Resource Management and

Optimization for Cognitive Radio

Networks

Jie Xiang

Doctoral Dissertation

Submitted to
the Faculty of Mathematics and Natural Sciences
at the University of Oslo
in partial fulfillment of the requirements for
the degree of Philosophiae Doctor

September 2011
II
This thesis work was carried out at Simula Research Laboratory
under the supervision of Dr. Yan Zhang, Prof. Olav Lysne and
Prof. Tor Skeie. Financial support is from the Resilient Wireless
Networks project at Simula (2007.9-2010.9).
To my family
Abstract

Wireless technologies have been greatly developed and used in the last decade,
such as Wi-Fi, Zigbee, Bluetooth, 3G, and 4G networks. The appearance
of new technologies leads to an increased request of the wireless spectrum.
However, the frequency bands for wireless communications are limited and
in danger of being exhausted owing to the fixed allocation regulation. Fortu-
nately, recent measurement campaign on spectrum usage showed that spec-
trum is not efficiently used by the licensed systems. For example, the recent
results released by FCC in US show that on average only 5% of the spectrum
from 30MHz to 30GHz is used. Cognitive radio, which can sense spectrum
usage, identify and intelligently access spectrum bands licensed to primary
systems, is thus a good candidate to improve spectrum utilization and system
performance.
In this thesis, we explore the resource management and optimization
problems for cognitive radio wireless networks including both one-hop and
multi-hop cases. In the one-hop case, we studied the cognitive radio cellu-
lar networks (CogCell) and femtocell networks (CogFem), while in multi-hop
case, we studied the cognitive radio mesh networks (CogMesh).
Firstly, we studied CogCell, where the cognitive radio enabled base sta-
tion is operated to provide service to secondary users (SUs) with the co-
existence of primary users (PUs). We investigated the uplink admission
and power control problem aiming to maximize the revenue received by op-
erators while guaranteeing interference constraints on PUs and Signal-to-
Interference-and-Noise-Ratio (SINR) requirements for SUs. We formulated
it as an instance of multidimensional knapsack problem in the one-channel
case, and as an instance of multidimensional multiple knapsack problem
in the multiple-channel case, respectively. Furthermore, we proposed low-
complexity heuristic algorithms which can achieve much more revenue than
the existing schemes, and are close to the optimal results obtained by MOSEK
optimization software.
Secondly, we proposed a radically new communication paradigm by in-
corporating cognitive radio in femtocell networks, where the cognitive radio

V
enabled femtocell base station (FBS) can opportunistically use the spectrum
from licensed systems, support all kinds of indoor communications, and im-
prove the quality of service for macrocell networks. We investigated the
downlink spectrum sharing and power allocation problem for CogFem aim-
ing to maximize the downlink capacity of each FBS while considering the
constraints on SINR measured by SUs and transmission power of FBS. We
employed a mixed primal and dual decomposition method, and proposed a
joint channel allocation and fast power control scheme to solve the problem.
Simulation results showed that CogFem with more spectrum opportunity
could achieve much higher capacity than normal femtocells. The proposed
channel allocation and power control scheme can converge very fast, achieve
much higher average capacity and lower user blocking rate than the tradi-
tional coloring method. We also found that even use the fixed power control
scheme together with our proposed channel allocation scheme, the capacity
is sacrificed only 2% comparing with dedicated power control schemes.
Finally, we studied CogMesh, where the cognitive radio enabled secondary
mesh routers (SMRs) can opportunistically utilize the primary licensed spec-
trum to deliver data from secondary mesh users. We investigated the route
and channel selection problem in CogMesh aiming to maximize the route
availability, while guaranteeing the end-to-end delay from SMRs to the gate-
way. We formulated it as a non-liner integer programming problem, and
transformed to a linear integer programming problem, which is further mod-
eled as a variant of multiple-choice knapsack problem. Then, we proposed
a low-complexity heuristic algorithm to solve it. Simulation results showed
that our proposed scheme achieve quite close successful solution ratio and
route availability to the results from MOSEK, and outperforms the channel
selection schemes based on best SINR and best channel availability schemes.
As a conclusion, we believe that our proposed resource management and
optimization schemes lay down a solid foundation for building cognitive radio
networks in future to efficiently use the invaluable spectrum resource.

VI
Acknowledgments

This dissertation could never have been written without help and support
from many people around me.
First of all, I would like to thank my supervisors Dr. Yan Zhang, Prof. Tor
Skeie and Prof. Olav Lysne for hiring me as a PhD student in Simula Research
Laboratory. I would also like to thank my master supervisor Dr. Jianhua He
for his continuous advising on my PhD research. I am greatly indebted to
Dr. Yan Zhang, whose guidance and stimulating suggestions helped me a
lot in all the time of my research. I am genuinely thankful to Prof. Tor
Skeie, who not only constantly fueled me with strength, hope, and support
to take on challenges and go through frustrations, but also hired me as a
research engineer after my PhD funding ended. Specially, I would like to
thank Dr. Frank Olaf Sem-Jacobsen and Dr. Amund Kvalbein for assigning
me not so heavy tasks in NaNoC and eValg projects that I could finish my
thesis writing.
I would like to thank Simula for providing me such a good working envi-
ronment, where I have got a lot of help from IT support and administration
group, and learnt a lot from many smart researchers and students in different
research groups.
I would also like to thank my collaborators during my PhD research.
They are Lang Xie, Dr. Qin Xin, Hai Ngoc Pham and Sabita Maharjan. We
had a lot of valuable discussions and glad cooperations. I would like to thank
Prof. Stein Gjessing and Prof. Paal E. Engelstad for their valuable suggestions
during cross project meetings. I am also grateful to Ralph Lorentzen for his
valuable suggestions on my optimization problems.
Special thank goes to Ernst Gunnar, my former officemate, whose special
humor made our office joyful. He was always helpful, and taught me the Nor-
wegian culture and working methods at the beginning of my life in Norway.
I also would like to thank my current officemate Ahmed Elmokashfi, who
helped me a lot during my thesis writing. I would like to thank Åshild Grøn-
stad Solheim and Halvard Moe for our perfect cooperation in eValg project.
I would like to thank many people in the former ND department. They are

VII
Audun Fosselie Hansen, Knut-Helge Vik, Ole Kristoffer Apeland, Sven-Arne
Reinemo, and Thomas Sødring. Special thank goes to Prof. Carsten Griwodz
who provided enough financial support for ND cakes when I was serving as
the “cake man”.
During my PhD study, I visited Hong Kong University of Science and
Technology and Princeton University, where I made many good friends.
Thanks Prof. Qian Zhang and Prof. Mung Chiang for enrolling me in their
project meetings and discussions. Special thank goes to Prof. Mung Chiang
for his instruction on optimization theory for wireless communications.
Finally, I owe an enormous debt of gratitude to my wife, my brother, and
my parents. Their unconditional love, understanding, and support are the
origin of the power for me to keep on moving in my research.

VIII
Contents

List of Figures XV

List of Tables XVII

List of Algorithms XIX

1 Introduction 1
1.1 Motivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 Cognitive radio cellular networks . . . . . . . . . . . . 3
1.1.2 Cognitive radio femtocell networks . . . . . . . . . . . 5
1.1.3 Cognitive radio mesh networks . . . . . . . . . . . . . . 7
1.2 Research methods . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.1 Research methods for computer science . . . . . . . . . 8
1.2.2 Research methods in this thesis . . . . . . . . . . . . . 9
1.3 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.1 Admission and power control for one-channel cognitive
radio cellular networks . . . . . . . . . . . . . . . . . . 10
1.3.2 Channel allocation with admission and power control
for multi-channel cognitive radio cellular networks . . . 11
1.3.3 Channel allocation and power control for cognitive ra-
dio femtocell networks . . . . . . . . . . . . . . . . . . 11
1.3.4 Channel selection for cognitive radio mesh networks . . 12
1.4 Thesis organization . . . . . . . . . . . . . . . . . . . . . . . . 12

2 Background and Related Work 15

2.1 Background of cognitive radio networks . . . . . . . . . . . . . 15
2.1.1 Definition of cognitive radio . . . . . . . . . . . . . . . 15
2.1.2 Cognitive cycle and key technologies . . . . . . . . . . 16
2.1.3 Deployment challenges . . . . . . . . . . . . . . . . . . 18
2.2 Research problems in our work . . . . . . . . . . . . . . . . . 21
2.2.1 Power control . . . . . . . . . . . . . . . . . . . . . . . 21

IX
 CONTENTS

2.2.2 Admission control . . . . . . . . . . . . . . . . . . . . . 22

2.2.3 Channel allocation . . . . . . . . . . . . . . . . . . . . 22
2.3 Related work . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3.1 Resource management in cognitive radio cellular net-
works . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3.2 Spectrum sharing in cognitive radio femtocell networks 24
2.3.3 Channel selection in cognitive radio mesh networks . . 25

3 Power and Admission Control for One-channel Cognitive Ra-

dio Cellular Networks 27
3.1 System model and problem formulation . . . . . . . . . . . . . 28
3.1.1 System model . . . . . . . . . . . . . . . . . . . . . . . 28
3.1.2 Interference power . . . . . . . . . . . . . . . . . . . . 29
3.1.3 QoS definition and requirement . . . . . . . . . . . . . 30
3.1.4 The operator problem . . . . . . . . . . . . . . . . . . 31
3.2 Joint admission and power control schemes . . . . . . . . . . . 32
3.2.1 JAPC-MRER . . . . . . . . . . . . . . . . . . . . . . . 32
3.2.2 JAPC-MSRA . . . . . . . . . . . . . . . . . . . . . . . 34
3.2.3 JAPC-Rand . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2.4 Simulation results and analysis . . . . . . . . . . . . . 36
3.3 Discussions on power control and pre-admission schemes . . . 40
3.4 Further improvements . . . . . . . . . . . . . . . . . . . . . . 45
3.4.1 Lower and upper bounds . . . . . . . . . . . . . . . . . 47
3.4.2 Multidimensional knapsack problem modeling and so-
lutions . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.4.3 Proposed efficiency based heuristics method . . . . . . 53
3.5 Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.5.1 Effect of the number of primary users . . . . . . . . . . 55
3.5.2 Effect of the number of secondary users . . . . . . . . . 57
3.5.3 Effect of the interference threshold on primary users . . 58
3.5.4 Effect of the interference to BS from primary systems . 59
3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

4 Resource Optimization for Multi-channel Cognitive Radio

Cellular Networks 61
4.1 System model and problem formulations . . . . . . . . . . . . 61
4.1.1 System model . . . . . . . . . . . . . . . . . . . . . . . 62
4.1.2 Interference from SUs to PUs . . . . . . . . . . . . . . 63
4.1.3 Uplink capacity of SUs . . . . . . . . . . . . . . . . . . 64
4.1.4 The operator problem . . . . . . . . . . . . . . . . . . 64
4.2 MMKP modeling . . . . . . . . . . . . . . . . . . . . . . . . . 65

X
CONTENTS

4.3 MKP modeling by matrix transformation . . . . . . . . . . . . 67

4.4 Proposed heuristic algorithm to MKP . . . . . . . . . . . . . . 70
4.5 Best SINR channel selection scheme . . . . . . . . . . . . . . . 75
4.6 Simulation results and analysis . . . . . . . . . . . . . . . . . 75
4.7 Conclusions and discussions . . . . . . . . . . . . . . . . . . . 84
4.7.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . 84
4.7.2 Discussions . . . . . . . . . . . . . . . . . . . . . . . . 85

5 Spectrum Sharing in Cognitive Radio Femtocell Networks 87

5.1 System model and assumptions . . . . . . . . . . . . . . . . . 88
5.1.1 System initialization . . . . . . . . . . . . . . . . . . . 88
5.1.2 Number of transceivers . . . . . . . . . . . . . . . . . . 88
5.1.3 Spectrum sensing and primary system protection . . . 89
5.1.4 Control channel . . . . . . . . . . . . . . . . . . . . . . 90
5.1.5 Handover between macrocell and femtocell . . . . . . . 90
5.1.6 Deployment example . . . . . . . . . . . . . . . . . . . 91
5.2 Problem formulations . . . . . . . . . . . . . . . . . . . . . . . 92
5.2.1 Channel model . . . . . . . . . . . . . . . . . . . . . . 92
5.2.2 Downlink capacity . . . . . . . . . . . . . . . . . . . . 92
5.2.3 Downlink spectrum sharing problem . . . . . . . . . . 93
5.3 Problem decompositions and solutions . . . . . . . . . . . . . 94
5.3.1 The master problem . . . . . . . . . . . . . . . . . . . 94
5.3.2 Subproblems . . . . . . . . . . . . . . . . . . . . . . . 95
5.4 Simulation results and discussion . . . . . . . . . . . . . . . . 101
5.4.1 Existing schemes . . . . . . . . . . . . . . . . . . . . . 102
5.4.2 Average capacity with different density of apartments . 103
5.4.3 System performance with different number of available
channels . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.4.4 The convergence . . . . . . . . . . . . . . . . . . . . . 106
5.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

6 QoS-aware Spectrum Access for Cognitive Radio Mesh Net-

works 109
6.1 System model and assumptions . . . . . . . . . . . . . . . . . 110
6.1.1 System model . . . . . . . . . . . . . . . . . . . . . . . 110
6.1.2 Channel model and adaptive modulation coding . . . . 111
6.1.3 Interference-avoid channel selection for adjacent links . 113
6.1.4 End to end delay . . . . . . . . . . . . . . . . . . . . . 113
6.1.5 Route availability . . . . . . . . . . . . . . . . . . . . . 115
6.2 Problem formulation . . . . . . . . . . . . . . . . . . . . . . . 116
6.2.1 Formulation of route and channel selection . . . . . . . 116

XI
 CONTENTS

6.2.2 Problem transformation . . . . . . . . . . . . . . . . . 117

6.3 Solutions from Lagrangian relaxations . . . . . . . . . . . . . . 119
6.4 Low-complexity heuristic channel selection schemes . . . . . . 121
6.4.1 Proposed channel selection scheme . . . . . . . . . . . 122
6.4.2 Alternative channel selection schemes . . . . . . . . . . 122
6.5 Matrix transformation for problem solving by optimization
software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
6.6 Simulation results and analysis . . . . . . . . . . . . . . . . . 125
6.6.1 Successful solution ratio . . . . . . . . . . . . . . . . . 126
6.6.2 Route availability . . . . . . . . . . . . . . . . . . . . . 128
6.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

7 Conclusion and Future Work 131

7.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
7.1.1 On research scenarios . . . . . . . . . . . . . . . . . . . 131
7.1.2 Comparison . . . . . . . . . . . . . . . . . . . . . . . . 133
7.1.3 On research methods . . . . . . . . . . . . . . . . . . . 133
7.2 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

Bibliography 136

Appendix A Publication list 149

Appendix B Acronym List 153

XII
List of Figures

1.1 The spectrum allocation table from USA [1] . . . . . . . . . . 2

1.2 An illustration of the coexistence between cognitive radio users
and and primary systems in cognitive radio cellular networks . 4
1.3 An illustration of the coexistence between cognitive radio fem-
tocells and primary systems such as macrocells and TV systems 6
1.4 An example of cognitive radio mesh network . . . . . . . . . . 7

2.1 Cognitive radio operation cycle [2] . . . . . . . . . . . . . . . . 16

3.1 System model of one-channel cognitive radio cellular networks 28

3.2 The secondary revenue in terms of the number of PUs . . . . . 38
3.3 The secondary revenue in terms of the number of SUs . . . . . 39
3.4 The secondary revenue in terms of the interference threshold . 40
3.5 The flow chart of joint admission and power control schemes
for one-channel CogCell . . . . . . . . . . . . . . . . . . . . . 43
3.6 Secondary Revenue in terms of number of PUs (ns = 50, Γ =
−100dBW , and Ip = −110dBW ) . . . . . . . . . . . . . . . . 56
3.7 Secondary revenue in terms of number of SUs (np = 50, Γ =
−100dBW , and Ip = −110dBW ) . . . . . . . . . . . . . . . . 57
3.8 Secondary Revenue in terms of Γ (ns = 50, np = 50, and
Ip = −110dBW ) . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.9 Secondary Revenue in terms of Ip (ns = 50, np = 50, and
Γ = −100dBW ) . . . . . . . . . . . . . . . . . . . . . . . . . . 59

4.1 System model of multi-channel cognitive radio cellular networks 62

4.2 Heuristic Algorithm to MKP . . . . . . . . . . . . . . . . . . . 74
4.3 Revenue in terms of number of channels (nmax pc = 10, Γ =
−70dBm, Ip = −80dBm, and ns = 100) . . . . . . . . . . . . 77
4.4 Percentage of optimal solution from MOSEK in terms of num-
ber of channels (nmax
pc = 10, Γ = −70dBm, Ip = −80dBm, and
ns = 100) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

XIII
 LIST OF FIGURES

4.5 Revenue in terms of number of SUs (nmax pc = 10, Γ = −70dBm,

Ip = −80dBm, and nc = 10) . . . . . . . . . . . . . . . . . . 79
4.6 Percentage of optimal solution from MOSEK in terms of num-
ber of SUs (nmax
pc = 10, Γ = −70dBm, Ip = −80dBm, and
nc = 10) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
4.7 Revenue in terms of different number of PUs per channel (Γ =
−70dBm, Ip = −80dBm, nSU = 100, and nc = 10) . . . . . . 80
4.8 Percentage of optimal solution from MOSEK in terms of num-
ber of maximum PUs per channel (nmax pc = 10, Γ = −70dBm,
Ip = −80dBm, and nc = 10) . . . . . . . . . . . . . . . . . . 81
4.9 Revenue in terms of different interference to BS from each
PU transmitter (nmaxpc = 10, Γ = −70dBm, nSU = 100, and
nc = 10) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.10 Percentage of optimal solution from MOSEK in terms of differ-
ent interference to BS from each PU transmitter (nmax pc = 10,
Γ = −70dBm, nSU = 100, and nc = 10) . . . . . . . . . . . . 82
4.11 Revenue in terms of different interference level per PU (nmax
pc =
10, Ip = −80dBm, nSU = 100, and nc = 10) . . . . . . . . . . 83
4.12 Percentage of optimal solution from MOSEK in terms of dif-
ferent interference level per PU (nmax
pc = 10, Ip = −80dBm,
nSU = 100, and nc = 10) . . . . . . . . . . . . . . . . . . . . . 84

5.1 An example of deploying CogFem networks with FBS controller 91

5.2 The decomposition of spectrum sharing problems in CogFem . 99
5.3 An illustration of the simulation scenario . . . . . . . . . . . . 102
5.4 Average capacity per femtocell in terms of number of floors in
each building. (3 rows, and 5 buildings per row) . . . . . . . . 103
5.5 Average capacity per femtocell in terms of number of buildings
in each row. (3 rows, and 5 floors per building) . . . . . . . . 104
5.6 Average capacity per femtocell in terms of number of rows of
buildings. (5 floors per building, and 5 buildings per row) . . . 104
5.7 Average capacity per femtocell in terms of number of channels.
(5 rows, 5 buildings per row, 10 floors per building) . . . . . . 105
5.8 Average blocking rate in terms of number of channels. (5 rows,
5 buildings per row, 10 floors per building) . . . . . . . . . . . 106
5.9 Average capacity per femtocell in terms of number of Itera-
tions. (10 available channels, 10 floors per building, 5 build-
ings per row, and 5 rows) . . . . . . . . . . . . . . . . . . . . . 107
5.10 Average user blocking rate in terms of number of Iterations.
(10 available channels, 10 floors per building, 5 buildings per
row, and 5 rows) . . . . . . . . . . . . . . . . . . . . . . . . . 108

XIV
LIST OF FIGURES

6.1 System model for route and channel selection in cognitive radio
mesh networks . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.2 An illustration of interfered links . . . . . . . . . . . . . . . . 113
6.3 Successful solution ratio from different channel selection schemes
(3D visualization) . . . . . . . . . . . . . . . . . . . . . . . . . 127
6.4 Successful solution ratio from different channel selection schemes128
6.5 Route availability from different channel selection schemes (3D
visualization) . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
6.6 Route availability from different channel selection schemes . . 130

XV
List of Tables

1.1 Scenarios addressed in each chapter . . . . . . . . . . . . . . . 13

2.1 Related works in cognitive radio cellular networks . . . . . . . 24

3.1 Table of notations for one-channel CogCell . . . . . . . . . . . 29

3.2 Simulation parameters for admission control in CogCell . . . . 37
3.3 Revenue allocation table with DTR and SINR mapping . . . . 37
3.4 Joint admission and power control schemes for one-channel
CogCell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

4.1 Table of notations for multi-channel CogCell . . . . . . . . . . 63

4.2 Simulation parameters for Multi-channel CogCell . . . . . . . 76
4.3 Revenue allocation table for Multi-channel CogCell . . . . . . 76

5.1 Table of notations for cognitive radio femtocell networks . . . 89

6.1 Table of notations for cognitive radio mesh networks . . . . . 111

6.2 Simulation parameters for CogMesh . . . . . . . . . . . . . . . 126
6.3 AMC code rate and SINR table for CogMesh . . . . . . . . . . 126

7.1 Summary of different scenarios . . . . . . . . . . . . . . . . . . 134

XVII
List of Algorithms

1 JAPC-MRER . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2 JAPC-MSRA . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3 JAPC-MKP . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4 Proposed MKP Heuristic Algorithm for MMKP: Part 1 . . . . 72
5 Proposed MKP Heuristic Algorithm for MMKP: Part 2 . . . . 73
6 Channel allocation algorithm for the master problem . . . . . 95
7 Power control algorithms for any FBS i . . . . . . . . . . . . . 98
8 Proposed joint channel allocation and fast power control algo-
rithm for FBS i . . . . . . . . . . . . . . . . . . . . . . . . . . 101
9 Channel selection scheme for route r with Lagrangian Methods 121
10 Proposed heuristic channel selection algorithm for a given route
r in CogMesh . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

XIX
Chapter 1

Introduction

Simplicity is the ultimate sophistication

— Leonardo da Vinci (1452- 1519)

During the last decade, wireless communication and networks have been
greatly developed including third generation (3G), fourth generation (4G) cel-
lular networks, IEEE 802.11 Wireless Local Area Networks (WLANs), IEEE
802.15.4/ZigBee Wireless Personal Area Networks (WPANs), Bluetooth, etc.
The radio spectrum ranging from 3KHz to 300GHz is the basic resource to
carry data in wireless networks. In each country, spectrum is regulated by
its radio regulatory agency, such as Federal Communications Commission
(FCC) in USA [3], Electronic Communications Committee (ECC) in Eu-
rope [4], The Norwegian Post and Telecommunications Authority (NPT) in
Norway [5], and Ofcom in UK [6]. Spectrum is traditionally assigned via a
fixed frequency allocation policy. For example, the spectrum allocation table
by FCC is shown in Fig. 1.1, where each portion of spectrum is exclusively
allocated to a specific wireless system, and all subscribers to a wireless system
should be granted to access the exclusive spectrum. Following this approach,
the spectrum resource is in danger of being exhausted. To get a license on a
spectrum band is being more and more difficult and expensive. For example,
to deploy 4G cellular networks, TeliaSonera pays SEK 563 million for the
15-year’s license in Sweden on four frequency blocks totaling 2 × 20 MHz in
the 2.6 GHz band [7], DKK 336.3 million for the 20-year’s license in Denmark
on 2 × 20 MHz paired spectrum and 10 MHz unpaired spectrum in the 2.5
GHz frequency band [8], EUR 819, 000 for the 20-year’s license in Finland
on five 2 × 5 MHz frequency band pairs in the 2.6 GHz band [9].
The Industrial, Scientific and Medical (ISM) spectrum band which is
mostly located around 2.4 and 5 GHz is the only spectrum that can be
shared by different networks. WLANs, WPANs, cordless phones, and even

1
 Introduction

Figure 1.1: The spectrum allocation table from USA [1]

microwave ovens are working in the ISM spectrum band, and suffering the
interference with each other. Thus, the performance of wireless networks
working in ISM spectrum band is highly limited by the coexistence of other
nearby wireless networks.
On the other hand, the licensed spectrum utilization is highly dependent
on the location and time. For instance, during some time periods in a cer-
tain geographic area, the allocated spectrum bands may be seldom used. In
November 2002, FCC published a report to indicate that for 90% of the time
many licensed frequency bands remain unused [10]. Furthermore, Shared
Spectrum Company (SCC) has published a bunch of spectrum measurement
results of US and some Europe Countries since 2004 [11]. From their spec-
trum reports in [12] [13] [14], we can see the utilization of many licensed
frequency band in many cities is less than 25%. This means that it is not an
actual spectrum scarcity that is worrisome, but rather the inefficient spec-
trum usage.
As a result, since 2004, FCC has recommended to consider authorizing
new devices in the TV broadcast spectrum at locations where TV channels
are not being used for authorized services, including broadcast television,
broadcast auxiliary services such as wireless microphones, and private land
mobile radio [15] [16]. The IEEE 802.22 Working Group on Wireless Regional
Area Networks (WRANs) was formed in October 2004, and has been working

2
1.1 Motivations

on the standardization for the rural broadband wireless access using the TV
broadcast spectrum by Cognitive Radio (CR) technologies [17].
The basic idea behind IEEE 802.22 is to exploit the unused or not fully
utilized licensed spectrum, which is called spectrum hole. Actually, this idea
was proposed in the concept of CR by Joseph Mitola III at Royal Institute of
Technology (KTH), Sweden, in 1999 [18]. With CR technologies, secondary
users (SUs) can work with primary users (PUs) in two different modes [19].
One is called underlay mode, where SUs can work on all of the channels if
the interference to the PUs is less than a predefined threshold. The other
one is called overlay mode, where SUs can only work on the channels which
are not occupied by the PUs. Both way can improve the spectrum utilization
significantly and solve the problem of spectrum shortage.
Resource management and optimization is one of the most important is-
sues in CR networks for both underlay and overlay spectrum sharing modes,
where the resource includes the spectrum bands (channels) and transmission
power. How to manage the resource and optimally allocate channels and
control the transmission power for the secondary systems is the main prob-
lem we investigate in this study. In this thesis, we apply CR technologies
in three major types of wireless networks including cellular networks, femto-
cell networks, and mesh networks, and formulate the resource optimization
problems accordingly.

1.1 Motivations
Although spectrum sharing brings opportunities for SUs to access the li-
censed channels, many new challenges come up when deploying CR in prac-
tice. For the application of CR technology, we investigate applications for
both underlay and overlay spectrum sharing modes from one-hop to multi-
hop topologies. In the case of one-hop scenarios, we study the problem of
spectrum sharing in cellular networks and femtocell networks. Regarding the
multi-hop application scenarios, we focus on cognitive radio mesh networks.

1.1.1 Cognitive radio cellular networks

Wireless cellular networks (also known as macrocell networks) have evolved
from 1st generation (1G) to 4G in the last three decades. The 1G mobile
communication system was introduced in the 1980s. It is analog and supports
the analog cell phones with the speeds up to 2.4kbps. The second generation
(2G) system was employed in 1992. It is the first digital communication sys-
tem with the speeds up to 64kbps. The 3G wireless communication systems

3
 Introduction

was employed from 2002 with the speeds up to 2Mbps. The first commercial
4G/LTE (Long Term Evolution) networks have been served by TeliaSonera
at Oslo and Stockholm since the end of 2009, which can achieve the speed up
to 1 Gbps in theory with normal speed of 100 Mbps [20]. However, operators
should pay an expensive license fee for the exclusively usage of the spectrum.
With CR, it is not only possible for operators to deploy cellular networks
without paying such an expensive license fee, but also can improve the sys-
tem performance. We call this kind of network Cognitive Radio Cellular
Networks (CogCell), where the CR-enabled SUs are able to sense the avail-
able spectrum holes, self-configure themselves to best fit with the specific
frequency, control the interference to PUs, and share the spectrum with the
licensed PUs efficiently.

Figure 1.2: An illustration of the coexistence between cognitive radio users and
and primary systems in cognitive radio cellular networks

Figure 1.2 shows a typical example of CogCell. Spectrum sharing brings

us into a great challenge that the SUs activity may cause severe interference
with the PUs specially for primary receivers (PRs). Admitting more SUs
will increase the interference power received by PRs. To obey the coexisting
rule, the interference with PUs from SUs should be not harmful and less than
a predefined threshold. Thus, admission control scheme at the BS plays an
indispensable role in CogCell. Although, the issue of admission control has
been extensively investigated in conventional cellular systems [21], admission
control in a power-controlled network is still an open issue [22]. Moreover,
conventional cellular networks are considerable different from CogCell. In
CogCell, more constraints have to be considered with respect to the admission

4
1.1 Motivations

control problem due to the presence of PUs. This makes our problem much
more complex than the open problem mentioned in [22].
The above problem is not the end of the story, even more challenging
problem follows when we consider multi-channel scenarios, where multiple
channels are available to allocate to the SUs. On each channel, there are
different PUs transmitting and receiving data. SUs which are transmitting
data on a channel will cause interference to the PRs on that channel. How to
allocate channels to SUs and control the transmission power to guarantee the
interference is an essential issue to deploy such a kind of CogCell in practice.

1.1.2 Cognitive radio femtocell networks

In mobile wireless networks, the demand for higher data rates and lower
power consumptions is continuously increasing, while the capacity provided
by the existing macrocell cellular networks is limited. Studies on wireless us-
age have shown that more than 50% voice calls and 70% data traffic originate
indoors [23]. Specificially, in healthcare environments, a recent study found
that 40% of all cellular minutes used by staff were between care providers
within the same building [24]. This phenomenon motivates the research and
development for femtocell networks, which require that each customer installs
a short-range low-cost low-power home base station. These femtocell base
stations (FBSs) can communicate with macrocell networks by a broadband
connection such as Digital Subscriber Line (DSL), cable modem, or a sepa-
rate wireless backhaul channel [25]. Femtocells can provide high data rates
and Quality of Service (QoS) with low transmission power for consumers.
For example, the study in [25] demonstrates that the transmission power can
be saved about 34dB and 77dB in different fading environments. As a result,
network operators may experience less traffic on their expensive macrocell
networks, and can focus their resources on the truly mobile users [25][26].
The spectrum allocated to femtocells is traditionally from the same li-
censed spectrum bands of macrocells, normally operated by the same mobile
network operator. In this case, the capacity of femtocell networks may be
highly limited due to the finite number of licensed spectrum bands and also
the interference with macrocells and other femtocells. It then inspires us
to incorporate the CR technology into femtocell networks, where the CR-
enabled femtocell users (FUs) and FBS can identify and utilize the spectrum
opportunities from the licensed systems such as macrocell networks and TV
broadcast systems as shown in Fig.1.3. In the following, we call this kind of
networks Cognitive Radio Femtocell Networks (CogFem). Besides the spec-
trum agility ability, CogFem has the following features: (a), the number of
users in each femtocell is small, e.g., 2, 4 [25]. (b), the size of the cell coverage

5
 Introduction

is about the house or apartment range, e.g., 100 m2 . (c), the availability of
licensed channels is similar in neighboring cells, this is the major difference
with CR macrocell networks, where the channel availability may vary a lot
between neighboring cells.

Figure 1.3: An illustration of the coexistence between cognitive radio femtocells

and primary systems such as macrocells and TV systems

Figure 1.3 shows an illustration of the coexistence between CogFem and

primary systems such as macrocells and TV systems. The problem of spec-
trum sharing emerges when deploying femtocell networks. Spectrum sharing
is not the unique problem for CogFem networks, but also an important prob-
lem for WLAN and macrocell networks. For the spectrum sharing problem in
WLAN, for example in industry such as Cisco, an AP placement strategy is
applied to reduce the interference between adjacent floors in a building [24].
In their AP placement strategy, they try to not “stack” APs in adjacent floors.
For example, in floor A, APs are placed in the living room, while in floor B,
APs are placed in the bedroom. This can increase the distance between APs
in adjacent floors, and reduce the interference, but still can not avoid the in-
terference. For macrocell networks, traditional spectrum allocation methods
are based on coloring methods that no neighboring cells can use the same
spectrum at the same time [27]. Since the number of femtocells could be
much higher than the number of macrocells in a certain area, this kind of
spectrum allocation requires more spectrum bands and will lead to inefficient
and unfair spectrum utilization. This motivates our study to further improve
the spectrum utilization and cell capacity in CogFem.

6
1.1 Motivations

1.1.3 Cognitive radio mesh networks

Wireless mesh networks is believed to be a highly promising technology to
extend the network access area in a cheap and convenient way [28] [29]. In
this context, there is a strong motivation to utilize the unused spectrum to
deliver the mesh network traffic [30]. Thus, Cognitive Radio Mesh Networks
(CogMesh), which combines CR and mesh technologies, is proposed with
the aim to improve the spectrum utilization and extend the network access
area simultaneously [30] [31] [32]. The CogMesh scenario we study is illus-
trated by a typical example in Fig. 1.4. There are several secondary mesh
routers (SMRs) and a Secondary Mesh Gateway (SMG) which connects to
the Internet. EachSMR and the SMG are equipped with one CR transceiver
and a normal radio transceiver with a dedicated control channel. The CR
transceiver in SMRs can sense and utilize the available spectrum holes unused
by the PUs [33]. Once PUs in this area return to that channel, the SMRs
should release these spectrum, and switch to another spectrum hole. Several
secondary mesh users (SMUs) can access their nearby SMRs to communicate
with the users not only in the CogMesh but also in the Internet through the
SMG.

