See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/385446024

Resource Allocation for RIS Assisted CoMP-NOMA Networks using

Reinforcement Learning Final Year Design Project

Preprint · April 2024

DOI: 10.13140/RG.2.2.29898.17606

CITATIONS READS

0 146

2 authors:

Muhammad Umer Muhammad Ahmed Mohsin

National University of Sciences & Technology National University of Sciences and Technology
8 PUBLICATIONS 3 CITATIONS 17 PUBLICATIONS 5 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Muhammad Ahmed Mohsin on 01 November 2024.

The user has requested enhancement of the downloaded file.

 Resource Allocation for RIS
Assisted CoMP-NOMA Networks
using Reinforcement Learning
Final Year Design Project

By
Muhammad Umer (CMS – 345834)
Muhammad Ahmed Mohsin (CMS – 333060)

Supervisor: Dr. Syed Ali Hassan

Co-supervisor: Dr. Huma Ghafoor

School of Electrical Engineering and Computer Science (SEECS)

National University of Sciences and Technology (NUST)
Islamabad, Pakistan

April 25, 2024

 Certificate

It is certified that the contents and form of the thesis entitled

“Resource Allocation for RIS Assisted CoMP-NOMA

Networks using Reinforcement Learning”

submitted by Muhammad Ahmed Mohsin and Muhammad Umer have been found
satisfactory for the requirements of the degree.

Advisor: Syed Ali Hassan, Ph.D.

Director, Information Processing and Transmission Lab.

Associate Professor, School of Electrical Engineering and Computer Science,
National University of Sciences and Technology (NUST)
Islamabad, 44000.

Co-Advisor: Huma Ghafoor, Ph.D.

Lecturer, School of Electrical Engineering and Computer Science,

National University of Sciences and Technology (NUST)
Islamabad, 44000.

i
 Abstract
This thesis delves into the forefront of wireless communication by exploring the syn-
ergistic integration of three transformative technologies: Simultaneously Transmit-
ting and Reflecting Reconfigurable Intelligent Surfaces (STAR-RIS), Coordinated
Multi-Point transmission (CoMP), and Non-Orthogonal Multiple Access (NOMA).
Driven by the ever-increasing demand for higher data rates, improved spectral ef-
ficiency, and expanded coverage in the evolving landscape of 6G development, this
research investigates the potential of these technologies to revolutionize future wire-
less networks.

The thesis analyzes the performance gains achievable through strategic deploy-
ment of STAR-RIS, focusing on mitigating inter-cell interference, enhancing signal
strength, and extending coverage to cell-edge users. Resource sharing strategies for
STAR-RIS elements are explored, optimizing both transmission and reflection func-
tionalities. Analytical frameworks are developed to quantify the benefits of STAR-
RIS assisted CoMP-NOMA networks under realistic channel conditions, deriving
key performance metrics such as ergodic rates and outage probabilities. Addition-
ally, the research delves into energy-efficient design approaches for CoMP-NOMA
networks incorporating RIS, proposing novel RIS configurations and optimization
algorithms to achieve a balance between performance and energy consumption. Fur-
thermore, the application of Deep Reinforcement Learning (DRL) techniques for
intelligent and adaptive optimization in aerial RIS-assisted CoMP-NOMA networks
is explored, aiming to maximize network sum rate while meeting user quality of ser-
vice requirements. Through a comprehensive investigation of these technologies and
their synergistic potential, this thesis contributes valuable insights into the future of
wireless communication, paving the way for the development of more efficient, reli-
able, and sustainable networks capable of meeting the demands of our increasingly
connected world.

ii
 Acknowledgements
We extend our sincere gratitude to Almighty Allah for granting us the strength,
guidance, and perseverance throughout the journey of completing this project. We
would like to express our heartfelt appreciation to our parents for their unwaver-
ing support, encouragement, and belief in our abilities. Their love, sacrifices, and
continuous encouragement have been the cornerstone of our success.

We are deeply grateful to our friends for their camaraderie, motivation, and support,
which made this journey memorable and enjoyable. We owe a debt of gratitude to
our supervisor, Dr. Syed Ali Hassan, for his invaluable guidance, support, and exper-
tise. His insightful feedback, encouragement, and patience have been instrumental
in shaping this thesis.

Special thanks are due to Dr. Haejoon Jung and Dr. Huma Ghafoor for their
assistance, valuable insights, and encouragement throughout this project. Their
expertise, encouragement, and willingness to help have been truly appreciated. We
are indebted to all those who have contributed in any way, no matter how small,
to the completion of this project. Thank you for your support, encouragement, and
belief in us.

iii
 Publication List
The main contributions of this research are either published or accepted or are
presently submitted for acceptance in journals and conferences as mentioned in the
following list:

Journal Articles
[1] M. Umer, M. A. Mohsin, M. Gidlund, H. Jung, and S. A. Hassan, “Analysis
of STAR-RIS Assisted Downlink CoMP-NOMA Multi-Cell Networks under
Nakagami-m Fading,” IEEE Communications Letters, 2024.

Conference Papers
[1] M. Umer, M. A. Mohsin, S. A. Hassan, H. Jung, and H. Pervaiz, “Performance
Analysis of STAR-RIS Enhanced CoMP-NOMA Multi-Cell Networks,” in 2023
IEEE Globecom Workshops (GC Wkshps), pp. 2000–2005, IEEE, 2023.

[2] M. Umer, M. A. Mohsin, S. A. Hassan, and H. Jung, “On the Energy Efficiency
and Passive Beamforming Design of RIS Assisted CoMP-NOMA Networks,”
Submitted to IEEE GLOBECOM 2024

[3] M. Umer, M. A. Mohsin, S. A. Hassan, and H. Jung, “Deep Reinforcement

Learning for Trajectory and Beamforming Optimization in CoMP-NOMA Net-
works with Aerial STAR-RIS,” Submitted to IEEE GLOBECOM 2024

iv
Table of Contents

Table of Contents vii

List of Figures viii

List of Tables x

1 Introduction 1
1.1 Project Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.4 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.5 Project Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.6 Organization of the Report . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Background 4
2.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.1.1 Challenges in Modern Wireless Communication . . . . . . . . 4
2.1.2 Emerging Technologies . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2.1 Non-Orthogonal Multiple Access (NOMA) . . . . . . . . . . . 6
2.2.2 Coordinated Multi-Point (CoMP) . . . . . . . . . . . . . . . . 7
2.2.3 Reconfigurable Intelligent Surfaces (RIS) . . . . . . . . . . . . 7
2.2.4 UAV-assisted Wireless Networks . . . . . . . . . . . . . . . . . 8
2.3 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3.1 STAR-RIS Enhanced CoMP-NOMA Networks . . . . . . . . . 8
2.3.2 DRL for Aerial RIS in CoMP-NOMA Networks . . . . . . . . 9
2.3.3 Energy Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4 Research Gaps and Opportunities . . . . . . . . . . . . . . . . . . . . 10
2.4.1 Performance Challenges in CoMP-NOMA Networks . . . . . . 10
2.4.2 Open Research Areas in STAR-RIS Technology . . . . . . . . 11
2.4.3 Statistical Gaps . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4.4 DRL in Aerial RIS Networks . . . . . . . . . . . . . . . . . . . 12

v
 Table of Contents

2.4.5 Energy Efficiency Assessment in Aerial RIS Networks . . . . . 14

2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3 Synergy of STAR-RIS, CoMP, and NOMA 17

3.1 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.1.1 Channel Model . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.1.2 STAR-RIS Configuration . . . . . . . . . . . . . . . . . . . . . 19
3.1.3 Rate Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.1.4 Outage Probability Analysis . . . . . . . . . . . . . . . . . . . 21
3.2 End-to-end SINR Statistics . . . . . . . . . . . . . . . . . . . . . . . 22
3.2.1 Effective Channel Characterization . . . . . . . . . . . . . . . 22
3.2.2 Probability Density Functions of SINRs . . . . . . . . . . . . . 24
3.3 Ergodic Rate (ER) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.4 Outage Probability (OP) . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.5 Simulation Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.6 Improvements in Ergodic Rate . . . . . . . . . . . . . . . . . . . . . . 29
3.7 Impact of RIS on Outage Probability . . . . . . . . . . . . . . . . . . 30
3.8 Exhaustive Search for Optimality Regions . . . . . . . . . . . . . . . 31

4 Energy Efficient Design for CoMP-NOMA Networks 32

4.1 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.1.1 Channel Model . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.1.2 RIS Configuration . . . . . . . . . . . . . . . . . . . . . . . . 35
4.2 Performance Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.2.1 Rate and Outage Probability Analysis . . . . . . . . . . . . . 35
4.2.2 Energy Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.3 Passive Beamforming Design . . . . . . . . . . . . . . . . . . . . . . . 37
4.4 Simulation Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.5 Impact of Cooperation on Energy Efficiency . . . . . . . . . . . . . . 38
4.6 Impact of RIS Elements on Energy Efficiency . . . . . . . . . . . . . 39
4.7 Outage Sum Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.8 Rate Threshold and Energy Efficiency . . . . . . . . . . . . . . . . . 42
4.9 Impact of PBF Design on Sum Rate . . . . . . . . . . . . . . . . . . 42

5 Deep Reinforcement Learning for Intelligent NOMA Networks 43

5.1 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.1.1 Channel Model & RIS Configuration . . . . . . . . . . . . . . 45
5.1.2 Signal Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5.2 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

vi
 Table of Contents

5.3 DRL-based Proposed Solution . . . . . . . . . . . . . . . . . . . . . . 48

5.3.1 MDP Formulation . . . . . . . . . . . . . . . . . . . . . . . . 48
5.3.2 MO-PPO Algorithm . . . . . . . . . . . . . . . . . . . . . . . 49
5.3.3 Complexity and Convergence Analysis . . . . . . . . . . . . . 51
5.4 Simulation Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.5 Learning and Convergence . . . . . . . . . . . . . . . . . . . . . . . . 52
5.6 Algorithmic Performance on Sum Rate . . . . . . . . . . . . . . . . . 53
5.7 MO-PPO and RIS Elements . . . . . . . . . . . . . . . . . . . . . . . 55
5.8 UAV Trajectory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

6 Conclusion 56
6.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
6.1.1 Performance Analysis . . . . . . . . . . . . . . . . . . . . . . . 56
6.1.2 Optimization Strategies . . . . . . . . . . . . . . . . . . . . . 57
6.1.3 Deep Reinforcement Learning . . . . . . . . . . . . . . . . . . 57
6.2 Limitations and Future Work . . . . . . . . . . . . . . . . . . . . . . 57

References 62

vii
List of Figures

3.1 An illustration of STAR-RIS-aided coordinated NOMA cluster. . . . . 18

3.2 The PDFs and CDFs of the SINRs at the center and edge user, with
K = 34 elements, mi,u = mi′ ,u = 1, and mi,R = mR,u = 2, ∀i ∈ I, i′ ∈
I \ {i}, ∀u ∈ U. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3 The PDFs and CDFs of the SINRs at the center and edge user, with
K = 34 elements, mi,u = mi′ ,u = 1, and mi,R = mR,u = 2, ∀i ∈ I, i′ ∈
I \ {i}, ∀u ∈ U. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4 Outage probability of network users versus Pt for equal amplitude
coefficients (β t = β r ), and element assignments (K1R = K2R ), when
K > 0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.5 Ergodic rate for varying RIS element assignments (K1R , K2R ) and am-
plitude adjustments (βt , βr ), with Pt = −10 dBm. . . . . . . . . . . . 31

4.1 An illustration of the RIS-assisted multi-cell CoMP-NOMA network. 33

4.2 Energy efficiency vs. number of cooperative BSs J at Pt = 0 dBm
and under various RIS configurations with K = 70 elements. . . . . . 39
4.3 Outage sum rate vs. transmit power Pt with J = 4 cooperative BSs
and K = 70 elements. . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.4 Energy efficiency contour plot for varying transmit power Pt and rate
threshold Rth with J = 4 cooperative BSs and K = 70 elements. . . . 40
4.5 Energy efficiency vs. number of RIS elements K at Pt = 0 dBm and
J = 4 cooperative BSs. . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.6 Outage sum rate vs. CO/EO split-ratio for different cooperative BSs
J with Pt = 0 dBm with K = 72 elements. . . . . . . . . . . . . . . . 41

5.1 Aerial RIS-assisted coordinated NOMA cluster. . . . . . . . . . . . . 44

5.2 Average cumulative reward vs. number of training episodes with
Pt = 20 dBm and K = 120 elements. . . . . . . . . . . . . . . . . . . 53
5.3 Sum rate vs. transmit power for different algorithms and configura-
tions with K = 120 elements. . . . . . . . . . . . . . . . . . . . . . . 53
5.4 Impact of the number of RIS elements on the achievable sum rate
with Pt = 10 dBm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

viii
 List of Figures

5.5 Top view of the UAV trajectory obtained by the MO-PPO algorithm
sampled every 25 time slots and averaged over 10 evaluation episodes 54

ix
List of Tables

3.1 Simulation Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . 29

4.1 Simulation Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . 39

5.1 Simulation Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . 52

x
Chapter 1

Introduction

This chapter provides an overview of the research project, outlining its motivation,
objectives, methodology, and expected outcomes. It also details the organization of
the report for clarity and navigation.

1.1 Project Overview

This research project delves into the synergistic potential of combining three cutting-
edge technologies: Simultaneously Transmitting and Reflecting Reconfigurable In-
telligent Surfaces (STAR-RIS), Coordinated Multi-Point transmission (CoMP), and
Non-Orthogonal Multiple Access (NOMA) to address the challenges of future wire-
less communication networks, particularly in the context of 6G development. The
project investigates the performance enhancements achievable through strategic de-
ployment of STAR-RIS for mitigating inter-cell interference and enhancing signal
strength, particularly for cell-edge users. Additionally, it explores resource shar-
ing strategies for STAR-RIS elements to optimize both transmission and reflection
functionalities within the network.

The project further analyzes the performance of STAR-RIS assisted CoMP-NOMA

networks under realistic channel conditions, deriving analytical expressions for key
performance metrics such as ergodic rates and outage probabilities. Moreover, the
project investigates energy-efficient design approaches for CoMP-NOMA networks
incorporating RIS, proposing different RIS configurations and optimization algo-
rithms for maximizing energy efficiency while maintaining desired performance lev-
els. Finally, the project explores the use of Deep Reinforcement Learning (DRL)
techniques, specifically multi-output proximal policy optimization (MO-PPO), to
optimize the joint configuration of unmanned aerial vehicles (UAVs), RIS phase

1
 Chapter 1. Introduction

shifts, and NOMA power control for maximizing network sum rate and meeting
user quality of service requirements.

1.2 Motivation
The ever-increasing demand for higher data rates, improved spectral efficiency, and
expanded coverage in wireless communication systems has pushed the boundaries
of existing technologies. Traditional approaches are struggling to keep pace with
these demands, necessitating the exploration of novel solutions. The limitations of
conventional RIS designs, the challenges of interference management in dense het-
erogeneous networks, and the need for more efficient spectrum utilization motivate
the investigation of STAR-RIS, CoMP, and NOMA as potential enablers for next-
generation wireless networks. By combining these technologies, we aim to achieve
significant performance improvements and pave the way for a more connected and
sustainable future.

1.3 Objectives
The primary objectives of this research project are:

• To analyze and quantify the performance gains achievable through the inte-
gration of STAR-RIS in CoMP-NOMA networks, particularly in terms of data
rates, coverage, and spectral efficiency.

• To develop efficient resource sharing strategies for STAR-RIS elements to op-

timize both transmission and reflection functionalities within the network.

• To design and evaluate energy-efficient solutions for RIS-assisted CoMP-NOMA

networks, considering different RIS configurations and optimization algorithms.

• To explore the application of DRL techniques for joint optimization of UAV

trajectory, RIS phase shifts, and NOMA power control in aerial RIS-assisted
CoMP-NOMA networks.

1.4 Methodology
The research methodology will involve a combination of theoretical analysis, simu-
lation modeling, and algorithm development. Analytical tools will be employed to
derive performance metrics and gain insights into the behavior of STAR-RIS assisted
CoMP-NOMA networks under various channel conditions. Simulation models will

2
 Chapter 1. Introduction

be developed to evaluate the performance of different system configurations and re-

source allocation strategies. Optimization algorithms will be designed to maximize
energy efficiency, spectral efficiency, and other key performance indicators. Addi-
tionally, DRL techniques will be explored for joint optimization in aerial RIS-assisted
CoMP-NOMA networks.

1.5 Project Outcomes

The expected outcomes of this research project include:

• Development of efficient resource sharing strategies for STAR-RIS elements.

• A comprehensive understanding of the performance benefits and limitations

of STAR-RIS assisted CoMP-NOMA networks.

• Design and evaluation of energy-efficient solutions for RIS-assisted CoMP-

NOMA networks.

• Implementation and analysis of DRL-based optimization techniques for aerial

RIS-assisted CoMP-NOMA networks.

• Contribution to the advancement of knowledge and understanding of RIS,

CoMP, and NOMA technologies for future wireless communication systems.

