adding

instances from the class which has only few instances multiple times to the result
dataset. Produces a random subsample of a dataset using either sampling with replace-
ment or without replacement. It balances the dataset by generating new data for the
minor class [12].

4 Experimental Results on Features Selection

The results show that the three algorithms have achieved almost same accuracy in
our test scenarios. However, the SMO classifier algorithm has the lowest mean of
accuracy in both feature selection and groups selection as it is plotted in Table 3.
The logistic regression has achieved overall the best/highest accuracy at almost
all the scenarios and features that had been tested. The logistic regression algorithm
has shown that when it comes to websites phishing detection with use of the WEKA
platform can detect the phishing attempts, and it can be considered as a reliable
algorithm. On the other hand, the algorithms SMO and Naïve Bayes their results
were in some cases very close but according to our dataset and our tests has shown
that the performance for both of these algorithms were not best results and SMO
algorithm consume a long time to process one test, this withdraw can affect the
detecting process.

5 Conclusion

This work principally comprises machine learning-based techniques to recognize

phishing websites. The used dataset contains 30 features, for more investigation
and analysis, the features of dataset had been divided into four groups, each group
contains related features, and then, their accuracy had been examined (precision and
recall). The groups merged, finally, six different groups had been obtained as a result
of this merging, and these groups had been analyzed to get the best results. All the
groups were tested with three classifiers, the logistic regression was found to be the
best regarding accuracy comparing with other classifiers. Then, a resample filter had
been added to balance the dataset, which in return helped to enhance the result of
logistic regression with accuracy rate of 93.8%.
Table 3 Overall accuracy
Algorithm All Address Abnormal HTML, JAVA Domain Group
G1 G2 G3 G4 G5 G6
Naïve Bayes 92.9 89.5 87.2 55.3 70.6 92.7 89.6 90 86.9 87.9 71.2
SMO 93.9 89.5 86.6 55.2 69.5 92.8 89.5 89.5 87.5 88.7 69.5
Detection of Phishing Websites Using Machine Learning

Logistic regression 93.8 89.5 87.3 55.9 70.8 93.1 89.6 91 87.3 88.8 71.1
1313
1314 A. R. Abbas et al.

References

1. Kirlappos I, Sasse MA (2012) Security education against phishing: a modest proposal for a
major rethink. IEEE Secur Priv 10(2):24–32
2. APWG (2017) APWG Reports| APWG. [Link] [Online]. Available
[Link]
3. Katz J, Aspden P (1997) Motivations for and barriers to internet usage: results of a national
public opinion survey. Internet Res 7(3):170–188
4. Parekh S, Parikh D, Kotak S, Sankhe S (2018, April) A new method for detection of phishing
websites: URL detection. In: 2018 second international conference on inventive communication
and computational technologies (ICICCT). IEEE, pp 949–952
5. Sahingoz OK, Buber E (2018) Machine learning based phishing detection from URLs. Expert
Syst Appl. p 1–32
6. Rao RS, Pais AR (2018) Detection of phishing websites using an efficient feature-based
machine learning framework. Neural Comput Appl. 1–23
7. Subasi A, Molah E, Almkallawi F, Chaudhery TJ (2017, November) Intelligent phishing web-
site detection using random forest classifier. In: 2017 international conference on electrical and
computing technologies and applications (ICECTA). IEEE, pp 1–5
8. Buber E, Demir Ö, Sahingoz OK (2017, September) Feature selections for the machine learning
based detection of phishing websites. In: 2017 international artificial intelligence and data
processing symposium (IDAP). IEEE, pp 1–5
9. Abunadi A, Akanbi O, Zainal A (2013, December) Feature extraction process: a phishing
detection approach. In: 2013 13th international conference on intellient systems design and
applications. IEEE, pp 331–335
10. Domingos P, A general method for making classifiers cost-sensitive. Artificial Inelligence
Group, Instituto Superior Técnico, Lisboa, 1049–001
11. Babagoli M, Aghababa MP, Solouk V (2018) Heuristic nonlinear regression strategy for
detecting phishing websites. Soft Comput 1–13
12. Jain AK, Gupta BB (2018) Towards detection of phishing websites on client-side using machine
learning based approach. Telecommun Syst 68(4):687–700
Smart Drip Irrigation Using IOT

Smita Deshmukh, Swati Chavan, Prajakta Zodge, Pooja Dalvi

and Akshaykumar Jadhav

Abstract As we can see in today’s world, only some devices like PCs and mobiles are
connected to the Internet. Nowadays, the world is copiously surpassed by the Internet
and the internet of things. The Internet is used for the rudimentary need of all human
beings [3]. IoT represents the concurrence of advances in minimization, wireless
connectivity, enhances batteries and data storage capacity, and without sensors, IoT
is not possible. It simply means to monitor a physical device or machine, or it is
inter-networking of physical devices which is entrenched with electronics, sensors,
software, and network connectivity to facilitate it to achieve greater value and services
by swapping data with the producer [1]. IoT permits objects to be sensed or controlled
remotely across the network infrastructure. The result improves accuracy, economic
benefits, efficiency and reduces the intervention of a human. In this paper, we are
going to deal with rudimentary and imperative perceptions of IoT and its scope in the
forthcoming future. This paper studies the need for IoT in day-to-day life for different
applications and gives fleeting information about IoT. IoT contributes significantly
toward revolutionary farming approaches. So, we are trying to demonstrate IoT in
the automatic watering system [2]. An automatic watering system monitors and
preserves the approximate moisture gratified in the soil. Raspberry Pi is used as a
microcontroller to implement the control unit. The setup uses the temperature sensor,
moisture sensor and humidity sensor which measure the approximate temperature,
moisture and humidity in the land. This value empowers the system to use belonging
quantity of water which avoids over/under irrigation.

S. Deshmukh · S. Chavan (B) · P. Zodge · P. Dalvi · A. Jadhav

IT Department, Terna Engineering College, Mumbai University, Plot no. 12, Sector-22, Phase 2,
Nerul(W), Navi Mumbai, Maharashtra 400706, India
e-mail: Swati98tc@[Link]
P. Zodge
e-mail: Prajaktazodge21@[Link]
P. Dalvi
e-mail: Dalvipoojav25@[Link]
A. Jadhav
e-mail: Aj773144@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1315

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1316 S. Deshmukh et al.

Keywords IoT · Moisture · Temperature · Humidity

1 Introduction

Agriculture is and will be the backbone of Indian economy [3]. Irrigation scheduling
is one of the solutions of the questions of “When do I water?” [4]. In our country,
monsoon rainfall is irregular and uneven. We have only 4% of the world’s fresh-
water resources to satisfy the agricultural needs for our 1.324 billion population,
and improper method of irrigation is the primary reason for water wastage in agri-
culture [5]. Automatic irrigation is easy to configure, and it helps to save energy
and resources. To make sustainable agriculture and prevent water wastage, smart
drip irrigation using IoT is proposed [6]. Sensors are physical device which con-
verts device parameter to electric signal on which we control whole system [6]. The
objective of this system is to render a reliable, robust, efficient and intelligent drip
irrigation controller device-based system which is smart enough to analyze distinct
parameters of a field like moisture, temperature, humidity, etc., and provides a water
delivering schedule in a targeted manner near the root zone of the crop to ensure all
the crops get enough water for their healthy growth, thereby reducing manual inter-
vention of farmer [7]. Sensors measure the field parameter which helps to reduce
wastage of water than other systems [8]. The system analyzes the soil quality to avoid
soil erosion [9]. It provides capability to farmer for control system through Android
application. The system uses emitter lines with different nozzles that can control
water flow so that plants like succulents can get less water, while plants with high
water requirements can get more and keep a check on the amount of water used for
irrigation [8]. The system gathers local weather information and some even factors
in the field landscape (types of plants, soil quality, slopes, etc.) to make irrigation
run-time adjustments, so the crops always receive the appropriate amount of water
[7]. In order to inform the farmer about exact field condition and provide manual
control over the system, system incorporates the concept of IoT (internet of things)
via mobile app [10]. The Android application has a user interface (UI) which will
show all the data to user in graphical format [11]. System provides benefit of both
reduction of labor cost and water wastage.

2 Existing System

The prior system consists of mainly two parts hardware and software [8]. In that, soft-
ware is a web page designed by using PHP, and soil content monitored by hardware
part [8]. The existing system consists of Arduino along with sensors which monitor
the soil moisture content and accordingly irrigating the fields as when needed [8].
This system introduced a GSM-SMS secluded measurement and control system for
Smart Drip Irrigation Using IOT 1317

farms based on PC-based database system which are connected with the base sta-
tion, which is developed by using a microcontroller, GSM module, actuators and
sensors. Many conditions like the status of electricity, running status of a motor,
increased/decreased temperature, changing a level of moisture will be informed to
the users via SMS on GSM network or by Bluetooth [8]. The system receives and
sends messages through GSM module, and parameters like temperature, air humidity
and moisture which are set by the central station are measured in every base station.
Parameter is exchanged between a far end and designed system via SMS on GSM
network. A SIM with 3G data pack inserted into a system which provides IoT fea-
tures to the system. This system sets the irrigation time depending on reading from
sensors and type of crop, and it can automatically irrigate the field when needed.
The sensor parameters regularly updated on a web page by using the GSM-GPRS
SIM900A parameter.

3 Proposed System

Features of Smart Drip Irrigation Using IOT:

1. Automated drip system collects real-time data of the water content in the root
zone of the crop as an input argument, correlates it with other parameters such as
temperature, humidity, insolation, light-intensity, barometric pressure of envi-
ronment and outputs the precise amount of water/fertilizers required for the
crop.
2. Water Source: The water level in the water reserve is kept in check by the SDIS,
and the farmer is informed accordingly.
3. Real-time monitoring: System comprises a mobile app and GSM messaging
facility which enable the farmer to remotely monitor the status of the field by
knowing the sensor values.
4. Customization: A farmer can monitor and control the valve/motor status and
set the desired crop moisture level by operating a mobile app or through GSM
messaging service.
5. Set Preferences: Through the mobile app, a farmer can select the crop type, soil
type preferences to adjust the system for a specific type of crop.
This is smart drip irrigation using IoT is the system which is designed to minimize
water wastage in agriculture. It is cost effective, any farmer can use it in farm fields,
and it increases productivity and saves water resources.

4 Working:

The smart drip irrigation using IoT comprises four major sections which are as
follows:
1318 S. Deshmukh et al.

Sensor section: System collects real-time data of the water content in the root zone of
the crop as an input argument, correlates it with other parameters such as temperature,
humidity, insolation, light-intensity, barometric pressure of environment and outputs
the precise amount of water/fertilizers required for the crop. The water level in the
water reserve is kept in check by the system, and the farmer is informed accordingly.
Control section: Raspberry Pi monitors the system based on approximate parameter
calibrated by various sensors. It defines the threshold value based on which it takes
decisions whether to irrigate or not irrigate the field. SIS provides offline communi-
cation via GSM Module, an online communication via mobile app to notify the user
exact condition of field.
GSM module (offline):
• Control the motor/valve status via SMS.
• Set the desired crop moisture level.
• Monitoring different parameters of the crops like moisture, temperature, humidity,
soil fertility, insolation, light-intensity, barometric pressure of the environment.
• Monitor the status of water reserve and weather patterns.
Android application (via the Internet):
A farmer monitors and controls the valve/motor status and sets the desired crop
moisture level by operating the mobile app.
IoT section:
Machine learning algorithm:
• Monitors different parameters of the crop, water reserve, weather patterns and
supervises the irrigation.
• Maintains records of different crop parameters and corresponding weather
conditions.
• Performs analysis on historical data to predict the performance and watering needs
of crop and recommends optimum practices to avoid any damage to the crops.
• Notifications to the farmer via mobile app and GSM.
E-Plant: (Mobile app)
• Controlling the motor/valve status.
• Set the desired crop moisture level if necessary.
• Monitoring different parameters of the crop, water reserve and weather patterns.
• Selection of plant type, soil type as per farmer requirement.
• Day-wise graphical analysis of crop data.
• Predictions and recommendations based on the performance of the crop.
Smart Drip Irrigation Using IOT 1319

Block Diagram
1320 S. Deshmukh et al.

Screenshot
Smart Drip Irrigation Using IOT 1321

5 Conclusion

In our India, most of the water is consumed in agriculture activity. The methodology
was designed in the way such that it monitors temperature, humidity, moisture in the
soil, and the project provides an occasion to learn the existing systems, along with
their topographies and downsides. The approach can be used to switch the motor
depending on the favorable condition of plants, i.e., sensor values, thereby system-
atizing the process of irrigation, which helps to anticipate over or under irrigation of
soil, thereby avoiding crop damage which is one of the most time-efficient activities
in farming. The farm owner can observe the development online through an Android
app. This research work determined that there can be a significant advancement in
farming with the use of IoT and automation.

Reference

1. [Link]
[Link]
2. [Link]
3. [Link]
farm-irrigation-systems
4. Sahu CK, Behera P (2015) A low cost smart irrigation control system. In: IEEE sponsored 2nd
international conference on electronics and communication system (ICECS)
5. Kansara K et al. (2015) (IJCSIT) Int J Comput Sci Inf Technol 6(6):5331–5333. [Link]
[Link]/publication/296396429_Sensor_based_Automated_Irrigation_System_
with_IOT_A_Technical_Review
6. Bhagyashree KC, Rana JG (2016) Smart irrigation system using raspberry pi In IRJET 03(05)
7. Automated Irrigation System Using a Wireless Sensor Network and GPRS Module Article
(PDF Available). In: IEEE transactions on instrumentation and measurement
8. [Link]
9. Tararani G, Shital G, Sofiya K, Gouri P, Vasekar SR, (IRJET) Smart drip irrigation system
using IOT. [Link]
10. Sahu T, Automated smart irrigation system using raspberry pi. MTech CSE Gyan Ganga
Institute of Technology and Sciences Jabalpur, India. [Link]
volume172/number6/[Link]
11. Nemali KS, van Iersel MW (2006) An automated system for controlling drought stress and
irrigation in potted plants. Sci Hortic 110(3):292–297, Nov 2006
An Efficient Scheduling Algorithm
for Sensor-Based IoT Networks

M. Deva Priya, T. Suganya, A. Christy Jeba Malar, E. Dhivyaprabha,

Prajith Kesava Prasad and L. R. Vishnu Vardhan

Abstract Internet of Things (IoT) based networks with sensors are energy and delay
stringent. Efficient scheduling algorithms for IoT-based networks are the need of the
hour. Nodes with selfish behavior degrade the performance of the network. Hence, a
scheduling algorithm that schedules packets based on their emergencies and priorities
yields better results. In this paper, M/M/1 and M/M/N scheduling scheme to schedule
Emergency packets (E-packets) and Regular packets (R-packets) is proposed. The
next-hop nodes are chosen based on the trust value of nodes. It is seen that the
proposed scheme yields better results in terms of Packet Delivery Ratio (PDR),
end-to-end delay, throughput and routing overhead.

Keywords IoT networks · Wireless sensor network (WSN) · Scheduling ·

Backpressure scheduling · Emergency packets (E-packets) · Regular packets
(R-packets)

M. Deva Priya (B) · T. Suganya · A. Christy Jeba Malar · E. Dhivyaprabha · P. K. Prasad ·

L. R. Vishnu Vardhan
Sri Krishna College of Technology, Coimbatore, Tamil Nadu, India
e-mail: [Link]@[Link]
T. Suganya
e-mail: suganya.t@[Link]
A. Christy Jeba Malar
e-mail: [Link]@[Link]
E. Dhivyaprabha
e-mail: [Link]@[Link]
P. K. Prasad
e-mail: 15tucs144@[Link]
L. R. Vishnu Vardhan
e-mail: 18tucs253@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1323

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1324 M. Deva Priya et al.

1 Introduction

Internet of Things (IoT) is a technology in which devices including sensors are

connected through the Internet. It comprises of a large network including people
and things which accumulate and share data depending on the surrounding envi-
ronment and their usage [1]. IoT systems are versatile and find their applications in
diverse fields due to their flexibility in adapting to any environment. It is a powerful
technology with devices that are capable of supporting data collection, automation,
operations, etc.
A Wireless Sensor Network (WSN) includes several hundreds and thousands of
tiny sensors that sense, monitor, measure, process and communicate the values of real-
world parameters such as heat, humidity and weather [2]. These networks are more
economical in contrast to traditional wired networks with sensors. The sensors are
energy stringent and have limited lifetime. The sensors are equipped with transceivers
involving less power that aid in collecting data. To preserve the cost effectiveness
of WSNs, they are made of small batteries. This becomes a challenge as energy is a
factor that determines the lifetime of a network [3, 4].

1.1 Selfish Node Attack

In WSN, nodes forward data packets to their neighbors consuming bandwidth, energy
and memory. Selfish nodes maintain communication with desired nodes and forward
packets, but decline to collaborate with other nodes. They do not share resources
with others but use them only for their own necessity. They conserve their own
resources, but drain others’. They do not forward or retransmit packets leading to
loss of communication and degradation of performance of network [5].

2 Related Work

In this section, the scheduling schemes proposed by various authors for WSNs are
discussed. Backpressure Scheduling (BS) is not suitable for Emergency IoT (EIoT)
as the queueing model is inefficient for scheduling Emergency packets (E-packets).
Tassiulas and Ephremides [6] have propounded a traditional BS scheme which
was later extended to Mobile Ad hoc Networks (MANETs) [7], cellular networks
[8], WSN [9], energy stringent sensor networks [10], WSNs for multimedia [11] and
multi-hop wireless networks.
Sridharan et al. [12] have designed a protocol for rate control based on the BS
scheme. The throughput in a network is improved by the propounded data aggrega-
tion scheme. The overall network stability is maintained by conserving energy in a
network [13].
An Efficient Scheduling Algorithm for Sensor-Based IoT Networks 1325

Moeller et al. [9] have proposed a BS scheme different from path-based schemes.
In path-based schemes, loops are absent and packets are transmitted through shorter
paths. In the modified BS scheme, the delay of BS is reduced and unnecessary long
path and loop problems are circumvented. These issues can be avoided by computing
the weights based on the difference in queue backlog.
Similarly, Dvir and Vasilakos [14] have proposed a hop count-based scheme
wherein packets are transmitted to a controlled area. Ying et al. [15] have designed
a clustering-based BS scheme wherein two types of queues namely, inter-cluster
queues and intra-cluster queues are maintained in each node.
The number of queues and the end-to-end delay are reduced when the traditional
BS scheme uses Last in First Out (LIFO) queue instead of First in First Out (FIFO)
queue. Alresaini et al. [16] have pre-computed the queue backlogs, enabling the net-
work to generate backlog difference gradients on low network loads. The end-to-end
delay is acceptable as the per-neighbor queue replaces the traditional backpressure
queue model.
Huang et al. [17] have examined the tradeoff between LIFO and FIFO queue
models. It is seen that the average energy consumption of BS scheme converges
gradually. They have not dealt with queuing emergency packets.
Ji et al. [18] have focused on reducing the forwarding packet delay by employing
a computation method for weighted queues. To deal with the packets at the end of
the queue, packet delay-based link weight assignment is done.
Hu and Gharavi [19] have found paths using a greedy approach. The hop count
is based on the lengths of the shortest paths.
It is seen that, in the recent past, BS schemes have gained much attention [20].
A distributed gradient-based backpressure framework [21] is proposed for WSNs
to deal with optimization of energy and throughput. Venkataraman et al. [22] have
designed a BS scheme that incorporates trust factor into link weights. Zheng et al. [23]
have propounded a network model, wherein the scheduling is dealt in the task layer.
Each task comes with a scheduling policy, followed by an application-dependent
transmission.