Figure 1.4: An example of cognitive radio mesh network

A significant challenge in CogMesh is the real-time service provision,

which has strict constraints on end-to-end delay, jitters, and packet loss
rate. We focus on the delay and packet loss caused by the bottleneck of the
CogMesh. In CogMesh, the end-to-end delay includes the channel contention
delay on each link, channel switching delay on each SMR, queueing delay on
each SMR, and the transmission delay from the SMU to the SMG. The
quality of a wireless channel varies because of spatial, time, bandwidth, and

7
 Introduction

central frequency fluctuations [34]. The modulation scheme can be adjusted

according to the channel quality to assure that the data can be transmitted
successfully, with the help of Adaptive Modulation Coding (AMC) technique.
High data rate modulation scheme is used when the channel quality is good,
while lower data rate modulation scheme is employed when the channel qual-
ity becomes bad. Therefore, the data rate of the channel can vary. Hence
the transmission delay will be different if we use different channel selection
strategies. On the other hand, channel stability (availability) also varies for
different channels due to PUs’ activities. During the data transmission, if
some PUs return, SUs should stop the transmission and vacate the chan-
nel. Thus, packet collisions may happen, and the data transmitted by SMUs
could be lost.
All the above challenges motivate us in this work to develop an efficient
and reliable channel selection scheme for CogMesh to fulfill the real-time
applications.

1.2 Research methods

In this section, we introduce the research methods for general computer sci-
ence and specify the method used in this thesis work.

1.2.1 Research methods for computer science

The author in [35] indicates the research methods for computer science ac-
cording to the following three categories.

• Theoretical computer science [35]

In theoretical computer science, it follows the very classical method-
ology of building theories as logical systems with stringent definitions
of objects and operations for deriving and proving theorems. Theories
do not compete with each other but explain the fundamental nature
of information. There is no history of critical experiments that decide
between the validity of various theories as in physical sciences.
The central topic in theoretical computer science is the design and
analysis of algorithms. The results are judged by the insights they
reveal about the mathematical nature of various models of computing
and/or by their utility to the practice of computing and their ease of
application, for example the upper- and lower- resource bounds for the
solutions of various problems.

8
1.2 Research methods

• Experimental computer science [35]

In experimental computer science, experiments are used for both theory
testing and exploration. The computer scientists must observe phenom-
ena, formulate explanations and theories, and test them, to understand
the nature of information processes. Besides, experiments can also be
used in areas where theory and deductive analysis are difficult to apply,
and can help scientists derive theories from observation. Examples in
experimental computer science are automatic theorem proving, plan-
ning, NP-complete problems, natural language, vision, games, neural
nets/connections, machine learning, and analyzing performance behav-
ior on networked environments in the presence of resource contention
from many users.

• Computer simulation [35]

Computer simulation comprises computer-based modeling and simula-
tion. It can efficiently handle large data sets, can access a variety of
distributed resources and collaborate with other experts over the Inter-
net, etc. It is very efficient to tackle problems of great complexity. It
can also provide good visualization.

In addition, modeling is a common way for all these three areas. Modeling
is the first step of abstraction, it is used to simplify the phenomenon of
interest [35].

1.2.2 Research methods in this thesis

In this thesis, we combine the methods of both theoretical computer science
and computer simulation.
Firstly, we model the problem into a mathematical optimization form.
Then we use the theory of optimization to analyze and solve this problem.
Specifically, we formulate the resource optimization problem in one-channel
CogCell as an instance of multidimensional knapsack problem, the resource
optimization problem in multi-channel CogCell as an instance of multidimen-
sional multiple knapsack problem, the downlink spectrum sharing problem in
CogFem as a mixed integer non-linear programming problem, and the chan-
nel and route selection problem in CogMesh as a multiple choice knapsack
problem.
Finally, we write simulation codes for computer simulation to evaluate
our proposed algorithms. In this study, we build simulation scenarios and
implement our algorithms and schemes on Matlab [36]. Other simulation
platforms and programming languages can also be used to verify our proposed

9
 Introduction

algorithms. The results are also compared with a well known optimization
software called MOSEK [37].

1.3 Contributions
In this work, we have explored the resource management and optimization
problem in one/multiple channel cognitive radio cellular networks, femtocell
networks, and mesh networks. Our contributions are as follows.

1.3.1 Admission and power control for one-channel cog-

nitive radio cellular networks
In this study, we address the joint admission and power control in CogCell
from the perspective of the network operator. We formulate this problem as
an optimization problem where the objective is to maximize the secondary
revenue achieved by the BS, while satisfying the QoS requirements on SUs
and interference constraints on PRs.
In our study in [38], we propose Joint Admission and Power Control
scheme using a Minimal Revenue Efficiency Removal algorithm (JAPC-MRER)
to address the operator problem. In order to compare the performance of dif-
ferent schemes, we also introduce Joint Admission and Power Control scheme
using a Minimal SINR Removal Algorithm (JAPC-MSRA) and Joint Admis-
sion and Power Control scheme using a Random removal algorithm (JAPC-
Rand), wherein JAPC-MSRA uses an algorithm proposed in [39] to remove
the SUs with minimal Signal to Interference and Noise Ratio (SINR), while
JAPC-Rand remove SUs randomly when the constraints are not satisfied.
The comparison indicates that our proposed JAPC-MRER can achieve much
higher secondary revenue for the operator than the other two schemes, while
it has the similar time complexity with the other two schemes.
In our study in [40], we further improve JAPC-MRER. Firstly, we find
a way to determine the value of the power scale factor in JAPC-MRER
and introduce two pre-admission control schemes. Secondly, we reformulate
the admission and power control problem as a Multidimentional Knapsack
Problem (MKP). Then, we propose a novel admission and power control
scheme called JAPC-MKP which is heuristic with low complexity. Finally,
simulation results show that our proposed JAPC-MKP can approach the
optimal results from the optimization software MOSEK [37], and greatly
outperform the previous fixed power scale JAPC-MRER schemes.

10
1.3 Contributions

1.3.2 Channel allocation with admission and power con-

trol for multi-channel cognitive radio cellular net-
works
In multi-channel CogCell, we further extend our study in one-channel Cog-
Cell for the operator problem to maximize the revenue while admitting and
allocating channels to SUs and control the power for the admitted SUs . Our
contributions are threefold.

• We formulate the joint channel allocation, admission and power control

problem as a mixed-integer non-linear programming problem which is
NP-hard in general. Then, we transfer it to a 0-1 integer linear program-
ming, and can be analogous to a Multidimentional Multiple Knapsack
Problem (MMKP).

• Based on the MMKP modeling, we propose a heuristic algorithm to

get an approximate solution for the operator problem.

• Simulation results show that our proposed algorithm can achieve quite
close revenue to optimal solution by MOSEK [37], and achieve much
better revenue than other schemes.

1.3.3 Channel allocation and power control for cogni-

tive radio femtocell networks
In this study, we address the spectrum sharing problem in CogFem to maxi-
mize the capacity of femtocell networks. In particular, our contributions are
fourfold.

• To our best knowledge in the literature, our study is the first to incor-
porate the concept of CR into femtocells, and formulate the downlink
spectrum sharing problems in overlay mode as a mixed integer non-
linear programming problem.

• We employ mixed primal and dual decomposition methods to solve the

spectrum sharing problem. We also study the robust optimization con-
sidering the worst case due the random movements of FUs. According
to the solution of the decomposed problem, we proposed a joint channel
allocation and fast power control scheme.

• Simulation results show that CogFem could achieve much higher ca-
pacity than normal femtocells. The proposed scheme achieved much

11
 Introduction

higher average capacity and lower user blocking rate than the coloring
method.

• Simulation results also show that the proposed joint channel allocation
and power control scheme can converge very fast. In addition, the
expense for fixed power control scheme with our channel allocation
strategies is only 2% less average capacity comparing to the dynamic
power control scheme .

1.3.4 Channel selection for cognitive radio mesh net-

works
In the study of cognitive radio mesh networks, our contributions are threefold.

• We jointly consider two major factors, channel availability and channel

quality, for CogMesh in a heterogeneous primary system environment.
We formulate the problem of maximizing the route availability, while
guarantee that the end-to-end packet delay is less than a predefined
requirement.

• We transform the original non-linear programming problem to a 0-1

integer linear programming, and model it as a variant of Multiple-
Choice Knapsack Problem (MCKP). Based on the MCKP modeling,
we propose a heuristic method to solve this problem.

• Simulation results show that our proposed heuristic method can achieve
close route availability and solution rate to the optimal result from
MOSEK. It outperforms the best SINR scheme and best channel avail-
ability scheme.

1.4 Thesis organization

This thesis is organized as follows. Chapter 2 introduces the background of
cognitive radio networks, and summaries the related works in resource man-
agement and optimization in cognitive radio networks. Chapter 3 describes
our proposed admission and power control schemes for cognitive radio cel-
lular networks sharing one channel with primary users. Chapter 4 describes
our proposed channel allocation with admission and power control schemes
for cognitive radio cellular networks sharing multiple channels with primary
users. Chapter 5 describes our proposed spectrum allocation with power con-
trol schemes for cognitive radio femtocell networks. Chapter 6 describes our

12
1.4 Thesis organization

proposed route and channel selection scheme for cognitive radio mesh net-
works. Chapter 7 concludes our study in this thesis and points out several
future directions in the research on cognitive radio networks.

Table 1.1: Scenarios addressed in each chapter

Chapter Spectrum Sharing Mode Hops Channels

Chapter 3 Underlay one one
Chapter 4 Underlay one multiple
Chapter 5 Overlay one multiple
Chapter 6 Overlay multiple multiple

The relationship between the major chapters from Chapter 3 to Chap-

ter 6 can be seen from Table 1.1, where we summarize the scenarios addressed
in different chapters. We study the underlay spectrum sharing problem for
one-hop scenarios in one and multiple channel cases in Chapter 3 and Chap-
ter 4 respectively. While we study the overlay spectrum sharing problem for
multiple channels in both one-hop and multiple-hop cases in Chapter 5 and
Chapter 6 respectively.

13
Chapter 2

Background and Related Work

In this chapter, we introduce the background of cognitive radio technologies

and present the related work. The background of cognitive radio technolo-
gies includes the definition, key technologies, and deployment challenges. We
organize the related work around three main themes of our research on cogni-
tive radio networks: (i.) admission and power control with channel allocation
in cognitive radio cellular networks, (ii.) channel allocation and power con-
trol in cognitive radio femtocell networks, (iii.) channel selection in cognitive
radio mesh networks followed by a discussion of previous research related to
our own.

2.1 Background of cognitive radio networks

2.1.1 Definition of cognitive radio
The term “cognitive radio” was firstly introduced by Joseph Mitola in his
paper in 1999, where he defined cognitive radio as: “ A radio that employs
model based reasoning to achieve a specified level of competence in radio
related domains.” [18].
In 2005, Professor Simon Haykin defined cognitive radio as: “An intel-
ligent wireless communication system that is aware of its surrounding envi-
ronment (i.e., outside world), and uses the methodology of understanding-
by-building to learn from the environment and adapt its internal states to
statistical variations in the incoming RF stimuli by making corresponding
changes in certain operating parameters (e.g., transmit-power, carrier fre-
quency, and modulation strategy) in real-time, with two primary objectives
in mind: (i.) highly reliable communications whenever and wherever needed;
(ii.) efficient utilization of the radio spectrum.” [2].

15
 Background and Related Work

On the other hand, the regulator FCC defined cognitive radio as: “A
radio that can change its transmitter parameters based on interaction with
the environment in which it operates.” [41].
There will be a lot of benefits from the new radio regulations, such as
getting more capacity, decreasing the cost of communications, improving
reliability, and reaching longer distances with wireless equipments.

2.1.2 Cognitive cycle and key technologies

The basic cognitive cycle for a cognitive radio is shown in Fig. 2.1, wherein
the receiver is required to do spectrum sensing, analysis, and estimation
before transmission in order to protect PUs. The transmitter will then select
an appropriate spectrum band (channel) and control the transmit-power to
guarantee the interference to PUs are not harmful.

Figure 2.1: Cognitive radio operation cycle [2]

Spectrum sensing
In the overlay spectrum sharing mode, SUs detect the activities of PUs
in real time, and use the spectrum bands which are not occupied by any
PU. Spectrum sensing is one of the most important procedures in CR net-
works. In literature, there are four major methods for spectrum sensing,
i.e., Matched filter, Energy detection[42], Cyclostationary detection[43], and

16
2.1 Background of cognitive radio networks

Wavelet detection[44]. Each method has its own advantages and disadvan-
tages in different scenarios. Detecting the event of PU transmission by a
single node is not effective when the SU is shadowed from the PU, or when
the SU is out of the PU ’s transmission range but it can still interfere with
the primary receiver inside the PU ’s transmission range[19]. Therefore,
cooperative sensing[45][46], which allows several nodes sense the spectrum
environment and make the decision in a cooperative manner, is thought to
be an efficient way to solve such problems.

Spectrum decision
SUs make decision on which spectrum to use based on the spectrum sensing
results. It is one of the most important procedures. A good spectrum decision
mechanism should gain as much as possible benefit for transmission, provided
the interference to PUs is not harmful or SUs work in a different channel from
the channel used by PUs.
Several dynamic spectrum access schemes such as [47], [48], [49], [50],
and [51] are proposed using the sensing-based opportunistic spectrum access
approaches. For instance, in [47], SUs utilize the past observations to build
predictive models for spectrum availability, and choose the channels with the
most availability metric. In [48], the authors consider that SUs can only sense
some of the available channels because of hardware and energy constraints,
and derive the spectrum access strategies under the formulation of finite-
horizon Partially observable Markov Decision Processes (POMDPs). In [51],
the authors extend the work in [48]. They model the channel occupancy
by PUs with a continuous-time Markov chain, and propose an opportunis-
tic spectrum access scheme via periodic channel sensing, while reducing the
complexity of the optimal solution in [48].

Transmission power control

After spectrum decision, SUs should decide the transmission power on the
transmitter.
In the underlay spectrum sharing mode, each SU needs to control its
transmission power in order to guarantee the interference to PU is not harm-
ful. The key issue for the underlay approach is how to measure the inter-
ference on PUs in an efficient way[52]. Several works have considered the
interference constraints for SUs e.g., [53] [54].
On the other hand, in case of the overlay spectrum sharing mode, trans-
mission power control is mainly to achieve the required QoS level in terms of
data rate, etc.

17
 Background and Related Work

2.1.3 Deployment challenges

There are many challenges in deploying cognitive radio into reality, including
channel definition, channel availability and heterogeneity, channel quality,
and the common control channel.

Channel definition
In the literature, a channel in CR networks is always assumed as a spectrum
unit. However, there has been no definition about the bandwidth of a chan-
nel yet. This issue was firstly addressed by Ian F. Akyildiz et al. in [19].
Later D. Xu et al. studied the optimal channel bandwidth problem in [55]
to maximize the SUs’ throughput. Generally, a channel can get more ca-
pacity when the bandwidth increases, but the channel switching probability
may increase because the probability for PUs returning to a wider range of
spectrum could be higher than that in a smaller one. The increased channel
switching operations will then cause additional overheads like switching delay
which would reduce the SUs’ throughput.
Another uncertainty in defining a channel is overlapping or not. When
the available spectrum is divided into several channels, these channels could
be non-overlapping or partially overlapping. Two channels are said to be
non-overlapping when they are separated by at least 25 MHz [56]. Using
non-overlapping channels can eliminate the interference between different
channels, but may be a waste of spectrum. On the contrary, using partially
overlapped channels can improve the spectrum utilization, which is not al-
ways harmful [57]. Although channel overlapping can increase the number of
available channels and improve spectrum utilization, the adjacent SUs that
are using the partially overlapped channels may cause interference to each
other. Moreover, in this case, the interference to any PU on a certain channel
should include the effort of all the SUs’ transmissions on the partially over-
lapped channels, which results in more complexity to model and estimate the
interference on PUs.
The aforementioned issues mainly focus on channels divided by continuous
spectrum. However, it is possible to construct a channel with discrete sub-
carriers, as done by Orthogonal Frequency-Division Multiplexing (OFDM)
modulation scheme in physical layer, which has been widely used in the
IEEE 802.11a/g and the IEEE 802.16 standards [58].

Channel availability and heterogeneity

A channel is said to be available for SUs when it is not occupied by any
PUs (in the overlay spectrum sharing mode) or the interference from SUs to

18
2.1 Background of cognitive radio networks

PUs is under a tolerable threshold (in the underlay spectrum sharing mode).
PUs’ arbitrary activities result in a dynamic nature of channel availability.
In the literature, most work assumes the channel usage pattern of PUs fol-
lows an independent and identically distributed ON/OFF random process,
such as [59][60][61][62][63]. Where the ON-period represents the channel is
occupied by PUs while the OFF-period represents the channel is available
for SUs.
The channel availability of SUs on different locations may be distinct from
each other because of different PU activities. Even in the same geolocation,
SUs may have different available channels because of hardware limitations
such as sensing constraints (different SUs may be capable of sensing different
range of spectrum) and transmission constraints (SUs may be capable of
transmitting on different range of spectrum). This phenomenon would result
in the problem of channel heterogeneity where SUs have different available
channels at a certain time [64]. In this heterogeneous situation, neighboring
SUs should negotiate a common channel to communicate with each other
before data transmission.

Channel quality
The quality of wireless channels varies over time, space, and frequency. Some
important parameters were addressed in [19] as follows.

• Interference: Since channels are shared by different SUs, some chan-

nels may be more crowded compared to others. Therefore, an SU using
the same transmission power on different channels may result in differ-
ent SINR on its intent SU receiver. Higher SINR would bring higher
throughput to the SU. Moreover, consider the protection of PUs in the
underlay spectrum sharing mode, the allowed interference on different
channels may be different. Therefore, the allowed transmission power of
an SU should be controlled and may be different on different channels.

• Path Loss: The path loss is related to the distance between the SU
transmitter and receiver, as well as the channel central frequency. The
path loss increases when the distance and frequency increase. There-
fore, an SU transmitter may increase its transmission power to com-
pensate for the increased path loss to its intent SU receiver. However,
this may cause higher interference to other SUs and PUs.

• Wireless link errors: The errors of links using different channels depend
on the modulation scheme as well as the interference at the SU receiver.

19
 Background and Related Work

• Holding time: The holding time of a channel refers to the expected time
duration that SUs can work on this channel. Because the activities of
PUs may be different on each channel, the holding time may change
accordingly.

The channel quality can be characterized by the above parameters jointly.

Common control channel problem

Neighboring SUs in a CR network can communicate with each other directly
only if they work on a common channel. But before the communication,
they do not know which channel can be used on each other. So, they need
to exchange messages to know the available channels on each other. Thus a
common channel can be chosen based on their agreement. But the exchanged
messages require a Common Control Channel (CCC). This is called the CCC
problem as addressed in [65]: “a channel is required to choose a channel”.
In [66], the authors analyzed the design requirement of CR networks, and
suggested to distinguish control channel and data traffic channels. A simple
solution is to have a dedicated CCC. This channel is a dedicated licensed
spectrum band to SUs for the exchange of control messages, thus it will
not be interrupted by any PUs. In the literature, many contributions are
based on this assumption such as [67] [68] [59] [69] [70] [71]. However, this
assumption has several following drawbacks.

• License fee: A license fee may be required to get the licensed spectrum
band. Therefore, it would be expensive to build and deploy such a CR
network.
• Saturation: This dedicated channel can be saturated easily if many SUs
contend the control channel for their own traffic. Therefore, it would
be the bottleneck of the network throughput.
• Security: It is possible for adversaries to attack SUs by forging control
messages to the control channel. It may cause saturation of the control
channel that results in Denial-of-Service (DoS). These forged control
messages can also cause communication disruptions and gain unfair
advantages in resource allocation [72].

Another solution is to choose a control channel among the available chan-

nels such as in [65] [73]. There are several challenges related to this case.
Firstly, SUs should vacate the channels (in the overlay mode) or reduce the
transmission power (in the underlay mode) when PUs are detected. There-
fore, the control channel should be the most reliable channel that can not

20
2.2 Research problems in our work

be interrupted frequently. Secondly, it is sometimes not feasible to select a

CCC for the whole network due to the channel heterogeneity problem we
have mentioned in 2.1.3.

Spectrum sensing problems

Spectrum sensing is not always perfect, thus it gives rise to false alarm and
miss detection.
False alarm happens when the spectrum sensing results report the return
of PUs which are actually not exist. But following the sensing result, SUs
may stop the current transmission and decide to switch to another channel.
It then causes additional channel access delay and reduction of throughput.
In contrast to false alarm, miss detection happens when SUs fail to detect
the active PUs, and continue working on that channel. Thus, it can cause
uncontrolled interference to PUs. It is not only harmful to PUs but also
harmful to SUs.

2.2 Research problems in our work

We study the following three major problems: power control, admission con-
trol, channel allocation. These problems are always considered together.

2.2.1 Power control

Power control in cognitive radio networks is much more complex than in
traditional wireless networks.
In cognitive radio networks, SUs control transmission power not only to
achieve required QoS level while saving power, but also to protect primary
systems. The interference generated by SUs to any PU should be carefully
considered, and should not exceed the tolerable threshold.
In our study, we consider the power control problem in cognitive radio
cellular networks and femtocell networks. In cognitive radio cellular networks,
we focus on the uplink transmission power control for all SUs. SUs, which are
allowed to access the Base Station (BS), are required to control transmission
power to achieve their QoS level while the interference to PUs is not harmful.
In the scenario of CogFem, we focus on the downlink transmission power
control for all secondary femtocell base stations. Secondary femtocell base
stations control transmission power to achieve the downlink QoS level to
secondary femtocell users, while the interference to neighbouring femtocell
base stations are minimized.

21
 Background and Related Work

2.2.2 Admission control

When users’ requirement exceeds network’s capacity, admission control is
usually used besides power control to guarantee the service for dedicated
users by rejecting service requests from other users.
In CogCell, a BS is deployed to serve SUs in its coverage area and utilize
spectrum from primary systems, when some PUs are in the interference range
of the CogCell. SUs are not allowed to transmit any data to the BS if
the interference caused by SUs to the PUs is higher than the pre-defined
threshold. In addition, different SUs may require different levels of QoS and
hence make different payment based on the provided QoS level. From the
perspective of operators, the admission problem is to maximize the secondary
revenue while the interference from admitted SUs to PUs is less than the
tolerated interference threshold, and the QoS level of admitted SUs can be
satisfied.

2.2.3 Channel allocation

Channel allocation is an important problem for coexistence between SUs and
PUs in cognitive radio networks. It is highly related to spectrum decisions.
In CogCell, we consider the uplink channel allocation from SUs to sec-
ondary BS. The channel allocation strategy is designed to not only control
the interference between SUs working in the same channel, but also control
the interference to PUs working in that channel. In addition, with the con-
straint of limited transmission power and required QoS level, the channel
allocation problem becomes more challenging.
In CogFem, we consider the intra-femtocell channel allocation and down-
link channel allocation for femtocell users in each femtocell. Where the intra-
femtocell channel allocation handles the interference between neighbouring
femtocells, while the downlink channel allocation for femtocell users tries to
save power providing the required QoS level is satisfied.
In CogMesh, channel allocation problem is the fundamental problem for
each link to select a channel to transmit and receive. However, channels have
different characteristics in terms of different channel availability and quality.
It is a fundamental requirement to provide a reliable route in CogMesh. In
addition, we also consider the real-time applications, where end-to-end delay
is required to be less than a threshold. To this end, we need to design a
metric for each link to select an appropriate channel so that the end-to-end
delay requirements are guaranteed and the route availability is maximized.

22
2.3 Related work

2.3 Related work

2.3.1 Resource management in cognitive radio cellular
networks
Resource management in cognitive radio cellular networks is related to chan-
nel allocation, admission control, and power control schemes.
In literature, a few attempts have been made on the resource allocation
and power control problems for CogCell in the underlay spectrum sharing
mode. They can be classified according to different number of PUs per chan-
nel, different number of channels, and different spectrum access schemes as
shown in Table 2.1. In the following, we will discuss the previous works
according to different number of PUs and different channels.
Related work according to different number of PUs are shown as follows.
• In the case of one PU system model, there are related works such
as [74] and [75]. In [74], Y. Xing et al. considered the scenario with
one PU, several SUs and separative receivers. The study proposed a
distributed constrained power control algorithm and found the optimal
link subset to achieve the maximum revenue with the help of a potential
game. In [75], L. Zhang et al. modeled a smooth optimization problem,
and proposed a minimal SINR removal algorithm (MSRA)to search the
optimal set of SUs.
• In the case of multiple PUs system model, there are related works such
as [76], [77], [78] and [79]. Specially, in [77], the authors studied the
problem of power allocation in a Single Input Multiple Output Multiple
Access Channel (SIMO-MAC) based CR network. Where channel is
divided into subchannels as the same number of the antennas of the
BS. They proposed a multi cap water-filling algorithms to allocate the
power for each SU.
Related work according to different number of channels are shown as
follows.
• Some efforts have been made in one-channel Code Division Multiple
Access (CDMA) CR networks. For example, in [80], the authors studied
the problem of power and rate allocation for a set of links sharing
only one channel in a CR network. The power allocation is either
0 or maximum power, where if an SU is in a bad channel it stops
transmission, otherwise it transmits with the maximum power. In [81],
the authors studied the rate and power allocation problem in a CDMA
cognitive radio networks sharing one channel.

23
 Background and Related Work

• For multiple channel scenarios, several efforts have been made in Or-
thogonal Frequency-Division Multiple Access (OFDMA) CR networks,
Where each subchannel can be allocated to only one SU. For example,
in [82], the authors studied the problem of frequency, rate and power
allocation in OFDMA CR networks. They decomposed the original
problem into subproblems to maximize the utility of every subchannel.
In [83], the authors studied the subcarrier allocation and power control
for OFDMA CR networks.

Table 2.1: Related works in cognitive radio cellular networks

Related Number Number of Medium

Work of PUs Channels Access
[74], [75] one one CDMA
SIMO-
[77] multiple one
MAC
[79], [81], [84]
multiple one CDMA
and [80]
[82], [83] one multiple OFDMA

2.3.2 Spectrum sharing in cognitive radio femtocell net-

works
Since our work on cognitive radio femtocell networks is the first as far as we
know. In literature, there is no study on spectrum sharing in cognitive radio
femtocell networks. However, there are few attempts on normal femtocell
networks. In [85], the authors studied a downlink case for WiMAX femto-
cell networks. In [86], the authors applied a finite-difference time-domain
(FDTD) method to predict the coverage of WiMAX femtocells. In [87], the
authors used a centralized method of dynamic frequency planning (DFP) to
minimize the overall femtocell network interference to allocate the spectrum
to femtocells. In [88], the authors studied the resource management problem
in Orthogonal Frequency-Division Multiple Access (OFMDA) femtocells and
proposed a location-based allocation scheme between macrocells and femto-
cells to adapt the varying user population.

24
2.3 Related work

2.3.3 Channel selection in cognitive radio mesh net-

works
In literature, there are some related work on channel assignment consider-
ing end-to-end delay requirements in wireless mesh networks . For example,
in [89], the authors proposed an interference avoidance channel assignment
scheme for different links based on graph coloring. In [90], the authors pro-
posed a channel selection scheme for 802.11 based wireless mesh networks.
In [91], the authors designed a routing protocol based on the (weighted) end-
to-end delay metric in order to minimize the end-to-end delay in wireless
mesh networks. However, in CogMesh, we should take into account chan-
nel availability due to the activities of primary systems, thus the channel
assignment problem is more challenging.
There are also some related work about channel selection and dynamic
spectrum access in CR networks. In [92], a two-person cooperative game the-
oretical approach is applied in the channel selection between two secondary
nodes. However, this work considers only a one-hop scenario with two sec-
ondary users. In [93], the authors proposed a channel selection scheme to se-
lect the channel with the highest channel weight which is defined as e−p (1−p),
where p is the occupancy rate of PUs. In [94], the authors studied the channel
selection and routing problems in multi hop CR networks, with the objec-
tive of minimizing the total bandwidth used in the network. However, they
did not consider the end-to-end delay requirement for each flow. In [95],
a stochastic channel selection algorithm based on learning automate tech-
niques is proposed. Each secondary node selects one of the channels in a
probability which is defined in a probability list. This probability list will be
updated according to the result of each selection. The packet will be sent
once the channel selected is available to use. However, the end-to-end service
requirement is not considered in this work. In [96], three channel selection
strategies are proposed for SUs to access heterogeneous channels. The first
two are based on the detection of PUs’ activities. The third one is based on
the monitoring of the throughput of secondary nodes. Thus, in their work,
channel selection strategy is made according to either the channel availability
(PU’s activity) or the channel quality (throughput can be achieved). These
two factors are not considered simultaneously. Moreover, only one-hop sec-
ondary systems are considered. In a multi-hop network, such as CogMesh,
the channel selection problem will be more complex.

25
Chapter 3

Power and Admission Control

for One-channel Cognitive
Radio Cellular Networks

We start with the resource management and optimization problem in one-

channel CogCell, where joint power and admission control is one of the most
important issues. In such a CogCell, a BS is deployed to serve SUs in its
coverage area and utilize spectrum from primary systems, when some PUs
are in the interference range of the CogCell. In the uplink, SUs can be
admitted to the BS provided that the interference caused by SUs to the PUs
is no higher than the pre-defined threshold. In addition, different SUs may
require different levels of QoS and hence make different payment based on
the provided QoS level.
In this chapter, we address the joint admission and power control problem
in CogCell from the perspective of the network operator to maximize the
revenue obtained from SUs subjected to the interference constraints on PUs
and QoS requirements of SUs.
The rest of this chapter is organized as follows. We introduce the CogCell
model and formulate the optimization problem in Section 3.1. In Section 3.2,
we describe and evaluate the performance of our proposed JAPC-MRER and
two other schemes, i.e. JAPC-MSRA, and JAPC-Rand. We then discuss the
power control schemes and propose three pre-admission control schemes in
Section 3.3. We further improve our results by reformulation the problem
and propose the Joint Admission and Power Control with Multidimensional
Knapsack Problem modeling (JAPC-MKP) scheme in Section 3.4. Then we
evaluate the performance of all the proposed schemes in Section 3.5. Finally,
Section 3.6 concludes this chapter.

27
 Power and Admission Control for One-channel Cognitive Radio
Cellular Networks

3.1 System model and problem formulation

In this section, we describe the system model, introduce the definitions of
interference constraints and QoS requirements, and finally formulate the op-
erator optimization problem.

3.1.1 System model

Figure 3.1 shows the system model of one channel CogCell. The BS is located
at the center of the cell and provides services for SUs. A number of PUs
including PRs and primary transmitters (PTs) are distributed in this cell.
PRs are receiving while PTs are transmitting. This CogCell employs Code-
Division Multiple Access (CDMA), so that SUs can access the same spectrum
band simultaneously. We consider the situation that the spectrum used by
SUs are licensed to PUs. Hence, PRs receive interference from SUs which are
transmitting data to the BS, while the BS receives interference from PTs.
We further assume that the BS can measure the interference from PTs.

Figure 3.1: System model of one-channel cognitive radio cellular networks

Table 3.1 lists the notations in this chapter. We denote the interference
generated by PTs to the BS as Ip , which can be dynamically changed ac-
cording to the movements and other activities of PTs. Let Ns denote the
set of SUs, Np represent the set of PRs, respectively. Let ns = |Ns | and
np = |Np |. Namely, ns and np denote the number of SUs and PRs in the cell,
respectively. The network service provider receives the revenue from the ac-
cumulated payment by every admitted SU. Suppose that SU i (i ∈ Ns ) pays

28
3.1 System model and problem formulation

ri for the operator with the QoS demand in terms of minimal Data Trans-
mission Rate (DTR) λ̄i . On the other hand, SU i generates interference τij
to PU j if SU i is allowed to access the channel. The interference to PU j
from all the active SUs cannot exceed the threshold Γj .

Table 3.1: Table of notations for one-channel CogCell

Symbol Meaning
Ns the set of SUs
Np the set of PRs
Is the interference received at the BS from all SUs
Ip the interference power received by the BS from PTs
ns the number of SUs
np the number of PRs
Pi the transmission power at SU i
P̂ the maximum transmission power at SUs
ri the revenue from SU i
τij the interference from SU i to PR j
Γj the threshold of interference power at PR j
hi the power attenuation from SU i to the BS
hij the power attenuation from SU i to PR j
di the distance between SU i and the BS
dij the distance between SU i and PR j
λ̄i the minimum uplink DTR required by SU i
ξ¯i the minimum uplink SINR required by SU i

3.1.2 Interference power

While SUs share the spectrum with PUs (including PTs and PRs), SUs causes
interferences to the PRs. Let TjI denote the interference power received by
PR j.
Xns
I
Tj = hij Pi xi (3.1)
i=1

where the indicator xi shows whether SU i is admitted or not. xi = 1

represents that SU i is admitted, zero otherwise. Pi refers to the transmission
power at SU i. hij denotes the power attenuation from SU i to PU j and is

29
 Power and Admission Control for One-channel Cognitive Radio
Cellular Networks

given by
Gsi Gpj
hij = (3.2)
(dij )n
where dij denotes the distance from SU i to PR j. The exponent n is the
path fading factor. Gsi and Gpj denote the antenna gains of SU i and PR j,
respectively. Therefore, according to (3.1) and (3.2), the interference power
caused by SU i is expressed as

Gsi Gpj Pi
τij = hij Pi = . (3.3)
(dij )n

3.1.3 QoS definition and requirement

In CogCell, different SUs may have different QoS requirements, and make
different payment (e.g., SU i pays ri to the operator at the BS). In this
chapter, we employ DTR as the major QoS metric. According to Shannon’s
channel capacity formula, the uplink maximum data transmission rate from
SU i to the BS is given by

λi = Blog2 (1 + ξi ) (3.4)

where ξi is the uplink SINR of SU i measured at the BS. In the CogCell

network, different SUs may require different traffic demands, e.g. voice, video
and web browsing. For different types of traffic, the required data rates are
different. Let λ̄i denote the minimum required DTR by SU i. Let ξ¯i denote
the required SINR for SU i. Therefore, based on (3.4), we can obtain the
required SINR as
λ¯i
ξ¯i = 2 B − 1 (3.5)

Let Is denote the accumulated interference at the BS caused by all active

ns
P
SUs, i.e., Is = hi Pi xi . According to the definition of SINR in [34], we can
i=1
calculate the uplink SINR of an active SU i as following

received power at the BS for SUi

ξi =
noise plus interference
hP
= Pins i (3.6)
N0 + Ip + j=1,j6=i hsb s
j Pj xj
hi Pi
=
N0 + Ip + Is − hi Pi

30
3.1 System model and problem formulation

where hi denotes the power attenuation from SU i to the BS. Similarly, we

have
Gs Gb
hi = i n (3.7)
(di )
where Gb stands for the antenna gain of the BS. di refers to the distance from
SU i to the BS.

