Received: 19 February 2021 Accepted: 25 March 2021
DOI: 10.1002/dac.4829
RESEARCH ARTICLE
An efficient adaptive threshold-based dragonfly
optimization model for cooperative spectrum sensing in
cognitive radio networks
Sivasankari Jothiraj1 | Sridevi Balu2 | Neelaveni Rangaraj3
1
Department of Electronics and
Communication, ULTRA College of
Summary
Engineering and Technology, Madurai, Effective utilization of spectrum resources is an important factor in wireless
India communication which reduces spectrum scarcity. Over the years, communica-
2
Department of Electronics and
tion systems use different frequency bands, and the users are categorized into
Communication, Velammal Institute of
Technology, Chennai, India licensed and unlicensed users. Most of the wireless bands are typically
3
Department of Electronics and licensed; as a result, accommodation of new technologies such as Internet of
Communication, MNM Jain Engineering Things and machine to machine communication becomes difficult. So it is
College, Chennai, India
essential to obtain a wireless spectrum to adopt new technologies. Cognitive
Correspondence radio technology is introduced to improve such spectrum utilization. Reports
Sivasankari Jothiraj, Department of state that most of the licensed spectrums are underutilized, and few spectrums
Electronics and Communication, ULTRA
College of Engineering and Technology, are overutilized. Cognitive radio networks help to exploit the licensed spec-
Madurai 625104, Tamilnadu, India. trum and access the spectrum without any interference to the licensed user.
Email: jsivash11@[Link]
Through its spectrum sensing and spectrum sharing process, cognitive radio
network gains more attention in wireless communication. This research work
proposed an efficient optimized spectrum sensing technique for cognitive radio
networks through dragonfly optimization algorithm along with the adaptive
threshold process. Proposed work performs better in terms of detection accu-
racy and efficiency when compared to conventional spectrum sensing schemes
such as linear support vector machine and particle swarm optimization
models.
KEYWORDS
classification, cognitive radio network, spectrum sensing, spectrum sharing
1 | INTRODUCTION
Spectrum access in the current wireless communication systems is stuck with fixed frequency allocation policies.
Due to this policy management, most of the licensed bands are underutilized, and other unlicensed bands are over-
utilized. Recent studies report that most of the licensed spectrum is not utilized properly in the time and space
domain. The average utilization of licensed bands will be in the range of 5–50%, and it proves that the utilization of
spectrum is handled inefficiently. Static policies followed in spectrum allocation, fixed radio functions, and rigid pol-
icies are the major reasons for this underutilization. To improve the spectrum utilization and reduce the unlicensed
band issues, these unused spectrums can be allocated to unlicensed users in an opportunistic manner. Cognitive
radio technology provides such possibilities to unlicensed users to utilize the licensed user spectral bands without
Int J Commun Syst. 2021;e4829. [Link]/journal/dac © 2021 John Wiley & Sons Ltd. 1 of 11
[Link]
2 of 11 JOTHIRAJ ET AL.
any interference. Through its dynamic spectrum allocation strategies, cognitive radio improves spectrum utilization.
This innovative concept is accepted by the Federal Communications Commission (FCC) and adopted in various
wireless communication models.
The main objective of the cognitive radio is to sense the wireless spectrum, identify the unused or underutilized
bands, and allocate the unused bands to unlicensed users without interference to the licensed users. A vast survey has
been made to analyze the spectrum sensing strategies in the cognitive radio network, and the observations in existing
models are as a simplified note. A novel spectrum sensing using matching game theory is reported in the literature1
which performs spectrum sensing by forming coalitions among the users in the fading channel. The proposed work is
derived based on matching theory, and target detection probability is used as collaborative distributed spectrum sens-
ing. The proposed model provides better matching performance and reduces the false alarm probability and improves
the network throughput. Energy harvesting in a cognitive radio network is reported in the Zhang et al.2
which improves spectrum utilization. Research work is derived based on centralized cooperative spectrum sensing,
and Markov decision process is utilized to attain maximum throughput and better sensing process. However, the
computation cost of the system increases due to its increased number of iterations.