3 Existing System

Two existing schemes are discussed in this section namely, Last in First out (LIFO)
Backpressure Scheduling (BS) scheme and Event-Aware Backpressure Scheduling
(EABS) scheme.
1326 M. Deva Priya et al.

3.1 Last in First Out (LIFO) Backpressure Scheduling (BS)

Scheme

Packets in a network may be regular or important. Some packets may be important

as they pertain to emergency events and demand faster transmission.
LIFO BS scheme focuses on changing the traditional FIFO queue to LIFO [9].
Routing decisions are based on the difference in queue backlogs which is acceptable
for network with low loads. This may lead to longer and loop path problem. The
packets transmitted in a network can either be Emergency packets (E-packets) or
Regular packets (R-packets). The end-to-end delay increases with the increase in the
number of packets. If a common scheduling scheme is proposed to handle both the
packet types under LIFO BS scheme, E-packets will not be effectively serviced [12].
The E-packets are scheduled through shortest paths in a non-congested network.
They are delivered to the destination at the earliest. However, long paths, even
paths with loops may be selected in the traditional BS scheme. As the network
load increases, the numbers of loops also increase, leading to more end-to-end delay.
Guaranteeing the performance of a scheduling algorithm for E-packets is challenging
[13].

3.2 Event-Aware Backpressure Scheduling (EABS)

Event-Aware Backpressure Scheduling (EABS) scheme chooses the next-hop node

by integrating shortest path selection with the BS scheme. The E-packets are sent
through shortest paths. The difference in queue backlogs is considered to circum-
vent congestion. The end-to-end delay is reduced with an increase in the forwarding
percentage [14]. E-packets are to be delivered to the destination within the stipu-
lated time. For example, in case of fire, packets with temperature details should be
forwarded quickly. The number of E-packets is limited. Distance-weighted method
reduces the end-to-end delay of packets. E-packet is given priority and forwarded to
the nodes closer to the destination. When multiple E-packets are to be forwarded,
the one with the least deadline is chosen to be forwarded [24].

4 Proposed System

Selfish nodes degrade the performance of a network. The proposed Secure Event-
Aware Backpressure Scheduling (SEABS) scheme considers the E-packets in a net-
work that is prone to selfish node attacks and assigns more weight to them. In EABS,
though the queuing delay is reduced for E-packets, the dropping of packets is not
taken into consideration. The proposed SEABS scheme deals with the selfish node
attack.
An Efficient Scheduling Algorithm for Sensor-Based IoT Networks 1327

Assume that some intermediate nodes act as selfish nodes and drop the Route
Request (RREQ) packets. The E-packets should be capable of traversing through the
network and reaching the destination quickly. In the proposed scheme, two queues
namely, M/M/1 or M/M/n are used. The RREQs that are broadcast are put into the
M/M/n queue, and the Route Responses (RREPs) are scheduled in an M/M/1 queue,
as the E-packets are unicast when selfish node behavior is sensed in the network.
The scheduler is supported by a packet classifier, which classifies the packets as R-
packets and E-packets. The E-packets are given priority, and the scheduler forwards
the packets through the nodes with better trust value.
The E-packets are scheduled using higher priority M/M/1 queue. R-packets are
scheduled using M/M/n Queue.
The backlog of queue ‘Q’ at a node at time slot ‘t + 1’ is given by ‘Q BT (t + 1)’.
Q Q
Q BT (t + 1) = Q TR (t) + RegDES (t) + EmgDES (t) (1)

where
Q BT (t) Queue backlog at time slot ‘t’
Q TR (t) Queue transmission rate at time slot ‘t’
Q
RegDES (t) Number of R-packets arriving at a queue
Q
EmgDES (t) Number of E-packets arriving at a queue
The selected queue ‘qsel ’ is given by,
 
HCDES − HCSRC DES
qsel = TR (t)
Q SRC − .Q TR (t) (2)
HCSRC

where

TR (t)
Q SRC Queue backlog at source at time slot ‘t’
TR (t)
Q DES Queue backlog at destination at time slot ‘t’
HCSRC The hop-count at source
HCDES The hop-count at destination
The trustworthiness of node ‘y’ by node ‘x’ is given by the following equation.

mc + µ ms
Trustxy = (3)
mt + µ ma

where
µ Probability that the link remains active and correct
mc Transmitted message found to be correct
ms Successful transmissions
mt Total number of messages not destined but transmitted by ‘x’ to ‘y’
ma Total number of attempted transmissions
1328 M. Deva Priya et al.

After selecting the forwarding queues, the distance weight ‘w’ can be expressed
as,
 qsel qsel 
HCDES − HSRC
w= qsel (4)
HSRC

where
q
sel
HCDES The hop-count at the source for the selected queue
qsel
HSRC The hop-count at destination for the selected queue
Weighted queue backlog difference ‘w(t)’ is given by,
 qsel qsel 
w(t) = Trustxy · Q BT (t) − w ∗ Q BT (t) . (5)

where
qsel
Q BT (t) Queue Backlog at time slot ‘t’ of the selected queue
The node with maximum ‘w(t)’ is selected as the next-hop node.

5 Performance Evaluation

Network Simulator-2 (Ns-2) is used for simulations. A total of 50 mobile nodes are
randomly deployed in a rectangular area 1000 m × 1000 m.
The capacity of the wireless channel capacity is taken as 2 Mbps. The simulation
period is taken as 50 sec.
The performance is analyzed by varying the number of selfish nodes. The packet
size is fixed to 512 bytes. The random waypoint model is used.
The performance of the proposed Secured Event-Aware Backpressure Scheduling
(SEABS) scheme is compared with the existing Last in First out (LIFO) Backpres-
sure Scheduling (BS) scheme and Event-Aware Backpressure Scheduling (EABS)
scheme. The performance is analyzed in terms of Packet Delivery Ratio (PDR),
throughput, end-to-end delay and routing overhead. It is seen that the proposed
SEABS offers better PDR and throughput, involving less end-to-end delay and
routing overhead in contrast to the existing BS and EABS schemes (Figs. 1, 2,
3, 4).
The proposed SEABS offers 1.5 times and 1.2 times better PDR in contrast to the
existing BS and EABS scheduling schemes respectively. Similarly, it yields 3 and
1.8 times better Throughput when compared to BS and EABS scheduling schemes
respectively. The existing schemes BS and EABS schemes respectively involve 2 and
1.5 times more delay when compared to the proposed SEABS scheme. Similarly,
they involve 1.8 and 1.6 times more routing overhead in contrast to SEABS.
An Efficient Scheduling Algorithm for Sensor-Based IoT Networks 1329

Fig. 1 Packet delivery ratio

Fig. 2 Throughput

6 Conclusion

In this paper, a Secured Event-Aware Backpressure Scheduling (SEABS) scheme is

propounded. The performance of the proposed system is compared with Last in First
out (LIFO) Backpressure Scheduling (BS) scheme and Event-Aware Backpressure
Scheduling (EABS) scheme. As the next-hop nodes are selected based on the weights
computed by taking the trust value and the weighted queue backlog difference, the
proposed scheme outperforms the existing schemes.
1330 M. Deva Priya et al.

Fig. 3 End-to-end delay

Fig. 4 Routing overhead

References

1. Chen H, Jia X, Li H (2011) A brief introduction to IoT gateway. In: Proceedings of the IET
International Conference on Communication Technology and Application, ICCTA, pp 610–613
2. Ian I, Su W, Sankarasubramaniam Y, Cayirci E (2002) Wireless sensor networks: a survey.
Comput Netw 38(4):393–422
3. Abbasi AA, Younis M (2011) A survey on clustering algorithms for wireless sensor networks.
Comput Commun 30(14–15):2826–2841
4. Boyinbode O, Le H, Takizawa M (2011) A survey on clustering algorithms for wireless sensor
networks. Int J Sp-Based Situat Comput 1(2–3):130–136
An Efficient Scheduling Algorithm for Sensor-Based IoT Networks 1331

5. Das SK, Saha BJ, Chatterjee PS (2014) Selfish node detection and its behavior in WSN.
In: Proceedings of the IEEE International Conference on Computing Communication and
Networking Technologies, ICCCNT, pp 1–6
6. Tassiulas L, Ephremides A (1992) Stability properties of constrained queueing systems and
scheduling policies for maximum throughput in multihop radio networks. IEEE Trans Autom
Control 37(12):1936–1948
7. Neely MJ, Modiano E, Rohrs CE (2005) Dynamic power allocation and routing for time-varying
wireless networks. IEEE J Sel Areas Commun 23(1):89–103
8. Andrews M, Kumaran K, Ramanan K, Stolyar A, Vijayakumar R, Whiting P (2000) CDMA
data QoS scheduling on the forward link with variable channel conditions. Bell Lab Tech Memo
4:1–45
9. Moeller S, Sridharan A, Krishnamachari B, Gnawali O (2010) Routing without routes: the back-
pressure collection protocol. In: Proceedings of the 9th ACM/IEEE International Conference
on Information Processing in Sensor Network, pp 279–290
10. Liu L, Jiang J, Shu L, Hancke G (2017) Analysis of energy-efficient connected target coverage
algorithms for industrial wireless sensor networks. IEEE Trans Industr Inf 13(1):135–143
11. Majidi A, Mirvaziri H (2014) BDCC: backpressure routing and dynamic prioritization for
congestion control in WMSNs. Int J Comput Netw Inf Secur 6(5):29
12. Sridharan A, Moeller S, Krishnamachari B (2005) Investigating backpressure based rate control
protocols for wireless sensor networks, vol 7. USC EE CENG technical report, CENG-2008
13. Neely MJ, Urgaonkar R (2009) Optimal backpressure routing for wireless networks with multi-
receiver diversity. Ad Hoc Netw 7(5):862–881
14. Dvir A, Vasilakos AV (2010) Backpressure-based routing protocol for DTNs. ACM SIGCOMM
Comput Commun Rev 40(4):405–406
15. Ying L, Srikant R, Towsley D, Liu S (2011) Cluster-based back-pressure routing algorithm.
IEEE/ACM Trans Netw (TON) 19(6):1773–1786
16. Alresaini M, Sathiamoorthy M, Krishnamachari B, Neely MJ (2012) Backpressure with adap-
tive redundancy. In: Proceedings of the IEEE International Conference on Backpressure with
Adaptive Redundancy, pp 2300–2308
17. Huang L, Moeller S, Neely MJ, Krishnamachari B (2013) LIFO-backpressure achieves near-
optimal utility-delay tradeoff. IEEE/ACM Trans Netw 21(3):831–844
18. Ji B, Joo C, Shroff NB (2013) Delay-based back-pressure scheduling in multihop wireless
networks. IEEE/ACM Trans Netw 21(5):1539–1552
19. Hu B, Gharavi H (2014) Greedy backpressure routing for smart grid sensor networks. In:
Proceeding of the 11th IEEE Consumer Communications and Networking Conference, pp
32–37
20. Jiao Z, Yao Z, Zhang B, Li C (2014) A distributed gradient-assisted anycast-based backpressure
framework for wireless sensor networks. In: Proceedings of the IEEE International Conference
on Communications (ICC), pp 2809–2814
21. Jiao Z, Zhang B, Gong W, Mouftah H (2015) A virtual queue-based back-pressure scheduling
algorithm for wireless sensor networks. EURASIP J Wirel Commun Netw 1(35)
22. Venkataraman R, Moeller S, Krishnamachari B, Rao TR (2015) Trust–based backpressure
routing in wireless sensor networks. Int J Sensor Netw 17(1):27–39
23. Zheng X, Cai Z, Li J, Gao H (2017) A study on application-aware scheduling in wireless
networks. IEEE Trans Mob Comput 16(7):1787–1801
24. Qiu T, Qiao R, Wu DO (2018) EABS: an event-aware backpressure scheduling scheme for
emergency Internet of Things. IEEE Trans Mob Comput 17(1):72–84
Cost-Based Meta-Task Scheduling
Algorithm for MultiCloud Computing
(CBMTSA)

B. J. Hubert Shanthan and L. Arockiam

Abstract MultiCloud plays vital role in providing the heterogeneous types of

resources to the user on-demand with minimum cost and time. Resource manage-
ment and scheduling act as an influential aspect in the improvement of the perfor-
mance of the MultiCloud environment. Scheduling deeds a considerable challenge
in the distributed heterogeneous multiple cloud systems. The existing min-min algo-
rithm is suitable for smaller number of tasks. The tasks are allocated to the VMs
with high-processing speed. The proposed cost-based meta-task scheduling algo-
rithm (CBMTSA) is feasible for the passive autonomous task-based scheduling for
MultiCloud systems. Scheduling and rescheduling are two stages involved in this
algorithm. Scheduling stage is used to allocate the tasks to the high-speed VMs.
Makespan value is computed in the scheduling stage. The computed makespan is an
outset value for rescheduling the tasks. The rescheduling stage is used to reschedule
the tasks from high-speed VMs to low-speed VMs. The CBMTSA surpasses the
existing min-min algorithm. Makespan, execution cost, and cloud server utilization
ratio are the metrics considered in this algorithm.

Keywords MultiCloud · CBMTSA · Makespan · Execution cost · Scheduling

stage · Rescheduling stage

1 Introduction

A cloud is an inevitable technology which provides the storage, computation, and

communication resources to the users connected to the Internet [1]. In MultiCloud
the cost is charged based on the usage of the VMs [2]. Platform as a Service (PaaS)
[3], Software as a Service (SaaS) [4] and Infrastructure as a Service (IaaS) are three
major models in the cloud [5]. The SaaS model is designed for the end users to access

B. J. Hubert Shanthan (B) · L. Arockiam

Department of Computer Science, St. Joseph’s College, Trichy, India
e-mail: hshanthan@[Link]
L. Arockiam
e-mail: larockiam@[Link]
© Springer Nature Singapore Pte Ltd. 2020 1333
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1334 B. J. Hubert Shanthan and L. Arockiam

through simple Web browser. IaaS service model is used for storage, computation,
and communication process among the virtual machines [6]. PaaS model is used for
the programmers to deploy and create their applications [7]. In the MultiCloud, there
are heterogeneous types of service providers with different pricing schemes based on
the SLA terms between CSPs and CSUs. MultiCloud is an advancement of the cloud
and it offers different types of cloud services from other cloud service providers [8].
The resource scheduling in MultiCloud is a tedious process. The solutions obtained
from the existing heuristics-based approaches do not reduce both cost and time
simultaneously [9]. The failure rate is comparatively less in the static scheduling,
when compared with the dynamic scheduling algorithms [10].
The main purpose of this article is to propose a cost-Based meta-task schedul-
ing algorithm (CBMTSA) curtails the execution rate of the virtual machines in the
MultiCloud environment and increase the resource utilization of the cloud server.
Section 2 narrates the existing literature works of the algorithm. Section 3 elucidates
the propounded CBMTSA. The Sect. 4 gives results and discussions and Sect. 5
concludes with research findings and list out the future directions of research.

2 Related Works

Lucas-Simarro et al. [11] proposed an optimized service deployment scheduling

strategy for MultiCloud environment. The modular broker architecture was suggested
based on cost optimization criteria. The user constraints such as budget, performance,
VM placement, and load balancing under different environmental conditions such
as static and dynamic were considered in their approach.
Kokilavani and Amalarethinam [12] articulated an independent and load-balanced
scheduling algorithm to dwindle the makespan and to use the underutilized cloud
servers. The algorithm comprised of two stages namely scheduling and rescheduling.
The scheduling stage was used to allocate and execute the tasks to the VM with
minimum execution time. The computed makespan value in the scheduling stage
was used as a threshold value.
Tang et al. [13] developed a mechanism for sharing the service provider resources.
The mechanism was designed to analyze the demands of the user and the pricing
policy was designed for the anticipating users. Nash equilibrium was a strategy based
on the game theory developed to solve the bidding problem among the users.
Amalarathinam et al. [14] enlightened a customer-oriented cost-based task
scheduling algorithm for cloud. The algorithm was designed with economical cost
of the user. The price was computed based on the duration the tasks were executed
in the virtual machines. Monetary cost and makespan were used to improve the
performance of the cloud systems.
Panda et al. [15] proffered a scheduling algorithm for allocation of the tasks in the
MultiCloud environment. Makespan and rate of utilization of the cloud server are
the parameters considered in this algorithm. Allocation, matching, and scheduling
are the stages used in that algorithm. The VMs with minimum competition time are
Cost-Based Meta-Task Scheduling Algorithm for MultiCloud … 1335

selected in the matching stage. The first-come-first-serve (FCFS) basis mechanism

was followed in the matching stage. The allocation stage acted as an intermediate
layer between matching and scheduling stage.
Elkader [16] expressed a novel computation-based load balance aware scheduling
algorithm for cloud. The algorithm comprised of two stages namely fill and spill.
The fill scheduler was used to schedule the task based on the VM capabilities in each
cloud. The spill scheduler was used to execute tasks with the VM.
Mei et al. [17] analyzed a mechanism to increase profit among the cloud service
providers. The pricing model was designed for short-term and long-term users. The
existing pricing model was emphasized on long-term users. The cloud broker acted
as a MultiCloud agent between service provider and the cloud clients. The optimized
multi server M/M/N/n queuing model was used to configure the VM for the user-
defined tasks. The queuing model helped to analyze the factor that affects the profit
in the cloud service providers.
Ali and Alam [18] suggested a mechanism for the allocation of the tasks with
minimum length to the VM with minimum completion time. The algorithm was
developed based on the standard min-min algorithm. The algorithm emphasized on
the smaller tasks and the waiting times for larger tasks were increased exponentially.
Mihailescu and Teo [19] extended dynamic pricing scheme for the MultiCloud
environment. Dynamic pricing scheme increased the efficiency of resource usage by
the customers. The pricing strategies were decided based on the resource demand of
the customers. The VMs in the cloud are charged high when the customer requests
are peak on demand. The VMs are charged with less cost when the customer requests
are low.
Ibrahim et al. [20] enlightened a task-based scheduling algorithm for the Multi-
Cloud. The pricing model for MultiCloud is designed in this algorithm. The algorithm
was designed to decrease the makespan and minimize the execution time and cost.
The pricing models of the companies such as Google Cloud and Amazon EC2 were
used in this algorithm.