3.1.4 The operator problem

The operator problem in CogCell is to maximize the revenue obtained by the
operator of the BS. We define the revenue received by the operator on the BS
from the SUs as the secondary revenue. The objective is to find the optimal
subset of admitted SUs such that the secondary revenue is maximized. At
the same time, both the interference power at PUs and the QoS requirements
of SUs should be guaranteed. The problem is formulated as follows
ns
X
maximize ri xi (3.8)
i=1
subject to:

ns
X
τij xi ≤ Γj , ∀j ∈ Np (3.9)
i=1
xi ∈ {0, 1}, ∀i ∈ Ns (3.10)
ξi ≥ ξ¯i , if xi = 1, ∀i ∈ Ns (3.11)
Pi ∈ [0, P̂ ], ∀i ∈ Ns (3.12)
where P̂ is the maximal transmission power for each SU. Constraint (3.9)
represents that the interference from all SUs to PUs cannot exceed the in-
terference threshold. Constraint (3.11) represents that the QoS (in terms of
SINR, which is determined by DTR) requirement of active SUs should be
satisfied. Constraint (3.12) represents the power limitation of SUs.
The defined optimization problem should solve the transmission power of
SUs and find out the optimal subset of SUs. Only considering the constraints
(3.9) and (3.10), the defined problem (3.8) is a typical 0-1 linear problem,
which is NP-Complete [97]. However, τij (∀i ∈ Ns , j ∈ Np ) is dynamically
changed with different power allocation schemes of SUs based on the con-
straints (3.11) and (3.12). Thereafter, the dynamics of τij (∀i ∈ Ns , j ∈ Np )
makes the original 0 − 1 linear problem even more challenging. In the follow-
ing section, we propose a joint admission and power control scheme to solve
this problem.

31
 Power and Admission Control for One-channel Cognitive Radio
Cellular Networks

3.2 Joint admission and power control schemes

In this section, we propose a joint admission and power control scheme using
minimal revenue efficiency removal algorithm called JAPC-MRER to address
the operator problem. In order to compare the performance of different
schemes, we also introduce JAPC-MSRA and JAPC-Rand, wherein JAPC-
MSRA uses an algorithm proposed in [39] to remove the SUs with minimal
SINR, while JAPC-Rand remove SU randomly when the constraints are not
satisfied.

3.2.1 JAPC-MRER
JAPC-MRER runs in a heuristic way by several iterative operations. Let Ns∗
and Np∗ denote the possible set of admitted SUs and the valid set of PUs,
respectively. Initially, all SUs are admitted by the BS, i.e., Ns∗ = Ns , and all
PUs should be taken into account, i.e., Np∗ = Np . Let Pimin (i ∈ Ns∗ ) denote
the minimum transmission power of SU i to achieve the required minimum
SINR ξ¯i . The ratio relationship of Pimin between all SUs can be represented
in the following: P1min : P2min : ... : Pnmin
s
= y1 : y2 : ... : yns , where yi (i ∈ Ns∗ )
can be calculated by (3.6) and (3.11) as follows
1
yi = −1 (3.13)
(1 + ξ¯i )hi
Therefore, we can temporally allocate the power β P̂ to the SU which has
the largest power ratio ŷ. Here, β is a power scaling factor (β ∈ (0, 1]).

ŷ = max{yi |∀i ∈ Ns∗ } (3.14)

The power used by other SUs can be assigned based on the ratio to the
SU with the transmission power β P̂ as follows

β P̂
Pimin = yi (3.15)
ŷ
We choose Pi equal to Pimin , for all i in the set of Ns∗ . The reasoning is as
follows. If there exists any i in the set of Ns∗ , wherein Pi is greater than Pimin ,
SU i causes more interferences to any other SU j (∀j ∈ Ns∗ , j 6= i) than using
the transmission power Pimin . According to (3.6), the SINR of SU j decreases
if SU j does not increase its transmission power accordingly. Therefore, all
SUs in the set of Ns∗ other than SU i should increase their transmission power
to keep their SINRs non-decreasing. On the other hand, if SUs increase the
transmission power, PUs will receive more interference. Due to the constraint

32
3.2 Joint admission and power control schemes

of interference threshold, fewer SUs can be admitted, which will result in

smaller secondary revenue. This will be worse than the situation when each
SU uses Pimin as the transmission power. As a consequence, the power of
SUs Pi should be set as Pimin (∀i ∈ Ns∗ ) to achieve the maximal secondary
revenue. After allocating the power to every active SU, we can calculate the
interference from every SU to every PU. We use ϕj (j = 1, ..., np ) to record
the difference between the total interference on PU j and its interference
threshold.
X
ϕj = τij − Γj (3.16)
i∈Ns∗

If ϕj is not greater than 0, the total interference experienced by PU j

is less than its threshold. In this situation, PU j should be removed from
Np∗ (the set of valid PUs) since the interference constraint has been already
satisfied on this PU. Otherwise, i.e. ϕj is greater than 0, the positive value
of ϕj can physically represent the importance of PU j to the admission set.
Consequently, we introduce the revenue efficiency factor as follows
ri
ei = P , ∀i ∈ Ns∗ (3.17)
ϕj τij
j∈Np∗

The SU with higher revenue efficiency factor is able to provide higher

revenue for the operator while generating lower interference to PUs. In order
to achieve higher secondary revenue with guaranteed interference at PUs, we
can remove the SU with the minimal revenue efficiency factor in the next
iteration. The detailed algorithm is shown in Algorithm 1. It terminates if
either Np∗ or Ns∗ becomes empty.
The time complexity is dominated by the operation of calculating Pi , ϕj
and ei . There are maximum ns iterations for the main loop. After each main
loop, the number of admitted SUs is reduced by one. The time complexity
can be calculated as follows.

T (ns , np ) = (ns + (ns − 1) + (ns − 2) + ...) calculate power

+(ns + (ns − 1) + (ns − 2) + ...)np calculate ϕj
+(ns + (ns − 1) + (ns − 2) + ...)np calculate ei
= (ns + (ns − 1) + (ns − 2) + ...)(2np − 1)
≤ ns (1+ns2)(2np −1)
= O(n2s np )
(3.18)
2
Therefore the time complexity is O(ns np ).

33
 Power and Admission Control for One-channel Cognitive Radio
Cellular Networks

Algorithm 1 JAPC-MRER
Input: Ns , Np , {λi }, {Γj }, {dij }, {di }.
Output: Ns∗ , {Pi }
1: Initialization: Ns∗ ← Ns , Np∗ ← Np .
2: Calculate yi , ∀i ∈ Ns∗ , according to (3.13)
3: while Np∗ 6= ∅ do
4: Select ŷ according to (3.14)
5: Calculate Pimin , ∀i ∈ Ns∗ , according to (3.15).
6: Pi ← Pimin , ∀i ∈ Ns∗ .
7: for j ∈ Np∗ do
8: Calculate ϕj according to (3.16)
9: if ϕj ≤ 0 then
10: Np∗ ← Np∗ − j
11: if Np∗ == ∅ then
12: “All the Interference constraints are satisfied”
13: Return;
14: end if
15: end if
16: end for
17: Calculate ei , ∀i ∈ Ns∗ , according to (3.17).
18: Choose an SU i, where ei = min{ej |∀j ∈ Ns∗ },
19: Pi ← 0
20: Ns∗ ← Ns∗ − i
21: if Ns∗ == ∅ then
22: “No SU can be admitted”
23: Return;
24: end if
25: end while

3.2.2 JAPC-MSRA
JAPC-MSRA is also a joint admission and power control scheme. Instead, it
uses a minimal SINR removal algorithm which is proposed in [39].
The details are shown in Algorithm 2, where, the set of SUs Ns∗ is updated
by removing the SU with the minimal SINR in each iteration of the main
loop.
The time complexity is dominated by the operation of calculating the
power Pi , ξi , and the verification about the interference threshold constraints
in each iteration of the main loop. The time complexity can be calculated as
follows.

34
3.2 Joint admission and power control schemes

Algorithm 2 JAPC-MSRA
Input: Ns , Np , {λi }, {Γj }, {dij }, {di }.
Output: Ns∗ , {Pi }
1: Initialization: Ns∗ ← Ns , Np∗ ← Np .
2: Calculate yi , ∀i ∈ Ns∗ , according to (3.13)
3: while Np∗ 6= ∅ do
4: Select ŷ according to (3.14)
5: Calculate Pimin , ∀i ∈ Ns∗ , according to (3.15).
6: Pi ← Pimin , ∀i ∈ Ns∗ .
7: if All interference threshold constraints are valid then
8: Break;
9: end if
10: Calculate ξi , ∀i ∈ Ns∗ , according to (3.6).
11: Choose an SU i, where ξi = min{ξj |∀j ∈ Ns∗ },
12: Pi ← 0
13: Ns∗ ← Ns∗ − i
14: if Ns∗ == ∅ then
15: Echo “No SU can be admitted”
16: Break;
17: end if
18: end while

T (ns , np ) = (ns + (ns − 1) + (ns − 2) + ...) calculate power

+(ns + (ns − 1) + (ns − 2) + ...)np verify interference thresholds
+(ns + (ns − 1) + (ns − 2) + ...) calculate ξi
= (ns + (ns − 1) + (ns − 2) + ...)(np + 2)
≤ ns (1+ns2)(np +2)
= O(n2s np )
(3.19)
2
Therefore the time complexity of this algorithm is O(ns np ), which is the
same as JAPC-MRER.

3.2.3 JAPC-Rand
JAPC-Rand is also a joint admission and power control scheme. It randomly
removes an SU in each iteration. The implementation of JAPC-Rand is
based on the Algorithm 2 with minor modification that the line 10 and 11
are modified to randomly select an SU. In consequence, the set of SUs is
updated by randomly removing an SU in each iteration operation.

35
 Power and Admission Control for One-channel Cognitive Radio
Cellular Networks

The time complexity is dominated by the operation of calculating power

Pi in each loop, and the verification of the interference threshold constraints.
The time complexity can be calculated as follows.

T (ns , np ) = (ns + (ns − 1) + (ns − 2) + ...) calculate power

+(ns + (ns − 1) + (ns − 2) + ...)np verify interference thresholds
≤ ns (1+ns2)(np +1)
= O(n2s np )
(3.20)
2
Therefore the time complexity of this algorithm is O(ns np ), which is the
same as JAPC-MSRA and JAPC-MRER.

3.2.4 Simulation results and analysis

In this section, we provide the simulation results to demonstrate the perfor-
mance of the three joint admission and power control schemes JAPC-MRER,
JAPC-MSRA, and JAPC-Rand.
We have implemented a CogCell simulator using Matlab [36]. In the
simulator, there is a BS located at the center of a cell. The parameters
used in this simulation are summarized in Table 3.2. Where Rmax denotes
the radius of the cell and is set as 1000m. The minimal distance from the
BS to any SUs or PUs is denoted by Rmin . In our simulation, we choose
Rmin = 100m. The topologies of SUs and PUs are generated in the way as
follows. The distance between SUs (or PUs) and the BS are randomly chosen
from [Rmin , Rmax ], The angles from any SUs (or PUs) to the BS are randomly
chosen from [0, 2π]. The uplink channel bandwidth B is set as 5M Hz. The
DTR demand required by every SU i (i ∈ Ns ) is randomly chosen from
Table 3.3. We also show the DTR and SINR mapping in Table 3.3, where
the required SINR demands are calculated by (3.5).
The path fading factor n is set as 4. The antenna gains of all SUs, PUs,
and the BS are equal to 1. The power scaling factor β is set as 1.
The revenue ri obtained from SU i (i ∈ Ns ) is dependent on the DTR. The
SU with higher DTR pays more and hence generate higher revenue for the
service provider. Without loss of generality and for the sake of illustration,
we allocate the revenue and DTR according to Table 3.3.
In the following, we evaluate the performance in terms of the secondary
revenues in three different cases, i.e., changing the number of PUs, changing
the number of SUs, and changing the interference threshold. We randomly
generate 100 topologies. In each topology, we randomly generate the DTR
demands. In each case, we run the three schemes, i.e., JAPC-MRER, JAPC-

36
3.2 Joint admission and power control schemes

Table 3.2: Simulation parameters for admission control in CogCell

Symbol Value Symbol Value

Rmax 1000m Rmin 100m
B 5 MHz P̂ 0.28 W
n 4 β 1
Gsb
i 1 Gb 1

Table 3.3: Revenue allocation table with DTR and SINR mapping

Revenue 1 2 4 8 16 32
DTR (kbps) 16 32 64 128 256 512
Required SINR 0.0022 0.0043 0.0087 0.0175 0.0353 0.0718

MSRA, and JAPC-Rand to obtain the secondary revenue. Then, we calculate

the average secondary revenue based on the results in these 100 topologies.

Effect of the number of PUs

In this case, we fix the number of SU s and the interference threshold while
changing the number of PUs. Particularly, there are 50 SUs in the CogCell.
The interference threshold for each PU is −90dBW .
Figure 3.2 shows the secondary revenue in terms of the number of PUs
with three different schemes JAPC-MRER, JAPC-MSRA, and JAPC-Rand.
The secondary revenue decreases with the increasing number of PUs.
Because more PUs in the cell result in more stringent interference constraints.
This leads to fewer admittable SUs and hence lower revenue. In Fig. 3.2, when
np = 1, the secondary revenue obtained by JAPC-MRER, JAPC-MSRA,
and JAPC-Rand is 512.6, 478.2, and 463.2, respectively. When np increases
to 50, the secondary revenue obtained by JAPC-MRER, JAPC-MSRA, and
JAPC-Rand decreases to 384.5, 68.96, and 43.17, respectively. The secondary
revenue obtained by JAPC-MRER, when np increases to 50, is more than
5 times of that obtained by JAPC-MSRA, and is almost 9 times of that
obtained by JAPC-Rand.
The JAPC-Rand generates the lowest revenue since the JAPC-Rand does
not consider the payment differentiation and may remove the SU with high
revenue for the operator. The operator can receive more revenue by employ-

37
 Power and Admission Control for One-channel Cognitive Radio
Cellular Networks

600
JAPC−MRER
JAPC−MSRA
500 JAPC−Rand

400
Secondary Revenue

300

200

100

0
0 10 20 30 40 50
Number of PUs

Figure 3.2: The secondary revenue in terms of the number of PUs

ing JAPC-MSRA than by using JAPC-Rand since JAPC-MSRA iteratively

removes SUs with the minimal SINR, which gives minimal payment to the
operator. However, only considering the payment is not enough to achieve
the maximum secondary revenue. The interference constraints should be also
taken into account. In JAPC-MRER, the introduced revenue efficiency factor
considers not only the generated revenue but also the interference to all PUs.
Following this advantage, we can see that the operator can obtain much more
secondary revenue by employing JAPC-MRER than the other two schemes.

Effect of the number of SUs

In this case, we change the number of SUs while fixing the interference thresh-
old and the number of PUs. Especially, the interference threshold for each
PU is the same as in the previous case, which is −90dBW . The number of
PUs is 6.
Figure 3.3 shows the secondary revenue in terms of the number of SUs
with three different schemes JAPC-MRER, JAPC-MSRA, and JAPC-Rand.
The secondary revenue increases with the increasing number of SUs.
When the number of SUs is less than 4, all these schemes achieve the same
revenue. In this situation, the number of SUs is so few that the interference on
all the PUs are too small to exceed the threshold. With the increasing num-
ber of SUs, the schemes perform differently. For instance, when ns increases

38
3.2 Joint admission and power control schemes

600

JAPC−MRER
500
JAPC−MSRA
JAPC−Random

400
Secondary Revenue

300

200

100

0
0 10 20 30 40 50
Number of SUs

Figure 3.3: The secondary revenue in terms of the number of SUs

to 50, the secondary revenue obtained by JAPC-MRER, JAPC-MSRA, and

JAPC-Rand is 495.4, 340.7, and 300.5, respectively. Again, JAPC-Rand
achieves the least secondary revenue because it may remove the SUs with
high revenue but low interference. JAPC-MSRA achieves more revenue than
JAPC-Rand for the operator because it can keep the SUs with high revenue.
However, the SUs with high revenue and also high interference to PUs may be
admitted. Our proposed scheme JAPC-MRER achieve the balance between
the revenue and the interference, and can achieve the highest revenue.

Effect of the interference threshold

In this case, we fix the number of PUs and SUs while changing the interference
threshold for PUs. Specially, there are 6 PUs and 50 SUs in the BS. The
interference thresholds of all PUs are changed from −150dBW to −10dBW .
The results are shown in Fig. 3.4, where JAPC-MRER achieves a slightly
lower revenue than the optimal when the interference threshold is less than
−100dBW . There are two extreme situations when all the schemes achieve
nearly same performance. In this example, when the threshold is less than
−130dBW , the interference thresholds are too small to admit any SU while
guaranteeing the limited interference to PUs. When the threshold is greater
than −50dBW , the interference thresholds are sufficiently large and hence all
SUs can be admitted while guaranteeing the interference to PUs. When the

39
 Power and Admission Control for One-channel Cognitive Radio
Cellular Networks

600

500

400
Secondary Revenue

JAPC−MRER
JAPC−MSRA
300 JAPC−Rand

200

100

0
−150 −100 −50 0
Interference Threshold (dBW)

Figure 3.4: The secondary revenue in terms of the interference threshold

interference threshold is greater than −130dBW and smaller than −50dBW ,

JAPC-MSRA and JAPC-Rand perform very similarly, because these two
schemes do not consider the interference to PUs. Since JAPC-MRER con-
siders the influence of the interference, it removes the SUs with high interfer-
ence to PUs and low revenue to the operator. Following this, JAPC-MRER
is capable of achieving the highest among these schemes.

3.3 Discussions on power control and pre-admission

schemes

In this section, we further discuss the power scaling factor β in the power con-
trol schemes in the previous section and propose three pre-admission control
schemes.
For any admitted SU i, according to the SINR constraints and the power
allocation strategy in (3.15), we have

40
3.3 Discussions on power control and pre-admission schemes

hi Pi
ξi =
N0 + Ip + Is − hi Pi
hi βŷP̂ yi
= (3.21)
yk βŷP̂ yk − hi βŷP̂ yi
P
N0 + Ip +
k∈Ns∗

≥ξ¯i
Then, we obtain
ŷ(N0 + Ip )
β≥ ! (3.22)
hi yi P
P̂ ξ¯i
− hk yk
k∈Ns∗ ,k6=i

where
hi yi X
− hk yk > 0
ξ¯i ∗
k∈N ,k6=i s

Moreover, since β ∈ (0, 1], we have

ŷ(N0 + Ip )
0< ! ≤1 (3.23)
hi yi P
P̂ ξ¯i
− hk yk
k∈Ns∗ ,k6=i

Then we can obtain

hi yi X
¯ − hk yk ≥ ŷ(N0 + Ip )P̂ −1 (3.24)
ξi k∈N ∗ ,k6=i s

which is equivalent to the following inequality

1 X 1 −1
¯ − −1 ≥ ŷ(N0 + Ip )P̂ (3.25)
1 + ξi 1 + ξ¯
k∈Ns∗ ,k6=i k

where the left side of the inequality can be further transformed as follow.
 
1 X 1 1 X 1
¯ − −1 = 1 − −1 −
¯
1 + ξi k∈N ∗ ,k6=i 1 + ξk ¯
1 + ξi ¯ −1
s k∈Ns∗ ,k6=i 1 + ξk
X 1
=1− (3.26)
¯ −1
∗ 1 + ξk k∈Ns
X 1
=1− −1
i∈Ns∗ 1 + ξ¯i

41
 Power and Admission Control for One-channel Cognitive Radio
Cellular Networks

Therefore, substituting (3.26) into (3.25), we have

 
1 −
X 1  ŷ −1 ≥ (N0 + Ip )P̂ −1
−1 (3.27)
¯
i∈Ns∗ 1 + ξi

If the inequality (3.27) is not true, the SINR constraints (3.11) cannot be
satisfied. Therefore, some pre-admission procedures should be carried out.
In this study, we consider the following three metrics.

• Maximum y removal: The maximum y removal scheme will remove SUs

with maximum value of y until the inequality (3.27) are true.

• Minimum SINR removal: The minimum SINR removal scheme will

remove SUs with minimum SINR ξ until the inequality (3.27) are true.

• Minimum channel gain removal: The minimum channel gain removal

scheme will remove SUs with minimum channel gain h until the in-
equality (3.27) are true.

After the above pre-admission procedures, we can have an updated set

of SUs Ns∗ . Then, according to (3.22) and (3.26) we can obtain the value of
power scale factor.

ŷ(N0 + Ip )
β≥ ! (3.28)
1
P
P̂ 1− 1+ξ¯i
−1
i∈Ns∗

Since the bigger β is, the higher the power is, which will result more inter-
ference to PUs. Therefore, we will choose β as small as possible.
ŷ(N0 + Ip )
β= ! (3.29)
1
P
P̂ 1− 1+ξ¯i
−1
i∈Ns∗

By applying different pre-admission control metrics (maximum y removal,

minimum SINR removal, or minimum channel gain removal), removal algo-
rithms (MRER or MSRA), power control scaling update strategies (fixed or
keep updating after each removal), we have in total twelve schemes as shown
in Table 3.4. The flow chart of all the joint admission and power control
schemes are shown in Fig. 3.5.
These schemes can be treated as variants from JAPC-MRER and JAPC-
MSRA described in Section 3.2. Specially, the schemes in Section 3.2 is a

42
3.3 Discussions on power control and pre-admission schemes

Figure 3.5: The flow chart of joint admission and power control schemes for
one-channel CogCell

43
 Power and Admission Control for One-channel Cognitive Radio
Cellular Networks

Table 3.4: Joint admission and power control schemes for one-channel CogCell

Main Sub Category Schemes Pre-admission met-

Category rics
JAPC-MRER- JAPC-MRER-y- Maximum y removal
fixed fixed
JAPC-MRER- Minimum SINR re-
SINR-fixed moval
JAPC- (β is fixed after JAPC-MRER- Minimum channel
MRER each removal) gain-fixed gain removal
JAPC-MRER- JAPC-MRER-y Maximum y removal
dynamic
JAPC-MRER- Minimum SINR re-
SINR moval
(β updates after JAPC-MRER- Minimum channel
each removal) gain gain removal
JAPC-MSRA- JAPC-MSRA-y- Maximum y removal
fixed fixed
JAPC-MSRA- Minimum SINR re-
SINR-fixed moval
JAPC- (β is fixed after JAPC-MSRA- Minimum channel
MSRA each removal) gain-fixed gain removal
JAPC-MSRA- JAPC-MSRA-y Maximum y removal
dynamic
JAPC-MSRA- Minimum SINR re-
SINR moval
(β updates after JAPC-MSRA- Minimum channel
each removal) gain gain removal

special case when we fix β after each removal. We denote the general fixed β
schemes as JAPC-MRER-fixed and JAPC-MSRA-fixed, respectively. On the
other hand, if β keeps updating after each removal, we denote such schemes as
JAPC-MRER-dynamic and JAPC-MSRA-dynamic, respectively. In the case
of fixing β, when we apply the maximum y removal pre-admission metric,
we get JAPC-MRER-y-fixed and JAPC-MSRA-y-fixed, when we apply the
minimum SINR removal pre-admission metric, we get JAPC-MRER-SINR-
fixed and JAPC-MSRA-SINR-fixed, when we apply the minimum channel
gain removal, we get JAPC-MRER-gain-fixed and JAPC-MSRA-gain-fixed.

44
3.4 Further improvements

Otherwise, if β keeps on updating after each removal, we get JAPC-MRER-y

and JAPC-MSRA-y (when applying maximum y removal pre-admission met-
ric), JAPC-MRER-SINR and JAPC-MSRA-SINR (when applying minimum
SINR removal pre-admission metric), JAPC-MRER-gain and JAPC-MSRA-
gain (when applying minimum channel gain removal pre-admission metric).

3.4 Further improvements

In this section, we improve the algorithm proposed in the previous sections by
transforming the formulation. We rewrite the original problem formulation
as follows.
ns
X
maximize ri xi (3.30)
i=1

subject to:

ns
X
hij Pi xi ≤ Γj , ∀j ∈ Np (3.31)
i=1
xi ∈ {0, 1}, ∀i ∈ Ns (3.32)
ξi ≥ ξ¯i , if xi = 1, ∀i ∈ Ns (3.33)
Pi ∈ [0, P̂ ], ∀i ∈ Ns (3.34)

Let Ns∗ denote the optimal solution for admitted SUs is a subset of Ns .
Thus,

1, i ∈ Ns∗

xi =
0, otherwise
From (3.33), we have

hi Pi
ξi = ≥ ξ¯i , ∀i ∈ Ns∗
N0 + Ip + Is − hi Pi
When any admitted SU i achieves the required SINR,

hi Pi
= ξ¯i , ∀i ∈ Ns∗
N0 + Ip + Is − hi Pi

Thus
−1
hi Pi = (N0 + Ip + Is )/(1 + ξ¯i ), ∀i ∈ Ns∗

45
Power and Admission Control for One-channel Cognitive Radio
Cellular Networks

−1
Let ai denote 1 + ξ¯i , we have

hi Pi = (N0 + Ip + Is )/ai , ∀i ∈ Ns∗ (3.35)

We add up all the admitted SUs as follows
X X
hi Pi = (N0 + Ip + Is )/ai
i∈Ns∗ i∈Ns∗

The left side is equal to Is , thus

X
Is = (N0 + Ip + Is ) a−1
i , ∀i ∈ Ns∗
i∈Ns∗

After solving the above equation for Is , we can get

N0 + Ip
Is = !−1 (3.36)
a−1
P
i −1
i∈Ns∗

Substituting (3.36) into (3.35), we can get the solution of power

N0 +Ip +Is
Pi =  hi ai 
N0 +Ip
= N0 + Ip +  /(hi ai )
 
!−1
P
a−1 −1 (3.37)
k
k∈Ns∗
N0 +Ip
= P −1
!
hi ai 1− ak
k∈Ns∗

Substituting (3.37) into the constraint in (3.31), we have

X N0 + Ip
hij ! ≤ Γj , ∀j ∈ Np
i∈Ns∗
a−1
P
hi ai 1− k
k∈Ns∗

Therefore,
 
X X
hij h−1 −1
i ai (N0 + Ip ) ≤ Γj
1 − a−1
k
, ∀j ∈ Np
i∈Ns∗ k∈Ns∗

Then, we can obtain the following constraint

46
3.4 Further improvements

X
hij h−1 −1

−1
i Γj (N0 + Ip ) + 1 ai ≤ 1, ∀j ∈ Np
i∈Ns∗

It is equal to the following constraint

X
hij h−1 −1

−1
i Γj (N0 + Ip ) + 1 ai xi ≤ 1, ∀j ∈ Np (3.38)
i∈Ns

To obey constraint (3.34), we have

N0 + Ip
0≤ ! ≤ P̂ , ∀i ∈ Ns∗
a−1
P
hi ai 1− k
k∈Ns∗

which turns out to be

X N0 + Ip
a−1
k ≤ 1− ∀i ∈ Ns∗ (3.39)
k∈Ns∗
hi ai P̂

3.4.1 Lower and upper bounds

We introduce z, where z = min{hi ai , ∀i ∈ Ns∗ }. According to (3.39) we have
X N0 + Ip
a−1
k ≤ 1−
k∈N ∗
z P̂
s

It is equal to the following constraint

X N0 + Ip
a−1
i xi ≤ 1 − (3.40)
i∈Ns
z P̂
In summary, the original optimization problem can be transformed into
the following formulated problem.
X
maximize ri xi
i∈Ns
X
hij h−1 −1

−1
subject to i Γj (N0 + Ip ) + 1 ai xi ≤ 1, ∀j ∈ Np (3.41)
i∈Ns
X N0 + Ip
a−1
i xi ≤ 1 − , z = min hi ai (3.42)
z P̂ xi =1
i∈Ns

xi ∈ {0, 1} ∀i ∈ Ns
The constraint (3.42) dynamically changes according to different admitted
SUs.

47
 Power and Admission Control for One-channel Cognitive Radio
Cellular Networks

Constraints Reduction

We can further examine the relationship between the two constraints in (3.41)
and (3.42).

X X
a−1
i xi ≤ 1 − hij h−1 −1 −1
i Γj (N0 + Ip )ai xi , ∀j ∈ Np (3.43)
i∈Ns i∈Ns
X N0 + Ip
a−1
i xi ≤ 1 − (3.44)
i∈Ns
z P̂

In the right of the above two constraints, we need to show the constraint
that has higher value. Since both are larger than 0, we can compare the ratio
to 1.

hij h−1 −1 −1
P
i Γj (N0 + Ip )ai xi
i∈Ns
R= N0 +Ip
z P̂ (3.45)
X
−1 −1 −1
=z P̂ hij hi Γj ai xi
i∈Ns

The lower bound of the ratio can be calculated as follows,

X
R =z P̂ hij h−1 −1 −1
i Γ j ai x i
i∈Ns
X hk ak
=P̂ hkj Γ−1
j + P̂ hij Γ−1
j , z = hk ak , k ∈ Ns∗
i∈Ns∗
hi ai
(3.46)
≥P̂ hkj Γ−1
j

≥P̂ Γ−1
j min∗ hij
i∈Ns

≥P̂ Γ−1
j min hij , ∀j ∈ Np
i∈Ns

Let RL represent the lower bound of R

 
RL = max Γ−1
j min hij P̂ (3.47)
j∈Np i∈Ns

48
3.4 Further improvements

The upper bound of the ratio can be calculated as follows,

X
R =z P̂ hij h−1 −1 −1
i Γj ai
i∈Ns∗
X hk ak
=P̂ Γ−1
j hij
∗
i∈Ns
h i ai
X (3.48)
−1
≤P̂ Γj hij
i∈Ns∗
X
≤P̂ Γ−1
j hij , ∀j ∈ Np
i∈Ns

Let RU represent the upper bound of R

!
X
RU = min Γ−1
j hij P̂ (3.49)
j∈Np
i∈Ns

Then we have the following three cases.

• If RL ≥ 1, then R ≥ 1, the problem formulation can be modified as
follows.
P
maximize ri xi
i∈Ns
hij h−1 −1
P
−1
subject to i Γ j (N0 + Ip ) + 1 ai xi ≤ 1, ∀j ∈ Np
i∈Ns
xi ∈ {0, 1}, ∀i ∈ Ns
(3.50)
• If RU ≤ 1, then R ≤ 1, the problem formulation can be modified as
follows.
P
maximize ri xi
i∈Ns
ai xi ≤ 1 − N0z+I
P −1 p
subject to P̂
, z = min hi ai (3.51)
i∈Ns xi =1
xi ∈ {0, 1}, ∀i ∈ Ns

• Otherwise, all the original constraints should be considered.

The lower bound with tightening of constraints

In (3.42), a tighter constraint can be obtained, if we get the value z from all
the SUs no matter it is admitted or not.
z ≥ min hi ai
i∈Ns

49
 Power and Admission Control for One-channel Cognitive Radio
Cellular Networks

Therefore, we can obtain the following tighter constraint when z is equal

to min hi ai .
i∈Ns
X N0 + Ip
a−1
i xi ≤ 1 − (3.52)
i∈Ns
min hi ai P̂
i∈Ns

Thus,
P
maximize r i xi
i∈N
Ps
hij h−1 −1

−1
subject to i Γj (N0 + Ip ) + 1 ai xi ≤ 1, ∀j ∈ Np
i∈Ns (3.53)
N0 +Ip
a−1
P
i xi ≤ 1 − min hi ai P̂
i∈Ns i∈Ns

xi ∈ {0, 1}, ∀i ∈ Ns

The upper bound with relaxation of constraints

In (3.42), a looser constraint can be obtained, if we get the value z as the
maximal value of the product of hi and ai from all the SUs no matter it is
admitted or not.
z ≤ max hi ai
i∈Ns

Therefore, we can obtain the following relaxed constraint when z is equal

to max hi ai .
i∈Ns
X N0 + Ip
a−1
i xi ≤ 1 − (3.54)
i∈Ns
max hi ai P̂
i∈Ns

Thus,
P
maximize r i xi
i∈N
Ps
hij h−1 −1

−1
subject to i Γj (N0 + Ip ) + 1 ai xi ≤ 1, ∀j ∈ Np
i∈Ns (3.55)
N0 +Ip
a−1
P
i xi ≤ 1 − max hi ai P̂
i∈Ns i∈Ns

xi ∈ {0, 1}, ∀i ∈ Ns

3.4.2 Multidimensional knapsack problem modeling and

solutions
According to (3.39) we have
X N0 + Ip
a−1
k + ≤ 1, ∀i ∈ Ns∗
k∈Ns∗
hi ai P̂

50
3.4 Further improvements

which is equal to the following inequality,

X N0 + Ip
a−1
k xk + h−1 −1
i ai xi ≤ 1, ∀i ∈ Ns
k∈Ns
P̂

Then we have
 
X N0 + Ip −1 −1
a−1
k xk + 1+ hi ai xi ≤ 1, ∀i ∈ Ns (3.56)
k∈Ns ,k6=i
P̂

Therefore, the reformulated problem is as follows.

P
maximize ri xi
i∈N
Ps
hij h−1 −1

−1
subject to i Γ j (N 0 + Ip ) + 1 ai xi ≤ 1, ∀j ∈ Np
i∈Ns   (3.57)
N0 +Ip −1
a−1 hi a−1
P
k xk + 1 + P̂ i xi ≤ 1, ∀i ∈ Ns
k∈Ns ,k6=i
xi ∈ {0, 1}, ∀i ∈ Ns

The above formulation can be further written in a canonical matrix form,

which is useful for computer aided optimization tools, such as MOSEK [37]
and CPLEX [98].

maximize Cx
subject to Ax ≤ 1 (3.58)
x ∈ {0, 1}

where
1 = (1, 1, ..., 1)T

x = (x1 , x2 , ..., xns )T

C = (r1 , r2 , ..., rns )

51
 Power and Admission Control for One-channel Cognitive Radio
Cellular Networks

 A(np +ns )×ns = 

h1,1 (N0 +Ip )+h1 Γ1 h2,1 (N0 +Ip )+h2 Γ1 hns ,1 (N0 +Ip )+hns Γ1
a1 h1 Γ1 a2 h2 Γ1
. . . ans hns Γ1
 h1,2 (N0 +Ip )+h1 Γ2 h2,2 (N0 +Ip )+h2 Γ2 hns ,2 (N0 +Ip )+hns Γ2 

 a1 h1 Γ1 a2 h2 Γ2
. . . ans hns Γ2



 . . . . . . 


