Depar tment of Information and Communications Engineering

Planning of Wireless
Networks for 5G/6G
Applications
Mohsen Abedi

DOCTORAL
THESES
Aalto University publication series
DOCTORAL THESES 31/2024

Planning of Wireless Networks for

5G/6G Applications

Mohsen Abedi

A doctoral thesis completed for the degree of Doctor of Science

(Technology) to be defended, with the permission of the Aalto
University School of Electrical Engineering, at a public examination
held at the lecture hall AS1 of the school on 1 March 2024 at 12
o'clock noon.

Aalto University
School of Electrical Engineering
Department of Information and Communications Engineering
Risto Wichman Group
Supervising professor
Professor Risto Wichman, Aalto university, Finland

Thesis advisor
Dr. Alexis Alfredo Dowhuszko, Aalto university, Finland

Preliminary examiners
Dr. Nobby Stevens, KU Leuven, Belgium
Professor Majid Safari, University of Edinburgh, UK

Opponent
Professor Marcos Katz, University of Oulu, Finland

Aalto University publication series

DOCTORAL THESES 31/2024

© 2024 Mohsen Abedi

ISBN 978-952-64-1669-4 (printed)

ISBN 978-952-64-1670-0 (pdf)
ISSN 1799-4934 (printed)
ISSN 1799-4942 (pdf)
[Link]

Unigrafia Oy
Helsinki 2024

Finland
N
SWA ECO
C
Publication orders (printed book):
DI

LA
NOR

BEL

[Link]
Printed matter
Printed matter
1234 5678
4041-0619
 Ab s t ra c t
Aalto University, P.O. Box 11000, FI-00076 Aalto [Link]

Author
Mohsen Abedi
Name of the doctoral thesis
Planning of Wireless Networks for 5G/6G Applications
P u b l i s h e r School of Electrical Engineering
U n i t Department of Information and Communications Engineering
Series Aalto University publication series DOCTORAL THESES 31/2024
Field of research Wireless Communication Systems
M a n u s c r i p t s u b m i t t e d 2 October 2023 D a t e o f t h e d e f e n c e 1 March 2024
P e r m i s s i o n f o r p u b l i c d e f e n c e g r a n t e d ( d a t e ) 18 December 2023 L a n g u a g e English
Monograph Article thesis Essay thesis
Abstract
The rise of 6G will allow wireless communication networks to achieve unprecedented levels of
connectivity, capacity, and coverage. As part of 6G, higher frequency bands such as mmWaves,
Terahertz, and visible light can be utilized to provide larger capacity than sub-6 GHz bands
currently used in 4G/5G. However, as frequencies increase, the signal range and obstructions may
limit the signal's coverage, creating a signifi cant challenge when it comes to ensuring seamless
coverage.
Wireless network planning aims to determine the minimum number of wireless access points
and their locations in the service areas. Planning is key to achieving seamless coverage, optimizing
bandwidth utilization, reducing energy consumption, and ensuring suffi cient quality of services.
Planning is an NP-hard problem that requires an exhaustive search in order to reach the optimal
solution. Due to the large dimensions, parameters, and diverse requirements of networks, it
isimpossible to conduct an exhaustive search for network planning. It can, however, be made
feasible by developing mathematical tools and optimization models.
The purpose of this dissertation is to examine the issue of wireless network planning in light
of the different wireless network requirements. Based on Voronoi diagrams and Delaunay
triangulation, a regularity algorithm is proposed for planning wireless cellular networks that
maximize coverage and balance cell loads outdoors. In addition, we propose a graph that models
the indoor areas taking into account the propagation limits imposed by signals at higher
frequencies. We then demonstrate that the deployment of access points for seamless Line-of-Sight
coverage of indoor areas can be achieved by partitioning this graph into cliques, each representing
a wireless access point. As a final step, we use this graph modeling to analyze network requirements
to ensure that access points are deployed in a fashion that meets various operational requirements,
for example, Line-of-Sight backhauling between access points and multiple Line-of-Sight coverage
for positioning.

K e y w o r d s wireless communications, reference node deployment, capacity, coverage

I S B N ( p r i n t e d ) 978-952-64-1669-4 I S B N ( p d f ) 978-952-64-1670-0
I S S N ( p r i n t e d ) 1799-4934 I S S N ( p d f ) 1799-4942
L o c a t i o n o f p u b l i s h e r Helsinki L o c a t i o n o f p r i n t i n g Helsinki Y e a r 2024
P a g e s 176 u r n [Link] /URN:ISBN: 978-952-64-1670-0
Preface

As this long and rewarding journey comes to its conclusion, I want to

express my gratitude to those who have been instrumental in supporting
me, both in my academic pursuits and on a personal level, which has
enabled the successful completion of this dissertation.
I would like to express my sincere appreciation to my supervisor, Profes-
sor Risto Wichman, for providing me with the invaluable opportunity to
be a part of his research team and consistently providing his intellectual
guidance and unwavering support throughout my extensive doctoral stud-
ies. Additionally, I wish to convey my thanks to my advisor and coauthor,
Dr. Alexis Dowhuszko, for his expert technical guidance and compassion-
ate support throughout this academic journey. His humble attitude and
dedication to my success have made a significant impact on my work.
Beyond the confines of academia, I am deeply grateful to those whose
contributions have been pivotal in every facet of this endeavor and in
nurturing my personal well-being. My sincere thanks go to my close
companion and friend, Alireza Jaribion, whose benevolent supports have
been inspiring in every aspect of my life. I would also like to extend
my appreciation to my dearest friends, Ramin and Pegah Hashemi, Sina
Moradian, Mohammad Yasoubi, Saeed Shahbazi, Arash Vahedniazi, and
Diyako Ghaderyan for sharing unforgettable memories throughout this
journey.
Lastly, but by no means least, I express my heartfelt gratitude to my
parents, Mehdi and Zohre, for their unconditional support, their love, and
encouragement. In a bittersweet moment, I bid a fond farewell to my
grandmother, Zahra, whose memory and love will forever remain in my
heart.

Helsinki, January 25, 2024,

Mohsen Abedi

1
Contents

Preface 1

Contents 3

List of Publications 5

Author’s Contribution 7

Abbreviations 9

Symbols 13

1. Introduction 17
1.1 Motivation and Scope of the Dissertation . . . . . . . . . . . 17
1.2 Network Planning in Wireless Cellular Systems . . . . . . 20
1.3 Contribution of the Dissertation . . . . . . . . . . . . . . . . . 21
1.4 Structure of the Dissertation . . . . . . . . . . . . . . . . . . . 23

2. Background 25
2.1 Communications . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1.1 Radio Frequency Channel Modeling . . . . . . . . . 25
2.1.2 Visible Light Communication Channel Modeling . 27
2.1.3 Calculation of the Data Rate . . . . . . . . . . . . . . 29
2.2 Positioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2.1 Power-Based Positioning Systems . . . . . . . . . . . 30
2.2.2 Time-Based Positioning Systems . . . . . . . . . . . 30
2.2.3 Positioning Principles . . . . . . . . . . . . . . . . . . 31

3. Planning of RF-Based Outdoor Cellular Wireless Networks 33

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2 Geometrical Tools . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2.1 Voronoi Diagram . . . . . . . . . . . . . . . . . . . . . 34
3.2.2 Delaunay Triangulation . . . . . . . . . . . . . . . . . 35
3.3 Requirements of Cellular Wireless Networks . . . . . . . . . 36

3
Contents

3.3.1 Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.3.2 Coverage . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.4 Planning for Load-Balancing . . . . . . . . . . . . . . . . . . . 38
3.5 Planning with Conflicting Network Requirements . . . . . 42
3.6 Planning for Coverage Maximization . . . . . . . . . . . . . . 45
3.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

4. Planning of New Spectrum-Based Indoor Wireless Networks 53

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.2 VLC Channel Model and Optical Wireless Backhauling . . 56
4.3 Illumination requirements of VLC networks . . . . . . . . . 58
4.4 Art Gallery problem and the State of the Art in Wireless
Indoor Planning . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.5 Graph Modeling of Indoor areas . . . . . . . . . . . . . . . . . 61
4.6 Indoor Planning of Optical Wireless Networks for Los Con-
dition in Access and Backhauling . . . . . . . . . . . . . . . . 64
4.6.1 Ensuring Los Condition in Access . . . . . . . . . . 64
4.6.2 Ensuring Los Condition in Access and Backhauling 67
4.6.3 Simulation Results . . . . . . . . . . . . . . . . . . . 70
4.7 Connection Between the Deployment of Positioning Access
Points and the Accuracy of Indoor Positioning . . . . . . . . 72
4.8 Network Planning for a Precise Indoor Positioning . . . . . 76
4.8.1 P-AP Deployment to Ensure 1-LoS Coverage . . . . 76
4.8.2 Access Point Deployment to Ensure 2-LoS Coverage 77
4.8.3 P-AP Deployment to Ensure 3-LoS Coverage . . . . 81
4.8.4 Simulation Results . . . . . . . . . . . . . . . . . . . . 83
4.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

5. Conclusions 89

References 93

Publications 103

4
List of Publications

This thesis consists of an overview and of the following publications which

are referred to in the text by their Roman numerals.

I Mohsen Abedi and Risto Wichman. Planning and Optimization of

Cellular Networks Using Load-based Voronoi Algorithm. In 52nd
Asilomar Conference on Signals, Systems, and Compute, Pacific Grove,
CA, US, 1589-159, October 2018.
II Mohsen Abedi and Risto Wichman. Cellular Network Planning
under Variable QoS Requirements Using Voronoi Algorithm. In
IEEE Wireless Communications and Networking Conference, Seoul,
South Korea, 1-6, May 2020.
III Mohsen Abedi and Risto Wichman. Cellular Network Planning via
Local Regularity Optimization. In 93rd IEEE Vehicular Technology
Conference, Helsinki, Finland, 1-7, April 2021.
IV Mohsen Abedi, Alexis Alfredo Dowhuszko and Risto Wichman. Vis-
ible Light Communications: A Novel Indoor Network Planning Ap-
proach. In IEEE Global Communications Conference, Madrid, Spain,
1-7, December 2021.
V Mohsen Abedi, Alexis Alfredo Dowhuszko and Risto Wichman. In-
door Planning of Optical Wireless Networks for LoS Condition in
Access and Backhauling. Submitted to IEEE Transactions on Com-
munications, September 2023.
VI Mohsen Abedi, Alexis Alfredo Dowhuszko and Risto Wichman. En-
suring line-of-sight condition for accurate positioning and sensing
over RF/OWC networks. Submitted to IEEE Journal on Selected
Areas in Communications, November 2023.

5
Author’s Contribution

Publication I: “Planning and Optimization of Cellular Networks Using

Load-based Voronoi Algorithm”

The author of this dissertation presented the idea, developed the techni-
cal solution, obtained the simulation results, redacted every draft of the
manuscript and its final version. The other author provided technical
feedback and assisted with the redaction of the manuscript.

Publication II: “Cellular Network Planning under Variable QoS

Requirements Using Voronoi Algorithm”

The author of this dissertation presented the idea, developed the technical
solution, obtained the simulation results, and redacted every draft of the
manuscript and its final version. The other author provided technical
feedback and assisted with the redaction of the manuscript.

Publication III: “Cellular Network Planning via Local Regularity

Optimization”

The author of this dissertation presented the idea, developed the technical
solution, obtained the simulation results, and redacted every draft of the
manuscript and its final version. The other author provided technical
feedback and assisted with the redaction of the manuscript.

7
Author’s Contribution

Publication IV: “Visible Light Communications: A Novel Indoor

Network Planning Approach”

The author of this dissertation presented the idea, developed the technical
solution, obtained the simulation results, and redacted every draft of the
manuscript and its final version. The remaining authors provided technical
feedback and assisted with the redaction of the manuscript.

Publication V: “Indoor Planning of Optical Wireless Networks for LoS

Condition in Access and Backhauling”

The author of this dissertation presented the idea, developed the technical
solution, obtained the simulation results, and redacted every draft of
the manuscript. The remaining authors provided technical feedback and
assisted with the redaction of the manuscript.

Publication VI: “Ensuring line-of-sight condition for accurate

positioning and sensing over RF/OWC networks”

The author of this dissertation presented the idea, developed the technical
solution, obtained the simulation results, and redacted every draft of
the manuscript. The remaining authors provided technical feedback and
assisted with the redaction of the manuscript.

8
Abbreviations

AoA angle of arrival

AoD angle of departure

APO average point optimization

BS base station

CAPEX capital expenditure

CDF cumulative distribution function

CH convex hull

CID Cell Identification

CoV coefficient of variation

CPVA channel-based power Voronoi algorithm

CTC connectivity tree construction

CVA centroidal Voronoi algorithm

dB decibel

DC direct current

EC enclosing circle

EE edge elimination

eMBB enhanced mobile broadband

EVA effective visibility angle

FD full duplex

FDD frequency division duplexing

9
Abbreviations

FoV field of view

FPO farthest point optimization

FSO free space optics

GNSS global navigation satellite system

GPS global positioning system

HPVA heterogeneous power Voronoi algorithm

IoT internet of things

IPS indoor positioning system

ISI inter symbol interference

LED light emitting diode

Li-Fi light fidelity

LoS line-of-sight

LPVA load-based power Voronoi algorithm

LTE long term evolution

MCC maximal clique clustering

MIMO multiple input multiple output

mMTC massive machine type communication

mmWave millimeter wave

MSA minimum separation angle

MSD minimum separation distance

NP Non-deterministic Polynomial-time

NLoS non line-of-sight

OFDM orthogonal frequency division multiplexing

OPEX operational expenditure

OW optical wireless

P-AP positioning access point

PD photo-detector

PDA potential deployment area

10
 Abbreviations

PLC power line communication

PoE power-over-ethernet

PRS positioning reference signal

PSO particle swarm optimization

RF radio frequency

RTT round-trip time

SDS service demand share

SINR signal to interference plus noise ratio

SIR signal to interference ratio

SMO spatial median optimization

SR server-side requirement

TDoA time difference of arrival

TDD time division duplexing

ToA time of arrival

THz terahertz

UAV unmanned aerial vehicle

UE user equipment

UR user-side requirement

URLLC ultra reliable low latency communication

VLC visible light communication

3GPP 3rd generation partnership project

11
Symbols

APD effective physical area of the photo-detector

A(1)
i primary area of index i

A(2)
i secondary area of index i

A(3)
i trinary area of index i

B total bandwidth

C (1)
i primary clique of index i

C (2)
i secondary clique of index i

C (3)
i trinary clique of index i

CR(A) connection region of area A

D T transpose of vector D

d s minimum separation distance

E illumination

E m illumination maintenance value

exp(.) exponential function

Fd s (.) forbidden region

f o (.) transmittance of the optical pass-band filter

G tia gain of transimpedance amplifier within a photo-detector

G(1) primary LoS graph

G(2) secondary LoS graph

G(3) trinary LoS graph

13
Symbols

g minimum number of cliques partitioning the primary LoS graph

g′ minimum number of cliques partitioning the secondary LoS graph

g′′ minimum number of cliques partitioning the trinary LoS graph

H i hidden point of index i in a layout

HT (R) hyper triangulation of a layout with respect to the parameter R

h size of connectivity tree

h i (x, y) channel gain from i th base station to a user at location point (x, y)

h LED-PD height between photo-detector and LED

I(.) indicator function

I interference

J whole service area

J i service area of the cell with index i

K total number of resource units in time-frequency domain

L(.) line of sight area

M number of triangles (polygons) partitioning a layout

M(A) mean value of vector A

m L Lambertian index of LED

M(S, C) clique mapping with respect to the set of cliques S and the clique C

N number of base stations

n number of vertices in a layout

P i array of nodes from the root node to the i th leaf in the connectivity tree.

P c coverage probability

P ib power of base station with index i

P i point of index i in a layout

PLED optical power of the light emitting diode

p i polygon of index i in a layout

R hyper triangulation parameter

R root node of the connectivity tree

14
 Symbols

R data data rate

R PD (.) responsivity of the photo detector

N
R+ positive real values of N dimension

r maximum range

S o(w) (.) spectral power distribution of the LED

S(1) set of primary cliques

S(2) set of secondary cliques

S(3) set of trinary cliques

s size of the set of hidden points in a layout

S a heuristic positive value

T connectivity tree

t independence number for a graph

U location point of a user equipment

U o illumination uniformity value

Vi traffic volume share at i th cell

v(L) set of vertices of layout L

W vector of Voronoi weights

w i Voronoi weight with index i

Wd s (S, C) Well-spaced area with respect to the set of cliques S and the
clique C

X set of base stations locations

x i base station location with index i

α a heuristic step size

γ signal to interference plus noise ratio

δ(x, y) service demand distribution

ϵ a positive small enough value

θE effective visibility angle

θmax source radiation semi-angle at half power of the LED

15
Symbols

θs minimum separation angle

κ source signal to user interference parameter

λ wavelength

λl lower cut-off wavelength

λu upper cut-off wavelength

µ mean value

ρ vector of cell loads

ρ i load at i th cell

σ standard deviation

φ angle of irradiance between photo-detector and LED

ψ incidence of the LoS link between photo-detector and LED

Ψ Field-of-View semi-angle of the photo-detector

∅ empty set

≜ definition

̸= not equal

⪯ partial ordering

|a| absolute value of a

|S| size of the set S

⃓⃓ ⃓⃓
⃓⃓ x − y⃓⃓ Euclidean distance between the points x and y

∩ intersection
⋂︁
i intersection over index i

∪ union
⋃︁
i union over index i
∑︁
i sum over index i
∫︁
A integration over set A

[.]+ positive part

16
1. Introduction

1.1 Motivation and Scope of the Dissertation

By leveraging 5G and 5G+ technology, wireless communication networks

will be able to achieve unprecedented levels of capacity, coverage, and
connectivity. Previous generations of wireless and mobile communication
networks have primarily provided connectivity for smartphones, tablets,
and laptops. Using them, individuals could communicate with one another,
send text messages, or access the internet from virtually anywhere and
at any time. The 5G target is to provide data rates of up to 20 Gbps,
capacity increases of up to 10000 times, radio latency as small as 1 ms, and
ultra-reliability of 99.9999% that enable new use cases and applications
on top of 5G mobile networks [1, 2, 3, 4]. The new services enabled by 5G
include industrial Internet of Things (IoT) connectivity and Ultra Reliable
Low Latency Communication (URLLC) connectivity, and massive Machine
Type Communication (mMTC). Massive IoT aims at optimizing networks
and devices for connectivity with billions of low-cost devices with long
battery life. URLLC will cross the chain for robust low latency commu-
nication, enabling, for example, wireless robots. Further, mMTC makes
it more economical to connect a very large number of objects wirelessly
to the network than with previous generations. These new services and
capabilities have enabled customers to digitalize and automate their pro-
cesses, such as manufacturing, construction, and logistics. In this regard,
industry verticals have shown strong interest in 5G and plan to set up
their own private dedicated wireless network or to use operator spectrum
and networks with slicing capabilities.
Three major approaches to increase data rate and/or throughput in 5G
wireless communication networks are listed in the following;
• Spectral efficiency enhancement: For eMBB usage, the mini-
mum requirements for peak spectral efficiency are 30 bit/s/Hz for
downlink and 15 bit/s/Hz for uplink, according to ITU report [5]. A

17
Introduction

key technology for boosting the spectral efficiency of 5G is the use of

massive multiple input multiple output (MIMO) beamforming. The
use of multiple antennas at the transmitter and receiver in a massive
MIMO system allows multiple data streams or users to share the
same frequency band [6, 7, 8]. In practice, massive MIMO is capable
of operating at frequencies above 1 GHz in the base stations, as well
as at millimeter waves in the devices [2, 9, 10]. The introduction of
Full Duplex (FD) is also a promising method of improving spectrum
efficiency and flexibility to overcome the limitations of conventional
duplex operation [11]. It allows both downlink and uplink connec-
tions to operate simultaneously at the same frequency, which could
theoretically double the spectrum efficiency.

• Ultra-densification: The ultra-densification of the network helps

alleviate congestion in cities with high densities of traffic demand
as well as during major events where a large number of users are
attempting to access the network simultaneously [12, 13]. In areas
where macrocells are overloaded or their signals might be weak,
blocked, or coexist with a high level of interference, a greater amount
of coverage will be provided by adding a large number of small cells.
Ultra-densification has contributed to improving the throughput of
communication networks throughout the 4G and 5G and is expected
to be adopted further by 5G+ networks. Some technologies can period-
ically or permanently form small cells that are adjustable to varying
traffic patterns, including moving relays, hotspots, and unmanned
aerial vehicles (UAVs). Therefore, a network planning approach is
essential for the placement of small cell Base Stations (BS)s in 5G
and 5G+ mobile communication networks.

• New spectrum: The 5G standard utilizes millimeter waves (mmWaves)

in addition to RF bands ranging from 0.4/GHz to 6/GHz. There is
much more spectrum available with mmWave (30 − 300 GHz) com-
pared to lower frequencies, which allows for ultrafast data rates
of up to a few terabits per second, as well as massive connectivity
[14, 15, 16]. The mmWaves, however, are affected by atmospheric
absorption and signal blocking resulting from obstructions. Further-
more, it is anticipated that 5G+ networks will utilize terahertz (THz)
and optical wireless (OW) frequency bands. THz bands (0.1-10/THz)
have a speed of one terabit per second, and they support a high
density of devices, making it a suitable technology for small cells.
In spite of this, THz bands have a limited propagation range and
are not capable of penetrating large obstacles [17, 18]. As a final
point, the OW band includes infrared (300 − 400 THz), visible light
(400 − 700 THz), and ultraviolet (700 − 30000 THz) frequencies that can
be used to transmit data at speeds exceeding several gigabits per sec-

18
 Introduction

Figure 1.1. (a) RF small cells outdoor and a the issue of interference level, (b) an optical
wireless small cell indoor and the issue of seamless coverage.

ond [19]. However, the OW band is less suited to environments with

a high level of mobility due to its short range and blocking caused by
opaque objects.

It is expected that outdoor small cells will be smaller and denser as a

result of the increasing demand for wireless communication services, as
well as using less power at the BSs. The 4G network utilizes RF to achieve
ultra-densification, so that macro-cells and small cells transmit over sub-
6GHz frequency bands to provide coverage for outdoor service areas. The
mmWave band, however, is envisioned to be an appropriate transmission
option for ultra-densification in 5G networks due to its shorter range
and higher data rate. The result is that 5G and 5G+ small cells are ex-
pected to experience a greater level of interference and, therefore, have
a lower signal-to-noise-plus-interference ratio (SINR) than 4G networks,
see Figure 1.1(a). It will then become increasingly difficult to determine
the optimal locations of small cell BSs that maximize coverage probabil-
ity, minimize latency, are compatible with the geometric parameters of
the outdoor environment, and are able to accommodate dynamic traffic
demand.
The use of new spectrum such as mmWaves, terahertz, and optical
wireless will provide unprecedented bandwidth and data rates in 5G and
5G+ wireless networks. Signals in these frequency bands have a short
range, making access points operating on these frequencies a good option
for forming femtocells, picocells, and ottocells (visible light bands), in which
the distance between the access point and the user is small, especially for
indoor applications. However, networks operating in these frequency bands
require Line-of-Sight (LoS) links between the access point and the users in
order to ensure reliable communication/accurate positioning. Figure 1.1(b)
illustrates a Light Emitting Diode (LED) transmitting data to users within
an indoor environment. It can be seen that the device on the table is able
to receive data, while the two users are unable to do so for various reasons,

19
Introduction

including one of the users falling outside the field of view of the LED, and
the other experiencing signal blockage. Therefore, while new spectrum may
provide high data rates and lower interference levels, ensuring seamless
coverage of indoor service areas remains a significant challenge.