3 Cost-Based Meta-Task Scheduling Algorithm (CBMTSA)

Figure 1 depicts the working mechanism of CBMTSA algorithm.

Scheduling and rescheduling are vital stages involved in this algorithm.
The input values are given in the form of ETC matrix for the scheduling stage.
The ETC matrix comprises of execution time of the tasks given by the user to the
virtual machines.
The formula for ETC calculation is given by the Eq. (1)

ETCij = ISi/PSj (1)

where ISi denotes millions of instructions (MI) which is also known as task length
and PSj denotes millions of instructions per second (MIPS) also known as virtual
1336 B. J. Hubert Shanthan and L. Arockiam

Fig. 1 Methodological diagram of CBMTSA algorithm

machine processing speed. The completion time of each task is computed by the
sum of expected execution time and the virtual machine ready time. The formula for
completion time is given below in the Eq. (2).

CTij = ETCij + Rti (2)

where ETCij denotes the ETC matrix and RTi denotes the ready time of each virtual
machine available in the MultiCloud.
The tasks from the ETC matrix are scheduled based on their minimum completion
time. The tasks with minimum execution time are allocated to the VM with minimum
completion time. Cloud manager acts as a centralized controller between cloud user
and cloud service provider. Cloud manager selects tasks from the ETC matrix based
on the first-come-first-serve (FCFS) order.
The manager computes the completion time of all the VMs in the MultiCloud
environment. After the computation process, cloud manager selects the VMs that
give minimum earliest completion time and assign the task to the VMs of the cor-
responding cloud. The ready time for all VMs in the clouds is updated after the
allocation of the tasks. The makespan value of VMs in each cloud is computed. The
makespan value is obtained from the scheduling stage of each VM and stored as an
outset value. The cost computed is based on the processing speed of the VMs. The
VMs with high-processing speed requires more cost and it completes the tasks within
minimum time. The low-speed VMs require very low cost and their completion time
is extremely high when compared to the high-speed VM.
Cost-Based Meta-Task Scheduling Algorithm for MultiCloud … 1337

The cloud VMs server utilization ratio is given below in the Eq. (3).

  n
Pr VM j = VM j ∗ Cost (3)
j=1

where VMj denotes the computation time of the jth virtual machine. Cost denotes
the price of the accessing VMs in the cloud. In the scheduling stage, the high-speed
VMs are emphasized and they are too expensive. It is not suitable for the users
who demand low-cost resources. So, to overcome this conflict, rescheduling stage is
introduced to reduce the execution cost of the VM and to increase the idle utilization
of VMs. The cloud manager selects the tasks with minimum execution time from the
high-speed VMs. The selected task is rescheduled to low-speed VMs with maximum
completion time. The computed makespan value in the rescheduling stage must not
exceed the threshold value computed in the scheduling stage. The rescheduling stage
balances the load among the virtual machines in the cloud and also reduces the overall
maximum completion time of the tasks. The formula for cloud resource utilization
ratio calculation is given by the Eq. (4)
suci
CRUi j = ∗ 100 (4)
Num_T

where Suci denotes the successive completion of tasks in each cloud, Num_T
represents the number of tasks available in the clouds.

3.1 Proposed CBMTSA Algorithm for MultiCloud

Computing

Input: Total Number of tasks, VMs, Clouds

Output: Makespan, Cost, Average Resource Utilization
Step 1: Start
Step 2: Input the VMs, clouds, processing speed, and task
Step 3: CALL CB SCHEDULE
Step 4: CALL CB RESCHEDULE
Step 5: Stop
CB-SCHEDULE
Step 1: Check the condition whether the queue lengths of the tasks are not empty
Step 2: Check the total number of VMs and clouds are not left empty
Step 3: Arrange the tasks in the increasing order
Step 4: The tasks with minimum completion time are selected and executed in the
VMs from each cloud
Step 5: Compute Makespan and execution cost of the Virtual machines in the clouds.
Step 6: Stop
1338 B. J. Hubert Shanthan and L. Arockiam

CB-RESCHEDULE
Step 1: The condition whether the Meta tasks and Virtual machines are not empty
Step 2: Compute the completion time of the tasks
Step 3: Arrange the tasks in decreasing order
Step 4: The tasks with Maximum Execution time are selected and executed in VMs
with high processing speed
Step 5: Reallot the selected task from high processing speed VM to low capacity
VMs
Step 6: The ready time of the VMs are updated frequently
Step 7: Calculate the Makespan and execution cost of VMs
Step 8: Calculate Cloud server utilization ration
Step 9: Stop

4 Results and Discussion

Table 1 represents the sample ETC matrix for 10 tasks, 4 virtual machines, and 2
clouds. The cloud C1 represents Amazon EC2 cloud and the cloud C2 represents
Google Cloud Console.
Table 2 details the scheduling stage of the CBMTSA algorithm. The stage uses
batch mode of scheduling mechanism where all the tasks are executed parallely in
the available virtual machines.
Table 3 shows the rescheduling stage of the CBMTSA algorithm. The rescheduling
process reduces the makespan, computation cost of the VMs, and also increases the
cloud server utilization.
Table 4 illustrates the makespan value of the rescheduling stage. The makespan
value in the rescheduling stage is 28.96. The total cost incurred in the rescheduling
stage is 191.13$.

5 Conclusion

A cost-based meta-task scheduling algorithm (CBMTSA) articulated to schedule

the independent tasks. The algorithm focuses mainly to reduce the computation
cost of the virtual machines. The existing min-min algorithm is suitable for smaller
number of tasks and it merely schedules the tasks to the VMs with high-processing
power. The min-min algorithm gives more importance to the high-speed VM cloud
service providers and it also costs high for the cloud consumers. The low-speed
VM or the virtual machine with minimum processing speed is not utilized in the
existing algorithm. To overcome the challenges, the proposed CBMTSA algorithm
is developed and the importance is given for both the low-speed VMs and the high-
speed VMs of the MultiCloud environment. The proposed CBMTSA comprised of
Table 1 Expected time to compute (ETC) matrix
Cloud Virtual machines (seconds) T1 T2 T3 T4 T5 T6 T7 T8 T9 T10
Cloud C1 VM1 11.6 6.92 0.61 6.01 1.99 15.63 72.7 16.17 1.2 12.66
VM2 8.21 4.82 0.43 4.18 1.4 10.87 50.86 11.44 0.84 8.78
Cloud C2 VM3 9.82 5.77 0.51 5 1.67 13.1 61.45 13.68 1.02 10.49
VM4 6.73 1.23 0.18 1.28 0.09 7.78 18.4 8.11 0.42 4.29
Cost-Based Meta-Task Scheduling Algorithm for MultiCloud …
1339
 1340

Table 2 Scheduling phase

Cloud Virtual machines (seconds) T1 T2 T3 T4 T5 T6 T7 T8 T9 T10
C1 (Amazon) VM1 (1S) 6.01
VM2 (3S) 4.82 0.84 8.78
C2 (Google cloud) VM3 (2S) 0.51 1.67 13.68
VM4 (4S) 6.73 7.78 18.4
B. J. Hubert Shanthan and L. Arockiam
Table 3 Rescheduling phase
Cloud Virtual machines (seconds) T1 T2 T3 T4 T5 T6 T7 T8 T9 T10
C1 (Amazon) VM1 (1S) 11.6 6.01 1.2
VM2 (3S) 4.82 8.78
C2 (Google cloud) VM3 (2S) 0.51 1.67 13.1 13.68
VM4 (4S) 18.4
Cost-Based Meta-Task Scheduling Algorithm for MultiCloud …
1341
1342 B. J. Hubert Shanthan and L. Arockiam

Table 4 Makespan results in the rescheduling phase

Clouds C1 C2
Virtual machines VM1 VM2 VM3 VM4
Makespan values (seconds) 18.81 13.6 28.96 18.4

two stages, namely scheduling stage and rescheduling stage. The scheduling stage
uses min-min scheduling algorithm. The rescheduling stage is introduced to underrate
the VMs computation cost and the makespan value of the VMs. The rescheduling
stage selects the task-VM pair with minimum execution time and it is rescheduled to
the task-VM pair with maximum completion time. The experimental results from the
simulator prove that the proposed algorithm surpasses the existing min-min algorithm
in terms of metrics such as computation cost, makespan, and cloud server utilization
rate. This algorithm enables the consumer to access the cloud resources at minimum
cost and time.

References

1. Mell P, Grance T (2009) The NIST definition of cloud computing. National Institute of
Standards and Technology, Information Technology Laboratory, Technical Report Ver. 15
2. Munteanu VI, Sandru C, Petcu D (2014) MultiCloud resource management: cloud service
interfacing. J Cloud Comput 3(1):1–23
3. Buyya R, Broberg J, Goscinski A (2012) Cloud computing principles and paradigms. Wiley
4. Lahmar F, Mezni H (2018) Multicloud service composition: a survey of current approaches
and issues. J Soft Evolut 1–24 (Wiley)
5. Shanthan BJH, Arockiam L (2018) Resource based load balanced min min algorithm
(RBLMM) for static meta task scheduling in cloud. In: International conference on advances
in computer science and technology. Int J Eng Technol Spec Issue 1–8
6. Jane VA, Shanthan BJH, Arockiam L (2018) Survey of algorithms for scheduling in the cloud:
in a metric perspective. Int J Comput Sci Eng 6(2):66–70
7. Shanthan BJH, VinothKumar DA, Karthigai Priya G, Arockiam L (2017) Scheduling for
Internet of Things applications on cloud: a review. Imp J Inter Discip Res (IJIR) 1(3):1649–1653
8. Shanthan BJH, Arockiam L (2018) Spell aware meta task scheduling algorithm (SAMTSA)
for MultiCloud. Int J Sci Res Comput Sci Appl Manag Stud 7(4):1–5
9. Stephen A, Shanthan BJH, Ravindran D (2018) Enhanced round Robin algorithm for cloud
computing. Int J Sci Res Comput Sci Appl Manag Stud 7(4):1–5
10. Shanthan BJH, Arockiam L (2018) Rate aware meta task scheduling algorithm for MultiCloud
computing (RAMTSA). IOP conference series. J Phys Conf Ser 1142:012001, pp 1–10. https://
[Link]/10.1088/1742-6596/1142/1/012001
11. Lucas-Simarro JL, Moreno-Vozmediano R, Montero RS, Llorente IM (2012) Scheduling
strategies for optimal service deployment across multiple clouds. Future generation computer
systems, Elsevier, pp 1431–1441
12. Kokilavani T, Amalarethinam DD (2011) Load balanced min-min algorithm for static meta-task
scheduling in grid computing. In: Int J Comput Appl 20(2):43–49
13. Tang L, He S, Li Q (2016) Double-sided bidding mechanism for resource sharing in mobile
cloud. IEEE J Cloud Comput 1–11
14. Amalarathinam DIG, Beena TLA (2015) Customer facilitated cost-based scheduling (CFCSC)
in cloud. Procedia, Elsevier, pp 660–667
Cost-Based Meta-Task Scheduling Algorithm for MultiCloud … 1343

15. Panda SK, Gupta I, Jana PK (2015) Allocation-aware task scheduling for heterogeneous
MultiCloud systems. In: Second international symposium on big data and cloud computing
challenges, Procedia computer science, Elsevier, pp 176–184
16. Elkader AA (2017) Enhancing the minimum average scheduling algorithm (MASA) based on
makespan minimizing. Artifi Intell Mach Learn J 17(1):9–13 (Delaware)
17. Mei J, Li K, Tong Z, Li Q, Li K (2018) Profit maximization for cloud brokers in cloud computing.
IEEE Trans Parallel Distrib Sys 1–18
18. Ali SA, Alam M (2018) Resource-aware min-min (RAMM) algorithm for resource allocation
in cloud computing environment. arXiv:1803.00045
19. Mihailescu M, Teo YM (2010) Dynamic resource pricing on federated clouds. In: 2010 10th
IEEE/ACM international conference on cluster, cloud and grid computing, IEEE, pp 513–518
20. Ibrahim E, El-Bahnasawy NA, Fatma A (2017) Task scheduling algorithm in cloud computing
environment based on cloud pricing models. In: IEEE 2016 world symposium on computer
applications and research, pp 65–71
Implementation of Low-Cost Mobile
Robot for Rescue Challenges

Rajesh Kannan Megalingam, Shree Rajesh Raagul Vadivel,

Prasant Kumar Yadav, Katta Nigam, Ravi Teja Geesala
and Ruthvik Chanda

Abstract One of the biggest challenges in today’s world in the field of robotics is
rescue robotics. This paper aims in the design and implementation of mobile robot
for the search and rescue operations in natural calamities such as earthquakes. These
rescue robots reduce the response time as compared to humans and help in getting
information to the rescue teams using sensors. The main issues concerned with the
present rescue robots are modularity, mobility, durability, and robustness. The robot
is designed considering all the required parameters in SOLIDWORKS CAD and
simulated in Rviz with the control interface as Robot Operating System (ROS). The
robot is designed in such a way that it can do all the mobility tasks like climbing
stairs, moving on uneven terrains, step fields, sand, and gravel, as well as exploring
tasks like finding the injured victims and hazardous signs.

Keywords Multi-terrain robot · Robot Operating System (ROS) · Graphical user

interface (GUI) · Solidworks · Tele-operated

R. K. Megalingam · S. R. R. Vadivel · P. K. Yadav · K. Nigam · R. T. Geesala (B) · R. Chanda

Department of Electronics and Communication Engineering, Amrita Vishwa Vidyapeetham,
Amritapuri, India
e-mail: ramsrimanasi@[Link]
R. K. Megalingam
e-mail: rajeshkannan@[Link]
S. R. R. Vadivel
e-mail: shreerajgul@[Link]
P. K. Yadav
e-mail: yprasant0@[Link]
K. Nigam
e-mail: 123nigam.k@[Link]
R. Chanda
e-mail: ruthvikchanda1999@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1345

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1346 R. K. Megalingam et al.

1 Introduction

The first use of the rescue robots was actually during the World Trade Center (WTC)
collapse in 2001 though research was going on this field in the past for many years.
The main goal of these robots is to reduce the number of deaths during disasters by
surveying the areas which humans are not permitted until the fire is off. Effective
communication without delay and errors between the rescue team and the robot plays
a key role in these situations. The rescue robot presented in this research has both
wired connectivity and wireless connectivity. Robot Operating System (ROS) estab-
lishes both wired and wireless connectivities through master–slave protocol between
the control system and the rescue robot. Both wired and wireless systems have its
own advantage and disadvantage in rescue operations. Wired connection provides
reliable connection and does not provide any interference as in wireless communi-
cation. The advantage of this is that it provides constant and faster speed compared
to wireless connection because of one-to-one connection. But the disadvantage of
wired connectivity is that they may not be so effective in disaster scenarios because
their reliability is not so good as compared to wireless. For example, if the cables
are damaged, the re-installation process would be difficult in those disaster areas.
This rescue robot carries several cameras, lidar, kinect, and other sensors. Several
sensors are used in this robot to know the environmental conditions of the unknown
disaster area. The sensors like IMU and encoders used are for locating or identifying
the robot motion. The robot is tele-operated from the controller station with a user
interface and a controller.
The innovative idea of the mechanical design with the flipper mechanism and
main drive mechanism is an advantage for this robot. The center of mass is well
adjusted so that it does not topple while climbing stairs and any other uneven rough
disaster terrain. The compactness and small size of the design allow the robot to enter
small voids and collect useful information where humans cannot enter. The agility of
the robot also holds an important aspect during these disaster situations where even
seconds of time is valuable.

2 Related Works

The paper [1] presents the development of crawler rescue robot with two flipper arms
which has the capability to move in all the four directions, i.e., not only up and down
(pitching motion) but also left and right (yawning motion). The flipper arm in this
robot is 2DOF. Paper [2] proposes the new design of the mobile robot which aims
to perform on different types of terrains and disaster areas in a balanced way. This
robot has the capability to run on any rocky and sandy area and to climb the stairs.
The paper [3] discusses an exploration algorithm for the rescue robots in which it
automatically maps the unknown environment while driving. Paper [4] presents the
Implementation of Low-Cost Mobile Robot for Rescue Challenges 1347

method for multi-hop communication in robots, i.e., in detail, a method by which

GUI using ROS can be constructed to operate the robot is introduced. Paper [5]
describes the machine learning techniques for quick training of robots for naviga-
tional tasks and facilitating remote operations. In paper [6], the authors described the
design and control of the four-flipper tracked robot. They even described the control
mechanisms and the multifunctionality capabilities of the robot. Paper [7] describes
the fixed path algorithm and the behavior of the autonomous wheelchair using sim-
ulation technique. In paper [8], they proposed the gesture-based wheel chair and its
unconventional methods of navigation which are simple and cost-effective. The two
methods include Line Following Navigation (LFN) and Location Aware and Remem-
bering Navigation (LARN). Paper [9] describes about the Interactive Remote Robot
Operation (IRRO) for reducing the ambiguities and abstracting them. The authors
used iterative closest points (ICP) algorithm for the results of collision detection in
the reconstructed 3D map. In paper [10], the authors developed a dummy robot for
the evaluation of the safety measures and features of the robot. With this testing
using the dummy robot, the safety concerns that exist with the humans are removed.
Paper [11] describes the study of risk assessment of the rescue robot which helps to
improve the reliability of the rescue robot. The assessment is done by considering
each subsection in the robot and then combining everything together. Paper [12] pro-
poses the configuration of sensors to be used in the robot which is used to measure the
environmental conditions of the disaster areas like CO2 sensor, temperature sensor,
and smoke particle density.

3 Architecture

Figure 1 gives the detailed process on how Robot Operating System (ROS) acts as
the main platform to control the robot. The ROS forms the base for the architecture
of our robot. Through ROS nodes and topics, the required data is exchanged. The
user can control the robot from workstation wirelessly. The control station system is
the master, and the robot is slave. The control is divided into different blocks.

3.1 User Interface

The user interface block consists of a graphical user interface. The GUI is built on
ROS plug-in RQT.
The GUI consists of two camera view blocks which give a good perception of the
robot surroundings from the cameras fixed on the robot. For establishing commu-
nication between the robot and control station, the computers should be connected
to an external router through ethernet cables. By running both the systems under
fixed ROS_MASTER_URI with an IP address, we can establish a bridge between
1348 R. K. Megalingam et al.