 . . . . . . 

hij (N0 +Ip )+hi Γj
. . . . .
 

 ai hi Γj 


 . . . . . . 

 . . . . . . 
h1,np (N0 +Ip )+h1 Γnp h2,np (N0 +Ip )+h2 Γnp hns ,np (N0 +Ip )+hns Γnp
 

 a1 h1 Γnp a2 h2 Γnp
. . . ans hns Γnp


N0 +Ip +h1 P̂
 

 a1 h1 P̂
a−1
2 . . . a−1
ns


N0 +Ip +h2 P̂
a−1 a−1
 

 1 a2 h2 P̂
. . . ns



 . . . . . . 


 . . . . . . 

N0 +Ip +hi P̂
. . . . .
 
 ai hi P̂

 

 . . . . . . 


 . . . . . . 

N0 +Ip +hns P̂
a−1
1 a−1
2 . . . ans hns P̂

The above formulation is a 0 − 1 integer linear programming, and can be

analogous to a multidimensional knapsack problem, where there are (ns +np )
constraints as the dimensions.
P
maximize r i xi
i∈N
Ps
subject to wij xi ≤ 1, j = 1, .., ns + np (3.59)
i∈Ns
xi ∈ {0, 1}, ∀i ∈ Ns
where
−1 −1

−1
 h
 ij hi Γj (N0 + Ip ) + 1 ai 1 ≤ j ≤ np
−1
wij = a
i  1 + np ≤ j ≤ ns + np , j 6= (i + np )
 1+ N 0 +I p −1 −1
hi ai j = i + np

P̂
(3.60)
The multidimensional knapsack problem is classified as an NP-hard op-
timization problem, which cannot be solved in polynomial time [99]. Many
algorithms, which can be categorized into two types: exact algorithms and
heuristic algorithms, have been proposed to solve this problem. The branch-
and-bound method [100], and dynamic programming [101] [102] can be used

52
3.4 Further improvements

to find the exact optimal solution of the problem. However, these meth-
ods have high computation load [99]. Heuristic algorithms, which aim to
compute feasible solutions of “reasonable quality” within “reasonable running
time” [99], is more feasible than the optimal algorithms. Typical heuristic al-
gorithms include greedy-type heuristic algorithm, relaxation-based heuristic
algorithm, etc.

3.4.3 Proposed efficiency based heuristics method

In this study, we propose an efficiency based heuristics algorithm called
JAPC-MKP. The heuristics method proceeds iteratively. Let Ncs denote
the original set of constraints index from 1 to ns + np , which follows the
same sequence of the row index for the matrix A. In each iteration, it will
exam the valid constraints, and remove one SU in each constraint. Let Ncs∗
denote the valid constraints, and Ns∗ represent the set of admitted SUs in
each iteration.
The order to remove one SU in each iteration is according to the efficiency
which is defined as follows.
ri
ei = P (3.61)
αj wij
∗
j∈Ncs

The smaller the efficiency is, the higher probability to remove that SU. In
each iteration, we remove the SU with the minimal efficiency. In equation
(3.61) αj is called relevance value of constraint j. It shows the importance to
every constraint. The higher the relevance value of a constraint, the higher
the scarcity of the corresponding resource is. It then becomes less attractive
to pack an SU which consumes a lot of that resource. In our study, we define
αj as follows. X
αj = wij − 1, j ∈ Ncs∗ (3.62)
i∈Ns∗

This method is shown in Algorithm 3. The complexity is dominated by

the calculation of αj in each iteration. In the worst case, it has maximal
ns + np iteration, the cardinality of |Ns∗ | reduces by one at each iteration
from ns to 1. The time complexity can be calculated as follows.

T (ns , np ) = (ns + (ns − 1) + (ns − 2) + ...)(ns + np ) calculate αj

≤ (ns + (ns − 1) + (ns − 2) + ... + 1)(ns + np )
= ns (1+n
2
s)
(ns + np )
= O(n2s (ns + np ))
(3.63)

53
 Power and Admission Control for One-channel Cognitive Radio
Cellular Networks

Thus the time complexity at worst case is O(n2s (ns + np )).

Algorithm 3 JAPC-MKP
Input: Ns , Np , Ncs
Output: Ns∗ , {Pi }
1: Initialization: Ns∗ ← Ns , Np∗ ← Np , Ncs ∗
← Ncs .
∗ ∗
2: Calculate wij , ∀i ∈ Ns , j ∈ Ncs , according to (3.60)
∗
3: while Ncs 6= ∅ do
4: for j ∈ Ncs∗ do
5: Calculate αj according to (3.62)
6: if αj ≤ 0 then
∗ ∗
7: Nnc ← Nnc −j
∗
8: if Nnc == ∅ then
9: “All constraints are satisfied”
10: Return;
11: end if
12: end if
13: end for
14: for i ∈ Ns∗ do
15: Calculate ei according to (3.61).
16: end for
17: Choose an SU i∗ , where ei∗ = min∗ ei
∀i∈Nnc
18: xi ∗ ← 0
19: Pi ∗ ← 0
20: Ns∗ ← Ns∗ − i∗
21: if Ns∗ == ∅ then
22: “No SU can be admitted”
23: Return;
24: end if
25: end while
26: for i ∈ Ns∗ do
27: xi ← 1
28: Calculate Pi according to (3.37).
29: end for

3.5 Simulation results

In this section, we evaluate the performance of all the proposed schemes in
Section 3.3 and Section 3.4. The schemes in Section 3.2 is a special case of the

54
3.5 Simulation results

fixed β updating schemes in Section 3.3. In the convenience for comparison

in this simulation, we denote the JAPC-MRER and JAPC-MSRA schemes
in Section 3.2 as JAPC-MRER-fixed and JAPC-MSRA-fixed, respectively.
Where JAPC-MRER-fixed includes JAPC-MRER-y-fixed, JAPC-MRER-SINR-
fixed, and JAPC-MRER-gain-fixed, while JAPC-MSRA-fixed includes JAPC-
MSRA-fixed includes JAPC-MSRA-y-fixed, JAPC-MSRA-SINR-fixed, and
JAPC-MSRA-gain-fixed for the three different pre-admission control met-
rics. We denote the schemes proposed in Section 3.3 as JAPC-MRER-
dynamic and JAPC-MSRA-dynamic, respectively. Where JAPC-MRER-
dynamic includes JAPC-MRER-y, JAPC-MRER-SINR, and JAPC-MRER-
gain, while JAPC-MSRA-dynamic includes JAPC-MSRA-y, JAPC-MSRA-
SINR, JAPC-MSRA-gain for the three different pre-admission control met-
rics.
We use the CogCell simulator specified in Section 3.2.4, where we generate
100 random topologies. In each topology, we randomly generate the DTR
demands. In addition, we use the same DTR and SINR mapping table as
Table 3.3. The noise power N0 is set as 10−14 W. The other simulation
parameters are the same as in Table 3.2 except the power scaling factor
β. In this simulation, the power scaling factor is calculated by (3.29) after
the pre-admission procedures for the schemes proposed in Section 3.3. For
JAPC-MKP proposed in Section 3.4, there is no need for this parameter.
In the following simulation, we evaluate the performance in terms of the
secondary revenues in four different cases, i.e., changing the number of PUs,
changing the number of SUs, changing the interference threshold (Γ), and
changing the interference to BS from primary systems (Ip ).
In each case, with the given number of PUs and SUs, the value of Γ and
and Ip , we solve the optimization problem by MOSEK software [37], and
run the joint admission and power control schemes, i.e., JAPC-MKP, JAPC-
MRER-dynamic, JAPC-MRER-fixed, and JAPC-MSRA-dynamic. Then, we
calculate the average secondary revenue based on the results in the 100 ran-
dom topologies.

3.5.1 Effect of the number of primary users

In the following, we evaluate the secondary revenue when changing the num-
ber of PUs. The other parameters are fixed as follows, ns = 50, Γ =
−100dBW , and Ip = −110dBW .
The results are shown in Fig. 3.6, where the secondary revenue decreases
with the increasing number of PUs. The reason is that more PUs will add
more constraints on interference generated by SUs. We can see that our
proposed JAPC-MKP can approach the optimal results obtained by the

55
 Power and Admission Control for One-channel Cognitive Radio
Cellular Networks

450
Optimal JAPC−MRER−gain
400 JAPC−MKP JAPC−MRER−y−fixed
JAPC−MRER−y JAPC−MRER−SINR−fixed
350 JAPC−MRER−SINR JAPC−MRER−gain−fixed
JAPC−MSRA−y
300 JAPC−MSRA−SINR
Secondary Revenue

JAPC−MSRA−gain
250

200

150

100

50

0
0 10 20 30 40 50
Number of PUs

Figure 3.6: Secondary Revenue in terms of number of PUs (ns = 50, Γ =

−100dBW , and Ip = −110dBW )

optimization software MOSEK, and achieve higher revenue than all other
schemes. JAPC-MRER-SINR achieves the least revenue than all the other
schemes, the reason is as follows. From (3.27), we see that if the constraints !
1
P
cannot be satisfied, parts of the left side of that inequality 1 − 1+ξ¯i
−1
∗
i∈Ns
!
1
P
is too large. To reduce the value of 1 − 1+ξ¯
−1 , we may need to re-
i
i∈Ns∗
move more
P SUs1 with minimal SINR metric, since the smaller SINR is, the
smaller 1+ξ¯
−1 is. JAPC-MRER-y achieves the highest revenue among
i
i∈Ns∗
all the three pre-admission control metrics. The reason can be found from
(3.27), where ŷ reduces after removing the SU with maximum y, thus the left
part of the inequality can increase more efficiently than other two metrics.
The dynamic β updating strategy can achieve higher revenue than the fixed
β strategy in all kinds of pre-admission control schemes. This is because
that with the updated β calculated by (3.29) after each removal, β decreases

56
3.5 Simulation results

and results in power decreasing for all prospective SUs, which further results
in less interference to PUs. Therefore, more SUs can be admitted compar-
ing to the fixed β scheme. JAPC-MRER can achieve higher revenue than
JAPC-MSRA for all kinds of pre-admission schemes in the case of dynamic
updating β strategy.

3.5.2 Effect of the number of secondary users

In the following, we evaluate the secondary revenue when changing the num-
ber of SUs. The other parameters are fixed as follows, np = 50, Γ =
−100dBW , and Ip = −110dBW .

200
JAPC−MRER−SINR Optimal
180 JAPC−MRER−gain JAPC−MKP
JAPC−MRER−y−fixed JAPC−MRER−y
160 JAPC−MRER−SINR−fixed
JAPC−MRER−gain−fixed
140
JAPC−MSRA−y
Secondary Revenue

120 JAPC−MSRA−SINR
JAPC−MSRA−gain
100

80

60

40

20

0
0 10 20 30 40 50
Number of SUs

Figure 3.7: Secondary revenue in terms of number of SUs (np = 50, Γ =

−100dBW , and Ip = −110dBW )

The results are shown in Fig. 3.7, where the secondary revenue increases
with the increasing number of SUs. The reason is that with the increasing
number of SUs, the number of SUs with trivial interference can increase and
contribute on the increasing revenue to BS. We can see that our proposed
JAPC-MKP can approach the optimal results obtained by the optimiza-

57
 Power and Admission Control for One-channel Cognitive Radio
Cellular Networks

tion software MOSEK, and achieve higher revenue than all other schemes.
With the same reason as explained in the previous case, JAPC-MRER-SINR
achieves the least revenue than all the other schemes, and JAPC-MRER-y
achieves the highest revenue among all the three pre-admission control met-
rics, and the dynamic β updating strategy can achieve higher revenue than
the fixed β strategy in all kinds of pre-admission control schemes. JAPC-
MRER can achieve higher revenue than JAPC-MSRA for all kinds of pre-
admission schemes in the case of dynamic updating β strategy.

3.5.3 Effect of the interference threshold on primary

users
In the following, we evaluate the secondary revenue when changing Γ, which
is the interference threshold on PUs. The other parameters are fixed as
follows, there are 50 SUs, 50 PUs, and Ip = −110dBW .

400
Optimal
350 JAPC−MKP
JAPC−MRER−y
JAPC−MRER−SINR
300
JAPC−MRER−gain
JAPC−MRER−y−fixed
Secondary Revenue

250 JAPC−MRER−SINR−fixed
JAPC−MRER−gain−fixed
200 JAPC−MSRA−y
JAPC−MSRA−SINR
JAPC−MSRA−gain
150

100

50

0
−150 −100 −50
Interference threshold on primary users (dBW)

Figure 3.8: Secondary Revenue in terms of Γ (ns = 50, np = 50, and Ip =

−110dBW )

The results are shown in Fig. 3.8. Again, JAPC-MKP achieves revenue
closely to the optimal result from MOSEK, and outperform all other schemes.
When Γ is less than −130dBW , the interference constraints are too strict
that no SUs can be admitted into the system. When using JAPC-MRER the

58
3.5 Simulation results

dynamic β updating schemes achieve higher revenue than the fixed β updat-
ing scheme when Γ is less than −70dBW and larger than −130dBW . The
maximum y removal pre-admission control scheme always outperform the
minimal SINR and channel gain removal schemes using either JAPC-MRER
or JAPC-MSRA. JAPC-MRER outperforms JAPC-MSRA when using the
same pre-admission control scheme by using either dynamic or fixed β up-
dating strategy.

3.5.4 Effect of the interference to BS from primary

systems
In the following, we evaluate the secondary revenue when changing Ip , which
is the interference to BS from primary systems. The other parameters are
fixed as follows, ns = 50, np = 50, and Γ = −100dBW .

450
Optimal
400 JAPC−MKP
JAPC−MRER−y
JAPC−MRER−SINR
350 JAPC−MRER−gain
JAPC−MRER−y−fixed
300 JAPC−MRER−SINR−fixed
Secondary Revenue

JAPC−MRER−gain−fixed
250 JAPC−MSRA−y
JAPC−MSRA−SINR
JAPC−MSRA−gain
200

150

100

50

0
−150 −100 −50
Interference to BS from primary systems (dBW)

Figure 3.9: Secondary Revenue in terms of Ip (ns = 50, np = 50, and Γ =

−100dBW )

The results are shown in Fig. 3.9. Again, JAPC-MKP achieves revenue
closely to the optimal result from MOSEK, and outperform all other schemes.
When Ip is greater than −80dBW , the interference to BS is too strong that
no SUs can be admitted into the system. When Ip is less than −80dBW , the
dynamic β updating schemes achieve higher revenue than the fixed β updat-
ing scheme using JAPC-MRER. The maximum y removal pre-admission con-

59
 Power and Admission Control for One-channel Cognitive Radio
Cellular Networks

trol scheme always outperform the minimal SINR and channel gain removal
schemes using either JAPC-MRER or JAPC-MSRA. JAPC-MRER outper-
forms JAPC-MSRA when using the same pre-admission control scheme by
using either dynamic or fixed β updating strategy.

3.6 Conclusion
In this chapter, we have investigated the problem of maximizing the sec-
ondary revenue of the CogCell, while satisfying the QoS (in terms of DTR) re-
quirements on SUs and guaranteeing the interference constraints on PUs. To
solve this optimization problem, we first introduced a revenue efficiency factor
to search for the SUs with high revenue and also low interference, and pro-
posed JAPC-MRER. The time complexity is O(n2s np ), which is the same as
the other two algorithms used in JAPC-MSRA and JAPC-Rand. Simulation
results indicated that our proposed JAPC-MRER can achieve much higher
secondary revenue for the operator than the other two schemes. We further
improved our proposed JAPC-MRER by pre-admission control schemes and
dynamic updating the power scale after each removal. Simulation results
showed that JAPC-MRER-dynamic schemes can achieve higher secondary
revenue than JAPC-MRER-fixed schemes with all kinds of pre-admission
control schemes. The minimal y removal pre-admission control scheme can
achieve higher secondary revenue than other pre-admission control schemes.
In the end, we transformed the operator problem to an instance of multidi-
mensional knapsack problem, and proposed a heuristic scheme called JAPC-
MKP with O(n2s (ns + np )) time complexity. Simulation results showed that
JAPC-MKP can approach the optimal results obtained by the optimization
software MOSEK, and achieve higher secondary revenue than all other pro-
posed schemes.

60
Chapter 4

Resource Optimization for

Multi-channel Cognitive Radio
Cellular Networks

In the previous chapter, we focused on the power and admission control for
one-channel CogCell. We now consider the multi-channel CogCell, where
channel allocation strategies should be taken into account. In this chapter,
we study the operator problem again to maximize the secondary revenue in
multi-channel CogCell. We formulate this problem as an instance of multi-
dimensional multiple knapsack problem, and proposed a heuristic method.
The rest of this chapter is organized as follows. We introduce the system
model and formulate the optimization problem in Section 4.1. In Section 4.2,
we model the operator problem as an MMKP. We then transfer the MMKP to
MKP in Section 4.3, and present our proposed heuristic scheme in Section 4.4.
In Section 4.5, we introduce a traditional channel allocation scheme based on
SINR together with JAPC-MKP which has been proposed in Chapter 3. In
Section 4.6, we evaluate the performance of different schemes by simulation.
Finally, we draw the conclusions in Section 4.7.

4.1 System model and problem formulations

In this section, we introduce the system model, the coexistence condition of
PUs and SUs, the definition of interference from SUs to PUs, and the uplink
throughput of SUs. At the end of this section, we formulate the operator
problem.

61
 Resource Optimization for Multi-channel Cognitive Radio
Cellular Networks

4.1.1 System model

We consider a multi-channel CDMA CogCell in a certain area with multiple
PUs, SUs, and channels as illustrated in Fig. 4.1. The set of channels is
denoted as Nc . These channels are licensed to the primary systems in that
area. A secondary BS is deployed to serve a set of SUs (denoted by Ns ) using
the channels from Nc . For a given channel m (m ∈ Nc ), a set of Npm PUs will
receive interference from SUs working in the same channel m. Those PUs are
not part of the CogCell users, but existing in the CogCell area. They may
be fixed or mobile. The interference power received by BS from the primary
transmitters can be measured by the BS, and we denote it as Ipm for any
channel m. BS will allocate a channel for each SU. The allocation rule should
guarantee that the interference to those PUs in any channel m should not
exceed the tolerable interference threshold.
This model is different with OFDMA mode as follows. In OFDMA mode,
wideband channels are divided into several sub-carriers, which logically form
subchannels. Each subchannel are further allocated to one user at one time.
The BS should decide how to allocate the subchannels to SUs, and each
SU can utilize several subchannels at the same time. But in multi-channel
CDMA mode, the BS should decide how to allocate the wideband channels
to each SU, which can only use one of the wideband channels.
Table 4.1 lists the notations used in this chapter. Channel gain infor-
mation between SUs and BS, between SUs and PUs in receiving mode is
assumed to be estimated by the same method as in [79].

Figure 4.1: System model of multi-channel cognitive radio cellular networks

62
4.1 System model and problem formulations

Table 4.1: Table of notations for multi-channel CogCell

Symbol Meaning
Ns the set of SUs
Nc the set of Channels
Nsm the set of SUs on channel m
Npm the set of PUs on channel m
ns the number of SUs, equal to |Ns |
nc the number of Channels, equal to |Nc |
npm the number of PUs in receiving mode on channel m
np the total number of PUs in receiving mode on all channels
Pmi the transmission power of SU i on channel m
P̂ the maximum transmission power of every SU
Ipm the interference on channel m measured at BS from primary systems
ξmi the SINR of SU i on channel m measured at the BS
ξ¯i the minimum uplink SINR requirement of SU i
λ̄i the data rate required by SU i
τmij the interference from SU i to PU j on channel m
Γmj the interference threshold of PU j on channel m
ri the payment from SU i (the revenue got from SU i)
gmi the channel gain from SU i to the BS on channel m
hmij the channel gain from SU i to PU j on channel m

4.1.2 Interference from SUs to PUs

While SUs and PUs are coexistent in the underlay mode, SUs cause interfer-
ence to the PUs which are receiving data from their primary transmitters. On
channel m (m ∈ Nc ), the interference power τmij received by PU j (j ∈ Npm )
caused by SU i (i ∈ Ns ) is given by

τmij = hmij Pmi (4.1)

where Pmi is the transmission power of SU i on channel m. hmij is the channel

gain from SU i to PU j on channel m.
According to the coexisting rule in the underlay mode, the total interfer-
ence power received by any PU j on channel m should be no more than the
predefined threshold Γmj . That is
X
hmij Pmi ≤ Γmj , ∀j ∈ Npm . (4.2)
i∈Ns

63
 Resource Optimization for Multi-channel Cognitive Radio
Cellular Networks

4.1.3 Uplink capacity of SUs

Let ξmi denote the Signal-Interference-plus-Noise-Ratio (SINR) at the BS for
the transmitter SU i (i ∈ Nsm ) on channel m. According to the definition of
SINR, we can obtain
ξmi = βmi Pmi (4.3)
where gmi
βmi = m
N0 +Ipm +I−i
gmi
= N0 +Ipm +
P
gmk Pkm (4.4)
k∈Nsm ,k6=i
gmi
= N0 +Ipm +Ism −gmi Pmi

where gmi is the channel gain from SU i to the BS on channel m. N0 denotes

the average power of background noise received by the BS at every channel.
m
Ipm represents the interference on channel m at the BS caused by PUs. I−i
denotes the interference power received by the BS from all the SUs on channel
m
m except SU i. In practice, as explained in [103], Ipm and I−i can be measured
at the BS.
Suppose that the bandwidth of the channel m is Bm , according to Shan-
non’s capacity theory, we can obtain the capacity for SU i on channel m as
follows.

λmi = Bm log2 (1 + ξmi ), ∀m ∈ Nc (4.5)

If SU i is allowed to access from channel m, the data rate is satisfied as
follows,
λmi ≥ λ̄i
It is then equivalent to the following SINR requirements according to
(4.5).

ξmi ≥ ξ¯i

4.1.4 The operator problem

We call the sum of revenue got from all admitted SUs Secondary Revenue.
The operator problem is how to admit a subset of SUs and allocate a chan-
nel for each SU and control its transmission power such that the secondary
revenue can be maximized. In the mean time, the solution should obey the
constraints of maximum transmission power, minimum SINR for SUs, and
maximum interference to every PU. In the following, we denote this problem
as Maximization of Revenue from SUs (MRS) problem for short. It can be
formulated as follows.

64
4.2 MMKP modeling

MRS: X X
maximize ri xmi (4.6)
i∈Ns m∈Nc

subject to:
X
hmij Pmi ≤ Γmj , ∀j ∈ Npm , m ∈ Nc (4.7)
i∈Ns
xmi ∈ {0, 1}, ∀i ∈ Ns , m ∈ Nc (4.8)
X
xmi ≤ 1, ∀i ∈ Ns (4.9)
m∈Nc

ξmi ≥ ξ¯i , if xmi = 0, ∀i ∈ Ns , m ∈ Nc (4.10)

Pmi ∈ [0, P̂ ], ∀i ∈ Ns , m ∈ Nc (4.11)
Pmi = 0, if xmi = 0, ∀i ∈ Ns , m ∈ Nc (4.12)

where xmi is a binary variable, xmi = 1 if SU i works on channel m, zero

otherwise. Constraint (4.7) means that the interference from SUs to PUs can
not exceed the predefined threshold. Constraint (4.8) and (4.9) indicate that
every SU can only occupy one channel at most. Constraint (4.10) represents
the minimum SINR requirement of all SUs working on channel m. Con-
straint (4.11) shows the power constraint of SUs, wherein P̂ is the maximum
transmission power that can be used at SUs.
The solution is to find the value of the binary variables xmi and every
SU’s transmission power Pmi , which may not be an integer. Moreover, the
objective function (4.6) is linear to the binary variables. Therefore, this
optimization problem is a Mixed Integer Non-linear Programming (MINLP)
problem, which is NP-hard in general.

4.2 MMKP modeling

For constraint (4.10) in MRS problem, we can use the result from Chapter 3.
According to (3.37) about the power calculation in one-channel CogCell, we
can have the similar calculation for an SU in any channel m.
N0 +Ipm +Ism
Pmi = gmi ai
 
N0 +Ipm
= N0 + Ipm +  /(gmi ai )
 
!−1
P
a−1 −1
(4.13)
k
k∈Nsm
N0 +Ipm
= !
a−1
P
gmi ai 1− k
k∈Nsm

65
 Resource Optimization for Multi-channel Cognitive Radio
Cellular Networks

−1
where ai denotes 1 + ξ¯i . Substitute (4.13) into the constraint in (4.7), we
have
X N + Ipm
hmij  0  ≤ Γmj , ∀j ∈ Npm , m ∈ Nc
P −1
i∈Nsm gmi ai 1 − ak
k∈Nsm

Therefore,
!
X X
−1 −1
hmij gmi ai (N0 + Ipm ) ≤ Γmj 1− a−1
k , ∀j ∈ Npm , m ∈ Nc
i∈Nsm k∈Nsm

Then,
X
−1 −1
Γmj (N0 + Ipm ) + 1 a−1


hmij gmi i ≤ 1, ∀j ∈ Npm , m ∈ Nc
i∈Nsm

It is equivalent to the following constraint

X
−1 −1
Γmj (N0 + Ipm ) + 1 a−1


hmij gmi i xmi ≤ 1, ∀j ∈ Npm , m ∈ Nc (4.14)
i∈Ns

On the other hand, to obey constraint (4.11), we have

N + Ipm
0≤  0  ≤ P̂ , ∀i ∈ Nsm , m ∈ Nc
P −1
gmi ai 1 − ak
k∈Nsm

It can be rewritten as
X N0 + Ipm
a−1
k + ≤ 1, ∀i ∈ Nsm , m ∈ Nc
k∈Nsm
gmi ai P̂

which is equivalent to the following constraint

X
a−1
k xkm + (N0 + Ipm )P̂
−1 −1 −1
gmi ai xmi ≤ 1, ∀i ∈ Ns , m ∈ Nc (4.15)
k∈Ns

Combining (4.14) and (4.15), we have the following constraint.

X
wmij xmi ≤ 1, ∀j = 1, 2, · · · , ns + npm , m ∈ Nc (4.16)
i∈Ns

66
4.3 MKP modeling by matrix transformation

where, for each i in Ns , and m in Nc , we have

−1 −1

−1
 h mij g Γ
mi mj (N0 + Ipm ) + 1 ai j = 1, 2, · · · , npm
 a−1

j = npm + 1, npm + 2, · · · ,

i
wmij = npm + ns ; j 6= npm + i

  
 1 + (N0 + Ipm )P̂ g −1 −1 −1
a j = npm + i

mi i
(4.17)
Therefore, the reformulated problem is as follows.
MMKP Formulation:
P P
maximize ri xmi
i∈N
Ps m∈Nc
subject to wmij xmi ≤ 1, j = 1, 2, · · · , ns + npm , m ∈ Nc
i∈N
P s (4.18)
xmi ≤ 1, ∀i ∈ Ns
m∈Nc
xmi ∈ {0, 1}, ∀i ∈ Ns , m ∈ Nc

The above formulation is a 0 − 1 integer linear programming, and can be

analogous to a multidimensional multiple knapsack problem (MMKP), where
there are nc knapsacks, while any knapsack m (m = 1, · · · , nc ) has (ns +npm )
constraints as the dimensions.

4.3 MKP modeling by matrix transformation

In order to solve the MMKP in (4.18) by MOSEK, we need to describe it in
a general matrix form as follows.

maximize RX
subject to AX ≤ U (4.19)
X∈ζ

where, ζ = {0, 1}, and U = (1, 1, · · · , 1)T . R = (R1×ns , R1×ns , · · · , R1×ns )

| {z } | {z }
(nc +1)ns +np nc
(wherein R1×ns = (r1 , r2 , · · · , rns )), and X = (X1 , X2 , · · · , Xnc )T (wherein
| {z } | {z }
ns nc
Xm = (xm,1 , xm,2 , · · · , xm,ns ), ∀m ∈ Nc ).
| {z }
ns
According to the constraints in (4.18) that every SU can only use maxi-
mum one channel, we have the following matrix form constraint.

67
 Resource Optimization for Multi-channel Cognitive Radio
Cellular Networks

0
  

11×nc 0 ··· 0X1
 0 11×nc ···   X0 
0
  2
 .  ..  ≤ U (4.20)

 .. .. .. ..
 . . . .
  . 
0
0 0 · · · 11×nc Xns
0
where, Xi = (x1,i , x2,i , · · · , xnc ,i )T , ∀i ∈ Ns .
| {z }
nc
The inequality of (4.20) is further equivalent to the following

(A0 , A0 , · · · , A0 )X ≤ U (4.21)
| {z }
nc

where,  
1 0 ··· 0
0 1 ··· 0
A0 =  ..
 
.. . . .. 
. . . .
0 0 ··· 1 n
s ×ns

Then, we define the following matrix A,

 
A1 0 · · · 0
 0 A2 · · · 0 
 
A =  ... .. . . .. 

. . . 
 
0 0 · · · Anc 
A0 A0 · · · A0 (n
s (nc +1)+np )×nc ns

where, for each m ∈ Nc (m 6= 0),

 (N0 +Ipm )hm,1,1 (N0 +Ipm )hm,2,1 (N0 +Ipm )hm,ns ,1

hm,1 Γm,1
+1 hm,2 Γm,1
+1 hm,ns Γm,1
+1
···
 (N0 +Ipma)h1 m,1,2 a2 ans
 
(N0 +Ipm )hm,2,2 (N0 +Ipm )hm,ns ,2 
 hm,1 Γm,2
+1 hm,2 Γm,2
+1 hm,ns Γm,2
+1 

 a1 a2
··· ans


 .. .. .. .. 
.
 (N0 +Ipm )h.m,1,n . .
 
(N0 +Ipm )hm,2,npm (N0 +Ipm )hm,ns ,npm

pm
 hm,1 Γm,npm +1 +1 +1
 
hm,2 Γm,npm hm,ns Γm,npm
a1 a2
··· ans

Am = 
 
N0 +Ipm 
+1
P̂ hm,1
 1 1


 a1 a2
··· ans


N0 +Ipm
 +1 
1 P̂ hm,2 1
···
 
a1 a2 ans
 
.. .. .. ..
 

 . . . .


 N0 +Ipm 
+1
1 1 P̂ hm,ns
a1 a2
··· ans (ns +npm )×ns

68
4.3 MKP modeling by matrix transformation

0 0
We employ i (i = 1, 2, · · · , nc ns ) to denote the index of any member in
0
X. i is a combination of m and i as follows.
0
i = (m − 1)ns + i, m = 1, 2, · · · , nc ; i = 1, 2, · · · , ns . (4.22)
0
Reversely, given i we get
 0
 m =  di /ns e
0
ns i mod ns = 0 (4.23)
 i = 0
i mod ns Otherwise

Then, the MMKP in (4.18) can be represented as the following MKP .

MKP Formulation:
nP
c ns
maximize r̂i0 yi0
0
i =1
nPc ns
subject to
0
ωj 0 i0 yi0 ≤ 1, j = 1, 2, · · · , (nc + 1)ns + np (4.24)
0
i =1
0
yi0 ∈ {0, 1}, ∀i = 1, 2, · · · , nc ns
0
where, r̂i0 = ri (i is calculated from i by (4.23)). ωj 0 i0 has different expres-
0
sions with different j .
m−1
0 0 P
• when j ≤ nc ns + np , let jm = (m − 1)ns + npk , for any m =
k=1
1, 2, · · · , nc ,
0 0 0 0
– if j = jm + 1, jm + 2, · · · , jm + npm , we have
( (N0 +Ipm )hmij
gmi Γmj
+1 0
ωj 0 i0 = ai
, i = (m − 1)ns + i; j = 1, 2, · · · , npm
0, Otherwise
(4.25)
0 0 0 0 0
– if j = jm + npm + 1, jm + npm + 2, · · · , jm + npm + ns , and j 6=
0
jm + npm + i, where i = 1, 2, · · · , ns , we have
 1 0
a
, i = (m − 1)ns + i
ωj i =
0 0 i (4.26)
0, Otherwise
0 0
– if j = jm + npm + i, where i = 1, 2, · · · , ns , we have
( N0 +Ipm
+1 0
P̂ gmi
ωj 0 i0 = ai
, i = (m − 1)ns + i (4.27)
0, Otherwise

69
 Resource Optimization for Multi-channel Cognitive Radio
Cellular Networks

0
• when j > nc ns + np , for any m = 1, 2, · · · , nc , we have
 0 0
1, i = (m − 1)ns + j − nc ns − np
ωj 0 i0 = (4.28)
0, Otherwise

In this MKP, there is only one knapsack. But it has nc ns items and
(nc + 1)ns + np dimensions.