A trustful cluster-based cooperative compressed spectrum sensing algorithm is proposed in the Wu3 to reduce the
spectrum sensing issues in the cognitive radio network. In order to obtain the trustful cluster, the maximum likelihood
function is used.4 It estimates the distance between the user and cluster, and it is compared with a predefined threshold
value to obtain a trustful cluster. Research work performs better in spectrum sensing and identifies the channel status
from the trusted cluster which effectively manages the malicious users in the environment. A combined predetection
and sleeping policy are introduced in research work5 to improve the spectrum sensing performance in cognitive radio
networks. Detection of a primary user (PU) by utilizing all the sensing nodes in the predetection subphase and sleeping
policy is introduced to decrease the energy consumption in the sensing process. Rayleigh fading channel is used in the
research work to minimize the energy consumption of nodes and attains better performance with reduced false alarm
probability.
A multistage spectrum sensing is reported in Mourougayane et al.6 which utilizes two parallel detectors for high
and low signal to noise ratio (SNR) bands. For high SNR bands, energy detector and singular value-based detector are
used. In the case of low SNR bands, autocorrelation-based detector and cyclostationary feature-based detector are used.
The optimal performance of the multistage detector is obtained through these SNR estimation process and adaptive
sampling process. Research model performs better in terms of detection accuracy and sensing time. However, the
design complexities of the proposed model are high compared to other spectrum sensing techniques which are consid-
ered as a major limitation. Discrete-time Markov chain model-based spectrum analysis is reported in Afzal et al.7 The
research model defines the association process of customer premises equipment and base station and analysis the spec-
trum status. The performance of the system is evaluated in terms of back-off stage, passage time. However, the complex-
ity of the proposed approach is high, and performance in spectrum analysis must be improved.
Spectrum monitoring and prediction of spectrum availability are the two major factors which need to be considered
to improve the spectrum mobility in cognitive radio network.8 The recent developments in spectrum monitoring and
parallel data transmission open the way for various reactive approaches; however, it faces issues such as single packet
loss and imperfect spectrum monitoring. To overcome these issues, spectrum prediction and monitoring techniques are
combined as a fusion process using AND–OR fusion rule. Research model efficiently detects the activities of the PU
during transmission and increases the system throughput and reduces data loss. A similar throughput enhancement
strategy is reported in Raj9 which detects the spectrum holes without introducing interference to the PU. Probabilistic
spectrum access method is used in the research work defines the channel status as spectrum access probability by intro-
ducing an optimal power allocation strategy, and the proposed work seems to increases the network throughput better
than conventional models.
Two state semi-Markov process is reported in Bagchi and Siddiqui10 to improve the throughput through dynamic
spectrum access. Proposed work considers the prior and posterior distributions for the transition states and estimates
the loss function as Bayes risk factor. Average availability is calculated based on the time frames, and frame length is
calculated based on detection rates and false alarm probabilities to improve the throughput. Using constant factors to
evaluate the system performance is considered as the limitation of the proposed work. Reinforcement learning-based
spectrum management is reported in Yin et al.11 which obtain a decision for channel allocation and improves the
throughput of the secondary user. Epsilon-greedy method is used to reduce the issues in the decision process of
the reinforcement model. Proposed model allows the user to switch the channel based on the decision and reduces the
unnecessary switching requirements.
JOTHIRAJ ET AL. 3 of 11
Reliability in the spectrum sensing process is considered in Adardour et al.,12 and the issues are addressed through
the weighted averaging method. The proposed model measures the signal strength of the secondary user using a simple
recursive estimator. All together makes the proposed model more reliable in the spectrum sensing process. Spectrum
sensing speed and accuracy are considered as the main objective in research work13 and obtain better performance
using discrete wavelet packet transform and principal component analysis along with convex optimization. The input
signal is decomposed through the discrete wavelet packet transform to generate the data matrix. The principal compo-
nent analysis is used to reduce the dimensionality of the data matrix, and it is further optimized through convex optimi-
zation. Proposed work preserves the data, and dimensionality reduction improves the sensing speed, and optimization
increases the accuracy of the process. Though the performance of this model is better, it has limitations in its design
complexity and implementation cost.
A hybrid spectrum accessing model is reported in Zhang et al.14 which utilizes nonorthogonal multiple access for
cooperative transmission in the cognitive radio network. The data transmission of secondary users is defined in the
research model where the secondary users can transmit data directly to the destination when the PU is not available.