Fig. 1 Architectural diagram of the system

the control station and the robot. By this bridge, data communication takes place. A
customized joystick is designed for the control of robot. The user can also control
the robot through keyboard.

3.2 ROS Platform

ROS platform deals the transmission of data through nodes. Nodes are processes
that perform computation. Our robot control system usually comprises many nodes.
Nodes exchange data through ROS-defined messages. These messages are routed
via a transport system with publish/subscribe semantics through topics. We used
USB_CAM ROS package to transmit camera data. These transmitted data are seen
in GUI. And the user input values through joystick are published by JOY_NODE
in topic names CMD_VEL and FLIPPER_VAL. These transmitted messages are
subscribed by SERIAL_NODE, and by establishing serial communication between
PC and Teensy, the received data is transmitted to Teensy board. Further computation
of data takes place in Teensy board which produces the corresponding PWM values
and drive signals to the motor drivers and flipper drivers.

3.3 Hardware

One of the most important aspects of navigating mobile robots is perception. The
robot was remotely operated with the help of cameras (logitech c310), encoders, and
IMU data. We placed a camera at the front end of the robot with a rotary base made
of servo to get a 180° view, +90° to the right and −90° to the left, while the robot is
in motion. We had one camera each at the corner end, at the back placed horizontally,
Implementation of Low-Cost Mobile Robot for Rescue Challenges 1349

mounted over a moving servo base with possible rotation of 90° (45° to each side).
This arrangement covers all the possible views around all the sides of the robots,
and most importantly, we can also view the positioning of the flippers. The camera
live feed was streamed back to the operating station with the help of ROS nodes and
ROS usb_cam package. Since the bandwidth required for the data was huge, there
was a delay in the communication network so we had to compress the video data to
reduce the video latency. The video data was compressed by a node written in C++
which dropped some of the unnecessary features of the images of each frame in a
way that the video quality is maintained to a standard.
The encoder data and IMU were used to keep track of the motion and the distance
moved by the robot in the GUI. The IMU data was used to measure if we have a rolling
on a inclined surface. The encoder data also helped to maintain our differential drive
of the main drive without any error. The data was transmitted from the Teensy 3.2
through a topic/odom to the GUI node to keep a note of the sensor (IMU, encoders)
readings.
The flippers were designed to perform a movement of 360° around the center
position of the main drive motor shaft. To enable an easy control of the flippers, they
were assigned with some poses. The first pose consists of all the four flippers pointing
vertically up and 90° from the ground. This pose was used to move it quickly through
the plane surfaces since the area of contact is minimum in this pose. The second pose
was where the front two flippers were vertically inclined at an angle 45° from the
ground, and others were vertically 90° from the ground. This position was mostly
used for climbing up inclined surfaces. The third position was where all the flippers
laid horizontally to the ground making an angle of zero degree. This position was
fully stretched position of the robot.

4 Design and Implementation

Body as shown in Figs. 2 and 3 was designed on SOLIDWORKS platform with two
drive wheels and four flippers, and mild steel was the material used in manufactur-
ing the robot. Chain sprocket power transmission mechanism is followed in robot
mobility where electrical power of the motor was converted into momentum which
was essential for the body movement. The dimensions of the robot body are 60 ×

Fig. 2 Left corner view

1350 R. K. Megalingam et al.

30 × 25 cm (length × width × height); the electronics for the body are placed from
opening designed to and fro of the robot. The center of the mass of the body was
sustained with care to avoiding the toppling of the robot; two E-bike motors with
24 V input supply and 350 W power help to drive the body. The shaft of the E-bike
motors was replaced with 10-mm shaft which was further coupled with 50-mm pitch
circle sprocket (Fig. 4). Worm-geared window motors (Fig. 5) with 12 V voltage
supply were used for the flipper movement, and 12-mm shaft was passed through
these motors which is coupled to 5 mm thickness and 150-mm-long flipper plate.
Two sprockets of 100 mm pitch circle diameter are coupled through welding concen-
trically to maintain common axis of rotation around the shaft that has been passed
through the center for the flipping mechanism. One sprocket was mated with main
drive chain where the other was mated for flipper chain. Main sprocket is parallel to
two more sprockets; one of them was attached to drive sprocket which was coupled
to E-bike motor, and the other sprocket was coupled to rotary encoder. The side plate
of body which was manufactured with 5 mm MS had the PCD of the E-bike and
encoder to couple. Sprocket-to-sprocket distance for the center chain was 54 cm,
and the length of the flipper was 15 cm (distance between main sprocket and dummy
sprocket of flipper). Flipper adjusting mechanism is taken care by the flipper plate
which was designed and manufactured with the PCD for the center shaft and 50-mm
pitch circle flipper dummy sprocket axis. Lidar placement and mountings for pan and
tilt for camera were taken care during the manufacturing. 1-mm metal sheet was used
for body cover of the robot, 3 cm ground clearance for the robot, and the 380 cm ×
320 cm area for placements of components (camera, lidar) above the body.

Fig. 3 Right-top corner

view
Implementation of Low-Cost Mobile Robot for Rescue Challenges 1351

Fig. 4 Coupled sprockets

Fig. 5 Worm-geared
window motor

5 Experiments and Results

The robot was tested in an arena which resembles the real-world scenario. The
operator controls the robot using a customized joystick board. The operator controlled
wirelessly through the video visual from the cameras fixed on the robot. The design
and solid metallic body gave extra stability. To inspect the feasibility of the proposed
design, we tested the robot under different terrain scenarios, A. 25° ramps, B. uneven
terrain, C. 45° steep terrain, and D. parallel rail bars.
1352 R. K. Megalingam et al.

Fig. 6 Robot tested on rail

bars

5.1 25° Ramps

This test scenario contains 25° continuous ramps. According to the tested results,
the robot was able to drive freely as shown in Fig. 6. The flipper mechanism gave
extra support to modify the pose to the altered terrain environments.

5.2 Uneven Terrains

This test scenario consists of varied angles and dimensions of ramps on the trot
continuously. Despite these harsh terrains, the robot drives freely.

5.3 45° Steep Terrain

The arena consists of a 45° steep track. With the high torque E-bike motor and the
flipper mechanism, it will be to go up the 45° steep track.

5.4 Parallel Rail Bars

This test consists of two parallel rail bars of breath equal to the dimension of the
main track width of the robot. With the PID control, speed parameter of the robot
is controlled by which we were able to drive the robot on the parallel rails without
deviating its path as shown in Fig. 7.
These scenarios will test the robustness and flexibility of the robot in harsh
environments.
Implementation of Low-Cost Mobile Robot for Rescue Challenges 1353

Fig. 7 Robot tested on

ramps

6 Future Works

Robot flipper mechanism needs to be modified so it adds more stability to reach

steepy terrains too. Gearbox should be redesigned to meet the required torque for
flipper. The drive chain mechanism needs to be modified since axial motion of the
main sprockets causing the chain to slip from the sprocket. This problem can be
restrained by installing cross-roller bearing mechanism. Efficiency of power trans-
mission through sprocket chain mechanism is way less than direct coupling of the
motor to sprocket so it is necessary to avoid chain drive mechanism. Chassis mate-
rial needs to be fabricated with aluminum to diminish the weight issues. Track belt
needs to be installed with nylon pulley instead of direct chain and sprocket since
robot is lacking friction in few situations. This also reduces the weight of the robot.
As the robot runs many modules, processor speed of the microprocessor needs to
be improved to decrease delay in transmission of camera footage. Communication
range of the robot also needs to be improved for extending the exploration limits
of the robot, needs to extend research on autonomous navigation which maps the
surrounding and navigate itself, and needs to include image-processing module for
detecting hazard signs and victims.

7 Conclusion

Although there is an enormous amount of research in the field of search and inspec-
tion robotics, this paper provides the working and design of a low-cost mobile robot
which can be used for search and inspection in disaster-hit areas. The mobility of the
1354 R. K. Megalingam et al.

robot through rough terrains or through debris left in the disaster hit areas is impact-
ful and considerable because of its flipper mechanism. The body design developed
is very suitable for disaster sites because of its rigidity and durability. The power
consumption of the robot is also less compared to other robots used for exploration
purposes. The paper also acknowledges the fact that during disaster situation, we
need a reliable communication network between the operator and the robot. Thus,
the Ubiquiti antenna deployed near the controller station provides strong and reliable
connection between the master and slave interface. The Wi-Fi adapter used is config-
ured with an operating bandwidth of 5 GHz so that we do face a transmission delay
in the network. We also focus on the setup time for the robot which is considerably
small in our case. Setup time needs to be very small for these rescue robots.

Acknowledgements We thank Amrita University and Humanitarian laboratory for providing us

this wonderful opportunity to direct our thoughts into a real prototype. We thank Dr. Rajesh Kannan
Megalingam for providing us space and infrastructure needed to complete this project. We thank
all the people who have helped us to accomplish this project in time.

References

1. Sato N, Torii K, Morita Y (2013) Development of crawler type rescue robot with 2 DOF flipper
arms. In: IEEE/SICE international symposium on system integration
2. Gupta VK, Gupta AK (2013) Design and development of six-wheeled multi-terrain robot. In:
International conference on control, automation, robotics and embedded systems (CARE)
3. Wirth S, Pellenz J (2007) Exploration transform: a stable exploring algorithm for robots in
rescue environments. In: IEEE international workshop on safety, security and rescue robotics
4. Shin S, Yoon D, Song H, Kim B, Han J (2017) Communication system of a segmented rescue
robot utilizing socket programming and ROS. In: 14th international conference on ubiquitous
robots and ambient intelligence (URAI)
5. Moridian B, Kamal A, Mahmoudian N (2018) Learning navigation tasks from demonstration
for semi-autonomous remote operation of mobile robots. In: IEEE international symposium
on safety, security, and rescue robotics (SSRR)
6. Kovačić Z, Cukon M, Brkić K, Vasiljević G, Mutka A, Miklić D, Vuglec F, Rajković I Design
and control of a four-flipper tracked exploration and inspection robot. In: 21st mediterranean
conference on control and automation
7. Megalingam RK, Vishnu GB, Pillai M (2015) Development of intelligent wheelchair simulator
for indoor navigation simulation and analysis. In: IEEE international women in Engineering
(WIE) conference on electrical and computer engineering 2015 (WIECON-ECE 2015), BUET,
Bangladesh
8. Megalingam RK, Manoj PS, Nammily NR (2011) Unconventional indoor navigation: ges-
ture based wheelchair control. In: International conference on indoor positioning and indoor
navigation (IPIN’2011), Guimarães, Portugal
9. Shin YD, Park JH, Jang GR, Yoon JS, Baeg MH (2013) Interactive remote robot operation
framework for rescue robot. In: IEEE International symposium on safety, security, and rescue
robotics (SSRR)
10. Nallathambi DJ (2018) Comprehensive evaluation of the performance of rescue robots using
victim robots. In: 4th international conference on control, automation and robotics (ICCAR)
Implementation of Low-Cost Mobile Robot for Rescue Challenges 1355

11. Phuengsuk R, Suthakorn J (2016) A study on risk assessment for improving reliability of rescue
robots. In: IEEE international conference on robotics and biomimetics (ROBIO)
12. Ferreira NL, Couceiro MS, Araújo A, Rocha RP (2013) Multi-sensor fusion and classification
with mobile robots for situation awareness in urban search and rescue using ROS. In: IEEE
international symposium on safety, security, and rescue robotics (SSRR)
Smog Detection and Pollution Alert
System Using Wireless Sensor Network

Manita Rajput, Mrudula Geddam, Priyanka Jondhale and Priyanka Sawant

Abstract Air pollution is a primary concern for humankind, causing asthma attacks,
wheezing, shortness in breath, etc. In the Indian subcontinent, air quality decreases,
increasing smog due to the burning of crops to prepare for the harvest and emis-
sions from the vehicles, etc. Driving becomes difficult due to this blinding smog.
To overcome these problems, a wireless network of sensors is proposed along the
expressways and in public places to measure the smog. A personal area network
(PAN) network of motes is set up. Using MQ135 and DSM501a sensors, smog
pollutants are measured. Contiki OS and Cooja simulator are used to analyze the
performance of network by emulating nodes using Tmote Sky. The nodes in this
PAN are mounted over the lamp posts on the bridge to collect information of various
pollutants and send the acquired data to a mote. This mote acts as a central node in
the wireless sensor network (WSN) and transmits data over user datagram protocol
(UDP). Finally, the parameters with warning messages are displayed at the entrance
of bridges for the drivers to create a safety alert. The future aspect of this project
may lead to displaying the messages at short intervals along the whole bridge with
updated values. Also, the recorded data can be stored on the server and after a certain
period of time, the data can be analyzed to observe the number of pollutants being
emitted, and measures can be taken accordingly.

Keywords Smog · TmoteSky TelosB motes · Cooja simulator · Contiki OS · AQI

M. Rajput · M. Geddam · P. Jondhale (B) · P. Sawant

Department of Electronics and Telecommunication Engineering, Fr. C.R.I.T, Vashi, Navi,
Mumbai, India
e-mail: priyankajondhale12@[Link]
M. Rajput
e-mail: [Link]@[Link]
M. Geddam
e-mail: mrudula17geddam@[Link]
P. Sawant
e-mail: piyasawant10@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1357

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1358 M. Rajput et al.

1 Introduction

Smog is a form of air pollution that mainly consists of ozone, along with harmful
substances like sulfur dioxide, nitrogen dioxide, carbon monoxide, and PM10s, which
can find their way deep into the lungs. Smog is a harmful mixture of fog, dust and air
pollutants such as nitrogen oxides and volatile organic compounds which combine
with sunlight to form a dense layer of ground-level ozone. Ozone when present high
in the atmosphere is beneficial, but when nearer to the ground, it can cause irritating
health effects. The name was constructed by putting together the words ‘smoke’ and
‘fog’ [1]. Smog can be caused by
• Large amounts of coal burning in an area.
• Slash-and-burning of crops (a major source in Delhi).
• Smog-forming pollutants generated from automobile exhausts, power plants,
fireworks, even paint, hair spray, charcoal starter fluid, and plastic popcorn
packaging.
• The formation of smog is also closely linked with temperature, sunshine, and calm
winds [2]. The most affected people by smog are children, infants, the elderly,
those with cardiac and respiratory problems like asthma, emphysema, chronic
bronchitis, those who stay outdoors for long periods of time, and people with
unusual susceptibility to ozone. Smog can cause minor issues like eye and throat
irritation, headaches. But when exposed to it over a long period of time, smog can
have a much worse effect on the body [3]. The basic structure of a WSN [4] is
shown in Fig. 1. It shows a cluster node to which many sensors are connected.
The data fetched by sensors is collected by the cluster node. It is then sent to the
sink node which can be a mote, Raspberry Pi, or an Arduino, etc. This sink node
collects, packetizes, and sends the data to the monitoring system (microprocessor).
It processes the data packets and displays them for the specified application.
In order to achieve the aim of the project, a detailed block diagram of the pro-
posed solution is shown in Fig. 1b which shows the actual interconnection of various
components. Motes collect and transfer data using four stages: collecting the data,
processing the data, packaging the data, and communicating the data. Each mote

Fig. 1 a Basic structure of a WSN [5]. b Implemented block diagram of system

Smog Detection and Pollution Alert System Using Wireless Sensor … 1359

collects data using various types of sensors connected to it. A network of sensors
which can measure the smog contents is proposed for the measurement of smog and
alerts the crowd along the expressways and public places.

2 Motivation

In India and its neighboring countries, air quality decreases just about every year
at the onset of winter as farmers begin burning their crops to prepare for their new
harvest and emissions from the vehicles, etc. The air quality index in Delhi—which
measures the concentration of toxic particulate matter in the air—had shot up to 451
in the year 2017, as compared to the ideal value of 100. This led to the motivation to
think about developing a system hoping to address this problem at an affordable price
and alert the drivers and pedestrians about the current environmental conditions.

3 Literature Review

3.1 Present Smog Detection Devices in India

Many devices are available in the market presently, to measure the level of smog.
Some of them were studied by us. Described below are the present smog detection
devices available in the market. A brief description of them and drawbacks, as to
why they cannot be used effectively, accompanies them.
(1) Air Pollution and Fog Detection through Vehicular Sensors: This paper demon-
strates that LIDAR technology, already onboard for the purpose of autonomous
driving, can be used to improve the weather condition recognition when com-
pared with a camera-only system. System has the combination of a front camera
and a LIDAR laser scanner used as a sensor instrument set for air pollution and
fog recognition [6].
(2) Air Pollution Detection Based On Head Selection Clustering And Average
Method From Wireless Sensor Network: The proposed method mainly focus on
longer sustain time period of sensor network, effective processing of collected
information, and less overhead in routing information between sensor nodes
[7].
(3) Smart Environment Monitoring Beacon-Cluj Napoca University, Romania: The
purpose of Smart Environment Monitoring Beacon is to present a beacon cre-
ated for monitoring the environmental conditions, like weather parameters, air
1360 M. Rajput et al.

pollution, sound levels, and UV radiation index using GSM module [8].
(4) 6LoWPAN Network Using Contiki Operating System: This paper focuses on
the working of 6LoWPAN network using Contiki tool with Cooja simulator and
various WSN’s simulation tools [9].
(5) Wireless Sensor Network-Based Pollution Monitoring System in Metropolitan
Cities [10]: This study proposes air pollution and monitoring model which
detects pollution in the air on the basis of data mining algorithm [11]. Bluetooth
module is used to connect the controller with client, and the client connects with
the server via web services. Wireless sensors are used to the amount of gases
[12].
(6) Urban Air Pollution Monitoring System with Forecasting Models: A system for
monitoring and forecasting urban air pollution is presented in this paper. The
system uses low-cost air quality monitoring motes that are equipped with an
array of gaseous and meteorological sensors [13].
(7) Into the SMOG: Stepping Stone to Centralized WSN Control: This paper illus-
trates that wireless sensor networks (WSNs) can lead to improve network life-
time, benefit reliability, help to diagnose and localize network failures. A com-
plete network topology model that scales and reacts to the network dynamics
that occur in low-power wireless networks is proposed [14].