4.4 Proposed heuristic algorithm to MKP

In this section , we will introduce two lemmas and definitions, then we will
describe our proposed heuristic algorithm to MKP.
0
Lemma 1. Given a set of constraints CS, we say the i -th decision variable
0
yi0 is equal to 0, if there exist a j ∈ CS that ωj 0 i0 > 1 .
0
Proof. We use proof by contradiction. Suppose y is equal to 1, and all
0
the
P constraints are satisfied. Assume there is a j where ωj i > 1, thus
0 0

ωj 0 i0 > ωj 0 i0 > 1. Which means the constraint will never be satisfied.

i0 ∈D
0
The contradiction happens, and the hypothesis is wrong. Thus y should be
equal to 0.
0
Lemma 2. Given a selected set of decision variables as D, the j -th constraint
0
can be removed if the constraint is satisfied when yi0 is equal to 1 for all i in
D.
0 0
Proof. Assume j -th constraint (j = 1, ...,P(nc + 1)ns + np ) is satisfied when
0
yi0 is equal to 1 for all i in D. It means ωj 0 i0 ≤ 1
i0 ∈D
Because ωj i is non-negative and yi is equal to either 0 or 1, we have,
0 0 0
P P P 0
ωj 0 i0 yi0 ≤ ωj 0 i0 . Thus ωj 0 i0 yi0 ≤ 1. It means that the j -th con-
0
i ∈D i0 ∈D i0 ∈D
straint is always satisfied whatever the final solution is for all decision vari-
ables.
From Lemma 2, we have the following definition for valid constraints.
Valid Constraints Given a selected set of decision variables as D, we say
0 0
the j -th constraint is valid, if for all i in D the constraints do not meet. The
set of all valid constraints is noted as CS. In another word,
 
 X 
0
CS := j | ωj 0 i0 > 1 (4.29)
 0 
i ∈D

70
4.4 Proposed heuristic algorithm to MKP

Relevance of a valid constraint We say αj 0 is the reference for a valid

0 0
constraint j (j ∈ CS). X
αj 0 := ω j 0 i0 − 1 (4.30)
i0 ∈D

Since j is a valid constraint, according to Definition 4.4, we have αj 0 > 0.

The higher the relevance of a valid constraint, the higher the “scarcity” of
the corresponding resource and the less attractive it becomes to pack an item
which consumes a lot of that resource [99].
0 0
Efficiency for an item i We say ei0 is the efficiency for the i -th decision
variable yi0 .

r̂i0 r̂i0
ei0 := P = ! (4.31)
α j 0 ω j 0 i0 P P
j 0 ∈CS ωj 0 i0 − 1 ωj 0 i0
j 0 ∈CS i0 ∈D

The core of our proposed heuristic algorithm is minimal efficiency removal.

The major steps are as follows.

• In the beginning, we set CS to {1, 2, · · · , (nc + 1)ns + np }, and D to

0
{1, 2, · · · , nc ns }. Every decision variable yi0 (∀i ∈ D) is equal to 1.
0
• Then, we go through all i ∈ D by Lemma 1 to set yi0 to 0, if there
0 0
exist j ∈ CS that ωj 0 i0 > 1. i is then removed from D where yi0 = 0.

• We then select a variable with index of i∗ where ei∗ = arg min

0
ei0 ,
i ∈D
remove i∗ from D, set yi∗ to 0, and update the valid constraints set CS.

• We repeat this selection and removal procedure until either D or CS is

empty. If D is empty, it means no SU can be admitted. The reason is
either the interference constraint or QoS constraint is too strict to all
SUs for all channels.

The flow chart of the algorithm is shown in Fig. 4.2. We show more
details in Algorithm 4 and 5.
The time complexity is dominated by the procedure of MinimalEfficien-
cyRemoval. It will call the function of UpdateConstraintsAndEfficiencies(CS,
D) maximum nc ns times in the worst case. In the function UpdateCon-
straintsAndEfficiencies(CS, D), updating efficiency dominates the procedure.
It takes maximum (nc +1)ns +np to calculate in the worst case if no constraint

71
 Resource Optimization for Multi-channel Cognitive Radio
Cellular Networks

Algorithm 4 Proposed MKP Heuristic Algorithm for MMKP: Part 1

Input: Nc , Ns , {Npm |∀m ∈ Nc }.
Output: {xmi }, {Pmi }
1: procedureInitialization
2: for m = 1 → nc do
3: for i = 1 → ns do
4: Pmi ← 0; xmi ← 0
0
5: i ← (m − 1)ns + i; yi0 ← 1
6: for j = 1 → (nc + 1)ns + np do
7: ω j 0 i0 ← 0
8: end for
9: end for
10: end for
11: end procedure
12: procedureCountWeight
13: for m = 1 → nc do
m−1
0 P
14: jm ← (m − 1)ns + npk
k=1
15: for i = 1 → ns do
0
16: i ← (m − 1)ns + i
17: for j = 1 → npm do
0 0
18: j ← jm+ j 
(N0 +Ipm )hmij
19: ω j 0 i0 ← gmi Γmj
+ 1 /ai . According to (4.25)
20: end for
21: for k = 1 → ns do
0 0
22: j ←j +1
23: if k == i then  
N0 +Ipm
24: ωj i ← P̂ g
0 0 + 1 /ai . According to (4.27)
mi
25: else
26: ωj 0 i0 ← 1/ai . According to (4.26)
27: end if
28: end for
29: end for
30: end for
31: for i = 1 → ns do
32: for m = 1 → nc do
0 0
33: i ← (m − 1)ns + i; j ← nc ns + np + i
34: ωj 0 i0 ← 1 . According to (4.28)
35: end for
36: end for
37: end procedure
72
4.4 Proposed heuristic algorithm to MKP

Algorithm 5 Proposed MKP Heuristic Algorithm for MMKP: Part 2

38: CS ← {1, 2, · · · , (nc + 1)ns + np }
39: D ← {1, 2, · · · , nc ns }
40: procedure PreAdmissionControl
0
41: for all i ∈ D do
0
42: for all j ∈ CS do
43: if ωj 0 i0 > 1 then
44: y i0 ← 0
45: Break
46: end if
47: end for
48: end for
49: end procedure
50: function UpdateConstraintsAndEfficiencies(CS, D)
0
51: for all j ∈ CS do
52: Update CS according to (4.29).
53: end for
0
54: for all i ∈ D do
55: Update ei0 according to (4.31).
56: end for
57: end function
58: procedure MinimalEfficiencyRemoval
59: UpdateConstraintsAndEfficiencies(CS, D)
60: while CS 6= ∅ and D = 6 ∅ do
∗
61: i ← arg min 0
ei0
i ∈D
62: xi ∗ ← 0
63: D ← D − i∗
64: UpdateConstraintsAndEfficiencies(CS, D)
65: end while
66: end procedure
67: procedure GetResult
68: if D 6= ∅ then
0
69: for all i ∈ D do
70: Calculate m and i according to (4.23).
71: xmi ← 1
72: end for
73: for m = 1 → nc do
74: for i = 1 → ns do
75: Calculate Pmi according to (4.13)
76: end for
77: end for
78: end if
73
79: end procedure
 Resource Optimization for Multi-channel Cognitive Radio
Cellular Networks

Figure 4.2: Heuristic Algorithm to MKP

removed. We use ti (i = 1, 2, ...) to denote the number of constraints removed

by the i-th calling. The time complexity can be calculated as follows.

T (nc , ns , np ) = nc ns ((nc + 1)ns + np ) + (nc ns − 1)((nc + 1)ns + np − t1 )

+(nc ns − 2)((nc + 1)ns + np − t2 ) + ...
≤ (nc ns + (nc ns − 1) + (nc ns − 2) + ... + 1) ((nc + 1)ns + np )
= nc ns (1+n
2
c ns )
((nc + 1)ns + np )
= O(n2c n2s (nc ns + np ))
(4.32)
Thus, the time complexity of our proposed scheme is O(n2c n2s (nc ns + np )).

74
4.5 Best SINR channel selection scheme

4.5 Best SINR channel selection scheme

In order to evaluate the performance of our proposed scheme in the previous
section, we, hereby, introduce a scheme using traditional channel allocation
based on SINR. Wherein, any SU i (∀i ∈ Ns ) select the channel m∗i which
can achieve the most SINR with a certain power P .

m∗i = arg max ξmi

∀m∈Nc
hmi P
= arg max (4.33)
∀m∈Nc N0 +Ipm
= arg max N0h+I
mi
pm
, ∀i ∈ Ns
∀m∈N c

After channel allocation, we then use our proposed joint admission and
power control scheme JAPC-MKP in Chapter 3. We denote this scheme as
BestSINR-JAPC-MKP.
From (3.63), the time complexity using JAPC-MKP on a given channel m
is O(n2sm (nsm + npm )), where nsm and npm represent the number of SUs and
PUs, respectively. The total time complexity can be calculated as follows.
nc
n2sm (nsm + npm )
P
T (nc , ns , np ) =
m=1
nc
n2sm (nsm + nmax
P
≤ pc )
m=1
nc nc
n3sm + n2sm nmax
P P
= pc ) (4.34)
m=1 m=1
nc nc
nsm )3 + ( nsm )2 nmax
P P
≤ ( pc
m=1 m=1
= n2s (ns + nmax
pc )
= O(n2s (ns + nmax
pc ))

Thus, the time complexity of our proposed scheme is O(n2s (ns + nmax
pc )).

4.6 Simulation results and analysis

In this section, we will describe the simulation parameters and present the
simulation results. To evaluate the simulation results, we use MOSEK to get
the optimal results.
In our CogCell simulator, we generate the topology for 100 times for a
given number of channels, SUs and PUs. Each time, the topology is generated
randomly as follows. The BS locates at the center of a cell with its radius
Rmax as 1000m. The minimal distance from the BS to any SUs or PUs, Rmin ,
is set as 100m. The distance between SUs (or PUs) and the BS are randomly

75
 Resource Optimization for Multi-channel Cognitive Radio
Cellular Networks

chosen from [Rmin , Rmax ], The angles from any SUs (or PUs) to the BS are
randomly chosen from [0, 2π]. The number of channels in this system ranges
from 1 to 10, and the bandwidth of each channel is 5M Hz. The number
of PUs per channel ranges from 1 to nmax pc . The average power of noise is
−110dBm. The maximum transmission power of all SUs is 280mW . For
the estimation of channel gain in our simulation, we consider a slow fading
channel, and the path loss is d14 , where d is the distance between a transmitter
and its receiver. For the interference on each channel m, we use Ipm = npm Ip ,
where Ip is the interference contribution from one primary transmitter.

Table 4.2: Simulation parameters for Multi-channel CogCell

Symbol Value Symbol Value

Rmax 1000m Rmin 100m
B 5 MHz nc [1, 10]
max
npc [2, 10] ns [1, 100]
N0 −110dBm P̂ 280mW

The revenue ri obtained from SU i (i ∈ Ns ) is dependent on the DTR. The

SU with higher DTR pays more and hence generate higher revenue for the
service provider. Without loss of generality and for the sake of illustration,
we allocate the revenue and DTR according to Table 4.3, which is the same
as 3.3 in Chapter 3.

Table 4.3: Revenue allocation table for Multi-channel CogCell

Revenue 1 2 4 8 16 32
DTR (kbps) 16 32 64 128 256 512
Required SINR 0.0022 0.0043 0.0087 0.0175 0.0353 0.0718

In the following simulation, we evaluate the performance in terms of the

secondary revenues in five different cases, i.e., changing the number of chan-
nels, changing the number of SUs, changing the number of PUs per channel,
changing the interference threshold (Γ), and changing the interference to BS
from primary systems (Ip ).
In each case, with the given number of PUs and SUs, the value of Γ and
and Ip , we solve the optimization problem by MOSEK software [37], and run
our proposed scheme, and the bestSINR-JAPC-MKP. Then, we calculate the

76
4.6 Simulation results and analysis

average secondary revenue based on the results in the 100 random topologies,
and 100 random data rate requirements according to Table 4.3 .

Effect of number of channels

In the following, we evaluate the secondary revenue when changing the num-
ber of channels from 1 to 10. The other parameters are fixed as follows,
ns = 100, nmax
pc = 10, Γ = −70dBm, and Ip = −80dBm. The results are
shown in Fig. 4.3 and Fig. 4.4.

900

800

MOSEK
700
Secondary Revenue

Proposed
BestSINR−JAPC−MKP

600

500

400

300
1 2 3 4 5 6 7 8 9 10
Number of Channels

Figure 4.3: Revenue in terms of number of channels (nmax

pc = 10, Γ =
−70dBm, Ip = −80dBm, and ns = 100)

Figure 4.3 shows the secondary revenue increases with the increasing num-
ber of channels. The reason is the more channels, the more SUs can be ad-
mitted. Thus, the revenue to BS increases. When nc is equal to 1, the results
of three schemes are almost the same. However, when ns is greater than 1,
the gap between our proposed scheme and BestSINR-JAPC-MKP becomes
bigger and bigger. When ns is greater than 7, our proposed scheme gets
more than twice the revenue got by BestSINR-JAPC-MKP. Moreover, our
proposed scheme approaches the results from MOSEK all the time.
Figure 4.4 shows the percentage of optimal solution from MOSEK for both
our proposed scheme and BestSINR-JAPC-MKP. BestSINR-JAPC-MKP de-
creases from 97% when nc is equal to 1 to as low as 47% when nc is equal

77
 Resource Optimization for Multi-channel Cognitive Radio
Cellular Networks

100%
97%

92% Proposed
Percentage of optimal solution from MOSEK

90% BestSINR−JAPC−MKP

80%

70%

60%

50%
47%

40%
1 2 3 4 5 6 7 8 9 10
Number of Channels

Figure 4.4: Percentage of optimal solution from MOSEK in terms of number

of channels (nmax
pc = 10, Γ = −70dBm, Ip = −80dBm, and ns = 100)

to 10. Our proposed scheme decreases from 97% when nc is equal to 1 to

92% when nc is equal to 3. However, when nc is greater than 3, our proposed
scheme starts to increase the revenue. When nc is great than 7, our proposed
scheme achieves more than 97%.

Effect of number of SUs

In the following, we evaluate the secondary revenue when changing the num-
ber of SUs from 1 to 100. The other parameters are fixed as follows, nc = 10,
nmax
pc = 10, Γ = −70dBm, and Ip = −80dBm. The results are shown in
Fig. 4.5 and Fig. 4.6.
Figure 4.5 shows the secondary revenue increases with the increasing num-
ber of SUs. The reason is channels in the system are not saturated with lower
number of SUs, and can serve more SUs when more SUs are available. Thus,
the revenue to BS increases. When ns is smaller than 20, the results of
three schemes are almost the same, since in that case, almost all the SUs
can be admitted. However, when ns is greater than 20, the gap between
our proposed scheme and BestSINR-JAPC-MKP becomes bigger and bigger.
When ns is 100, our proposed scheme gets more than twice the revenue got
by BestSINR-JAPC-MKP. Moreover, our proposed scheme approaches the
results from MOSEK all the time.

78
4.6 Simulation results and analysis

900

800 MOSEK
Proposed
700 BestSINR−JAPC−MKP

600
Secondary Revenue

500

400

300

200

100

0
0 10 20 30 40 50 60 70 80 90 100
Number of SUs

Figure 4.5: Revenue in terms of number of SUs (nmax

pc = 10, Γ = −70dBm,
Ip = −80dBm, and nc = 10)

100%
97%
Percentage of optimal solution from MOSEK

90% Proposed
BestSINR−JAPC−MKP

80%

70%

60%

50%
46%

40%
0 10 20 30 40 50 60 70 80 90 100
Number of SUs

Figure 4.6: Percentage of optimal solution from MOSEK in terms of number

of SUs (nmax
pc = 10, Γ = −70dBm, Ip = −80dBm, and nc = 10)

79
 Resource Optimization for Multi-channel Cognitive Radio
Cellular Networks

Figure 4.6 shows the percentage of optimal solution from MOSEK de-
creases for both our proposed scheme and BestSINR-JAPC-MKP with the
increasing number of SUs. BestSINR-JAPC-MKP decreases to as low as
46%, while our proposed scheme can still achieve at least 97% of the optimal
result from MOSEK.

Effect of number of PUs per channel

The number of PUs per channel is randomly generated from 1 to nmaxpc . We
max
change npc from 2 to 10, while other parameters are fixed as follows: Γ =
−70dBm, Ip = −80dBm, nSU = 100, and nc = 10. The results are shown in
Fig. 4.7 and Fig. 4.8.

1000

900

800
Secondary Revenue

MOSEK
700 Proposed
BestSINR−JAPC−MKP

600

500

400
2 3 4 5 6 7 8 9 10
nmax
pc

Figure 4.7: Revenue in terms of different number of PUs per channel (Γ =

−70dBm, Ip = −80dBm, nSU = 100, and nc = 10)

Figure 4.7 shows the secondary revenue decreases with the increasing
nmax
pc . The reason is the more PUs in a channel, the more interference con-
straints should be considered. Thus, fewer SUs can be admitted. It then
results in less secondary revenue to BS. Our proposed scheme approaches
MOSEK in all the cases, and achieves more than twice the revenue got by
BestSINR-JAPC-MKP.
Figure 4.8 shows the percentage of optimal solution from MOSEK for
both our proposed scheme and BestSINR-JAPC-MKP. BestSINR-JAPC-

80
4.6 Simulation results and analysis

100%
97%
Percentage of optimal solution from MOSEK

90%

Proposed
BestSINR−JAPC−MKP

70%

50%
47%
44%
40%
2 3 4 5 6 7 8 9 10
max
npc

Figure 4.8: Percentage of optimal solution from MOSEK in terms of number

of maximum PUs per channel (nmax
pc = 10, Γ = −70dBm, Ip = −80dBm, and
nc = 10)

MKP achieves between 44% and 47%, while our proposed scheme achieves
more than 97% of the secondary revenue from MOSEK.

Effect of interference from primary system

We change Ip from −140dBm to −40dBm, while other parameters are fixed
as follows: nmax
pc = 10, Γ = −70dBm, nSU = 100, and nc = 10. The results
are shown in Fig. 4.9 and Fig. 4.10.
Figure 4.9 shows the secondary revenue decreases with the increasing
interference to BS from each PU transmitter. The reason is the more in-
terference from PU transmitters to BS in a channel, the higher power SUs
should use to achieve a certain SINR level. It then results in more inter-
ference to PUs. Because the interference to PUs is limited, fewer SUs can
be admitted. Thus, the secondary revenue to BS decreases. Our proposed
scheme achieves more than twice the revenue got by BestSINR-JAPC-MKP
when the interference to BS per PU transmitter is less than −80dBm, and
approaches MOSEK in all cases.
Figure 4.10 shows the percentage of optimal solution from MOSEK for
both our proposed scheme and BestSINR-JAPC-MKP. BestSINR-JAPC-
MKP achieves between 40% and 50%, when Ip is less than −80dBm, and

81
 Resource Optimization for Multi-channel Cognitive Radio
Cellular Networks

1100

1000

900

800 MOSEK
Proposed
Secondary Revenue

700 BestSINR−JAPC−MKP

600

500

400

300

200

100

0
−140 −130 −120 −110 −100 −90 −80 −70 −60 −50 −40
Interference to BS from each PU transmitter (dBm)

Figure 4.9: Revenue in terms of different interference to BS from each PU

transmitter (nmax
pc = 10, Γ = −70dBm, nSU = 100, and nc = 10)

100%
97%
Percentage of optimal solution from MOSEK

90%

Proposed
80%
BestSINR−JAPC−MKP

70%

60%

50%

40%
−140 −130 −120 −110 −100 −90 −80 −70 −60 −50 −40
Interference to BS from each PU transmitter (dBm)

Figure 4.10: Percentage of optimal solution from MOSEK in terms of different

interference to BS from each PU transmitter (nmax
pc = 10, Γ = −70dBm, nSU =
100, and nc = 10)

82
4.6 Simulation results and analysis

increase to 90% when Ip increases to −40dBm. On the other hand, our

proposed scheme achieves more than 97% of the secondary revenue from
MOSEK, it achieves the same revenue with MOSEK when Ip is less than
−90dBm.

Effect of interference threshold on PUs

We change Γ from −140dBm to −40dBm, while other parameters are fixed
as follows: nmax
pc = 10, Ip = −80dBm, nSU = 100, and nc = 10. The results
are shown in Fig. 4.11 and Fig. 4.12.

900

800
MOSEK
700 Proposed
BestSINR−JAPC−MKP
600
Secondary Revenue

500

400

300

200

100

0
−140 −130 −120 −110 −100 −90 −80 −70 −60 −50 −40
Interference Threshold on each PU (dBm)

Figure 4.11: Revenue in terms of different interference level per PU (nmax

pc = 10,
Ip = −80dBm, nSU = 100, and nc = 10)

Figure 4.11 shows the secondary revenue increases with the increasing
interference threshold on PUs. The reason is the higher interference threshold
on PUs, the more SUs can be allowed to transmit. Thus, the revenue to BS
increases. When Γ is smaller than −130dBm, the results of three schemes are
almost the same, since in that case, almost no SUs can be admitted. However,
when Γ is greater than −130dBm, the gap between our proposed scheme
and BestSINR-JAPC-MKP becomes bigger and bigger until Γ is equal to
−70dBm. When Γ is greater than −70dBm, the gap between our proposed
scheme and BestSINR-JAPC-MKP stays almost the same. It also shows,
when Γ is greater than −90dBm, our proposed scheme gets more than twice

83
 Resource Optimization for Multi-channel Cognitive Radio
Cellular Networks

100%

90%
Percentage of optimal solution from MOSEK

Proposed
BestSINR−JAPC−MKP

70%

57%

50%

40%

35%
−140 −130 −120 −110 −100 −90 −80 −70 −60 −50 −40
Interference Threshold on each PU (dBm)

Figure 4.12: Percentage of optimal solution from MOSEK in terms of different

interference level per PU (nmax
pc = 10, Ip = −80dBm, nSU = 100, and nc = 10)

the revenue got by BestSINR-JAPC-MKP. Moreover, our proposed scheme

approaches the results from MOSEK very closely.
Figure 4.12 shows the percentage of optimal solution from MOSEK for
both our proposed scheme and BestSINR-JAPC-MKP. Our proposed scheme
achieves more than 90% of optimal solution from MOSEK, while BestSINR-
JAPC-MKP achieves only 40% to 57%.

4.7 Conclusions and discussions

4.7.1 Conclusions
In this chapter, we have investigated the operator problem of maximization
the secondary revenue, while satisfying the power limitation, minimum SINR
and interference constraints. We modeled the problem as MMKP, and then
transfer it to MKP. Then, we proposed a heuristic algorithm based on the
MKP formulation. Simulation results showed our proposed heuristic scheme
archive much more secondary revenue than BestSINR-JAPC-MKP, and is
close to the optimal results from MOSEK.

84
4.7 Conclusions and discussions

4.7.2 Discussions
The problem we studied in this chapter can be extended and applied in many
ways.

• Different revenue formulations. In our simulation, revenue is propor-

tional to the data rate of each SU. In practice, the revenue can be
customized by the service providers.

• Different path loss models. In our simulation, we consider a slow fading

channel, and the path loss is d14 , where d is the distance between a
transmitter and its receiver. In practice, we can use any suitable path
loss models.

• When SU can use multiple channels in the same time.

In case every SU has the ability to use multiple channels at the same
time, say maximum K, we can modify our formulations as follows.

P P
maximize ri xmi
i∈N
Ps m∈Nc
subject to wmij xmi ≤ 1, j = 1, 2, · · · , ns + npm , m ∈ Nc
i∈N
Ps
xmi ≤ K, ∀i ∈ Ns
m∈Nc
xmi ∈ {0, 1}, ∀i ∈ Ns , m ∈ Nc
(4.35)

85
Chapter 5

Spectrum Sharing in Cognitive

Radio Femtocell Networks

In the previous chapters, we have investigated the resource optimization prob-

lems in the underlay spectrum sharing mode. From this chapter on, we study
the resource optimization in the overlay spectrum sharing mode. Specially in
this chapter, we study the resource optimization problem in cognitive radio
femtocell networks.
Femtocell is envisioned as a highly promising solution for indoor wireless
communications. The spectrum allocated to femtocells is traditionally from
the same licensed spectrum bands of macrocells. In this case, the capacity of
femtocell networks may be highly limited due to the finite number of licensed
spectrum bands and also the interference with macrocells and other femto-
cells. In this chapter, we propose a radically new communication paradigm
by incorporating cognitive radio in femtocell networks. The cognitive radio
enabled femtocells are able to access licensed spectrum bands not only from
macrocells but also from other licensed systems (e.g. TV systems). Thus, the
co-channel interference in femtocells can be greatly reduced and the network
capacity can be significantly improved.
The rest of this chapter is organized as follows. In section 5.1, we intro-
duce the system model and assumptions. Then we formulate the downlink
spectrum sharing problem for CogFem in section 5.2. In section 5.3, we
employ a mixed primal and dual decomposition method to decompose the
problem into a master problem with channel allocation, and several subprob-
lems with power control at each femtocell. Then, we propose a joint channel
allocation and fast power control schemes. In section 5.4, we evaluate the
performance of our proposed scheme for normal femtocells and CogFem, and
compare it to some existing methods. Simulation results also showed that
our proposed scheme without any iteration can achieve almost twice of the

87
 Spectrum Sharing in Cognitive Radio Femtocell Networks

average capacity by coloring method when the number of available channels

is less than 5. Moreover, our proposed scheme can converge very fast with a
typical value of only 5 iterations, and it can achieve around 2% extra average
capacity than fixed power control scheme. Finally, we draw conclusions in
section 5.5.

5.1 System model and assumptions

In this section, we introduce the system model and assumptions for CogFem
in the overlay spectrum sharing mode.
The notations used in this chapter are shown in Table 5.1. Suppose that
there are a set of F femtocells in the coverage of a macrocell. For any FBS i
(i ∈ F), there are a set of Mi FUs. Normally the number of FUs is between
2 and 4 as indicated in [25]. In this chapter, we also use i as the ID of the
femtocell where FBS i is located. There are a set of Ni licensed channels
that can be used for femtocell i. Ni may change dynamically depending on
the activities of nearby primary systems. These channels are not only from
macrocells but also from other licensed systems. We claim that the definition
of this channel could be adapted according to the particular access method.
For example, if femtocells use CDMA, the channel is a wide band like 5MHz,
10MHz. If femtocells use OFDMA, the channel could be a narrow band
subchannel containing several subcarriers typically 100KHz similar in IEEE
802.22 draft standard [104].

5.1.1 System initialization

Whenever an FBS turns on, it will first sense the spectrum environment to
initialize an available spectrum list. The FBS will be responsible to allo-
cate spectrum to its users, and inform them the suitable uplink transmission
power. The uplink power control is out of the topic in this chapter. Synchro-
nization between neighboring FBSs is not obligatory in CogFem, but it is an
option if any FBS wants to synchronize with its neighbors. The synchroniza-
tion can be implemented by listening to neighboring femtocells information
to obtain the frame length and structure.

5.1.2 Number of transceivers

We do not specify the stringent requirement on the number of transceivers on
FBS. One transceiver for each FBS is possible. In this situation, each FBS
will perform both spectrum sensing and data transmission on this transceiver

88
5.1 System model and assumptions

Table 5.1: Table of notations for cognitive radio femtocell networks

Symbol Meaning
F the set of FBSs
Mi the set of FUs in FBS i
Ni the set of available channels in FBS i
i the index of FBS
j the index of FU
c the index of channel
mi the number of FUs in FBS i
xijc the binary indicator of channel c on FU j in FBS i
pijc the transmission power for FBS i at channel c on FU j
hijc the channel gain on channel c for FBS i and FU j
Iijc the interference at FU j in FBS i on channel c
ψ the minimum required SINR for FUs

at different time. To reduce the complexity and improve the throughput,

two transceivers for each FBS would be better. In this situation, each FBS is
equipped with two transceivers. One is called sensing radio used for spectrum
sensing, while the other one is called cognitive radio used for data commu-
nication of both intra-femtocell and inter-femtocell on the selected channels.
So that, FBS can do spectrum sensing and data transmission simultaneously.

5.1.3 Spectrum sensing and primary system protec-

tion
Each FBS is able to sense the available spectrum. The available spectrum
list can be stored into a local database or a database in the Internet for the
future use. FBSs from other femtocells can access these information from
the database, and negotiate with the neighboring FBSs with the available
spectrum.
For spectrum sensing, both FBS and FU can support spectrum sensing
if the hardware expense is not an issue. In this situation, whenever an FU
detects the return of a primary user (PU), it will stop transmission and inform
the FBS in the control channel. The FBS will do fine spectrum sensing by
itself, and determine the real existence of the PU. If a real PU exists, the FBS
will inform the FU to switch to another channel. Otherwise, it will inform
FU to continue using the current channel.

89
 Spectrum Sharing in Cognitive Radio Femtocell Networks

To save hardware expense and battery lifetime for FU, we can suppose
only FBS would do spectrum sensing. In this situation, whenever an FBS
detects the return of a PU, it will stop transmission, and then inform its FUs
and the neighboring FBSs about the existence of the PU. It then updates
the available channel list, and runs the spectrum sharing algorithms to select
new channels and allocates new time-subchannel blocks for its FUs.
The main challenge is the accuracy of spectrum sensing. Due to hardware
limitation and spectrum detection schemes, false alarm and miss detection
may happen. Specifically, false alarm happens when the spectrum sensing
results show that primary signal exists but actually there is no primary signal.
Miss detection happens when the spectrum sensing results show that there
is no primary signal but actually primary signal do exist. False alarm will
cause unnecessary channel switching which results in increased delay and
packet loss. Miss detection will cause interference to the primary systems.
Cooperative spectrum sensing and decision is a good candidate to reduce the
probability of false alarm and miss detection. In this chapter, the details of
spectrum sensing is out of the range of this topic, we assume perfect spectrum
sensing. For more details on spectrum sensing, please refer to [105].

5.1.4 Control channel

There are two kinds of control channels. One is called inter-femtocell control
channel, whereby each FBS can communicate with each other. The other
one is called intra-femtocell control channel, whereby each user in a femto-
cell can communicate with its FBS to obtain the channel information and
allowed transmission power. These control channel could be a dedicated con-
trol channel or a rendezvous channel which can be selected according to some
metrics such as channel availability. Since every FBS has a broadband con-
nection to the Internet, in spite of using the inter-femtocell control channel,
neighboring FBSs can communicate with each other through the broadband
connection. Similarly, an additional FBS controller in the Internet can be
helpful for the management of FBSs.

5.1.5 Handover between macrocell and femtocell

Whenever an FU moves into a femtocell from a macrocell, it can detect the
existence of an FBS by listening to the control channel information, and
decide to switch into the femtocell network.
By contrast, whenever an FU moves out of a femtocell, it can detect that
the strength from FBS is weaker than the strength from macrocell BS (MBS),
then it decides to switch into the macrocell network.

90
5.1 System model and assumptions

In traditional mobile macrocell networks, handover can be either soft or

hard. In soft handover, the mobile user will communicate with the two BSs
at the same time, until the signal strength is higher than a threshold. In hard
handover, the mobile user will switch to the new BS as soon as the signal is
stronger than the old BS.
In CogFem, we can employ both soft and hard handover. Simply, we can
use hard handover for instance in this chapter. When the FU detects that
the pilot signal is much stronger than the traffic channel, it will switch to
the femtocell network. If it detects that the pilot signal from macrocell BS
is much higher than current femtocell data traffic channel, it will switch to
the macrocell network.

5.1.6 Deployment example

In practice, we can deploy CogFem in a flexible way. As an example shown
in Fig. 5.1, an FBS controller can be added in the system architecture to
improve the management ability for all the FBSs.

Figure 5.1: An example of deploying CogFem networks with FBS controller

91
 Spectrum Sharing in Cognitive Radio Femtocell Networks

5.2 Problem formulations

In this chapter, we consider the downlink spectrum sharing problem in the
overlay mode. Femtocells use the licensed channels when they are not occu-
pied by primary systems. Thus, there is no co-channel interference between
primary systems and femtocells. The only interference should be managed is
amongst femtocells. Suppose each femtocell user in a femtocell i requires one
channel. We consider the worst case when all neighboring femtocells are in
downlink transmission. In the following, we analyze the downlink capacity
and then formulate the spectrum sharing problem.

5.2.1 Channel model

The indoor path loss model in dB is based on the ITU and COST 231 indoor
model [106] [107] as follows.
 
nij +2
nij +1
−0.46
Gij = 37 + 30log10 dij + 18.3nij (5.1)

where dij is the distance between the transmitter i and the receiver j. nij
denotes the number of floors in the path. We introduce hijc as follows to
represent the channel gain between FBS i and its FU j on channel c.
Gij
hijc = 10(− 10
)
(5.2)

5.2.2 Downlink capacity

Any femtocell user will receive interference from neighboring femtocells using
the same channel. We consider an Additive White Gaussian Noise (AWGN)
channel. The SINR of the received signal from FBS i at femtocell user j can
be denoted as
ξijc = gijc pijc xijc (5.3)
where
hijc
gijc = (5.4)
N0 + Iijc
where N0 denotes the background noise power. Iijc represents the interference
measured at user j on channel c from femtocells other than i. pijc is the
downlink transmission power for FBS i on channel c. xijc is a binary indicator.
If xijc = 1, user j in femtocell i works on channel c, zero otherwise.

92
5.2 Problem formulations

The downlink capacity of any femtocell user j in femtocell i can be cal-

culated according to Shannon’s capacity theory as follows.
X
Cij = Bc log2 (1 + ξijc ) (5.5)
c∈Ni

where Bc denotes the bandwidth of channel c. ξijc is defined in (5.3). Then,

we can calculate the downlink capacity of femtocell i as follows.
P
Ci = Cij
j∈M
Pi P (5.6)
= Bc log2 (1 + ξijc ), ∀i ∈ F
j∈Mi c∈Ni

5.2.3 Downlink spectrum sharing problem

The spectrum sharing problem in CogFem downlink transmission is to max-
imize the downlink capacity of all FBSs while guaranteeing the channel allo-
cation, SINR, and power constraints.
P1
X
maximize Ci (5.7)
i∈F

s.t.

xijc ∈ {0, 1}, ∀i ∈ F, j ∈ Mi , c ∈ Ni (5.8)

X
xijc = 1, ∀i ∈ F, j ∈ Mi (5.9)
c∈Ni
X X
xijc = mi , ∀i ∈ F (5.10)
j∈Mi c∈Ni

ξijc ≥ ψ, if xijc = 1, ∀i ∈ F, j ∈ Mi , c ∈ Ni (5.11)

pijc = 0, if xijc = 0, ∀i ∈ F, j ∈ Mi , c ∈ Ni (5.12)

pijc ≥ 0, ∀i ∈ F, j ∈ Mi , c ∈ Ni (5.13)

X X
pijc ≤ Pimax , ∀i ∈ F (5.14)
j∈Mi c∈Ni

93
 Spectrum Sharing in Cognitive Radio Femtocell Networks

where ψ denotes the minimum required SINR for FUs. Constraint (5.9)
means every user in a CogFem can only use one channel. Constraint (5.10)
means the total number of channels can be used in one femtocell is equal to
the number of users in that femtocell mi . Constraint (5.11) represents that if
channel c is allocated to user j in femtocell i for downlink transmission, the
SINR received on user j should be higher than the predefined threshold ψ.
Constraint (5.12) means any FBS i will not allocate any power on channel
c, if channel c is not allocated to FBS i. Constraint (5.13) represents the
transmission power of any FBS i should be no less than 0, while constraint
(5.14) indicates the total transmission power of any FBS i on its FUs can
not exceed the maximum power budget Pimax .
The solution of the formulated spectrum sharing problem is the channel
allocation vector x and power vector p, and the objective function is non-
linear. Thus it is a MINLP problem, which is NP-hard in general. In the
following sections, we will use decomposition methods to solve it.