1.2 Network Planning in Wireless Cellular Systems

In the literature, planning a mobile communication network mainly in-

volves determining the location of base stations and antenna configura-
tions in consideration of a variety of requirements and objectives. There
have been extensive efforts for network planning in the previous genera-
tions of mobile communication networks, such as in the Universal Mobile
Telecommunications System UMTS (3G) and Long Term Evolution LTE
(4G). Network planning is known as an NP-hard problem. Due to this,
researchers unanimously modeled the network area as demand nodes,
representing the center of an area characterized by a given traffic demand
[20]. Here, we investigate several research works on the field of mobile
communication network planning in the literature.
As for UMTS, authors in [21] considered a candidate set of locations along
with their configuration, forming a mixed integer programming problem
and solving it using Tabu Search. According to [22], the BS configuration
problem can be modeled as a linear integer programming problem, closely
related to the Minimum Dominating Set Problem, and can be solved with
linear programming software. Researchers in [23] developed nonlinear
optimization algorithms to locate BSs, and studied the solutions using
Hooke and Jeeves’ method, quasi-Newton, and conjugate gradient search
methods. Additionally, genetic algorithms are used to optimize antenna
and system parameters in [24]. The authors in [25] analyze a co-siting
problem, which consists of adding BSs on top of an existing UMTS network,
and formulate it using both integer and continuous optimization variables.
Network planning approaches have become increasingly sophisticated
as mobile communication networks evolve into LTE. The authors in [26]
used the simulated annealing (SA) algorithm to optimize the positions
of LTE base stations. According to [27], the authors propose a software-
based network planning approach, namely, Atoll, that uses accurate and
approximate algorithms to ensure high coverage and high quality of ser-
vice. Based on the uncertainty of the traffic distribution over time, a
stochastic approach is proposed in [28] for LTE BS locations. Finally, sta-
tistical geometry and Voronoi diagrams have been extensively employed
by researchers for planning, analyzing, and optimizing wireless cellular
networks [29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40]. However, a network
planning approach that takes into account dynamic traffic distribution
with different network requirements, such as capacity, coverage, and in-

20
 Introduction

terference, remains to be developed. Moreover, the literature has not yet

examined network planning that takes LoS visibility into consideration as
a stringent requirement of higher frequency networks.

1.3 Contribution of the Dissertation

In this dissertation, we examine and develop mathematical tools for wire-

less network planning based on a wide range of performance objectives,
requirements, and limitations common to wireless networks, including
seamless coverage, adequate capacity, a non-uniform distribution of users,
interference, and physical obstacles. In addition, we propose methods for
assessing the optimality of the resulting network planning. Wireless net-
work planning includes determining the number of access points and their
placements to simultaneously achieve the network objectives and to meet
its requirements, such as ensuring seamless coverage, maximizing band-
width utilization, reducing energy consumption, and lowering CAPEX and
OPEX of 5G and 5G+ mobile communication networks [41, 42]. In general,
network planning is constrained by these competing requirements, which
require the definition of trade-off parameters as part of the optimization
process. A primary benefit of wireless network planning involves ensuring
seamless coverage, i.e., a reliable and strong signal is received at all points
in the service area. Wireless network coverage mainly depends on factors
such as antenna power, distances between users and access points, channel
characteristics, and interference. Since wireless networks have limited
capacity, load balancing is another objective of wireless network planning
that involves sharing traffic equally between cells. In densely populated
areas, such planning facilitates the efficient use of bandwidth to allow
multiple simultaneous connections without compromising performance.
Additionally, wireless network planning can minimize electromagnetic
interference caused by neighboring access points, network slices, or other
sources.
There are three major phases in the network planning process, namely
dimensioning (identifying how many access points will be installed), de-
tailed planning (determining where the access points will be placed), and
post-planning (optimization and configuration). Planning a network is
a Non-deterministic Polynomial-time (NP)-hard problem, which entails
an exhaustive search to achieve the optimal solution [43]. Due to the
large dimensions, parameters, and diverse requirements of networks, it is
impossible to conduct an exhaustive search to determine optimal network
planning. Nevertheless, mathematical tools and optimization models can
help identify near-optimal solutions. It is through the use of these tools
that network planning complexity is reduced significantly, and network
operators can configure network parameters in real time.

21
Introduction

This dissertation contributes in two directions:

• Planning of RF-based outdoor wireless cellular networks. A major
advantage of the sub-6GHz bands is that they offer a wide coverage
area, and the signals penetrate through walls and obstacles, mak-
ing these bands suitable for outdoor wireless network areas. As a
downside, these frequency bands are limited in bandwidth, resulting
in low data rates, and the poor SINR further reduces the data rate
around the edges of the cells. It is possible to overcome these limita-
tions in sub-6GHz bands through densification, which is the use of a
large number of low power small cells, such as microcells femtocells,
and picocells, to offload macrocells and to provide higher data rates
and improved Quality of Services (QoS). This will result in 5G and
5G+ networks having a large dimension, with irregular and dynamic
demands and requirements, resulting in a risk of overloading and
high interference levels. In order to address these challenges effec-
tively, we developed optimization methods based on Voronoi diagrams
and Delaunay triangulation that are capable of placing BSs that are
compatible with non-uniform spatial distributions of users with high
levels of quality of service. As a result of these optimization methods,
the coverage probability is maximized across the entire service area
while the cell loads are balanced.

• Planning of new spectrum-based indoor wireless networks. When us-

ing new spectrum such as mmWaves, THz, or OW for indoor wireless
networks, seamless LoS coverage is required, which means maintain-
ing a LoS link between access points and users within the service
area. When planning an indoor wireless network, it is necessary to
consider the size and geometric properties of the indoor environment,
such as walls and obstacles. To address this challenge, we develop
a graph that models LoS propagation in an indoor environment. In
light of this graph model, we determine the minimum number and
location of access points to ensure LoS coverage in an indoor environ-
ment. Additionally, we extend the graph model to form a connectivity
tree, by which we are able to determine the minimum number of ac-
cess points and their placements to guarantee LoS conditions in both
access and backhaul. With the help of this graph modeling technique,
it is also possible to determine the deployment of positioning access
points that will ensure precise positioning. Finally, graph modeling
is used to evaluate indoor network planning optimality.
It is critical to note that this dissertation focuses solely on developing the-
oretical tools and evaluating their results via simulations. Work regarding
experimental verification is left open for future research.
In publications I, II, and III, contributions are made to the planning of
RF-based outdoor wireless cellular networks with irregular service demand

22
 Introduction

distributions. In publication I, we propose an iterative method based on

power Voronoi algorithm to determine the optimal location of BSs. In
this method, we update the power Voronoi weights to achieve uniform
volume shares and load distributions. In publication II, a variety of server-
side and user-side requirements are taken into account when optimizing
BS location, which is accomplished by defining trade-off parameters for
the optimization problem. In publication III, a method for optimizing
the BS locations is presented that utilizes both Voronoi diagrams and
Delaunay triangulations in order to simultaneously balance the volume
shares and maximize the coverage. For coverage maximization, we define
local regularity optimization problem and solve it through particle swarm,
farthest point, average point, and spatial median methods.
Publications IV, V, and VI contribute to the planning of new spectrum-
based indoor wireless networks. According to publication IV, a visibility
graph modeling approach is proposed to determine the minimum number
and location of Visible Light Communication (VLC) access points to provide
reliable indoor LoS coverage. In publication V, we determine the minimum
number of VLC access points and their locations required for access and
backhauling to meet LoS requirements. Consequently, a connectivity
tree is considered to ensure that each VLC access point is capable of
communicating with (another) access point via LoS infrared wireless links.
Publication VI considers the deployment of positioning access points as
a method of guaranteeing an LoS condition to one, two, or three access
points in order to improve the precision of indoor positioning systems that
utilize proximity, triangulation, or trilateration principles to determine a
user’s location. As two configuration parameters, a minimum distance and
a minimum angle between the access points will guarantee the desired
positioning accuracy.

1.4 Structure of the Dissertation

This dissertation consists of five chapters, including this introductory part

as well as six original publications that contribute to wireless network
planning for 5G and beyond. Chapter 2 presents the background to channel
modeling in wireless communications and positioning methods. The pro-
posed methods for planning RF-based outdoor cellular wireless networks
are discussed in Chapter 3. In this chapter, mathematical tools such as
Voronoi diagrams and Delaunay triangulation are introduced, along with
major cellular network requirements such as capacity and coverage. Then,
strategies to simultaneously achieve load balancing and maximize cover-
age are described. By modeling the indoor environment with a visibility
graph, Chapter 4 proposes network planning methods for new spectrum-
based indoor wireless networks. This chapter begins by describing the

23
Introduction

VLC channel model and providing an overview of the Art Gallery Problem.
Thereafter, it describes the construction of the visibility graph and the
planning method for ensuring LoS conditions indoors. During the final
sections of this chapter, the visibility graph is used to determine the de-
ployment of positioning access points for a precise positioning objective.
Finally, Chapter 5 presents the conclusions drawn from this dissertation
and future research directions.

24
2. Background

With the development of many emerging applications, accurate positioning

and reliable communication have become more crucial. For industrial
automation, robotics, and autonomous vehicles to operate safely and ef-
ficiently, real-time access to both communication and positioning data
is essential. Future 6G wireless networks are expected to implement
joint communication and positioning, also known as communications and
positioning or communications and localization techniques. From the per-
spective of communication, we present an overview of channel modeling
and how its elements contribute to estimating achievable data rates. As a
next step, we will examine the background of accurate positioning methods.

2.1 Communications

Due to the ultra-fast data speeds and low latency requirements of 5G and
5G+, a thorough understanding of the propagation environment is essen-
tial for designing, planning, and optimizing wireless networks. Channel
modeling can be achieved through different techniques such as extensive
measurements, statistical methods, machine learning, and ray tracing.
We present an overview of RF channel models. Then, we present channel
modeling for visible light communication using a LED. It is noteworthy
that researchers are still working on channel modeling for mmWaves and
THz frequency bands [44, 45, 46].

2.1.1 Radio Frequency Channel Modeling

Wireless communication channels in RF bands are characterized by fading,

which is the change in signal attenuation as a function of variables such
as time, geographical location, and frequency. Fading is often modeled as
a random process. In sub-6GHz, fading channels are generally divided
into three categories, namely mean path loss, shadowing, and multi-path
fading, see Figure 2.1.

25
Background

Figure 2.1. RF channel modeling.

• Mean path loss models provide a statistical expectation of the fading

process. The single-slope model is the most commonly used mean
path loss model in homogeneous environments. The mean path loss
ion the dB scale is given by

L p = L 0 + 10n log10 (d/d 0 ), (2.1)

where L 0 represents the mean path loss at reference distance d0 , d

refers to the distance between the transmitter and the receiver, and n
represents the path loss exponent, which depends on antenna height,
carrier frequency, and propagation environment. In the case of free
space communication, we have d0 = 1, L 0 = 32.44 + 20 log10 ( f MH z ), and
n = 2, i.e., the received power drops with the square of the distance.
Therefore, a stronger mean path loss is experienced by the signal as
the carrier frequency increases, resulting in a smaller cell coverage
area.

• Shadowing occurs when obstacles with size from tens to hundreds

of wavelengths cause a variation in path loss around the mean path
loss L p . Shadowing is random, but it is correlated when measured
at nearby locations. Shadowing can be described as a log-normal
distribution with the probability density function as

L2s
−
1 (2.2)
e 2σs ,
2
f (L s ) = ⎷
2πσs

wherein L s denotes the shadowing effect in [dB].

• Multi-path fading occurs when the transmitted signal travels to

the receiving antenna through different paths. When some elements

26
 Background

in the propagation elements move, the pattern of constructive and

destructive interference created by the different paths varies, thereby
affecting the strength of the received signal [47]. Multipath fading
can affect all frequencies, starting at low frequency to microwave
and beyond. The components of these waves are formed by reflection,
scattering, and diffraction causing phase distortions in the signal and
inter-symbol interference (ISI). Thus, the strength of the transmitted
signal is determined by the sum of the component signals having
different amplitudes and phases. Multi-path fading models are char-
acteristic of dense urban environments since there is less likelihood
of establishing a LoS link between the transmitter and receiver. It is
possible to experience variations in the range of tens of decibels when
the receiver (or obstacles) shift its location. When there is no LoS
link, the multi-path fading is modeled as the Rayleigh distribution
given by
a2
2a − σ2 (2.3)
f (a) = 2 e a , a≥0
σa
where a and E[a2 ] = σ2a denote the amplitude, and the mean power of
the channel gain P r /P t , respectively. In case the component signals
include the LoS link, multi-path fading follows the Rician distribu-
tion.

More advanced channel models in sub-6GHz bands include the 3GPP

spatial channel model [48], WINNER II [49], WINNER+ [50], 3GPP three-
dimensional (3D) cellular channel models [51], IMT-A [52], and COST 2100
[53]. Fading channel modeling has been applied to publications I, II, and
III.

2.1.2 Visible Light Communication Channel Modeling

Visible Light Communication (VLC) is primarily intended for LoS links

between transmitters and receivers in an indoor environment [54]. While
there is considerable literature that addresses NLoS in VLC channel
modeling [55], the LoS signal is typically much stronger than the reflected
signal [56, 57]. The LoS link is also essential to facilitate an accurate and
unbiased Visible Light Positioning (VLP) system. Therefore, we are mainly
focusing on LoS channel characteristics of VLC systems.
Based on the Lambertian radiation pattern, the DC gain of the optical
channel between a Phosphor-Converted PC-LED transmitter and a photo-
detector (PD) is given by
{︄
(m L+1)A PD
dir 2π d 2
cosm L (φ) cos (ψ), 0 ≤ ψ ≤ Ψ,
HLED,PD (0) = (2.4)
0, ψ > Ψ,

27
Background

Figure 2.2. VLC channel model via Lambertian radiation pattern [58].

[︁ ]︁
where m L = −1/ log2 cos(θmax ) denotes the Lambert index of the LED, in
which θmax [rad] defines the source radiation semi-angle at half power of
the LED. Besides, φ [rad] and ψ [rad] refer to the angle of irradiance and
incidence of the LoS link, respectively, see Figure 2.2. Furthermore, Ψ [rad]
denotes the Field of View (FoV) semi-angle of the PD with an effective
physical area of A PD [m2 ]. Thus, this network has a maximum range of
r = h LED-PD × tan(Ψ), where h LED-PD and d are the height and distance,
respectively, between LED and PD. Therefore, a PD located beyond this
range may receive no data from the LED. Then, the DC current at the
output of the PD is calculated as
∫︂ λu
i dir
PD (0) =
dir
PLED HLED,PD (0) S o(W) (λ) R PD (λ) f o (λ) d λ. (2.5)
λl

where, PLED and So(w) (λ) are the total radiant power and spectral power
distribution of the LED, respectively. Furthermore, RPD (λ) denotes the
responsivity of the PD, and f o (λ) is the transmittance of the optical pass-
band filter with lower (λl ) and upper (λu ) cutoff wavelengths. Then, the
power of the electrical signal at the output of PD is given by
⃓ ⃓2
ΛPD (0) = ⃓ i dir ⃓
PD (0)G tia , (2.6)

where G tia refers the gain of transimpedance amplifier within a PD. Tran-
simpedance amplifiers convert the current output of PDs, which are light-
sensitive devices, into voltage outputs that are later processed by the
receiver.

28
 Background

2.1.3 Calculation of the Data Rate

Generally, the achievable data rate is upper-bounded by Shannon capacity.

It may be possible to achieve data rates close to Shannon capacity by using
suitable modulation, coding, equalization, and near-ideal filters. In this
regard, the data rate is calculated as
(︁ )︁ (︂ Pr )︂
R data < B log2 1 + SINR = B log2 1 + , (2.7)
B(I + N0 )
where P r [Watt] denotes the received power, B refers to the channel band-
width, and N0 [Watt/Hz] and I [Watt/Hz] represent the power spectral
density of the white Gaussian noise and the interference, respectively.
Thus, in wireless networks, data rates are low at cell edges where received
power is weak and interference from neighboring cells is strong. Maxi-
mizing SINR near the cell edge is therefore an essential part of network
planning success.
Data rate in VLC channels can be calculated by substituting P r in (2.7)
with optical received power (2.6). Here, it can be observed that when the
PD is outside the field of view, no data can be received, as the received
power is zero. Indoor scenarios have fewer users, less interference, and
larger bandwidth available. Therefore, in the VLC case, the data rate
received from the LED is high if the user falls within its field of view.
It is, therefore, a necessity to ensure that the indoor wireless networks
that use the new spectrum provide seamless coverage in such a way that
a user anywhere within the indoor space may establish a LoS link to a
transmitter (LED) within its range.

2.2 Positioning

In recent years, Global Navigation Satellite Systems (GNSS), such as

GPS and/or Galileo, have been used standalone or in conjunction with
cellular networks to provide positioning services [59, 60]. Unfortunately,
satellite-based positioning systems provide only reliable coverage outdoors
and cannot be used for indoor 5G applications, including warehouse logis-
tics and smart manufacturing [61]. For this purpose, a new Positioning
Reference Signal (PRS) was defined to be used by various 5G positioning
techniques, such as Roundtrip Time (RTT), Angle of Arrival (AoA), Angle
of Departure (AoD), and Time Difference of Arrival (TDOA) [62]. The posi-
tioning accuracy of all these methods is notably affected by the existence of
a LoS condition between the positioning access points and User Equipment
UE [63]. Specifically, when it comes to indoor positioning networks, the
deployment of access points becomes more challenging due to walls and
other large obstacles blocking the LoS component [60]. This is because,

29
Background

in the case of a non-line-of-sight (NLoS) condition, both time and angle

measurements become biased, notably increasing positioning uncertainty.
On the other hand, positioning via higher frequency bands and band-
widths provides much better time resolution when calculating propagation
time (RTT) and time differences (TDoA) over-the-air [64]. In addition, at
high carrier frequencies, packing large-size antenna arrays becomes more
feasible, enabling finer resolution in the angular domain [65].

2.2.1 Power-Based Positioning Systems

Received Signal Strength (RSS) at the UE is used to estimate distance

in power-based positioning systems. This type of positioning system can
take advantage of existing infrastructure, thereby reducing costs and
enhancing sustainability. Power-based positioning systems also benefit
from not synchronizing clocks. Despite this, these systems are susceptible
to errors due to changes in the medium, fluctuations in the power supply,
as well as channel parameters such as shadowing and multi-path fading.

2.2.2 Time-Based Positioning Systems

Using time-based positioning systems, the signal’s travel time is multi-

plied by light speed to determine its distance. The distances between the
UE and the access points are then used to estimate the UE’s location.
With time-based measurements, a high degree of precision is achieved,
which is inversely proportional to the signal bandwidth. Nevertheless,
several factors affect the accuracy of the measurements, including multi-
path propagation, NLoS links between the access points and the UE, and
inaccuracies in device clocks.
Time of Arrival (ToA) technique uses the time at which the signal is trans-
mitted and the time a receiver detects the signal has been received. The
distance between the access point (Tx) and the UE (Rx) is then calculated
by
d = (τTx − τRx ) × c, (2.8)
where c denotes the speed of light, τTx refers to the transmission time at
the access point, and τRx is the receiving time at the UE. To obtain the exact
times τTx and τRx , the ToA technique requires accurate synchronization
between the internal clocks at the access point and the UE, which entails
complex hardware. Alternatively, the Round-Trip-Time (RTT) technique
calculates the total propagation time from the access point to the UE and
vice versa. Then, RTT is given by

t RTT = 2(τTx − τRx ) + δ, (2.9)

where δ is the response time of the UE, which is assumed to be known. So,

30
 Background

Figure 2.3. 2-dimensional tri-lateration positioning principle; (a) circles centered in each
access point used in RTT method and (b) hyperbolas focused in each pair of
access points used in TDoA method.

by measuring tRTT , the distance can be calculated as

t RTT − δ
d= × c, (2.10)
2

which does not need exact measurements of τTx and τRx .

Time Difference of Arrival (TDoA) is the third technique in the time-based
positioning system. This technique involves measuring the difference
between the ToAs of several different access points. In this regard, let two
access points simultaneously send signals, and the UE receive them at the
measured times τ(1) Rx (2)
ToA = τ1 + ∆ and τToA = τ2 + ∆, where τ1 and τ2 can be
Rx Rx Rx

interpreted as the real received signal times from the access points 1 and 2,
in case the UE was synchronized with the access points. Here, we assume
that the access points are well synchronized with each other, for example,
by being connected through a wired network to synchronize the clocks. Due
to this, the same synchronization time value ∆ between the access points
and the UE appears in the time measurements τ(1) (2)
ToA and τToA . Therefore,
the distances between the UE and the access points 1 and 2 are given by
d 1 = (τTx − τRx Tx Rx Rx
1 ) × c and d 2 = (τ − τ2 ) × c, respectively. However, τ1 and τ2
Rx

are unknown and therefore d1 and d2 cannot be calculated independently.

Alternatively, the distance difference in the TDoA technique is derived as
(︁ )︁ (1) (2)
d 2 − d 1 = (τTx − τRx Tx Rx Rx Rx
2 ) − (τ − τ1 ) × c = (τ1 − τ2 ) × c = (τToA − τToA ) × c, (2.11)

wherein the both measurement values τ(1) (2)

ToA and τToA are known at the UE.
Therefore, in the TDoA technique, the UE calculates the differences of the
distances to a pair of access points with no need for time synchronization.

2.2.3 Positioning Principles

Regardless of the method used to estimate the distance or direction to-

wards the UE, positioning techniques are broadly classified according to

31
Background

the fundamental principles employed to estimate the position: proximity,

triangulation, and multilateration [59]. The 3GPP Cell Identification CID
method uses the proximity principle and assigns the UE to the known
position of the nearest access points [66]. In contrast, the 3GPP positioning
method relies on the triangulation principle and AoA (AoD) measurements;
here, the UE’s position is assigned to the intersection of the directions from
where the reference signal is being received (transmitted) at two access
points [67]. Finally, the multilateration principle is used in 3GPP RTT and
TDoA methods, which assigns the UE’s position to the intersection of at
least three geometrical objects. For RTT, as shown in Figure. 2.3(a), these
objects represent circles centered on each access point, while for TDoA, as
shown in Figure 2.3(b), they represent hyperbolas with focuses at each
pair of access points [59].

32
3. Planning of RF-Based Outdoor
Cellular Wireless Networks

3.1 Introduction

In the years leading up to 5G, the number of mobile users has grown
rapidly, along with the wide variety of mobile applications seamlessly
integrated into everyday life, contributing to the rapid development of new
radio access technologies. Base stations in cellular networks are typically
installed at predetermined locations, while data traffic is highly diverse,
non-uniform, dynamic, and irregular in nature [41]. For instance, it is
possible that the location of hot spots within the target area could change
as a result of the events taking place in the area. As a consequence, certain
cells may experience overloading while others may experience low loads,
which leads to the inefficient use of capacity and energy.
Planning a cellular network is designed to determine the number and
location of BSs needed to meet users’ traffic demands while maintaining a
high level of QoS. Ideally, all BSs should be equally loaded while maximiz-
ing coverage probability. If traffic demand is uniformly distributed within
the target area, it may be possible to achieve this goal by placing the BSs
in a regular grid or triangular lattice (hexagonal pattern) [68]. There is,
however, no analytical method for determining the optimal location of BSs
when traffic does not follow a uniform pattern. The planning of wireless
cellular networks is generally recognized as an NP-hard problem, mean-
ing that it is not possible to find the optimal solution in polynomial time.
There are two major factors to consider when designing a near-optimal
cellular network planning. As part of the network planning process, the
capacity at each cell must be met, which is a function of demand distri-
bution, interference, and resource availability. Second, network planning
requires ensuring coverage, i.e., a user anywhere within the service area
experiences a sufficient level of quality of service [5]. For the planning,
analysis, and optimization of cellular wireless networks, researchers have
extensively employed mathematical tools such as statistical geometry and

33
Planning of RF-Based Outdoor Cellular Wireless Networks

Voronoi diagrams [69, 29, 30, 31, 32, 33, 70, 34, 35, 36, 37, 38, 39, 40].
Throughout this chapter, we propose algorithms for determining near-
optimal cell planning for cellular networks, taking different requirements
into account, such as capacity and coverage.
The structure of the rest of this chapter is as follows. A brief introduction
to Voronoi diagrams and Delaunay triangulation is presented in section
3.2. We investigate the major requirements of wireless cellular networks
in Section 3.3 including capacity and coverage. We then propose different
methods of cell planning in sections 3.4, 3.5, and 3.6 for meeting capacity,
coverage, or variable network requirements. Finally, a summary of the
chapter is given in section 3.7.