1.6 Organization of the Report

This report is structured to provide a comprehensive and logical flow of information.
Chapter 1 serves as the introduction, outlining the research project’s motivation,
objectives, methodology, and expected outcomes. Chapter 2 delves into the back-
ground, offering a detailed review of relevant literature on NOMA, CoMP, RIS, and
STAR-RIS technologies. Chapter 3 explores the synergistic potential of combining
STAR-RIS with CoMP-NOMA networks, analyzing their performance under real-
istic channel conditions. Chapter 4 investigates energy-efficient design approaches
for RIS-assisted CoMP-NOMA networks. Chapter 5 explores the application of
DRL techniques for intelligent optimization in aerial RIS-assisted CoMP-NOMA
networks. Finally, Chapter 6 concludes the report by summarizing the key findings
and discussing potential future research directions.

3
Chapter 2

Background

2.1 Preliminaries
The relentless expansion of wireless communication necessitates continuous inno-
vation to address escalating demands for data rates, spectral efficiency, and cover-
age. This thesis delves into the exploration of cutting-edge technologies, namely
Non-Orthogonal Multiple Access (NOMA), Coordinated Multi-Point (CoMP), and
Reconfigurable Intelligent Surfaces (RIS), with a particular focus on Simultaneously
Transmitting and Reflecting RIS (STAR-RIS), to tackle the limitations of conven-
tional wireless systems.

This introductory section establishes the foundation for understanding these tech-
nologies and their potential impact. It begins by outlining the key challenges faced
by modern wireless communication systems and subsequently introduces the afore-
mentioned solutions, highlighting their operating principles and potential benefits.

2.1.1 Challenges in Modern Wireless Communication

The proliferation of data-intensive applications, such as high-definition video stream-
ing, virtual reality, and the Internet of Things (IoT), has placed immense pressure
on existing wireless infrastructure. Key challenges include:

• Data Rate Demands: The need for faster data transfer speeds to support
bandwidth-hungry applications is ever-increasing.

• Spectral Efficiency: Efficient utilization of the limited radio spectrum is

crucial to accommodate the growing number of users and devices.

• Coverage: Ensuring ubiquitous and reliable wireless coverage across diverse

4
 Chapter 2. Background

environments, including urban canyons and remote areas, remains a significant

challenge.

Traditional approaches, such as increasing carrier frequencies or deploying additional

base stations, offer limited scalability and often introduce new complexities. Conse-
quently, exploring and developing novel technologies is essential to overcome these
hurdles and pave the way for next-generation wireless communication systems.

2.1.2 Emerging Technologies

Several promising technologies have emerged to address the limitations of conven-
tional wireless systems:

Non-Orthogonal Multiple Access (NOMA)

NOMA departs from the traditional orthogonal multiple access (OMA) paradigm by
leveraging power-domain multiplexing to serve multiple users on the same resource
block. By allocating higher power levels to users experiencing weaker channel con-
ditions, NOMA ensures successful decoding at the receiver while simultaneously
improving spectral efficiency and user fairness. This technique offers significant ad-
vantages over OMA, including increased system capacity and enhanced service for
cell-edge users.

Coordinated Multi-Point (CoMP)

CoMP enhances network performance through coordinated transmission and recep-

tion among multiple base stations. This coordination effectively mitigates inter-cell
interference (ICI), leading to improved signal quality, increased network capacity,
and extended coverage, particularly at cell edges. However, implementing CoMP
presents challenges related to channel state information (CSI) acquisition and syn-
chronization among base stations.

Reconfigurable Intelligent Surfaces (RIS)

RIS introduces a novel paradigm by employing software-controlled metasurfaces to

manipulate electromagnetic waves. These surfaces, comprised of numerous reconfig-
urable elements, can dynamically adjust the phase and amplitude of incident waves
to achieve desired reflection, refraction, or scattering. While offering significant po-
tential for enhancing wireless environments, traditional RIS designs suffer from the
limitation of only operating in the reflection or transmission mode, known as the
“half-space” problem.

5
 Chapter 2. Background

Simultaneously Transmitting and Reflecting RIS (STAR-RIS)

STAR-RIS overcomes the limitations of traditional RIS by incorporating active el-

ements alongside passive reflectors. This enables simultaneous transmission and
reflection of signals, providing greater flexibility in manipulating electromagnetic
waves and further enhancing wireless communication capabilities. STAR-RIS holds
immense potential for improving signal coverage, boosting system capacity, and in-
creasing spectral efficiency, particularly in challenging propagation environments.

The subsequent sections of this thesis will delve deeper into these technologies,
exploring their theoretical foundations, practical implementations, and potential
applications in the context of future wireless communication systems.

2.2 Related Work

This section explores existing research on key technologies relevant to this thesis:
Non-Orthogonal Multiple Access (NOMA), Coordinated Multi-Point (CoMP), Re-
configurable Intelligent Surfaces (RIS), and their application in multi-cell networks.
These technologies address critical challenges in modern wireless communication
systems, aiming to increase data rates, improve spectral efficiency, and enhance
coverage.

2.2.1 Non-Orthogonal Multiple Access (NOMA)

The work in [1] delves into the potential of NOMA for 5G and beyond networks.
Recognizing the limitations of traditional orthogonal multiple access (OMA) in ac-
commodating the escalating demands of data-intensive applications, the authors
highlight the advantages of NOMA in enhancing bandwidth efficiency by serving
multiple users within the same resource block. The paper provides a comprehen-
sive overview of power-domain multiplexing aided NOMA, encompassing theoreti-
cal foundations, multiple antenna designs, interaction with cooperative transmission
schemes, resource management strategies, and co-existence with other 5G technolo-
gies. Additionally, it identifies existing challenges and proposes potential solutions,
offering valuable design guidelines and outlining future research directions in this
domain.

Further exploring NOMA, the survey in [2] offers a detailed analysis of its progress in
5G systems, focusing primarily on power-domain NOMA employing superposition
coding at the transmitter and successive interference cancellation at the receiver.
The paper delves into various aspects of NOMA, including capacity analysis, power

6
 Chapter 2. Background

allocation strategies, user fairness considerations, and integration with other wireless
technologies such as cooperative communications, MIMO, beamforming, and net-
work coding. Implementation challenges and potential avenues for future research
are also discussed, providing a holistic understanding of NOMA and its potential
impact on future wireless networks.

2.2.2 Coordinated Multi-Point (CoMP)

The study presented in [3] investigates CoMP techniques for LTE-Advanced sys-
tems, focusing on coordinated transmission and reception among multiple points to
mitigate interference and enhance signal quality. The authors evaluate the potential
performance benefits of CoMP across various deployment scenarios with varying
traffic loads. Furthermore, the study delves into implementation aspects, practical
challenges, and deployment considerations for CoMP in LTE-Advanced networks.
Building upon the concept of CoMP, the work in [4] explores base station coopera-
tion in the downlink of heterogeneous cellular networks, specifically focusing on joint
transmission scenarios. Using stochastic geometry, the authors derive expressions
for network coverage probability, considering a typical user receiving data from a
pool of cooperating base stations. The analysis demonstrates significant gains in
coverage probability, particularly when cooperation involves multiple base stations,
and emphasizes the importance of coherent joint transmission in achieving diversity
gain.

2.2.3 Reconfigurable Intelligent Surfaces (RIS)

The study in [5] investigates the use of RIS for enhancing downlink multi-user com-
munication in a multi-antenna base station scenario. The authors focus on develop-
ing energy-efficient resource allocation strategies for transmit power and RIS phase
shifts while guaranteeing individual link budget requirements for users. To address
the resulting non-convex optimization problems, two computationally efficient algo-
rithms are proposed. The paper also introduces a realistic power consumption model
for RIS-based systems and evaluates the proposed methods in a realistic outdoor
environment. Results demonstrate significant energy efficiency gains compared to
conventional relaying techniques.

Further exploring RIS technology, the work in [6] provides a comprehensive overview
of its potential for revolutionizing wireless communication networks. The paper
delves into the applications, advantages, hardware architecture, and signal model of
RIS, highlighting its ability to intelligently manipulate the wireless propagation en-
vironment using passive reflecting elements. The authors also discuss the challenges

7
 Chapter 2. Background

associated with designing and implementing hybrid wireless networks incorporating

both active and passive components. Numerical results showcase the performance
improvements achievable with RIS in typical wireless network scenarios.

2.2.4 UAV-assisted Wireless Networks

While not directly related to RIS, CoMP, or NOMA, the inclusion of UAV-assisted
wireless networks in the original literature review presents an opportunity to dis-
cuss the potential integration of these technologies. The survey in [7] discusses the
increasing role of unmanned aerial vehicles (UAVs) in enhancing transmission effi-
ciency and coverage in wireless communication systems, particularly in the context
of 5G and beyond networks utilizing millimeter wave (mmWave) frequencies. The
paper offers a comprehensive overview of integrating 5G mmWave communications
into UAV-assisted networks, presenting a taxonomy of research issues and solutions,
and discussing technical advantages, challenges, and potential applications. This
opens avenues for exploring the synergistic benefits of combining UAV-assisted net-
works with RIS, CoMP, and NOMA to further enhance coverage, capacity, and
spectral efficiency in future wireless systems.

Furthermore, the work in [8] proposes a novel framework for designing trajectories
of multiple UAVs by predicting users’ mobility information. This framework aims
to maximize the instantaneous sum transmit rate while satisfying user rate require-
ments. Integrating such trajectory design and user mobility prediction techniques
with RIS, CoMP, and NOMA could pave the way for highly efficient and adaptable
wireless networks capable of dynamically responding to user demands and environ-
mental changes.

2.3 Related Work

This section explores existing research on the integration of STAR-RIS, CoMP, and
NOMA technologies in various wireless communication scenarios. The focus is on
understanding how these technologies are combined to address challenges related to
coverage, spectral efficiency, and energy efficiency.

2.3.1 STAR-RIS Enhanced CoMP-NOMA Networks

The work in [9] introduces a novel design for STAR-RIS within a NOMA-enhanced
CoMP network. Building upon existing signal enhancement and cancellation ap-
proaches, the authors propose a simultaneous signal enhancement and cancellation

8
 Chapter 2. Background

based (SSECB) design that leverages a large number of RIS elements to concur-
rently eliminate inter-cell interference and boost desired signals. Simulation results
demonstrate the effectiveness of SSECB in outperforming conventional designs and
achieving superior performance in CoMP-NOMA networks.

Further exploring the potential of STAR-RIS in NOMA systems, the study in [10] in-
vestigates its application in enhancing coverage quality and spectral efficiency. The
authors analyze the secrecy performance of a STAR-RIS aided downlink NOMA
system employing the energy splitting protocol. Analytical expressions for secrecy
outage probability (SOP) are derived, and asymptotic performance analysis is con-
ducted to gain insights into system behavior. The results demonstrate the superior
secrecy performance of STAR-RIS-NOMA compared to conventional OMA systems.

2.3.2 Deep Reinforcement Learning (DRL) for Aerial RIS

in CoMP-NOMA Networks
The integration of UAVs with RIS in CoMP-NOMA networks presents exciting op-
portunities for further enhancing coverage and capacity. The work in [11] proposes
a novel approach for maximizing communication efficiency in a multi-UAV system
assisted by RIS. A hybrid learning scheme combining multi-agent DRL and alternat-
ing optimization is employed to optimize UAV trajectories, cooperative beamform-
ing, and RIS passive beamforming (PBF). The proposed framework demonstrates
superior performance compared to conventional systems and heuristic algorithms,
achieving higher communication rates and fast convergence.

Further exploring the use of RIS in UAV-assisted communication, the study in [12]
investigates the joint optimization of UAV placement and RIS phase-shift to max-
imize data transfer rates while minimizing UAV energy consumption. DRL algo-
rithms, specifically DDQN and DDPG, are employed to address the optimization
challenges. Numerical results demonstrate the effectiveness of these algorithms in
improving the energy efficiency of RIS-assisted UAV systems compared to bench-
mark solutions.

The work in [13] introduces an energy harvesting (EH) scheme for UAV-RIS sys-
tems operating in communication-disabled areas. The proposed EH-RIS scheme
utilizes SWIPT technology to simultaneously transport information and harvest
energy through a split passive reflecting array. A robust DRL-based algorithm is
developed for efficient resource allocation in dynamic environments with pedestrian
mobility and rapid channel changes. Simulation results showcase the effectiveness
and efficiency of the proposed EH-RIS system, surpassing existing solutions and

9
 Chapter 2. Background

achieving near-optimal performance.

2.3.3 Energy Efficiency of RIS-assisted Multi-Cell NOMA

Networks
Energy efficiency (EE) is a critical consideration in the design of future wireless
networks. The paper in [14] focuses on maximizing EE in multi-cell multi-carrier
NOMA networks while considering hardware impairments. The authors propose a
two-stage approach, employing the BWOA algorithm for user association and sub-
channel assignment, and SPCA for power allocation. Simulation results demonstrate
the effectiveness of the proposed algorithm in achieving comparable performance to
existing methods while surpassing benchmarks for both NOMA and OMA systems.

Another study on EE maximization in multi-cell multi-carrier NOMA networks is

presented in [14]. The authors propose a matching-based framework for user associ-
ation and a two-stage quadratic transform approach for power allocation to address
the non-convex EE maximization problem. Numerical results demonstrate the su-
perior EE performance of the proposed method compared to existing approaches for
NOMA and OMA systems.

Expanding on the concept of EE, the work in [15] explores secure energy efficiency
maximization in an RIS-assisted uplink wireless communication system involving a
UAV acting as a mobile relay. The authors propose an algorithm for joint opti-
mization of UAV trajectory, RIS phase shift, user association, and transmit power
to maximize system efficiency. Simulation results demonstrate significant gains in
secure energy efficiency compared to traditional schemes without RIS integration.

2.4 Research Gaps and Opportunities

This section identifies key research gaps and opportunities in the existing litera-
ture on RIS-enhanced CoMP-NOMA networks. Addressing these gaps is crucial for
realizing the full potential of these technologies in future wireless communication
systems.

2.4.1 Performance Challenges in CoMP-NOMA Networks

While CoMP-NOMA networks offer significant potential for enhancing spectral ef-
ficiency, several challenges hinder their practical implementation:

• Channel State Information (CSI) Acquisition: Accurate CSI is crucial

10
 Chapter 2. Background

for effective coordination among base stations in CoMP systems. However,

acquiring accurate CSI in dynamic wireless environments with RIS deploy-
ments is a complex task due to the additional channel paths introduced by
the RIS. Developing efficient and robust CSI acquisition techniques tailored
for RIS-assisted CoMP-NOMA networks is essential.

• Synchronization Issues: Precise synchronization among base stations is

critical for CoMP operation to avoid inter-cell interference. The introduction
of RIS further complicates synchronization requirements due to the potential
for additional delays and phase shifts introduced by the reflecting elements.
Developing robust synchronization protocols that account for the unique char-
acteristics of RIS is necessary.

• Signal Processing Complexity: NOMA requires more complex signal pro-

cessing compared to traditional OMA techniques, both at the base stations
and user equipment. This increased complexity can lead to higher power
consumption and computational costs. Efficient signal processing algorithms
optimized for RIS-assisted CoMP-NOMA systems are needed to mitigate these
challenges.

2.4.2 Open Research Areas in STAR-RIS Technology

STAR-RIS presents a promising evolution of RIS technology, but its integration into
wireless communication systems requires further investigation in several key areas:

Channel Modeling: Accurate channel models that capture the unique characteristics
of STAR-RIS, including its simultaneous transmitting and reflecting capabilities,
are crucial for system design and performance evaluation. These models should
account for mutual coupling between elements, the interaction between transmitted
and reflected signals, and the impact of different RIS configurations on channel
characteristics.

• Capacity Analysis: Determining the theoretical limits of STAR-RIS systems

in terms of achievable data rates is essential for understanding its potential
benefits and limitations. This analysis should consider various factors such as
RIS size, element spacing, operating frequency, and channel conditions.

• Beamforming Optimization: Designing efficient beamforming algorithms

for STAR-RIS that jointly optimize reflection and transmission while consid-
ering practical constraints is a challenging task. Advanced optimization tech-
niques are needed to achieve optimal performance while accounting for factors
like power limitations, hardware imperfections, and channel uncertainties.

11
 Chapter 2. Background

• Performance Comparison: A comprehensive comparison of STAR-RIS

with existing technologies such as conventional RIS and relaying systems under
various scenarios is crucial for evaluating its advantages and identifying suit-
able application areas. This comparison should consider metrics like energy
efficiency, spectral efficiency, coverage, and cost.

2.4.3 Statistical Analysis Gaps in RIS-based CoMP-NOMA

Networks
While existing research has explored the integration of STAR-RIS into CoMP-
NOMA networks, there is a lack of a comprehensive statistical framework for ana-
lyzing and optimizing their performance. Key gaps include:

• Performance Analysis and Optimization: Existing studies often rely on

simplified assumptions and specific scenarios when analyzing outage probabil-
ity and rate coverage. More comprehensive analyses incorporating diverse user
distributions, varying channel conditions, and realistic RIS configurations are
needed. Additionally, developing statistically robust algorithms for optimal
user pairing, power allocation, and joint beamforming/phase shift design is
crucial.