5.3 Problem decompositions and solutions

In this section, we use mixed primal and dual decomposition methods to
solve the downlink spectrum sharing problem based on the decomposition
theories in [108] and [109].

5.3.1 The master problem

Given a feasible power pijc for each FU i, we have the master problem in
charge of updating the channel allocation variables {xijc }, by solving the
following problem.
XX X
maximize Bc log2 (1 + gijc pijc xijc ) (5.15)
i∈F j∈Mi c∈Ni

s.t.

xijc ∈ {0, 1}, ∀i ∈ F, j ∈ Mi , c ∈ Ni (5.16)

X
xijc = 1, ∀i ∈ F, j ∈ Mi (5.17)
c∈Ni

X X
xijc = mi , ∀i ∈ F (5.18)
j∈Mi c∈Ni

94
5.3 Problem decompositions and solutions

This problem can be solved heuristically in polynomial times. We can

observe that the objective function (5.15) is concave and monotonously in-
creasing with gijc pijc xijc . Intuitively, we can find a solution by assigning 1 to
xijc with the maximum gijc pijc , and 0 to other channel vector for the same FU
j in femtocell i. We repeat this process until all the constraints are satisfied.
The detailed implementation is shown in Algorithm 6.
Algorithm 6 Channel allocation algorithm for the master problem
Input: {Ni }, {Mi }, {pijc }, {gijc }.
Output: {xijc }.
0 0
1: Initialization: Mi ← Mi , Ni ← Ni .
0
2: while ∀Mi 6= ∅ do
0
3: if Ni = ∅ then
4: Break; . not enough channels for femtocell i.
5: else
6: {i∗ , j ∗ , c∗ } ← arg max0 0
gijc pijc
∀i∈F ,j∈Mi ,c∈Ni
0 0
7: Ni∗ ← Ni∗ − c∗
0 0
8: Mi∗ ← Mi∗ − j ∗
9: end if
10: end while

The complexity of the Algorithm 6 is O(|F||M||N |), where | · | denotes

the cardinal of the set within. |M| and |N | are the maximum number of
FUs and channels per FBS, i.e., |M| = max |Mi |, and |N | = max |Ni |.
∀i∈F ∀i∈F

5.3.2 Subproblems
problem formulation
Given a solution of channel allocation {xijc }, we can get the following power
control subproblem to obtain the transmission power for every FBS i to any
of its FU j on the allocated channel cj .
XX
maximize Bc log2 (1 + gijc pijc∗ ) (5.19)
i∈F j∈Mi
s.t.

pijc ≥ 0, ∀i ∈ F, j ∈ Mi (5.20)
X
pijc ≤ Pimax , ∀i ∈ F (5.21)
j∈Mi

95
 Spectrum Sharing in Cognitive Radio Femtocell Networks

gijc pijc ≥ ψ, ∀i ∈ F, j ∈ Mi (5.22)

pijc = 0, if c 6= cj , ∀i ∈ F, j ∈ Mi , c ∈ Ni (5.23)

The Lagrangian
We form the Lagrangian function as follows
P P
L(p, λ, ν) = Bc log2 (1 + gijc pijc )
i∈F j∈Mi !
λi Pimax −
P P
+ pijc
i∈F
P P j∈Mi (5.24)
+ νij (gijc pijc − ψ)
Pi∈F j∈Mi
= Li (pi , λi , νi )
i∈F

where the Lagrangian multiplier vectors of λ and ν are non-negative. (λ =

(λ1 , λ2 , ...)T , ν = (ν1 , ν2 , ...)T , νi = (νi1 , νi2 , ...)T ). pi is the power vector for
FBS i. And Li (pi , λi , νi ) is defined as follows.
P
Li (pi , λi , νi ) = Bc log2 (1 + gijc pijc )
j∈Mi !
+λi Pimax −
P
pijc (5.25)
j∈Mi
P
+ νij (gijc pijc − ψ)
j∈Mi

Thus, the Lagrangian dual can be decomposed into |F| subproblems for
each FBS i (∀i ∈ F). For each given λi and νi , the dual is to solve pi

p∗i = arg max Li (pi , λi , νi ) (5.26)

pi ≥0

The decomposed Lagrangian dual function (5.25) is concave on pi , accord-

ing to Karush-Kuhn-Tucker (KKT) condition [110], we have the following
equations for any FBS i.

∂Li (pi , λi , νi )
=0 (5.27)
∂pijc
!
X
λi Pimax − pijc = 0 (5.28)
j∈Mi

96
5.3 Problem decompositions and solutions

νij (gijc pijc − ψ) = 0 (5.29)

max
P
where (5.28) means if pijc 6= Pi , the Lagrangian multiplier λi
j∈Mi
should be zero. Similarly, (5.29) means if gijc pijc 6= ψ, the Lagrangian multi-
plier νij should be zero.
According to (5.27), we have
Bc gijc
− λi + νij gijc = 0 (5.30)
(1 + gijc pijc )ln2
We can obtain pijc as follows
 Pimax
Bc 1
pijc = − (5.31)
(λi − νij gijc )ln2 gijc P min
ijc

where [·]badenotes the projection onto the area in [a, b]. This solution is
min
only valid when xijc = 1. If xijc = 0, pijc = 0. Pijc is the minimum
min
transmission power at FBS i for FU j on channel c. Pijc can be determined
by substituting (5.3) into (5.11) as follows.
min ψ
Pijc = (5.32)
gijc
This minimum value may be changed according to the environment, for ex-
ample the movement of FUs and the interference from other FBSs. Moreover,
the first part in (5.31) should be non-negative, so we have
λi > νij gijc , ∀j ∈ Mi (5.33)
min
In addition, we observe that Pimax should be larger than the sum of Pijc
in any FBS i. Otherwise, there will be no feasible solution for the problem,
which results in not all FUs P in the FBS can be served. Therefore, Pimax
min
should be configured at least Pijc in the CogFem deployment.
∀j,c

Solution with subgradient methods

In the following, we discuss the updating of Lagrangian multipliers when
xijc = 1. We employ the projected subgradient method as follows.
" !#+
X
λi (t + 1) = λi (t) − α(t) Pimax − p∗ijc (5.34)
j∈Mi
+
νij (t + 1) = νij (t) − α(t)(gijc p∗ijc − ψ)

(5.35)
+
where [·] denotes the projection onto the non-negative area. α(t) is a posi-
tive stepsize for the t times iteration.

97
 Spectrum Sharing in Cognitive Radio Femtocell Networks

Convergence Theoretically, α(t) can be chosen in a manner of either con-

stant or diminishing. The method using constant stepsize α(t) = α(0),
(α(0) > 0) can not guarantee the convergence, it may iterate repeatedly
near the optimal solution. By contrast, using the diminishing stepsize, where
∞
P
α(t) > 0, lim α(t) = 0, and α(t) = ∞, for example, α(t) = α(0)/t, the
t→∞ t=1
solution will be finally converged.
In our scenario, we can try α(0) = 1, and α(t) = 1/t, t = 1, 2, .... For the
initial value of λ and ν, we can choose λi (0) = Pimax . To obey the constraints
in (5.33), we can assign

θλi (0)
νij (0) = , ∀j ∈ Mi
gijc

where θ is a scale in the range of [0, 1).

Distributed implementation This scheme can work distributively, where

each FBS i updates its own λi and νij until convergence. The details of the
power control algorithm for any FBS i is shown in Algorithm 7. We can
summarize our decomposition methods in Fig. 5.2, where the Lagrangian
multipliers λi and νij serve as the prices for each FBS i.

Algorithm 7 Power control algorithms for any FBS i

Input: i, {xijc }, Ni , Mi , θ.
Output: {pijc }.
1: Initialization: α ← 1, λi ← Pimax , and νij ← gθλi , t ← 0, pijc ← Pijc
min
ijc
2: while not converged do
3: Update λi according to (5.34)
4: for all j ∈ Mi do
5: Update νij according to (5.35)
6: Calculate pijc according to (5.31)
7: end for
8: t←t+1
9: α ← 1/t
10: end while

Proposed scheme
Although the subgradient method can converge to the optimal solution, it
highly depends on the stepsize and initial values. Therefore, it may require

98
5.3 Problem decompositions and solutions

Figure 5.2: The decomposition of spectrum sharing problems in CogFem

lots of iterations and the knowledge of stepsize and Lagrangian multipliers

crossing different FBSs. In this study, we are interested in finding a more
suitable approach for CogFem. We can rewrite (5.31) as follows

1 1
pijc = 0
− (5.36)
λi gijc

where
ln2
λ0i = (λi − νij gijc ) (5.37)
Bc
where the Lagrangian multiplier νij can change accordingly with gijc , so
that (5.37) is only changed with i instead of both i and j.
Substituting (5.36) to (5.28) when xijc = 1, we have
mi
λ0i = 1 (5.38)
Pimax
P
+ gikck
k∈Mi

where ck is the selected channel for FU k. Substituting (5.38) to (5.36) when

xijc = 1, we have

 0,
  Pimax
xijc = 0
pijc = 1 max
+ m1i
P 1 1
− gijc (5.39)
 mi Pi

k∈Mi
gikc k
, xijc = 1
min
Pijc

min
Similarly, pijc should be no less than Pijc . This can be guaranteed by
max
the configuration of Pi .
In this scheme, we assume Channel State Information (CSI) can be ob-
tained by each FBS. One of the possible way to obtain CSI is as follows: each
FU can report the interference measurement result to its FBS. Therefore,
each FBS will make the decision of channel selection and power allocation

99
 Spectrum Sharing in Cognitive Radio Femtocell Networks

according to the measurements not only on FBS but also on its users. In prac-
tice, each femtocell user is required to negotiate a control channel with its
FBS, and reports its measurements to the FBS through this channel. Based
on these information, FBS then characterizes the channels with the accurate
interference levels for each user, and chooses mi channels with lowest inter-
ference levels. The channel allocation metric is based on gijc . FBS can also
use its own measured interference as approximate interference on FUs. Our
study in [111] has shown that the network performance in terms of average
capacity is quite close to each other by either obtaining the interference from
FBS or FUs. The reason behind it is that FBS and its FUs are in the same
apartment. Other FBSs, where the interference comes from, get power decay
not only because of distance but also because of the penetration of floors and
walls.
This scheme goes as follows.

• Channel c∗ is allocated to user j ∗ , if gij ∗ c∗ has the maximum value in

the available channels and users.

• Then the allocated channel and user will be removed from the sets of
channels and users.

• We repeat the channel and user selection until there is no user or chan-
nel left.

• After channel allocation, the power for each user j in femtocell i is

calculated according to (5.39) for the worst case, and according to (5.31)
for normal case, respectively.

The details of the joint channel allocation and fast power control are
shown in Algorithm 8. This scheme is distributed since each FBS work inde-
pendently. Moreover, each FBS can periodically updating the joint channel
allocation and fast power control schemes by the changing of gijc because of
the change of other FBSs’ interference, the movements of FUs, and so on.
We will show the convergence in the simulation results in the next section.
The complexity of this scheme depends on the channel selection and power
allocation. For any FBS i, it is bounded by O((|Ni | × |Mi |)2 ) by employing
quicksort in channel and user selection. Similar as the first scheme, in this
scheme, whenever an FBS detects a return of primary users on the licensed
channel, it will perform the following procedures sequentially, i.e., inform its
user to switch to another channel with the least interference on the available
channel list, decide a transmission power according to (5.39), and update the
transmission power on other active channels.

100
5.4 Simulation results and discussion

Algorithm 8 Proposed joint channel allocation and fast power control al-
gorithm for FBS i
Input: i, Ni , Mi .
Output: {xijc }, {pijc }.
0 0
1: Initialization: Mi ← Mi , Ni ← N .
0
2: while Mi 6= ∅ do
0
3: if Ni = ∅ then
4: Break; . not enough channels for femtocell i.
5: else
6: {j ∗ , cj ∗ } ← arg max
0 0
gijc
j∈Mi ,c∈Ni
0 0
7: Ni ← Ni − cj ∗
0 0
8: Mi ← Mi − j ∗
9: end if
10: end while
11: for j ∈ Mi do
12: Calculate pijcj by (5.39).
13: Calculate ξijc by (5.3).
14: if ξijc < ψ then
15: pijcj ← 0 . power allocation for user j in cell i is failed.
16: else
17: xijc ← 1
18: end if
19: end for

5.4 Simulation results and discussion

In this section, we evaluate our proposed downlink spectrum sharing schemes.
We have implemented a CogFem simulator based on MATLAB, where we
create a dense urban apartment topology as shown in Fig. 5.3. There are
maximal 10 rows of apartment buildings. Each row has maximal 10 buildings,
while each building has maximal 10 floors. The length, width, height of an
apartment are 10, 10, and 3 meters, respectively. We call the gap between
neighboring buildings in a row side gap, and set it as 1 meter, while we
call the gap between neighboring rows row gap, and set it as 5 meters. Each
apartment has an FBS, and has 2 to 4 users suggested by [26]. These users sit
randomly in each apartment. Without loss of generality, each FBS is located
at the middle of the apartment. The minimum required SINR is 0.01. The
average power of noise is −110dBmW . The bandwidth of each channel is
100KHz. For the estimation of channel gain in our simulation, we consider

101
 Spectrum Sharing in Cognitive Radio Femtocell Networks

Figure 5.3: An illustration of the simulation scenario

a slow fading channel, and use the COST 231 path loss model in (5.1). We
run each case 10 times with different random seeds for the number of users
in each femtocell and the number of available channels, and then calculate
the average capacity per femtocell.

5.4.1 Existing schemes

In this simulation study, we compare our proposed scheme to the existing
channel allocation and power allocation schemes. The most popular chan-
nel allocation scheme for cellular networks is coloring methods by which no
neighboring cells can use the same spectrum at the same time, e.g. in [27].
Regarding the power control method, we consider the fixed power control
method by which the total power budget is equally divided by the number

102
5.4 Simulation results and discussion

of users in each CogFem.

5.4.2 Average capacity with different density of apart-

ments
In the simulation for this purpose, we use power 10mW for the maximum
power budget in each FBS. We apply our proposed scheme in normal fem-
tocell networks and cognitive radio femtocell networks, respectively. In the
case of normal femtocells, the number of available channels is fixed to 10,
while the number of available channels is randomly changed from 10 to 20 in
the case of CogFem.

1.8
Proposed scheme for normal femtocell
average capacity per femtocell (Mbps)

Proposed scheme for CogFem

1.6

1.4

1.2

0.8

0.6

0.4

0.2
1 2 3 4 5 6 7 8 9 10
number of floors

Figure 5.4: Average capacity per femtocell in terms of number of floors in each
building. (3 rows, and 5 buildings per row)

Figure 5.4, 5.5, and 5.6 show the variation of average capacity per fem-
tocell while changing the number of floors, buildings, and rows, respectively.
We can see the average capacity per femtocell decreases while increasing the
number of floors, buildings, and rows, respectively. The reason is as follows.
When the number of floors, buildings, and rows increases, the number of
FBSs also increases. It then leads to more interference amongst femtocells
given a limited number of available channels. From the results in Fig. 5.4, 5.5,
and 5.6, CogFem achieved almost twice the average capacity of normal femto-
cells without CR capability by using our proposed scheme. This is essentially

103
 Spectrum Sharing in Cognitive Radio Femtocell Networks

1.8

1.6
average capacity per femtocell (Mbps)

1.4 Proposed scheme for normal femtocell

Proposed scheme for CogFem

1.2

0.8

0.6

0.4
1 2 3 4 5 6 7 8 9 10
number of buildings

Figure 5.5: Average capacity per femtocell in terms of number of buildings in

each row. (3 rows, and 5 floors per building)

0.9 Proposed scheme for normal femtocell

average capacity per femtocell (Mbps)

Proposed scheme for CogFem

0.8

0.7

0.6

0.5

0.4

1 2 3 4 5 6 7 8 9 10
number of rows

Figure 5.6: Average capacity per femtocell in terms of number of rows of

buildings. (5 floors per building, and 5 buildings per row)

104
5.4 Simulation results and discussion

due to more channel opportunities in CogFem than normal femtocells.

5.4.3 System performance with different number of

available channels
In the simulation for this purpose, we use power 20mW for the maximum
power budget in each FBS.

5
average capacity per femtocell (Mbps)

3 Fixed power control without iteration

Proposed scheme without iteration
Coloring
2

0
0 5 10 15 20 25 30
number of channels

Figure 5.7: Average capacity per femtocell in terms of number of channels. (5

rows, 5 buildings per row, 10 floors per building)

Figure 5.7 shows the variation of average capacity per femtocell while
changing the available channels. Here, we fix the topology as 5 rows, 5 build-
ings per row, and 10 floors per building. It shows that the average capacity
per femtocell increases while the number of available channels increases. That
is because more channel candidates can reduce the interference from neigh-
boring femtocells by allocating different channels to neighboring femtocells.
The fixed power control scheme using our channel allocation strategy in Al-
gorithm 6 without any iteration achieved almost the same average capacity
of our proposed scheme also without any iteration. Both of these schemes
achieved much higher average capacity than coloring method. Specifically,
when the number of channels is less than 5, the fixed power control scheme
and our proposed scheme can achieve almost twice of the capacity of the
coloring method. The performance gap reduced slightly until the number of
channels approaches 30.

105
 Spectrum Sharing in Cognitive Radio Femtocell Networks

0.9

0.8

Fixed power control without iteration

0.7 Proposed scheme without iteration
Coloring method
average user blocking rate

0.6

0.5

0.4

0.3

0.2

0.1

0
0 5 10 15 20 25 30
number of channels

Figure 5.8: Average blocking rate in terms of number of channels. (5 rows, 5

buildings per row, 10 floors per building)

The average blocking rate is defined as the ratio of total failed FUs to all
FUs. Those failed FUs exist if the SINR at any FU is lower than the required
minimum SINR, thus this FU will be not served. Figure 5.8 shows the average
blocking rate by using different channel allocation and power control schemes.
When the number of available channels is less than 5, the blocking rate for
the coloring method is higher than 50%. The blocking rate for fixed power
control scheme and our proposed scheme become zero when the number of
available channels turns to no less than 5, while the blocking rate for the
coloring method stops blocking when the available channels is more than 15.
The reason is that the coloring method requires the neighboring CogFem
can not use the same channels at the same time. It requires much more
channels to allocate all the neighboring FUs. On the contrary, the fixed
power control scheme and our proposed scheme are based on the interference
related channel allocation, so that neighboring CogFem can utilize the same
channel as long as the interference is not unberable for neighboring FBSs and
FUs.

5.4.4 The convergence

In the simulation for this purpose, we use power 20mW for the maximum
power budget in each FBS. We study a scenario with 10 available channels.
Figure 5.9 and 5.10 show the convergence of our proposed scheme and the

106
5.5 Conclusion

fixed power control scheme. In each iteration, the channel allocation mech-
anism will update according to the new interference measured. Figure 5.9
shows that both schemes can converge by a few iterations, e.g., around 5.
Our proposed scheme outperforms the fixed power control scheme by obtain-
ing around 2% higher average capacity. Both schemes have zero blocking
rate as shown in 5.10.

4.3

4.25
average capacity per femtocell (Mbps)

4.2

4.15
Fixed power control
Proposed scheme
4.1

4.05

3.95
0 5 10 15
number of iterations

Figure 5.9: Average capacity per femtocell in terms of number of Iterations.

(10 available channels, 10 floors per building, 5 buildings per row, and 5 rows)

5.5 Conclusion
In this chapter, we have investigated the spectrum sharing problem in down-
link transmission while applying CR technology into femtocell networks. We
formulated this problem as a MINLP problem and then use decomposition
methods to solve this problem. According to the solution of the decom-
posed problem, we proposed a joint channel allocation and fast power control
scheme. Simulation results showed that CogFem with more spectrum oppor-
tunities could achieve much higher capacity than normal femtocells depend-
ing on the number of available of channels. Our proposed scheme converges
very fast, and achieves much higher average capacity and lower user block-
ing rate than the coloring method. Using fixed power control together with

107
 Spectrum Sharing in Cognitive Radio Femtocell Networks

0.8
Fixed power control
0.6
Proposed scheme
average user blocking rate

0.4

0.2

−0.2

−0.4

−0.6

−0.8

−1
0 5 10 15
number of iterations

Figure 5.10: Average user blocking rate in terms of number of Iterations. (10
available channels, 10 floors per building, 5 buildings per row, and 5 rows)

our proposed channel allocation scheme only sacrifices 2% average capacity

comparing to using dynamic power control scheme.

108
Chapter 6

QoS-aware Spectrum Access for

Cognitive Radio Mesh
Networks

So far, we have investigated the resource optimization problem in one-hop

wireless network topology, including macrocell networks in chapter 3 and 4,
femtocell networks in chapter 5. In this chapter, we study the optimal channel
and route selection problems in multi-hop cognitive radio mesh networks.
In CogMesh, secondary mesh routers (SMRs) can opportunistically uti-
lize the primary licensed spectrum for the traffic of the secondary mesh
users (SMUs). We study the QoS problems for real-time communication
in CogMesh, where end-to-end delay should be less than a threshold. More-
over, different spectrum bands may have different quality in terms of SINR,
due to the spatial, time, and frequency selective fading. SMRs may select an
appropriate channel to achieve maximum data rate and minimum transmis-
sion latency. However, because of the uncertainty of the primary systems,
the channels on use may have to be released frequently, and will cause packet
loss and lots of retransmission incidents.
In this chapter, we formulate the optimization problem to select a route
and determine the channels on each link to maximize the route availability,
while guaranteeing the end-to-end transmission delay. We transform the
non-linear programing problem as a variant of Multiple-Choice Knapsack
Problem. We propose a channel and route selection scheme based on the
Lagrangian methods, and a low-complexity heuristic scheme.
Numerical results show that our proposed method acts much better than
the best SINR scheme and best channel availability scheme, in terms higher
successful solution ratio, higher route availability. Our proposed method
achieves quite close results to the optimization software MOSEK.

109
 QoS-aware Spectrum Access for Cognitive Radio Mesh Networks

The rest of the chapter is organized as follows. We introduce the system

model in Section 6.1 and formulate the optimization problem in Section 6.2.
In Section 6.3, we introduce the solution by Lagrangian relaxation. Then, we
propose a low-complexity heuristic scheme in Section 6.4. In Section 6.5, we
use matrix transformation so that optimization software such as MOSEK can
be used to solve our problem. In Section 6.6, we evaluate the performance of
our proposed scheme. Finally, we draw the conclusions in Section 6.7.

6.1 System model and assumptions

In this section, we describe the system model, introduce the channel model,
analyze the end-to-end delay, and route availability due to PU’s activities.

6.1.1 System model

Suppose an SMU wants to transmit some data through the CogMesh to a user
in the Internet as shown in Fig. 1.4. Since the communication bottleneck is
CogMesh, we focus on the QoS from the SMR by which the SMU is attached
to the secondary mesh gateway (SMG).

Figure 6.1: System model for route and channel selection in cognitive radio
mesh networks

Figure 6.1 illustrates the system model, while Table 6.1 lists the main
notations used in this chapter. In this model, there are sets of R routes from
the source SMR S to the gateway SMG G. For each route r (r = 1, ..., |R|),
the set of nodes can be denoted as Nr while the set of links can be denoted
as Lr (|Lr | = |Nr | − 1).

110
6.1 System model and assumptions

The available spectrum can be characterized in a set of M (M = 6 ∅)

channels. For each link l on route r, the available channels form a subset
of M, which can be denoted as Mrl (Mrl ∈ M). At any moment, each
channel can be either free or occupied by the primary system. Thus, these
M channels are not always be available for every link, and the SMRs will
stop working on the channel when PUs return. The SMRs will then either
wait until the channel becomes free or switch to another free channel. We
assume all SMRs and SMG are equipped with two radio transceivers. One is
CR which can dynamically choose working channels. The other one is used
for the control information exchange, wherein a dedicated narrow spectrum
band may be allocated for this purpose.

Table 6.1: Table of notations for cognitive radio mesh networks

Symbol Meaning
r a route
l a link
n an SMR
m a channel
R the set of routes
Lr the set of links for a given route r ∈ R
Irl the set of interfered links of l
Mrl the set of channels of link l on route r
Nr the set of SMRs on route r
L the packet length
D the packet end-to-end delay threshold
Bm the spectrum bandwidth of channel m
di,j the distance between SMR i and j
λrlm the transmission rate on channel m of link l on route r
xrlm the binary indicator of channel m at link i on route r
vrlm the channel availability of channel m at link l on route r
Γk the k-th SINR threshold

6.1.2 Channel model and adaptive modulation coding

Different channels at the same link may have different channel fading param-
eters, interference, and spectrum bandwidth, while the same channel may
have different channel fading parameters and interference on different links.

111
 QoS-aware Spectrum Access for Cognitive Radio Mesh Networks

We assume that the transmission power is fixed during the data transmission.
Therefore, the received SINR on each link of the CogMesh will be a variable.
According to different SINR, the modulation scheme used in this channel can
be different. Higher modulation scheme and wider spectrum bandwidth can
bring out higher data transmission rate and low transmission latency. We
assume that the channel fading is slow fading, which means that the chan-
nel quality will not change fast in a certain area. Therefore, each channel
between two neighbor SMRs has a fixed quality for a holding time.
We consider the time and frequency selective slow fading channels, and
use the Finite-State Markov Channel (FSMC) model [112] to represent of
the dynamic state of the wireless channel. Assume that all channels have
K + 1 states. In each state, the received SINR is different. We define Γk
(k = 0, 1, ..., K) as the lower bound threshold of the state k, where 0 = Γ0 <
Γ1 < ... < ΓK < ΓK+1 = ∞. We say link ei is in state k, if the SINR is
between Γk and Γk+1 .
AMC technique is used in our system model to adaptively change the
modulation scheme according to the quality of the channel. Where, chan-
nel’s quality can be estimated by the SINR measured on the receiving node.
Different modulation schemes can bring out different data transmission rate.
For a K + 1 state wireless channel with the bandwidth of B0 , we can
employ K types of modulation schemes. For any modulation scheme k (k =
1, ..., K), the data transmission rate is λk , while the SINR threshold is Γk .
Without loss of generality, let the sequence of Γ1 to ΓK be of increasing order.
Therefore, the data rate function can be defined as follows in equation (6.1).


 0 if ξ < Γ1
λ1 if Γ1 ≤ ξ < Γ2




 λ2 if Γ2 ≤ ξ < Γ3


f (ξ, B0 ) = . ... (6.1)
. ...




. ...




λk if ξ ≥ Γk


According to Shannon’s channel capacity formula, the maximum data

transmission rate is in directly proportional to the bandwidth. For any chan-
nel m with bandwidth Bm at any link l on route r, the data rate can be
achieved as follows.
Bm
λrlm = f (ξrlm , B0 ) (6.2)
B0
We employ a binary variable xrlm (xrlm ∈ {0, 1}, ∀r ∈ R; ∀l ∈ Lr ; ∀m ∈
Mrl ) to indicate whether channel m is selected for the link l on route r or
not. If xrlm is equal to 1, channel m is selected for the link l on route r, 0

112
6.1 System model and assumptions

otherwise. Therefore, the transmission data rate for link l on route r, can be
described as follows.
X
λrl = λrlm xrlm , ∀r ∈ R; ∀l ∈ Lr (6.3)
m∈Mrl

6.1.3 Interference-avoid channel selection for adjacent

links
We assume the antenna equipped at each SMR is half-duplex, which means
it can either transmit or receive data, but not both at the same time. The
1-hop neighbors can work in the same link, since they will not transmit data
at the same time. But there exist other links which are in the interference
range of the link. For example, suppose link li and lj work on the same
channel, we call link li is interfered by lj if SMR i + 1 is in the interference
range of SMR j (as shown in Fig. 6.2).

Figure 6.2: An illustration of interfered links

Suppose the distance between i + 1 and j is di+1,j , the interference range

of j is Rj . Thus, if di+1,j ≤ Rj , node i + 1 is in the interference area of
j. Following the scheduling method in [94], we do not allow the interference
links work on the same channel at the same time. Specially, we can assign
different channels for link li and lj .

6.1.4 End to end delay

We focus on the delay caused inside CogMesh, without considering the delay
from the source SMU to its nearby source SMR, and the delay from SMG to
the user in the Internet. The average end to end delay for route r consists of
the queueing and transmission delay Dr,t on each link along the route, and

113
 QoS-aware Spectrum Access for Cognitive Radio Mesh Networks

channel switching delay Dr,s at each intermediate SMRs. Thus, the end to
end delay is as follows.
Dr = Dr,t + Dr,s (6.4)
Subsequently, we give the expressions of Dr,t and Dr,s .

Queueing and transmission delay

Suppose the average queue length (number of packets buffered) is Qrl for link
l on route r, and the average delay for for a packet to deliver on channel m
at link l on route r is Drlm .
X X
Dr,t = (Qrl + 1) Drlm (6.5)
l∈Lr m∈Mrl

In [91], the authors presented a closed form for the queueing and transmis-
sion delay for wireless mesh networks using 802.11 distributed coordination
function (DCF) MAC protocol, considering inter-flow and intra-flow inter-
ference. According to [91], the average queueing and transmission delay in
wireless mesh networks can be expressed as follows.
" #
Nrlm
L 1 − αrlm
Drlm = + Brlm (6.6)
λrlm 1 − αrlm

where, for any channel m at link l on route r, Nrlm is the maximum number
of retransmissions, αrlm is the transmission failure probability, and assume it
is stable during the retransmissions of the packet [91], and Brlm is the back
off delay.
!
Nrlm
Wmin [1 − (2αrlm )Nrlm +1 ] 1 − αrlm
Brlm = − (6.7)
2(1 − 2αrlm ) 2(1 − αrlm )

where Wmin is the minimal contention window.

Channel switching delay

Channel switching delay caused at intermediate SMRs where the upstream
link and downstream link work on different channels. Let Ds0 denote the
switching delay at one SMR from one channel to another. For example, in
IEEE standard 802.11-2007, the switching delay is defined as 224µs [113].
Assume each SMR can finish receiving all buffered packets before it starts
forwarding to its next SMR. Thus, channel switching only happens at most
once for a given SMR during a packet travels on the route. Then, the total

114
6.1 System model and assumptions

switching delay accumulated by all the SMRs on the route r to SMG can be
calculated as follows.
P
Dr,s = Ds,l
l∈Lr
P P |xrim −xrjm | (6.8)
= Ds0 2
.
j=i+1;i,j∈Lr m∈M

where Ds,l denotes the switching delay on the SMR where link l is the in-
coming link.

6.1.5 Route availability

We consider the route availability in the CogMesh from the source SMR to the
SMG. In CogMesh, routes become unavailable mostly in the case when SMRs
are working on channels where PUs return. We employ channel availability
to distinguish the difference of PUs’ activities on different channels. The
channel availability of channel m at link l on route r can be defined as
follows.
I
E(Trlm )
vrlm = I B
, ∀l ∈ Lr , m ∈ M. (6.9)
E(Trlm ) + E(Trlm )

I B
where Trlm and Trlm represent the idle and busy time of primary user on
channel m at link l on route r, respectively. Higher channel availability also
I B
indicates lower transmission error and packet loss rate. E(Trlm ) and E(Trlm )
represent the mean value of idle and busy time, respectively. The channel
availability is measured periodically, and will update accordingly.
Assume each link can only work on one channel. Let vr denote the end
to end route availability from SMR to the SMG on route r, we can obtain
!
Y X
vr = vrlm xrlm (6.10)
l∈Lr m∈Mrl

where xrlm is a 0-1 binary variable, which indicates the channel selection
strategy for channel m at link l on route r. Since we assume only one channel
is used for a given link, we have
X
xrlm = 1, ∀l ∈ Lr ; r ∈ R
m∈Mrl

115
 QoS-aware Spectrum Access for Cognitive Radio Mesh Networks

6.2 Problem formulation

In this section, we formulate the optimization problem to maximize route
availability while considering the end-to-end delay and interference constraints
between neighbouring links.

6.2.1 Formulation of route and channel selection

For any route r (r ∈ R), we formulate a channel selection problem to maxi-
mize route availability while guaranteeing the end-to-end delay and interfer-
ence constraints. We assume that any link l (l ∈ Lr ) has at least one channel
that can be used. The problem we study is how does every link choose an
optimal channel for data transmission. This decision process may happen in
the following cases. 1), When the source SMR wants to start a new session of
data transmission. 2), When any SMR has to stop transmission on a channel
because of PUs’ return. 3), When the channel quality varies and affect the
end-to-end delay. Hence, the problem is formulated as follows:
P1

maximize vr (6.11)
subject to:

Dr ≤ D (6.12)

X
xrlm = 1, ∀l ∈ Lr (6.13)
m∈Mrl

X
xrim + xrlm ≤ 1, ∀l ∈ Lr ; m ∈ Mrl ; (6.14)
i∈Irl

xrlm ∈ {0, 1}, ∀l ∈ Lr ; m ∈ Mrl . (6.15)

where D is the required delay threshold. Constraint (6.12) represents the
end-to-end delay can not exceed the threshold D. Constraint (6.13) indicates
that each link should work on one and only one data channel. Constraint
(6.14) means the link in interference range area can not work on the same
channel. The solution is to find out every xrlm (∀l ∈ Lr , m ∈ Mrl ), so that
all the constraints are satisfied and the objective function vr is maximized.
Since vr is nonlinear function to xrlm , the formulated problem is a nonlinear
integer problem.