3.2 Geometrical Tools

The study of wireless cellular networks requires the use of mathematical

tools to model associations between BSs and users. These associations can
be represented using Voronoi diagrams and Delaunay triangulations.

3.2.1 Voronoi Diagram

Voronoi diagrams are mathematical concepts originally found in computa-

tional geometry and geographic information systems. A Voronoi diagram
divides a plane into regions based on distances between each point and
the generators, seeds, or sites. These regions are distinguished by Voronoi
edges, which define the tessellation of the plane, dividing it into a collection
of non-overlapping cells, each associated with a distinct site.
In the context of cellular wireless networks, a Voronoi diagram is defined
as a method of assigning points within a region to BSs based on their
location X = { x1 , x2 , ..., x N } in the area, where N refers to the number of BSs.
Then, power Voronoi diagram in two dimensions is formulated as

J i ≜ {(x, y) ∈ J|∥(x, y) − x i ∥2 − w i ≤ ∥(x, y) − xk ∥2 − wk , ∀ i ̸= k}, (3.1)

where x i denotes the location of i th BS with weight w i , see [71]. Also, J i

⋃︁
refers to the i th cell such that J = N i =1 J i denotes the whole service area
and J i ∩ Jk = ∅ for i ̸= k. By assuming unweighted power Voronoi w i = 0 for
all i in (3.1), we determine the cells of the ordinary Voronoi diagram as

J i ≜ {(x, y) ∈ J|∥(x, y) − x i ∥ ≤ ∥(x, y) − xk ∥, 1 ≤ i ≤ N, ∀ i ̸= k}. (3.2)

Therefore, each cell of ordinary Voronoi diagram represents the set of

points closest to the associated BS. In Figure 3.1(a), a Voronoi diagram
is shown for a set of N = 23 sites (BSs), resulting in 23 Voronoi cells, as
defined in (3.2). We associate a mobile user located at each Voronoi cell to
the associated BS.

34
 Planning of RF-Based Outdoor Cellular Wireless Networks

(a) (b)
Figure 3.1. A given set of sites (BSs) generating (a) Voronoi diagram and (b) Delaunay
triangulation, [72].

3.2.2 Delaunay Triangulation

Delaunay triangulation is a geometric technique used to triangulate a

set of sites in the plane such that no site lies within the circumscribed
circle of any triangle [71]. Using Delaunay triangulation, the triangles
formed between the sites are as equilateral as possible, making it an
attractive choice for computational geometry applications such as mesh
generation, pattern recognition, and computer graphics. As Figure 3.1(b)
shows, a Delaunay triangulation is derived by connecting any two sites
that share a Voronoi edge, and thus Delaunay triangulations are dual to
Voronoi diagrams. The most important properties of Voronoi diagrams and
Delaunay triangulations are listed as follows.

• Uniqueness: For any set of sites X , there exists a unique Delaunay

triangulation.

• Duality with Voronoi diagrams: For Delaunay triangles formed by

X , the center points of circumscribed circles are the Voronoi vertices
of X .

• Co-linearity: Co-linear sites result in unbounded Voronoi cells.

• Convex hull: The union of Delaunay triangles, generated from X ,

forms the convex hull (CH) of the set of the sites CH(X ).

• Boundary of convex hull: If two sites from X with unbounded

Voronoi cells share a Voronoi edge, then the Delaunay edge between
them partially creates the boundary of CH(X ) and vice versa.

• Empty circumscribed circle: All circumscribed circles of Delaunay

triangles of X are empty, i.e., no site in X lies inside any circum-
scribed circle.

35
Planning of RF-Based Outdoor Cellular Wireless Networks

• Largest empty circle: The largest empty circle inside CH(X ) is the
largest circumscribed circle of Delaunay triangles for X . In other
words, the radius of the largest empty circle represents the greatest
possible distance between a point in the area and its associated site
(BS).

Later, we use Delaunay Triangulation to maximize coverage through the

whole service area.

3.3 Requirements of Cellular Wireless Networks

This section discusses two major requirements for wireless networks,

namely capacity and coverage.

3.3.1 Capacity

The capacity of a wireless cellular network refers to the total amount of

traffic that can be handled at any given time [73]. RF spectrum availability,
network infrastructure, network architecture, and antenna technology all
play a role in determining wireless network capacity.
The service demand distribution δ(x, y) is defined as the amount of data
to be transmitted to the user with location coordinates (x, y) within the time
interval under consideration. An assumption is made that the distribution
of service demand in the target area is known statistically, i.e. a probability
density function can be used to determine its spatial distribution. The
volume of traffic within the service area, defined as the average traffic, is
an important factor in determining the number of BSs to be installed, a
process known as dimensioning. Depending on the network topology and
the demand distribution, the volume share is determined at each cell as
∫︂
Vi = δ(x, y)dxd y. i = 1, ..., N (3.3)
Ji

According to (3.3), the volume share is determined by the demand distri-

bution, regardless of available resources within the cellular network, the
distance between BSs and users, interference, and BS power.
In cellular wireless networks, cell load refers to the amount of traffic or
data demand within a particular cell or coverage area compared to the
network’s capacity. Cell loads vary based on the number of users and
devices accessing the network and the amount of data they transmit at any
given time. Therefore, if the cell load exceeds the capacity of the network,
it can lead to network congestion, decreased performance, and slower data
speeds, a situation known as a cell overloading. Based on the load-coupling
model, cell load values are calculated as the fraction of time-frequency
resources consumed over total resources at each cell [74]. Furthermore,
cell load is interpreted as the probability that the signals transmitted by a

36
 Planning of RF-Based Outdoor Cellular Wireless Networks

BS interfere with the receiving signals of the user devices in other cells.
Taking the advantage of the connection between the two interpretations,
the coupling relation between cells is formulated as
∑︂ δ(x, y)
ρi = , i = 1, ..., N
P ib h i (x, y) (3.4)
(x,y)∈J i K B log2 (1 + ∑︁ )
b
k∈N\ i P k h k (x, y)ρ k + σ
2

where P ib and ρ i denote the transmit power and cell load value at i th BS,
h i (x, y) refers to the channel gain from i th BS to a user at location point
(x, y), and σ2 represents the noise power. The values B and K refer to the
system bandwidth and total number of resource units in time-frequency
domain, respectively. As (3.4) shows, load values are non-linear functions
of the load values in the other cells, demand distribution, power allocated
to the BSs, system bandwidth, amount of resource units, and channel
coefficient between the users and BSs. So, the load coupling system (3.4) is
written in a compact form as

ρ = f (ρ ), (3.5)

where ρ = [ρ 1 , ρ 2 , ..., ρ N ]T and the right-hand-side terms in (3.4) are ex-

pressed as f (ρ ) = [ f 1 (ρ ), f 2 (ρ ), ..., f N (ρ )]T . Since the solution to (3.5) is not
straightforward to derive, it is derived through a convex optimization
problem as
N
∑︂
minimize ρi (3.6a)
ρ
i =1

s.t.: ρ − f (ρ ) ⪰ 0, (3.6b)
N
ρ ∈ R+ , (3.6c)

where the convexity of the problem, conditions for the solution existence
and the solution uniqueness of (3.6) are studied in [74]. For (3.6), its
solution space Sρ ⪰ f (ρ ) is also referred to as the feasible load region. From
(3.4), it follows that f (ρ ) is strictly increasing with f (0) ⪰ 0 in case the
noise power is non-zero. Hence if Sρ ⪰ f (ρ ) ̸= ∅, then for any optimal solution
to (3.6), (3.6b) holds with equality, as otherwise (3.6a) can be improved,
contradicting that the solution is optimal. In conclusion, any optimum of
(3.6) is a solution to (3.5).
The optimal solution to (3.6), denoted by ρ ∗ , is determined using an
iterative method ρ m+1 = f (ρ m ) for m → ∞ with ρ 0 = 0, where ρ m denotes the
load vector in m th update. Therefore, ρ ∗ exists if and only if ρ m converges.
From the network performance standpoint, the cellular network planning
is feasible with allocated resources if ρ ∗ falls within the interval 0 ⪯ ρ ∗ ⪯ 1.
Otherwise, the case that ρ ∗i ≥ 1 for some i implies that the network is
overloaded and thus, it is infeasible. Then, reconfiguration of the cell
planning is needed. Roughly, if ρ m diverges for large m, the network may
need increasing resources as well as network re-planning.

37
Planning of RF-Based Outdoor Cellular Wireless Networks

3.3.2 Coverage

Coverage in cellular wireless networks refers to the geographic area served

by a network BSs and can provide wireless services to mobile devices.
Network operators aim to provide as much coverage as possible while con-
sidering cost and network capacity. The coverage area of a cellular network
is a complicated function of the placement, number, and configuration of its
base stations, as well as factors such as terrain, building structures, and
atmospheric conditions. Therefore, no single formula represents coverage.
However, several methods can be used to estimate the coverage of a cellular
wireless network including:
• Propagation modeling: Based on factors such as the height of
the antennas at the base station and the frequency of the signal,
mathematical models are used to estimate the strength of radio
signals at different locations [68].

• Drive testing: The strength of the radio signal at various locations

is measured by a vehicle equipped with measurement equipment
that is driven through the coverage area [75].

• Simulation: An analysis of the network coverage is performed using

computer simulations based on parameters such as the location of
the base stations, antenna characteristics, and the radio environment
[76].

• Field testing: Tests are conducted at different locations within the

coverage area by making calls, sending texts, and using data services
[77].
Depending on a network’s particular needs and requirements, one of
these methods may be suitable, since each has its own strengths and
limitations. In this dissertation, given a signal to interference plus noise
ratio (SINR) value γ0 as a parameter, we calculate the coverage probability
in an abstract form as
∫︁
J I(γ(x, y) > γ0 )dxd y (3.7)
P c (γ0 ) = ,
A
where γ(x, y) denotes the SINR value at point of coordinates (x, y), I(.)
denotes the indicator function, and A refers to the area of total service
network.

3.4 Planning for Load-Balancing

The load balancing optimization problem in a non-uniform service demand

area is given by
minimize max ρ i . (3.8)
X 1≤ i ≤ N

38
 Planning of RF-Based Outdoor Cellular Wireless Networks

Algorithm 3.1. Centroidal Voronoi Algorithm

1: X (0) , X (1) ←− Random(J, N)

2: J(1) ←− Voronoi(X (1) ) % (3.2)
3: z ←− 2
4: While ∥ X (z−1) − X (z−2) ∥ > ϵ
5: X (z) ← Centroid(J(z−1) , δ) % demand centroids (3.9)
6: J(z) ← Voronoi(X (z) )
7: z ← z+1
8: End While

To find a solution to (3.8), we firstly evaluate the Centroidal Voronoi Al-

gorithm CVA for load balancing. Then, we propose two iterative methods,
namely, Channel-based Power Voronoi Algorithm CPVA, and Load-based
Power Voronoi Algorithm LPVA. The steps of CVA is illustrated in Algo-
rithm 3.1. Here, we choose BS locations X randomly within service area
and find the Voronoi diagram using (3.2). Then, we move each BS step by
step to the demand centroid of each cell calculated as
∫︁∫︁
J i (x, y)δ(x, y)dxd y
ci ≜ ∫︁∫︁ , i = 1, ..., N (3.9)
J i δ(x, y)dxd y

and determine the Voronoi diagram with the updated BS locations. We

continue this procedure until the BS locations converge. Although CVA is
fast, it fails to determine the optimal locations for the BSs.
We take into account the fact that the load imposed on a BS is determined
by the channel gain between users and the serving BS. Due to the relatively
low received power, users at the edge of the cell have a significant impact
on load values. So, we approximate the load values as
∫︂
δ(x, y)
ρ˜i = dxd y, i = 1, ..., N (3.10)
Ji K B log2 (1 + κ h i (x, y))
where κ denotes the source signal to user interference parameter. By
substituting ρ i in (3.8) by (3.10), we balance the approximate load values
in CPVA. As Algorithm 3.2 shows, we first run CVA as a starting point.
Then, we determine the power Voronoi diagram with the vector of weights
such that the cells with high load values capture larger weights and,
therefore, their sizes shrink. Similarly, the cells with smaller approximate
loads enlarge. Finally, we update the weights based on the approximated
loads. Although CPVA is fast, it fails to flatten the load values accurately.
Alternatively as a solution to (3.8), Load-based Power Voronoi Algorithm
LPVA determines the load values at each step using (3.4). Solving (3.4)
at each step makes LPVA a heavy task, especially for the service areas

39
Planning of RF-Based Outdoor Cellular Wireless Networks

Algorithm 3.2. Channel/Load-based Power Voronoi Algorithm

1: X (0) ←− Random(J, N)
2: X (1) ←− CVA % Algorithm 3.1
3: J1 ←− Voronoi(X (1) ) % (3.2)
4: z ←− 2, W(2) ←− 0
5: While ∥ X (z−1) − X (z−2) ∥ > ϵ
6: X (z) ←− Centroid(J(z−1) , δ) % (3.9)
7: J(z) ←− PowerVoronoi(X (z) , W(z) ) % (3.1)
8: ρ̃ (z) ←− Load(J(z) , X (z) , δ)
% (3.10) for CPVA or (3.4) for LPVA
(z)
M [ ρ̃ ] − ρ̃ (z)
9: W(z+1) ←− W(z) + S ×
M[ρ̃ (z) ]
10: z ←− z + 1
11: End While

4
10

50 2 50
10
100 100
y-position (m)

150 1.5 150 5

200 200
1 0
250 250

300 300
0.5 -5
350 350

400 0 400 -10

100 200 300 400 500 600 100 200 300 400 500 600
x-position (m) x-position (m)
(a) (b)
x y
Figure 3.2. Service area with (a) demand distribution δ(x, y) = exp(− ) and (b) mini-
100 100
− 4
mum SINR requirement distribution γ(x, y) = −10 + 10 x y[dB].

with a large number of BSs. However, as (3.4) is accurate, obtaining cell

planning with uniform load pattern is guaranteed. In Algorithm 3.2, the
function M[.] and S represent the mean value and a heuristic parameter
value, respectively. If S is set too large, algorithms will diverge, and if it
is set too small, processes will run slowly. Moreover, W(z) = [w1 , w2 , ..., w N ]
refers to the vector of Voronoi weight at step z, and ϵ > 0 denotes a small
enough value.
Figure 3.2(a) illustrates the demand distribution for a 600m × 400m area
with a higher demand density in the upper-right corner. Figure 3.3 com-
pares three cell planning approaches for the area with demand distribution
as in Figure 3.2(a) with N = 30 BSs. Using CVA as in Figure 3.3(a), it is
observed that two cells are too large and overloaded, see Figure 3.3(b). In
Figure 3.3(c), we apply CPVA with κ = 80dB. One can observe that although
CPVA avoids overloading among all the cells, it fails to obtain accurately
uniform load pattern due to load approximation. Finally, using LPVA in

40
 Planning of RF-Based Outdoor Cellular Wireless Networks

1.2

y-position (m)
100 1

Load value
0.8
200
0.6 0.5
300
0.4
400 0
100 200 300 400 500 600 0 10 20 30
(a) (b)
y-position (m)

100 1

Load value
200
0.5
300

400 0
100 200 300 400 500 600 0 10 20 30
(c) (d)
y-position (m)

100 1
Load value

200
0.5
300

400 0
100 200 300 400 500 600 0 10 20 30
x-position (m) BS index
(e) (f)
Figure 3.3. BS deployment and load patterns in a service area with demand
distribution, as in Figure 3.2(a) with N = 30 BSs, using (a,b): CVA,
(c,d): CPVA with κ = 80dB, (e,f): LPVA; the colors show the cell load
values.

Figure 3.3(e), all the cells have ideally the same load values. However, the
running time in LPVA heavily increases with the number of BSs and the
number of steps.
Here, we explain the ground for choosing the power to noise parameter
value κ. Since the load approximation values in (3.10) are the estimations
[︁ P ib ]︁
of the real load values in (3.4), we set κ = E x,y ∑︁ =
k∈N\ i P k h k (x, y)ρ k + σ
2

P ib
, where we assumed that the communication system is thermal noise
σ2
limited. We assume that the BS is transmitting with power P ib = 3 mW
per user communicating over 15 KHz bandwidth with noise power density
σ2 = −177 dBm/Hz at the receiver. Then, the BS transmission power to
noise parameter value is derived as κ = 80dB.

41
Planning of RF-Based Outdoor Cellular Wireless Networks

Finally, we provide a brief comparison of the complexity of the algorithms

presented. In two-dimensional space, it is noteworthy that calculating
Voronoi diagrams using a set of N points is of complexity O(N log(N)), [78].
Assuming that the complexity of computing the demand centroid of a re-
gion (3.9) is fixed, denoted by O(1), then the complexity of calculating the
demand centroids for the entire service area is O(N). Therefore, the com-
plexity of CVA and CPVA is O(N log(N)). In contrast, LPVA is dominated
by the calculation of the load vector with a complexity of O(N 3 ), [74]. As a
result, CPVA has much less computational complexity than LPVA.

3.5 Planning with Conflicting Network Requirements

In wireless cellular networks, demand and QoS requirements of the con-

nected devices are typically non-uniformly distributed in the service area.
We classify these requirements into two categories, user-side and server-
side requirements. User-side requirements include minimum SINR re-
quirements, latency limits, etc., imposed by the users in the target area. On
the other hand, server-side requirements include network requirements
such as cell load factor, and resource consumption limits [74]. Therefore,
we formulate the optimization problem as

minimize max Vi (3.11a)

X i

s.t.: URk (x, y) ≥ γk (x, y), (x, y) ∈ J, 1 ≤ k ≤ K (3.11b)

SRli ≥ θ il , 1 ≤ i ≤ N, 1 ≤ l ≤ L (3.11c)

where URk (x, y) and SRli refer to the k th requirement of the user at the loca-
tion point (x, y) and l th requirement of i th BS, respectively. It is noted that
meeting all user-side and server-side requirements might be conflicting
and impossible at the same time. Here, we reformulate (3.11) as

minimize max f i (δ(x, y), UR1 (x, y), ..., URK (x, y), SR11 , ..., SRLN ) (3.12a)
X i

s.t.: (x, y) ∈ J, (3.12b)

wherein the penalty functions are defined as

f i (δ(x, y), UR1 (x, y), ..., URK (x, y), SR11 , ..., SRLN ) (3.13a)
∫︂ K
∑︂ L
∑︂
k (x,y)−UR k (x,y)]+ l l +
= δ(x, y) + κk (e[γ − 1)dxd y + λl (e[θ i −SR i ] − 1) (3.13b)
Ji k=1 l =1

for 1 ≤ i ≤ N . Here, the positive part function is defined as

{︄
x, x≥0
[x]+ =
0. x<0

42
 Planning of RF-Based Outdoor Cellular Wireless Networks

Algorithm 3.3. Heterogeneous Power Voronoi Algorithm

1: X (0) ←− Random(J, N)
2: X (1) ←− CVA % Algorithm 3.1
3: J(1) ←− Voronoi(X (1) ) % (3.2)
4: z ←− 2, W(2) ←− 0
5: While ∥ X (z−1) − X (z−2) ∥ > ϵ
6: X (z) ←− Centroid(J(z−1) , δ, UR1 , ..., URK , κ1 , ..., κK , γ1 , ...., γK ) % (3.14)
7: J(z) ←− PowerVoronoi(X (z) , W(z) ) % (3.1)
8: f 1 , ..., f N ←− fcal(J(z) , δ, UR1 , ..., URK , SR11 , ..., SRLN , κ1 , ..., κK ,
λ1 , ..., λL , γ1 , ..., γK , θ 1i , ..., θ iL ) % Balancing functions (3.13)
M[[ f 1 , ..., f N ]] − [ f 1 , ..., f N ]
9: W(z+1) ← W(z) + S ×
M[[ f 1 , ..., f N ]]
10: z ← z+1
11: End While

The performance of the optimization problem (3.12) depends on the con-

straint weighting factors κk ≥ 0 and λl ≥ 0 for all k and l , which are derived
empirically. Specifically, the objective function of (3.12) increases drasti-
cally with a small violation of the user-side or server-side constraints.
Here, we reformulate the cell centroids as
∫︁∫︁ (︂ ∑︁K )︂
[γk (x,y)−URk (x,y)]+ − 1) dxd y
J i (x, y) δ(x, y) + k=1 κ k (e
c′i ≜ ∫︁∫︁ ∑︁K , (3.14)
[γk (x,y)−URk (x,y)]+ − 1)dxd y
J i δ(x, y) + k=1 κ k (e

which is a function of the demand distribution and the user-side con-

straints.
We determine the solution to (3.12) via Heterogeneous Power Voronoi
Algorithm (HPVA) shown in Algorithm 3.3. The steps of HPVA are as
follows; First, we determine the locations of BSs in the target area via
CVA. Second, we move the BSs to the centroids calculated by (3.14) and the
new tessellation is derived using (3.1). Here, by calculating f i for all i , the
weight vector are updated such that the cells with higher f i increments w i
and vice versa. This procedure is repeated until the BS locations converge.
In this manner, we have a cell planning that satisfies uniform balancing
values f 1 = f 2 = ... = f N . Ideally, the equality Vi = f k for all i and k indicates
that all the user-side and server-side constraints are met and the volume
shares at cells are balanced.
To evaluate the performance of HPVA, we consider a target area with
x y
spatial demand distribution δ(x, y) = exp(− ) and spatial minimum
100 100
SINR requirement distribution γ(x, y) = −10 + 10−4 x y[dB], shown in Figure
3.2(a) and Figure 3.2(b), respectively. We illustrate the performance of
HPVA over target area in Fig. 3.4.

43
Planning of RF-Based Outdoor Cellular Wireless Networks

107
12
10

Balancing value
y-position (m)

100
8
200 6
4
300
2
400 0
100 200 300 400 500 600 0 10 20 30
(a) (b)
7
10
12
10

Balancing value
y-position (m)

100
8
200 6
4
300
2
400 0
100 200 300 400 500 600 0 10 20 30
(c) (d)
107
12
10
Balancing value
y-position (m)

100
8
200 6
4
300
2
400 0
100 200 300 400 500 600 0 10 20 30
x-position (m) BS index
(e) (f)
Figure 3.4. BS deployment using HPVA to satisfy equal volume shares among
the cells in the service demand area of Figure 3.2(a) and (a,b) ne-
glecting the user-side requirements, (c,d) minimum SINR requirement
γ(x, y) = 0 [dB] for all x and y, and (e,f) the minimum SINR requirement
distribution as in Figure 3.2(b).

• Figure 3.4(a) illustrates the resulting BS location using HPVA for

the given demand distribution regardless of the minimum SINR
requirement, i.e., κk = 0 for all k. Here, we observe uniform balancing
values f 1 = ... = f N , where f i = Vi , see Figure 3.4(b).

• The resulting BS locations given the demand distribution of Figure

3.2(a) and a uniform minimum SINR requirement γ(x, y) = 0[dB] for
all x and y is shown in Figure 3.4(c) with balancing values in Figure
3.4(d). Here, BSs are distributed uniformly to satisfy the minimum
SINR requirements. It is observed here that although balancing
values are uniform, but the volume shares are unevenly distributed

44
 Planning of RF-Based Outdoor Cellular Wireless Networks

among BSs.

• The resulting volume shares and the balancing values in the cell
planning for the demand distribution as in Figure 3.2(a) and the
minimum SINR requirement distribution as in Figure 3.2(b) is rep-
resented in Figure 3.4(e) and (f). It is observed that the BSs are
accumulated both in the up-right and down-right sides to satisfy
demand and minimum SINR requirements.