Otherwise, secondary users need to obtain optimal relay for data transmission. Beamforming technology is introduced
in the research work used to receives the signal from relay node and source of secondary users, helps to reduce the
necessity of retransmission for the secondary users. The channel interference due to secondary users is analyzed in
Wang et al.15 that presents an efficient transmission system with an intelligent cancelation technique based on the sta-
tus of the PUs. Based on the PU time slots, the transmission system is defined into different cases and increases the per-
formance in terms of energy efficiency and less retransmission.
The power control issues in the cognitive radio network are reported in He et al.16 that presents an energy
harvesting approach to meet the energy requirement for data transmission. Hybrid game theory-based routing and
power control algorithm is developed to maximize the path throughput and quality of services. Longer distance, high
throughput, higher hop counts, low energy consumption, and reduced delay are the advantages of the research model.
Machine learning-based spectrum sensing is reported in the literature Saber et al.17 which uses four machine learning
models such as support vector machine,18 decision tree, k-nearest neighbor, and artificial neural network to detect the
signals. The comparative analysis evaluates the performance of all four models and identifies that the support vector
machine is much better in spectrum sensing compared to other models. Adaptive threshold-based models are reported
in the literature,19,20 which obtains high estimation accuracy and less computational complexity. Through the adaptive
process, the spectrum detection performance is increased even for the uncertain noise and reduces interference to
the PUs.
1.1 | Motivation of the research
The research motivation is framed based on the analysis of existing research models, and the observations are summa-
rized as follows.
• Energy-based models are widely used in the cognitive radio spectrum sensing process.
• Limited research models are available for cooperative and compressive spectrum sensing process.
• Existing research works focused on throughput while the performance of the system degrades in terms of quality of
services and computation cost.
1.2 | Objective of the research
Based on the above findings, the objective of the proposed work is framed.
• Energy efficiency is considered as the main objective where it was achieved by an efficient dragonfly optimization
algorithm along with the adaptive threshold process.
• To improve the detection probability, accuracy, and efficiency of the system through an optimization approach.
• To achieve a low complexity spectrum sensing algorithm for a high detection rate with the minimum number of
samples.
4 of 11 JOTHIRAJ ET AL.
2 | P R O P O S ED W O R K
The mathematical model for the proposed cooperative spectrum sensing algorithm is presented in this section. The sys-
tem consists of m samples of signals which are received from the secondary users. The secondary user in the network
computes the signal power and a decision about PU is obtained based on the decisions of the secondary user. Fusion
center takes the responsibility to provide the decision based on the secondary user observations, and in some cases, the
decision can be made by the secondary user itself. But most of the cases prefer the first condition as the decision is
obtained through the fusion center. The given issue is framed as a hypothesis as H(0) for noise and H(1) for noise
affected signal. Here, the noise-affected signal represents the signal presence, and binary hypothesis for sensing model
based on time instant is derived as
(
n i ðl Þ : H ð0Þ
yi ðlÞ = for i = 1, 2, 3…:n ð1Þ
gi sðlÞ + ni ðlÞ : H ð1Þ
where ni(l) represents the Gaussian noise, gi represents the channel gain, and s(l) represents the primary signal. The
SNR is given as
φs jgi j2
SNRðSi Þ = ð2Þ
ρ2i
P
where φs is the energy of the transmitted signal, and it is given as φs = m 2
i − 1 jsðlÞj , and the energy of the received signal
Pm 2
is obtained as δi = l = 1 jyðlÞj , where m is the number of sensed samples. The received signal is forwarded to the fusion
center through a control channel. Figure 1 depicts the cooperative spectrum sensing process.