3.2 Census Data

According to the World Health Organization (WHO) global air pollution database
released in Geneva, India has 14 out of 15 most polluted cities in the world—Kanpur
being the worst with a PM 2.5 concentration of 173 µg/m3 . Soot, dust, ozone, and
sulfur oxides are a growing threat for billions of people around the world. A whopping
nine in ten people on Earth breathe highly polluted air, and more than 80% of urban
dwellers have to endure outdoor pollution that exceeds health standards, according
to the WHO’s World Global Ambient Air Quality Database. The measurements and
calculations as of 2016 revealed that 11 of the 12 cities with the highest levels are
located in India. Figure 2 shows Kanpur, India, population 3 million, tops the list with
a yearly average of 319 µg/m3 of PM 2.5, the most hazardous particle commonly
measured [15].
Smog Detection and Pollution Alert System Using Wireless Sensor … 1361

Fig. 2 Air pollution in Delhi, India, with PM 2.5 calculation, Survey 2016 [15]

4 Software Implementation Using Cooja

4.1 Simulated Results of InBuilt Sensors

The Tmote Sky MTM-CM5000 consists of inbuilt temperature, light, and humidity
sensors. SHT-11 sensor measures both humidity and temperature. Initially, the sen-
sors are activated, and then the current temperature and humidity values are fetched.
The obtained values are then displayed on an LCD display.
Using the above flow process shown in Fig. 3, we implemented the software model
for measuring temperature and humidity by setting up two nodes in Cooja software,
and the obtained results are shown in Fig. 4a, b.

4.2 Simulated Results of External Sensor Interfacing

For measuring levels of PM10 and CO2 pollutants, external sensors, namely
DSM501a and MQ135 are interfaced with the mote. These sensors are powered
using the VCC and GND pins of motes itself. The sensor senses the analog values
and gives it to the mote via ADC 0/1 pin. The acquired measures are voltage values,
are later converted to ppm or µg/m3 , and are then displayed as the sensor output.
The values are updated after a particular delay provided in the code. Figure 5 shows
the flowchart for interfacing external sensors with the motes. It illustrates the port
configuration and the way to proceed with this interfacing. The software model for
the same and obtained results are given in Fig. 6a, b where CO2 pollutant is measured
by implementing a software representation.
1362 M. Rajput et al.

Fig. 3 Flowchart for temperature and humidity measurement using inbuilt sensors

Fig. 4 a Setting two nodes in Cooja to check temperature and humidity. b Obtained node output
showing ideal temperature and light values

In Figure 7a, b, PM10 pollutant’s software simulation is done using Cooja

simulator.
Smog Detection and Pollution Alert System Using Wireless Sensor … 1363

Fig. 5 Flowchart for measuring pollutants values from external sensors (MQ135 and DSM501a)

Fig. 6 a Setting up a node in Cooja for measuring CO2 . b Node output showing CO2 value in ppm

4.3 Creation of a PAN Network in Cooja

After obtaining the measured values, they are required to be transmitted to the central
mote in order to get displayed. For this data transmission, a PAN network of two
or more motes is created, and using unicast or multicast routing, the obtained data
packets are routed in the network. In this case, a network of only two motes is
1364 M. Rajput et al.

Fig. 7 a Setting up a node in Cooja for measuring PM10 pollutant value. b Node output showing
PM10 value in µg/m3

created, and so unicasting is used. Figure 8a shows the flowchart for creating a
virtual simulation of motes and sensors in PAN configuration. In this configuration,
all the pollutant values are obtained simultaneously.
The two motes work in client–server configuration, their PAN arrangement is done
in VMware software, and the obtained simulated results are shown in Fig. 9a, b.

4.4 Hardware Implementation with Motes and Sensors

Taking the analog output from the sensor nodes, we got the following results. We
considered various real-life inputs that may be received by the sensors under different
weather conditions. Mote1 is working as a client while Mote2 is working as a server;
their individual transmissions and PAN configuration outputs with internal as well
as external sensors are provided in Fig. 10.
Figure 10a depicts the function of one of the motes as a client sensing and trans-
mitting data packet to the mote at other end while Fig. 10b shows another mote
working as a server receiving all the data packets from the client and routing them
in the PAN configuration.
Figure 10c shows the MQ135 sensor’s measured CO2 output when interfaced
with the mote. Its obtained results are shown in ppm while Fig. 10d gives DSM501a
sensor’s real-time measured PM10 pollutant value in µg/m3 . This arrangement itself
gave fairly reliable output and creates a PAN configuration with two motes in it, one
sending data packets and the other receiving it making a client–server configuration.
These sensors send their data to mote via the ADC pin available on MTM-5000 mote
and are processed with the help of MSP430 microprocessor’s timers and counters
utilities.
We have actually implemented the real-time PAN configuration of two Tmotes
and sensors powered using USB slots of laptop. Figure 10a shows the real-time
hardware implementation of the same. In this implementation, Mote1 acts as a client
and has MQ135 and DSM501a sensors interfaced with it. Sensors collect the data
and give data packets to client mote. These packets are then processed by onboard
Smog Detection and Pollution Alert System Using Wireless Sensor … 1365

Fig. 8 Flowchart for creating a PAN consisting of motes and sensors

Fig. 9 a Setting up two motes in PAN configuration using Cooja software. b Data transmission
and reception between both the motes in Cooja output window
1366 M. Rajput et al.

(a) Mote1 functioning as Client (b) Mote2 as Server/PAN configuration

of motes

(c) MQ135 sensor output measuring CO2 (d) DSM501a sensor output measuring
PM10 value

Fig. 10 Hardware sensor and mote’s output in VMware workstation terminal window

MSP430 microcontroller and sent to the server Mote2 at the other end over UDP.
The acquired data is then displayed.
Figure 10b shows the output window of client mote which sends the data packets
to the server mote. It first fetches the server data, packetizes them, and then routes the
packets over UDP. Figure 10c describes the operation of server mote which receives
the data packets sent by client mote and displays them on the terminal window as
shown in Fig. 11.

5 Conclusion

Overall, the development for this project has been discussed covering hardware
design and software development. This project demonstrated the possibility of imple-
menting a system that will help in monitoring the pollutant level in the atmosphere,
thus creating a social alert. The project is developed especially keeping in mind
the vehicle accident prevention application, and hence, emphasizes the creation of
a driver alert system with system information being displayed on the big displays
along highways. The idea to implement the system on street light poles has made it
cost-effective by cutting off the installation costs, and also the usage of motes has
incorporated the use of wireless communication for data transmission and increased
Smog Detection and Pollution Alert System Using Wireless Sensor … 1367

Fig. 11 a Actual implementation of PAN configuration with motes and sensors

node connectivity limit. Powering motes using batteries leads to increased efficiency,
less power consumption, and long system life.

References

1. Corbalan P, Marfievici R, Cionca V, O’Shea D, Pesch D (2016) Into the SMOG: the stepping
stone to centralized WSN control. In: 2016 IEEE 13th international conference on mobile ad
hoc and sensor systems (MASS)
2. Kampa M, Castanas E (2007) Human health effects of air pollution–laboratory of experimental
endocrinology. University of Crete, School of medicine, Heraklion, Greece, 10 June 2007
3. Online: Effects of smog–India today available: [Link]
4. Online: [Link]
5. Genc S (2012) The adverse effects of air pollution on the nervous system. J Toxicol 23. Article
ID 782462
6. Sallis P, Dannheim C, Icking C, Maeder M (2015) Air pollution and fog detection through
vehicular sensors [16]. In: IEEE Xplore conference on 06 April
7. Goel AK, Ray S, Agarwal P, Chandra N (2012) Air pollution detection based on head selection
clustering and averaging method from wireless sensor network. In: IEEE Xplore 2012 2nd inter-
national conference on advanced computing and communication technologies (SICACCT), 7–8
Jan 2012
8. Alexandru P, Andrei M, Cristina-Madalina S, Stan O (2018) Smart environment monitoring
beacon. Cluj Napoca University, Romania—IEEE international conference on automation,
quality and testing, robotics (AQTR)—05 July 2018
1368 M. Rajput et al.

9. Kavyashree ED (2018) 6LoWPAN network using contiki operating system. In: 3rd national
conference on image processing, computing, communication, networking and data analytics
(NCICCNDA 2018), 28 April 2018
10. Ayele TW, Mehta R (2018) Air pollution monitoring and prediction using IoT. In: 2018 sec-
ond international conference on inventive communication and computational technologies
(ICICCT)
11. Raipure S, Mehetre D (2015) Wireless sensor network based pollution monitoring system
in metropolitan cities. In: International conference on communications and signal processing
(ICCSP)
12. Raipure S, Mehetre D (2015) Wireless sensor network based pollution monitoring system in
metropolitan cities. In: IEEE international conference on communications and signal processing
(ICCSP) November 2015, Melmaruvathur, India
13. Shaban KB, Kadri A, Rezk E (2016) Urban air pollution monitoring system with forecasting
models. In: IEEE xplore IEEE sensors journal conference on 15 April 2016
14. Corbalan P, Marfievici R, Cionca V, O’Shea D, Pesch D (2016) Into the SMOG: the stepping
stone to centralized WSN control. In: IEEE 13th international conference on mobile ad hoc
and sensor system
15. Online: The times of india available: [Link]
world-in-bad-air-quality-kanpur-delhi-among-top-15-mumbai-4th-most-polluted-megacity/
articleshow/[Link]
IoT-Based Women Security System

Megalingam Rajesh Kannan, K. Jyothsna, T. S. Aparna, T. Anjali, M. Meera

and S. D. Amrutha

Abstract In the current global scenario, the prime question in every woman’s mind
is about her safety and security. The only thought haunting every women is when they
will be able to move freely on the streets even in odd hours without worrying about
their security. This work suggests a new perspective to use technology to protect
women. The wearable system resembles a normal watch with a button. Women can
press the button when they feel discomfort and activate the system. The system
can also be activated by changes in sensor setup output which is part of the system.
When activated, the system tracks the location of the woman using Global Positioning
System (GPS) sensor and sends an emergency email to the person who can help or
save her. The system also incorporates a screaming alarm that uses real-time clock, to
call out for help. The main advantage of this system is that the user does not require
a smartphone unlike other applications that have been developed earlier. The use
of sophisticated components ensures accuracy of the system and makes it reliable.
Uneven terrains, step fields, sand and gravel, as well as exploring tasks like finding
the injured victims and hazardous signs.

Keywords IoT · Women security · GPS · Email alert

M. Rajesh Kannan · K. Jyothsna (B) · T. S. Aparna · T. Anjali · M. Meera · S. D. Amrutha

Department of Electronics and Communication Engineering, Amrita Vishwa Vidyapeetham,
Amritapuri, India
e-mail: kjyothsnareddy117@[Link]
M. Rajesh Kannan
e-mail: rajeshkannan@[Link]
T. S. Aparna
e-mail: aparnats90@[Link]
T. Anjali
e-mail: anjalitsuresh98@[Link]
M. Meera
e-mail: meeramurali3498@[Link]
S. D. Amrutha
e-mail: amruthasdevan17@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1369

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1370 M. Rajesh Kannan et al.

1 Introduction

In this research work, we are developing a women security watch for helping woman
in distress by using IoT. Using this device, women can send the details of location,
an emergency mail and switch on the screaming alarm by simply pressing a button
in the watch. Or the system can also be automatically activated by observing the
variations in the sensors like the temperature sensor and heartbeat sensor. Even in
the twenty-first century, women cannot step out of their house in night time as they
cannot be assured of their physical safety. Now a days, the prime question in every
girls mind, considering the ever rising increase of issues on women harassment in
recent past, is mostly about her safety and security. This research work suggests a
new perspective to use technology for women safety. Approximately, 848 Indian
women are harassed, raped or killed every day. That is a way beyond huge number.
We propose an idea which changes the way everyone thinks about women safety.

2 Motivation

An undeniable reality that has not changed and is still prevailing, not only in our
country but all around the world, is the safety of women. Whether at home, working
place or outside the home, safety of women matters a lot. Its a sad truth that every
minute and every second some women, let it be mother, sister, wife, young girls,
infants are getting harassed, assaulted and molested at various places all over the
world. Even though there are many laws and worldwide organisations working for
the welfare of women, the number of unlawful or illegal acts against women is rapidly
increasing day by day. The fact is that there are cases which are not even filed or
reported. Now, the situation has reached to a stage where parents are scared to allow
their daughters to step out of the house. This has inherited a fear in every girl’s mind
and decreased the confidence level. This has become a matter requiring a resolution.
Thus, it is indeed a need to call for a security system for the safety of women.

3 Related Works

A smart watch for women security based on IoT concept ‘watch me’ is proposed
in [1]. It works automatically based on the heartbeat rate which triggers the sensor,
produces a high pitch alarm sound and sends an alert signal to the nearby police
station. The Personal Stun-A Smart device for women safety, when activated, sends
GPS location, pulse rate and body temperature of the person to the ice contacts
and police control room. The band which is part of this device when thrown with a
force, the force sensor gets activated and sends the GPS location [2]. An innovative
approach for women and children security-based location tracking system using
IoT-Based Women Security System 1371

GPS and GSM is presented in [3]. Paper [4] proposes a system which sends message
along with location to the family members, nearest police station and to people in
near vicinity. In [5], the authors propose a tracking system that receives GPS signals,
processes and transmits data to the tracking center.
A novel IoT access architecture for vehicle monitoring system is proposed in [6].
Even though this is not relevant to the problem we address, we can learn about the
IoT and sensor integration. Paper [7] presents the design and implementation of an
Ethernet-based Smart Home intelligent system for monitoring the electrical energy
consumption based upon the real-time tracking of the devices at home. In paper [8],
the authors present a system which gathers sensor data, translates, sends data to the
server with 3G/4G network, stores and retrieves sensor data using IoT. An IoT smart
home architecture for long-term care of people with special needs is presented in [9].
Paper [10] presents a system which sends alerts to the user about the health risks if
any, ensures better utilization of energy and resources and sends an email notification
to the user account if higher light intensity is detected in the room. We see that papers
[6–10] are not related to women security but used IoT technology for the intended
cause.
Paper [11] proposes a smart IoT-based agriculture system to control CO2 , soil
moisture, temperature and light, based on the soil moisture. The objective is to
increase the yield and to provide organic farming. A vehicle tracking system using
GPS is proposed in [12]. Human tracking in certain outdoor and indoor areas by
combining the use of RFID and GPS is proposed in [13]. A real-time GPS-based
tracking system based on GSM mobile phone which tracks the location of the vehicle
and its speed based on mobile phone is presented in [14]. A system that is used for
parenting purpose to monitor the movement of children outside the home is proposed
in [15]. In all these research papers, we see that either IoT or GPS or both technologies
are used to achieve the goal. In our system, we propose to use IoT, GPS and Wi-Fi
technologies to achieve our goal.

4 Design and Implementation

The architecture diagram of the system we designed and implemented is shown in

Fig. 1. This section describes about each of the blocks in the architecture diagram.

4.1 Design Units

Sensor Unit. The sensor unit consists of a heart rate sensor to sense and monitor the
heart rate of the user wearing the device. The heart rate signals are passed on to the
Microcontroller Unit (MCU) to process and take a decision.
MCU. Microcontroller Unit is Arduino uno with Atmega328 microcontroller.
Even though it is not the heart of the system, it can take some good decisions as to
1372 M. Rajesh Kannan et al.

Fig. 1 Architecture diagram of the propose women safety system

find out irregular heart rates and report to the Microprocessor Unit (MCU). It works
according to the program stored in its read-only memory (ROM). The Atmega328 is
a 28 pin microcontroller.
MPU and Wi-Fi I/F. The microprocessor unit (MPU) is the heart of the system.
It interacts with the MCU, Global Positioning System (GPS), Wi-Fi, email, cloud
storage and emergency switch. This is the central controller and the algorithm to
interact with all the above-mentioned modules are programmed into the MPU. The
Raspberry Pi 2 board is used as the MPU. The programming is done in Python. The
Raspberry Pi 2 board uses a powerful ARM Cortex-A7-based quad-core processor
which runs at 900 MHz. The Wi-Fi I/F, i.e., the wireless LAN is part of the MPU which
is supported by the Raspberry Pi 2. Using the wireless interface, the user can connect
the proposed wearable device to the Internet via tethering with the smartphone Wi-Fi
interface. By this way, the device gets continuous access to Internet thereby enabling
the access to cloud storage.
GPS Module and USB TTL. The GPS module is connected to the MPU via USB
TTL. USB TTL is used as level shifter to translate the 5 V signaling of USB signals
to the TTL level and vice versa. The GPS tracks the location (latitude and longitude)
of the user wearing the device. The MPU gets the location information of the user
and passes it to the cloud. If the user is in distress (either abnormal heart rate or the
emergency switch is pressed), the cloud will immediately send an emergency email
to the relative or the caretaker of the user.
Emergency Switch. The user in distress, when presses the emergency switch
which is directly connected to the MPU, the MPU will initiate the process of sending
an emergency email to the relative or the caretaker of the user without any delay. The
MPU gets the location of the user from the GPS data and sends to the cloud which
will then initiate an email.
IoT-Based Women Security System 1373

Cloud Storage (ThingSpeak). We used ThingSpeak for our project which is free
to use for research purposes. For IoT environment, ThingSpeak enables communi-
cation capabilities. It is very helpful to design and build applications based on the
data obtained from the sensors. Eight fields of data can be stored by each channel,
and it allows to use 255 alphanumeric characters. The best thing is that it offers four
dedicated fields for positional data which includes latitude, longitude, description
and elevation. The data that is received at the ThingSpeak end and is time stamped
with a sequential ID. For our purpose, we need to create one channel to pump our
sensor data. Once the channel is created, ThingSpeak allows to publish the data with
an API. It also creates a ‘write key’ for authentication.
Email. When the person or the user is in distress, it causes increase in heart rate,
and the MPU can process this data and decide to send an emergency email to the
user’s relative. The latitude and the longitude coordinates of the user’s location are
collected via GPS and passed to ThingSpeak which in turn notified the user’s relative.
Alternatively, if the user presses the emergency switch, the MPU does repeat the same
process to email the relative of the user, for help.
Screaming Alarm. The screaming alarm gets triggered whenever the user in
distress presses the emergency switch or abnormal heart rate is detected by the MPU.
Along with sending the email to the relative of the user, the screaming alarm will
also be triggered so that nearby people can hear and notice and come to the rescue
of the user in distress.

4.2 Functional Flow of the Proposed System

Figure 2 shows the work flow for the proposed security system for women. During
a critical condition or accident, the user’s heart rate increases. The sensor unit is
continuously monitored by the MCU, and the sensor data is processed. The heart
rate is determined, and this too is a continuous process.
The MCU passes the sensor data to the MPU. The Raspberry Pi-based MPU
processes this data and can take a decision to contact the ThingSpeak if the heart rate
is abnormal. A serial communication port is used between the MCU and the MPU
to pass the information. Alternatively, if the user can press the emergency switch
under stressful condition or distress, the MPU repeats the same workflow as when
the heart rate increases. The heart rate condition is detected automatically by the
system without the need for user intervention. The emergency switch press requires
human intervention. Both modes of operation of the devices enhance the user-friendly
nature of the device. Once the MPU decides that an email notification has to be sent, it
collects the GPS data from the GPS module via USB TTL and passes this information
to the IoT-based ThingSpeak. The ThingSpeak can then send an emergency email
notification to the relative or any other person that was predetermined by the user and
registered with the ThingSpeak. Along with sending email to the user, the screaming
alarm will also be triggered.
1374 M. Rajesh Kannan et al.