Thus, by tuning the parameters κk and λl for all k and l , respectively,

HPVA can simultaneously meet user-side and server-side requirements.

3.6 Planning for Coverage Maximization

A cellular network with maximum coverage has improved SINR, particu-

larly near its edges. The literature has shown that BS topologies such as
the triangular lattice (hexagonal) BS deployment provide the best coverage
[79, 37]. Nevertheless, such a deployment does not suit nonuniform de-
mand distributions, which leads to unbalanced loads and possible overloads.
To determine a network topology that is compatible with the non-uniform
distribution of demand while maximizing coverage, we introduce a local
regularity optimization problem.
Let a BS be at the point x i with Delaunay edges to a set of m i BSs
located at the points y1 , y2 , ...ym i . We define the local regularity optimization
problem for a BS as
σD i (x′i , y1 , y2 , ..., ym i )
minimize , (3.15)
′
xi µD i (x′i , y1 , y2 , ..., ym i )

where σD i and µD i denote standard deviation and mean of the distances

from x′i to the set of points y1 , y2 , ..., ym i . The objective function of (3.15) is
known as coefficient-of-variation (CoV)-based regularity metric [35]. We
give a generalized form of (3.15) as
(︃ ⃓⃓⃓ ∑︁m i ⃓⃓ ⃓⃓
′ ⃓⃓ ⃓ n )︃ n
1
∑︁m i ⃓⃓ ⃓⃓
⃓⃓⃓ ′ ⃓⃓ l =1 yl − x i ⃓
k=1 ⃓ yk − x i − ⃓
mi
minimize (︁∑︁m i ⃓⃓ ⃓⃓ )︁ 1 (3.16)
′
xi ⃓⃓ yk − x′ ⃓⃓ p p
k=1 i
mi
where n and p denote the n-norm and p-norm distance parameters, re-
spectively. As a special case, (3.16) with n = 2 and p = 1 corresponds to
(3.15).
For any positive p and n, the solution to the optimization problem (3.16)
is the set of points whose coordinate equals ±∞. In other words, the
solution to (3.16) is non-unique and falls in infinity. So, we change the

45
Planning of RF-Based Outdoor Cellular Wireless Networks

Algorithm 3.4. Local Regularity Algorithm

1 X ′ ←− CVA % Algorithm 3.1

2 For iter = 1, ..., S :
3 For i = 1, ..., N :
4 (y1 , ..., ym i ) ← Delaunay(x′i ) % Delaunay neighbors
5 x′∗
i ← LocalRegularity (y1 , ..., ym i ) % Solving (3.17) via PSO,

% (3.20I) via FPO, (3.20II) via APO, or (3.20III) via SMO

6 x′i ← (1 − α)x′i + α x′∗
i

7 End For
8 End For

objective function of (3.16) by removing the denominator and constraining

x′i as
mi ⃓
(︂ ∑︂ ∑︁m i ⃓⃓⃓⃓ ⃓⃓
′ ⃓⃓ ⃓ n )︂
1
⃓⃓⃓⃓⃓ ⃓⃓
′ ⃓⃓ l =1 yl − x i ⃓ n
minimize ⃓ yk − x i − ⃓
′ xi mi (3.17)
k=1

s.t.: x′i ∈ CH(y1 , y2 , ..., ym i ).

The objective function of (3.17) is non-convex, and therefore, we solve it
using Particle Swarm Optimization (PSO), [80]. To relax the non-convexity
of (3.17), we remove the concave term as
mi ⃓ 1
(︂ ∑︂
⃓⃓⃓⃓⃓ ⃓⃓⃓n )︂
′ ⃓ ⃓⃓
minimize ⃓ k n
′
y − x i ⃓
xi
k=1 (3.18)

s.t.: x′i ∈ CH(y1 , y2 , ..., ym i ),

which is convex problem, for n ≥ 1. Then, we equivalently rewrite (3.18) as
mi ⃓ 1
(︂ ∑︂
⃓⃓⃓⃓⃓ ⃓⃓⃓n )︂
′ ⃓⃓⃓
minimize ⃓ k y − x i ⃓
n , (3.19)
′
xi
k=1

The optimization problem (3.19) has well-known solutions in literature

with n = ∞, 2, and 1. So, we rewrite (3.19) as
mi ⃓ 1
(︂ ∑︂
⃓⃓⃓⃓⃓ ⃓⃓⃓⃓n )︂
minimize
′
⃓ yk − x′i ⃓⃓⃓ n ≡
xi
k=1
⎧ (︂⃓⃓ ⃓⃓)︂
⎪
⎪minimize
⎪ max ⃓⃓ yk − x′i ⃓⃓ , n = ∞ (I) (3.20)
⎪
⎪ x′i
⎨ ∑︁m i ⃓⃓⃓⃓ ⃓ ⃓
′ ⃓⃓2
minimize k=1 yk − x i , n = 2 (II)
⎪ x′i
⎪
⎪ ∑︁m i ⃓⃓⃓⃓ ⃓ ⃓
⎪
⎪ ′ ⃓⃓
⎩minimize
′ k=1 yk − x i , n = 1 (III)
xi

In (3.20-I), we minimize the maximum distance to the set of points

y1 , y2 , ..., ym i which is determined as the center of the Enclosing Circle (EC)

46
 Planning of RF-Based Outdoor Cellular Wireless Networks

Figure 3.5. Real target area with N = 529 BSs; (a) The map of original BS locations and
(b) the spatial demand density.
y-position

y-position

x-position x-position
(a) (b)

Figure 3.6. (a) Original BS locations as shown in Figure 3.5(a), (b) BS deployment
via CVA compatible with the demand density shown in Figure 3.5(b).

for the set of points y1 , y2 , ..., ym i in a worst-case linear time [81]. These
problems are also solved via Farthest Point Optimization (FPO) using
a finite number of candidate FP Voronoi vertices [71]. Also in (3.20-II),
we minimize the sum squared distances to the set of points y1 , y2 , ..., ym i .
The solution to (3.20-II) is derived using Average Points Optimization
(APO), known as facility localization via Euclidean norm squared criterion
[82]. Finally in (3.20-III), we minimize the sum distances to the set of
points y1 , y2 , ..., ym i . With the assumption that y1 , y2 , ..., ym i are non-collinear,
the solution to (3.20-III) is unique and derived using Spatial Median
Optimization (SMO) [83]. This is due to the fact that if we assume that
Delaunay neighbors y1 , y2 , ..., ym i are collinear, the Voronoi area of BS x i is
unbounded which is not the case in this dissertation.
Algorithm 3.4 describes the step by step procedure of local regularity
algorithm; Given the demand distribution in a service area, we start from
the set of BS locations resulting from the CVA. Then, we draw Delaunay
triangulation and move each BS to a new location using (3.17) via PSO,
(3.20-I) via FPO, (3.20-II) via APO, or (3.20-III) via SMO. Taking a few
iterations, we finally end up with a BS locations that are compatible with

47
 Planning of RF-Based Outdoor Cellular Wireless Networks

y-position

(a) (b)
y-position

x-position x-position
(c) (d)

Figure 3.7. BS deployment by applying 3 iterations of local regularity algorithm

using (a) PSO, (b) FPO, (c) APO, and (d) SMO.

the non-uniform demand and maintains the local regularity.

The performance of the local regularity algorithm is evaluated by con-
sidering real data obtained from the original deployment of N = 529 BS
locations in a 14 km × 18 km area, see Figure 3.5(a). Also, Figure 3.5(b) illus-
trates its spatial demand density by calculating downlink volume shares
within 35 days averaged over the cell areas. Plotting the Voronoi diagram
of original BS locations in Figure 3.6(a), we determine a BS placement via
CVA in Figure 3.6(b) to be compatible with the demand density of Figure
3.5(b). The reason for this is that in areas with higher demand density, we
have a greater concentration of BSs with smaller cell sizes and vice versa.
It is also evident from the visual observation that this BS placement is
non-regular, which will result in a low coverage probability.
In Figure 3.7(a) , we show the new BS locations as a result of applying
three iterations of the local regularity algorithm starting from the place-
ment shown in Figure 3.5(b). Here, we iteratively solve the optimization
problem (3.17) with n = 2 using PSO. In Figure 3.7(b), the optimization
problem (3.20-I) is iteratively solved via FPO. Additionally, Figure 3.7(c)

48
 Planning of RF-Based Outdoor Cellular Wireless Networks

1 1 Original BS deployment
BS deployment via PSO

CoV-based Regularity
BS deployment via FPO

CoV-based SDS
0.8 0.8 BS deployment via APO
BS deployment via SMO
0.6 0.6
Original BS deployment
0.4 BS deployment via PSO 0.4
BS deployment via FPO
0.2 BS deployment via APO 0.2
BS deployment via SMO
0 0
0 2 4 6 8 0 2 4 6 8
Number of L-regularity iterations Number of L-regularity iterations
(a) (b)

1 1
Original BS deployment
BS deployment via PSO
Coverage Probability

0.8 0.8

0.6 0.6

CDF
0.4 0.4

0.2 Original BS deployment 0.2

BS deployment via PSO
0 0
-20 -10 0 10 20 -20 -10 0 10 20 30
SINR threshold (dB) SINR (dB)
(c) (d)
Figure 3.8. (a) SDS and regularity metric comparison of the local regularity al-
gorithm via PSO, FPO, APO, and SMO, (b) Coverage probability for
the original versus the resulting BS deployments, (c) SIR CDF for the
original versus the resulting BS deployments.

shows the result of the regularity algorithm with iteratively solving the
optimization problem (3.20-II) using APO. Finally, Figure 3.7(d) presents
the topology as a result of the regularity algorithm by solving optimization
problem (3.20-III) via SMO.
As can be seen from Figure 3.8(a) and Figure 3.8(b), the line graph com-
pares the Coefficient of Variation (CoV)-based service demand share (SDS)
metric and local regularity metric, respectively, for the local regularity
algorithm, when using PSO, FPO, APO, and SMO. Here, we define the
CoV-based SDS metric as the standard deviation of volume shares over its
mean. Several iterations of the local regularity algorithm using PSO, FPO,
and APO have only a slight effect on the CoV-based SDS metric, while
significantly improving regularity by reducing the CoV-based local regu-
larity metric to even one third for PSO. Figures 3.8(c) and 3.8(d) evaluate
coverage probability and SINR Cumulative Distribution Function (CDF),
respectively, for the original BS deployment shown in Figure 3.5(a) and
the optimal BS deployment using three iterations of the local regularity

49
Planning of RF-Based Outdoor Cellular Wireless Networks

algorithm via PSO depicted in Figure 3.8(a). The coverage probability for
the original BS deployments closely resembles the coverage probability
for the Poisson Point Process deployment, which is the lowest coverage
probability for the service area [37]. However, as it is observed in Fig-
ures 3.8(c) and 3.8(d), the resulting BS deployment using PSO increases
coverage probability by as much as 13% and SINR CDF by 3 [dB], which
matches perfectly with the coverage probability of hexagonal deployment.
Therefore, local regularity algorithm via PSO results in a BS deployment
that gives the highest coverage probability [37].

3.7 Summary

To provide high quality of service, a wireless cellular network must have

adequate coverage, capacity, and reliability. In order to achieve optimal
performance and reduce interference, it is necessary to deploy adequate
infrastructure, including towers, antennas, and backhaul connectivity, as
well as efficient network management and optimization. A network should
also support a variety of services and applications, such as voice, data,
and video. It should prioritize traffic in accordance with user needs and
network conditions to ensure a seamless user experience. In this regard,
optimal deployment of BSs plays a significant role in improving network
quality.
With the aid of geometrical tools such as Voronoi diagrams and Delaunay
triangulation, we proposed three algorithms for optimal deployment of
BSs in wireless cellular networks with non-uniform traffic demand dis-
tributions. First, we applied iterative power Voronoi algorithms for load
balancing. Second, we proposed a heterogeneous power Voronoi algorithm
taking into account a set of network requirements. Finally, we proposed
a local regularity algorithm to balance loads and maximize coverage. It
was observed that with the proposed deployment methods, the traffic
shares were evenly distributed, and the coverage probability was ideal as
a hexagonal cell deployment.
Although the proposed algorithms have been shown to improve wireless
cellular network performance, our theoretical analysis and abstract deploy-
ment algorithms impose some limitations and simplifications. Our first
assumption was that the signals received by users are only influenced by
mean path loss, ignoring shadowing effects, scattering, and reflection from
obstacles, such as buildings and mountains, which are typical scatterers
in dense urban areas. Secondly, we assumed that BSs could be deployed
anywhere within a service area. Nevertheless, there is a limited set of
possible candidate sites for BSs in practice, which may constrain the ap-
plication of the algorithms presented in this chapter. Finally, we assumed
that we were provided with data about the spatial distribution of traffic

50
 Planning of RF-Based Outdoor Cellular Wireless Networks

demand in the service area, which may not be readily available in practice.
It is therefore possible to improve the proposed deployment algorithms by
taking into account each one of these practical constraints.

51
4. Planning of New Spectrum-Based
Indoor Wireless Networks

4.1 Introduction

The coverage issues associated with wireless networks in RF bands have

been addressed by a variety of analytical tools. However, indoor wireless
networks that employ higher frequencies like mmWave, THz, and visible
light still face significant challenges in providing seamless coverage in
targeted areas such as rooms and offices. In comparison with those that can
be found in the bands below 6 GHz, higher frequency bands provide a much
wider range of signal bandwidths. According to 3GPP standardization [84,
85], mmWave signals can have a range of a few hundred meters [86], which
is much larger than a typical distance between two points indoors.
THz band communication (0.1–10 THz) may alleviate the spectrum scarcity
that may result from the migration from 5G to 6G [87]. However, the trans-
mission distance in THz bands is shorter than in mmWave bands due to
high path loss attenuation [88] and the difficulty in generating strong trans-
mission power when the carrier frequency exceeds 100 GHz [89]. A similar
range-constrained situation occurs when visible light signals replace RF
signals. As LEDs are extremely efficient sources of optical power [90],
visible light signals are not limited by their optical power. Rather, the
range of visible light signals is determined by the directivity of the light
beams and the selective field-of-view response of PDs in the receivers [91].
Like mmWaves and THz, visible light signals are susceptible to link block-
ages since they cannot pass through large obstacles such as walls, doors,
and furniture. The combination of this feature and the range constraint
makes it difficult to deploy access points indoors that ensure LoS coverage.
Despite the fact that we could use mmWaves and THz frequency bands for
communication and positioning, we primarily focused on VLC networks.
The reason for this is that VLC links are shorter in range and their tech-
nology maturity level is currently much higher than that of THz links. So,
from now on, we will discuss VLC, but the discussion could be generalized

53
Planning of New Spectrum-Based Indoor Wireless Networks

to embrace mmWaves or THz.

The Internet of Things and machine-type communication networks
present continuing challenges to wireless networks, placing a great deal
of strain on their performance, particularly in terms of data rates. Due
to the larger bandwidth available in visible light frequency bands, VLC
can enhance the data rate and system capacity of 5G/6G networks cur-
rently constrained by highly saturated radio frequencies [92, 93, 94]. The
widespread use of LEDs, initially designed for illumination, further con-
tributes to the adoption of VLCs. In addition, VLC provides interference
confinement, sustainability, and ultra-densification capabilities [95].
Despite these benefits, VLC networks are susceptible to limitations in
terms of signal propagation. The presence of obstructing objects such as
walls, doors, and curtains can cause link blockages between a LED and
users due to weaker reflection power compared to the LoS link [96, 97, 98].
Additionally, the optical power received by users decreases as they move
toward the edge of the FoV, thus restricting the maximum range of an LED
[56, 57]. As a result of these features, coverage is limited in VLC networks,
but privacy and security are improved.
Indoor deployment has recently been examined with a focus on VLC ap-
plications for both illumination and communication purposes. To conduct
a coverage analysis of the entire indoor environment, LEDs are typically
placed in a uniform [99] or random [100] distribution. In [101] and [102],
authors used statistical tools to analyze LED deployment using the proba-
bility density function (pdf) per user and the central Voronoi tessellation,
respectively. Authors in [103] addressed optimal locations for two LEDs in
a 2.5m × 5m × 3m room in order to maximize average area spectral efficiency.
The authors of [104] investigated the effect of a relay-assisted VLC network
as a solution to the challenges associated with seamless coverage in an
indoor environment, while the authors of [105] examined the coexistence of
RF and VLC, commonly referred to as crowds in HetNet. However, the use
of LEDs for LoS coverage indoors has not been widely studied analytically
[106].
A significant challenge, however, remains in developing high-speed, re-
liable, and scalable backhaul infrastructure for 5G/6G, including VLC
networks. Traditionally, fiber optics and Ethernet have been used for
backhaul in Long-Term Evolution (LTE) networks [107, 108]. As far as
data transmission speed is concerned, fiber optic is the fastest option.
Besides, the Power-over-Ethernet (PoE) standard allows LEDs to obtain
both electrical power and data over a single cable [109, 110]. It would be
prohibitive, however, to use fiber optic or PoE cables to connect every LED
due to the density of LEDs required to cover an indoor area [111]. Power
line communications (PLC) has also been suggested as another possibility
for backhauling in VLC networks [112]. Despite the widespread use of
power lines, PLC backhauling causes a tremendous amount of interference.

54
 Planning of New Spectrum-Based Indoor Wireless Networks

Besides, PLCs require additional equipment in order to interface with

VLCs, which results in a greater level of complexity. In accordance with
[107], Free Space Optics (FSO) is used for backhaul communications due
to its narrow, license-free, and point-to-point beams, which make optical
wireless backhauling an inexpensive and promising technology for VLC
networks. For reliable optical wireless backhauling, we need to ensure LoS
condition among LEDs.
In order to minimize systematic errors and to improve the accuracy of
the indoor Positioning System (IPS), it is essential that positioning access
points are deployed to guarantee the LoS condition to one (proximity
principle), two (triangulation principle), three (multilateration principle),
and possibly more access points. As a result of this deployment, the
access points that serve the same UE are sufficiently separated in both
the distance and angular domains. By doing so, we ensure a high level of
positioning accuracy throughout the layout. Noteworthy is the fact that
the deployment approaches presented are frequency agnostic so that access
points may utilize any type of technology for positioning.
This chapter describes how to determine a minimum number of APs, e.g.,
LEDs, and their location in order to meet indoor environment performance
requirements such as LoS coverage, LoS backhauling, and multiple LoS
coverage. As a means of meeting seamless LoS coverage indoors, we model
the geometrical properties of indoor environments with a LoS graph and
partition this graph into a minimum number of cliques. To satisfy LoS
coverage and backhauling, we model the LoS graph with a tree. Further,
we utilize a multiple partitioning approach for the LoS graph in order to
provide multiple LoS coverage for accurate indoor positioning. The LoS
graph is also shown to be an effective tool for assessing the optimality
of indoor deployments. For simplicity, it is assumed that the layouts
in this dissertation consist only of walls. However, blocking caused by
furniture, columns, and other obstacles can also be easily incorporated into
the proposed graph modeling.
This chapter is structured as follows. Section 4.2 discusses the VLC
channel model and optical wireless backhauling. Sections 4.3 presents
the illumination requirements of VLC networks. Section 4.4 describe the
methods used to address the art gallery problem [113], as well as the
state of the art in wireless indoor planning. We present a graph model of
indoor areas in Section 4.5 and propose deployment methods for ensuring
a LoS condition in access and backhaul in Section 4.6. In Section 4.7, the
connection between the location of access points and the accuracy of indoor
positioning is discussed. Indoor network planning method for a precise
indoor positioning is presented in Section 4.8, and lastly, we summarize
the chapter in Section 4.9.

55
Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.1. Optical wireless network; (a) System model. (b) VLC channel model.

4.2 VLC Channel Model and Optical Wireless Backhauling

In optical wireless networks, the access points (LEDs here) transmit data
using visible light signals, also known as optical wireless access [54, 114].
Optical wireless networks provide good security, bandwidth, as well as
sustainability due to their capability of simultaneously illuminating and
transmitting data. It can be seen from Figure 4.1(a) that the spatial
distribution of illumination changes more smoothly than the receiving
data rate, where data rates are significantly reduced near cell boundaries
[57].
Based on Figure 4.1(a), it appears that wired backhaul requires a more
complex and expensive network infrastructure than wireless backhaul.
Also, wired backhauling restricts the scalability and mobility of the net-
work. In contrast to wired backhaul, wireless backhaul is more cost-
effective since it is easier to install, more reliable, and requires fewer
infrastructure assets [115]. Considering a Visible Light Communication
Access Point (VLC-AP), LEDs are used both for transmission and back-
haul. Radio is more commonly used for wireless backhaul, however optical
wireless offers a future-proof, low-cost, and reliable backhaul when the
distance between access points does not exceed tens of meters [116, 117].
Thus, we assumed that the backhauling data would be transmitted via an
infrared LED employed in a VLC-AP, which would then be relayed to other
VLC-APs until all VLC-APs were covered. The LoS conditions between any
two VLC-APs, either directly or through a group of VLC-APs, are therefore
critical for a reliable optical wireless backhaul.
The direct illumination framework is used to model VLC channels be-
cause the direct optical signal is much stronger than the reflected optical
signal [56, 57]. In this model, Phosphor-Converted PC-LEDs radiate ac-

56
 Planning of New Spectrum-Based Indoor Wireless Networks

cording to Lambertian radiation patterns, see Figure 4.1(a). Accordingly,

the DC gain of the optical channel between the LED transmitter and the
PD receiver is given by
{︄
(m L+1)A PD
dir 2π d 2
cosm L (φ) cos (ψ), 0 ≤ ψ ≤ Ψ,
HLED,PD (0) = (4.1)
0, ψ > Ψ,
[︁ ]︁
where m L = −1/ log2 cos(θmax ) denotes the Lambert index of the LED, in
which θmax [rad] defines the source radiation semi-angle at half power
of the LED. Besides, φ [rad] and ψ [rad] refer to the angle of irradiance
and incidence of the LoS link, respectively. Furthermore, Ψ [rad] denotes
the FoV semi-angle of the PD with an effective physical area of A PD [m2 ].
In order to simplify our analysis, we assume that the PD always points
upward. Thus, this network has a maximum range of r = hLED-PD × tan(Ψ),
where hLED-PD and d are the height and distance, respectively, between
LED and PD. Therefore, a PD located beyond this range cannot receive
data from the LED.
The spectral optical power that reaches the PD at wavelength λ is given
by
pdir dir (W)
o,PD (λ) = PLED HLED,PD (0) S o (λ), (4.2)
and the DC current at the output of the PD is calculated as
∫︂ λu
i dir
PD (0) = pdir
o,PD (λ) R PD (λ) f o (λ) d λ. (4.3)
λl

Here, PLED and So(w) (λ) are the total radiant power and spectral power
distribution of the LED, respectively. Furthermore, RPD (λ) denotes the
responsivity of the PD, and f o (λ) is the transmittance of the optical pass-
band filter with lower (λl ) and upper (λu ) cutoff wavelengths. We then
derive the power of the electrical signal at the output of PD as
⃓ ⃓2
ΛPD (0) = ⃓ i dir ⃓
PD (0)G tia , (4.4)

where G tia refers to the gain of transimpedance amplifier within the PD.
Then, the received data rate can be approximated as
ΛPD (0)
R data = B log(1 + ), (4.5)
I + σ2

where B, I , and σ2 refer to LED transmission bandwidth, the power of the

interfering electrical signal, and the noise power of the received signal,
respectively. In (4.5), AWGN thermal noise was assumed to dominate the
noise power, and non-linear distortion of LEDs was neglected. Therefore,
Shannon capacity was used as a measure of the achievable data rate in
VLC networks [118].
For VLP systems based on RSS, it is necessary to model channels very
precisely in order to achieve accurate positioning. In this case, the real LED

57
Planning of New Spectrum-Based Indoor Wireless Networks

Table 4.1. Standard illumination metrics for indoor environments

Type of indoor environment Em Uo

Underfloor tunnel and cellar 50 0.4
Parking areas 75 0.4
Building corridors 100 0.4
Store and stock rooms 100 0.4
Control stations 150 0.6
Dressing rooms 200 0.4
Waiting rooms 200 0.4
Libraries: book shelves 200 0.6
Rooms for physical exercises 300 0.4
Bakeries 300 0.6
Ticket office 300 0.6
Classrooms 300 0.6
Rooms for medical attentions 500 0.6
Post rooms 500 0.6
Laboratories 500 0.6
Staff offices 500 0.6
Shoe making 500 0.6
Libraries: reading area 500 0.6
Manufacturing of jewellery 1000 0.7
Electronic workshops, testing, and adjusting 1500 0.7

channel gain may not follow (32) exactly, as there exists a bias depending
on the location of the area [119]. In light of the difficulty of measuring
the true radiation pattern of an LED, the development of dedicated VLP
LEDs with Lambertian radiation patterns may contribute to the growth of
RSS-based VLP.