• Resource Management and Interference Mitigation: Dynamically adapt-

ing RIS configuration to varying channel conditions and user demands requires
intelligent resource management strategies. Statistical methods can provide
valuable tools for predicting channel variations and optimizing RIS phase shifts
accordingly. Furthermore, statistical techniques can aid in developing effi-
cient interference cancellation and mitigation schemes tailored for RIS-assisted
CoMP-NOMA networks.

• Security and Privacy: The broadcast nature of RIS reflections introduces

security and privacy concerns. Statistical approaches can be leveraged to de-
sign secure transmission protocols and privacy-preserving mechanisms that
mitigate potential vulnerabilities in RIS-assisted CoMP-NOMA networks.

2.4.4 DRL in Aerial RIS Networks: Challenges and Oppor-

tunities
While DRL has shown promising results in optimizing aerial RIS networks, several
research gaps and challenges need to be addressed for practical real-world deploy-
ment:

12
 Chapter 2. Background

Expanding the Scope of Scenarios and Environments

• Dynamic Environments: Most existing research focuses on static environ-

ments, neglecting the dynamic nature of real-world scenarios with moving
objects and changing channel conditions. Investigating DRL techniques in
such dynamic environments is crucial for practical applications.

• Complex Network Topologies: Current studies primarily consider simple

network topologies with limited aerial platforms and RIS elements. Exploring
DRL algorithms in more complex networks with multiple aerial layers, diverse
user distributions, and heterogeneous platforms (e.g., drones and high-altitude
platforms) is essential for expanding the applicability of these techniques.

Enhancing Algorithm Efficiency and Scalability

• Computational Complexity: Many DRL algorithms suffer from high com-

putational complexity and long training times, hindering real-time implemen-
tation. Research into efficient DRL algorithms with reduced complexity and
faster convergence is needed to overcome this limitation.

• Scalability: The performance of DRL algorithms often degrades as network

size and complexity increase. Developing scalable DRL solutions capable of
efficiently handling large-scale aerial RIS networks is vital for practical deploy-
ment.

Joint Optimization and Hybrid Approaches

• Integration with Path Planning: Investigating the joint optimization of

aerial platform trajectories and RIS configurations using DRL has the potential
to significantly improve overall network performance.

• Hybrid DRL Approaches: Combining DRL with other optimization tech-

niques, such as evolutionary algorithms or convex optimization methods, could
leverage the strengths of each approach and lead to improved performance and
convergence speed.

Security and Privacy Considerations

• Adversarial Attacks: The vulnerability of DRL algorithms to adversarial

attacks is a growing concern. Research into robust DRL techniques that are
resilient to malicious attacks is necessary for secure and reliable operation of
aerial RIS networks.

13
 Chapter 2. Background

• Privacy Preservation: DRL algorithms may require access to sensitive user

data, raising privacy concerns. Investigating privacy-preserving DRL tech-
niques that protect user information while maintaining performance is critical
for ethical and responsible implementation of these technologies.

2.4.5 Energy Efficiency Assessment in Aerial RIS Networks:

A Holistic Approach
Accurately assessing and optimizing energy efficiency in aerial RIS networks requires
a holistic approach that considers various factors and challenges:

Developing Comprehensive Energy Consumption Models

• Holistic Network Perspective: Existing models often focus on individual

components, neglecting the interconnected nature of the entire network. Com-
prehensive models considering UAVs, RIS, communication links, and ground
infrastructure are needed for accurate energy efficiency assessments.

• Dynamic Energy Consumption: Models should account for real-time fac-

tors such as UAV trajectory optimization, varying channel conditions, and RIS
configuration changes to accurately capture the dynamic energy consumption
of aerial RIS networks.

Optimization Algorithms with Energy Efficiency Focus

• Balancing Performance and Energy Efficiency: Algorithms should be

designed to optimize network performance with energy efficiency as a primary
or secondary objective, effectively balancing these often conflicting goals.

• Joint Optimization: Exploring joint optimization of UAV trajectory, RIS

phase shifts, and communication resource allocation with an emphasis on en-
ergy efficiency holds significant potential for improving overall network sus-
tainability.

Addressing Practical Considerations and Implementation Challenges

• Hardware Limitations: Realistic models and optimization algorithms need

to account for the limitations of RIS elements, UAVs, and communication
systems to ensure practical feasibility and avoid overestimating potential gains.

• Environmental Factors: Environmental conditions, such as weather and

wind speed, can significantly impact energy consumption. These factors should

14
 Chapter 2. Background

be considered in energy efficiency assessments and optimization strategies.

• Cost Analysis: Evaluating the economic feasibility of aerial RIS networks

through cost-benefit analysis and comparison with terrestrial networks is cru-
cial for determining their viability and potential for large-scale deployment.

Standardization and Protocols

• Developing Standardized Protocols: As a relatively new technology, aerial

RIS networks lack standardized protocols for energy management and commu-
nication. Establishing such protocols is essential for ensuring interoperability
and facilitating large-scale deployment.

• Energy-efficient Communication Protocols: Existing communication pro-

tocols may not be optimized for energy efficiency in aerial RIS networks. Re-
search on adapting existing protocols or developing new energy-aware proto-
cols is necessary to minimize energy consumption while maintaining reliable
communication.

Security and Privacy in Energy-Efficient Systems

• Energy-efficient Security Mechanisms: Implementing security mecha-

nisms can increase energy consumption. Research is needed on designing
energy-efficient security solutions for aerial RIS networks that balance security
requirements with energy efficiency goals.

• Privacy Concerns: The dynamic nature of aerial RIS networks raises privacy
concerns. Studies should explore privacy-preserving mechanisms that protect
user information while maintaining energy efficiency and network performance.

2.5 Summary
This chapter has laid the foundation for exploring the potential of cutting-edge
technologies in addressing the ever-growing demands of modern wireless communi-
cation. The escalating need for higher data rates, improved spectral efficiency, and
expanded coverage has exposed the limitations of traditional approaches, necessi-
tating the investigation of novel solutions.

Non-Orthogonal Multiple Access (NOMA) was introduced as a promising candi-

date for enhancing spectral efficiency by enabling multiple users to share the same
resource block, offering advantages over conventional Orthogonal Multiple Access
(OMA) schemes. Additionally, Coordinated Multi-Point (CoMP) techniques were

15
 Chapter 2. Background

presented as a means to mitigate inter-cell interference and improve network perfor-

mance through coordinated transmission and reception among base stations. How-
ever, challenges related to channel state information acquisition, synchronization,
and signal processing complexity remain hurdles for the practical implementation of
both NOMA and CoMP.

Reconfigurable Intelligent Surfaces (RIS) emerged as a revolutionary technology ca-

pable of manipulating electromagnetic waves through software-controlled metasur-
faces, offering a new paradigm for shaping wireless environments. The limitations of
traditional RIS designs, particularly the “half-space” problem, were addressed with
the introduction of Simultaneously Transmitting and Reflecting RIS (STAR-RIS) as
a more versatile and powerful solution.

By establishing a firm understanding of the fundamental principles and limitations

of NOMA, CoMP, RIS, and STAR-RIS, this chapter has paved the way for further
exploration and analysis of their potential in shaping the future of wireless commu-
nication systems. Subsequent chapters will delve deeper into their applications and
evaluate their impact on the evolution of wireless networks, striving to overcome
existing challenges and unlock their full potential.

16
Chapter 3

Synergy of STAR-RIS, CoMP, and

NOMA: Paving the Path for 6G
Wireless Networks

The evolution towards sixth-generation (6G) wireless networks demands significant

advancements in spectral efficiency and coverage to accommodate the ever-growing
demand for data-driven applications and ubiquitous connectivity. Reconfigurable
Intelligent Surfaces (RIS) have emerged as a transformative technology with the po-
tential to revolutionize wireless communication by intelligently manipulating elec-
tromagnetic waves. RIS can dynamically shape the propagation environment to en-
hance signal strength and mitigate interference, thus, facilitating higher data rates
and improved spectral efficiency.

In parallel, the increasing deployment of small, low-power base stations within cel-
lular networks has led to challenges related to cross-tier interference and increased
energy consumption. Coordinated Multi-Point (CoMP) techniques offer a solution
by enabling base stations to coordinate transmissions through high-speed fronthaul
links, thereby mitigating interference and enhancing overall network performance.
Integrating CoMP with Non-Orthogonal Multiple Access (NOMA) further improves
spectral efficiency by allowing multiple users to share the same resource block.

This chapter explores the synergistic potential of combining STAR-RIS with CoMP-
NOMA networks to address the challenges of future wireless communication systems.
By leveraging the unique capabilities of each technology, we aim to pave the path
towards achieving the ambitious goals of 6G networks, providing enhanced coverage,
improved spectral efficiency, and a more sustainable and connected future.

17
 Chapter 3. Synergy of STAR-RIS, CoMP, and NOMA

Performance Analysis

3.1 System Model

As shown in Figure 3.1, we consider a multi-cell STAR-RIS assisted CoMP-NOMA
network, in which each base station (BS) serves a NOMA pair consisting of its
corresponding center and edge users. Consequently, the edge user is part of two
NOMA pairs, each served by a different BS.

Uf
STA
R-R
IS
Cell
edge

BS1 U1,c
NOMA Pair
of BS1

Converage NOMA Pair

boundary of BS1 of BS2

U2,c
Con
nLoS links boun verage
dary
of BS
Interfering links
2 BS2

Figure 3.1: An illustration of STAR-RIS-aided coordinated NOMA cluster.

We define the index set I = {1, 2} for the two BSs, whereas C = {1, 2, . . . , C}
is defined for the cell-center users, and F = {1, 2, . . . , F } for the cell-edge users,
respectively. Additionally, let U = C ∪ F, denote the set of all system users. For
the sake of simplicity, we assume C = 2, where each center user Uci is served by its
corresponding BSi , and F = 1, with C and F representing the cardinality of C and
F, respectively.

The BSs employ power-domain NOMA techniques to communicate with the users.
Specifically, ∀i ∈ I, c ∈ C, and f ∈ F, BSi forms the NOMA pair (Uci , Uf ), where
Uci is the cell-center user of BSi . Consequently, the Uf is part of two NOMA
pairs, each cluster served by a different BS. To mitigate the strong ICI experienced
by Uf , CoMP is adopted between the two BSs. In addition, it is assumed that
the BSs are connected to a central processing unit (CPU) via high-speed fronthaul
links, facilitating seamless information sharing and coordinated transmissions among
them.

In this work, perfect CSI is assumed to be available at the BSs. While this is a

18
 Chapter 3. Synergy of STAR-RIS, CoMP, and NOMA

challenging assumption in practice, recent advances in channel estimation techniques

for RIS-enabled wireless networks have shown that it is possible to achieve accurate
CSI [9, 16, 17] with a reasonable amount of overhead.

3.1.1 Channel Model

For each communication link in the system, we take into account both large-scale
fading and small-scale fading effects. Due to the relatively large propagation dis-
tances and the scattering effect of the links between Bi and Uu , ∀i ∈ I and u ∈ U,
the channels are assumed to follow Rayleigh fading, expressed as:

ρo
r
hi,u = vi,u , (3.1)
P L(di,u )

where vi,u is a complex Gaussian random variable that follows a Rayleigh distribution
with zero mean and unit variance, ρo is the reference path-loss at a distance of 1

α
m, P L(di,u ) is the large scale path-loss, modeled as P L(di,u ) = di,u i→u , where
di,u is the distance and αi→u is the path-loss exponent between the BSi and Uu ,
respectively.

On the contrary, the link between the STAR-RIS, hereafter represented by R, and
BSi is assumed to exhibit a dominant line-of-sight (LoS) path [18]. Therefore, these
links are subject to the Rician fading, where their channel coefficients are expressed
as:  s 
ρo κi,R 1
r r
hi,R =  vˆi,R + vi,R  , (3.2)
P L(di,R ) κi,R + 1 κi,R + 1

where di,R is the distance between the BSi and R, κi,R represents the Rician factor,
vˆi,R represents the deterministic LoS components, and vi,R denotes the complex
Gaussian random variables, each following a Rayleigh distribution with zero mean
and unit variance, thus representing the NLoS components. The links between R
and Uu , ∀u ∈ U, are also modeled similarly.

3.1.2 STAR-RIS Configuration

The energy splitting (ES) model of the STAR-RIS array can be mathematically
characterized by the following respective transmission- and reflection-coefficient ma-
trices [19]:

t t t
p
Θr = β r diag(ejθ1 , ejθ2 , . . . , ejθK ), (3.3)
jθ1r jθ2r r
p jθK
Θt = β t diag(e ,e ,...,e ), (3.4)

19
 Chapter 3. Synergy of STAR-RIS, CoMP, and NOMA

where β t , β r ∈ [0, 1] and θkt , θkr ∈ [0, 2π), ∀k ∈ K ≜ {1, 2, . . . , K}. The phase shifts
for transmission and reflection (i.e., θkt and θkr ) can generally be chosen indepen-
dently of each other [20]. However, the amplitude adjustments for transmission and
reflection are coupled by the law of conservation of energy. Assuming the STAR-RIS
does not impose any power loss, the relation between the amplitude coefficients (i.e.,
β t and β r ) is expressed as β t + β r = 1. To reduce the signaling overhead between
the STAR-RIS and the BSs, all elements are adjusted to have the same transmission
and reflection coefficients.

3.1.3 Rate Analysis

To analyze the rates achieved for users in the system model shown in Figure 3.1,
we first present the signal model. Specifically, ∀i ∈ I, c ∈ C, and f ∈ F, let
the tuple (Uc1 , Uc2 , Uf ) represent the coordinated NOMA cluster, and let P1 and
P2 denote the transmit powers of BS1 and BS2 , respectively. This signal model
serves as the foundation for evaluating the achieved rates and optimizing the system
performance in the considered cluster. Additionally, let x1,c , x2,c , and xf represent
the message signal intended for Uc1 , Uc2 , and Uf , respectively. Each BSi broadcasts
a superimposed signal of the messages intended for users within its coverage region,
Uci and Uf , and expressed as [21]:
p p
xi = ζi,c Pi xi,c + ζi,f Pi xf , (3.5)

where ζi,c and ζi,f are the power allocation (PA) factors assigned by BSi to users
Uci sand Uf , respectively. It is important to note that Uci experiences stronger
channel conditions compared to Uf , making it the dominant NOMA user in the pair
(Uci , Uf ) formed by BSi . Following the principle of NOMA, Uci should be capable
of detecting and decoding the message intended for Uf . This principle also implies
that ζi,c < 0.5, or 0.5 < ζi,f < 1 [22, 23].

For brevity, we only define the rate achieved by Uc1 from the set of cell-center users
C, as the same steps could be extended to define the rate of Uci , ∀i ∈ I and c ∈ C.
The received signal at Uc1 can be written as:

yc1 = h1,c x1 + h2,c′ x2 + No , (3.6)

where No is an additive white Gaussian noise (AWGN), i.e., No ∼ CN (0, σ 2 ).

Further, h2,c′ is the channel corresponding to the link between BS2 and Uc1 , which is
the cell-center user of BS1 , and represents the ICI experienced by Uc1 . By utilizing
successive interference cancellation (SIC) techniques, Uc1 first decodes the message

20
 Chapter 3. Synergy of STAR-RIS, CoMP, and NOMA

signal of Uf (i.e., xf ) and then removes it from yc1 to decode its own message (i.e.,
x1,c ). Based on this approach, the signal-to-interference-and-noise ratio (SINR) and
the corresponding achievable rate at Uc1 for decoding the message of Uf can be
expressed as:
2
ζ1,f P1 H1,c
γ1,c→f = 2 2 , (3.7)
ζc1 P1 H1,c + P2 h2,c′ + σ2


R1,c→f = log2 1 + γ1,c→f , (3.8)

where H1,c = h1,c + hH

R,c Θr h1,R represents the combined channel from BS1 to Uc1 .
Furthermore, the SINR and the corresponding achievable rate of Uc1 for decoding
its own message can be expressed as:
2
P1 H1,c
γc1 = ζc1 2 , (3.9)
P2 h2,c′ + σ2
Rc1 = log2 (1 + γc1 ) . (3.10)

On the contrary, Uf , belonging to two NOMA pairs, receives its signal through the
broadcasts from each BSi , ∀i ∈ I. Thus, the received signal at Uf can be expressed
as:
yf = H1,f x1 + H2,f x2 + N0 , (3.11)

where H1,f and H2,f represent the combined channels from BS1 to Uf and from
BS2 to Uf , and can be expressed as H1,f = h1,f + hH R,f Θt h1,R and H2,f = h2,f +
H
hR,f Θt h2,R , respectively. Given that non-coherent JT-CoMP is taken into consid-
eration, the SINR and the corresponding achievable rate at Uf can be expressed
as [24, 25]:

2 2
ζ1,f P1 H1,f + ζ2,f P2 H2,f
γf = 2 2 , (3.12)
ζc1 P1 H1,f + ζ2,c P2 H2,f + σ2


Rf = log2 1 + γf . (3.13)

3.1.4 Outage Probability Analysis

To further investigate the efficacy of strategically placing the STAR-RIS in im-
proving the system performance, we analyze the outage probability experienced by
cellular users. Following the principles of NOMA, ∀i ∈ I, c ∈ C, and f ∈ F, if Ui,c
cannot decode xf , or is capable of decoding xf but not xi,c , an outage occurs, the

21
 Chapter 3. Synergy of STAR-RIS, CoMP, and NOMA

probability of which is expressed as [26]:

Pi,c = Pr (γi,c→f < γthf ) + Pr (γi,c→f > γthf , γc < γthc ), (3.14)

where γthf and γthc represent the outage thresholds for Uf and Ui,c , respectively.
Similarly, in regard to Uf , an outage occurs when it fails to decode xf , and the
corresponding outage probability is expressed as:

Pf = Pr (γf < γthf ). (3.15)

Analytical Analysis

3.2 End-to-end SINR Statistics

3.2.1 Effective Channel Characterization
√ P
Let Zi,u = |Hi,u |2 = (|hi,u | + βn K 2
k=1 |hR,u ||hi,R |) , where n ∈ {t, r} represents the
transmission and reflection regions of STAR-RIS, respectively. The distribution of
Zi,u is derived in the following lemma.