Fusion center evaluates the statistics from cognitive radio and obtains a decision. Global test statistics are used in
the fusion center, and variance under different hypothesis is given as follows:
Xm
VarðFCjH ð0ÞÞ = i=1
2Nσ 4i + ϑ2i = wt γH ð0Þw ð3Þ
and for H(1)
Xm
VarðFCjH ð1ÞÞ = i=1
2Nσ 4i + ϑ2i + Si σ 4i w2i = wt γH ð1Þw ð4Þ
Based on the energy detection, the decision rule is formulated, and it is given as
Fc > τ for H ð0Þ
ð5Þ
Fc < τ for H ð1Þ
FIGURE 1 Cooperative spectrum sensing
JOTHIRAJ ET AL. 5 of 11
where τ represents the decision threshold, and it is observed through the adaptive threshold process. The decision
threshold is calculated as
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
τi = q − 1 Pf wt γH ð0Þw + Nσ 2 w ð6Þ
where the false alarm probability is represented as Pf. The complexity of the spectrum sensing process is reduced due to
the adaptive threshold process which increases the detection probability. The cooperative spectrum sensing detection
probability is given as
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi !
q − 1 Pf wt γH ð0Þw −φs gt w
Pd = q pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð7Þ
wt γH ð1Þw
where w is the optimal weight factor, and it is used to maximize the detection probability.
Figure 2 depicts an illustration of dragonfly characteristics which is formulated as an optimization model. In order
to optimize the weight factor and to improve the detection performance in cognitive radio, proposed work introduces
the dragonfly optimization algorithm. Dragonfly algorithm (DFA) produces efficient results for real-world applications
under different domains such as image processing, wireless sensor networks, and cloud computing. Compared to other
optimization algorithms, the convergence time is quite acceptable. Moreover, the DFA is steady and easy to merge with
other algorithms.
The optimization problem starts with a random set of solutions which is observed from the behavior of dragonflies.
The best solution is obtained as an iterative process, and it is given as
Xn
Sbest = − i=1
x − xi ð8Þ
where the individual current position is represented as x and the neighbor individual is represented as xi. The position
of the ith individual considering the velocity (vi) is given as follows:
Pn
i = 1 vi
Pi = ð9Þ
n
FIGURE 2 Dragonfly optimization
6 of 11 JOTHIRAJ ET AL.
Pn
xi
The position consistency is obtained as ci = in= 1 −x, and the attraction is given as fi = x+ − x. The position of the drag-
+
onfly is represented as x, and x represents the food source position. The distraction factor is obtained as
ei = x − + x ð10Þ
where the position of enemy is represented as x−. The behavior of dragonflies is obtained based on the above parame-
ters. The position vector and step vector
Δx t + 1 = ðsSi + pPi + Cci + Ff i + Eei Þ ð11Þ
where the separation is represented as Si, Pi is the alignment, ci is the consistency, fi is the food source, and ei is the dis-
traction factor. The coefficients in all the terms represent the weight functions. The position update in the optimization
model is given as follows.
x t + 1 = x t + Δx t + 1 ð12Þ
The overall optimization process is illustrated as a process flow chart in Figure 3. The process starts by randomly initial-
izing the dragonflies’ population, and a step vector is assigned for each flies. Then using the adaptive threshold func-
tion, the maximum iteration is compared to continue the process of calculating the fitness function of each dragonfly.
Once the fitness function is obtained, the food source and opponent details are updated, which helps the systems to get
aware of other users.
FIGURE 3 Process flow of proposed optimization model
JOTHIRAJ ET AL. 7 of 11
The characteristic features of dragonflies such as separation, alignment, cohesion, attraction, and distraction from
enemies are updated, and new values are obtained if there is a change in the environment. Then the secondary user
presence is checked and updated similar to the velocity and position update of the dragonflies. Finally, the process is
incremented and repeated to obtain the next position which reflects the spectrum sensing process in the cognitive radio
network. The pseudocode for the proposed adaptive double threshold dragonfly optimization for cooperative spectrum
sensing is summarized as follows.
3 | R E S U L T A N D DI S C U S S I O N
The proposed adaptive threshold dragonfly optimization is experimentally verified, and its result performances are ana-
lyzed in this section. Experimentation is progressed in MATLAB R2018 installed on Windows 10 operating system of
2.20-GHz frequency with 16-Gb memory. The simulation parameters are depicted in Table 1. To validate the better per-
formance, conventional linear support vector machine and particle swarm optimization models are compared with the
results of the proposed work.
The detection probability of the proposed model for secondary users is estimated and depicted in Figure 4. Based on
the SNR, this detection probability is observed for three sets of secondary users. The detection range varies if the num-
ber of secondary users increases due to spectrum requests, and detection probability reaches maximum if the SNR
crosses the zero point.