4.3 Illumination requirements of VLC networks

The primary objective of the VLC network is to provide suitable illumina-

tion throughout the entire area beneath it. The illumination per square
meter is given by
∫︂ λu dir
p o,PD (λ)
E = 200 × d λ, (4.6)
λl A PD
wherein λl and λu represent the lowest and highest wavelengths in visible
light electromagnetic signals, respectively. Also, the numeric coefficient

58
 Planning of New Spectrum-Based Indoor Wireless Networks

denotes the average Watt to Lumen conversion ratio. The illumination at

each point is a summation of the illumination factors received from each
visible LED.
It is also necessary that the VLC network satisfies illumination require-
ments based on the type of environment and the task objective in terms of
the maintenance value E m and the uniformity value U o . In the visual task
area, the maintenance value represents the minimum illuminance level
that should be satisfied. Moreover, uniformity value is the proportion of the
lowest illumination over the mean illuminance. Large enough uniformity
value ensures that there is no great differences in brightness. In Table
4.1, examples of indoor areas with varying tasks are shown along with
their minimum maintenance and uniformity requirements1 . It is therefore
essential to ensure that the illumination requirements of an indoor envi-
ronment are not greatly violated when deploying LEDs/access points for
communication and positioning.

4.4 Art Gallery problem and the State of the Art in Wireless Indoor
Planning

Assuming a layout with n vertices, the Art Gallery Problem involves de-
termining the minimum number of stationary omni-directional camera
guards required to ensure that every point on the layout is visible from
at least one of them [113]. Deployment of access points for LoS coverage
in indoor areas, like LEDs in indoor VLC networks, poses a challenge
similar to that of the art gallery problem, particularly when the range
is unconstrained (r = ∞). In this scenario, a single camera guard or AP
with unlimited range can cover any point within a convex layout [120].
However, when dealing with non-convex layouts, multiple guards might be
required for LoS coverage.
A set of possible approaches to tackle the art gallery problem are pre-
sented in Figure 4.2. Assume placing camera guards on every third vertex
of the layout, Figure 4.2(a) shows that the guards do not cover the red cross-
marked point, and therefore, they fail to cover the whole layout. Moreover,
covering the layout’s edges does not guarantee covering the whole layout,
such as the red cross-marked point in Figure 4.2(b) is left uncovered. As in
Figure 4.2(c), placing guards on every reflex vertices of a layout guarantees
covering the whole layout. However, this approach might be inefficient in
case there are a large number of reflex vertices forming the layout.
Chvatal suggested that the upper bound for the minimum number of
⌊︁ ⌋︁
guards required for the art gallery problem is n/3 , which was then verified
by Fisk by demonstrating that the vertices of every layout are 3-colorable
[121, 113]. With this proof, each vertex is labeled with one of three different
1 [Link]

59
Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.2. Approaches to tackle art gallery problem via (a) placement of the
guards on every third vertex points, (b) placement of the guards such
that all the layout edges (walls) are visible, (c) placement of the guards
on the reflex vertices, and (d) 3-coloring method.

colors so that no adjacent vertices share the same color. Thus, placing
guards on the vertices with the minimum number of color uses, such as
six blue vertices in Figure 4.2(d), provides full coverage of the layout.
Furthermore, several heuristic solutions have been proposed to the art
gallery problem with a focus on layout vertices, intersections between edge
extensions, and center points of polygons [122]. Another proposed solution
involves iteratively weighting a selection from a large set of grid points
[123]. However, none of these approaches address the unique requirements
of wireless networks, such as limited range, backhauling, and precise
positioning.
In a boundless area without obstacles, hexagonal cells are the most
efficient architecture for deploying APs with limited range, just like RF
cellular networks use them outdoors [124, 125]. It is advantageous to
deploy hexagonal cells since they ensure coverage of the LoS by deploying
a minimum number of APs. However, when it comes to a limited-sized
area with walls, deploying APs becomes particularly challenging. For a
square-shaped room with side length s 0 as in Figure 4.3(a), deploying a
single AP or LED, shown by a triangle mark in the center, with maximum
⎷
range r = 2s 0 /2 or larger is enough for a LoS coverage. Furthermore, as
⎷ ⎷
in Figure 4.3(b), two APs with maximum range 5s 0 /4 ≤ r < 2s 0 /2 are
enough to cover the area. Interestingly, Figure 4.3(c) illustrates that with
⎷ ⎷ ⎷
the maximum range ( 6 − 2)s 0 /2 ≤ r < 5s 0 /4, three APs are necessary
and sufficient for a LoS coverage. Lastly, Figure 4.3(d) indicates that

60
 Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.3. LoS coverage of the square-shaped indoor area with side length s 0 by
the optimal deployment of (a) a single LED (triangle mark) with the
⎷
maximum range r 1 = 2s 0 /2, (b) two LEDs with the maximum ranges
⎷ ⎷ ⎷
r 2 = 5s 0 /4, (c) three LEDs with the maximum ranges r 3 = ( 6− 2)s 0 /2,
⎷
(d) and four LEDs with the maximum ranges r 4 = 2s 0 /4.

four APs are required to fully cover the area if the maximum range is
⎷ ⎷ ⎷
limited to 2s 0 /4 ≤ r < ( 6 − 2)s 0 /2. Consequently, it is not an easy task to
determine an optimal LED deployment to provide LoS coverage for even a
square-shaped room. An irregular layout further complicates this problem.
To establish the minimum number of access points and their placement
required to fulfill various constraints, such as LoS coverage, backhauling,
and visibility angles for precise positioning, we represent the floor plan
as a graph and formulate the minimum clique cover as the corresponding
optimization problem. Furthermore, we determine lower bounds for the
minimum number of access points necessary to evaluate the proposed
approaches’ effectiveness.

4.5 Graph Modeling of Indoor areas

We characterize LoS graphs as a tool for modeling the geometrical charac-

teristics of layouts in order to construct and analyze equivalent optimiza-
tion problems. Given a point P and a polygon p inside the layout, we define
LoS areas in the following.

61
Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.4. (a) LoS area of a point and a polygon with the maximum range r . (b)
Connection region of an area.

Figure 4.5. Hyper triangulation with two different parameter values; (a) HT (R =
R 1 ) and (b) HT (R = R 2 ).

• LoS area of a point L(P) denotes the set of all the points X such
that the line segment X P lies entirely inside the layout, and ∥ X P ∥ ≤ r.

In other words, there is no layout edge between P and L(P) while L(P) falls
within the maximum range from P .

• LoS area of a polygon L(p) refers to the set of all the points X such
that X ∈ L(P i ) for i = 1, 2, ..., l , where P1 , P2 , ..., P l denote the vertices of
p.

So, we immediately conclude that L(p) = L(P1 ) ∩ L(P2 )... ∩ L(P l ). Besides,
given an area A within a layout, we define connection region as follows.

• Connection region of an area CR(A) refers to the locus Y for which

there is at least one point X ∈ A such that the segment X Y entirely
lies inside the layout.

Given a point P and a quadrilateral p within the sample layout, Figure

4.4(a) visualizes L(P) and L(p) in red and yellow, respectively, with the
maximum range r . It shows that all vertices of p are openly visible and lie
within r of any point in L(p). Furthermore, given the area A, Figure 4.4(b)
indicates CR(A) in green.
Creating an equivalent LoS graph starts with partitioning the layout.
Partitioning has a significant effect on both the complexity of the graph and
the optimization problems it leads to. Due to their convexity and ability to

62
 Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.6. Construction of a LoS graph; (a) a hyper triangulation of the layout
and (b) the LoS graph

partition any layout, triangles are particularly attractive as layout compo-

nents. Triangulation is the process of dividing a layout into triangles by
adding diagonals. However, to meet the requirements of wireless networks,
the triangles may have to be smaller than in the triangulation.
• Hyper Triangulation HT (R) denotes a triangulation process fol-
lowed by connecting the midpoint of the largest side to the opposite
vertex of each triangle until no triangle side is larger than R .
Figure 4.5 exhibits HT (R = R1 ) and HT (R = R2 ) over the sample layout,
where R1 > R2 . Starting from the triangulation HT (R = ∞) as in Figure
4.2(d), Figure 4.5(a) is derived by bisecting three triangles whose largest
sides were longer than R1 . Similarly, in Figure 4.5(b), all the triangles’
sides are smaller than R2 . When R decreases in HT (R), the number of
triangles M increases with an approximate rate of 1/R 2 for small values
of R . As an important property, let p be a triangle in HT (R). If there is a
point X such that X ∈ L(p), then X ∈ L(Q), wherein Q refers to any point
inside p. This property lets us define the equivalent graph for the indoor
areas.

• LoS Graph G(1) refers to a simple unweighted graph whose nodes

represent the triangles p 1 , ...., p M of the layout. Two nodes p i and
p j in the LoS graph are adjacent if and only if L(p i ) ∩ L(p j ) ̸= ∅ for
1 ≤ i, j ≤ M and i ̸= j .

A LoS graph is also referred to a primary LoS graph. Figure 4.6(a) shows
HT (R = r) with M = 25 triangles, for which Figure 4.6(b) represents the LoS
graph with the maximum range r . For instance, since L(p 3 ) ∩ L(p 20 ) ̸= ∅
and L(p 3 ) ∩ L(p 22 ) = ∅, the nodes p 3 and p 20 are adjacent, while p 3 and p 22
are non-adjacent. Three factors influence the structure of the LoS graph:
the shape of the layout, the value of R in the HT (R) it is constructed by,
and the value of r . LoS graphs provide valuable information, including an
infinite set of possible APs deployments for LoS coverage as well as a tool
for assessing optimality. They also allow for the consideration of a variety
of requirements in wireless communication and positioning systems.

63
Planning of New Spectrum-Based Indoor Wireless Networks

Algorithm 4.2. Maximal Clique Clustering (MCC)

Input G(1) [p 1 , p 2 , ..., p M ] % The LoS graph

1 While G(1) ̸= ∅:

2 G(1) [p′1 , p′2 , ..., p′M ] ←− Ascending-Sort(G(1) )

3 c ←− ∅

4 For i = 1 to M :
5 If c ∪ p′i forms a clique in G(1) :

6 If L(c ∪ p′i ) ̸= ∅:
7 c ←− c ∪ p′i

8 G(1) ←− G(1) − c

9 M ←− |G(1) |

10 Return

• LoS area of a clique L(c) refers to the intersection of the LoS areas
⋂︁
of all the nodes in the clique c, i.e., L(c) := p i ∈ c L(p i ).

As an important property, deploying an AP within L(c) in the layout

provides LoS condition to any point inside each polygon p i ∈ c.

4.6 Indoor Planning of Optical Wireless Networks for Los Condition

in Access and Backhauling

We address the deployment of LEDs in this section in order to ensure LoS

access throughout the layout. Secondly, we consider LoS backhaul as an
additional requirement.

4.6.1 Ensuring Los Condition in Access

We aim to determine the minimum number of VLC-APs (LEDs) and their

locations in order to ensure LoS coverage (LoS condition in access) for an
indoor environment. To achieve LoS coverage, it is necessary to cover every
point in a layout with at least one VLC-AP. Hence, we consider the use of
clique partitioning as a feasible deployment method of VLC-APs to enable
LED-based wireless access.
Let us take a LoS graph and partition it into g cliques c 1 , c 2 ,..., c g , such
that L(c j ) ̸= ∅ for all j . Following this, deploying a set of g VLC-APs, one
located within each L(c j ), will ensure LoS coverage for the entire layout.
Figure 4.7 illustrates a non-optimal deployment of VLC-APs to provide LoS
coverage in the sample layout with the maximum range r using a feasible
clique partitioning of the LoS graph. Figure 4.7(a) displays the partition

64
 Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.7. A non-optimal LED deployment; (a) A feasible clique clustering of the
LoS graph and (b) LoS areas of the cliques.

of the LoS graph into seven cliques distinguished by seven different colors,
whose LoS areas are illustrated in Figure 4.7(b) with similar colors. For
example, the purple area in the layout represents the LoS area of the
clique with the nodes p 4 , p 9 , p 10 , and p 21 . Hence, deploying g = 7 VLC-APs
anywhere inside each clique LoS area, as in Figure 4.7(b), guarantees LoS
coverage for the sample layout.
For LoS coverage, the minimum number of VLC-APs and their locations
are determined by partitioning the LoS graph into the minimum number
of cliques or, equivalently, by minimizing g. NP-hardness is known to
exist for the minimum clique cover problem [126, 127]. So, we propose
the Maximal Clique Clustering (MCC) method shown in Algorithm 4.2.
Initially, the MCC method sorts the nodes of the LoS graph in ascending
degree order. In turn, we add the nodes to a cluster (with preference given
to the nodes with the smallest degree) as long as the cluster continues to
form a clique for the LoS graph with a non-empty LoS area. Thus, adding
each node increases the size of the cluster. Upon sweeping all nodes in the
LoS graph, we may refer to the cluster of nodes as the maximal clique. As
a final step, we remove the maximal clique from the LoS graph and repeat
the steps until no nodes remain in the LoS graph.
Figure 4.8 visualizes the steps of the MCC method to identify the mini-
mum number of VLC-APs for LoS coverage in the sample layout when the
maximum range is r . Starting with the LoS graph shown in Figure 4.6(b),
we select p 14 as the node with the lowest degree in Figure 4.8(a) and the
cluster connected to it as p 11 p 12 p 13 p 14 . Then, by removing this clique, we
select p 25 as the minimum degree node of the remaining LoS graph in
Figure 4.8(b). Continuing the similar steps, we end up with the five cliques
shown in Figure 4.8(e), whose LoS areas are displayed in Figure 4.8(f). As
a result, deploying g = 5 VLC-APs, one inside each LoS area, ensures the
LoS coverage of the sample layout. However, an effective benchmark to
assess the optimality of the number of VLC-APs is still required.
Consider a set of s points in a layout, H1 , H2 ,..., H s , which have pairwise
disjoint LoS areas, i.e., L(H i ) ∩ L(H j ) = ∅ for 1 ≤ i, j ≤ s and i ̸= j , referred to

65
Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.8. Optimal deployment of VLC-APs for a LoS coverage via MCC method:
(a) The LoS graph and determining the first maximal clique; (b) the
second maximal clique; (c) the third maximal clique; (d) the forth
maximal clique; (e) the fifth maximal clique; (f) LoS area of the five
maximal cliques that were identified in the LoS graph;

Figure 4.9. Identifying a lower bound for the minimum number of VLC-APs that
are required for LoS coverage via LoS graph: a) An independent set of
nodes with size t = 5 in the LoS graph; b) Determination of a hidden
set of t = 5 points within each corresponding triangles.

as a hidden set of points. Then, s specifies a lower bound to the minimum

number of VLC-APs required for LoS coverage of the layout, i.e., s ≤
g. A hidden set of points with a maximum size, however, is not easily
determined. Therefore, we may also utilize the LoS graph in accordance
with the procedure described in the following: Let us assume that a set of t
nodes q1 , q2 , ..., q t in the LoS graph forms an independent set, i.e., no two
nodes are adjacent to each other. If there exists a hidden set of t points H1 ,
H2 ,..., H t in the layout lying inside the triangles q 1 , q 2 , ..., q t , respectively,
then t provides a lower bound to the number of VLC-APs for LoS coverage,
i.e., t ≤ g. Therefore, it is sufficient to find an independent set of nodes in

66
 Planning of New Spectrum-Based Indoor Wireless Networks

Algorithm 4.3. Potential Deployment Area Updating

Input T[A1 , A2 , ..., Ah ] % connectivity tree.

R % The root node in T.

1 P1 , P2 , ..., Pl ←− Paths-to-all-leaves(T, R)

% P i : Array of nodes from R to the i th leaf in T.

2 For i = 1 to l :
3 For j = 1 to |P i | − 1 : % A ij : j th node in the path P i .
4 A ij+1 ←− CR(A ij ) ∩ A ij+1 % A1i is set as R.
5 Return

the LoS graph and search over the corresponding triangles rather than
searching over the entire layout for a hidden set of points. In this case, the
procedure is simpler due to the discrete nature of the LoS graph and the
necessity to search for a point in a smaller area, such as inside a triangle.
Using the LoS graph, Figure 4.9 illustrates how a lower bound can
be determined for g. Figure 4.9(a) represents the similar graph as in
Figure 4.6(b), marked with an independent set of t = 5 nodes p 1 , p 8 , p 13 ,
p 19 , p 24 . Then, we find a hidden set of t = 5 points H1 ,..., H5 , shown by
cross-marks in Figure 4.9(b), lying inside the triangles p 1 , p 8 , p 13 , p 19 ,
p 24 , respectively. So, we conclude that t = 5 ≤ g. Therefore, five VLC-APs
determined in Figure 4.8(f) is optimal.

4.6.2 Ensuring Los Condition in Access and Backhauling

We determine the minimum number of VLC-APs ( h) required to provide

both LoS coverage and wireless connectivity. The purpose of wireless con-
nectivity in this context is to make each pair of VLC-APs visible to each
other, either directly or through intermediary VLC-APs. Wireless connec-
tivity enables optical wireless backhauling since data can be transmitted
via optical wireless technology from one VLC-AP to another towards the
core network. A Potential Deployment Areas (PDA) is defined as a region
in which a VLC-AP can be placed to maintain LoS coverage and connec-
tivity. It is noteworthy that any changes made to one PDA may affect the
others. As a result, we define a tree diagram as a means of modeling the
interaction among PDAs.

• Connectivity Tree (T) is a free tree in which each node represents

a PDA in the layout. Two nodes Aa1 and Aa i in T are connected by a
path of nodes Aa2 ,...., Aa i−1 if and only if the two VLC-APs deployed in
the PDAs Aa1 and Aa i have connectivity via the array of i − 2 VLC-APs
deployed in Aa2 ,..., Aa i−1 .

67
Planning of New Spectrum-Based Indoor Wireless Networks

Using T, the PDA-updating method iteratively removes areas within the

PDAs that are infeasible, and therefore, maintaining connectivity among
the VLC-APs, as shown in Algorithm 4.3. Finding the directional paths
from a root node R to the leaves yields a rooted version of T. We interpret
R as the highest priority and degree of freedom PDA. As the PDA-updating
method proceeds from R to all the leaves, the overlapped area with the
connection region of the previous node replaces the PDA at each node.
Consequently, we ensure that a VLC-AP can be deployed at every point
inside the resulting PDAs.
As described in Algorithm 4.4, PDAs and T are simultaneously con-
structed using the Connectivity Tree Construction (CTC) method, which
uses the PDA-updating as an inner algorithm. According to the CTC
method, the LoS area of the first maximal clique L(c) in the LoS graph is
assigned as the first PDA A1 , which also forms the first node in T. Remov-
ing c from the LoS graph, the rest of the method can be summarized as
follows;

I. Among the current nodes in T, find the node AC such that CR(AC )
has a non-empty overlap with the LoS area of the node in the LoS
graph with the smallest degree,

II. Create the node AC within the remaining LoS graph and connect it
to the nodes whose LoS areas have a non-empty overlap with CR(AC ),

III. While searching the nodes over the remaining LoS graph in an as-
cending degree order, find the largest clique c that is fully connected
to AC while satisfying CR(AC ) ∩ L(c) ̸= ∅;

IV. Create the new PDA A|T|+1 := CR(AC ) ∩ L(c) as a node in T and connect
it to AC ,

V. Run the PDA-updating algorithm for the current T by assigning

R := A|T|+1 ;

VI. Remove AC and c from the LoS graph and start over until no node in
the LoS graph remains.

The CTC steps are illustrated in Figure 4.10 when they are applied to the
sample layout with maximum range r . Figure 4.10(a) shows the LoS graph
with the first maximal clique c highlighted in green while creating the
node A1 := L(c) in T as well as creating the node AC := A1 in the remaining
LoS graph, see Figure 4.10(b). As a result, we find the new maximal clique
c in purple by connecting AC to those nodes whose visible areas overlap
with CR(AC ). In Figure 4.10(c), we create the node A2 := CR(A1 ) ∩ L(c) ̸= ∅
in T and connect it to the node AC (A1 here). The PDA-updating algorithm
is now run with the root node R := A2 over T, and therefore, A1 is updated
according to A1 := A1 ∩ CR(A2 ). Since CR(A2 ) overlaps with the LoS area of

68
 Planning of New Spectrum-Based Indoor Wireless Networks

Algorithm 4.4. Connectivity Tree Construction

Input G[p 1 , p 2 , ..., p M ] % The PV graph

Initialize T ←− Ø % Connectivity tree

c % The first maximal clique in G, Algorithm 4.2

1 A1 ←− V(c) % The first PDA

2 Create the first node A1 in T

3 G ←− G − c

4 M ←− |G|

5 While G ̸= ∅:
6 G[p′1 , p′2 , ..., p′M ] ← Asc-Sort(G)
(︁⋃︁ )︁
7 k ←− min i, s.t. V(p′i ) ∩ A j ∈T CR(A j ) ̸= ∅
∑︁ M
8 AC ←− argmax i =1 sgn(|V(p k ) ∩ CR(A j ) ∩ V(p i )|)
′ ′
A j ∈T

9 Create the node AC in G

10 For i = 1 to M:

11 If CR(AC ) ∩ V(p′i ) ̸= ∅:
12 Connect(G, p′i , AC ) % Connect p′i and AC in G

13 c ←− ∅

14 For i = 1 to M:

15 If AC , p′i , and c form a clique:

16 If CR(AC ) ∩ V(p′i ) ∩ V(c) ̸= ∅:
17 c ← c ∪ p′i

18 A|T|+1 ←− CR(AC ) ∩ V(c)

19 Create the node A|T|+1 in T

20 Connect(T, A|T|+1 , AC )

21 A1 , ..., A|T| ←− PDA-Updating(T, R = A|T|+1 ) % Algorithm 4.3

22 G ← G − (AC ∪ c)

23 M ← |G|

24 Return

the smallest degree node in the remaining LoS graph, we create the node
AC := A2 as in Figure 4.10(c). Figure 4.10(g) illustrates the final T, where
Figure 4.10(h) illustrates the five PDAs associated with it.
As T is created, the number of VLC-APs required to maintain LoS cover-
age and connectivity equals the number of PDAs, i.e., h = |T|. Therefore, it
is sufficient to deploy h = |T| = 5 VLC-APs in some of the points within the
five PDAs in Figure 4.10(h) in order to achieve LoS coverage and connec-

69
Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.10. Procedure to find the minimum number of VLC-APs and their PDAs
to satisfy LoS coverage with connectivity via CTC method: (a) Deter-
mination of first maximal clique in the LoS graph as the first node
A1 in the connectivity tree; determination of the (b) second; (c) third;
(d) forth; and (e) fifth node of the connectivity tree; (g) Structure
of the connectivity tree; and (h) Representation of the nodes in the
connectivity tree (PDAs)

tivity for the sample layout. The following steps describe how VLC-APs
are deployed:

I. Deploy a VLC-AP anywhere within an arbitrary PDA from the set of

PDAs A1 , A2 , ..., Ah . Let the PDA and the VLC-AP deployment point
be A i and P i , respectively;

II. Assign the PDA by the single point A i := P i ;

III. Assign the root node R := A i in T and update all the PDAs via Algo-
rithm 4.3;

IV. Repeat the process until each PDA contains a single VLC-AP.

Accordingly, the five VLC-APs that result from using the CTC and PDA
updating methods will have LoS links both in access and backhaul through-
out the whole layout. Noneheless, the five VLC-APs using MCC method,
as shown in Figure 4.8(f), ensure LoS coverage only in access and not
necessarily in backhaul.