Lemma 1. Assuming a large K, and by applying MoM, the distribution of Zi,u

is approximated as a Gamma distribution, Zi,u ∼ Γ kZi,u , θZi,u , with the following
probability density function (PDF).
x
−θ
xkZi,u −1 e Zi,u

fZi,u (x) = kZ
, x > 0, (3.16)

θZi,ui,u Γ kZi,u

(2)
µ2Z µZ −µ2Z
i,u i,u i,u
where kZi,u = (2) and θZi,u = µZi,u
are the shape and scale parameters of
µZ −µ2Z
i,u i,u
√
2K βΩiu ΩiR ΩRu Γm ( 12 , 12 , 21 )
the Gamma distribution, with µZi,u = √
miR mRu miu
+βK 2 ΩiR ΩRu +Ωiu and
√ 3/2 √
(2) 4β 3/2 K 3 Ωiu (ΩiR ΩRu )3/2 Γm ( 21 , 32 , 32 ) 2 4KΩiu βΩiR ΩRu Γm ( 32 , 12 , 12 )
µZi,u = √
miu (miR mRu )3/2
+6βK Ω Ω Ω
iR iu Ru + √ 3/2 +
miR mRu miu
β 2 K 4 Ω2iR (miR +1)(mRu +1)Ω2Ru (m +1)Ω2
miR mRu
+ iumiu iu as the first and second moments of Zi,u , respec-
tively, and Γm (a, b, c) = Γ(miuΓ(m + a)Γ(miR + b)Γ(mRu + c)
iu )Γ(miR )Γ(mRu )
.
√ P
Proof. For brevity, let the combined channel be Gi,u = βn K k=1 |hR,u ||hi,R |. By
applying CLT, and noting that it is a scaled double-Nakagami random variable (RV),

22
 Chapter 3. Synergy of STAR-RIS, CoMP, and NOMA

the distribution of Gi,u can be approximated as a Gamma distribution, i.e.,

 
(2)
µ2Gi,u µGi,u − µ2Gi,u
Gi,u ∼ Γ  (2)
, ,
µGi,u − µ2Gi,u µGi,u

(2)
where µGi,u and µGi,u are the first and second moments of Gi,u , respectively, with
the p-th moment of Gi,u given by [27]
√
(K βn )p (ΩiR ΩRu )p/2 Γ mRu + p2 Γ miR + p2



(p)
µGi,u = , (3.17)
(mRu miR )p/2 Γ (miR ) Γ (mRu )

and as |hi,u | ∼ N akagami(mi,u , Ωi,u ), the p-th moments are known to be given by
(p) Γ(miu + p2 )Ωiu p/2
µ|hi,u | = Γ(miu )m p/2 . Since |hi,u | and Gi,u are independent, the p-th moment of
iu

|Hi,u | can be obtained via the moments of its summands, i.e., |hi,u | and Gi,u , by
applying the binomial theorem. Hence, the p-th moment of |Hi,u | is given by
p  
(p)
X p (q) (p−q)
µ|Hi,u | = µ|hi,u | µGi,u . (3.18)
q=0
q

(2) (2) (4)

Knowing that only the first two moments of Zi,u , µZi,u = µ|Hi,u | and µZi,u = µ|Hi,u | ,
are necessary to approximate its distribution as a Gamma distribution, the first two
moments of Zi,u are, therefore, given by

(2) (2)
µZi,u = µ|hi,u | + 2µ|hi,u | µGi,u + µGi,u , (3.19)
(2) (4) (3) (2) (2) (3) (4)
µZi,u = µ|hi,u | + 4µ|hi,u | µGi,u + 6µ|hi,u | µGi,u + 4µ|hi,u | µGi,u + µGi,u . (3.20)

The final expression of moments can thus be obtained through the means of substi-
tution in (3.19) and (3.20).

Using the moment matching-based Gamma approximation approach, the distribu-

tion of the sum of a Gamma RV and the square of a Nakagami-m RV with different
shape and scale parameters, further weighted by different constant terms, i.e., path
loss and power allocation factors, is derived in the following lemma.
(a,b)
Lemma 2. Let Bi,u = aZi,u + b|hi′ ,u |2 , where a, b ∈ R+ , i′ ∈ I \ {i}, then the distri-
(a,b) (a,b)

bution of Bi,u is approximated as a Gamma distribution, Bi,u ∼ Γ kB(a,b) , θB(a,b) ,
i,u i,u

23
 Chapter 3. Synergy of STAR-RIS, CoMP, and NOMA

with the following PDF

x
−θ
k (a,b) −1 (a,b)
B B
x i,u e i,u
fB(a,b) (x) = k (a,b)  , x > 0, (3.21)
i,u B

i,u
θ (a,b) Γ kB(a,b)
Bi,u i,u

(2)
µ2 (a,b) µ (a,b)
−µ2 (a,b)
B B B
i,u i,u i,u
where kB(a,b) = (2) and θB(a,b) = µ
, with µB(a,b) = aµZi,u + bΩi,u
i,u µ (a,b)
−µ2 (a,b) i,u B
(a,b) i,u
B B i,u
i,u i,u
(2) (2)
and µ (a,b) = a2 µZi,u + 2abµZi,u Ωi,u + b2 Ω2i,u (1 + m1i,u ) as the first and second moments
Bi,u
(a,b)
of Bi,u , respectively.

Proof. As |hi,u | ∼ N akagami(mi,u , Ωi,u ), the square of |hi,u | is known to be Gamma

Ω
distributed, i.e., |hi,u |2 ∼ Γ ki,u , θi,u , where ki,u = mi,u and θi,u = mi,u
i,u
are the shape
and scale parameters of the Gamma distribution, respectively. Further, by using the
scaling property of Gamma distribution, i.e., X ∼ Γ (k, θ) =⇒ aX ∼ Γ (k, aθ), the
(2) (2)
first and second moments of aZi,u are given by µaZi,u = aµZi,u and µaZi,u = a2 µZi,u ,
(a,b)
respectively. Finally, the first and second moments of Bi,u can be obtained by
applying the binomial theorem in (3.18) and substituting the moments of aZi,u and
b|hi,u |2 .

3.2.2 Probability Density Functions of SINRs

Lemma 3. The PDF of the SINR at Uci to decode the signal of Uf , i.e., γi,c→f , is
given by  αi,c  νi,c→f
xθWi,c,f xθWi,c,f
θWi,c,f ρζi,f θZ ρζi,f θZi,c
+1
i,c
fγi,c→f (x) = , (3.22)
ρζi,f θZi,c B(kZi,c , kWi,c,f )
for x > 0, where αi,c = kZi,c − 1, νi,c→f = −(kZi,c + kWi,c,f ), B (· , ·) is the Euler
(2)
µ2W µW −µ2W
i,c,f i,c,f i,c,f
Beta function, kWi,c,f = (2) , and θWi,c,f = µWi,c,f
, with µWi,c,f =
µW −µ2W
i,c,f i,c,f
(2) (2)
µ (ρζ , ρ) + 1 and µWi,c,f = µ + 2µ (ρζ , ρ) + 1.
Bi,c i,c (ρζ
Bi,c i,c
, ρ)
Bi,c i,c

ρζi,f Zi,c
Proof. The expression in (3.7) can be rewritten as γi,c→f = Wi,c,f
, where Wi,c,f =
(ρζ , ρ) (ρζ , ρ)
Bi,c i,c+ 1, with Zi,c and Bi,c i,c
both being Gamma RVs based on the statistics
derived earlier. Then, the distribution of Wi,c,f can also be approximated by an


equivalent Gamma RV, i.e., Wi,c,f ∼ Γ kWi,c,f , θWi,c,f . As Zi,c and Wi,c,f are two
independent Gamma RVs, the ratio of two Gamma RVs is known to follow a Beta
ρζ Z
prime distribution, i.e., Wi,fi,c,fi,c ∼ β ′ kZi,c , kWi,c,f , 1, ρζi,f θZi,c /θWi,c,f , corresponding

to the PDF in (3.23).

24
 Chapter 3. Synergy of STAR-RIS, CoMP, and NOMA

Corollary 1. As γi,c→f and γi,c are closely related, differing only by weighting con-
stants, the PDF of γi,c is given by
 αi,c  νi,c
xθWi,c xθWi,c
θWi,c ρζi,c θZi,c ρζi,c θZi,c
+1
fγi,c (x) = , x > 0, (3.23)
ρζi,c θZi,c B(kZi,c , kWi,c )

(2)
µ2W µW −µ2W
i,c i,c i,c
where νi,c = −(kZi,c +kWi,c ), kWi,c = (2) , and θWi,c = µWi,c
, with µWi,c =
µW −µ2W
i,c i,c
(2) (2)
µB(0, ρ) + 1 and µWi,c = µ (0, ρ) + 2µB(0, ρ) + 1 as the first and second moment of
i,c Bi,c i,c
(0, ρ)
Wi,c = Bi,c + 1, respectively.

Lemma 4. The PDF of the SINR at Uf , i.e., γf , is given by

 αf  νf
xθWf xθWf
θWf θVf θVf
+1
fγf (x) = , x>0 (3.24)
θVf B(kVf , kWf )

µ2V
f
where B (· , ·) is the Euler Beta function, νf = −(kVf + kWf ), kVf = (2) , θVf =
µV −µ2V
f f
(2) (2)
µV −µ2V µ2W µW −µ2W
f f f f f
µVf
, kWf = (2) , and θWf = µ Wf
, with µVf = ρ(ζi,f µZi,f + ζi′ ,f µZi′ ,f ),
µW −µ2W
f f
(2) (2) (2)
µVf = ρ2 (ζi,f
2
µZi,f + 2ζi,f ζi′ ,f µZi,f µZi′ ,f + ζi2′ ,f µZi′ ,f ), µWf = ρ(ζi,c µZi,f + ζi′ ,c µZi′ ,f ) + 1,
(2) 2 2
and µWf = 2µZi,f (ρ2 ζi,c ζi′ ,c µZi′ ,f + ρζi′ ,c + ρζi,c ) + ρ2 ζi2′ ,c µ2Zi′ ,f + ρ2 ζi,c µZi,f + 1.

V
Proof. The expression in (3.12) can be rewritten as γf = Wff , where Vf = ρζi,f Zi,f +
ρζi′ ,f Zi′ ,f and Wf = ρζi,c Zi,f + ρζi′ ,c Zi′ ,f + 1, with Zi,f and Zi′ ,f both being Gamma
RVs. The rest of the proof is similar to that of Figure 3.

In Figure 3.2, we compare the analytical and simulated PDFs and CDFs of SINRs
of network users. The close alignment between the analytical approximations and
Monte Carlo (MC) simulations affirms the accuracy of the derived expressions. Ad-
ditionally, we conduct the Kolmogorov-Smirnov (KS) goodness-of-fit test, however,
the details are omitted due to space constraints.

3.3 Ergodic Rate (ER)

The ER of the edge user is defined as
Z ∞
Rf = log2 (1 + x)fγf (x)dx, (3.25)
0

25
 Chapter 3. Synergy of STAR-RIS, CoMP, and NOMA

(a) The PDF of γi,c (b) The CDF of γi,c

(c) The PDF of γi,c→f (d) The CDF of γi,c→f

(e) The PDF of γf (f) The CDF of γf

Figure 3.2: The PDFs and CDFs of the SINRs at the center and edge user, with K = 34
elements, mi,u = mi′ ,u = 1, and mi,R = mR,u = 2, ∀i ∈ I, i′ ∈ I \ {i}, ∀u ∈ U.

where fγf (x) is the PDF of γf in (3.24). Noting that the PDF of γf is a Beta prime
distribution, the ER of the edge user is derived in the following theorem.

Theorem 1. In the proposed system model, the ER of the edge user is given by
 
1  θWf 0, 1 − kWf , 1 
Rf = G3,2
3,3  , (3.26)
ln(2)Λf θVf 0, 0, kVf

 a1 , . . . , ap 
where Gm,n
p,q z is the Meijer G-function, Λf = B(kVf , kWf ) Γ(κf ) and
b1 , . . . , b q
κf = kVf + kWf .

Proof. Substituting the PDF of γf in (3.24) into (3.25), we obtain

!αf !νf
∞
θWf xθWf xθWf
Z
Rf = log2 (1 + x) × 1+ dx. (3.27)
θVf B(kVf , kWf ) 0 θ Vf θVf

From [28, Eq. (11)] and [28, Eq. (10)], the logarithmic and power functions can be

26
 Chapter 3. Synergy of STAR-RIS, CoMP, and NOMA

 1, 1 .
expressed in terms of a Meijer G-function, i.e., log2 (1 + z) = G1,2
2,2 z ln(2)
1, 0
 1 − ν .
and (1 + z)ν = G1,1
1,1 z Γ(ν), respectively. Furthermore, using the analyt-
0
ical continuation of the Meijer G-function, the integral in (3.27) can be rewritten
as
 
Z ∞
θWf 1,2  1, 1 
Rf = G2,2 x
ln(2)θVf B(kVf , kWf )Γ(κf ) 0

1, 0
 
 xθWf −kWf 
× G1,1
1,1   dx. (3.28)
θVf kVf − 1

Finally, using the integral representation of the Meijer G-function, we obtain (3.26).

To gain further insight, we express the high-SNR approximation for the ER of the
edge user as  
1 3,2  0, 1 − kṼf , 1 
R∞
f ≈ G3,3 θf˜ , (3.29)
ln(2)Λf˜ 0, 0, kVf

θṼ
where Λf˜ = B(kVf , kṼf ) Γ(κf˜), κf˜ = kVf + kṼf , and θf˜ = θVf . The approximate
f
parameters, denoted as kṼf and θṼf , can be computed using the first and second mo-
(2) 2 (2)
ment, that is, µṼf = ρ(ζi,c µZi,f +ζi′ ,c µZi′ ,f ) and µṼ = ρ2 (ζi,c µZi,f +2ζi,c ζi′ ,c µZi,f µZi′ ,f +
f
(2)
ζi2′ ,c µZi′ ,f ), respectively.

Likewise, the ER of the center users is defined as

Z ∞
Ri,c = log2 (1 + x)fγi,c (x)dx, (3.30)
0

where fγi,c (x) is the PDF of γi,c in (3.23). The ER of the center users can then be
derived as follows.

Theorem 2. The ER for the center users is given by

 
1 3,2  θWi,c 0, 1 − kWi,c , 1 
Ri,c = G3,3 , (3.31)
ln(2)Λi,c ρζi,c θZi,c

0, 0, kZi,c

27
 Chapter 3. Synergy of STAR-RIS, CoMP, and NOMA

where Λi,c = B(kZi,c , kWi,c ) Γ(κi,c ) and κi,c = kZi,c + kWi,c .

Proof. The proof closely follows that of Figure 1.

Similar insights can be derived for the ER of the center users as of the edge user,
however, the details are omitted due to space constraints.

3.4 Outage Probability (OP)

The OP for the edge user is defined as the probability that the instantaneous SINR
at the edge user to decode its own message is below a certain threshold, and can be
expressed as Pf = Pr (γf < γthf ), where γthf = 2Rthf − 1 is the target SINR with
Rthf being the target rate for edge users. As γf ∼ β ′ kVf , kWf , 1, θVf /θWf , and the

CDF of a Beta prime distribution is known to be an incomplete Beta function, the

OP for the edge user can be expressed as

Γ(kVf + kWf )
Pf = Bψ (kV , kWf ), (3.32)
Γ(kVf )Γ(kWf ) f f

λth θW
where ψf = θV +λfth θfW , and Bz (· , ·) is the incomplete Beta function. As the thresh-
f f f
old (λthf ) tends towards infinity, the incomplete Beta function in (3.32) converges
to the Euler Beta function, i.e., Bψf (kVf , kWf ) → B(kVf , kWf ).