116
6.2 Problem formulation

After we select channels for all possible routes, we would finally select one
route with the maximum route availability with PUs, while the end-to-end
delay is guaranteed.
r∗ = arg max vr (6.16)
r∈R

We can also use the above metric to select multiple routes as backup routes.
Note, the channel and route selection scheme should update periodically
according to periodical measure results of availability on each channel as well
as the channel quality.

6.2.2 Problem transformation

The formulated nonlinear problem can be transformed into linear by the
following methods.

Transform the nonlinear objective function into linear form

Using the monotonously
Q P increased function ln, we can transform the objective
function from to . The deduction is as follows.

!
Y X
maximize vr ⇔ maximize vrlm xrlm
l∈Lr m∈Mrl
!
Y X
⇔ maximize ln vrlm xrlm
l∈Lr m∈Mrl
!
X X
⇔ maximize ln vrlm xrlm
l∈Lr m∈Mrl
X X
⇔ maximize vrlm xrlm
l∈Lr m∈Mrl

In consequence, we can reformulate the objective function (6.11) in P1 as

follows:
X X
maximize vrlm xrlm (6.17)
l∈Lr m∈Mrl

Transform the nonlinear constraint into linear form

Substituting (6.4) and (6.5) into (6.12), we have

117
 QoS-aware Spectrum Access for Cognitive Radio Mesh Networks

X X
(Qr,l + 1) Drlm xrlm + Dr,s ≤ D (6.18)
l∈Lr m∈Mrl

For the channel switching delay Dr,s along the route r, we consider the
worst case where each adjacent link works in different channels. Therefore,

Dr,s = Ds0 (|Lr | − 1) (6.19)

which is a constant for the variable xrlm . The benefit of this worst case
consideration can result in a more reliable solution for channel selection.
The reason is as follows: channels may change during the flow transmission,
the number of switching channels may vary. If we can guarantee the worst
case channel switching delay, it is believed that our solution is feasible for all
the cases.
Substituting (6.19) into (6.18), we have
X X
(Qrl + 1) Drlm xrlm ≤ D − Ds0 (|Lr | − 1) (6.20)
l∈Lr m∈Mrl

Moreover, we introduce positive variable wrlm to denote the coefficient in

the modified objective function and constraint function as follows.

(Qrl + 1)Drlm
wrlm = , ∀l ∈ Lr ; m ∈ Mrl (6.21)
D − Ds0 (|Lr | − 1)
For any channel m in link l on route r, the analog meaning of vrlm is the
value (profit), while the meaning of wrlm is the weight (cost). Therefore, the
nonlinear constraint (6.12) changes into a linear constraint. Thus, the original
non-linear problem can be reformulated as the following linear programming
problem. The re-transformed problem can be defined as follows.
P2
X X
Maximize vrlm xrlm (6.22)
l∈Lr m∈Mrl

Subject to:
X X
wrlm xrlm ≤ 1 (6.23)
l∈Lr m∈Mrl

X
xrlm = 1, ∀l ∈ Lr (6.24)
m∈Mrl

118
6.3 Solutions from Lagrangian relaxations

X
xrim + xrlm ≤ 1, ∀l ∈ Lr ; m ∈ Mrl (6.25)
i∈Irl

xrlm ∈ {0, 1}, ∀l ∈ Lr ; m ∈ Mrl . (6.26)

This is a 0−1 integer linear problem, which is in general NP-complete [114].
Moreover, without constraint (6.25), this problem can be viewed as an in-
stance of Multiple-Choice Knapsack Problem, where we have |Lr | mutually
disjoint classes (links) of items (channels) to be packed into a knapsack of
capacity 1. Each item m (m ∈ Mrl ) has a profit vrlm and a cost wrlm . The
problem is to choose exactly one item from each class such that the total
profit is maximized without exceeding the capacity. In addition, the item in
interfered classes should be varied from each other.

6.3 Solutions from Lagrangian relaxations

By introducing the Lagrangian multipliers p (p ≥ 0), we can get the La-
grangian as follows
!
P P P P
L1 (p) = vrlm xrlm + p 1 − wrlm xrlm
l∈Lr m∈Mrl
P P l∈Lr m∈Mrl (6.27)
= βrlm xrlm + p
l∈Lr m∈Mrl

where
βrlm = vrlm − pwrlm (6.28)
In this chapter, we call βrlm the Lagrangian price. The Lagrangian relaxed
problem is
P3

ZL1 (p) = max L1 (p) (6.29)

s.t. (6.24), (6.25), and (6.26).
All feasible solutions to P2 are also feasible solutions to P3. Given an
optimal solution set of {x∗rlm } for P2, we have
!
vrlm x∗rlm + p 1 − wrlm x∗rlm
P P P P
ZL1 (p) =
l∈Lr m∈Mrl l∈Lr m∈Mrl
P P ∗ (6.30)
≥ vrlm xrlm
l∈Lr m∈Mrl
=Z

119
 QoS-aware Spectrum Access for Cognitive Radio Mesh Networks

Therefore, ZL1 (p) is an upper bound for Z. This bound can be derived by
choosing the channel with the maximum Lagrangian price βrlm in each link,
and by checking the constraint (6.25). The corresponding solution is xrlm = 1
for m = arg max βrlm and 0 otherwise. The tightest bound can be found by
m∈Mrl
solving the Lagrangian dual problem.

ZLD = min ZL1 (p) (6.31)

p≥0

The Lagrangian dual problem ZLD yields the minimum upper bound from all
Lagrangian relaxations. It can be stated as a linear programming problem
to find an optimal vector of multipliers p.
For the Lagrangian multiplier v, we can use a subgradient method to
update it as follows.
" !#+
X X
pk+1 = pk + tk 1 − wrlm x∗rlm (6.32)
l∈Lr m∈Mrl

where xrl∗ m is an optimal solution to P3 and tk is a positive scalar step size,

which can be set as follows according to [115],
∆k (Z ∗ − ZL1 (pk−1 ))
tk = 2 (6.33)

wrlm x∗rlm
P P
1 −

l∈Lr m∈Mrl

where ∆k is a positive scalar satisfying ∆k ∈ (0, 2]. Z ∗ is a upper bound on

ZL1 . Since any feasible solution of P2 serves as an upper bound on P2, and
can also used as the upper bound of ZL1 . ∆k is often determined by setting
∆0 = 2 and halving ∆k whenever ZL1 (p) has failed to decrease in some fixed
number of iterations. p0 = 0.
For each link l, the channel selection procedure can work in a sequential
and cooperative way. It can also be extended to simultaneous and indepen-
dent ways by different interference avoidance schemes.
In our strategy, the link with maximum βrlm will be selected to use channel
m, any other link l0 in the interfered range will remove its channel m, and
delete the βrl0 m accordingly. Subsequently, we follow this process until all
the links allocate a channel without interfering with each other. The detailed
algorithm is shown in Algorithm 9.
Let M̂r denote the sum of number of channels for all links on route r.
X
M̂r = |Mrl |
l∈Lr

120
6.4 Low-complexity heuristic channel selection schemes

Algorithm 9 Channel selection scheme for route r with Lagrangian Methods

Input: {Lr }, {Mrl }, {vrlm }, {wrlm }.
Output: {xijm }.
1: Initialize k = 0, ∆0 = 2 p0 = 0
2: Calculate the upper bound Z ∗
3: while Not Converged do
4: Update p according to (6.32).
5: while Lr 6= ∅ do
6: for each link l ∈ Lr do
7: for each available channel m do
8: βrlm ← vrlm − pwrlm . According to (6.28)
9: end for
10: Find the maximum βrlm on each link l.
11: end for
12: Find the link l∗ and channel m∗ with maximum βrlm over all links.
13: xrl∗ m∗ ← 1 . Assign channel m∗ for link l∗
14: Remove channel m∗ for the interfering links
15: Lr ← Lr − l∗
16: end while
17: Get the value of ZL1 (p)
18: Update tk according to (6.33).
19: end while

Assume it can be terminated in NI iterations. In each iterative, it spends

O(M̂r ) to get an optimal solution. Therefore, the total time complexity is
O(NI M̂r ).
This is the solution for the Lagrangian dual problem P3. There may be
a duality gap between P 3 and P 2. Therefore, to obtain the optimal solution
for P 2, we need to combine the Lagrangian methods with other techniques,
such as branch-and-bound, and dynamic programming. For example, the La-
grangian methods can be used to calculate the lower bound for each branching
in branch-and-bound method.

6.4 Low-complexity heuristic channel selec-

tion schemes
In this section, we propose a heuristic channel selection scheme for a given
route r, and then choose the route with maximum route availability. We also
introduce two alternative channel selection schemes in the end of this section.

121
 QoS-aware Spectrum Access for Cognitive Radio Mesh Networks

6.4.1 Proposed channel selection scheme

For any link l in route r, we sort the channels according to increasing weights
wrlm , and derive M∗rl . Therefore, the index of channels in M∗rl is different
from that in Mrl . We then construct an instance of knapsack by setting
ṽrlm = vrlm − vrl,m−1 , ∀l ∈ Lr , m = 2, 3, ..., |M∗rl |.
and
w̃rlm = wrlm − wrl,m−1 , ∀l ∈ Lr , m = 2, 3, ..., |M∗rl |.
and the residual capacity is
X
c̄ = 1 − wrl1
l∈Lr
If c̄ is less than 0 at this step, it indicates there is no solution, because
wrl1 is already the minimal weight from every link l.
If c̄ is non-negative, we need to check the interference constraints. The
method is to see if any two interfering links use the same channel. We need
to replace the channels by interfering links and check c̄ again. After we get
rid of the interference constraints and c̄ is still non-negative, we sort all the
link-channel pair according to decreasing incremental efficiencies defined as
follows.
ṽrlm
η̃rlm = (6.34)
w̃rlm
We then fill the knapsack up to capacity c̄ according to the order of
the link-channel pair sorting in terms of incremental efficiencies η̃rlm . Ca-
pacity constraint and interference constraints are checked before adding a
link-channel pair. After adding a link-channel pair, the channel m∗ used in
this link l∗ is marked as inactive from any interfering links Irl∗ , and the pre-
vious channel m0 from the same link l∗ in the knapsack is taken out, which
means xrl∗ m0 = 0, and channel m0 in interfering links Irl∗ is marked as active.
The residual capacity c̄ updates as follows.
c̄ = c̄ − w̃rlm
Following this approach until either the capacity constraint is broken or no
different channels can be assigned for interfering links. The channel selection
for all links on route r is then finished. The details of this scheme is shown
in Algorithm 10. The time complexity is O(M̂r )

6.4.2 Alternative channel selection schemes

In order to demonstrate the performance of our proposed channel selection
scheme, we introduce the following two channel selection schemes.

122
6.4 Low-complexity heuristic channel selection schemes

Algorithm 10 Proposed heuristic channel selection algorithm for a given

route r in CogMesh
Input: r, Lr , Mrl .
Output: {xijc }.
1: Initialization: c̄ ← 1
2: Calculate wrlm , vrlm , µrlm for all route link and channels.
3: for l ∈ Lr do
4: Remove the channels where wrlm > 1.
5: Sort the channels according to increasing wrlm and derive M∗rl .
6: for i = 2; i ≤ |M∗rl |; i + + do
7: ṽrli ← vrli − vrl,i−1
8: w̃rli ← wrli − wrl,i−1
9: end for
10: c̄ ← c̄ − wrl1
11: end for
12: if c̄ < 0 then
13: Return; . No solution in this case.
14: end if
15: Sort the link-channel pairs according to decreasing incremental efficien-
cies η̃ according to (6.34).
16: while Interfering links use the same channel do
17: Adjust channels for interfering links, update c̄
18: end while
19: if c̄ < 0 then
20: Return; . No solution in this case.
21: end if
22: while 1 do
23: while Interfering links use the same channel do
24: Reselect another link-pair {l∗ , m∗ } .
25: end while
26: c̄ ← c̄ − w̃rl∗ 1
27: if c̄ < 0 then
28: Return; . Has solution in this case.
29: else
30: Record the old channel index m0 for link l∗ in the knapsack.
31: Mark channel m0 for all the interfering links as active.
32: xrl∗ m0 ← 0
33: Mark channel m∗ for all the interfering links as inactive.
34: xrl∗ m∗ ← 1
35: {l∗ , m∗ } ← arg max ∗ η̃rlm . Get the index of link-channel
l∈Lr ,m∈Mrl
pair with the maximal η̃
36: end if
123
37: end while
 QoS-aware Spectrum Access for Cognitive Radio Mesh Networks

Channel selection with best SINR

In this method, every node selects the channel with maximum SINR which
can achieve highest data transmission rate, without considering the channel
availability. The similar scheme was used in [90] to verify their proposed
scheme.
In our scenario, we describe the scheme with best SINR as follows. For
each route r, the link l∗ with maximum SINR among all other links select
its channel m∗ first. The interfering links will remove m∗ from their channel
table. Follow this procedure, until every link selects a channel. We repeat
it for all other routes. Finally, the route r∗ with maximum sum of channel
availability will be selected. In the later performance evaluation, we denote
this scheme as best-SINR for short.

Channel selection with best availability

In this method, every node selects the channel with maximum channel avail-
ability, without considering SINR and other factors. For each route r, the
link l∗ with maximum channel availability vrlm is selected to use the channel
m∗ . The interfering links will remove m∗ . Follow this procedure until every
link selects a channel. We repeat it for all other routes. Finally, the route
r∗ with maximum sum of channel availability will be selected. In the later
performance evaluation, we denote this scheme as best-availability for short.

6.5 Matrix transformation for problem solv-

ing by optimization software
Matrix form is used in calculation the optimal solution using optimization
tools such as MOSEK [37] and CPLEX [98].
For any given route r, V1×M̂r denotes the profit matrix for all links and
channels, while W1×M̂r denotes the cost matrix for all links and channels.
Let HM̂r ×M̂r denote the interference matrix for all channels on all links on
route r.
 
(xr11 , xr12 , ..., xr1|Mr1 | )T ,
 (xr21 , xr22 , ..., xr2|Mr2 | )T , 
XM̂r ×1 = 
 
.. 
 . 
(xr|Lr |1 , xr|Lr |2 , ..., xr|Lr ||Mr|Lr | | )T

124
6.6 Simulation results and analysis

 
(1, 1, ..., 1)1×|Mr1 | (0, 0, ..., 0)1×|Mr2 | · · · (0, 0, ..., 0)1×|Mr|Lr | |
(0, 0, ..., 0)1×|M | (1, 1, ..., 1)1×|M | · · · (0, 0, ..., 0)1×|M
r|Lr | | 

r1 r2
G|Lr |×M̂r =

.. .. ... .. 
 . . . 
(0, 0, ..., 0)1×|Mr1 | (0, 0, ..., 0)1×|Mr2 | · · · (1, 1, ..., 1)1×|Mr|Lr | |

We introduce A to denote the joint matrix of W1×M̂r and HM̂r ×M̂r

 
HM̂r ×M̂r
A(M̂r +1)×M̂r =
W1×M̂r

maximize V1×M̂r XM̂r ×1

subject to A(M̂r +1)×M̂r XM̂r ×1 ≤ 1
(6.35)
G|Lr |×M̂r XM̂r ×1 = 1|Lr |×1
x ∈ {0, 1}
In the following simulation study, we use MOSEK to get the optimal
solution for the above problem.

6.6 Simulation results and analysis

We have implemented a CogMesh simulator on MATLAB platform. We
consider a grid topology similar to the topology used in [91], where SMRs
are uniformly placed. The interference range of any SMR is one hop.
Detailed simulation parameters are shown in Table 6.2. The number of
available routes is 10. For each route, the number of hops changes from 2 to
10. For each link, the number of available channels changes from 2 to 10. We
follow the AMC table in [90], where there are four modes shown in Table 6.3.
Each channel on every link selects a data rate from {11, 5.5, 2, 1} Mbps
according to the quality of that channel. We assume the traffic follows the
constant bit rate with the packet size of 512 bytes. The maximum number of
retransmissions Krlm is randomly generated in [1, 5]. The slot time is 20µs,
while the minimum contention window size Wmin is 0.2ms which is 10 slots.
The channel availability vrlm is randomly generated in [0.5,1]. For any given
number of routes, hops, and channels, we randomly generated the parameters
for 200 times.
To verify and compare the performance of our proposed scheme with best-
SINR scheme, best-availability scheme, and optimal solution from MOSEK,
we introduce the following definition of successful solution ratio.

125
 QoS-aware Spectrum Access for Cognitive Radio Mesh Networks

Table 6.2: Simulation parameters for CogMesh

Symbol Value Symbol Value

B0 10 MHz vrlm [0.5, 1]
Ts0 0.224ms Nrlm [1, 5]
D 40ms L 512 Bytes
Wmin 0.2ms α [0, 0.5]
Ql [1, 10]

Table 6.3: AMC code rate and SINR table for CogMesh

Data rate (Mbps) 11 5.5 2 1

SINR (dB) [8, +∞) [6, 8) [4, 6) [0, 4)

Successful solution ratio Assume the total number of seeds for the sim-
max
ulation is Nseed . In each seed, there are |R| different routes. We call a
seed has a valid solution, if there is at least one solution from its |R| different
valid
routes. Let Nseed denote the sum of seeds with valid solution. The successful
valid max
solution ratio is the ratio between Nseed and Nseed .
valid
Nseed
successful solution ratio := max
(6.36)
Nseed
max
In our simulation, Nseed is equal to 200, and |R| is equal to 10. A higher
successful solution ratio indicates a better channel selection scheme. In addi-
tion, we also present the results of route availability from all different channel
selection schemes. Route availability is set to 0 for those seeds which can not
get valid solutions. We calculate and compare the average route availability
from all 200 seeds for all schemes.

6.6.1 Successful solution ratio

The successful solution ratio for all schemes are shown in Fig. 6.3 (3D visu-
alization) and Fig. 6.4. Both figures show that best-availability scheme has
the lowest successful solution ratio when the number of hops is more than 6.
The reason is the best-availability scheme selects the channel with maximal
channel availability for each link without considering the delay constraints.
The solutions from best-availability scheme have higher chance to break the

126
6.6 Simulation results and analysis

delay constraints and result in invalid solutions. Our proposed scheme has
similar successful solution ratio with best-SINR scheme, since both scheme
take the delay constraints into account.

1 1
Successful Solution Ratio

Successful Solution Ratio

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
10 10
8 10 8 10
6 8 6 8
Nu 6 Nu 6
mb 4 mb 4
er o 4 nels er o 4 nnels
f hop f chan f hop f cha
s 2 2 ber o s 2 2 ber o
Num Num

(a) Best-SINR (b) Best-availability

1 1
Successful Solution Ratio

Successful Solution Ratio

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
10 10
8 10 8 10
6 8 6 8
Nu 6 Nu 4 6
mb 4 mb
er o 4 nels er o 4 nnels
f hop f chan f hop 2 f cha
s 2 2 ber o s 2 ber o
Num Num

(c) MOSEK (d) Proposed

Figure 6.3: Successful solution ratio from different channel selection schemes
(3D visualization)

Figure 6.4 shows that when there are 2 or 3 channels, our proposed scheme
and best-SINR scheme achieve a bit lower successful solution ratio than the
optimal solution from MOSEK. For both our proposed scheme and best-SINR
scheme, the successful solution ratio starts dropping from 6 hops when there
are only 2 channels, while it starts dropping from 9 hops when there are 3
channels. From the optimal solution, it starts dropping from 7 hops instead
of 6 hops when there are 2 channels, and it starts dropping from 10 hops
instead of 9 hops when there are 3 channels. When there are more than 3
channels, our proposed scheme and best-SINR scheme always have the same
100% successful solution ratio as the optimal solution from MOSEK.

127
 QoS-aware Spectrum Access for Cognitive Radio Mesh Networks

1 1

0.9 0.9

0.8 0.8

0.7 0.7
Successful Solution Ratio

Successful Solution Ratio

0.6 2 channels 0.6
3 channels
0.5 0.5 2 channels
4 channels
3 channels
5 channels
0.4 0.4 4 channels
6 channels
5 channels
7 channels
6 channels
0.3 8 channels 0.3
7 channels
9 channels
8 channels
0.2 10 channels 0.2
9 channels
10 channels
0.1 0.1

0 0
2 3 4 5 6 7 8 9 10 2 3 4 5 6 7 8 9 10
Number of Hops Number of Hops

(a) Best-SINR (b) Best-availability

1 1

0.9
0.9
0.8

0.8 0.7
Successful Solution Ratio

Successful Solution Ratio

2 channels 2 channels
0.6
0.7 3 channels 3 channels
4 channels 4 channels
0.5
5 channels 5 channels
0.6 6 channels 6 channels
7 channels 0.4
7 channels
8 channels 8 channels
0.5 9 channels 0.3 9 channels
10 channels 10 channels
0.2
0.4
0.1

0
2 3 4 5 6 7 8 9 10 2 3 4 5 6 7 8 9 10
Number of Hops Number of Hops

(c) MOSEK (d) Proposed

Figure 6.4: Successful solution ratio from different channel selection schemes

6.6.2 Route availability

The route availability for all schemes are shown in Fig. 6.5 (3D visualization)
and Fig. 6.6. Both figures show that best-SINR scheme has the lowest route
availability in all the cases. The reason is the best-SINR scheme selects
the channel with maximal channel SINR for each link without considering
the channel availability. The solutions from best-SINR scheme have higher
chance to select channels with lower channel availability and result in lower
route availability solutions.
From Fig. 6.5, we can see the route availability got from our proposed
scheme has the closest pattern of the optimal solution from MOSEK com-
paring with the route availability from other two schemes.
From Fig. 6.6, we can see our proposed scheme achieves a bit lower route
availability than the best-availability when there are less than 6 hops. How-
ever, the route availability got by the best-availability scheme drops faster

128
6.7 Conclusion

1 1

0.8 0.8
Route Availability

Route Availability
0.6 0.6

0.4 0.4

0.2 0.2

0 0
10 10
8 10 8 10
6 8 6 8
Nu 6 Nu 6
mb 4 mb 4
er o 4 nels er o 4 nnels
f hop f chan f hop f cha
s 2 2 ber o s 2 2 ber o
Num Num

(a) Best-SINR (b) Best-availability

1 1

0.8 0.8
Route Availability

Route Availability

0.6 0.6

0.4 0.4

0.2 0.2

0 0
10 10
8 10 8 10
6 8 6 8
Nu 6 Nu 6
mb 4 mb 4
er o els er o nnels
f hop
4
f chann f hop
4
f cha
s 2 2 ber o s 2 2 ber o
Num Num

(c) MOSEK (d) Proposed

Figure 6.5: Route availability from different channel selection schemes (3D
visualization)

than that got by our proposed scheme when the number of hops keeps in-
creasing from 6 to 10. Our proposed scheme outperforms the best-availability
scheme in case of longer hops (larger than 6). Moreover, the route availabil-
ity achieved by our proposed scheme is closest to the optimal solution from
MOSEK among all other three schemes in long hop cases.

6.7 Conclusion
In this chapter, we have investigated the real-time communication problem
in CogMesh. We have formulated this problem of maximization the route
availability, while guaranteeing the end-to-end delay from SMR to the gate-
way. We transformed the original non-liner integer programming problem
to a linear integer programming problem. Then we modeled it as a vari-

129
 QoS-aware Spectrum Access for Cognitive Radio Mesh Networks

0.9 1

2 channels
0.8 0.9
3 channels
4 channels
0.8
0.7 5 channels
6 channels
0.7
0.6 7 channels
Route Availability

Route Availability
8 channels
0.6
9 channels
0.5
10 channels
0.5
0.4 2 channels
0.4 3 channels
4 channels
0.3
0.3 5 channels
6 channels
0.2 7 channels
0.2
8 channels
0.1 0.1 9 channels
10 channels
0 0
2 3 4 5 6 7 8 9 10 2 3 4 5 6 7 8 9 10
Number of Hops Number of Hops

(a) Best-SINR (b) Best-availability

1 1

0.9 0.9

0.8 0.8

0.7 0.7
Route Availability

Route Availability

0.6 0.6

0.5 0.5
2 channels 2 channels
0.4 3 channels 0.4 3 channels
4 channels 4 channels
0.3 5 channels 0.3 5 channels
6 channels 6 channels
0.2 7 channels 0.2 7 channels
8 channels 8 channels
0.1 9 channels 0.1 9 channels
10 channels 10 channels
0 0
2 3 4 5 6 7 8 9 10 2 3 4 5 6 7 8 9 10
Number of Hops Number of Hops

(c) MOSEK (d) Proposed

Figure 6.6: Route availability from different channel selection schemes

ant of the multiple-choice knapsack problem, and proposed a low-complexity

heuristic algorithm to solve it. Simulation results showed that our proposed
scheme achieved quite close successful solution ratio and route availability to
the results from MOSEK, and outperformed the channel selection schemes
based on best SINR and best channel availability schemes in terms of higher
successful solution ratio and better route availability in most cases.

130
Chapter 7

Conclusion and Future Work

Throughout this thesis, we have studied the resource optimization problems

in CogCell, CogFem, and CogMesh. Our contributions include the problem
modeling, formulation, and solutions. Our work sheds a light on the future
deployment of cognitive radio technology into cellular networks, femtocell
networks, and mesh networks. In this chapter, we will draw our conclusions
on both research scenarios and methods, and present the future work.

7.1 Conclusion
7.1.1 On research scenarios
Cognitive radio cellular networks
Firstly, we have addressed the admission and power control problem in one-
channel CogCell. The objective is to maximize the secondary revenue to
the operator, while guaranteeing the interference constraints on primary re-
ceivers. In addition, the QoS level in terms of data rate is satisfied for
admitted SUs. In our earlier study, we proposed a joint admission and power
control scheme using minimal revenue efficiency removal algorithm called
JAPC-MKP to address the operator problem. In our later study, we further
improved JAPC-MRER by reformulation and remodeling. The admission
and power control problem is reformulated and remodeled as a MKP. Then,
we propose a novel admission and power control scheme called JAPC-MKP
which is heuristic with low complexity. Finally, simulation results show that
our proposed JAPC-MKP can approach the optimal results from the opti-
mization software MOSEK [37], and greatly outperform the previous fixed
power scale JAPC-MRER schemes.
Secondly, we have studied the multi-channel CogCell scenario, where

131
 Conclusion and Future Work

channel allocation strategies need to consider in addition to admission and

power control. We have the same objective as the one-channel CogCell case.
But we have more resource in terms of number of channels. We formulate the
joint channel allocation, admission and power control problem as a MINLP
problem which is NP-hard in general. Then, we modeled it as a MMKP,
and proposed a heuristic method to solve it. Our solution is quite close to
MOSEK from the simulation results.

Cognitive radio femtocell networks

In the scenario of CogFem, we have studied the spectrum sharing problem

to maximize the total capacity of femtocell networks. We employed mixed
primal and dual decomposition methods to solve the spectrum sharing prob-
lem. We also studied the robust optimization considering the worst case due
to FUs’ random movements. According to the solution of the decomposed
problem, we proposed a joint channel allocation and power control scheme.
Simulation results showed our proposed channel allocation scheme achieved
much higher capacity and lower user blocking rate than traditional coloring
method. Our proposed dynamic power control scheme can converge very
fast. Using the fixed power control scheme together with our proposed chan-
nel allocation scheme achieves only 2% less capacity than the dynamic power
control scheme.

Cognitive radio mesh networks

In the scenario of CogMesh, we have investigated the real-time communica-

tion problem. We formulated this problem of maximization the route avail-
ability, while guaranteeing the end-to-end delay from SMR to the gateway.
We transformed the original non-liner integer programming problem to a
linear integer programming problem. Then we modeled it as a variant of
MCKP. Based on the MCKP modeling, we proposed a heuristic method to
solve this problem. Simulation results showed that our proposed scheme
achieved quite close successful solution ratio and route availability to the re-
sults from MOSEK, and outperformed the channel selection schemes based
on best SINR and best channel availability schemes in terms of higher suc-
cessful solution ratio and better route availability in most cases.

132
7.1 Conclusion

7.1.2 Comparison
Overlay vs underlay spectrum sharing modes
Basically, the spectrum utilization efficiency is higher in the underlay spec-
trum sharing mode than that in the overlay spectrum sharing mode, because
SUs in the underlay spectrum sharing mode can use the spectrum even PUs
exist. The problem to solve in the underlay spectrum sharing mode is to
carefully control the interference to primary receivers, so that the interfer-
ence is not harmful to primary receivers, such as our study in CogCell. On
the other hand, channel allocation strategy is very important in the overlay
spectrum sharing mode. Where we need to take channel availability into
account to design a most reliable route in CogMesh.

One-hop vs Multi-hop scenarios

In one-hop scenarios, we have two scenarios: CogCell and CogFem.
In CogCell, we need to consider the interference from SUs to primary
receivers in the underlay spectrum sharing mode. In CogFem, we need to
consider the interference between secondary femtocell base stations.
In multi-hop scenarios, the interference between neighbouring links in
interference range should take into account. The end-to-end performance is
more important than one hop performance in multi-hop scenarios.

One-channel vs Multiple-channel scenarios

In one-channel scenarios, optimal power and admission control can achieve
optimal revenue to secondary base station operators, while in multiple-channel
scenarios, channel allocation strategies dominates power and admission con-
trol.

7.1.3 On research methods

We have done our research by problem modeling and formulations. Then
we design algorithms according to the solution of the optimization problem,
and use our simulator based on MATLAB to verify the performance of our
proposed algorithms. We also compare our solutions to optimization software
MOSEK.
We have learnt from Chapter 3 that a proper problem modeling and for-
mulation dominates the algorithm design and final performance. It is very
important for problem solving and algorithm design. According to our ear-
lier problem formulation, we proposed a joint admission and power control

133
 Conclusion and Future Work

Table 7.1: Summary of different scenarios

Scenarios Objective Constraints Modeling Solutions

One-Channel Secondary Date rate, Inter- Heuristic algo-
MKP
CogCell revenue ference, power rithm
Multi-
Secondary Date rate, Inter- Heuristic algo-
Channel MMKP
revenue ference, power rithm
CogCell
Sum of ca- primal and dual
CogFem SINR, power MINLP
pacity decomposition
Route Delay, interfer- Variant Heuristic algo-
CogMesh
availability ence, power MCKP rithm

scheme using minimal revenue efficiency removal algorithm called JAPC-

MKP to address the operator problem. We further improve JAPC-MRER by
reformulation and remodeling our problem as a MKP. According to the solu-
tion of the reformulated problem, we proposed a novel admission and power
control scheme called JAPC-MKP which is heuristic with low complexity.
Simulation results show that our proposed JAPC-MKP can approach the
optimal results from the optimization software MOSEK, and greatly outper-
form the previous fixed power scale JAPC-MRER schemes.
We summarize our methods as follows: Modeling and Formulation ->
Re-Formulation -> Problem re-modeling to a kind of well studied problems
(such as knapsack problems) -> design algorithms based on solutions to well
studied problems (such as knapsack problems).

7.2 Future work

In future, there are several directions in the research of resource optimization
in cognitive radio networks.

• Efficient way to get information of primary receivers

The information of primary receivers includes the geo-locations and
interference threshold is quite important in resource optimization in
cognitive radio networks. In [116], the authors introduced a primary re-
ceiver detection method by exploiting the local oscillator leakage power
emitted by the RF front end of primary receivers.
In practice, this kind of information may be done by dedicated pri-
mary receiver detection server, cooperated with lots of sensors. Those

134
7.2 Future work

information can be stored and the operator can use it while doing op-
timization. And this information should be updated regularly.

• Extension to more objective functions

The objective we studied in this thesis includes secondary revenue,
capacity, and route availability. In the future, we want to extend our
studies into more objective functions, e.g., fairness.

• Demonstration on testbeds
So far we have verified our algorithms and schemes by simulations.
In the future, it will be more helpful to implement our algorithms
on testbeds. The testbed platforms can be GNU Radio [117] with
USRP [118], WARP [119], and ASRP2 [120].

135
Bibliography

[1] “United States radio spectrum frequency allocations chart as of 2003.”

[Online]. Available: http://www.ntia.doc.gov/osmhome/allochrt.pdf

[2] S. Haykin, “Cognitive radio: brain-empowered wireless communica-

tions,” IEEE Journal on Selected Areas in Communications, vol. 23,
no. 2, pp. 201 – 220, feb. 2005.

[3] (2010) Fcc. [Online]. Available: http://www.fcc.gov/

[4] (2010) Ecc. [Online]. Available: http://www.ero.dk/ECC1/

[5] (2010) The norwegian post and telecommunications authority (npt).

[Online]. Available: http://http://www.npt.no/

[6] (2010) Ofcom. [Online]. Available: http://www.ofcom.org.uk/

[7] (2008) TeliaSonera to build 4G - next generationmobile net-

work. http://www.teliasonera.com/News-and-Archive/Press-
releases/2008/TeliaSonera-to-build-4G–next-generation-mobile-
network/.

[8] (2010) TeliaSonera has won a 4G license in Den-

mark. http://www.teliasonera.com/News-and-Archive/Press-
releases/2010/TeliaSonera-has-won-a-4G-license-in-Denmark/.

[9] (2009) Sonera wins 4G frequencies in the Finnish auc-

tion. http://www.teliasonera.com/News-and-Archive/Press-
releases/2009/Sonera-wins-4G-frequencies-in-the-Finnish-auction/.

[10] Federal Communications Commission Spectrum Policy Task Force.

(2002, November) FCC Report of the Spectrum Efficiency Working
Group. http://www.fcc.gov/sptf/files/SEWGFinalReport 1.pdf.