4.6.3 Simulation Results

Based on the proposed deployment methods, we generate a visual represen-

tation of the optimal number of VLC-APs and their locations. We compare
our results to to those obtained by conventional methods with maximum
ranges of long ( r = 10 m), medium ( r = 3 m), and short (r = 2 m). We applied

70
 Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.11. Deployments of LED-based VLC-APs for the given museum layout
with three maximum ranges to satisfy: (a) LoS coverage via MCC
deployment; (b) LoS coverage with connectivity via CTC method.

Figure 4.12. Conventional LED-based VLC-AP deployments applied to the mu-

seum layout with the maximum cell range r = 2 m via: (a) HEX
deployment; and (b) HEX+ deployment with LoS coverage.

the different deployment methods to the ground floor plan of the Jewish
Museum in London2 . This layout contains n = 67 vertices arranged within
a surface area of approximately 30 × 35 m.
The range of VLC channel in access is limited by the FoV of the LEDs
as a result of the incidence angle of the LoS link between the LED and
the UE. However, due to the fixed location of the VLC-APs, the infrared
transmitters and the receivers can be considered well adjusted to the LoS
link in backhaul. VLC-APs can, therefore, communicate beyond the size of
the museum with optical wireless backhaul if there is a LoS link between
them. As a result, we assume that optical wireless backhauling has an
unlimited range given the layout.
Figure 4.11(a) illustrates the deployment of the minimum number of VLC-
APs necessary to provide LoS coverage throughout the museum by using
MCC method. With r = 10 m, the MCC method suggests g = 8 VLC-APs,
2 See [Link]

71
Planning of New Spectrum-Based Indoor Wireless Networks

which are shown in red squares at the center points of the eight clique LoS
areas resulting from MCC method. We then reduce the range to r = 3 m and
r = 2 m, resulting in g = 24 and g = 48 VLC-APs, which are represented by
green circles and yellow triangles, respectively. Consequently, we observe
more VLC-APs with a smaller cell range in order to ensure LoS coverage.
It should be noted, however, that the dashed red lines between the four red
squares indicate that the MCC method does not guarantee connectivity
among VLC-APs. Figure 4.11(b) shows the results of the CTC method in
terms of determining the minimum number of VLC-APs to ensure LoS
coverage with connectivity. With r = 10 m, we obtain a total of h = 8 VLC-
APs covering the entire area of the projected area, maintaining connectivity
throughout (solid red lines) despite the distance r = 10 m. By setting
r = 10 m, we obtain a total of h = 8 VLC-APs covering the entire area and
maintaining connectivity (solid red lines). Furthermore, the CTC method
allocates h = 28 and h = 51 VLC-APs when r = 3 m and r = 2 m, respectively.
Based on this, the number of VLC-APs is not significantly different between
MCC and CTC. Additionally, it is observed that relocation of the VLC-APs
is sometimes sufficient to ensure LoS coverage and connectivity, as in the
case of cell ranges of r = 10 m.
As the next step, we illustrate two representative conventional indoor
VLC-AP deployments, in which HEX deployment describes a method for
shifting the layout over hexagonal cells so that the placement of VLC-
APs minimizes the indoor areas experiencing outages. According to Fig-
ure 4.12(a), 35 VLC-APs are deployed by the HEX deployment with r = 2 m;
gray triangles indicate VLC-APs that were dropped out. Therefore, the
white areas within the layout represent the zones in outage. Aside from
this, in the HEX+ deployment method, the HEX-deployed VLC-APs are
paired with as few additional VLC-APs as possible to provide coverage
of the entire indoor service area. The 32 red triangles in Figure 4.12(b)
represent additional VLC-APs in addition to the HEX-deployed ones, which
together provide LoS coverage for the entire layout. Accordingly, 35+32 = 67
VLC-APs with r = 2 m are required when using HEX+ to achieve LoS cov-
erage. It follows that the HEX deployment leaves a large number of zones
uncovered, whereas the HEX+ deployment increases the total number of
VLC-APs by a substantial amount. Finally, the high interference caused
by more VLC-APs degrades the data rate of users.

4.7 Connection Between the Deployment of Positioning Access

Points and the Accuracy of Indoor Positioning

The positioning of a User Equipment (UE) in an Angle of Arrival (AoA)-

based Indoor Positioning System (IPS) takes place via triangulation and is
possible only when it has LoS conditions involving at least two Positioning

72
 Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.13. Effects of P-AP (triangle marks) placement over positioning accuracy
of a UE (cross marks) based on triangulation principle; (a) An error-
free angle measurements, (b) a small positioning uncertainty due to
a well P-AP placement, (c) a large positioning uncertainty due to a
tight P-AP placement.

Access Points (P-AP)s [128]. As opposed to a VLC-AP, which is used

for optical wireless communications in access and/or backhaul, a P-AP
is used as a general term for a positioning reference node regardless of
technology. The measurement of angles is, however, typically subject to
errors, which may result in a significant level of positioning uncertainty
regardless of the technology used [129]. It is assumed that the heights of
the UE and Positioning Access Points (P-AP)s are known, and therefore
two-dimensional analysis is sufficient for positioning.
Figure 4.13 illustrates how positioning uncertainty in two dimensions
is affected by the location of two P-APs when triangulation is used. Ac-
cording to Figure 4.13(a), accurate positioning can be achieved with angle
measurements that are error-free when LoS conditions exist between the
UE and P-AP. Figure 4.13(b) illustrates a relatively low level of positioning
uncertainty (gray area) due to angle measurement errors. When the P-APs
are located too closely together, despite the LoS conditions between the UE
and the P-APs, the uncertainty can be large and even unbounded, 4.13(c).
In general, positioning uncertainty is minimized when the visibility angle
between the UE and the pair of P-APs is close to the right angle. As a
result, a twin P-APs are defined as a pair of P-APs with a distance larger
than the Minimum Separation Distance (MSD), denoted by d s . Using too
small MSD values results in too small visibility angles for most UEs in
the layout, resulting in a high degree of positioning uncertainty. On the
other hand, setting too large MSD values may be infeasible due to layout
geometry and may require too many P-APs to cover the entire layout.
It is essential that the UE is capable of measuring distances from three
P-APs simultaneously in order to perform a precise positioning using the
trilateration principle, which is similar to techniques based on Time of
Arrival (ToA) [128]. Therefore, it is imperative that LoS conditions are
established between the UE and the minimum of three P-APs. Distance
measurements are subject to errors in practice, which may lead to substan-
tial uncertainties in positioning. Figure 4.14 illustrates how positioning

73
Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.14. Effects of P-AP (triangle marks) placement over positioning accuracy
of a UE (cross marks) based on trilateration principle; (a) An error-
free distance measurement, (b) a small positioning uncertainty due
to noisy measurements in case of a good P-AP placement, (c) a tight
P-AP placement, (d) a co-linear P-AP placement.

uncertainty is affected by the location of three P-APs in two dimensions

in accordance with the trilateration principle. Figure 4.14(a) shows an
accurate positioning, and Figure 4.14(b) illustrates the error in the dis-
tance measurement for the same P-AP placement. Because of the good
placement of P-APs here, the positioning uncertainty caused by the dis-
tance measurement error is the smallest. There is a large positioning
uncertainty when P-APs are located at small distances, as shown in Figure
4.14(c). Figure 4.14(d) illustrates a large positioning uncertainty caused
by the placement of co-linear P-APs. In addition, it is still unclear whether
this placement will resolve the positioning ambiguity since both UEs (blue
cross-marks) are located on the normal to the line connecting the P-APS.
Similarly to triangulation, the positioning uncertainty of trilateration is
relatively small as long as one of the three visibility angles between the
UE and P-APs is close to the right angle. This is why we define a triplet
regarding two configuration parameters, MSD and Minimum Separation
Angle (MSA), represented by θs .

• Triplet of P − APs consists of three P-APs whose location points form

the vertices of a triangle q such that the three sides of q are greater
than d s and the three inner angles of q are greater than θs .

It is noteworthy that the MSD and MSA values have to be selected within
feasible intervals 0 ≤ d s ≤ 2r and 0◦ ≤ θs ≤ 60◦ , respectively, where r repre-

74
 Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.15. Three visibility angles between a UE (cross mark) and the P-APs
(triangle marks) and identifying the EVA.

sent the maximum range of P-APs.

In the context of an IPS based on trilateration, a P-AP deployment
task is required to ensure that LoS conditions are maintained between
the UEs and triplets, resulting in relatively low positioning uncertainty
regardless of the location of the UEs. Figure 4.15 clarifies the definition
of a triplet, where the vertices of q are formed by three P-APs placed at
the points Q i , Q j , and Q k . The set of P-APs is a triplet with d s and θs , if
d s ≤ ∥Q i Q j ∥, ∥Q i Q k ∥, ∥Q j Q k ∥ and θs ≤ Q̂ i ,Q̂ j ,Q̂ k . Consider a UE at point U
with LoS conditions to the triplet. We define non-reflex visibility angles
θ i j := Qˆ︂ ˆ︂ ˆ︂
i UQ j , θ ik := Q i UQ k , and θ jk := Q j UQ k , among which the Effective
Visibility Angle (EVA), denoted by θE , refers to the closest one to 90 degrees.
The EVA confinement is characterized as a function of the MSD and MSA
parameters.
Assume a triplet of P-APs, forming a triangle q, has LoS conditions to a
UE with the location point U . Then, the EVA is bounded around the right
angle by

|90◦ − θE | ≤ 90◦ − θs , (4.7)

if U lies inside q, and

(︁ d s θs )︁
|90◦ − θE | ≤ 90◦ − 2 × arctan tan ( ) , (4.8)
2r 2
if U lies outside q.
According to (4.7) and (4.8) when it comes to P-AP deployment, setting
larger MSD and MSA values brings the EVA closer to the right angle for
any UE anywhere on the layout. Due to physical restrictions, this type of
setting may not be feasible and would require a large number of P-APs.
The EVA will be far from the right angle if the MSD is small, as shown
in Figure 4.14(c). Similarly, an example of a small MSA value is given
in Figure 4.14(d). In light of this, setting appropriate values for these
two parameters may be difficult due to the shape of the layout and the
maximum range available.

75
Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.16. Determining the minimum number of P-APs and their placement
to ensure 1-LoS coverage with r = ∞; (a) partitioning of the sample
layout, (b) the primary LoS graph, (c) clique clusters derived by the
primary MCC method, and (d) the primary areas.

4.8 Network Planning for a Precise Indoor Positioning

A P-AP deployment satisfies 1-LoS coverage if and only if at least one

P-AP covers every point in the layout. Moreover, 2-LoS coverage is met in
case every point in the layout has LoS condition to at least a twin P-AP.
Finally, a P-AP deployment satisfies 3-LoS coverage in case every point in
the layout has LoS condition to at least a triplet of P-APs.

4.8.1 P-AP Deployment to Ensure 1-LoS Coverage

Several indoor positioning systems rely on the proximity principle to iden-

tify Cell IDs for a number of purposes, including location-aware services
and asset tracking, requiring a LoS condition between the UE and a single
P-AP [130]. We thus intend to determine the minimum number and loca-
tion of P-APs that can provide 1-LoS coverage, that is, an UE anywhere
in an arbitrary layout can establish a LoS condition to at least one P-AP.
For the P-AP deployment in this case, it is sufficient to apply similar steps
as the (primary) MCC method described in Algorithm 4.2. Figure 4.16
illustrates determining the minimum number of P-APs and their locations

76
 Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.17. Edge elimination method.

in a new sample layout via the primary MCC method with the unlimited
range r = ∞. Figure 4.16(a) shows partitioning the sample layout into
M = 14 polygons, and Figure 4.16(b) illustrates the equivalent (primary)
LoS graph. Using primary MCC method, Figure 4.16(c) represents par-
titioning the primary LoS graph into g = 4 (primary) cliques C1(1) , C2(1) , C3(1)
and C4(1) whose visibility areas, referred to as primary areas, are denoted
by A(1) (1) (1) (1)
1 , A2 , A3 and A4 , respectively, see Figure 4.16(d). Therefore, de-
ploying one P-AP inside each primary area guarantees 1-LoS coverage.
Furthermore, the s = 4 number of hidden points H2 , H2 , H3 , and H4 implies
that the s = 4 ≤ g = 4 number of P-APs found in Figure 4.16(d) is optimal
for 1-LoS coverage.

4.8.2 Access Point Deployment to Ensure 2-LoS Coverage

Given an MSD value, we determine the minimum number of P-APs and

their placement, which ensures 2-LoS coverage, which means one UE
anywhere in the layout has LoS conditions to at least a twin P-APs. The
AoA technique represents an example of an IPS that uses the triangulation
principle, where LoS conditions between a UE and two P-APs are crucial
to position a UE. As a naive approach to satisfying 2-LoS coverage, two
P-APs should be placed within each primary area, e.g., Figure 4.16(d). This
deployment, however, may be considered infeasible if the MSD value is
relatively large. Therefore, we introduce the following definition to ensure
enough distance between twin P-APs deployed.

• Forbidden region Fd s (A) refers to the locus whose distance to any

point within the input area A is shorter than the MSD value, i.e.,
⋂︁ ⃓
Fd s (A) := O∈A {Q ′ ⃓∥Q ′ O ∥ < d s }.

The region Fd s (A) is entirely covered by any circle with radius d s and its
center located within A. The forbidden region is a self-inverse area-area
mapping, i.e., Fd s (Fd s (A)) = A and therefore a larger A results in a smaller
Fd s (A). Specifically in case A is too large, we have Fd s (A) = ∅. Also, the

77
Planning of New Spectrum-Based Indoor Wireless Networks

Algorithm 4.5. Edge Elimination (EE)

Inputs:
G(1) [p 1 , p 2 , ..., p M ] % The primary LoS graph

S(1) , A(1) (1) (1)

1 , A2 , ..., A g % The outputs of Algorithm 4.2

ds % The MSD value

1 G(2) ←− G(1)

2 For i from 1 to M :
3 For j from 1 to M :
4 If p i , p j ∈ C (1)
k : % C (1)
k from S
(1)

5 If L(p i ) ∩ L(p j ) − Fd s (A(1)

k )==∅:

6 Edge-Eliminate(G(2) , p i , p j )

7
8 Return

Output:
G(2) [p 1 , p 2 , ..., p M ] % The secondary LoS graph

farthest point of a bounded area coincides with the boundary. Thus, we

can equivalently substitute O ∈ A with O ∈ ∂(A), where ∂(A) denotes the
boundary of A. Furthermore, in case ∂(A) forms a polygon with straight
edges, we can further limit the center points only to the vertices of A,
⋂︁ ⃓
i.e., Fd s (A) := O∈v(A) {Q ′ ⃓∥Q ′ O ∥ < d s }, where v(.) stands for the set of vertex
points.
We modify the primary LoS graph so that any pair of P-APs serving the
same UE forms a twin.

• Secondary LoS Graph, denoted by G(2) , refers to a modified version

of the primary LoS graph using Edge Elimination EE method. EE
method removes the edges whose both ending nodes fall within a
common primary clique C (1)
k for some 1 ≤ k ≤ g, and the overlapping
LoS area of the ending nodes entirely lies inside Fd s (A(1)
k ).

To clarify the EE method in Algorithm 4.5, the four primary areas in

Figure 4.16(d) are represented in Figure 4.17(a) whose forbidden regions
are indicated by dotted patterns. Representing G(1) in Figure 4.17(b),
we eliminate three edges, shown by red dashed lines, to attain G(2) . As
an example, considering the nodes p 6 and p 12 both within C4(1) , the area
L(p 6 ) ∩ L(p 12 ) totally lies inside Fd s (A(1)
4 ), and thus the edge between p 6 and
p 12 is [Link] deploying P-AP for 2-LoS coverage, the secondary
LoS graph serves as the starting point.

• Clique mapping M(S(1) , C) takes the set of primary cliques and an

arbitrary clique C ∈ G(2) as inputs and returns all the primary areas

78
 Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.18. Determining the minimum number of P-APs and their placement
to satisfy 2-LoS coverage with r = ∞ and the given MSD value d s ;
(a) Forbidden regions of the primary areas, (b) the secondary LoS
graph, (b-f) the steps of secondary MCC method, (f) g′ = 5 secondary
cliques, (g) the secondary areas, and (h) the P-AP (triangle marks)
deployment.

whose associated primary cliques share at least one node with C , i.e.,
M(S(1) , C) := {A(1) (1) (1)
k |C k ∈ S , C ∩ C k ̸= ∅, 1 ≤ k ≤ g}.
(1)

The definition of clique mapping is used to define well-spaced areas:

• Well−spaced area Wd s (S(1) , C) takes the set of primary cliques and an

arbitrary clique C ∈ G(2) as inputs and returns the area that lies inside
L(C) while outside the union of forbidden regions for the primary
⋃︁
areas in M(S(1) , C), i.e., Wd s (S(1) , C) := L(C) − u∈M(S(1) ,C) Fd s (u).

Deploying a P-AP within Wd s (S(1) , C) for a clique C ∈ G(2) ensures a dis-

tance larger than d s to the P-APs placed in M(S(1) , C). Given the secondary
LoS graph, the primary cliques, and the MSD value, the secondary MCC
method partitions G(2) step by step into the minimum number of cliques
C 1(2) , C 2(2) , ..., C (2) (2)
g′ , labeled as secondary cliques, such that W d s (S , C k ), with
(1)

1 ≤ k ≤ g′ , are all non-empty. Then, we label them as secondary areas

A(2) (2)
k := W d s (S , C k ) for 1 ≤ k ≤ g . Following the steps in Algorithm 4.6,
(1) ′

the secondary MCC approach sorts the nodes of G(2) in ascending degree
order and adds them to the first cluster C1(2) , provided the clique forms a
non-empty Wd s (S(1) , C1(2) ). After removing the maximal clique from G(2) , we
repeat the algorithm until the LoS graph contains no nodes.
Using the secondary MCC method, Figure 4.18 illustrates how to de-
termine the minimum number of P-APs and their placement for 2-LoS
coverage. Figure 4.18(a) shows the forbidden regions of the four primary
areas, and Figure 4.18(b) represents G(2) derived in Figure 4.17. An illus-

79
Planning of New Spectrum-Based Indoor Wireless Networks

Algorithm 4.6. Secondary Maximal Clique Clustering

Inputs:
G(2) [p 1 , p 2 , ..., p M ] % Output of Algorithm 4.5

S(1) % Output of Algorithm 4.2

ds % The MSD value

1 k ←− 0

2 While G(2) ̸= ∅:

3 G(2) [q 1 , q 2 , ..., q M ] ←− Ascending_Sort(G(2) )

4 k ←− k + 1

5 C k ←− ∅

6 For i from 1 to M :
7 If Is_Clique (G(2) , C k ∪ q i ) ==1:
8 If Wd s (S(1) , C k ∪ q i ) ̸= ∅
9 Ck ← Ck ∪ q i

10
11 G(2) ←− G(2) − C k % C k : The maximal clique

12 M ←− |G(2) |

13 C (2)
k ←− C k % The secondary clique

14 A(2) (1) (2)

k ←− W d s (S , C k ) % The secondary area

15 Return

Outputs:
S(2) := {C1(2) , C2(2) , ..., C (2)
g′
} % Set of secondary cliques

A(2) (2) (2)

1 , A2 , ..., A g′ % Secondary areas

tration of the step-by-step procedure in the secondary MCC method can

be seen in Figures 4.18(b-f). As shown in Figure 4.18(b), p 5 is selected
as the node with the minimum degree. Here, the nodes p 4 and p 5 form
a maximal clique in G(2) wherein both belong to the primary clique C1(1) ,
see Figure 4.16(c). Thus, we form the maximal clique [p 5 , p 4 ], labeled as
the first secondary clique C1(2) , because we have M(S(1) , C1(2) ) = {A(1) 1 } and the
requirement Wd s (S(1) , C1(2) ) = L(C1(2) ) − Fd s (A(1)
1 ) =
̸ ∅ is met. Figure 4.18(e)
depicts three secondary cliques derived through similar steps. Taking p 1
as the node with the lowest degree, we are able to construct C4(2) [p 1 , p 2 , p 3 ].
Despite the fact that the set of nodes [p 1 , p 2 , p 3 , p 9 ] in the remaining graph
form a clique, its well-spaced area is empty; thus, we do not add p 9 to
C 4(2) . Finally, we end up with g′ = 5 secondary cliques C 1(2) , C 2(2) , ..., C 5(2) , as in
Figure 4.18(g), whose secondary areas A(2) (2)
k := W d s (S , C k ) for 1 ≤ k ≤ 5 are
(1)

shown in Figure 4.18(h) with the same colors. Therefore, a total number of

80
 Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.19. Well-spaced area of the points Q i and Q j with the given MSD and
MSA values.

g + g′ = 9 P-APs is necessary and sufficient for 2-LoS coverage of the layout.

Here, we place each P-AP within the areas A(1) (1) (2) (2)
1 , ..., A4 , A1 , ..., A5 as in Fig-
ure 4.18(f) and Figure 4.18(g), such that the P-AP associated to A(2) k has the
maximum distance to the P-APs placed within the areas in M(S(1) , C (2) k ). It is
(2)
because the P-APs associated with the areas in M(S , C k ) have partly the
(1)

common areas to serve with the P-AP for A(2) (1)

k . For instance, since A1 is the
(2)
only element in M(S(1) , C1 ), two P-APs are placed on the corner points of
A(1) (2) (2) (1)
1 and A1 , as in Figure 4.18(h). Similarly, since M(S , C 5 ) = {A3 , A4 },
(1) (1)
(2)
we place the P-AP inside A5 on the corner point to have the maximum
distance to the P-APs associated with A(1) (1)
3 and A4 . Finally, the nine P-APs,
placed as in Figure 4.18(h), guarantee 2-LoS coverage of the sample layout.
It is worth mentioning that the minimum number of P-APs required to
provide 2-LoS coverage is equal to or greater than twice the hidden set
of points, i.e., 2s ≤ g + g′ . This is because at least two P-APs are required
inside the visibility area of each hidden point to ensure 2-LoS coverage. If
the MSD is set to a small value, equality may be maintained.

4.8.3 P-AP Deployment to Ensure 3-LoS Coverage

We specify the minimum number of P-APs and their locations in order

to ensure 3-LoS coverage. That is, a UE anywhere in a layout has LoS
conditions to at least a triplet of P-APs. The trilateration principle is
applied by ToA-based IPS applications that require 3-LoS coverage. In
order to establish 3-LoS coverage, a deployment for 2-LoS coverage, such
as the one shown in Figure 4.18(h), is a good starting point. We deploy a
set of additional P-APs on top of the set already placed for 2-LoS coverage
using the following definition.

• Well − spaced area Wd s ,θs (Q i ,Q j ) refers to the locus Q k that forms a

triplet with a twin P-APs located at Q i and Q j .