Similarly, with regards to center users, the OP is defined as the probability that
the instantaneous SINR for decoding the user’s own message or the message of the
edge user falls below a certain threshold. Mathematically, it can be expressed as
Pi,c ≈ Pr (γi,c→f < γthf ) + Pr (γi,c→f > γthf , γc < γthc ), where γthc = 2Rthc − 1 is the
target SINR with Rthc being the target rate for center users, and the approximate
symbol is due to the fact that the detection sequence is not of fully independent
(1)
events. The first term in the sum expression, denoted here onwards as Pi,c , takes
on the same form as that of OP for the edge user in (3.32), except for parameters,
i.e., kVf → kZi,c , kWf → kWi,c,f , θVf → ρζi,f θZi,c , and θWf → θWi,c,f . Furthermore,
(2)
let Pi,c = Pr (γi,c→f > γthf , γc < γthc ), then, the second term in the sum expression
becomes
(2)
Pi,c = Iψi,c→f (kWi,c,f , kZi,c ) Iψi,c (kZi,c , kWi,c ), (3.33)
ρζi,f θZ λth θW
where ψi,c→f = ρζi,f θZ +λthi,cθW , ψi,c = ρζi,c θZ c+λthi,c θW , and Iz (· , ·) is the reg-
i,c f i,c,f i,c c i,c
ularized incomplete Beta function. The OP for the center user is then given by
(1) (2)
Pi,c ≈ Pi,c + Pi,c . Further improvement in approximation can be made by mak-
ing use of the fact that outage performance cannot be better than that of the

28
 Chapter 3. Synergy of STAR-RIS, CoMP, and NOMA

interference-free noise-only case. Therefore, the final expression of the OP can be

expressed as the maximum of the two cases, i.e., Pi,c ≈ max{Pi,c , Pr (γc < γthc )}.

Numerical Results

3.5 Simulation Setup

We consider an outdoor environment where the transmission bandwidth of the net-
work is set to B = 1 MHz, and the power of AWGN is set to σ 2 = −174+10 log10 (B)
(dBm) with a noise figure NF of 12 dB. For simplicity, we assume that the transmit
powers of both BS1 and BS2 are identical, expressed as P1 = P2 = Pt . Moreover,
the PA factors for U1,c , U2,c , and Uf are fixed to ζ1,c = ζ2,c = 0.3 and ζf = 0.7,
respectively.

In the three-dimensional Cartesian coordinate system, the locations of BS1 and BS2 ,
each with a coverage radius of 60 m, are set to (-50m, 0m, 25m) and (50m, 0m, 25m)
respectively. The STAR-RIS is strategically placed at the intersection of the two
cells, near Uf , specifically, at the coordinates (0m, 25m, 5m). Additionally, the
cellular users U1,c , U2,c , and Uf are positioned at (-40m, 18m, 1m), (30m, 22m, 1m),
and (0m, 35m, 1m), respectively. Some specific parameters used for simulation are
outlined in Table 4.1.

3.6 Improvements in Ergodic Rate

The impact of the number of RIS elements on the ergodic rate of the network is
shown in Figure 3.3. We observe that the ergodic rate increases with the number
of RIS elements as the RIS elements amplify the channel links. Again, the STAR-

Table 3.1: Simulation Parameters

Parameter Value
Path-loss exponent of BSi -(Uci , RIS) links αi→c = 3
Path-loss exponent of BSi -Uf link αi→f = 3.5
Path-loss exponent BSi -RIS links αi→R = 3
Path-loss exponent of RIS-Uci links αR→c = 2.7
Path-loss exponent of RIS-Uf link αR→f = 2.3
Path-loss exponent of Interfering links αi→c′ = 4
Rician factor of RIS-Uci links κR→c = 3 dB
Rician factor of RIS-Uf link κR→f = 4 dB

29
 Chapter 3. Synergy of STAR-RIS, CoMP, and NOMA

14
K = 120 Elements
K = 40 Elements
12 Without RIS
Non-CoMP, without RIS

Ergodic rate (bits/s/Hz)

10

0
-25 -20 -15 -10 -5
Transmit power per BS, Pt (dBm)

Figure 3.3: The PDFs and CDFs of the SINRs at the center and edge user, with K = 34
elements, mi,u = mi′ ,u = 1, and mi,R = mR,u = 2, ∀i ∈ I, i′ ∈ I \ {i}, ∀u ∈ U.

RIS assisted CoMP-NOMA network outperforms other networks, due to substantial

diversity gains at Uf .

3.7 Impact of RIS on Outage Probability

In Figure 3.4, we assess the OP for all users under various transmit power levels (Pt )
and system configurations, with fixed thresholds λthf = λthc = 0 dB. The STAR-RIS
assisted CoMP-NOMA network demonstrates significant OP enhancements for Uf ,
attributed to the formation of vLOS paths. Uc1 and Uc2 show marginal improve-
ments due to their predominant reliance on LOS paths from their corresponding
BS. Notably, in the absence of CoMP, Uf contends with elevated ICI and consis-

100

10!1
Outage probability

Analytical, U1;c , K = 34 Elements

Analytical, U2;c , K = 34 Elements
Analytical, Uf , K = 34 Elements
10!2 Analytical, Uf , without RIS
Analytical, Uf , Non-CoMP, without RIS
Simulation, U1;c , K = 34 Elements
Simulation, U2;c , K = 34 Elements
Simulation, Uf , K = 34 Elements
Simulation, Uf , without RIS
Simulation, Uf , Non-CoMP, without RIS
10!3
-40 -35 -30 -25 -20 -15 -10 -5 0
Transmit power per BS, Pt (dBm)

Figure 3.4: Outage probability of network users versus Pt for equal amplitude coefficients
(β t = β r ), and element assignments (K1R = K2R ), when K > 0.

30
 Chapter 3. Synergy of STAR-RIS, CoMP, and NOMA

tently high outage probabilities across all power levels. Moreover, we attribute the
convergence of outage probability of Uf towards the asymptotes to the problem of
saturation inherent in NOMA.

3.8 Exhaustive Search for Optimality Regions

Finally, in Figure 3.5, we demonstrate the effect of varying the RIS element as-
signments (K1R , K2R ) to BS1 and BS2 , respectively, and the amplitude adjustments
(βt , βr ) in an exhaustive fashion. Notably, the ergodic rate peaks when βt > βr
as a result of close proximity of the STAR-RIS to Uf , located within the transmis-
sion region of the RIS, thereby defining the optimal configuration for the network.
Investigating optimization techniques for STAR-RIS resources can provide further
insights to enhance spectral efficiency.
1/0
Ergodic rate (bits/s/Hz)
0.9/0.1
Amplitude adjustments ratio -t =-r

0.8/0.2 9.5
9.54

9.56
9.54
1

9.51
9.5

0.7/0.3
9.453
9

9.49
9.4

9.45
0.6/0.4
9.4

7
46
9.
67

0.5/0.5 9.438
9.4
0.4/0.6
9.412
0.3/0.7 9.35
9.39
0.2/0.8 9.36
9.3
0.1/0.9 9.32
9.26
0/1
0/72 9/63 18/54 27/45 36/36 45/27 54/18 63/9 72/0
Element splitting ratio K1A / K2A

Figure 3.5: Ergodic rate for varying RIS element assignments (K1R , K2R ) and amplitude
adjustments (βt , βr ), with Pt = −10 dBm.

31
Chapter 4

Energy Efficient Design for

CoMP-NOMA Networks

The convergence of CoMP and NOMA techniques has emerged as a promising so-
lution to enhance both coverage and capacity within cellular networks. Recent
research has further explored the integration of Reconfigurable Intelligent Surfaces
(RIS) into CoMP-NOMA systems, demonstrating the potential for significant per-
formance gains. By strategically deploying RIS, particularly at the cell edge, signal
quality for edge users can be dramatically improved, leading to enhanced network
coverage and capacity. However, despite these promising advancements, a crucial
question remains: how do varying RIS configurations, the number of cooperating
base stations (BSs), and the number of RIS elements influence the overall energy
efficiency of the network? This chapter delves into this critical aspect, investigating
the energy efficiency and passive beamforming (PBF) design within RIS-assisted
CoMP-NOMA networks. We propose and analyze two distinct RIS configurations:
Enhancement-only PBF (EO) and Enhancement & Cancellation PBF (EC). The EO
configuration focuses on optimizing RIS phases to solely enhance the desired signal
quality for edge users, while the EC configuration aims to optimize RIS phases for
both signal enhancement and interference suppression.

Through a comprehensive analysis, we explore the impact of these configurations,

along with the number of cooperating BSs and RIS elements, on the network’s en-
ergy efficiency. Additionally, we formulate a PBF design problem with the objective
of maximizing energy efficiency through optimized RIS phase shifts. Our findings
demonstrate that the integration of RIS into CoMP-NOMA networks offers substan-
tial improvements in energy efficiency and overall network performance, highlighting
its potential for the future of wireless communication systems.

32
 Chapter 4. Energy Efficient Design for CoMP-NOMA Networks

Towards Generalizability

4.1 System Model

We consider a downlink transmission scenario in an RIS-assisted multi-cell CoMP-
NOMA network, as illustrated in Figure 4.1. The network is comprised of I cells,
each modeled as a disk with radius Ri and served by a BS located at its center,
denoted by BSi , where i ∈ I ≜ {1, 2, . . . , I}. Each single-antenna BS utilizes two-
user NOMA to serve user clusters, also equipped with single antennas, within its
coverage area.∗

Figure 4.1: An illustration of the RIS-assisted multi-cell CoMP-NOMA network.

We define two user classes based on their location: the cell-center users and the
edge users. The cell-center users reside within the disk of their associated cell, while
the edge users fall outside it. Let C (i) ≜ {1, 2, . . . , Ci } represent the set of indices
for cell-center users associated with BSi , and F ≜ {1, 2, . . . , F } represent the set of
indices for the shared edge users cooperatively served by multiple BSs. We denote
cell-center users as Uic (superscript i indicates the serving BS) and edge users as Uef
(superscript e represents the edge class), with c ∈ C (i) and f ∈ F. Furthermore,
U = i∈I C (i) ∪ F represents the index set for all users in the network. For ease of
S

exposition and without loss of generality, we consider a single edge user and a single
cell-center user per cell in this work, i.e., Ci = F = 1, ∀i ∈ I.

For coordinated operation, the BSs are assumed to be interconnected via a high-
∗
Due to processing complexity, latency of SIC at the receivers, and the practical limitations
resulting in SIC error propagation, two-user NOMA pairs are considered in this work. Moreover,
two-user configurations are of practical interest and have been standardized in 3GPP Release
15 [29].

33
 Chapter 4. Energy Efficient Design for CoMP-NOMA Networks

speed backhaul network to a central processing unit (CPU). BSs participating in

CoMP are referred to as cooperative BSs and are denoted by the set J ≜ {1, 2, . . . , J},
where J ≤ I. To further improve the signal quality for edge users, each BSi is
equipped with an RIS, denoted Ri , strategically placed at the cell edge.

4.1.1 Channel Model

We consider both large-scale and small-scale fading effects for each communication
link in the system. Due to large propagation distances and the presence of numerous
scatterers, the direct links, i.e., the channels between the BSs and the users, are
modeled as Rayleigh fading channels. We denote the channel between BSi and user
Unu as hni,u , where n ∈ {I, e} for cell-center or edge user, and u ∈ U. Mathematically,
this can be expressed as
ρo
r
n
hi,u = vn , (4.1)
P L(dni,u ) i,u
where ρo is the reference path loss at 1 m, P L(dni,u ) is the large-scale path loss
modeled as P L(dni,u ) = (dni,u )−αn , such that αn is the path loss exponent, dni,u is the
distance between BSi and Unu , and vi,u n
∈ C1×1 is the small-scale Rayleigh fading
coefficient with zero mean and unit variance.

In contrast to the Rayleigh fading experienced on direct links, the channels between
BSs and RIS are modeled as Rician fading channels due to the presence of a dominant
line-of-sight (LoS) component. The channel between BSi and Ri is denoted hi,Ri and
can be expressed as
r r !
ρo κ 1
r
hi,Ri = gLoS + gNLoS , (4.2)
P L(di,Ri ) 1 + κ i,Ri 1 + κ i,Ri

LoS
where κ is the Rician factor, gi,Ri
∈ CK×1 is the LoS channel vector given by
h iT
LoS j(k−1)π sin(ωi ) j(K−1)π sin(ωi )
gi,Ri
= 1, . . . , e , . . . , e ,

where k ∈ {1, 2, . . . , K} indexes elements of Ri and ωi is the angle of arrival (AoA)

NLoS
of the LoS component at Ri while gi,R i
∈ CK×1 is the NLoS component which
follows Rayleigh fading as previously described. The channel between RIS and edge
users can be modeled similarly using Rician fading.

For the sake of simplicity, we assume perfect channel state information (CSI) at
the BSs. While achieving perfect CSI in practice can be challenging, recent ad-
vancements in channel estimation techniques for RIS-assisted wireless networks have
demonstrated the potential for accurate CSI acquisition with a reasonable over-

34
 Chapter 4. Energy Efficient Design for CoMP-NOMA Networks

head [16, 30–32].

4.1.2 RIS Configuration

Each BSi utilizes a passive RIS Ri at the cell edge to improve signal quality or
suppress interference for Uef , ∀f ∈ F. RIS elements can independently adjust the
phase shift of the incident signal and are assumed to be controlled by the CPU.
Furthermore, the phase shift (PS) matrix associated with Ri is expressed as Θi =
i i i
diag(l1 ejθ1 , l2 ejθ2 , . . . , lK ejθK ), where lk ∈ (0, 1] is the amplitude adjustment factor
and θki ∈ [−π, π) is the phase shift of the k-th element. In this work, we assume
an ideal RIS with perfect phase control and all reflection elements having a unit
amplitude (lk = 1, ∀k).†

4.2 Performance Analysis

4.2.1 Rate and Outage Probability Analysis
According to the NOMA principle, the BSs serve multiple users simultaneously
by superimposing their signals. Specifically, the signal transmitted by BSi can be
√ p
expressed as xi = ζi Pi se + (1 − ζi )Pi sci , where Pi is the transmit power of BSi ,
se and sci are the signals intended for Uef and Uic , respectively, and ζi is the power
allocation factor for the edge users. To ensure successful decoding by Uic , i.e., the
strong user, ζi is constrained to 0.5 < ζi < 1 [22].

The received signal at Uef can be written as

X p X q X
yfe = e
Hj,f ζj Pj se + e
Hj,f (1 − ζj )Pj scj + e
Hm,f xi +no , (4.3)
j∈J j∈J m∈I\J
| {z } | {z } | {z }
CoMP gain intra-cluster interference inter-cell interference

e
where Hj,f = hej,f + hTRj ,f Θj hj,Rj and Hm,f
e
= hem,f + hTRi ,f Θi hm,Ri are the effec-
tive channels between BSj and Uef and between BSi and Uef , respectively, and
no ∼ CN (0, σ 2 ) is the additive white Gaussian noise. To minimize synchroniza-
tion overhead, we employ non-coherent JT-CoMP, where the edge user Uef combines
signals from cooperative BSs without CSI exchange [24]. Therefore, the signal-to-
†
While practical RIS implementations are subject to phase quantization errors, this work pri-
oritizes establishing a proof-of-concept for the benefits of RIS-assisted CoMP-NOMA in terms of
energy efficiency and performance.

35
 Chapter 4. Energy Efficient Design for CoMP-NOMA Networks

interference-plus-noise ratio (SINR) at Uef can be expressed as

e 2
P
ζj Pj |Hj,f
j∈J |
γfe = P e 2 2
, (4.4)
j∈J (1 − ζj )Pj |Hj,f | + Ye + σ

e
|2 represents the inter-cell interference term.
P
where Yf = m∈I\J Pm |Hm,f

On the other hand, the received signal at Uic is given by

p X p
yci = hii,c (1 − ζi )Pi sci + hij,c ζj Pj se
j∈J
X q X
+ hij,c (1 − ζj )Pj scj + him,c xm + no . (4.5)
j∈J ,j̸=i m∈I\J

Based on the SIC principle, the SINR at Uic for decoding the signal intended for Uef
is given by
i 2
P
i j∈J ζj Pj |hj,c |
γc→f = P i 2 2
, (4.6)
j∈J (1 − ζj )Pj |hj,c | + Yi + σ

and the SINR at Uic for decoding its own signal is

(1 − ζi )Pi |hii,c |2
γci = P i 2 2
. (4.7)
j∈J ,j̸=i (1 − ζj )Pj |hj,c | + Yi + σ

It is worth noting that due to their placement at the cell edge, the impact of RIS
on the channels experienced by Uic is negligible. Thus, the SINR expressions for Uic
only consider the direct links between the BSs and the users. Finally, the achievable
rates for Uef and Uic can be calculated as

Ref = log2 (1 + γfe ), (4.8)

and

Ric = log2 (1 + γci ). (4.9)

An outage event occurs for cell-center users if Uic fails to decode se or is capable of
decoding se but fails to decode sci . The corresponding outage probability can be
e i
defined as Pic = 1 − P(γc→f
i
> γˆf , γci > γˆc ), where γˆf = 2Rth − 1 and γˆc = 2Rth − 1
represent the target SINR thresholds for Uef and Uic , respectively, corresponding to
their target rates Reth and Rith . For edge user Uef , an outage occurs if it fails to

36
 Chapter 4. Energy Efficient Design for CoMP-NOMA Networks

decode se , and the outage probability can be formulated as

Pef = P(γfe < γˆf ). (4.10)

4.2.2 Energy Efficiency

We define the energy efficiency of the network as the ratio of the outage sum rate to
the total power consumption. This metric essentially assesses whether the increase
in outage sum rate resulting from CoMP outweighs the corresponding increase in
total power consumption. Mathematically, the energy efficiency is formulated as

Reoutf
P
i
P
c∈C (i) Routc f ∈F
X X
ηEE = 1 + 1 , (4.11)
i∈I
P + PQ
λ i
P
j∈J λ j
+ PQ + PR

where Rioutc = (1−Pic )Ric and Reoutf = (1−Pef )Ref represent the effective outage rate
for Uic and Uef , respectively. Moreover, PQ represents the static power consumption
of a cell, PR = KPele denotes the total power consumption of Ri , where Pele is
the power consumption of k-th element, and λ ∈ (0, 1] signifies the power amplifier
efficiency.