Fig. 2 Work flow diagram

4.3 MPU Algorithm

The algorithm that is implemented in Python programming and used in the Raspberry
Pi MPU is shown as flowchart in Fig. 3. The program waits for trigger from either
the emergency switch or the sensor unit which uses the heart rate sensor. Once
triggered from either of these sources, the program immediately sounds the screaming
along. Simultaneously, it gets the latitude and longitudinal values from the GPS via
USBttl and passes these values to the IoT-based ThingSpeak. An email is immediately
sent to the relative of the user. For this entire system to work, the MPU should be
continuously connected to the Internet via mobile phone or smartphone. The email
ID of the relative of the user is stored in advance in ThingSpeak.
Figure 4 shows the implementation of the system for women security. The figure
clearly shows the MPU with Wi-Fi I/F, MCU, GPS, USBttl, screaming alarm and
sensor unit.
IoT-Based Women Security System 1375

Fig. 3 Flowchart for the Python program implemented in Raspberry Pi

5 Experiment and Results

5.1 Experimental Setup

Seven users were chosen to test the device. Each of the users was given the device and
asked to press the emergency switch from six different locations near to our Amrita
Vishwa Vidyapeetham University (AVVU). Two users from Vallikavu near our uni-
versity, one user at Mata Amritanandamayi Math, one user at Mata Amritanandamayi
Math (MAM) bajan hall, another user at Mata Amritanandamayi Math hospital, one
user at Amrita Vishwa Vidyapeetham university main gate and the last user at Amrita
Vishwa Vidyapeetham university canteen pressed the emergency switch.
1376 M. Rajesh Kannan et al.

Fig. 4 Women security device

5.2 Test Results

Table 1 shows the latitude and longitude values of the each of the seven users along
with their locations. While the second column in the table lists the latitude and
longitudinal values obtained through GPS, the third column lists the latitude and
longitudinal values obtained through Google map for the same locations. The last

Table 1 System tests carried out at various locations

Location GPS receiver results Actual location Deviation (m)
Latitude Longitude Latitude Longitude
Vallickavu 9.0940° N 76.4915° E 9.0960° N 76.4926° E 1.199
Vallickavu 9.0938° N 76.4915° E 9.0950° N 76.4925° E 2.028
MAM 9.0872° N 76.4915° E 9.0892° N 76.4871° E 2.462
MAM bajan hall 9.0895° N 76.4915° E 9.0897° N 76.4859° E 0.562
MAM hospital 9.0889° N 76.4915° E 9.0899° N 76.4849° E 2.521
AVVU main gate 9.0936° N 76.4915° E 9.0938° N 76.4917° E 0.6916
AVVU canteen 9.0932° N 76.4915° E 9.0939° N 76.4919° E 0.6908
IoT-Based Women Security System 1377

Fig. 5 Sample email details sent to the relative

column lists the error in meters, i.e., the latitude and longitudinal values obtained
through the GPS fitted with our system and the Google map. We can see that we
were able to achieve the error as minimum as 0.5 m, and hence, can conclude that
the proposed system could accurately pinpoint the location of the user in distress.
A sample email that was sent to the relative, when one of the users pressed the
emergency button of the proposed women safety system at the location Vallikavu,
is given in Fig. 5. In these emails, we can see the request for ‘HELP’ from the user,
the latitude and the longitude values and the date and time at which the request was
generated. These emails were sent using the IoT-based ThingSpeak.

6 Conclusion

This project analyzes the emergency response system which guards and helps the
women to save herself when she is in danger. The main objective is to model a low-
cost system which can accumulate the data of the lady and provide urgent alert in case
of emergency. Our effort behind this project is to construct and formulate a gadget
that acts like a compact weapon. It is certainly a precise and protective methodology
which will probably be very useful for the women. The establishment of a hardware
and software prototype has accomplished two objectives: affirmation of the design
and verifying whether the idolized technology is suitable for the system. The device
is easy to handle, highly responsive and assures safety to the user.

Acknowledgements We thank Amrita Vishwa Vidyapeetham University and Humanitarian Tech-

nology (HuT) labs for providing us this wonderful opportunity to direct our thoughts into a real
prototype. We thank all the people who have helped us to accomplish this project in time.
1378 M. Rajesh Kannan et al.

References

1. Helen A, Fathila MF, Rijwana R, Kalaiselvi VKG (2017) Watch me. In: 2017 second
international conference on computing and communication technologies (ICCCT), Chennai,
India
2. Ahir S, Kapadia S, Chauhan J, Sanghavi N (2018) The personal stun—A smart device for
women’s safety. In: 2018 international conference on smart city and emerging technology
(ICSEET), Mumbai, India
3. Velayutham R, Sabari M, Rajeswari MS (2016) An innovative approach for women and children
security based location tracking system. In: 2016 international conference on circuit, power
and computing technologies (ICCPCT), Nagarcoil, India
4. Harikiran GC, Menasinkai K, Shirol S (2016) Smart security solution for women based on
IoT. In: 2016 international conference on electrical, electronics and optimization techniques
(ICEEOT), Chennai, India
5. Dafallah HAA (2014) Design and implementation of an accurate real time GPS tracking sys-
tem. In: 2014 third international conference on e-technologies and networks for development
(ICeND), Beirut, Lebanon
6. Wang S, Hou Y, Gao F, Ji X (2016) Novel IoT access architecture for vehicle monitoring
system. In: 2016 IEEE 3rd world forum on internet of things (WF-IoT), Reston, VA, USA
7. Gupta P, Chhabra J (2016) International conference on innovation and challenges in cyber
security (ICICCSINBUSH), Noida, India
8. Suyama T, Kishine Y, Naya F (2014) Abstracting IoT devices using virtual machine for wireless
sensor nodes. In: 2014 IEEE world forum on internet of things (WF-IoT)
9. Coelho C, Coelho D, Wolf M (2015) An Iot smart home architecture for long-term care of
people with special needs. In: 2015 IEEE, 2nd world forum on internet of things (WF-IoT),
Milan, Italy
10. Malche T, Maheshwari P (2017) Internet of things (IoT) for building smart home system.
In: 2017 international conference on I-SNAC (IoT in Social, Mobile, Analytics and Cloud),
Palladom, India
11. Pallavi S, Mallapur JD, Bendigeri KY (2017) Remote sensing and controlling of greenhouse
agriculture parameters based on IoT. In: 2017 international conference on big data, IoT and
data science, Pune, India
12. Kumar R, Kumar H (2014) Availability and handling of data received through GPS device:
In tracking a vehicle. In: 2014 IEEE international advance computing conference (IACC),
Gurgaon, India
13. Hutabarat DP, Hendry H, Pranoto JA, Kurniawan A (2016) Human tracking in certain indoor
and outdoor area by combining the use of RFID and GPS. In: 2016 IEEE Asia Pacific conference
on wireless and mobile (APWiMob), Bandung, Indonesia
14. AI Rashad MA, Ourmar OA, Singh D (2013) A real time GSM/GPS based tracking system
based on GSM mobile phone. In: 2013 second international conference on future generations
communication technologies (FGCT), London, UK
15. Bilgic HT, Alkar AZ (2011) A secure tracking system for GPS-Enabled mobile phones. In: Pro-
ceedings of fifth international conference on information technology and multimedia (ICIMU),
Kuala Lumpur, Malaysia
IoT-based Wearable Micro-Strip Patch
Antenna with Electromagnetic Band Gap
Structure for Gain Enhancement

M. Ameena Banu, R. Tamilselvi, M. Rajalakshmi and M. Pooja Lakshmi

Abstract Wearable antennas used for various applications such as telemedicine,

firefighting, and navigation purpose are integrated into fabrics. The antenna per-
formance is described with the integration of Electromagnetic Band Gap (EBG)
structure in which the micro-strip patch antenna (MSP) consists of flexible substrate
material. The main goal of using EBG structure in micro-strip patch antenna is to
overcome the limitations of patch antenna and to achieve better gain and efficiency,
lower side lobes and back lobes level, better isolations among array elements, and by
suppressing surface wave modes. The design of micro-strip patch antenna is proposed
to resonate the antenna at 2.45 GHz using Jeans as substrate material for wearable
applications which supports Industrial, Scientific and Medical (ISM) applications.
The various characteristics of antenna such as return loss and VSWR are analyzed.
The simulation of antenna is done by CST studio software. Internet of Things (IoT)
may be used for the development of antennas, which supports multi-standard services
within a single design.

Keywords Wearable antenna · Jeans · Electromagnetic Band Gap · ISM band ·

CST studio software · IoT

1 Introduction

A device that designed to transmit or receive electromagnetic waves in free space is

known as antenna and also antenna called as a transducer. Comparing to all types
of antenna, the micro-strip patch antenna has low profile, low cost, lightweight,
and conveniently to be integrated with RF devices, because that is widely used
in various applications. The micro-strip antennas can be designed to have many
geometrical shapes and dimensions. The simplest configuration of micro-strip patch

M. Ameena Banu (B) · R. Tamilselvi · M. Rajalakshmi · M. Pooja Lakshmi

Department of Electronics and Communication Engineering, Sethu Institute of Technology,
Pulloor, Virudhunagar District 626115, Tamilnadu, India
e-mail: ameenabanuece@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1379

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1380 M. Ameena Banu et al.

Fig. 1 Basic structure of micro-strip patch antenna

antenna consists of a radiating patch, dielectric substrate, and ground plane. The
basic structure of micro-strip patch antenna is shown in Fig. 1.
Ground plane is called as lower conductor, the radiating patch is called as upper
conductor, and the dielectric substrate is placed between the two conductors. In micro-
strip patch antenna, the ground plane is a flat horizontal conducting surface, and it
used to reflect the radio waves from the other antenna elements. The shape and size
of the ground plane play a vital role in determining antenna radiation characteristics
including return loss, gain, and VSWR.
The most aspect of designing micro-strip patch antenna is to choose the suit-
able dielectric substrate. For particular applications, the various types of substrates
are available in the market that provides considerable flexibility in the choice of a
substrate. The dielectric substrate characteristics including dielectric constant (per-
mittivity) and dielectric medium are most considered to design a micro-strip patch
antenna. The electrical characteristics of antenna are determined by substrate thick-
ness and permittivity. The radiating patch is used to improve the performance of
antenna. The micro-strip patch antenna consists of various shapes. The common
shapes of patches are rectangular, circular, and triangular. Many shapes of patch
design are combined to improve the antenna performance parameters [1, 2].
The drawbacks of micro-strip patch antenna are narrow bandwidth, low efficiency,
and low gain. To overcome the drawbacks of micro-strip patch antenna, Electromag-
netic Band Gap (EBG) is used. EBG structure is used to enhance the gain of the
micro-strip patch antenna, reduce the backward radiation, and also achieve better
radiation efficiency. The EBG technology plays a major role in the radio frequency
and microwave applications because of their unique band gap characteristics at certain
frequency ranges. The periodic arrangements of dielectric or metallic elements are
known as EBG structure. The various EBG structures such as one-dimensional (1D),
two-dimensional (2D), three-dimensional (3D), mushroom and uni-planar EBG may
be used [3–5].
IoT is the network of physical devices and other appliances with embedded elec-
tronics, sensors, connectivity, software, and actuators, which is used to connect the
IoT-based Wearable Micro-Strip Patch Antenna … 1381

device and exchange data. Data is transferred over a network without requiring
human-to-computer or human-to-human interaction [6].

2 Literature Survey

Many researchers have designed the micro-strip patch antenna with flexible materials.
Kumar et al. [7] developed the hexagonal shaped body wearable textile antenna
on EBG substrate material, and they used felt as substrate material. The simulated
antenna had achieved the return loss of −18.8526 dB and gain of 6.551 dB.
Researchers Purohit and Raval [8] developed the wearable textile patch antenna
using jeans as substrate at 2.45 GHz. According to the simulated results, return loss
is −32.57 dB at 2.45 GHz, and gain is 7.26 dB. But the fabricated antenna resonated
at 2.4945 GHz with return loss of −30 dB.
Roland et al. [9] developed a patch antenna with EBG structure for WLAN appli-
cations using FR 4 as substrate. They used the mushroom structure for designing an
EBG structure because of its high impedance characteristics. They achieved better
return loss of −23.18 dB and gain of 14.2 dB at 2.42 GHz.
The above studies on design of micro-strip patch antenna with different substrate
materials and different EBG structure techniques elaborated the advantages of use
of flexible and multiple substrate materials. Hence, we authors are getting interested
in designing an antenna with flexible substrate material which can extend its use
for wearable applications and combination of different substrate materials as double
substrate layer with EBG structure to enhance the performance of the antenna at the
desired resonant frequency.

3 Antenna Design

The patch antenna, substrate, and ground layer dimensions are calculated as follows
[10]:
1. To calculate width of the patch (W )

C
W =  (1)
(εr +1)
2 f0 2

where εr —relative permittivity

2. To calculate effective dielectric constant (εreff )

 −1/
εr + 1 εr − 1 12h 2
εr (eff) = + 1+ (2)
2 4 W
1382 M. Ameena Banu et al.

where

h—height of the substrate

W —width of the patch

3. To calculate effective length of the patch (L eff )

C
L(eff) = √ (3)
2 f 0 εr (eff)

4. To calculate actual length of patch (L)

L = L(eff) − 2L (4)

where —small deviation if required.

5. Substrate

For wearable antenna, the substrate material may be cotton, jeans, felt, teflon, etc.
Two-layered substrates are used to design the micro-strip patch antenna. EBG ele-
ments are placed between the two-layered substrate to reduce the back radiations.
In this work, the dielectric material jeans with dielectric constant εr = 1.6 is used
as substrate. The thickness of each substrate material is 1.6 mm for two-layered
substrates with the dimensions of 54(L) × 47(W ) mm2 .

6. Ground

The copper material is used as ground plane with the dimension of 54(L) ×
47(W ) mm2 (Figs. 2, 3, 4).
All dimensions of different layers of micro-strip patch antenna are given in Table 1.

7. Electromagnetic Band Gap structure

EBG structures are always used as a part of microwave devices in order to improve
the performance of devices especially to improve the radiation/gain patterns and to
decrease the noise/losses in transmissions [11]. EBG structures are also known as

Fig. 2 Side view of antenna

IoT-based Wearable Micro-Strip Patch Antenna … 1383

Fig. 3 Top view of antenna

Fig. 4 Bottom view of

antenna

high impedance surface due to their ability to suppress the surface wave at certain
operational frequencies. In recent years, there has been rapid increase in the utiliza-
tion of Electromagnetic Band Gap (EBG) structures in electromagnetic and antenna
community [12]. In this proposed design, to achieve better gain and radiation effi-
ciency, different EBG structures have been implemented as shown in Figs. 5, 6, 7, 8,
and 9.

4 Results and Discussion

(i) Analysis of patch antenna without EBG structure

In this case, the micro-strip patch antenna with single substrate without EBG struc-
ture is considered where the thickness of the substrate material is 1.6 mm. While
1384 M. Ameena Banu et al.

Table 1 Dimensions of
Parameters Values (mm)
antenna
Patch length L 38
Patch width W 22
Patch height H 0.035
Ground length L g 47
Ground width W g 54
Ground height H g 0.035
Substrate1 length 47
Substrate1 width 54
Substrate1 height 1.6
Substrate2 length 47
Substrate2 width 54
Substrate2 height 1.6
EBG layer length L e 47
EBG layer width W e 54
EBG layer height H e 0.035

Fig. 5 EBG structure 1

simulating this design with the help of CST software, the following results shown in
Figs. 10, 11, 12, and 13 are obtained.
(ii) Analysis of patch antenna with EBG structure
In order to improve the performance of the micro-strip patch antenna, the EBG
structure is introduced in between the two substrate layers. The antenna parameters
such as return loss, VSWR, gain, and directivity of designed patch antenna with
different EBG structures are observed and compared. The proposed patch antenna
performance with and without EBG is analyzed using CST Studio Suite software.
EBG structure 1
IoT-based Wearable Micro-Strip Patch Antenna … 1385

Fig. 6 EBG structure 2

Fig. 7 EBG structure 3

Fig. 8 EBG structure 4

1386 M. Ameena Banu et al.

Fig. 9 EBG structure 5

Fig. 10 Return loss plot without EBG

The designed micro-strip patch antenna has the linked hexagonal EBG structure as
shown in Fig. 5, which has the gain and directivity of 5.225 dB and 6.802 dBi,
respectively. The antenna parameters like return loss, VSWR, gain, and directivity
are observed as depicted in Figs. 14, 15, 16, and 17.
EBG structure 2
The designed micro-strip patch antenna has modified EBG structure as shown in
Fig. 6, which has the gain and directivity of 4.813 dB and 6.849 dBi, respectively. As
compared with the previous design, the modified design produced high return loss
with a frequency of 2.998 GHz. The performance parameters are shown in Figs. 18,
19, 20, and 21.
IoT-based Wearable Micro-Strip Patch Antenna … 1387

Fig. 11 VSWR plot without EBG

Fig. 12 Gain without EBG

Fig. 13 Directivity without EBG

1388 M. Ameena Banu et al.

Fig. 14 Return loss plot with EBG structure 1

Fig. 15 VSWR plot with EBG structure 1

Fig. 16 Gain with EBG structure 1

IoT-based Wearable Micro-Strip Patch Antenna … 1389

Fig. 17 Directivity with EBG structure 1

Fig. 18 Return loss plot with EBG structure 2

Fig. 19 VSWR plot with EBG structure 2

1390 M. Ameena Banu et al.

Fig. 20 Gain with EBG structure 2

Fig. 21 Directivity with EBG structure 2

EBG structure 3
For the modified EBG structure 3 as shown in Fig. 7, gain and directivity are observed
as 4.898 dB and 6.316 dBi, respectively, at the resonant frequency of 1.746 GHz.
For this new MSP antenna, the performance parameters are shown in Figs. 22, 23,
24, and 25.
EBG structure 4
For the modified EBG structure 4 as shown in Fig. 8, gain and directivity are observed
as 5.063 dB and 6.728 dBi, respectively, with a resonant frequency of 1.884 GHz for
which the characteristics are observed as depicted in Figs. 26, 27, 28, and 29.
EBG structure 5
The gain and directivity of 4.930 dB and 6.711 dBi, respectively, are obtained for the
micro-strip patch antenna with EBG structure 5 which is resonated at the frequency
of 2.848 GHz. The plots of various parameters are given in Figs. 30, 31, 32, and 33.
IoT-based Wearable Micro-Strip Patch Antenna … 1391

Fig. 22 Return loss plot with EBG structure 3

Fig. 23 VSWR plot with EBG structure 3

Fig. 24 Gain with EBG structure 3

1392 M. Ameena Banu et al.

Fig. 25 Directivity with EBG structure 3

Fig. 26 Return loss plot with EBG structure 4

Fig. 27 VSWR plot with EBG structure 4

IoT-based Wearable Micro-Strip Patch Antenna … 1393

Fig. 28 Gain with EBG structure 4

Fig. 29 Directivity with EBG structure 4

The performance parameters such as return loss, VSWR, gain, directivity, and
resonant frequency observed for designed micro-strip patch antenna with different
EBG structures are listed in Table 2 which is used to compare the performance of
them.