In other words, Wd s ,θs (Q i ,Q j ) is the locus of Q k such that d s ≤ ∥Q i Q k ∥, ∥Q j Q k ∥

and θs ≤ Q̂ i ,Q̂ j ,Q̂ k for the triangle q =: △Q i Q j Q k . Figure 4.19 illustrates the

81
Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.20. Determining the minimum number of P-APs and their placement
to achieve 3-LoS coverage with r = ∞ and the given MSD and MSA
values; (a) The P-AP deployment as in Figure 4.18(h), (b) the trinary
LoS graph, (c-f) the steps of the trinary MCC method, (f) the resulting
g′′ = 8 trinary cliques, (g) the trinary areas, and (h) the P-AP (triangle
marks) deployment.

well-spaced area of two given points Q i and Q j . By selecting Q k within R1 ,

we ensure that d s ≤ ∥Q i Q k ∥, ∥Q j Q k ∥, where R1 indicates the exterior of the
circles with the center points Q i and Q j at radius d s , see Figure 4.19(a). By
referring the shaded region in Figure 4.19(b) as R2 , choosing Q k within R2
ensures θs ≤ Q̂ i ,Q̂ j . As in Figure 4.19(c), we draw two circles that take the
segment Q i Q j as a chord with radius ∥Q i Q j ∥/2 sin θs , where R3 denotes the
inner space. Therefore, placing the third P-AP inside R3 ensures θs ≤ Q̂ k .
Finally, in Figure 4.19, we have Wd s ,θs (Q i ,Q j ) := R1 ∩ R2 ∩ R3 .
A secondary LoS graph is modified by taking into account the polygons
in the layout that are already covered by a triplet of P-APs, as described in
the following definition.

• Trinary LoS Graph, denoted by G(3) , is constructed by removing the

nodes in the secondary LoS graph whose LoS area contains at least a
triplet of P-APs placed for 2-LoS coverage.

By creating the trinary LoS graph, we increase the likelihood that the
resulting number of P-APs for 3-LoS coverage is the minimum. An il-
lustration of the steps of the trinary MCC method for determining the
minimum number and placement of P-APs for a 3-LoS coverage is shown
in Figure 4.20. Figure 4.20(a) displays the set of P-APs already placed for
2-LoS coverage as in Figure 4.18(h) while Figure 4.20(b) represents G(3) .
Due to the fact that P-APs located at Q 5 , Q 6 , and Q 7 form a triplet within
L(p 7 ), node p 7 has been removed. Figures 4.20(c-f) illustrate the steps
of the trinary MCC method for finding the minimum number of cliques

82
 Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.21. Minimum number of P-APs (triangle marks) and their placements
for 1-LoS coverage.

that partition G(3) . The single node p 6 is clustered solely as a maximal

clique, which is referred to as the first trinary clique C1(3) [p 6 ]. In this
case, since points Q 5 and Q 6 fall within L(p 6 ), the first trinary area can
be calculated as A(3) (3)
1 := L(C 1 ) ∩ W d s ,θs (Q 5 ,Q 6 ). The next minimum degree
node p 5 is grouped solely as the second trinary clique C2(3) [p 5 ], see Figure
4.20(c). Despite the fact that p 5 is capable of forming a clique with p 4
containing the twin P-APs located at Q 1 and Q 2 in their LoS area, we
avoid adding p 4 to C2(3) because L(p 5 ) ∩ L(p 4 ) ∩ Wd s ,θs (Q 1 ,Q 2 ) is empty. As a
result of this procedure, we are left with g′′ = 8 trinary cliques in Figure
4.20(f), whose trinary areas are shown in Figure 4.20(g). For instance, the
twin P-APs located at Q 4 and Q 7 fall inside L(p 9 ), while the twin P-APs
located at Q 6 and Q 7 lie inside L(p 8 ) ∩ L(p 10 ) ∩ L(p 12 ). Therefore, forming
the last trinary clique C8(3) [p 8 , p 9 , p 10 , p 12 ], the last trinary area is defined
as A(3) (3)
8 = L(C 8 ) ∩ W d s ,θs (Q 6 ,Q 7 ) ∩ W d s ,θs (Q 4 ,Q 7 ). The final step is to place
additional P-APs, one for each trinary area as in Figure 4.20(h), resulting
in a total of g + g′ + g′′ = 17 P-APs for the sample layout, thus ensuring
3-LoS coverage.
It is noteworthy that the minimum number of P-APs required for 3-LoS
coverage can be lower-bounded by three times the number of hidden points,
i.e., 3s ≤ g + g′ + g′′ . If the MSD and MSA values are set at small values,
equality may be maintained.

4.8.4 Simulation Results

The proposed algorithms are evaluated in terms of their ability to satisfy

1-LoS, 2-LoS, and 3-LoS coverage using a 22m × 22m sample layout. For
r = ∞ and r = 6m, we partition the layout by HT (R = 3m) and HT (R = 2m),
resulting in M = 387 and M = 762 triangles, respectively, representing the
number of nodes in the primary LoS graph.
Based on the primary MCC method, Figure 4.21 illustrates the minimum

83
Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.22. Minimum number of P-APs (triangle marks) and their placements
for 2-LoS coverage.

number and location of P-APs required to achieve a 1-LoS coverage. Based

on the representation in Figure 4.21(a), placing only 4 P-APs with an
unlimited maximum range ensures 1-LoS coverage. A 1-LoS coverage is
obtained through 9 P-APs when r is set to 6m as in Figure 4.21(b). In
order to achieve the same performance with a smaller range, a larger
number of P-APs is required. In accordance with the secondary MCC
method, Figure 4.22 illustrates the minimum number of P-APs, along

84
 Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.23. Minimum number of P-APs (triangle marks) and their placements
for 3-LoS coverage.

with their placement, required to ensure 2-LoS coverage. Setting r = ∞,

Figures 4.22(a-c) indicate the resulting 8, 8, and 9 P-APs for MSD values
d s = 1, 3, and 5m, respectively. For instance, Figure 4.22(c) shows that any
UE within the layout has LoS condition to at least two P-APs located at
a distance of at least five meters from each other. The deployment of 18,
20, and 34 P-APs are illustrated in Figures 4.22(d-f) when r = 6m, using
the same MSD values. A noteworthy point is that in Figure 4.22(f), even
though there are many pairs of P-APs within five meters of one another,

85
Planning of New Spectrum-Based Indoor Wireless Networks

Figure 4.24. Comparison of EVA cumulative distribution for a) the P-AP deploy-
ments in Figures 4.23(a-c) and (b) the P-AP deployments in Figures
4.23(d-f).

each UE falls within the LoS of at least a twin P-AP with an MSD value
equal to five meters. The number of P-APs required for 2-LoS coverage
may be dramatically increased as a result of a higher MSD value.
A trinary MCC method is used in Figure 4.23 to determine the minimum
number of P-APs that are required for 3-LoS coverage. Here, our MSD
value is supplemented by an MSA value. For r = ∞, Figures 4.23(a-c)
illustrate the location of the 13, 14, and 16 P-APs, when (d s , θs ) = (1m, 40◦ ),
(3m, 30◦ ), and (5m, 20◦ ), respectively. It can be observed that there are LoS
conditions associated with at least a triplet of P-APs for a UE located
anywhere in the layout. With r = 6m and MSD values of d s = 1, 2, and 3m,
Figures 4.23(d-f) outline the 34, 36, and 37 P-APs, respectively. UEs are not
necessarily served by the triplet of P-APs that are closest to them because
of the spatial constraints imposed by MSD and MSA values.
The CDF of EVA for UEs distributed uniformly within the sample layout
was calculated as a performance metric for positioning accuracy in Figure
4.24. Firstly, we find the triplets within the LoS area of the UE and select
the EVA that is closest to 90 degrees. Figure 4.24(a) compares the CDF of
EVAs for the P-APs placed as in Figures 4.23(a-c). It is observed in Figure
4.24(a) that deploying 13 P-APs as in Figure 4.23(a) results in only 40% of
the layout area having an EVA within the interval |90◦ − θE | < 30◦ , while the
deployment of 16 P-APs as in Figure 4.23(c) yields almost 80% of the layout
within the EVA interval |90◦ − θE | < 30◦ . Similarly, Figure 4.24(b) shows
that deploying 37 P-APs, as in Figure 4.23(f), provides the EVA satisfying
|90◦ − θE | < 30◦ for more than 95% of the layout area. As a result, setting

86
 Planning of New Spectrum-Based Indoor Wireless Networks

higher MSD values for UEs in the layout improves the EVA distribution
despite an increase in P-APs required.

4.9 Summary

For reliable communication and precise positioning, we examined the issue

of indoor deployment throughout this chapter. Indoor communication and
positioning are becoming more challenging as high frequency bands become
available. These challenges include obstructions and limited ranges, which
result in the loss of LoS link. According to the literature, reflected signals
from walls and obstacles in higher frequency bands are much weaker than
LoS links. For this reason, it is essential that the indoor service area be
deployed in a carefully planned manner in order to ensure high levels
of coverage and connectivity. Therefore, we proposed the MCC method
to determine the minimum number and locations of VLC-APs within an
indoor area in order to promote LoS connectivity both in optical wireless
access and in backhauling. In this method, the indoor area is modeled
using a visibility graph, and then the graph is divided into as few cliques as
possible. The number of cliques determines the minimum number of access
points, and the visibility areas of the cliques determine their deployment
areas. Based on the proposed CTC method, a connectivity tree was defined
and generated to ensure reliable optical wireless backhauling. Employing
MCC and CTC methods over a real floor plan showed a similar number of
VLC-APs that outperformed conventional deployment methods in terms of
data rates and illumination.
A LoS link between P-APs and the user is also required for unbiased
and precise positioning on higher frequency bands. Positioning accuracy
is limited by the effective visibility angle between a user and the P-APs.
In other words, positioning becomes more accurate as the visibility angle
approaches the right angle. Hence, we defined a minimum separation
distance and a minimum separation angle as configuration parameters for
the P-AP deployment process. Using these two parameters, we proposed
primary, secondary, and trinary MCC methods that guarantee 1-LoS, 2-
LoS, and 3-LoS coverage, respectively. As a result of the deployments, the
effective visibility angles of the layout area were close to 90 degrees for
a user within greater than 95 percent of the layout area, which in turn
guarantees small positioning uncertainty in these areas.
It is important to note that our theoretical analysis and abstract deploy-
ment algorithms have several limitations and simplifications. Our initial
assumption was that indoor channel modeling is dominated by line-of-
sight links; however, non-line-of-sight signals can contribute significantly
to communication and positioning when using mmWaves, THz, and optical
wireless signals. Secondly, we considered two-dimensional floor plans for

87
Planning of New Spectrum-Based Indoor Wireless Networks

deployment indoors. By considering the three-dimensional shapes of obsta-

cles and layouts, the analysis could be more accurate. Consequently, it may
be possible to modify the algorithm proposed in this chapter to take into
account both line-of-sight and non-line-of-sight links, as well as leverage
the algorithms for three-dimensional analysis.

88
5. Conclusions

This dissertation focuses on the development of mathematical tools for

optimizing the deployment of access points under a variety of wireless
network requirements and constraints. Due to the nature of the deploy-
ment problem as an NP-hard problem and the large dimensions of wireless
networks, determining the optimal number and location of access points
would require an exhaustive search, which is often infeasible. However,
mathematical tools such as Voronoi diagrams, Delaunay triangulation, and
graph modeling will significantly decrease the complexity of the deploy-
ment process. As a consequence of these approaches, mathematical models
are adapted to account for wireless network requirements, resulting in
significant improvements in performance and quality of service.
To achieve the maximum performance and minimize interference, it is
necessary to deploy adequate infrastructure, including towers, antennas,
and backhaul connectivity, as well as to manage and optimize the network
effectively. For a seamless user experience, a network should be able to
provide a variety of services and applications, including voice, data, and
video, and prioritize traffic according to user needs and network conditions.
This is where network planning plays a key role in improving the network
quality. In the case of a uniform traffic demand distribution, hexagonal
deployments of BSs are ideal on both load balancing and maximum cov-
erage. Nevertheless, in practice, traffic in the service area tends to be
irregular, heterogeneous, and changing over time. In Chapter 3, we first
proposed three algorithms for planning wireless cellular networks with
uneven traffic demand distributions. For load balancing, we applied power
Voronoi-based iterative algorithms. Then, we proposed heterogeneous
power Voronoi algorithm to take into account a set of network require-
ments using a penalty function. Finally, we proposed a local regularity
algorithm via Delaunay triangulation to maximize the coverage while
maintaining balanced loads. It was observed that the coverage probabil-
ity as a result of the proposed algorithm is as ideal as a hexagonal cell
deployment.
As discussed in Chapter 4, we addressed the issue of providing coverage

89
Conclusions

to indoor areas using high frequencies such as mmWaves, THz, and visible
light. Despite their unprecedented advantages, such as unlicensed spec-
trum and privacy, signals in higher frequency bands suffer from LoS link
blockage due to obstacles as well as a limited range. As a means of leverag-
ing these constraints into the optimization problem, graphs representing
common visibility areas of floor plan partitions were proposed. Taking into
account VLC networks, we proposed MCC and CTC methods, as tools for
graph partitioning that were able to minimize the number of VLC-APs
required to ensure LoS on both access and backhaul links. Using MCC will
give the VLC network better coverage and reduce co-channel interference
since fewer VLC-APs will be required. This method achieves a maximum
average data rate while a high minimum data rate is maintained at the
edges. Additionally, MCC plays a vital role in illumination comfort, which
correlates with KPIs related to energy efficiency.
Chapter 4 also studies deployment methods for indoor positioning. We
first showed that the placement of access points directly affects the uncer-
tainty of positioning. Then, we developed three algorithms to determine
the minimum number of access points and their precise placement in order
to guarantee LoS visibility regardless of the user’s position within the
indoor service area. Similar to the MCC method, a graph divided into a
minimum number of cliques was created for positioning methods based
on proximity. To improve the accuracy of positioning methods based on
triangulation (AoA/AoD) and trilateration (TDOA/RTT), two design pa-
rameters, i.e., minimum separation distance and minimum separation
angle, were introduced. These parameters control the placement of access
points which is linked to the accuracy of the positioning. As technology
evolved from mmWave (unrestricted indoors) to THz and optical wireless
(restricted indoors), we examined access points from a variety of service
ranges. We observed that as the range of the positioning system decreases,
the number of access points required for LoS visibility rapidly increases.
As a result of increasing the minimum separation distance between access
points, we observed an improvement in positioning accuracy over large
ranges. Finally, to ensure positioning precision/accuracy and reduce the
number of access points required, the minimum separation angle becomes
increasingly critical as the range decreases.
There are several directions for research in the area of wireless network
planning in the future:

• Optimal placement of UAVs: A sophisticated real-time algorithm for

optimizing UAV placement in three dimensions may ensure suffi-
cient coverage, capacity, and energy efficiency that can be applied to
targeted areas with dynamic traffic density. In the event that the
ground network fails, this placement can be integrated with ground
infrastructure to provide a reliable communication system.

90
 Conclusions

• Optimal placement of relays and IRS: In order to establish accurate

communication links, directional antennas can be placed and aligned
according to the geographic layout and the orientation of the coverage
area. The elevation and tilt of directional antennas is also important
for optimizing coverage for specific areas.

• Optimal indoor planning taking into account NLoS links: Consider-

ing both LoS and NLoS propagation when deploying access points
indoors is more effective than considering only LoS links. To achieve
this, a more representative channel model for indoor scenarios is
needed to be available.

• Optimal deployment of directional antennas: In order to establish

accurate communication links, directional antennas must be placed
as well as aligned and azimuthed according to the geographic layout
and the orientation of the coverage area. The elevation and tilt of
directional antennas is also important for optimizing coverage for
specific areas.

• Machine Learning for Deployment Optimization: In order to optimize

the deployment of access points, machine learning techniques can
be applied to predict user behavior, network dynamics, and other
relevant parameters.

• Sensing and IoT Integration: Investigate the possibilities for leverag-

ing access points for sensing applications integrated with the Internet
of Things (IoT). Sensor-based services such as environmental moni-
toring, smart city applications, and smart cities can be included in
this category.

• Security and Privacy: Investigate the security and privacy impli-

cations of access point deployment for several objectives such as
securing communication, protecting user privacy, and preventing
unauthorized access.

• Edge Computing Integration: Consider ways to optimize access points

deployment for edge computing, enabling applications such as aug-
mented reality, virtual reality, and real-time data analysis to process
information with low latency.

• Cross-Layer Design: Consider a cross-layer approach that takes into

account interactions between the physical layer, MAC layer, and
application layer to optimize access point deployment.

The items on this list address potential research directions and opportuni-
ties in wireless communication networks. A more sophisticated approach
to network planning will be required as technologies, use cases, and re-
quirements evolve beyond 5G networks.

91
References

[1] S. Alraih, I. Shayea, M. Behjati, R. Nordin, N. F. Abdullah, A. Abu-Samah,

and D. Nandi, “Revolution or evolution? technical requirements and consid-
erations towards 6g mobile communications,” Sensors, vol. 22, no. 3, p. 762,
2022.
[2] H. Holma, A. Toskala, and T. Nakamura, 5G technology: 3GPP new radio.
John Wiley & Sons, 2020.
[3] J. G. Andrews, S. Buzzi, W. Choi, S. V. Hanly, A. Lozano, A. C. Soong,
and J. C. Zhang, “What will 5g be?,” IEEE Journal on selected areas in
communications, vol. 32, no. 6, pp. 1065–1082, 2014.
[4] X. Lin and N. Lee, “5g and beyond,” Cham, Switzerland: Springer Nature
Switzerland AG, 2021.
[5] ITU, “Minimum requirements related to technical performance for IMT-
2020 radio interface(s),” tech. rep., ITU-R M.2410-0, 2017.
[6] R. Chataut and R. Akl, “Massive mimo systems for 5g and beyond net-
works—overview, recent trends, challenges, and future research direction,”
Sensors, vol. 20, no. 10, p. 2753, 2020.
[7] A. Kammoun, M.-S. Alouini, et al., “Elevation beamforming with full dimen-
sion mimo architectures in 5g systems: A tutorial,” IEEE Communications
Surveys & Tutorials, vol. 21, no. 4, pp. 3238–3273, 2019.
[8] E. G. Larsson, O. Edfors, F. Tufvesson, and T. L. Marzetta, “Massive mimo
for next generation wireless systems,” IEEE communications magazine,
vol. 52, no. 2, pp. 186–195, 2014.
[9] S. A. Busari, K. M. S. Huq, S. Mumtaz, L. Dai, and J. Rodriguez, “Millimeter-
wave massive mimo communication for future wireless systems: A survey,”
IEEE Communications Surveys & Tutorials, vol. 20, no. 2, pp. 836–869,
2017.
[10] A. Inoue, “Millimeter-wave gan devices for 5g: Massive mimo antenna
arrays for sub-6-ghz and mm-wave bandwidth,” IEEE Microwave Magazine,
vol. 22, no. 5, pp. 100–110, 2021.
[11] S. K. Sharma, T. E. Bogale, L. B. Le, S. Chatzinotas, X. Wang, and B. Otter-
sten, “Dynamic spectrum sharing in 5g wireless networks with full-duplex
technology: Recent advances and research challenges,” IEEE Communica-
tions Surveys & Tutorials, vol. 20, no. 1, pp. 674–707, 2017.
[12] X. Ge, S. Tu, G. Mao, C.-X. Wang, and T. Han, “5g ultra-dense cellular
networks,” IEEE Wireless Communications, vol. 23, no. 1, pp. 72–79, 2016.

93
References

[13] I. F. Akyildiz, S. Nie, S.-C. Lin, and M. Chandrasekaran, “5g roadmap: 10

key enabling technologies,” Computer Networks, vol. 106, pp. 17–48, 2016.
[14] T. S. Rappaport, Y. Xing, G. R. MacCartney, A. F. Molisch, E. Mellios, and
J. Zhang, “Overview of millimeter wave communications for fifth-generation
(5g) wireless networks—with a focus on propagation models,” IEEE Trans-
actions on antennas and propagation, vol. 65, no. 12, pp. 6213–6230, 2017.
[15] W. Hong, Z. H. Jiang, C. Yu, D. Hou, H. Wang, C. Guo, Y. Hu, L. Kuai, Y. Yu,
Z. Jiang, et al., “The role of millimeter-wave technologies in 5g/6g wireless
communications,” IEEE Journal of Microwaves, vol. 1, no. 1, pp. 101–122,
2021.
[16] S. Tripathi, N. V. Sabu, A. K. Gupta, and H. S. Dhillon, “Millimeter-wave
and terahertz spectrum for 6g wireless,” in 6G Mobile Wireless Networks,
pp. 83–121, Springer, 2021.

[17] D. Serghiou, M. Khalily, T. W. Brown, and R. Tafazolli, “Terahertz channel

propagation phenomena, measurement techniques and modeling for 6g
wireless communication applications: A survey, open challenges and future
research directions,” IEEE Communications Surveys & Tutorials, 2022.

[18] M. Inomata, W. Yamada, N. Kuno, M. Sasaki, K. Kitao, M. Nakamura,

H. Ishikawa, and Y. Oda, “Terahertz propagation characteristics for 6g
mobile communication systems,” in 2021 15th European Conference on
Antennas and Propagation (EuCAP), pp. 1–5, IEEE, 2021.
[19] N. Chi, Y. Zhou, Y. Wei, and F. Hu, “Visible light communication in 6g:
Advances, challenges, and prospects,” IEEE Vehicular Technology Magazine,
vol. 15, no. 4, pp. 93–102, 2020.
[20] K. Tutschku, N. Gerlich, and P. Tran-Gia, “An integrated approach to
cellular network planning,” in Proceedings of the 7th International Network
Planning Symposium (Networks 96), vol. 185190, 1996.

[21] E. Amaldi, A. Capone, and F. Malucelli, “Radio planning and coverage

optimization of 3g cellular networks,” Wireless Networks, vol. 14, no. 4,
pp. 435–447, 2008.
[22] R. Mathar and T. Niessen, “Optimum positioning of base stations for cellular
radio networks,” Wireless Networks, vol. 6, no. 6, pp. 421–428, 2000.
[23] H. D. Sherali, C. M. Pendyala, and T. S. Rappaport, “Optimal location of
transmitters for micro-cellular radio communication system design,” IEEE
Journal on Selected areas in communications, vol. 14, no. 4, pp. 662–673,
1996.
[24] Z. Altman, J. Picard, S. B. Jamaa, B. Fourestie, A. Caminada, T. Dony,
J. Morlier, and S. Mourniac, “New challenges in automatic cell planning of
umts networks,” in Proceedings IEEE 56th Vehicular Technology Conference,
vol. 2, pp. 951–954, IEEE, 2002.
[25] L. Al-Kanj, Z. Dawy, and G. Turkiyyah, “A mathematical optimization
approach for cellular radio network planning with co-siting,” Wireless Net-
works, vol. 18, pp. 507–521, 2012.
[26] E. Yaacoub and Z. Dawy, “Lte radio network planning with hetnets: Bs
placement optimization using simulated annealing,” in MELECON 2014-
2014 17th IEEE Mediterranean Electrotechnical Conference, pp. 327–333,
IEEE, 2014.