The detection probability is analyzed with respect to false alarm probability and depicted in Figure 5. False alarm
probability defines the occurrence of false alarms when the spectrum is not available for the secondary user. Detection
probability and false alarm probability are directly proportional to each other, and it reaches a maximum for the highest
probability value. False alarm probability should be minimum in the spectrum sensing process.
To evaluate the detection probability for different threshold values, the detection probability and false alarm proba-
bility functions are measured. In Figure 5A, the threshold value is selected as 0.2, and its responses are plotted. Simi-
larly, for threshold value 0.5, the responses are obtained and plotted in Figure 5B. The proposed model attains better
performance for both threshold values compared to linear support vector machine and particle swarm optimization.
Due to the adaptive threshold selection and optimized weight functions, the detection probability of the proposed
model is considered to be better compared to other models.
8 of 11 JOTHIRAJ ET AL.
TABLE 1 Simulation parameters
S. no Parameter Range
1 No of primary users 100
2 No of secondary users 50
3 Threshold value 0.5
4 Bandwidth 3 MHz
5 Acceptable interference limit 10−2 to 10−1 W
6 Transmission BER 4 dB
7 Sensing time 2.5 ms
8 Noise spectral density power 10−6 W/Hz
FIGURE 4 Detection range of secondary users
FIGURE 5 False alarm probability for different threshold values (A) 0.2 and (B) 0.5
To obtain the network performance in spectrum sensing, the transmitted power from the PU is considered for anal-
ysis along with detection probability and depicted in Figure 6. It is observed that the proposed DFA attains better detec-
tion probability compared to other models even the power increases. The detection probability increases gradually as
similar to transmitted power, and it reaches maximum probability for 500 mW. Considerably the existing models
exhibit less probability of detection when it reaches the maximum power function. This demonstrates the efficient per-
formance of the proposed spectrum sensing process.
The detection accuracy of the proposed model is obtained with respect to SNR, and it is depicted in Figure 7. It is
observed that the proposed DFA attains 95%, and particle swarm optimization attains 93%, and SVM has 91% average
JOTHIRAJ ET AL. 9 of 11
FIGURE 6 Transmitted power and detection probability
FIGURE 7 Accuracy comparison
FIGURE 8 Efficiency comparison
10 of 11 JOTHIRAJ ET AL.
accuracy. The accuracy of the system reflects the detection probability which is generally high compared to other algo-
rithms, with less false alarm probability and less interference to the PUs.
The efficiency of the proposed model is validated based on the detection probability and detection accuracy and
false alarm ratio. It is observed from Figure 8, the proposed model attains maximum efficiency compared to support
vector machine and particle swarm optimization due to reduced false alarm ratio while detecting spectrum in the cogni-
tive radio network. From the experimental analysis, it is obvious that the proposed adaptive threshold dragonfly optimi-
zation model performs better compared to other models.
4 | C ON C L U S I ON
An efficient cooperative spectrum sensing for cognitive radio networks is proposed in this research work using an adap-
tive threshold-based dragonfly optimization algorithm. Spectrum sensing is an important process in cognitive radio net-
work which improves the spectrum utilization by allocating unused spectrum to the secondary or unlicensed users in
the network. Cooperative spectrum sensing is widely adopted, and various techniques are evolved to improve the detec-
tion accuracy. The proposed work introduces an adaptive threshold-based dragonfly optimization algorithm for cooper-
ative spectrum sensing which efficiently performs and identifies the spectrum holes. Existing spectrum sensing
techniques such as linear support vector machine-based sensing model and particle swarm optimization-based sensing
models are used to compare with the proposed model. Experimental results demonstrate that the proposed dragonfly
optimization model performs better in terms of detection accuracy and efficiency.
A C K N O WL E D G M E N T
This study has no funding.
CONFLICT OF INTEREST
There is no conflict of interest.
DATA AVAILABILITY STATEMENT
We used our own data. We used our own coding.
E TH IC A L S TA T EM EN T
Humans and animals are not involved in this research work.