5 Conclusion

While designing an antenna, the main consideration is its radiation characteristics.

In this proposed design, the aim of introducing EBG structure is for improving the
gain of the MSP antenna. When the performance parameters listed in Table 2 are
1394 M. Ameena Banu et al.

Fig. 30 Return loss plot with EBG structure 5

Fig. 31 VSWR plot with EBG structure 5

Fig. 32 Gain with EBG structure 5

IoT-based Wearable Micro-Strip Patch Antenna … 1395

Fig. 33 Directivity with EBG structure 5

Table 2 Comparison of performance parameters of MSP with various EBG structures and without
EBG
Parameters Return loss S11 VSWR Gain (dB) Directivity Frequency
(dB) (dBi) (GHz)
EBG −16.319 1.360 5.225 6.820 1.788
structure 1
EBG −17.435 1.310 4.813 6.849 2.998
structure 2
EBG −18.673 1.263 4.898 6.316 1.746
structure 3
EBG −22.970 1 5.063 6.723 1.884
structure 4
EBG −27.592 1.087 4.930 6.711 2.848
structure 5
Without EBG −15.691 1.392 1.765 5.143 2.12

considered for analysis, it is obviously observed that the gain of the patch antenna with
any of the EBG structure is more than the gain obtained without using EBG structure.
While integrating different structures of the EBG layer, their resonances couple each
other, and, as a result, a wider bandwidth will be generated and characteristics of
the patch antenna such as return loss, gain, VSWR, and directivity vary. At the
same time, if importance is given to return loss of the antenna which describes
the radiating efficiency of the antenna, the MSP antenna with EBG structure 5 has
improved return loss of −27.59 dB at the resonant frequency of 2.848 GHz. Right
now, we have designed the antenna using EBG. In future, the proposed method may
be further implemented using IoT for improving the overall performance.
1396 M. Ameena Banu et al.

References

1. Chougule JSA, Wali UV (2015) Design of flexible microstrip antenna for wearable application.
Int J Res Emerg Sci Technol 2(6)
2. Rais NHM, Soh PJ, Malek F et al (2009) A review of wearable antenna. In: 2009 Loughborough
antennas & propagation conference. [Link]
3. Tanaka M, Jae-Hyeuk J (2003) Wearable micro-strip antenna. In: The antennas and propagation
society international symposium
4. Santas JG, Alomainy A, Yang H (2007) Textile antennas for on body communications:
techniques and properties. In: The antennas and propagation, 2007. EuCAP 2007
5. Errifi H, Baghdad A, Badri A, Sahel A (2014) Improving micro-strip patch antenna directivity
using EBG superstrate. Am J Eng Res 3(11):125–130 (e-ISSN: 2320-0847 p-ISSN: 2320-0936 )
6. Nate K, Tentzeris MM (2015) A novel 3-D printed loop antenna using flexible NinjaFlex
material for wearable and IoT applications
7. Kumar R, Singh J, Sohi BS (2016) Hexagonal shaped body wearable textile antenna on EBG
substrate material. IJCSMC 5(6):260–266
8. Purohit S, Raval F (2014) Wearable-textile patch antenna using Jeans as substrate at 2.45 GHz.
Int J Eng Res Technol 3(5). ISSN: 2278-0181
9. Karpagavalli S, Shaaru Nivetha R, Roland DS (2016) Enhancement of gain in micro-strip
patch antenna using EBG structure for WLAN Application. Glob Res Dev J Eng. In: Inter-
national conference on innovation in engineering and technology (ICIET)-2016, July 2016,
e-ISSN:2445-5703
10. Balanis CA (2016) Antenna theory: analysis and design. Wiley
11. Amsaveni BM, Phavithra PJ (2018) Gain enhancement of a square patch antenna using EBG
structure. Int J Innov Technol Explor Eng 8(2S). ISSN: 2278-3075
12. Shital L, Vaishali D (2017) Micro strip antenna array with square EBG structure. IOSR
J Electron Commun Eng 12(6):58–62 e-ISSN: 2278-2834, p-ISSN: 2278-8735 (Ver. I
(November–December 2017)
Design and Implementation of a 3 DOF
Arm for Disaster Management

Rajesh Kannan Megalingam, Shree Rajesh Raagul Vadivel, Vamsi Gontu,

Deepak Nagalla, Ravi Kiran Pasumarthi and Phanindra Kumar Allada

Abstract Current trend in technology has, in turn, necessitated the amelioration in

automation and the ease of performing a task. Robotic arms are playing a decisive
role in all aspects of human life irrespective of the application. Not only designing
but also understanding them is strenuous. This paper discusses designing and imple-
mentation of a transparent 3-degree of freedom (DOF) arm which can be analyzed
and controlled to perform divergent dexterity tasks with simple commands in a plain
sailing way. The arm illustrated in this paper is mounted on a mobile robot and ana-
lyzed performing varied dexterity tasks with motley complexity. A lucid graphical
user interface (GUI) is developed using Qt-designer to generate commands to control
this arm. In discordance to heavy industrial or complex robotic arms, it can be shifted
between places and can be mounted on any platform depending on the application.

Keywords Qt-designer · RQT · GUI · Robot operating system (ROS) · DOF ·

Universal robot description format (URDF) · Planar workspace

R. K. Megalingam (B) · S. R. R. Vadivel · V. Gontu · D. Nagalla · R. K. Pasumarthi · P. K. Allada

Department of Electronics and Communication Engineering, Amrita Vishwa Vidyapeetham,
Amritapuri, India
e-mail: rajeshkannan@[Link]
S. R. R. Vadivel
e-mail: shreerajgul@[Link]
V. Gontu
e-mail: vamsi.gontu2308@[Link]
D. Nagalla
e-mail: [Link]@[Link]
R. K. Pasumarthi
e-mail: ravikiran.p3637@[Link]
P. K. Allada
e-mail: alladaphanindrakumar@[Link]

© Springer Nature Singapore Pte Ltd. 2020 1397

G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1398 R. K. Megalingam et al.

1 Introduction

Robotic arms are deployed in varied sectors like manufacturing, medicine, and many
more. Major intention behind robotic arms is to replace human beings who are
involved in dangerous work and also the work that need more accuracy and high
precision. A robotic arm is expected to do tasks a human arm is capable of performing
but with increased concentration and automation. By implementing robotic arms in
different sectors, a task can be accomplished with high accuracy, reduced threat to
human life, and in a short time span. In rescue operations, involving humans, the life
of rescuer is also at risk as the affected area is unknown and just sending a mobile
platform or a robot base is not an efficient way of rescuing. A robotic arm can be
mounted on a mobile platform which can be used to clear debris and obstacles on its
way and to interact with victims.
The triumphant deployment of a robotic arm needs a better understanding and
precise analysis in first place. Several aspects like designing, fabricating, manufac-
turing, programming, and many more are involved in making a robotic arm work
properly. The efficiency of arm is dependent on the workspace, accuracy, collision
avoidance, ease of control, etc. More the DOF more is the work space and efficiency.
Complexity in deployment increases simultaneously. Controlling the arm is also a
key aspect as a task can be accomplished as expected only when it can be controlled
with ease and when it responds as the controller expects it to respond.

2 Motivation and Problem Statement

A Robotic arm helps in the search and rescue operations. It helps us to locate victims in
location where a person or a robot itself cannot enter in search and rescue operations.
The robotic arm helps us in the cluttered areas during the natural disasters. But the
problem of developing an arm with more degrees of freedom is, it is complex to design
and build. It is also very expensive. It is risky for a human to save the life other in
complicated situations. So in order to help the people in such type of situations, we
can use the robotic arm. When the robotic arm is mounted on the robot it helps the
controller to carry the rubble piles at the situation.

3 Related Works

Paper [1] discusses about developing of a mechanical design using CAD and sim-
ulating it using RoboAnalyzer software. It explains that the control of the robotic
arm depends on the dexterity and manipulability of the arm. Paper [2] describes a
new method for sliding mode control mechanism to reduce the disturbance in the
control without affecting the performance or robustness of the system. It explains a
Design and Implementation of a 3 DOF Arm for Disaster … 1399

law which adapts to the changes in the functionality of the system without affecting
its performance. Paper [3] discusses about interfacing of a robotic arm using motion
planning to control the arm to pick and place the objects using Robotic Operating
System (ROS). Paper [4] explains about manually controlling of a robotic arm with
a miniature version of the robotic arm using Bluetooth wireless technology con-
trolled by human operator. Paper [5] discusses about controlling of a robotic arm
by mimicking the human arm using “Kinect Skeletal Project” of Kinect SDK. It
explains about controlling of the robotic arm using serial communication of Arduino
microcontroller. Paper [6] discusses about the development of a robotic arm which
is controlled using Bluetooth wireless interface that is attached to a pole, which can
be used for pruning and harvesting fruits. Paper [7] describes about a new manip-
ulation strategy to grasp various objects stably using a dual-arm robotic system in
the ROS environment. To grasp an object, an operability index of the dual-arm robot
(OPIND) has been defined. The manipulability index of both arms has been derived
using the Jacobian matrix. Paper [8] discusses about a new approach with enough
flexibility which can be potentially applicable for different scenarios in pick and
place of objects using Robotic Operating System (ROS). Paper [9] presents Xinxin,
an intelligent interaction service robot which can simultaneously localize and map
the area using SLAM mapping which can be interacted with the robotic arm using
voice command, remote control using ROS. Paper [10] describes about the design
and implementation of a robotic arm on a wheel chair system to control it using
teleoperation controls or voice command using ROS. Paper [11] describes about
developing of a 6 DOF arm using counterbalance mechanism (CBM) to maintain
high performance and reduce the torque required to move the arm. Paper [12] dis-
cusses about developing of a robotic arm using non-linear spring mechanism which
reduces the torque on the joints thereby providing position accuracy and collision
safety of the robotic arm.

4 Architectural Diagram

Robotic arm discussing in this paper follows the master–slave configuration. This is
one of the finest configurations that helps. It means the slave performs the tasks as
per the master commands that run the master system.

4.1 Control Station

Control station works as the ROS-master that refer to the controller and that commu-
nicates between the different nodes like joystick, keyboard, and GUI. These interfaces
are used for different purposes in controlling the arm. The time instances of the pub-
lished and the received data between ROS-master and ROS-slave should be in sync
1400 R. K. Megalingam et al.

else the data will be ignored. These GUI values are published by the topics /cmd_vel
and /joint_states.
Joystick. A customised Joystick is used for establishing a user friendly control
of the arm. The signals from joystick are published over the /joint_state topic by the
ROS-master running in the system on controller side. A topic is a virtual connection
between two nodes, which acts as a link between them as the data is transferred.
Keyboard. Another mode of control is using a keyboard, where the characters on
keyboard are used to control the arm. Each character clicked on the keyboard will be
published to ROS-slave which is running on the robot via a topic named /cmd_vel.
GUI. GUI is designed to provide the user a visual feed from the camera on the arm.
Qt-designer a package for designing GUI, integrated with ROS is used for displaying
the received feed from cameras. The GUI can also be designed to send data to the
ROS-slave for controlling the arm.

4.2 Base Station

From Fig. 1 base station refers to the robot. It includes both software and hard-
ware. ROS-slave subscribes the topics from the ROS-master and they are sent to the

Fig. 1 Architectural diagram

Design and Implementation of a 3 DOF Arm for Disaster … 1401

microcontroller, joints, and the camera. When the camera and the microcontroller
subscribe, they execute their tasks and send back the information to the ROS-master.

4.2.1 Software

Serial_node. Serial_node is a ROS node running as part of ROS-slave running in the

base station. It establishes a serial communication with the microcontroller which
controls the arm based on the commands sent from the control station. It also monitors
the connection between microcontroller and base station throughout the run.
Camera Node. This node is responsible for initialising the camera and transmit-
ting the feed to ROS-slave, which in turn publishes this to the GUI of ROS-master. The
data from cameras is converted to different formats and the controller can choose any
based on different factors like transmission power of antenna, transmission strength,
noise in the transmission medium, and delay caused during transmission.

5 Design and Implementation

The robotic arm has a planar workspace in X–Z plane. Three linear actuators are
used at all the three DOF’s. The design is made in such a way that it had a vast planar
workspace of −30° to 120° (Fig. 2). The first DOF is limited to 0°–125°, the second
DOF is limited from 0° to 145°, and the third DOF is limited from 0° to 138°. The
payload of the arm is 4–5 kg. Maximum reach length of the arm is 158 cm. It is very
easy to fix the arm on a stable surface or on a movable wheeled robot; hence, it can
also work as a spherical jointed robotic arm according to the base is mounted to. As
the payload of the arm is of 5kgs and the reach length is 158 cm, this can act as a
pick and place robotic arm. The design all can be manufactured by laser cutting and
it is very easy to implement. The compressed dimension of the arm is very less that
it can easily accommodate in the very less space of 15 × 80 × 8 cm (Fig. 3). The
weight of the arm is 6 kg such that it can be carried easily.

5.1 Software

This paper is mainly concerted on the graphical user interface (GUI) that works on
the Qt-platform which is an object-oriented application framework. This platform is
used in many softwares in order to perform multiple applications that run on many
of the desktops. Qt-designer is a GUI tool in Qt which forms a .ui file when widgets
and objects are added to it. RQT is a software shell of ROS that enacts the various
GUI tools in the form of plugins. When these plugins are enabled in the RQT, the
respective package should run in order to work the GUI appropriately.
1402 R. K. Megalingam et al.

Fig. 2 Workspace of the arm

Fig. 3 Compressed position

The three DOFs of the robotic arm is a joint-by-joint control algorithm using mul-
tiple GUIs in different ways. Using sliders (Qslider) and radio buttons (Qradiobutton)
widgets in a single GUI in Fig. 4, and the other GUI is an integration of line edit
(Qlineedit) and a push button that can be seen in Fig. 5.
Figure 4 shows the control of the first three DOFs of the arm using sliders and
the grippers are also being controlled by sliders. In Fig. 5, it was explained about the
values of the angles that the respective DOF should rotate which can be given in the
edit line that can be sent using the push button.
Arduino. The 3-DOF arm is programmed using the Arduino software. Since
Arduino provides UART TTL serial communication, the whole arm is implemented
Design and Implementation of a 3 DOF Arm for Disaster … 1403

Fig. 4 Slider-based control

Fig. 5 Angle-based control

1404 R. K. Megalingam et al.

Fig. 6 ROS rqt_graph

using this communication. The actuators that used in this robotic arm work as same
as the normal dc motor. End effector of the arm is also programmed using Arduino.
ROS Platform. ROS includes the mastering of nodes, multiplexing of data, and
allows distributed performance over the multicore and multiprocessor. ROS lines
are added to Arduino code to communicate between the nodes. These nodes publish
and subscribe using different topics in order to establish the connections between
the DOFs. In Fig. 6, it shows that the different topics have been published on
the serial node. Here, the topics (/joint_0, /joint_1, /joint_2) are subscribed by the
/subscriber_node and published on the /serial_node.
When camera is used to know the position of the gripper, camera node is also
attached to Fig. 6 Usb_cam is the package used to run the camera on the arm.
When RQT is launched, selecting the plugins icon → visualisation → image_view
a window is popped on the GUI in which the usb_cam_compressed topic is being
selected to the view the camera.

5.2 Hardware

The robotic arm that mentioned in the above context is employed with Arduino
ATmega328 microcontroller using serial communication between the three motors.
Two bullet and lact4 linear actuators had been used in the three DOFs. Dual motor
drivers with 6–18 V compatible 20 capable specifications were used in this arm. A
window shaft motor (wiper motor) is used to control the gripper. The position of
Design and Implementation of a 3 DOF Arm for Disaster … 1405

Fig. 7 Arm with gripper and

camera

the gripper can be known from the camera attached to the end effector. There is an
emergency switch that is attached at the gripper, if the gripper has extended to its
maximum position it triggers and make the working of the gripper to stop.
In Fig. 7, it is shown how the actuators are mounted and assembled to one joint
to another. It is also shown that the placement of camera on the gripper and how the
gripper moves in Fig. 8.
Gripper. The griper of the above robotic arm is a unique one, which is capable
of holding a block of 10 cm. The lever mechanism in Fig. 9 is used in this gripper.
A high torque worm geared dc motor Fig. 10 (window lift motor) is used for the
implementation of the lever mechanism. The holding torque of the gripper is 35 kg
cm. There is very less backlash in the gripper and which is helpful in the precise
working of gripper.

6 Experiments and Results

This section lists the results of experiments that we performed with the 3 DOF robotic
arm.
1406 R. K. Megalingam et al.

Fig. 8 Arm without camera mount

Fig. 9 Lever mechanism

Design and Implementation of a 3 DOF Arm for Disaster … 1407

Fig. 10 Motor used for

gripper

In the first case, in Table 1, the base is free to move toward any direction; hence
in that case, it can reach any point in the X–Y plane without any restrictions. But in
the case toward the Z-axis which is normal to the arm it cannot extend more than

Table 1 X, Y, and Z positioning of the arm

Arena condition Location Test result
X Y Z
Free movement of base is allowed 0.1 0 0 Pass
0.56 0 0 Pass
1.68 0 0 Pass
0 0.1 0 Pass
0 0.56 0 Pass
0 1.68 0 Pass
0 0 0.1 Pass
0 0 0.56 Pass
0 0 1.68 Fail
Base is fixed at point of start 0.1 0 0 Fail
0.56 0 0 Pass
1.68 0 0 Fail
0 0.1 0 Fail
0 0.56 0 Pass
0 1.68 0 Fail
0 0 0.1 Fail
0 0 0.56 Pass
0 0 1.68 Fail
1408 R. K. Megalingam et al.