94
 References

[27] M. E. Mohammed and K. H. Bilal, “Lte radio planning using atoll radio
planning and optimization software,” International Journal of Science and
Research (IJSR), vol. 3, no. 10, pp. 1460–146, 2014.
[28] U. Challita, L. Al-Kanj, and Z. Dawy, “On lte cellular network planning
under demand uncertainty,” in 2014 IEEE Wireless Communications and
Networking Conference (WCNC), pp. 2079–2084, IEEE, 2014.
[29] A. Landström, H. Jonsson, and A. Simonsson, “Voronoi-based isd and
site density characteristics for mobile networks,” in 2012 IEEE Vehicular
Technology Conference (VTC Fall), pp. 1–5, IEEE, 2012.
[30] X. Ge, B. Yang, J. Ye, G. Mao, and Q. Li, “Performance analysis of poisson-
voronoi tessellated random cellular networks using markov chains,” in 2014
IEEE Global Communications Conference, pp. 4635–4640, IEEE, 2014.
[31] D. González and J. Hamalainen, “Planning and optimization of cellular¨
networks through centroidal voronoi tessellations,” in 2015 IEEE 82nd
Vehicular Technology Conference (VTC2015-Fall), pp. 1–2, IEEE, 2015.
[32] D. González, H. Hakula, A. Rasila, and J. Hämäläinen, “Spatial mappings
for planning and optimization of cellular networks,” IEEE/ACM Transac-
tions On Networking, vol. 26, no. 1, pp. 175–188, 2017.
[33] W. El-Beaino, A. M. El-Hajj, and Z. Dawy, “A proactive approach for lte radio
network planning with green considerations,” in 2012 19th International
Conference on Telecommunications (ICT), pp. 1–5, IEEE, 2012.
[34] S. M. Yu and S.-L. Kim, “Downlink capacity and base station density in
cellular networks,” in 2013 11th international symposium and workshops on
modeling and optimization in mobile, ad hoc and wireless networks (WiOpt),
pp. 119–124, IEEE, 2013.
[35] M. Mirahsan, R. Schoenen, and H. Yanikomeroglu, “Hethetnets: Heteroge-
neous traffic distribution in heterogeneous wireless cellular networks,”
IEEE Journal on Selected Areas in Communications, vol. 33, no. 10,
pp. 2252–2265, 2015.
[36] F. Lagum, Q.-N. Le-The, T. Beitelmal, S. S. Szyszkowicz, and
H. Yanikomeroglu, “Cell switch-off for networks deployed with variable
spatial regularity,” IEEE Wireless Communications Letters, vol. 6, no. 2,
pp. 234–237, 2017.
[37] F. Lagum, S. S. Szyszkowicz, and H. Yanikomeroglu, “Quantifying the regu-
larity of perturbed triangular lattices using cov-based metrics for modeling
the locations of base stations in hetnets,” in 2016 IEEE 84th Vehicular
Technology Conference (VTC-Fall), pp. 1–5, IEEE, 2016.
[38] A. Brochard, B. Błaszczyszyn, S. Mallat, and S. Zhang, “Statistical learning
of geometric characteristics of wireless networks,” in IEEE INFOCOM 2019-
IEEE Conference on Computer Communications, pp. 2224–2232, IEEE,
2019.
[39] F. Baccelli and A. Giovanidis, “A stochastic geometry framework for analyz-
ing pairwise-cooperative cellular networks,” IEEE Transactions on Wireless
Communications, vol. 14, no. 2, pp. 794–808, 2014.
[40] K. Huang and V. K. Lau, “Enabling wireless power transfer in cellular
networks: Architecture, modeling and deployment,” IEEE Transactions on
Wireless Communications, vol. 13, no. 2, pp. 902–912, 2014.
[41] S. Wang and C. Ran, “Rethinking cellular network planning and optimiza-
tion,” IEEE Wireless Communications, vol. 23, no. 2, pp. 118–125, 2016.

95
References

[42] A. Taufique, M. Jaber, A. Imran, Z. Dawy, and E. Yacoub, “Planning wireless

cellular networks of future: Outlook, challenges and opportunities,” IEEE
Access, vol. 5, pp. 4821–4845, 2017.
[43] M. L. Brandeau and S. S. Chiu, “An overview of representative problems in
location research,” Management science, vol. 35, no. 6, pp. 645–674, 1989.
[44] I. A. Hemadeh, K. Satyanarayana, M. El-Hajjar, and L. Hanzo, “Millimeter-
wave communications: Physical channel models, design considerations,
antenna constructions, and link-budget,” IEEE Communications Surveys &
Tutorials, vol. 20, no. 2, pp. 870–913, 2017.
[45] C. Han, Y. Wang, Y. Li, Y. Chen, N. A. Abbasi, T. Kürner, and A. F. Molisch,
“Terahertz wireless channels: A holistic survey on measurement, modeling,
and analysis,” IEEE Communications Surveys & Tutorials, vol. 24, no. 3,
pp. 1670–1707, 2022.
[46] S. Salous, K. Haneda, and V. Degli-Esposti, “5g to 6g: A paradigm shift in
radio channel modeling,” Radio Science, vol. 57, no. 7, p. e2022RS007447,
2022.
[47] D. Tse and P. Viswanath, Fundamentals of wireless communication. Cam-
bridge university press, 2005.
[48] D. S. Baum, J. Hansen, J. Salo, G. Del Galdo, M. Milojevic, and P. Kyösti, “An
interim channel model for beyond-3g systems: extending the 3gpp spatial
channel model (scm),” in 2005 IEEE 61st Vehicular Technology Conference,
vol. 5, pp. 3132–3136, IEEE, 2005.
[49] P. Kyosti, “Winner ii channel models,” IST, Tech. Rep. IST-4-027756 WIN-
NER II D1. 1.2 V1. 2, 2007.
[50] J. Meinilä, P. Kyösti, L. Hentilä, T. Jämsä, E. Suikkanen, E. Kunnari, and
M. Narandžić, “Document title: D5. 3: Winner+ final channel models,” S
president. His research interest includes millimeter wave antennas, beam-
forming network, dielectric resonator antennas, microstrip antennas, EBG,
etc. He has published over, 2010.
[51] B. Mondal, T. A. Thomas, E. Visotsky, F. W. Vook, A. Ghosh, Y.-H. Nam,
Y. Li, J. Zhang, M. Zhang, Q. Luo, et al., “3d channel model in 3gpp,” IEEE
Communications Magazine, vol. 53, no. 3, pp. 16–23, 2015.
[52] A. Ghazal, Y. Yuan, C.-X. Wang, Y. Zhang, Q. Yao, H. Zhou, and W. Duan,
“A non-stationary imt-advanced mimo channel model for high-mobility wire-
less communication systems,” IEEE Transactions on Wireless Communica-
tions, vol. 16, no. 4, pp. 2057–2068, 2016.
[53] L. Liu, C. Oestges, J. Poutanen, K. Haneda, P. Vainikainen, F. Quitin,
F. Tufvesson, and P. De Doncker, “The cost 2100 mimo channel model,”
IEEE Wireless Communications, vol. 19, no. 6, pp. 92–99, 2012.
[54] D. Karunatilaka, F. Zafar, V. Kalavally, and R. Parthiban, “Led based indoor
visible light communications: State of the art,” IEEE communications
surveys & tutorials, vol. 17, no. 3, pp. 1649–1678, 2015.
[55] B. Turan, O. Narmanlioglu, O. N. Koc, E. Kar, S. Coleri, and M. Uysal,
“Measurement based non-line-of-sight vehicular visible light communication
channel characterization,” IEEE Transactions on Vehicular Technology,
vol. 71, no. 9, pp. 10110–10114, 2022.
[56] T. Komine and M. Nakagawa, “Fundamental analysis for visible-light com-
munication system using LED lights,” IEEE Trans. Consumer Electronics,
vol. 50, pp. 100–107, Feb. 2004.

96
 References

[57] A. A. Dowhuszko, M. C. Ilter, and J. Hämäläinen, “Visible light communica-

tion system in presence of indirect lighting and illumination constraints,” in
ICC 2020-2020 IEEE International Conference on Communications (ICC),
pp. 1–6, IEEE, 2020.
[58] A. Dowhuszko and A. Pérez-Neira, “Achievable data rate of coordinated
multi-point transmission for visible light communications,” in Proc. IEEE
Int. Symp. Personal Indoor and Mobile Radio Commun., pp. 1–7, Oct. 2017.
[59] J. del Peral-Rosado, R. Raulefs, J. López-Salcedo, and G. Seco-Granados,
“Survey of cellular mobile radio localization methods: From 1G to 5G,” IEEE
Commun. Surveys & Tutorials, vol. 20, pp. 1124–1148, 2Q 2018.
[60] S. Modarres Razavi, F. Gunnarsson, H. Rydén, A. Busin, X. Lin, X. Zhang,
S. Dwivedi, I. Siomina, and R. Shreevastav, “Positioning in cellular net-
works: Past, present, future,” in Proc. IEEE Wireless Commun. and Net-
working Conf., pp. 1–6, Apr. 2018.

[61] E. Dahlman, S. Parkvall, and J. Sköld, 5G NR: The Next Generation Wireless
Access Technology. USA: Academic Press, Inc., 2nd ed., 2021.

[62] R. Keating, M. Säily, J. Hulkkonen, and J. Karjalainen, “Overview of posi-

tioning in 5G New Radio,” in Proc. Int. Symp. Wireless Commun. Systems,
pp. 320–324, Aug. 2019.
[63] B. Ren, R. Fang, X. Ren, G. Li, H. Li, Z. Zhao, J. Li, R. Da, and S. Sun,
“Progress of 3GPP Rel-17 standards on New Radio (NR) positioning,” in
Proc. Int. Conf. Indoor Positioning and Indoor Navigation, pp. 1–14, Nov.
2021.
[64] H. Holma, A. Toskala, and T. Nakamura, 5G Technology: 3GPP New Radio.
John Wiley & Sons, Ltd, 2020.

[65] A. Shahmansoori, G. E. Garcia, G. Destino, G. Seco-Granados, and

H. Wymeersch, “Position and orientation estimation through millimeter-
wave MIMO in 5G systems,” IEEE Trans. Wireless Commun., vol. 17, no. 3,
pp. 1822–1835, 2018.
[66] H. Soganci, S. Gezici, and H. V. Poor, “Accurate positioning in ultra-
wideband systems,” IEEE Wireless Commun., vol. 18, pp. 19–27, Apr. 2011.
[67] T. T. Brooks, H. H. Bakker, K. A. Mercer, and W. H. Page, “A review of
position tracking methods,” in 1st International conference on sensing tech-
nology, pp. 54–59, 2005.
[68] A. Goldsmith, Wireless communications. Cambridge university press, 2005.
[69] G. Fan and J. Zhang, “A novel geometric diagram and its applications in
wireless networks,” in IEEE INFOCOM 2004, vol. 1, IEEE, 2004.
[70] J. N. Portela and M. S. Alencar, “Cellular coverage map as a voronoi dia-
gram,” Journal of Communication and Information Systems, vol. 23, no. 1,
2008.
[71] B. Boots, K. Sugihara, S. N. Chiu, and A. Okabe, “Spatial tessellations:
concepts and applications of voronoi diagrams,” 2009.

[72] D. Attali and J.-D. Boissonnat, “A linear bound on the complexity of the
delaunay triangulation of points on polyhedral surfaces,” in Proceedings of
the seventh ACM symposium on Solid modeling and applications, pp. 139–
146, 2002.

97
References

[73] R. N. Clarke, “Expanding mobile wireless capacity: The challenges pre-

sented by technology and economics,” Telecommunications Policy, vol. 38,
no. 8-9, pp. 693–708, 2014.
[74] I. Siomina and D. Yuan, “Analysis of cell load coupling for lte network plan-
ning and optimization,” IEEE Transactions on Wireless Communications,
vol. 11, no. 6, pp. 2287–2297, 2012.
[75] M. Charitos, D. Kong, J. Cao, D. Berkovskyy, A. A. Goulianos, T. Mizutani,
F. Tila, G. Hilton, A. Doufexi, and A. Nix, “Lte-a virtual drive testing for ve-
hicular environments,” in 2017 IEEE 85th Vehicular Technology Conference
(VTC Spring), pp. 1–5, IEEE, 2017.
[76] K. Haneda, R. Rudd, E. Vitucci, D. He, P. Kyösti, F. Tufvesson, S. Salous,
Y. Miao, W. Joseph, and E. Tanghe, “Radio propagation modeling methods
and tools,” in Inclusive Radio Communications for 5G and Beyond, pp. 7–48,
Elsevier, 2021.

[77] A. Ferreira, L. Mendes, W. Dias, T. Marins, D. Gaspar, A. Matos, C. Silva,

and B. Sokal, “5g-range project field trial,” in 2019 European Conference on
Networks and Communications (EuCNC), pp. 490–494, IEEE, 2019.

[78] F. Aurenhammer and R. Klein, “Voronoi diagrams.,” Handbook of computa-

tional geometry, vol. 5, no. 10, pp. 201–290, 2000.
[79] A. Guo and M. Haenggi, “Spatial stochastic models and metrics for the
structure of base stations in cellular networks,” IEEE Transactions on
Wireless Communications, vol. 12, no. 11, pp. 5800–5812, 2013.
[80] M. Clerc, Particle swarm optimization, vol. 93. John Wiley & Sons, 2010.

[81] D. J. Elzinga and D. W. Hearn, “The minimum covering sphere problem,”

Management science, vol. 19, no. 1, pp. 96–104, 1972.
[82] S. Boyd, S. P. Boyd, and L. Vandenberghe, Convex optimization. Cambridge
university press, 2004.
[83] C. Bajaj, “The algebraic degree of geometric optimization problems,” Discrete
& Computational Geometry, vol. 3, pp. 177–191, 1988.

[84] 3GPP TSG RAN, “Study on channel model for frequencies from 0.5 to 100
Ghz,” TR 38.901, 3GPP, May 2017. Version 14.0.0.
[85] T. Jiang, J. Zhang, P. Tang, L. Tian, Y. Zheng, J. Dou, H. Asplund,
L. Raschkowski, R. D’Errico, and T. Jämsä, “3GPP standardized 5G channel
model for IIoT scenarios: A survey,” IEEE Internet of Things Journal, vol. 8,
pp. 8799–8815, June 2021.
[86] T. S. Rappaport, S. Sun, R. Mayzus, H. Zhao, Y. Azar, K. Wang, G. N.
Wong, J. K. Schulz, M. Samimi, and F. Gutierrez, “Millimeter wave mobile
communications for 5G cellular: It will work!,” IEEE Access, vol. 1, pp. 335–
349, 2013.
[87] A. Dowhuszko and A. Pérez-Neira, “Achievable data rate of coordinated
multi-point transmission for visible light communications,” in Proc. IEEE
Int. Symp. Personal Indoor and Mobile Radio Commun., pp. 1–7, Oct. 2017.
[88] A. Moldovan, M. Ruder, I. Akyildiz, and W. Gerstacker, “LOS and NLOS
channel modeling for terahertz wireless communication with scattered
rays,” in Proc. IEEE Global Commun. Conf. Workshops, pp. 388–392, Dec.
2014.

98
 References

[89] C. M. Armstrong, “The truth about terahertz,” IEEE Spectrum, vol. 49,
no. 9, pp. 36–41, 2012.
[90] A. Dowhuszko and B. Genovés Guzmán, “Closed form approximation of
the actual spectral power emission of commercial color LEDs for VLC,” J.
Lightwave Tech., vol. 40, no. 13, pp. 4311–4320, 2022.
[91] A. Dowhuszko, M. Ilter, and J. Hämäläinen, “Visible light communication
system in presence of indirect lighting and illumination constraints,” in
Proc. IEEE Int. Conf. Commun., pp. 1–6, June 2020.
[92] H. Abuella, M. Elamassie, M. Uysal, Z. Xu, E. Serpedin, K. A. Qaraqe,
and S. Ekin, “Hybrid rf/vlc systems: A comprehensive survey on network
topologies, performance analyses, applications, and future directions,” IEEE
Access, vol. 9, pp. 160402–160436, 2021.

[93] L. Cheng, H.-M. Tsai, W. Viriyasitavat, and M. Boban, “Comparison of radio

frequency and visible light propagation channel for vehicular communica-
tions,” in Proceedings of the First ACM International Workshop on Smart,
Autonomous, and Connected Vehicular Systems and Services, pp. 66–67,
2016.

[94] A. E. Ibhaze, P. E. Orukpe, and F. O. Edeko, “High capacity data rate system:
Review of visible light communications technology,” Journal of Electronic
Science and Technology, vol. 18, no. 3, p. 100055, 2020.
[95] A. Yadav and O. A. Dobre, “All technologies work together for good: A glance
at future mobile networks,” IEEE Wireless Communications, vol. 25, no. 4,
pp. 10–16, 2018.
[96] C. Chen, D. A. Basnayaka, and H. Haas, “Downlink performance of optical
attocell networks,” Journal of Lightwave Technology, vol. 34, no. 1, pp. 137–
156, 2015.

[97] K. Du, O. Ozdemir, F. Erden, and I. Guvenc, “Sub-terahertz and mmwave

penetration loss measurements for indoor environments,” in 2021 IEEE
International Conference on Communications Workshops (ICC Workshops),
pp. 1–6, IEEE, 2021.
[98] L. E. M. Matheus, A. B. Vieira, L. F. Vieira, M. A. Vieira, and O. Gnawali,
“Visible light communication: concepts, applications and challenges,” IEEE
Communications Surveys & Tutorials, vol. 21, no. 4, pp. 3204–3237, 2019.
[99] A. Vavoulas, H. G. Sandalidis, T. A. Tsiftsis, and N. Vaiopoulos, “Coverage
aspects of indoor vlc networks,” Journal of Lightwave Technology, vol. 33,
no. 23, pp. 4915–4921, 2015.
[100] L. Yin and H. Haas, “Coverage analysis of multiuser visible light communi-
cation networks,” IEEE Transactions on Wireless Communications, vol. 17,
no. 3, pp. 1630–1643, 2017.
[101] A. M. Vegni and M. Biagi, “Optimal led placement in indoor vlc networks,”
Optics express, vol. 27, no. 6, pp. 8504–8519, 2019.
[102] J. Li, X. Bao, and W. Zhang, “Led adaptive deployment optimization in
indoor vlc networks,” China Communications, vol. 18, no. 6, pp. 201–213,
2021.
[103] I. Stefan and H. Haas, “Analysis of optimal placement of led arrays for
visible light communication,” in 2013 IEEE 77th Vehicular Technology
Conference (VTC Spring), pp. 1–5, IEEE, 2013.

99
References

[104] L. Feng, R. Q. Hu, J. Wang, and Y. Qian, “Deployment issues and perfor-
mance study in a relay-assisted indoor visible light communication system,”
IEEE Systems Journal, vol. 13, no. 1, pp. 562–570, 2018.
[105] M. Rahaim, I. Abdalla, M. Ayyash, H. Elgala, A. Khreishah, and T. D. Little,
“Welcome to the crowd: Design decisions for coexisting radio and optical
wireless deployments,” IEEE Network, vol. 33, no. 5, pp. 174–182, 2019.
[106] M. Abedi, A. A. Dowhuszko, and R. Wichman, “Visible light communica-
tions: A novel indoor network planning approach,” in 2021 IEEE Global
Communications Conference (GLOBECOM), pp. 1–7, IEEE, 2021.
[107] M. Alzenad, M. Z. Shakir, H. Yanikomeroglu, and M.-S. Alouini, “Fso-based
vertical backhaul/fronthaul framework for 5g+ wireless networks,” IEEE
Communications Magazine, vol. 56, no. 1, pp. 218–224, 2018.
[108] E. Boch, “High-capacity ethernet backhaul radio systems for advanced
mobile data networks [application notes],” IEEE Microwave Magazine,
vol. 10, no. 5, pp. 108–114, 2009.
[109] G. Mendelson, “All you need to know about power over ethernet (poe) and
the ieee 802.3 af standard,” Internet Citation,[Online] Jun, p. 13, 2004.
[110] L. S. Shen, “Power over ethernet (poe) technical overview,” Center for Energy
and Environment, vol. 1, 2019.
[111] H. Kazemi, M. Safari, and H. Haas, “A wireless optical backhaul solution for
optical attocell networks,” IEEE Transactions on Wireless Communications,
vol. 18, no. 2, pp. 807–823, 2018.
[112] T. Komine and M. Nakagawa, “Integrated system of white led visible-
light communication and power-line communication,” IEEE transactions on
Consumer Electronics, vol. 49, no. 1, pp. 71–79, 2003.
[113] M. De Berg, M. Van Kreveld, M. Overmars, and O. Schwarzkopf, “Computa-
tional geometry,” in Computational geometry, pp. 1–17, Springer, 1997.
[114] Y. Tanaka, T. Komine, S. Haruyama, and M. Nakagawa, “Indoor visible light
data transmission system utilizing white led lights,” IEICE transactions on
communications, vol. 86, no. 8, pp. 2440–2454, 2003.
[115] T. Sharma, A. Chehri, and P. Fortier, “Review of optical and wireless back-
haul networks and emerging trends of next generation 5g and 6g technolo-
gies,” Transactions on Emerging Telecommunications Technologies, vol. 32,
no. 3, p. e4155, 2021.
[116] V. Jungnickel, D. Schulz, J. Hilt, C. Alexakis, M. Schlosser, L. Grobe,
A. Paraskevopoulos, R. Freund, B. Siessegger, and G. Kleinpeter, “Opti-
cal wireless communication for backhaul and access,” in 2015 European
Conference on Optical Communication (ECOC), pp. 1–3, IEEE, 2015.
[117] D. Schulz, M. Schlosser, C. Alexakis, K. Habel, J. Hilt, R. Freund, and
V. Jungnickel, “Optical wireless led link for the backhaul of small cells,” in
Optical Fiber Communication Conference, pp. M2F–8, Optica Publishing
Group, 2015.
[118] K. G. Rallis, V. K. Papanikolaou, P. D. Diamantoulakis, S. A. Tegos, A. A.
Dowhuszko, M.-A. Khalighi, and G. K. Karagiannidis, “Energy efficient
cooperative communications in aggregated vlc/rf networks with noma,”
IEEE Transactions on Communications, 2023.
[119] N. Stevens, “Bias introduced by true radiation patterns in rss-based visible
light positioning,” in 2019 IEEE SENSORS, pp. 1–4, IEEE, 2019.

100
 References

[120] M. G. Dobbins, A. Holmsen, and T. Miltzow, “Smoothed analysis of the art

gallery problem,” arXiv preprint arXiv:1811.01177, 2018.
[121] S. Fisk, “A short proof of chvátal’s watchman theorem,” Journal of Combi-
natorial Theory, Series B, vol. 24, no. 3, p. 374, 1978.
[122] Y. Amit, J. S. Mitchell, and E. Packer, “Locating guards for visibility cov-
erage of polygons,” International Journal of Computational Geometry &
Applications, vol. 20, no. 05, pp. 601–630, 2010.
[123] A. Efrat and S. Har-Peled, “Guarding galleries and terrains,” Information
processing letters, vol. 100, no. 6, pp. 238–245, 2006.
[124] C. Chen, S. Videv, D. Tsonev, and H. Haas, “Fractional frequency reuse in
dco-ofdm-based optical attocell networks,” Journal of Lightwave Technology,
vol. 33, no. 19, pp. 3986–4000, 2015.
[125] S. R. Chaudhry and H. S. Al-Raweshidy, “Consolidating multi-criteria
visible-light personal network coordinator selection process in attocells,”
Wireless Personal Communications, vol. 115, pp. 1935–1954, 2020.
[126] R. M. Karp, Reducibility among combinatorial problems. Springer, 2010.

[127] V. E. Alekseev, R. Boliac, D. V. Korobitsyn, and V. V. Lozin, “Np-hard graph

problems and boundary classes of graphs,” Theoretical Computer Science,
vol. 389, no. 1-2, pp. 219–236, 2007.
[128] M. A. C. Duran, A. A. D’Amico, D. Dardari, M. Rydström, F. Sottile, E. G.
Ström, and L. Taponecco, “Terrestrial network-based positioning and navi-
gation,” in Satellite and terrestrial radio positioning techniques, pp. 75–153,
Elsevier, 2012.
[129] Y. He, A. Behnad, and X. Wang, “Accuracy analysis of the two-reference-
node angle-of-arrival localization system,” IEEE Wireless Communications
Letters, vol. 4, no. 3, pp. 329–332, 2015.
[130] J. Armstrong, Y. A. Sekercioglu, and A. Neild, “Visible light positioning: A
roadmap for international standardization,” IEEE Communications Maga-
zine, vol. 51, no. 12, pp. 68–73, 2013.

101
 Aalto-DT 31/2024

BUSINESS +
ECONOMY
ISBN 978-952-64-1669-4 (pr int ed)
9 HSTFMG*ebggje+

ART +
ISBN 978-952-64-1670-0 (pdf)
DESIGN +
ISSN 1799-4934 (pr int ed) ARCHITECTURE
ISSN 1799-4942 (pdf)
SCIENCE +
T E C H N O L O GY
Aalto University
C R O S S OV E R
School of Electrical Engineering
Department of Information and Communications Engineering DOCTORAL
[Link].fi THESES