4.3 Passive Beamforming Design

The primary objective of the PBF design is to optimize the energy efficiency of the
network by strategically adjusting the RIS phase shifts. This involves configuring
Rj associated with cooperative BSj to enhance the signal quality for Uef , while
simultaneously utilizing Rk , ∀m ∈ I \ J , to suppress the inter-cell interference
experienced by Uef . The optimization problem is formulated as

max energy efficiency ηEE in (4.11) (4.12a)

Φ
s.t. θkj ∈ [−π, π), ∀k ∈ [1, K], j ∈ J , (4.12b)
θkm ∈ [−π, π), ∀k ∈ [1, K], m ∈ I \ J (4.12c)

where Φ = {Θ1 , Θ2 , . . . , ΘI } represents the set of phase shift matrices for all RIS.

For cooperative BSs, the phase shifts of Ri in constraint (4.12b) are optimized to
e 2
maximize the effective channel gain |Hj,f | . From [33], the optimal phase shift for
each element of Rj is given by

(k) (k)
θkj = arg(hej,f ) − arg(hRj ,f · hj,Rj ), (4.13)

37
 Chapter 4. Energy Efficient Design for CoMP-NOMA Networks

(k) (k)
where arg(·) denotes the argument function, and hRj ,f and hj,Rj represent the k-th
elements of the channel vectors hRj ,f and hj,Rj , respectively.

In contrast to signal enhancement, the phase shifts of Rm for non-cooperating BSs

e
in constraint (4.12c) are adjusted to minimize the effective channel gain |Hm,f |2 ,
thus mitigating interference. Thus, the optimal phase shift for each element of Rm
can be calculated as
θkm = mod [ϕm
k + π, 2π] − π, (4.14)
(k) (k)
where ϕm e
k = arg(hm,f )−arg(hRm ,f ·hm,Rm ) and mod[·] denotes the modulo operation.
Adding π ensures a 180° phase shift, effectively suppressing interference.

It should be noted that incorporating cell-center users into the optimization problem,
or increasing the number of users per cell, significantly increases the complexity of
the beamforming design. To maintain tractability, this work focuses on a single edge
user and a single cell-center user per cell.

Numerical Results

4.4 Simulation Setup

The performance of the proposed design is evaluated in a network consisting of
I = 6 cells, each with a radius of Ri = 75 m. All BSs transmit at the same power
level, i.e., Pi = Pt dBm, ∀i ∈ I, and the power allocation factor for edge users is
set to ζi = 0.7. The RIS are positioned at the cell edge, resulting in a distance
of di,Ri = 75 m between BSi and Ri . The distances between BSs and users are
configured as follows: dii,c = 50 m, dei,f = 150 m, and dik,c = 200 m for k ̸= i, where
i ∈ I, c ∈ C (i) , and f ∈ F. Similarly, the distance between RIS and edge users is
set to dei,f = 75 m.

The path loss exponents are set to αR = 2.7, αi = 3, αe = 3.5, and αici = 4
for RIS, BS, edge user, and inter-cell interference links, respectively. The network
operates at a carrier frequency of fc = 2.4 GHz, and the noise power is defined as
σ 2 = −174 + 10 log10 (B), with a bandwidth B = 10 MHz. Table 4.1 summarizes
the remaining simulation parameters.

4.5 Impact of Cooperation on Energy Efficiency

Figure 4.2 illustrates the energy efficiency of the network as a function of the num-
ber of cooperating BSs J for various RIS configurations. The investigated scenarios

38
 Chapter 4. Energy Efficient Design for CoMP-NOMA Networks

Table 4.1: Simulation Parameters

Parameter Value
Power amplifier efficiency λ 0.4
Reference path loss ρo −30 dBm
Static power consumption PQ 30 dBm
Power dissipated at k-th RIS element Pele 5 dBm
Target cell-center user rate Rith 1 bps/Hz
Target edge user rate Reth 0.5 bps/Hz
Rician factor κ 3 dB
# of channel realizations Nmc 104

17
EC-based Phase Shifts
Energy e/ciency (bps/Hz/Joule)

16.8 EO-based Phase Shifts

Random Phase Shifts
No RIS
16.6

16.4

16.2

16

15.8

1 2 3 4 5 6
Number of cooperative BSs

Figure 4.2: Energy efficiency vs. number of cooperative BSs J at Pt = 0 dBm and under
various RIS configurations with K = 70 elements.

include: no RIS, RIS with random phase shifts, EO-based RIS, and EC-based RIS.
We observe that for the EO and EC configurations, the energy efficiency initially
increases with a growing number of cooperating BSs, reaching a peak at J = 4.
Beyond this point, the efficiency experiences a decline due to the saturation of both
the achievable rate and the outage probability, leading to diminishing returns. It
should be noted that the EC configuration consistently outperforms other scenarios
∀J, except when J = I. In this particular case where all BSs are cooperative, inter-
ference cancellation becomes redundant, leading to equivalent performance between
the EO and EC configurations.

4.6 Impact of RIS Elements on Energy Efficiency

The impact of the number of RIS elements K on the energy efficiency is depicted
in Figure 4.5. A clear trend emerges across all configurations: energy efficiency

39
 Chapter 4. Energy Efficient Design for CoMP-NOMA Networks

56

Outage sum rate (bits/s/Hz)

54

52

50

48 CoMP-NOMA, EC-based Phase Shifts

CoMP-OMA, EC-based Phase Shifts
CoMP-NOMA, without RIS
46 No-CoMP-NOMA, without RIS

8 9 10 11 12 13 14
Transmit power per BS, Pt (dBm)

Figure 4.3: Outage sum rate vs. transmit power Pt with J = 4 cooperative BSs and
K = 70 elements.
1.8
Energy e/ciency (bps/Hz/Joule) 1.4 30
1.6 2.6
4.
1
Rate threshold Rth (bps/Hz)

5.

1.4 25
6
7.
8
10

13
.9

1.2 16 .6
.4 20
19
1 .1
21
.8
0.8 15
24.4
0.6 26.8
10
0.4 29.2
31.6 5
0.2

0
-10 -5 0 5 10 15
Transmit power per BS, Pt (dBm)

Figure 4.4: Energy efficiency contour plot for varying transmit power Pt and rate threshold
Rth with J = 4 cooperative BSs and K = 70 elements.

improves as the number of RIS elements increases, reaching a peak at K = 90.

Beyond this point, the gains in outage sum rate are counterbalanced by the increased
power consumption associated with additional elements, resulting in a decline in
energy efficiency. Notably, the EC configuration consistently outperforms the EO
configuration across all values of K. Furthermore, the No-CoMP scenario exhibits
the lowest energy efficiency for all values of K, underscoring the significance of
CoMP in enhancing overall network performance.

40
 Chapter 4. Energy Efficient Design for CoMP-NOMA Networks

19 CoMP-NOMA, EC-based Phase Shifts

CoMP-NOMA, EO-based Phase Shifts

Energy e/ciency (bps/Hz/Joule)

CoMP-NOMA, Random Phase Shifts
18
No-CoMP-NOMA, EC-based Phase Shifts

17

16

15

14

13

18 54 90 126 162
Number of RIS elements K

Figure 4.5: Energy efficiency vs. number of RIS elements K at Pt = 0 dBm and J = 4
cooperative BSs.

CoMP-NOMA, J = 1 (Cooperative BSs)

29
CoMP-NOMA, J = 3 (Cooperative BSs)
Outage sum rate (bits/s/Hz)

28.8 CoMP-NOMA, J = 6 (Cooperative BSs)

28.6
28.4
28.2
28
27.8
27.6
27.4

0/72 9/63 18/54 27/45 36/36 45/27 54/18 63/9 72/0

CO/EO split-ratio

Figure 4.6: Outage sum rate vs. CO/EO split-ratio for different cooperative BSs J with
Pt = 0 dBm with K = 72 elements.

4.7 Outage Sum Rate

The analysis of outage sum rate concerning transmit power Pt is illustrated in Fig-
ure 4.3. To provide a comparative context, orthogonal multiple-access (OMA) is
included in the evaluation. As anticipated, the outage sum rate demonstrates an
increasing trend with a rise in transmit power across all considered configurations.
Notably, the implementation of CoMP-NOMA with EC-based RIS consistently sur-
passes all other scenarios, including OMA, throughout the entire range of Pt values.
An interesting observation is that the CoMP-OMA configuration exhibits steeper
increases in outage sum rate per unit of power compared to the CoMP-NOMA
configurations. This characteristic can be attributed to the more pronounced rate
saturation effect in NOMA networks, wherein the achievable rate is constrained by

41
 Chapter 4. Energy Efficient Design for CoMP-NOMA Networks

the SIC capabilities of the users.

4.8 Rate Threshold and Energy Efficiency

Figure 4.4 presents a contour plot illustrating the relationship between energy effi-
ciency, transmit power Pt , and joint rate threshold Rth , where we set Rith = Reth =
Rth for simplicity. Although the highest levels of energy efficiency are found in
regions with low Pt and Rth , these operating points are often impractical due to
minimum user rate requirements. As we move diagonally across the plot, a clear
trend emerges: energy efficiency decreases as the rate threshold increases. This in-
verse relationship highlights the inherent trade-off between achieving higher data
rates and maintaining energy efficiency. To meet the demand for higher Rth , the
network necessitates higher Pt , leading to increased power consumption and subse-
quently, reduced energy efficiency.

4.9 Impact of PBF Design on Sum Rate

Lastly, to summarize the trade-off between PBF designs, we analyze the outage sum
rate as a function of the ratio of Cancellation-only (CO) and EO elements in Fig-
ure 4.6. As expected, increasing CO elements led to a decrease in outage sum rate
for the fully cooperative scenario, J = I. However, with half of the BSs cooperat-
ing, the outage sum rate remained relatively stable regardless of the CO/EO ratio,
demonstrating a balanced contribution from both designs. Interestingly, the highest
gains were observed when J = 1, with all elements employing the CO scheme. This
emphasizes the critical role of interference cancellation in such scenarios. Moreover,
these findings highlight the importance of optimizing network parameters through
a robust optimization framework, a direction we intend to explore in the future.

42
Chapter 5

Deep Reinforcement Learning for

Intelligent NOMA Networks

While the previous chapter explored the potential of static RIS configurations in
CoMP-NOMA networks, this chapter delves into the dynamic adaptation of these
networks through the integration of Unmanned Aerial Vehicles (UAVs) and Aerial
RIS (ARIS). This combination of technologies presents a unique opportunity to
further enhance network performance by dynamically adjusting the propagation en-
vironment, enabling flexible coverage, and optimizing resource allocation. Existing
research has demonstrated the potential of incorporating UAVs within RIS, CoMP,
and NOMA frameworks, showcasing notable improvements in various performance
metrics as evidenced in [34, 35].

However, many of these studies rely on static RIS deployments, limiting the adapt-
ability and flexibility of the network. Although some works have investigated ARIS-
assisted CoMP-NOMA networks and optimized UAV trajectory and RIS phase shifts
for sum rate maximization [36], the optimization methods employed, such as double-
layer alternating optimization, often face scalability issues due to their complexity
and potential convergence challenges.

To overcome these limitations, this chapter introduces a novel approach based on

Deep Reinforcement Learning (DRL) for optimizing ARIS-assisted CoMP-NOMA
networks. Our DRL framework jointly optimizes UAV trajectory, RIS phase shifts,
and NOMA power control, aiming to maximize the network sum rate while adhering
to user Quality of Service (QoS) constraints. Through extensive simulations, we
evaluate the effectiveness of our proposed approach, assess the convergence behavior
of MO-PPO, and highlight the significant benefits of integrating CoMP-NOMA and
RIS within UAV-assisted networks.

43
 Chapter 5. Deep Reinforcement Learning for Intelligent NOMA Networks

Towards Optimization

5.1 System Model

As shown in Figure 5.1, we consider a multi-cell CoMP-NOMA network assisted
by a UAV-mounted RIS in a downlink transmission scenario. The network consists
of I cells, each modeled as a circular disk of radius R with a single-antenna BS at
its center, denoted as BSi , where i ∈ I ≜ {1, 2, . . . , I}. Each BSi invokes two-user
downlink NOMA to serve its respective cell-center and edge user, each also equipped
with a single-antenna. The cell-center users are defined as users that lie within the
disk of their associated cell and are denoted as Uci , ∀i and ci ∈ C i ≜ {1, 2, . . . , Ci },
where Ci is the number of cell-center users in cell i. Conversely, the edge users
are defined as users that do not lie within any cell and are denoted as Uf , ∀f
and f ∈ F ≜ {1, 2, . . . , F }, where F is the number of edge users in the network.
Furthermore, let U ≜ i∈I C i ∪ F be the set of all the users in the network. Without
S

loss of generality and for ease of exposition, we assume I = 2, and Ci = F = 1, ∀i.

Aerial
RIS

NOMA pairs

Desired links Interference links

Reflected links Blocked links

Figure 5.1: Aerial RIS-assisted coordinated NOMA cluster.

For coordinated operation, the BSs are assumed to be interconnected via a high-
speed backhaul network to a central processing unit (CPU). Moreover, to improve
the signal quality for edge users, an ARIS, denoted as R, is deployed at a fixed
altitude H over area A to create reflection links between the BSs and the users,
and is equipped with K passive elements. For tractability, we discretize the entire
system operation into time slots of equal length τ , where each time slot is indexed by
t ∈ T ≜ {1, 2, . . . , T }, such that T is the total flight time of the UAV. Furthermore,
we assume the presence of O obstacles in the network, denoted as O ≜ {1, 2, . . . , O},
where each obstacle Oo , o ∈ O has its own forbidden zone represented as a circular
disk of radius dmin , centered at the obstacle’s location.

44
 Chapter 5. Deep Reinforcement Learning for Intelligent NOMA Networks

Before proceeding with the channel and signal model, we define the positions of
the various entities in the network. Specifically, ∀i ∈ I, u ∈ U, and o ∈ O, the
positions of BSi , Uu , and Oo are represented by pi = (xi , yi , HB ), pu = (xu , yu , 0),
and po = (xo , yo , HO ), respectively, where HB and HO are the heights of the BSs
and obstacles, respectively. Moreover, the position of R at time slot t is denoted
as pR [t] = (xR [t], yR [t], HR ). In this work, we assume that the users are stationary,
and the UAV is capable of adjusting its horizontal position in the xy-plane, while
maintaining a fixed altitude HR .

5.1.1 Channel Model & RIS Configuration

In our analysis, both the large-scale path loss and small-scale fading effects are
considered. Similar to [34], we assume the presence of numerous scatterers in the
environment, and thus the direct links between BSi and Uu , denoted as hi,u , are
modeled as Rayleigh fading channels. Mathematically, the channel hi,u at time slot
t is given by
ρo
r
hi,u [t] = vi,u [t], (5.1)
P L(di,u )
where ρo is the reference path loss at 1 m, P L(di,u ) = (di,u )−αi,u is the large-scale path
loss, such that αi,u is the path loss exponent, and di,u = ∥pi − pu ∥ is the distance
between BSi and Uu and ∥·∥ denotes the Euclidean norm. Moreover, vi,u [t] ∈ C1×1
is the small-scale Rayleigh fading coefficient with zero mean and unit variance, and
is assumed to be independent and identically distributed (i.i.d) across different time
slots and users. In this work, as a special case, we assume that the direct link
between BSi and Uf is blocked due to the presence of obstacles, thus hi,f [t] = 0,
∀i, f .

Contrary to the direct links, the reflection links between BSi and R are modeled
as Rician fading channels, denoted as hi,R [t], due to the presence of a dominant
line-of-sight (LoS) component. At time slot t, the channel hi,R [t] is given by
r r !
ρo κ 1
r
hi,R [t] = gLoS [t] + gNLoS [t] , (5.2)
P L(di,R [t]) 1 + κ i,R 1 + κ i,R

where κ is the Rician factor representing the ratio of the power of the LoS component
to the power of the scattered components, and di,R [t] = pi − pR [t] is the distance
LoS
between BSi and R. Moreover, the deterministic LoS represented, i.e., gi,R [t] ∈
K×1
C , is given by
h iT
LoS j(k−1)π sin(ωi ) j(K−1)π sin(ωi )
gi,R = 1, . . . , e ,...,e ,

45
 Chapter 5. Deep Reinforcement Learning for Intelligent NOMA Networks

where k ∈ K ≜ {1, 2, . . . , K} indexes elements of R and ωi is the angle of arrival

NLoS
(AoA) whereas gi,R i
∈ CK×1 is the NLoS component following Rayleigh fading as
previously described. Similarly, the channel between R and Uu , denoted as hR,u ,
can also be modeled as a Rician fading channel.