Table 2 Boundaries in X, Y,
Axis Boundaries of the arm
and Z axes
Min Max
X-axis 0.48 1.42
Y-axis 0.48 1.42
Z-axis 0.54 1.63

the restricted value so the task has failed. In the second case, in Table 2, the base
movement is restricted so the arm can only reach the points which are in its boundary.
Hence, the arm fails to complete the tasks which are out of boundary.

7 Conclusion

The successful deployment of a 3 DOF robotic arm is described in this work. The
arm is mounted on a mobile base which is tested for rescue operations in simulated
arenas. The arm is able to do the dexterity check tasks like pick and place, turning
valves, and few other. As the arm has planar workspace in x- and z-planes, an object
in y-plane can be altered only by rotating the mobile platform. But, when the arena
is narrow, and if the object is in y-plane, the task cannot be performed as the body
cannot be moved. The complexity in deploying this arm is very less and it is well
suited for running basic tests for evaluating the complexity of task.

Acknowledgements We express our gratitude and thankfulness to Humanitarian Technology Labs

and the Department of Electronics and Communication Engineering, Amrita Vishwa Vidyapeetham,
Amritapuri Campus for helping us to complete this work.

References

1. Hussain SB, Kanwal F (2016) Design of a 3 DoF robotic arm. In: The sixth international
conference on innovative computing technology (INTECH 2016) on 24–26 August in Dublin,
Ireland
2. Fallaha C, Saad M, Kanaan H (2007) Sliding mode control with exponential reaching law
applied on a 3 DOF modular robotic arm. In: Proceedings of the European control conference
2007 on July 2–5 in Kos, Greece
3. Hernandez-Mendez S, Maldonado-Mendez C, Marin-Hernandez A, Rios-Figueroa HV,
Vazquez-Leal H, Palacios-Hernandez ER (2017) Design and implementation of a robotic arm
using ROS and Moveit! In: IEEE international autumn meeting on power, electronics and
computing (ROPEC) on 8–10 November 2017 in Ixtapa, Mexico
4. Megalingam RK, Boddupalli S, Apuroop KGS (2017) Robotic arm control through mim-
icking of miniature robotic arm. In: International conference on advanced computing and
communication systems (ICACCS) on Jan 06–07 2017 in Coimbatore, India
5. Megalingam RK, Saboo N, Ajithkumar N, Unny S, Menon D (2013) Kinect based gesture
controlled Robotic arm: a research work at HuT labs. In: IEEE international conference in
Design and Implementation of a 3 DOF Arm for Disaster … 1409

MOOC, innovation and technology in education (MITE) on 20–22 December 2013 in Jaipur,
India
6. Megalingam RK, Vignesh N, Sivanantham V, Elamon N, Sharathkumar MS, Rajith V (2016)
Low cost robotic arm design for pruning and fruit harvesting in developing nations. In: Inter-
national conference on intelligent systems and control (ISCO) on 7–8 Jan 2016 in Coimbatore,
India
7. Kim D-E, Park D-J, Moon J-H, Kim K-S, Park J-H, Lee J-M (2017) Development of a robot
manipulation technology in ROS environment. In: IEEE international conference on multisen-
sor fusion and integration for intelligent systems (MFI) on 16–18 November 2017 in Daegu,
South Korea
8. Tavares P, Sousa A (2015) Flexible pick and place architecture using ROS framework. In:
Iberian conference on information systems and technologies (CISTI) on 17–20 June 2015 in
Aveiro, Portugal
9. Zhaohui Z, Xuesong M, Xu B, Hanghang C, Jian T (2016) Development of an intelligent inter-
action service robot using ROS. In: IEEE advanced information management, communicates,
electronic and automation control conference (IMCEC) on 3–5 October 2016 in Xi’an, China
10. Tremblay T, Padir T (2013) Modular robot arm design for physical human–robot interaction.
In: IEEE international conference on systems, man and cybernetics on 13–16 Oct 2013 in
Manchester, UK
11. Lee W-B, Lee S-D, Song J-B (2017) Design of a 6-DOF collaborative robot arm with coun-
terbalance mechanisms. In: IEEE international conference on robotics and automation (ICRA)
on 29 May–3 June 2017 in Singapore
12. Park J-J, Kim H-S, Song J-B (2009) Safe robot arm with safe joint mechanism using nonlinear
spring system for collision safety. In: IEEE international conference on robotics and automation
on 12–17 May 2009 in Kobe, Japan
Author Index

A B
Abbas, Ahmed Raad, 1307 Balaji, S., 43
Abirami, S., 205 Balakrishnan, Vignesh, 1217
Adhvaryu, Kinjal, 79 Balakumar, A., 1089
Afridi, Shahid, 727 BalaSundaram, Ananthakrishnan, 197
Agarwal, Harshit, 1003 Banakar, R. M., 493, 1151
Agarwal, Vimal, 53 Barve, Amit, 121
Ahmed, Rekib Uddin, 461 Basak, Sujit Kumar, 1179
Ajay Kumar, S., 1233 Basil, K. Y., 665
Ajay Raj, B., 1233 Baweja, Vanshita R., 673
Ajith, Aparna, 143 Bawiskar, Sushant, 345
Akhil Sri Harsha, Suthapalli, 619 Bekal, Chaithra, 781
Alam, Mohammed Shamsul, 1273 Bélanger, Paul, 1179
Allada, Phanindra Kumar, 1397 Bhagat, Sudhanshu, 1191
Al-Mousawi, Ali Jameel, 87 Bhairavi, R., 569
Al Noman, Abdullah, 1285 Bhalaji, N., 1059
Ameena Banu, M., 1379 Bhandari, Rajiv R., 293
Amrutha, S. D., 1369 Bhandarkar, Rekha, 1039
Anand, R., 1223 Bharadwaj, Madiraju Akshay, 153
Ani, R., 187 Bharath, S., 105
Anisha, B. S., 1255, 1263 Bhaskar, N., 1207, 1217
Anitha, P., 275 Bhatia, Rajesh, 673
Anjali, T., 1369 Bhavathankar, Prasenjit, 687
Anusha, Kompalli, 1015 Bhavikatti, Sujata, 1151
Anusha, K. S., 1233 Bhuvana, K., 947
Aparna, T. S., 1369 Bhuvaneshwar, M., 1233
Arockiam, L., 1333 Biswal, Bhawani Sankar, 529
Arulaalan, M., 369 Biswas, Ranjit, 881
Arulkumar, V., 1059 Borade, Tanmay, 451
Aruna, P., 1245 Bowya, M., 1109
Asher, Vinit, 687
Ashokkumar, N., 113
Ashokkumar, Sigamani, 197 C
Aswin, A. V., 665 Chaithanya, P. S., 921
Avani, H. S., 773 Chanda, Ruthvik, 1345
Azam, Khairul Islam, 1285 Chandrasekaran, K., 325
© Springer Nature Singapore Pte Ltd. 2020 1411
G. Ranganathan et al. (eds.), Inventive Communication and Computational
Technologies, Lecture Notes in Networks and Systems 89,
[Link]
1412 Author Index

Charles Babu, G., 177 Gowrishankar, B. N., 1201

Chavan, Amruta, 61 Gupta, Aman, 439
Chavan, Swati, 1315 Gupta, Saurabh, 439
Chigateri, Keerthana, 1039 Gupta, Sorjeeta, 1223
Christy Jeba Malar, A., 1323

H
D Haran, V. J., 699
Dakhane, Dhananjay, 451 Harini, D., 805
Dalvi, Pooja, 1315 Harish, A., 1255
Dayal, Neelam, 735 Hasankar, Abhilash P., 33
Debbarma, Tribid, 325 Hoomod, Haider K., 87
Deeksha, S., 947 Hossain, Mohammad Emdad, 1273
Deepthi, M., 947 Hubert Shanthan, B. J., 1333
Deshmukh, Smita, 1315 Hublikar, Shivaraj, 483
Deshpande, Santhosh L., 583
Deva Priya, M., 1323
Devi Priya, R., 849, 971 I
Dhanya, K., 1089 Indu, V., 353
Dhivyaprabha, E., 1323
Dhoulath Beegum, J., 335
Dhummal, Ashwini, 865 J
Divya, C. D., 773 Jadhav, Akshaykumar, 1315
Dongre, Nilima, 61 Jadhav, Dipti, 61
Jadhav, Sandeep, 237
Jain, Ashish, 53
E Jain, Bhumika, 781
Edwankar, Prutha, 963 Jaisimha, Aashish, 1263
Eligar, Sanjay, 493 Janet, B., 591
Eligar, Vijaya, 483 Jariwala, Gaurav, 1003
Eswari, B. Maragatha, 153 Jayapandian, N., 71
Eswari, R., 591 Jeyalakshmi, C., 1089
Evangeline, C. Suganthi, 829 Jhawar, Saksham, 1255
Ezhilarasie, R., 1029 Jiji, Sasidharan, 1127
Jondhale, Priyanka, 1357
Jose, Christy James, 955
F Jose, Deepa, 873
Franklin, K. Ben, 381 Joseph Abraham Sundar, K., 1081
Joshi, Bharti, 471, 1239
Jyothsna, K., 1369
G
Garg, Roopali, 399
Geddam, Mrudula, 1357 K
Geesala, Ravi Teja, 1345 Kadam, Poonam, 983
Geetha, K., 805 Kadam, Subhas, 865
George, A., 1245 Kakhandki, Arun, 483
George Dharma Prakash Raj, E., 755 Kalim Amzad Chy, Md., 1273
Gharat, Jatin, 121 Kanagaraj, G., 275
Ghorpade, Tushar, 451 Kandasamy, Kamalanathan, 1139
Golam Rabiul Alam, Md., 1273 Kanimozhi, J., 901
Gontu, Vamsi, 1397 Kannammal, A., 563
Gopalakrishnan, Nivetha, 541 Karthick, V., 361
Gopalakrishnan, Vinodhini, 541 Karthika, K., 849, 971
Gotter, Priyanka, 23 Karthikeyan, A., 317, 767
Author Index 1413

Karthikeyan, B., 1081 Misra, Alok, 1295

Karthikeyan, V., 1109 Mohammed Muddasir, N., 721, 799
Kau, Mandeep, 1307 Mohan, Poornima, 153
Kaur, Jaspreet, 471 Mohapatra, Anjali, 529
Kaur, Kiranbir, 1, 13, 23 Mohapatra, Shubhankar, 529
Kaur, Manjot, 1 Mokammel Haque, Md., 707
Kaur, Preetjot, 399 Mota, Jash, 963
Kaur, Tanveer, 13 Muthulakshmi, R., 901
Kavipriya, R., 1119
Keerthana, B., 901
Khairnar, Vaishali, 1191 N
Khan, Salman, 1263 Naaz, Sameena, 881
Khan, Shahidul Islam, 1273 Nagalla, Deepak, 1397
Kidd, Andrew, 1161 Nagarajan, P., 113, 1171
King, Akeem T. L., 187 Naif, Jolan Rokan, 87
Kirubakaran, Steward, 391 Nair, Aswathy K., 43
Kolhe, Likhesh, 1191 Namitha, K., 153
Krishna, K. S. Ananth, 699 Narayanan, Gayathri, 105, 143, 215
Krishnan, Arya, 143 Narayanan, P. S., 187
Krishnan, Venkatalakshmi, 541 Naveena, Ambidi, 1015
Krisshna, P. Anand, 215 Naveen Balaji, G., 931
Kubde, Vidya, 653 Naveen Kumar, M. R., 1201
Kumar, Ankit, 1191 Nayak, C., 381
Kumaravelu, Vinoth Babu, 829 Naz, Insha, 881
Kumari, Anshu, 431 Neema, N., 799
Kumar, Manish, 673 Neha, N., 911
Kumar, Rakesh, 381, 549 Nigam, Katta, 1345
Kunte, Ashwini S., 599 Nirmal, S., 819
Kuriakose, Bineeth, 665 Nithya, N., 423
Niyas, Khadheeja, 1223
L
Lakshmi, K. N., 799
Lansy, Alan, 699 P
Loganathan, Vignesh, 1207 Palaniappan, Sambandam, 197, 361
Lourde, R. Mary, 789 Pandhre, Tanvi, 1239
Parameswaran, Vignesh, 307
Parmar, Navrattan, 257
M Parthipan, V., 391
Mahadevan, Revathi, 1239 Pasumarthi, Ravi Kiran, 1397
Mahaveerakannan, R., 1071 Pati, Kunal Sekhar, 767
Mahendran, K., 369 Patil, Deepti, 1239
Maheswari, M., 1119 Patil, Pratik, 317
Manikannan, G., 369 Patil, Rahul, 865
Manju, B. R., 503, 993 Patil, Rohan, 635
Manjula, R., 839 Patra, Narayan, 1097
Marathe, Nilesh, 61 Patwary, Muhammad Kamrul Hossain, 707
Mary Lourde, R., 745 Paul Jefferson, P., 873
Masum, Abdul Kadar Muhammad, 1273, PavanKumar, S. P., 781
1285 Ponmani Raja, M., 1127
Meera, M., 1369 Pooja Lakshmi, M., 1379
Megalingam, Rajesh Kannan, 1345, 1397 Pooja, M. R., 727
Meghana, B. U., 859 Prabakaran, P., 369
Meghanathan, Natarajan, 227, 891 Pradeep, R., 163
1414 Author Index

Pradheepa, R., 1233 Sanjana, M. S., 721

Prajitha, C., 361 Sanjeev, Kumar, 591
Prakash, B. Bhanu, 43 Sanjith, S. L., 755
Pramod, S., 353 Sanju, V., 865
Pranesh, 583 Sankara Babu, B., 177
Prasad, Prajith Kesava, 1323 Saran, J. Vaishnu, 153
Prashanth, M. V., 799, 859 Sarkar, Moumita, 529
Pravinraj, S., 921 Sashi Rekha, K., 275
Prema, N. S., 1201 Sasi Kiran, J., 177
Primya, T., 275 Sathishkumar, S., 849, 971
Priya, D., 647 Savaliya, Viraj, 983
Priyanga, S., 911, 921 Sawant, Dattatray, 635, 963
Puneeth, Kuppili, 353 Sawant, Priyanka, 1357
Sawant, Rohan, 1161
Sawarkar, Sudhir, 653
R Saxena, Pratishtha, 735
Radhika, A. D., 947 Sekar, Manjubala, 1139
Rahman, Md A., 891 Sekhar, Y. Chandra, 43
Rajalakshmi, M., 1379 Senthilnathan, R., 911
Raj, Amrita, 549 Seshan, Suresh, 911
Rajasekhar, K., 293 Sethupathi, M., 931
Rajasekhar Reddy, M., 619 Shah, Dharmil, 983
Rajasree, M. S., 955 Shah, Henish, 983
Rajendran, Sreeja, 745, 789 Shah, Kalp, 983
Rajesh Kannan, Megalingam, 1369 Shaikh, Sohail, 1191
Raj, M. M. Anishin, 1081 Shan-A-Alahi, Ahmed, 1285
Rajput, Manita, 1357 Shankar Sriram, V. S., 911, 921
Rakhee, M., 699 Shanmugasundaram, M., 307, 345
Rakib, Mohammed Golam Sarwar, 1285 Sharma, Ravindra Kumar, 53
Rama Devi, R., 1071 Sharma, Sunil, 53
Ramakanth Kumar, P., 1255, 1263 Shedge, Rajashree, 471
Ranga, Virender, 237, 257 Shetye, Yash, 963
Rasane, Krupa, 135 Simhachalam Naidu, B., 257
Ravi, V., 163 Sinchana, C., 727
Reddy, O. Rajasekhar, 153 Sinchana, K., 727
Reji, Basil, 665 Singhal, Madhvi, 431
Rethishkumar, S., 1047 Singh, Sachin Kumar, 345
Revathy, S., 1223 Singh, Sukhvir, 1307
Rinesh, S., 361, 391 Sivagamasundari, S., 205
Rohith, Narjala, 353 Sivakumar, V., 901
Sivaramakrishnan, N., 931
Soman, K. P., 819
S Someshwaran, M., 873
Sagar, G. R., 71 Soni, Anshu, 237
Sahana, H. P., 721 Sonnagi, Siddaram, 865
Saha, Prabir, 461 Sowmya, V., 819
Sahu, Ronak, 1191 Sreedharan, Rajkumar P., 993
Sai Hanuman, A., 177 Sreejith, A. G., 699
Sai Siddartha Reddy, P., 619 Sreenivas, K. Vandith, 215
Sambandam, Palaniappan, 391 Sridevi, R., 409, 423
Sampat, Manisha, 963 Srilakshmi, A., 805
Sandilya, Kambhampati Sai, 153 Srimathi, S., 409
Sangeetha, T. R., 143 Srinivas, M. Sai, 43
Author Index 1415

Sruthi, S., 335 Venkatramana, P., 113

Subbiah, Swaminathan, 197 Verma, Sushma, 163
Subramani, Gopinath, 1207 Vidhate, Amarsinh, 121
Sudha, Gnanou Florence, 569 Vidyashree, K. P., 721
Suganya, T., 1323 Vijayakumar, R., 1047
Sujatha, R., 517 Vijaya, M. S., 839
Sundresh, H. D., 647 Vijayan, Rajilal Manathala, 1029
Sunil, B., 1201 Vinay, Putchala, 353
Sunitha, N. R., 163 Vinod, Athira, 503
Supe, Neha, 1239 Vinothini, C., 275
Suresh Gnana Dhas, C., 1071 Viraktamath, S. V., 33
Suresh, S., 563 Vishnu Vardhan, L. R., 1323
Syed, Azhar, 789 Vishnu Vashista, D., 619
Viswan, Vimal P., 665
Vivek Sridhar Mallya, P. V., 143
T
Tambe, Suyog, 687
Tamilselvi, R., 1379 W
Thakkar, Hardik, 687 Warrier, Prasanth M., 993
Thakur, Neha Rupesh, 599 Wotto, Marguerite, 1179
Thakur, Shivam, 345
Theeijitha, S., 931
Turkar, Ayushi, 773 Y
Yadav, Divakar Singh, 1295
Yadav, Nishi, 431, 439
U Yadav, Prasant Kumar, 1345
Uddin, Mohammad Nazim, 1285 Yadav, Samarjeet, 735
Umamakeswari, A., 1029 Yakatpure, Surekha, 135
Umapriya, A., 1171 Yallappagoudar, Vinayakgouda, 33
Umarfarooq, A. S., 33 Yeswanth, Kaki, 43
Yuvaraj, D., 1081

V
Vadivel, Shree Rajesh Raagul, 1345, 1397 Z
Vandana Mansur, 517 Zodge, Prajakta, 1315