ORCID
Sivasankari Jothiraj [Link]
R EF E RE N C E S
1. Tahir M, HadiHabaebi M, Rafiqul Islam M. Novel distributed algorithm for coalition formation for enhanced spectrum sensing in cogni-
tive radio networks. AEU-Int J Electron C. 2017;77:139-148.
2. Zhang F, Jing T, Ma L. Optimal spectrum sensing-access policy in energy harvesting cognitive radio sensor networks. Procedia Comput
Sci. 2018;129:194-200.
3. Wu Y. Localization algorithm of energy efficient radio spectrum sensing in cognitive internet of things radio networks. Cogn Syst Res.
2018;52:21-26.
4. Shbat M, Tuzlukov V. Primary signal detection algorithms for spectrum sensing at low SNR over fading channels in cognitive radio. Digit
Signal Process. 2019;93:187-207.
5. Gao Y, Deng Z, Choi C. Combined pre-detection and sleeping for energy-efficient spectrum sensing in cognitive radio networks.
J Parallel Distrib Comput. 2018;114:85-94.
6. Mourougayane K, Amgothu B, Bhagat S, Srikanth S. A robust multistage spectrum sensing model for cognitive radio applications. AEU-
Int J Electron C. 2019;110:1-11.
7. Afzal H, Rafiq Mufti M, Yousaf M. Performance analysis of radio spectrum for cognitive radio wireless networks using discrete time
Markov chain. J Syst Softw. 2019;151:1-7.
8. Thakur P, Kumar A, Satashia SN. Spectrum mobility in cognitive radio network using spectrum prediction and monitoring techniques.
Phys Commun. 2017;24:1-8.
9. Raj JS. Machine learning implementation in cognitive radio networks with game-theory technique. J: IRO J Sustain Wirel Syst Jun. 2020;
2020(2):68-75.
JOTHIRAJ ET AL. 11 of 11
10. Bagchi S, Siddiqui JY. Throughput optimization using availability analysis based spectrum sensing for a cognitive radio. AEU-Int J Elec-
tron C. 2018;85:12-22.
11. Yin Z, Wang Y, Wu C. Reinforcement learning spectrum management paradigm in cognitive radio using novel state and action sets. Proc
Comput Sci. 2018;129:433-437.
12. Adardour HE, Meliani M, Hachemi MH. Improved local spectrum sensing in cluttered environment using a simple recursive estimator.
Comp Mater Continua. 2017;61:208-222.
13. Raj JS. Energy efficient sensed data conveyance for sensor network utilizing hybrid algorithms. J: IRO J Sustain Wirel Syst Dec. 2019;
2012(04):235-246.
14. Zhang Y, Liu J, Peng Y, Dong Y, Sun G, Huang H. Performance analysis of relay based noma cooperative transmission under cognitive
radio network. Comp Mater Continua. 2020;63(1):197-212.
15. Wang L, Liu J, Yang A. Analysis of underlay cognitive radio networks based on interference cancellation mechanism. Comp Mater Con-
tinua. 2020;63(1):401-416.
16. He X, Jiang H, Song Y, Owais M. Power control and routing selection for throughput maximization in energy harvesting cognitive radio
networks. Comp Mater Continua. 2020;63(3):1273-1296.
17. Saber M, el Rharras A, Kharraz HA. Spectrum sensing for smart embedded devices in cognitive networks using machine learning algo-
rithms. Procedia Comput Sci. 2020;176:2404-2413.
18. Jothiraj S, Balu S. A novel linear SVM-based compressive collaborative spectrum sensing (CCSS) scheme for IoT cognitive 5G network.
Soft Comput. 2019;23(18):8515-8523.
19. Hassan ES. Adaptive threshold to guarantee both detection and false alarm probabilities in multi-taper based spectrum sensing.
J Franklin Inst. 2018;356:1640-1657.
20. Sarala B, Rukmani Devi S, Joselin Jeya Sheela J. Spectrum energy detection in cognitive radio networks based on a novel adaptive
threshold energy detection method. Comput Commun. 2020;152:1-7.
How to cite this article: Jothiraj S, Balu S, Rangaraj N. An efficient adaptive threshold-based dragonfly
optimization model for cooperative spectrum sensing in cognitive radio networks. Int J Commun Syst. 2021;
e4829. [Link]