For the RIS configuration, we assume that the phase shift of the k-th element can
be set independently of other elements and that both the UAV trajectory and the
phase response are controlled by the CPU. Furthermore, the phase shift (PS) matrix
at time slot t is expressed as
 
jθ1 [t] jθ2 [t] jθK [t]
Θ[t] = diag a1 e , a2 e , . . . , ak e , (5.3)

where ak ∈ (0, 1] is the amplitude coefficient and θk [t] ∈ [−pi, pi) is the phase shift
of the k-th element. In this work, we assume an ideal RIS with perfect phase control
and all reflection elements having a unit amplitude, i.e., ak = 1, ∀k. Furthermore,
we assume the availability of perfect channel state information (CSI) at the CPU.
While this is a challenging assumption in practice, recent advancements in chan-
nel estimation techniques for RIS-assisted wireless networks have demonstrated the
feasibility of achieving accurate CSI with reasonable overhead [16, 30, 32].

5.1.2 Signal Model

In accordance with the NOMA principle, each BSi serves two users, Uci and Uf ,
simultaneously, by superimposing their signals. Let xi,ci [t] and xi,f [t] be the desired
signals intended for Uci and Uf at time slot t, then the transmitted signal from
p √
BSi can be expressed as xi [t] = (1 − λi )Pi xi,ci [t] + λi Pi xi,f [t], where Pi is the
transmit power of BSi and λi is the power allocation factor assigned to Uf , such
that λi ∈ (0.5, 1) to ensure successful decoding at Uci [22, 25].

The received signal at Uf can be expressed as

yf [t] = Hi,f [t]xi [t] + Hi′ ,f [t]xi′ [t] + no [t] (5.4)

where i′ ∈ I \ {i}, no [t] ∼ CN (0, σ 2 ) is the additive white Gaussian noise (AWGN),
and Hi,f [t] = hTR,f [t]Θ[t]hi,R [t] represents the effective channels between BSi and Uf
through R, respectively. To minimize synchronization overhead, we employ non-
coherent JT-CoMP, thus, the signal-to-interference-plus-noise ratio (SINR) is given
by
2 2
λi Hi,f [t] + λi′ Hi′ ,f [t]
γf [t] = 2 2 , (5.5)
(1 − λi ) Hi,f [t] + (1 − λi′ ) Hi′ ,f [t] + ρ1

46
 Chapter 5. Deep Reinforcement Learning for Intelligent NOMA Networks

where ρ = Pt /σ 2 is the transmit SNR and Pt = Pi , ∀i is the transmit power of each

BS.

On the other hand, the received signal at Uci can be expressed as

yci [t] = Hi,ci [t]xi [t] + hi′ ,ci [t]xi′ [t] + no [t], (5.6)

where Hi,ci [t] = hi,ci [t] + hTR,ci [t]Θ[t]hi,R [t] represents the effective channels between
BSi and Uci through R, respectively. Also, the term hi′ ,ci [t]xi′ [t] represents the ICI
caused by the transmission of BSi′ at Uci . Based on the SIC principle, Uci first
decodes xi,f [t] and then cancels it from yci [t] to decode xi,ci [t]. The SINR at Uci for
decoding xi,f [t] is given by

2
λi Hi,ci [t]
γci →f [t] = 2 2 , (5.7)
1
(1 − λi ) Hi,ci [t] + hi′ ,ci [t] + ρ

whereas the SINR at Uci for decoding xi,ci [t] is

2
(1 − λi ) Hi,ci [t]
γci [t] = 2 . (5.8)
1
hi′ ,ci [t] + ρ

Finally, the achievable sum rate of the network at time slot t can be expressed as
X X
Rsum [t] = Rci [t] + Rf [t]. (5.9)
i∈I f ∈F

where Rci [t] = log2 (1 + γci [t]) and Rf [t] = log2 (1 + γf [t]) are the achievable rates of
Uci and Uf , respectively.

5.2 Problem Formulation

In this work, our primary objective is to maximize the sum rate achieved over T
time slots. To achieve this goal, we jointly optimize three key control variables: the
UAV trajectory denoted as P ≜ {pR [t], ∀t}, the RIS phase shifts represented by
Θ ≜ {Θ[t], ∀t}, and the power allocation factors denoted as Λ ≜ {λi , ∀i}.

47
 Chapter 5. Deep Reinforcement Learning for Intelligent NOMA Networks

The optimization problem can be mathematically formulated as

X
max Rsum [t] (5.10a)
P, Θ, Λ t∈T

s.t. xR [t], yR [t] ∈ A, ∀t ∈ T , (5.10b)

pR [t] − po ≥ dmin , ∀o ∈ O, t ∈ T , (5.10c)
θk [t] ∈ [−π, π), ∀k ∈ K, t ∈ T , (5.10d)
Rci [t] ≥ Rcmin
i
, ∀i ∈ I, t ∈ T , (5.10e)
Rf [t] ≥ Rfmin , ∀f ∈ F, t ∈ T , (5.10f)
λi ∈ (0.5, 1), ∀i ∈ I, (5.10g)

where constraint (5.10b) restricts the UAV trajectory to lie within A, and con-
straint (5.10c) enforces a minimum safety distance between the UAV and any obsta-
cles present, thus guaranteeing the UAV’s safety. Constraint (5.10d) limits the phase
shifts applied by the RIS elements. To meet the quality of service (QoS) require-
ments, constraints (5.10e) and (5.10f) impose minimum rate thresholds, denoted by
Rmin
ci and Rmin
f , for Uci and Uf , respectively. Lastly, constraint (5.10g) defines the
permissible range for power allocation factors, ensuring successful SIC. The opti-
mization problem in (5.10) is non-convex due to the coupled variables {P, Θ, Λ}.
To address this, we propose a DRL-based solution in the next section.

5.3 Deep Reinforcement Learning-based Proposed

Solution
5.3.1 MDP Formulation
Before proceeding with the DRL-based solution, we model it as a single-agent
Markov Decision Process (MDP) with discrete time steps. This MDP is repre-
sented by the tuple ⟨S, A, P, R, γ⟩, where S denotes the set of possible environment
states, A represents the action space, P defines the state transition probabilities,
R is the reward function guiding the agent’s learning, and γ is the discount factor
that determines the importance of future rewards. At each time slot t, the agent
observes the current state st , selects an action at based on its policy, transitions to
a new state st+1 , and receives a reward R(st , at ). We define S, A, and R as follows

1. State Space S: The environment state at time slot t consists of the UAV’s cur-
rent position pR [t], the distance from the UAV to the center of obstacles dR [t] =
{ pR [t] − po , ∀o ∈ O}, the power allocation factors Λ, and the achievable rates

48
 Chapter 5. Deep Reinforcement Learning for Intelligent NOMA Networks

R[t] = {Rci [t], Rf [t], ∀i, f }. Thus, the state space can be expressed as

st = {pR [t], dR [t], Λ, R[t]} ∈ RdimS . (5.11)

P
where dimS = 2 + O + I + i∈I Ci + F is the dimension of the state space.

2. Action Space A: The action space of the formulated MDP consists of the UAV’s
movement in the horizontal xy-plane, the phase shifts of the RIS elements, and the
power allocation factors. Specifically, the action space at time slot t contains the
manuevering actions aR [t] ∈ {(−1, 0), (1, 0), (0, −1), (0, 1), (0, 0)}, representing left,
right, down, up, and hover, respectively, the phase shifts aΦ [t] = {ϕk [t], ∀k}, and
the power allocation factors aΛ = {λi , ∀i}. Thus, the action space can be expressed
as
at = {aR [t], aΦ [t], aΛ } ∈ RdimA . (5.12)

where dimA = 2 + K + I is the dimension of the action space.

3. Reward Function R: The reward function plays a crucial role in shaping the
learning behavior of the RL agent. Our design encourages maximizing the sum rate
while ensuring UAV safety and meeting QoS requirements by penalizing constraint
violations. The reward function is defined as
P !
u∈U ζu [t]
R(st , at ) = Rsum [t] 1 − − ξR [t]Kviol , (5.13)
|U|

where Kviol is the penalty factor for constraint violation, and ζu [t] = I{Ru [t] ≤ Rumin }
is the indicator function for the QoS constraints, i.e., ζu [t] = 1 if QoS constraints
are violated, and 0 otherwise. Similarly, ξR [t] = I{xR [t], yR [t] ∈
/ A ∧ pR [t] − po <
dmin , ∀o ∈ O} is the indicator function for UAV’s safety constraints.

5.3.2 MO-PPO Algorithm

In this work, the considered action space is a hybrid continous-discrete space, which
poses a challenge for traditional RL algorithms. While discretization of continuous
actions is a possibility, it can lead to a large action space, significantly increasing
computational complexity and potentially hindering performance. To address this
challenge, we propose employing a multi-output Proximal Policy Optimization (MO-
PPO) algorithm. MO-PPO extends the standard PPO [38] framework by employing
two parallel actor networks, each responsible for generating the discrete action aR
and the continuous actions aΦ and aΛ , respectively. The actor networks share the
first few layers, allowing for the extraction of common features and encoding the

49
 Chapter 5. Deep Reinforcement Learning for Intelligent NOMA Networks

Algorithm 1: MO-PPO Algorithm

1 Initialize the policy parameters θd and θc
2 for episode = 1, 2, . . . , N do
3 Receive initial state s0
4 for time step t = 0, 1, . . . , T do
5 Generate discrete action aR using πθd (at |st )
6 Generate continuous actions aΦ and aΛ using πθc (at |st )
7 Execute actions at = {aR , aΦ , aΛ }
8 if UAV violates (5.10b) or (5.10c) then
9 Set ξR [t] = 1, cancel the UAV’s movement, and update the state
st+1
10 end
11 Observe reward R as (5.13) and next state st+1
12 Collect a set of partial trajectories D with T̂ transitions
13 Compute the advantage estimate Ât as (5.14)
14 end
15 for epoch = 1, 2, . . . , E do
16 Sample a mini-batch of transitions B from D
17 Compute the clipped surrogate objectives LCLIP d (θd ) and LCLIP
c (θc )
as (5.15)
18 Optimize overall objective and update the policy parameters θd and
θc using Adam [37]
19 end
20 Synchronize the sampling policies as
θdold ← θd and θcold ← θc
21 Clear the collected trajectories D
22 end

state information. Furthermore, a single critic network is employed to estimate

the value function V (st ), which is used to compute a variance-reduced advantage
function estimate Ât for policy optimization. Following the implementation details
used in [39], the policy is executed for T̂ time steps, and Ât is computed as

T̂ −1
X
Ât = γ k rt+k + γ T̂ V (st+T̂ ) − V (st ), (5.14)
k=0

where T̂ is much smaller than the length of the episode T .

To generate the stochastic policy πθd (at |st ) for the discrete actions, the correspond-
ing actor network outputs |aR | logits, which are then passed through a softmax
function to obtain a probability distribution over the available discrete actions. Con-
versely, the continuous actor network generates the continuous actions aΦ and aΛ by
sampling from Gaussian distributions parameterized by the mean and standard de-

50
 Chapter 5. Deep Reinforcement Learning for Intelligent NOMA Networks

viation outputs of the network, as dictated by the stochastic policy πθc (at |st ). Both
πθd (at |st ) and πθc (at |st ) are optimized independently using their respective clipped
surrogate objective functions. For the discrete actions, the objective function is
given by h i
LCLIP
d (θd ) = Êt min(rt
d
(θd )Ât , ℵ(rt
d
, θd , ϵ)Ât , (5.15)

where ℵ(rtd , θd , ϵ) = clip(rtd (θd ), 1 − ϵ, 1 + ϵ), rtd (θd ) = πθd (at |st )/πθold
d
(at |st ) is the
importance sampling ratio, and ϵ is the clipping parameter. The objective function
for the continuous actions can be expressed in a similar manner but is left out for
brevity.

It is important to note that while both policies collaborate within the environment,
their optimization objectives remain decoupled, i.e., πθd (at |st ) and πθc (at |st ) are
treated as independent distributions during policy optimization, rather than a joint
distribution encompassing both action spaces. The MO-PPO algorithm is summa-
rized in Algorithm 1.

5.3.3 Complexity and Convergence Analysis

The complexity of DRL algorithms is commonly measured in terms of the number
of multiplications per iteration, which is a function of the number of parameters
in the policy and value networks. For MO-PPO, the overall complexity can be
expressed as O[ Q
P s PQd PQc
q=1 nq · nq−1 + q=1 nq · nq−1 + q=1 nq · nq−1 ], where Qs , Qd , and
Qc are the number of layers in the shared, discrete, and continuous actor networks,
respectively, and nq and nq−1 are the number of neurons in the q-th and (q − 1)-th
layers, respectively. In this work, we consider the same number of neurons in each
hidden layer, i.e., nq = nq−1 = n, ∀q, and the number of neurons in the output
layer is equal to the dimension of the action space. Thus, the overall complexity of
MO-PPO is O[n2 (Qs + Qd + Qc )].

Similar to other DRL algorithms, the convergence of MO-PPO is mathematically

difficult to analyze [40] since neural networks are highly dependent on the choice of
hyperparameters. However, the convergence of MO-PPO can be empirically verified
by monitoring the agent’s performance over multiple episodes and ensuring that the
reward function converges to a stable value. Moreover, the convergence of MO-PPO
can be accelerated by tuning the learning rate, clipping parameter, and penalty
factor for constraint violation.

51
 Chapter 5. Deep Reinforcement Learning for Intelligent NOMA Networks

Numerical Results

5.4 Simulation Setup

To evaluate the efficacy of the proposed MO-PPO algorithm, we construct a simu-
lated urban environment spanning an area of 150 × 150 m2 with I = 2 BSs, U = 3
users, and O = 2 obstacles. The initial position of the UAV is set to (0, 35, 50) m,
while BS1 and BS2 are located at (−35, −35, 25) m and (35, 35, 25) m, respectively.
All remaining entities are randomly placed within the environment.

Both BSs are assumed to transmit at an identical power level, i.e., P1 = P2 =

Pt . Furthermore, the network operates at a carrier frequency of fc = 2.4 GHz,
utilizing a bandwidth of BW = 10 MHz and the noise power is set to σ 2 = −174 +
10 log10 (BW ) dBm. To model the signal propagation characteristics, we employ
path loss exponents of αi,u = 3, αi,R = αR,u = 2.2, and αi′ ,u = 3.5, for direct,
reflection, and interference links, respectively. Table 5.1 summarizes the remaining
simulation parameters.
Table 5.1: Simulation Parameters
Parameter Value Parameter Value
Reference path loss ρo −30 dBm Rician factor κ 3 dB
Target data rate Rfmin 0.2 bps/Hz Learning rate 2.75e − 4
Target data rate Rcmin
i
0.5 bps/Hz Clipping parameter ϵ 0.1
Penalty constant Kviol 7 Discount factor γ 0.98
Minimum distance dmin 10 m Number of episodes N 750
Time slots per episode T 250 Number of epochs E 20
Number of neurons 64 Batch size B 128

5.5 Learning and Convergence

Figure 5.2 illustrates the average cumulative reward achieved by the MO-PPO algo-
rithm with different network configurations. As shown, the algorithm consistently
converges to a stable reward value after approximately 500 episodes, indicating the
successful acquisition of an effective policy. Notably, MO-PPO with random PS
exhibits a faster convergence rate compared to its counterpart with optimal PS.
This observation can be attributed to the increased complexity associated with op-
timizing the PS within the action space, leading to a slower convergence process.
Moreover, the CoMP-NOMA configuration achieves a superior average cumulative
reward compared to the CoMP-OMA configuration, underscoring the benefits of

52
 Chapter 5. Deep Reinforcement Learning for Intelligent NOMA Networks

35
MO-PPO, CoMP-NOMA Optimal PS
30 MO-PPO, CoMP-OMA Optimal PS
convergence

Average cumulative reward

MO-PPO, CoMP-NOMA Random PS
25 Exhaustive Search, CoMP-NOMA

20

15

10

0
100 200 300 400 500 600 700
Episodes

Figure 5.2: Average cumulative reward vs. number of training episodes with Pt = 20 dBm
and K = 120 elements.

MO-PPO, CoMP-NOMA Optimal PS Trajectory

25 MO-PPO, CoMP-OMA Optimal PS gain
H-PPO, CoMP-NOMA Optimal PS
MO-PPO, No-CoMP-NOMA Optimal PS
Sum rate (bits/s/Hz)

20 MO-PPO, CoMP-NOMA Random PS

15 NOMA gain
RIS gain
gain
C oM P
10

5
0 2 4 6 8 10 12 14 16 18 20
Transmit power per BS, Pt (dBm)

Figure 5.3: Sum rate vs. transmit power for different algorithms and configurations with
K = 120 elements.

NOMA in enhancing overall network performance. A comparison with the optimal

solution obtained through exhaustive search reveals that the proposed MO-PPO al-
gorithm achieves near-optimal performance, effectively demonstrating its capability
to solve the formulated problem.

5.6 Algorithmic Performance on Sum Rate

Next, we investigate the sum rate achieved by the network as a function of the
transmit power Pt as shown in Figure 5.3. As expected, the sum rate exhibits
an upward trend with increasing transmit power, emphasizing the crucial role of
power control in optimizing network performance. The results clearly showcase the

53
 Chapter 5. Deep Reinforcement Learning for Intelligent NOMA Networks

22
Exhaustive Search
20 MO-PPO, Optimal PS

Sum rate (bits/s/Hz)

H-PPO, Optimal PS
18 MO-PPO, Random PS

16

14

12

10
18 54 90 126 162
Number of RIS elements K

Figure 5.4: Impact of the number of RIS elements on the achievable sum rate with Pt = 10
dBm.

45
Obstacles Start
30
BS1

15 dmin
R
Uc 1
y (m)

-15

Uf
-30
BS2 Uc 2 Hover
-45
-45 -30 -15 0 15 30 45
x (m)

Figure 5.5: Top view of the UAV trajectory obtained by the MO-PPO algorithm sampled
every 25 time slots and averaged over 10 evaluation episodes

advantages of incorporating CoMP, RIS, and NOMA techniques to enhance spectral

efficiency across all power levels. Additionally, we benchmark the hover PPO (H-
PPO) algorithm, which maintains a fixed UAV position at the center of the user
clusters, against the proposed MO-PPO algorithm, and highlight the improvement
in network performance achieved due to the trajectory optimization.

54
 Chapter 5. Deep Reinforcement Learning for Intelligent NOMA Networks

5.7 MO-PPO and RIS Elements

Our investigation extends to analyzing the impact of the number of RIS elements on
the network’s achievable sum rate, providing further insights into the performance
of the MO-PPO algorithm Figure 5.4 illustrates the positive correlation between the
sum rate and the number of RIS elements, emphasizing the advantages of utilizing
a larger RIS to enhance network performance. However, we observe a subtle, yet
noteworthy trend: the difference in sum rate between the exhaustive search baseline
and the MO-PPO algorithm, while remaining small, increases with the number of
RIS elements. This observation highlights the importance of carefully considering
the trade-off between performance and complexity when determining the optimal
number of operational RIS elements, a challenge we aim to address in future work.

5.8 UAV Trajectory

Finally, we visualize the UAV trajectory generated by the MO-PPO algorithm in
Figure 5.5. It is observed that the UAV adopts a cautious approach, navigating
around obstacles while minimizing its distance to Uf . Such a trend is commonly ob-
served in DRL algorithms that operate on the principle of exploration-exploitation.
The agent learns to strike a balance between exploring the environment and ex-
ploiting its current knowledge to maximize the cumulative reward. The generated
trajectory further highlights the agent’s ability to adapt to the dynamic environment
and optimize network performance by effectively leveraging both RIS and NOMA
techniques.

55
Chapter 6

Conclusion

Our research explores the potential of utilizing reconfigurable intelligent surfaces

(RIS), specifically STAR-RIS and aerial RIS (ARIS), within CoMP-NOMA networks
to enhance performance and energy efficiency. Across four papers, we investigate
various aspects of network design, optimization, and performance analysis.

6.1 Summary
This thesis has investigated the potential of integrating STAR-RIS, CoMP, and
NOMA technologies to address the challenges of future wireless communication
systems. The research explored various aspects of these technologies, including
performance analysis, optimization strategies, and practical implementation consid-
erations.

6.1.1 Performance Analysis

A novel framework was introduced for integrating STAR-RIS into CoMP-NOMA
multi-cell networks, highlighting the unique ability of STAR-RIS to serve multiple
cells simultaneously. The impact of STAR-RIS on key performance metrics such as
achievable rates and outage probability was analyzed, demonstrating its superiority
over conventional systems in improving coverage and performance, particularly for
cell-edge users. Additionally, the thesis developed a tractable analytical framework
to evaluate the performance of STAR-RIS-assisted CoMP-NOMA networks under
Nakagami-m fading channels, deriving closed-form expressions for ergodic rate and
outage probability for each user.

56
 Chapter 6. Conclusion

6.1.2 Optimization Strategies

The research investigated various optimization strategies for maximizing network
performance. This included exploring the impact of CoMP cooperation, STAR-
RIS element allocation to base stations based on channel conditions, and ampli-
tude adjustments for transmission and reflection on achieving optimal network sum-
rate. Furthermore, the thesis explored energy-efficient design approaches for CoMP-
NOMA networks incorporating RIS, proposing different RIS configurations and op-
timization algorithms for maximizing energy efficiency while maintaining desired
performance levels.

6.1.3 Deep Reinforcement Learning

The application of Deep Reinforcement Learning (DRL) techniques was explored
for joint optimization of UAV trajectory, RIS phase shifts, and NOMA power con-
trol in aerial RIS-assisted CoMP-NOMA networks. The proposed DRL framework
demonstrated effectiveness in maximizing network sum rate while satisfying user
QoS constraints, highlighting the potential of combining CoMP-NOMA and RIS in
UAV-assisted networks for improved spectral efficiency and coverage.

6.2 Limitations and Future Work

While this thesis has made significant contributions to the understanding of RIS-
assisted CoMP-NOMA networks, there are limitations that pave the way for future
research:

• Limited Scope of Analysis: The analysis primarily focused on achievable

rates and outage probability, neglecting other crucial aspects such as energy
efficiency, spectral efficiency, and user fairness.

• Simplified Assumptions: The research often relied on static scenarios and

simplified channel models, which may not fully capture the dynamic nature
and complexities of real-world networks.

• Specific Technology Implementations: The analysis focused on specific

RIS configurations and DRL algorithms, potentially overlooking other promis-
ing options for optimizing performance.

Future research directions include:

• Investigating more realistic and dynamic scenarios: This includes con-

sidering user mobility, complex channel models, and dynamic interference en-

57
 Chapter 6. Conclusion

vironments.

• Expanding performance metrics: Future work should analyze a wider

range of metrics, including energy efficiency, spectral efficiency, user fairness,
and delay, to provide a more comprehensive evaluation of network perfor-
mance.

• Exploring distributed optimization algorithms: Developing distributed

optimization algorithms for RIS control is crucial for scalability and robustness
in large-scale networks.

• Addressing practical implementation aspects: Further research is re-

quired to address challenges related to hardware limitations, channel estima-
tion, and standardization of RIS technology.

• Incorporating machine learning techniques: Utilizing machine learning

for tasks such as channel prediction, resource allocation, and interference man-
agement can further enhance the performance and adaptability of RIS-assisted
CoMP-NOMA networks.

By addressing these limitations and pursuing these future research directions, we

can continue to advance the development of RIS-assisted CoMP-NOMA networks
and unlock their full potential for shaping the future of wireless communication.

58
References

[1] Y. Liu, Z. Qin, M. Elkashlan, Z. Ding, A. Nallanathan, and L. Hanzo, “Non-

orthogonal multiple access for 5g and beyond,” 2018.

[2] S. M. R. Islam, N. Avazov, O. A. Dobre, and K.-s. Kwak, “Power-domain non-

orthogonal multiple access (noma) in 5g systems: Potentials and challenges,”
IEEE Communications Surveys & Tutorials, vol. 19, no. 2, p. 721–742, 2017.

[3] D. Lee, H. Seo, B. Clerckx, E. Hardouin, D. Mazzarese, S. Nagata, and

K. Sayana, “Coordinated multipoint transmission and reception in lte-
advanced: deployment scenarios and operational challenges,” IEEE Commu-
nications Magazine, vol. 50, no. 2, pp. 148–155, 2012.

[4] G. Nigam, P. Minero, and M. Haenggi, “Coordinated multipoint joint trans-

mission in heterogeneous networks,” IEEE Transactions on Communications,
vol. 62, no. 11, pp. 4134–4146, 2014.

[5] C. Huang, A. Zappone, G. C. Alexandropoulos, M. Debbah, and C. Yuen,

“Reconfigurable intelligent surfaces for energy efficiency in wireless communi-
cation,” IEEE Trans. Wirel. Commun., vol. 18, no. 8, pp. 4157–4170, 2019.

[6] Q. Wu and R. Zhang, “Towards smart and reconfigurable environment: Intel-

ligent reflecting surface aided wireless network,” IEEE communications maga-
zine, vol. 58, no. 1, pp. 106–112, 2019.

[7] L. Zhang, H. Zhao, S. Hou, Z. Zhao, H. Xu, X. Wu, Q. Wu, and R. Zhang,
“A survey on 5g millimeter wave communications for uav-assisted wireless net-
works,” IEEE Access, vol. 7, pp. 117460–117504, 2019.

[8] X. Liu, Y. Liu, Y. Chen, and L. Hanzo, “Trajectory design and power control
for multi-uav assisted wireless networks: A machine learning approach,” IEEE
Transactions on Vehicular Technology, vol. 68, no. 8, pp. 7957–7969, 2019.

[9] T. Hou, J. Wang, Y. Liu, X. Sun, A. Li, and B. Ai, “A joint design for STAR-
RIS enhanced NOMA-CoMP networks: A simultaneous-signal-enhancement-

59
 References

and-cancellation-based (SSECB) design,” IEEE Trans. Veh. Technol., vol. 71,

no. 1, pp. 1043–1048, 2021.

[10] S. Xu, C. Liu, H. Wang, M. Qian, and J. Li, “On secrecy performance analysis
of multi-antenna star-ris-assisted downlink noma systems,” EURASIP Journal
on Advances in Signal Processing, vol. 2022, no. 1, p. 118, 2022.

[11] J. Chen, K. Zhai, Z. Wang, Y. Liu, J. Jia, and X. Wang, “Comp and ris-
assisted multicast transmission in a multi-uav communication system,” IEEE
Transactions on Communications, 2024.

[12] H. Mei, K. Yang, Q. Liu, and K. Wang, “3d-trajectory and phase-shift design for
ris-assisted uav systems using deep reinforcement learning,” IEEE Transactions
on Vehicular Technology, vol. 71, no. 3, pp. 3020–3029, 2022.

[13] H. Peng and L.-C. Wang, “Energy harvesting reconfigurable intelligent surface
for uav based on robust deep reinforcement learning,” IEEE Transactions on
Wireless Communications, vol. 22, no. 10, pp. 6826–6838, 2023.

[14] A. B. Adam, X. Wan, and Z. Wang, “Energy efficiency maximization in down-

link multi-cell multi-carrier noma networks with hardware impairments,” IEEE
Access, vol. 8, pp. 210054–210065, 2020.

[15] H. Long, M. Chen, Z. Yang, Z. Li, B. Wang, X. Yun, and M. Shikh-Bahaei,

“Joint trajectory and passive beamforming design for secure uav networks with
ris,” in 2020 IEEE Globecom Workshops (GC Wkshps, pp. 1–6, IEEE, 2020.

[16] L. Wei, C. Huang, G. C. Alexandropoulos, C. Yuen, Z. Zhang, and M. Debbah,

“Channel estimation for RIS-empowered multi-user MISO wireless communi-
cations,” IEEE Trans. Commun., vol. 69, no. 6, pp. 4144–4157, 2021.

[17] A. Taha, M. Alrabeiah, and A. Alkhateeb, “Enabling large intelligent surfaces

with compressive sensing and deep learning,” IEEE access, vol. 9, pp. 44304–
44321, 2021.

[18] Y. Guo, Z. Qin, Y. Liu, and N. Al-Dhahir, “Intelligent reflecting surface aided
multiple access over fading channels,” IEEE Transactions on Communications,
vol. 69, no. 3, pp. 2015–2027, 2020.

[19] X. Mu, Y. Liu, L. Guo, J. Lin, and R. Schober, “Simultaneously transmitting

and reflecting (star) ris aided wireless communications,” IEEE Transactions on
Wireless Communications, vol. 21, no. 5, pp. 3083–3098, 2021.

60
 References

[20] J. Xu, Y. Liu, X. Mu, and O. A. Dobre, “Star-riss: Simultaneous transmitting

and reflecting reconfigurable intelligent surfaces,” IEEE Commun. Lett., vol. 25,
no. 9, pp. 3134–3138, 2021.

[21] Y. Saito, Y. Kishiyama, A. Benjebbour, T. Nakamura, A. Li, and K. Higuchi,

“Non-orthogonal multiple access (noma) for cellular future radio access,” in
2013 IEEE 77th vehicular technology conference (VTC Spring), pp. 1–5, IEEE,
2013.

[22] M. Obeed, H. Dahrouj, A. M. Salhab, S. A. Zummo, and M.-S. Alouini, “User

pairing, link selection, and power allocation for cooperative NOMA hybrid
VLC/RF systems,” IEEE Trans. Wirel. Commun., vol. 20, no. 3, pp. 1785–
1800, 2020.

[23] A. Salem and L. Musavian, “Noma in cooperative communication systems with

energy-harvesting nodes and wireless secure transmission,” IEEE Transactions
on Wireless Communications, vol. 20, no. 2, pp. 1023–1037, 2020.

[24] R. Tanbourgi, S. Singh, J. G. Andrews, and F. K. Jondral, “A tractable model

for noncoherent joint-transmission base station cooperation,” IEEE Trans.
Wirel. Commun., vol. 13, no. 9, pp. 4959–4973, 2014.

[25] M. Elhattab, M. A. Arfaoui, C. Assi, and A. Ghrayeb, “RIS-assisted joint

transmission in a two-cell downlink NOMA cellular system,” IEEE J. Sel. Areas
Commun., vol. 40, no. 4, pp. 1270–1286, 2022.

[26] X. Yue, J. Xie, Y. Liu, Z. Han, R. Liu, and Z. Ding, “Simultaneously transmit-
ting and reflecting reconfigurable intelligent surface assisted noma networks,”
IEEE Transactions on Wireless Communications, vol. 22, no. 1, pp. 189–204,
2023.

[27] B. Tahir, S. Schwarz, and M. Rupp, “Analysis of uplink IRS-assisted NOMA

under Nakagami-m fading via moments matching,” IEEE Wireless Commun.
Lett., vol. 10, no. 3, pp. 624–628, 2021.

[28] V. S. Adamchik and O. Marichev, “The algorithm for calculating integrals of

hypergeometric type functions and its realization in reduce system,” in Proceed-
ings of the international symposium on Symbolic and algebraic computation,
pp. 212–224, 1990.

[29] Z. Ding and H. V. Poor, “A simple design of IRS-NOMA transmission,” IEEE

Commun. Lett., vol. 24, no. 5, pp. 1119–1123, 2020.

61
 References

[30] B. Zheng, C. You, W. Mei, and R. Zhang, “A survey on channel estima-

tion and practical passive beamforming design for intelligent reflecting surface
aided wireless communications,” IEEE Commun. Surv. Tutor., vol. 24, no. 2,
pp. 1035–1071, 2022.

[31] E. Shtaiwi, H. Zhang, S. Vishwanath, M. Youssef, A. Abdelhadi, and Z. Han,

“Channel estimation approach for RIS assisted MIMO systems,” IEEE Trans.
Cogn. Commun. Netw., vol. 7, no. 2, pp. 452–465, 2021.

[32] G. Zhou, C. Pan, H. Ren, P. Popovski, and A. L. Swindlehurst, “Channel esti-

mation for RIS-aided multiuser millimeter-wave systems,” IEEE Trans. Signal
Process., vol. 70, pp. 1478–1492, 2022.

[33] Q. Wu and R. Zhang, “Intelligent reflecting surface enhanced wireless net-

work via joint active and passive beamforming,” IEEE Trans. Wirel. Commun.,
vol. 18, no. 11, pp. 5394–5409, 2019.

[34] J. Zhao, L. Yu, K. Cai, Y. Zhu, and Z. Han, “RIS-aided ground-aerial NOMA
communications: A distributionally robust DRL approach,” IEEE J. Sel. Areas
Commun., vol. 40, no. 4, pp. 1287–1301, 2022.

[35] I. Budhiraja, V. Vishnoi, N. Kumar, D. Garg, and S. Tyagi, “Energy-efficient

optimization scheme for RIS-assisted communication underlaying UAV with
NOMA,” in ICC 2022 - IEEE Int. Conf. Commun., pp. 1–6, IEEE, 2022.

[36] S. Lv, X. Xu, S. Han, and P. Zhang, “UAV-RIS assisted coordinated multipoint
finite blocklength transmission for MTC networks,” IEEE Internet Things J.,
2023.

[37] D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” arXiv
preprint arXiv:1412.6980, 2014.

[38] J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov, “Proximal

policy optimization algorithms,” arXiv preprint arXiv:1707.06347, 2017.

[39] V. Mnih, A. P. Badia, M. Mirza, A. Graves, T. Lillicrap, T. Harley, D. Silver,

and K. Kavukcuoglu, “Asynchronous methods for deep reinforcement learning,”
in Int. Conf. Mach. Learn., pp. 1928–1937, PMLR, 2016.

[40] U. Challita, W. Saad, and C. Bettstetter, “Interference management for

cellular-connected UAVs: A deep reinforcement learning approach,” IEEE
Trans. Wirel. Commun., vol. 18, no. 4, pp. 2125–2140, 2019.

62

View publication stats