137
 BIBLIOGRAPHY

[11] Shared Spectrum Company, “Spectrum reports,”

http://www.sharedspectrum.com/papers/spectrum-reports/, Jan-
uary 2011.
[12] M. A. McHenry and D. McCloskey, “Spectrum occupancy measure-
ments: Chicago, illinois, november 16-18, 2005,” Shared Spectrum
Company, Tech. Rep., 2005.
[13] T. Erpek, M. Lofquist, and K. Patton, “Spectrum occupancy measure-
ments: Loring commerce centre, limestone, maine, september 18-20,
2007,” Shared Spectrum Company, Tech. Rep., 2007.
[14] S. S. Company, “Spectrum occupancy measurements: Chicago, illinois,
september 1-5, 2009,” Tech. Rep.
[15] (2003) Cognitive Radio Technologies Proceeding (CRTP) ET Docket
No. 03-108. http://www.fcc.gov/oet/cognitiveradio/.
[16] (2004) ET Docket No. 04-186. http://www.fcc.gov/oet/projects/tvbanddevice/.
[17] “IEEE 802.22/D1.0 Draft Standard for Wireless Regional Area Net-
works Part 22: Cognitive Wireless RAN Medium Access Control
(MAC) and Physical Layer (PHY) Specifications: Policies and Pro-
cedures for Operation in the TV Bands,” IEEE Draft Standard 802.22,
April 2008.
[18] J. Mitola and G. Maguire, “Cognitive radio: making software radios
more personal,” IEEE Personal Communications, vol. 6, no. 4, pp. 13–
18, 1999.
[19] I. F. Akyildiz, W.-Y. Lee, M. C. Vuran, and S. Mohanty, “Next gen-
eration/dynamic spectrum access/cognitive radio wireless networks: a
survey,” Computer Networks, vol. 50, no. 13, pp. 2127–2159, 2006.
[20] (2009) TeliaSonera first in the world with 4G ser-
vices. http://www.teliasonera.com/News-and-Archive/Press-
releases/2009/TeliaSonera-first-in-the-world-with-4G-services/.
[21] M. Ahmed, “Call admission control in wireless networks: a comprehen-
sive survey,” Communications Surveys & Tutorials, IEEE, vol. 7, no. 1,
pp. 49–68, First Qtr. 2005.
[22] M. Chiang, P. Hande, T. Lan, and C. W. Tan, Power Control in Wire-
less Cellular Networks Power Control in Wireless Cellular Networks.
now Publishers Inc., May 2008.

138
BIBLIOGRAPHY

[23] (2007) Presentaions by ABI Research, Picochip, Airvana,

IP.access, Gartner, Telefonica Espana, the 2nd Interna-
tional Conference on Home Access Points and Femtocells.
http://www.avrenevents.com/dallasfemto2007/purchase presentations.htm.

[24] S. Higgins, C. Mah, and T. Anderson. (2009) Cisco medical-

grade network (mgn) 2.0 – wireless architectures. [Online].
Available: http://www.cisco.com/en/US/docs/solutions/Verticals/
Healthcare/MGN wireless adg.pdf

[25] V. Chandrasekhar, J. Andrews, and A. Gatherer, “Femtocell networks:

a survey,” IEEE Communications Magazine, vol. 46, no. 9, pp. 59–67,
September 2008.

[26] H. Claussen, L. T. W. Ho, and L. G. Samuel, “An overview of the

femtocell concept,” Bell Lab. Tech. J., vol. 13, no. 1, pp. 221–245,
2008.

[27] C. Peng, H. Zheng, and B. Y. Zhao, “Utilization and fairness in spec-

trum assignment for opportunistic spectrum access,” Mobile Networks
and Applications, vol. 11, no. 4, pp. 555–576, 2006.

[28] Y. Zhang, J. Luo, and H. Hu, Wireless Mesh Networking: Architec-

tures, Protocols and Standards (Wireless Networks and Mobile Com-
puting). Auerbach Publications, December 2006.

[29] B. H. Walke, S. Mangold, and L. Berlemann, IEEE 802 Wireless Sys-

tems: Protocols, Multi-Hop Mesh/Relaying, Performance and Spec-
trum Coexistence. Wiley, January 2007.

[30] K. R. Chowdhury and I. F. Akyildiz, “Cognitive wireless mesh

networks with dynamic spectrum access,” IEEE Journal on Selected
Areas in Communications, vol. 26, no. 1, pp. 168–181, 2008. [Online].
Available: http://dx.doi.org/10.1109/JSAC.2008.080115

[31] T. Chen, H. Zhang, G. Maggio, and I. Chlamtac, “Topology manage-

ment in cogmesh: A cluster-based cognitive radio mesh network,” in
IEEE International Conference on Communications (ICC ’07), June
2007, pp. 6516–6521.

[32] R. C. Pereira, R. D. Souza, and M. E. Pellenz, “Overlay

cognitive radio in wireless mesh networks,” 2008. [Online]. Available:
http://www.citebase.org/abstract?id=oai:arXiv.org:0805.3643

139
 BIBLIOGRAPHY

[33] T. Weiss and F. Jondral, “Spectrum pooling: an innovative strategy for

the enhancement of spectrum efficiency,” Communications Magazine,
IEEE, vol. 42, no. 3, pp. S8–14, Mar 2004.

[34] D. Tse and P. Viswanath, Fundamentals of Wireless Communication.

Cambridge University Press, June 2005.

[35] G. Dodig-Crnkovic, “Scientific methods in computer science,” in

Conference for the Promotion of Research in IT at New Universities
and at University Colleges in Sweden, April 2002. [Online]. Available:
http://www.mrtc.mdh.se/index.php?choice=publications&id=0446

[36] (2009) MATLAB. http://www.mathworks.com/products/matlab/.

[37] “The mosek optimization software.” [Online]. Available: http:

//www.mosek.com/

[38] J. Xiang, Y. Zhang, and T. Skeie, “Joint admission and power control
for cognitive radio cellular networks,” 11th IEEE Singapore Interna-
tional Conference on Communication Systems (ICCS 2008), pp. 1519–
1523, Nov. 2008.

[39] L. Zhang, Y.-C. Liang, and Y. Xin, “Joint Admission Control and
Power Allocation for Cognitive Radio Networks,” Acoustics, Speech and
Signal Processing, 2007. ICASSP 2007. IEEE International Conference
on, vol. 3, pp. III–673–III–676, 15-20 April 2007.

[40] J. Xiang, Y. Zhang, and T. Skeie, “Admission and power control for
cognitive radio cellular networks: A multidimensional knapsack solu-
tion,” CogART 2010, Nov. 2010.

[41] “Cognitive Radio Technologies Proceeding (CRTP) ET Docket No. 03-

108.” [Online]. Available: http://transition.fcc.gov/oet/cognitiveradio/

[42] F. F. Digham, M.-S. Alouini, and M. K. Simon, “On the energy detec-
tion of unknown signals over fading channels,” Communications, IEEE
Transactions on, vol. 55, no. 1, pp. 21–24, Jan. 2007.

[43] P. Sutton, K. Nolan, and L. Doyle, “Cyclostationary signatures in prac-

tical cognitive radio applications,” Selected Areas in Communications,
IEEE Journal on, vol. 26, no. 1, pp. 13–24, Jan. 2008.

[44] Y. Youn, H. Jeon, H. Jung, and H. Lee, “Discrete wavelet packet trans-
form based energy detector for cognitive radios,” Vehicular Technology

140
BIBLIOGRAPHY

Conference, 2007. VTC2007-Spring. IEEE 65th, pp. 2641–2645, 22-25

April 2007.

[45] A. Ghasemi and E. Sousa, “Collaborative spectrum sensing for op-

portunistic access in fading environments,” New Frontiers in Dynamic
Spectrum Access Networks, 2005. DySPAN 2005. 2005 First IEEE In-
ternational Symposium on, pp. 131–136, 8-11 Nov. 2005.

[46] M. Gandetto and C. Regazzoni, “Spectrum sensing: A distributed ap-

proach for cognitive terminals,” Selected Areas in Communications,
IEEE Journal on, vol. 25, no. 3, pp. 546–557, April 2007.

[47] P. A. K. Acharya, S. Singh, and H. Zheng, “Reliable open spectrum

communications through proactive spectrum access,” in TAPAS ’06:
Proceedings of the first international workshop on Technology and policy
for accessing spectrum. New York, NY, USA: ACM, 2006, p. 5.

[48] Q. Zhao, L. Tong, A. Swami, and Y. Chen, “Decentralized cogni-

tive mac for opportunistic spectrum access in ad hoc networks: A
pomdp framework,” Selected Areas in Communications, IEEE Journal
on, vol. 25, no. 3, pp. 589–600, April 2007.

[49] S. Geirhofer, L. Tong, and B. Sadler, “Cognitive radios for dynamic

spectrum access - dynamic spectrum access in the time domain: Mod-
eling and exploiting white space,” Communications Magazine, IEEE,
vol. 45, no. 5, pp. 66–72, May 2007.

[50] K. Chowdhury and I. Akyildiz, “Cognitive wireless mesh networks with

dynamic spectrum access,” Selected Areas in Communications, IEEE
Journal on, vol. 26, no. 1, pp. 168–181, Jan. 2008.

[51] Q. Zhao, S. Geirhofer, L. Tong, and B. Sadler, “Opportunistic spectrum

access via periodic channel sensing,” Signal Processing, IEEE Trans-
actions on [see also Acoustics, Speech, and Signal Processing, IEEE
Transactions on], vol. 56, no. 2, pp. 785–796, Feb. 2008.

[52] A. E. Leu, M. McHenry, and B. L. Mark, “Modeling and analysis of in-

terference in listen-before-talk spectrum access schemes,” Int. J. Netw.
Manag., vol. 16, no. 2, pp. 131–147, 2006.

[53] M. Sharma, A. Sahoo, and K. D. Nayak, “Channel selection under

interference temperature model in multi-hop cognitive mesh networks,”
The 2nd IEEE International Symposium on New Frontiers in Dynamic

141
 BIBLIOGRAPHY

Spectrum Access Networks, 2007 (DySPAN 2007), pp. 133–136, 17-20

April 2007.

[54] J. Bater, H.-P. Tan, K. N. Brown, and L. Doyle, “Modelling inter-

ference temperature constraints for spectrum access in cognitive radio
networks,” Communications, 2007. ICC ’07. IEEE International Con-
ference on, pp. 6493–6498, 24-28 June 2007.

[55] D. Xu, E. Jung, and X. Liu, “Optimal bandwidth selection in multi-

channel cognitive radio networks: How much is too much?” in The
3rd IEEE Symposium on New Frontiers in Dynamic Spectrum Access
Networks (DySPAN 2008), Oct. 2008, pp. 1–11.

[56] P. Fuxjager, D. Valerio, and F. Ricciato, “The myth of non-overlapping

channels: interference measurements in ieee 802.11,” in The 4th An-
nual Conference on Wireless on Demand Network Systems and Services
(WONS ’07), Jan. 2007, pp. 1–8.

[57] A. Mishra, V. Shrivastava, S. Banerjee, and W. Arbaugh, “Partially

overlapped channels not considered harmful,” ACM SIGMETRICS
Performance Evaluation Review, vol. 34, no. 1, pp. 63–74, 2006.

[58] “IEEE Standard for Local and metropolitan area networks Part 16: Air
Interface for Broadband Wireless Access Systems,” IEEE Std 802.16-
2009 (Revision of IEEE Std 802.16-2004), pp. C1–2004, 29 2009.

[59] H. Su and X. Zhang, “Cross-Layer Based Opportunistic MAC Protocols

for QoS Provisionings Over Cognitive Radio Wireless Networks,” IEEE
Journal on Selected Areas in Communications, vol. 26, no. 1, pp. 118–
129, Jan. 2008.

[60] D. Djonin, Q. Zhao, and V. Krishnamurthy, “Optimality and com-

plexity of opportunistic spectrum access: A truncated markov decision
process formulation,” in IEEE International Conference on Communi-
cations 2007 (ICC ’07), June 2007, pp. 5787–5792.

[61] Q. Zhao, L. Tong, and A. Swami, “A Cross-Layer Approach to Cogni-

tive MAC for Spectrum Agility,” in Conference Record of the Thirty-
Ninth Asilomar Conference on Signals, Systems and Computers (AC-
SSC ’05), 28 2005-Nov. 1 2005, pp. 200–204.

[62] H. Su and X. Zhang, “Cognitive Radio Based Multi-Channel MAC Pro-

tocols for Wireless Ad Hoc Networks,” in IEEE Global Telecommunica-
tions Conference 2007 (GLOBECOM ’07), Nov. 2007, pp. 4857–4861.

142
BIBLIOGRAPHY

[63] H. Kim and K. Shin, “Efficient Discovery of Spectrum Opportunities

with MAC-Layer Sensing in Cognitive Radio Networks,” IEEE Trans-
actions on Mobile Computing, vol. 7, no. 5, pp. 533–545, May 2008.

[64] M. Ma and D. Tsang, “Impact of channel heterogeneity on spectrum

sharing in cognitive radio networks,” IEEE International Conference
on Communications 2008 (ICC ’08), pp. 2377–2382, May 2008.

[65] Y. Kondareddy, P. Agrawal, and K. Sivalingam, “Cognitive radio net-

work setup without a common control channel,” IEEE Military Com-
munications Conference, 2008 (MILCOM 2008), pp. 1–6, Nov. 2008.

[66] P. Pawelczak, R. Venkatesha Prasad, L. Xia, and I. Niemegeers, “Cog-

nitive radio emergency networks - requirements and design,” The 1st
IEEE International Symposium on New Frontiers in Dynamic Spec-
trum Access Networks, 2005 (DySPAN 2005), pp. 601–606, Nov. 2005.

[67] V. Brik, E. Rozner, S. Banerjee, and P. Bahl, “Dsap: a protocol for co-
ordinated spectrum access,” in The 1st IEEE International Symposium
on New Frontiers in Dynamic Spectrum Access Networks (DySPAN
2005), Nov. 2005, pp. 611–614.

[68] A. Chia-Chun Hsu, D. Weit, and C.-C. Kuo, “A Cognitive MAC Pro-
tocol Using Statistical Channel Allocation for Wireless Ad-Hoc Net-
works,” in IEEE Wireless Communications and Networking Conference
(WCNC 2007), March 2007, pp. 105–110.

[69] J. Jia, Q. Zhang, and X. Shen, “HC-MAC: A Hardware-Constrained

Cognitive MAC for Efficient Spectrum Management,” IEEE Journal
on Selected Areas in Communications, vol. 26, no. 1, pp. 106–117, Jan.
2008.

[70] Q. Zhao, L. Tong, A. Swami, and Y. Chen, “Decentralized cogni-

tive MAC for opportunistic spectrum access in ad hoc networks: A
POMDP framework,” IEEE Journal on Selected Areas in Communica-
tions, vol. 25, no. 3, pp. 589–600, April 2007.

[71] B. Hamdaoui and K. Shin, “OS-MAC: An Efficient MAC Protocol

for Spectrum-Agile Wireless Networks,” IEEE Transactions on Mobile
Computing, vol. 7, no. 8, pp. 915–930, Aug. 2008.

[72] K. Bian and J.-M. Park, “MAC-Layer Misbehaviors in Multi-Hop Cog-

nitive Radio Networks,” 2006 US - Korea Conference on Science, Tech-
nology, and Entrepreneurship (UKC2006), Aug. 2006.

143
 BIBLIOGRAPHY

[73] C. Cordeiro and K. Challapali, “C-MAC: A Cognitive MAC Protocol

for Multi-Channel Wireless Networks,” in The 2nd IEEE International
Symposium on New Frontiers in Dynamic Spectrum Access Networks
2007 (DySPAN 2007), April 2007, pp. 147–157.

[74] Y. Xing, C. N. Mathur, M. Haleem, R. Chandramouli, and K. Subbal-

akshmi, “Dynamic spectrum access with qos and interference temper-
ature constraints,” IEEE Transactions on Mobile Computing, vol. 6,
no. 4, pp. 423–433, April 2007.

[75] L. Zhang, Y.-C. Liang, and Y. Xin, “Joint admission control and power
allocation for cognitive radio networks,” IEEE International Conference
on Acoustics, Speech and Signal Processing (ICASSP 2007), vol. 3, pp.
III–673–III–676, 15-20 April 2007.

[76] J. Xiang, Y. Zhang, and T. Skeie, “Joint admission and power control
for cognitive radio cellular networks,” The 11th IEEE International
Conference on Communication Systems (ICCS 2008), pp. 1519–1523,
Nov. 2008.

[77] L. Zhang, Y.-C. Liang, and Y. Xin, “Joint beamforming and power
allocation for multiple access channels in cognitive radio networks,”
IEEE Journal on Selected Areas in Communications, vol. 26, no. 1,
pp. 38–51, Jan. 2008.

[78] J. Xiang, Y. Zhang, T. Skeie, and J. He, “Qos aware admission and
power control for cognitive radio cellular networks,” Wiley Wireless
Communications and Mobile Computing, vol. 9, no. 11, pp. 1520–1531,
2009.

[79] D. Kim, L. Le, and E. Hossain, “Joint rate and power allocation for cog-
nitive radios in dynamic spectrum access environment,” IEEE Trans-
actions on Wireless Communications, vol. 7, no. 12, pp. 5517–5527,
December 2008.

[80] Y. Zhu, Z. Sun, W. Wang, T. Peng, and W. Wang, “Joint power and
rate control considering fairness for cognitive radio network,” in IEEE
Wireless Communications and Networking Conference (WCNC 2009).,
April 2009, pp. 1–6.

[81] L. B. Le and E. Hossain, “Resource allocation for spectrum underlay

in cognitive radio networks,” IEEE Transactions on Wireless Commu-
nications, vol. 7, no. 12, pp. 5306–5315, December 2008.

144
BIBLIOGRAPHY

[82] P. Cheng, Z. Zhang, H.-H. Chen, and P. Qiu, “Optimal distributed joint
frequency, rate and power allocation in cognitive ofdma systems,” IET
Communications, vol. 2, no. 6, pp. 815–826, July 2008.

[83] W. Wang, W. Wang, Q. Lu, and T. Peng, “An uplink resource alloca-
tion scheme for ofdma-based cognitive radio networks,” International
Journal of Communication Systems, vol. 22, no. 5, pp. 603–623, 2009.

[84] H.-Y. Gu, C.-Y. Yang, and B. Fong, “Low-complexity centralized joint
power and admission control in cognitive radio networks,” IEEE Com-
munication Letters, vol. 13, no. 6, pp. 420–422, 2009.

[85] D. Lopez-Perez, A. Valcarce, G. De La Roche, E. Liu, and J. Zhang,

“Access methods to wimax femtocells: A downlink system-level case
study,” in 11th IEEE Singapore International Conference on Commu-
nication Systems (ICCS 2008), Nov. 2008, pp. 1657–1662.

[86] A. Valcarce, G. D. L. Roche, álpar Jüttner, D. López-Pérez, and

J. Zhang, “Applying fdtd to the coverage prediction of wimax femto-
cells,” EURASIP Journal on Wireless Communications and Network-
ing, pp. 555–576, 2009.

[87] D. Lopez-Perez, G. de la Roche, A. Valcarce, A. Juttner, and J. Zhang,

“Interference avoidance and dynamic frequency planning for wimax
femtocells networks,” in 11th IEEE Singapore International Conference
on Communication Systems (ICCS 2008), Nov. 2008, pp. 1579–1584.

[88] K. Sundaresan and S. Rangarajan, “Efficient resource management in

ofdma femto cells,” in MobiHoc ’09: Proceedings of the tenth ACM
international symposium on Mobile ad hoc networking and computing.
New York, NY, USA: ACM, 2009, pp. 33–42.

[89] P. Dutta, S. Jaiswal, D. Panigrahi, and R. Rastogi, “A new channel

assignment mechanism for rural wireless mesh networks,” in IEEE IN-
FOCOM 2008, April 2008, pp. 2261–2269.

[90] D. Niyato and E. Hossain, “Cognitive radio for next-generation wireless

networks: an approach to opportunistic channel selection in ieee 802.11-
based wireless mesh,” IEEE Wireless Communications, vol. 16, no. 1,
pp. 46–54, February 2009.

[91] H. Li, Y. Cheng, C. Zhou, and W. Zhuang, “Minimizing end-to-end

delay: A novel routing metric for multi-radio wireless mesh networks,”
in IEEE INFOCOM 2009, April 2009, pp. 46–54.

145
 BIBLIOGRAPHY

[92] H. N. Pham, J. Xiang, Y. Zhang, and T. Skeie, “Qos-aware chan-

nel selection in cognitive radio networks: A game-theoretic approach,”
in IEEE Global Telecommunications Conference (GLOBECOM 2008),
Nov. 30 - Dec. 4 2008, pp. 1–7.

[93] M. H. Rehmani, A. C. Viana, H. Khalife, and S. Fdida, “Adaptive

and occupancy-based channel selection for unreliable cognitive radio
networks,” in Rencontres Francophones sur les Aspects Algorithmiques
des Telecommunications (ALGOTEL) 2009, Juni 2009.

[94] Y. Hou, Y. Shi, and H. Sherali, “Optimal spectrum sharing for multi-
hop software defined radio networks,” in IEEE INFOCOM 2007, May
2007, pp. 1–9.

[95] Y. Song, Y. Fang, and Y. Zhang, “Stochastic channel selection in cogni-

tive radio networks,” in Global Telecommunications Conference, 2007.
GLOBECOM ’07. IEEE, Nov. 2007, pp. 4878–4882.

[96] P. N. Anggraeni, N. H. Mahmood, J. Berthod, N. Chaussonniere,

L. My, and H. Yomo, “Dynamic channel selection for cognitive ra-
dios with heterogenous primary bands,” Wireless Personal Commu-
nications, vol. 45, no. 3, pp. 369–384, 2008.

[97] J. Matousek and B. Gärtner, Understanding and Using Linear Pro-

gramming. Springer, November 2006.

[98] (2010) ILOG CPLEX. [Online]. Available: http://www-01.ibm.com/

software/integration/optimization/cplex-optimizer/

[99] H. Kellerer, U. Pferschy, and D. Pisinger, Knapsack Problems.

Springer, Berlin, Germany, 2004.

[100] W. Shih, “A branch and bound method for the multiconstraint zero-
one knapsack problem,” Journal of the Operational Research Society,
vol. 30, no. 4, pp. 369–378, Apr., 1979.

[101] P. C. Gilmore and R. E. Gomory, “The theory and computation of

knapsack functions,” Operations Research, vol. 14, no. 6, pp. 1045–
1074, Nov. - Dec., 1966.

[102] H. M. Weingartner and D. N. Ness, “Methods for the solution of the

multidimensional 0/1 knapsack problem,” Operations Research, vol. 15,
no. 1, pp. 83–103, Jan. - Feb., 1967.

146
BIBLIOGRAPHY

[103] P. Setoodeh and S. Haykin, “Robust transmit power control for cogni-
tive radio,” Proceedings of the IEEE, vol. 97, no. 5, pp. 915–939, May
2009.
[104] C. Stevenson, G. Chouinard, Z. Lei, W. Hu, S. Shellhammer, and
W. Caldwell, “IEEE 802.22: The first cognitive radio wireless regional
area network standard,” IEEE Communications Magazine, vol. 47,
no. 1, pp. 130–138, January 2009.
[105] T. Yucek and H. Arslan, “A survey of spectrum sensing algorithms for
cognitive radio applications,” IEEE Communications Surveys Tutori-
als, vol. 11, no. 1, pp. 116 –130, first 2009.
[106] “Guidelines for evaluation of radio transmission technologies for imt-
2000,” RECOMMENDATION ITU-R M.1225, 1997.
[107] “Digital mobile radio towards future generation systems cost 231 final
report,” COST Action 231, 1999.
[108] D. Palomar and M. Chiang, “A tutorial on decomposition methods
for network utility maximization,” Selected Areas in Communications,
IEEE Journal on, vol. 24, no. 8, pp. 1439 –1451, aug. 2006.
[109] ——, “Alternative distributed algorithms for network utility maximiza-
tion: Framework and applications,” IEEE Transactions on Automatic
Control, vol. 52, no. 12, pp. 2254 –2269, dec. 2007.
[110] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge Uni-
versity Press, March 2004.
[111] J. Xiang, Y. Zhang, and T. Skeie, “Dynamic spectrum sharing in cogni-
tive radio femtocell networks,” Lecture Notes of the Institute for Com-
puter Sciences, Social Informatics and Telecommunications Engineer-
ing, vol. 52, no. 12, pp. 164 – 178, dec. 2010.
[112] H. S. Wang and N. Moayeri, “Finite-state markov channel-a useful
model for radio communication channels,” Vehicular Technology, IEEE
Transactions on, vol. 44, no. 1, pp. 163–171, Feb 1995.
[113] “IEEE Standard for Information technology-Telecommunications and
information exchange between systems-Local and metropolitan area
networks-Specific requirements - Part 11: Wireless LAN Medium Ac-
cess Control (MAC) and Physical Layer (PHY) Specifications,” IEEE
Std 802.11-2007 (Revision of IEEE Std 802.11-1999), pp. C1–1184, 12
2007.

147
 BIBLIOGRAPHY

[114] D. P. Bertsekas, Network Optimization: Continuous and Discrete Mod-

els. Athena Scientific, May 2008.

[115] ——, Nonlinear Programming. Athena Scientific, September 1999.

[116] B. Wild, “Detecting primary receivers for cognitive radio applications,”

in in Proc. of the First IEEE Symposium on New Frontiers in Dynamic
Spectrum Access Networks, 2005, pp. 124–130.

[117] “GNU Radio.” [Online]. Available: http://www.gnuradio.org/

[118] “USRP by Ettus Research LLC.” [Online]. Available: http:

//www.ettus.com/

[119] “Rice University WARP Project.” [Online]. Available: http://warp.

rice.edu

[120] “Agile SDR Solutions.” [Online]. Available: http://www.

agile-sdr-solutions.com/

148
Appendix A

Publication list

Most of the material in this work has previously been published in the pro-
ceedings of various international conferences and journals. Here we provide
a list of publications during this thesis work.

Book Chapters
1. Jie Xiang and Yan Zhang, “Medium Access Control Protocols for Cog-
nitive Radio Networks,” in Cognitive Radio Networks: Architectures,
Protocols and Standards, edited by Y. Zhang, J. Zheng, H.H. Chen,
CRC Press, USA, 2010.

2. Jianhua He, Jie Xiang, Zuoyin Tang and Yan Zhang, “Dynamic and
Fair Spectrum Access for Autonomous Communications,” in Autonomic
Computing and Networking, edited by M. Denko, Laurence T. Yang and
Y. Zhang, Springer, USA, 2009.

3. Jianhua He, Xiaoming Fu, Jie Xiang, Yan Zhang and Zuoyin Tang,
“Routing and Scheduling for WiMAX Mesh Networks,” in WiMAX
Network Planning and Optimization, edited by Y. Zhang, CRC Press,
USA, 2009.

Journal Articles
1. Jie Xiang, Yan Zhang, Tor Skeie, and Lang Xie. “Downlink Spectrum
Sharing for Cognitive Radio Femtocell Networks,” IEEE Systems Jour-
nal, special issue on Broadband Access Networks, 4(4):524-534, 2010.

2. Jie Xiang, Yan Zhang and Tor Skeie. “Medium Access Control Proto-
cols in Cognitive Radio Networks,” Wireless Communications and Mo-

149
 Publication list

bile Computing, Wiley, special issue on Recent Advances in Wireless

Communications and Networks, 10(1):31-49, 2010.

3. Jie Xiang, Yan Zhang, Tor Skeie, and Jianhua He. “QoS Aware Ad-
mission and Power Control for Cognitive Radio Cellular Networks,”
Wireless Communications and Mobile Computing, Wiley, special issue
on Cognitive Radio and Advanced Spectrum Management, 9(11):1520-
1531, 2009.

4. Supeng Leng, Yan Zhang, Hsiao-Hwa Chen, Jie Xiang, and Mohsen
Guizani. “Power-fixed and Power-aware MAC Protocols for Multi-hop
Wireless Networks with Large Interference Area ,” IEEE Transactions
on Vehicular Technology, 58(6):2966-2976, July 2009.

Conference Proceedings
1. Jie Xiang, Yan Zhang, and Tor Skeie, “Admission and Power Control
for Cognitive Radio Cellular Networks: A Multidimensional Knapsack
Solution,” In Proc. of CogART 2010.

2. Lang Xie, Poul E. Heegaard, Yan Zhang, and Jie Xiang. “Reliable
Channel Selection and Routing for Real-time Services over Cognitive
Radio Mesh Networks,” In Proc. of Qshine 2010.

3. Sabita Maharjan, Jie Xiang, Yan Zhang and Stein Gjessing, “Delay Re-
duction for Real Time Services in IEEE 802.22 Wireless Regional Area
Network,” In Proc. of the 21st Annual IEEE International Symposium
on Personal, Indoor and Mobile Radio Communications (PIMRC), Septem-
ber 26 - 29, 2010, Istanbul Turkey.

4. Lang Xie, Jin Zhang and Jie Xiang, “Cognitive Networking for Wireless
Mesh Networks using Swarm Intelligence,” In Proc. of the 6th IEEE In-
ternational Conference on Wireless Communications, Networking and
Mobile Computing (WiCOM 2010), September 23 - 25, 2010, Chengdu,
China.

5. Lang Xie, Poul E. Heegaard, Jin Zhang, Jie Xiang, “System State aware
Admission Control scheme for IEEE 802.16 WiMAX-based healthcare
system,” In Proc. of the 6th International Wireless Communications
and Mobile Computing Conference (IWCMC 2010), June 28 - July 2,
2010, Caen, France.

150
 6. Jie Xiang, Yan Zhang and Tor Skeie. “Dynamic Spectrum Sharing in
Cognitive Radio Femtocell Networks,” In Proc. of the 4th International
Conference on Access Networks (ACCESSNETS 2009), Nov. 1 - 3,
2009, Hongkong, China. (Invited Paper)

7. Qin Xin and Jie Xiang. “Joint QoS-aware Admission Control, Channel
Assignment, and Power Allocation for Cognitive Radio Cellular Net-
works,” In Proc. of the 6th IEEE International Conference on Mobile
Ad-hoc and Sensor Systems (MASS 2009), pp. 294-303, Oct. 12 - 15,
2009, Macau, China.

8. Lang Xie, Jie Xiang, Yan Zhang and Jin Zhang. “Joint Bandwidth
Reservation and Admission Control in IEEE 802.16e based Networks,”
In Proc. of the 2009 IEEE International Conference on Communica-
tions (ICC 2009), pp. 1-6, June 14 - 18, 2009, Dresden, Germany.

9. Qin Xin, Yan Zhang and Jie Xiang. “Minimum-Latency Gossiping

in Multi-hop Wireless Mesh Networks,” In Proc. of the 2009 IEEE
International Conference on Communications (ICC 2009), pp.1-5, June
14 - 18, 2009, Dresden, Germany.

10. Yan Zhang, Jie Xiang, Qin Xin and Geir Egil Øien. “Optimal Sens-
ing Cooperation for Spectrum Sharing in Cognitive Radio Networks,”
Accepted by European Wireless 2009 (EW 2009), May 17 - 20, 2009,
Aalborg, Denmark.

11. Hai Ngoc Pham, Jie Xiang, Yan Zhang and Tor Skeie. “QoS-Aware
Channel Selection in Cognitive Radio Networks: A Game-Theoretic
Approach,” In Proc. of the 2008 IEEE Global Communications Con-
ference (GLOBECOM 2008), Nov. 30 - Dec. 3, 2008, New Orleans,
LA, USA.

12. Jie Xiang,Yan Zhang and Tor Skeie. “Joint Admission and Power Con-
trol for Cognitive Radio Cellular Networks,” In Proc. of the 11th IEEE
International Conference on Communication Systems 2008 (ICCS 2008),
pp.1519-1523, Nov. 19 - 21, 2008, Guangzhou, China. (Invited Paper)

13. Lang Xie and Jie Xiang. “A Novel Bandwidth Degradation Scheme
for Admission Control in IEEE 802.16e Networks,” In Proc. of the
4th IEEE International Conference on Wireless Communications, Net-
working and Mobile Computing (WiCOM 2008), pp.1-4, Oct. 12-14,
2008, Dalian, China.

151
 Publication list

14. Lang Xie, Jie Xiang and Yan Zhang. “Revenue-based Admission Con-
trol for Cognitive Radio Cellular Systems”. Presented in the 2008
International Workshop on Cognitive Networks and Communications
(COGCOM 2008), In Proc. of the 2008 International Conference on
Communications and Networking in China (ChinaCom 2008), pp.1200-
1204, Aug. 25-27, 2008, Hangzhou, China.

15. Qin Xin, Yan Zhang and Jie Xiang. “Optimal Spectrum Scheduling in
Cognitive Wireless Mesh Networks”, In Proc. of the 2008 International
Wireless Communications and Mobile Computing Conference (IWCMC
2008), pp.724-728, Aug. 6-8, 2008, Crete Island, Greece.

16. Yan Zhang and Jie Xiang. “A New Adaptive Energy Management
Scheme in IEEE 802.16e Mobile WiMAX,” In Proc. of the 2007 Nor-
wegian Informatics Conference (NIK 2007), pp. 111-114, Nov. 19-21,
2007, Oslo, Norway.

152
Appendix B

Acronym List

AMC Adaptive Modulation Coding. 8, 112

AWGN Additive White Gaussian Noise. 92

BS Base Station. 21, 22, 27, 28, 30–32, 36, 39, 57,
59

CCC Common Control Channel. 20, 21

CDMA Code Division Multiple Access. 23, 24, 62, 88
CogCell Cognitive Radio Cellular Networks. 4, 5, 9,
10, 22, 23, 27, 28, 30, 31, 37, 55, 60–62, 65,
131–133
CogFem Cognitive Radio Femtocell Networks. 5, 6, 9,
11, 21, 22, 87, 88, 91, 93, 94, 97, 99, 101, 103,
106, 107, 131–133
CogMesh Cognitive Radio Mesh Networks. 7–9, 12, 22,
25, 109, 110, 112, 113, 115, 125, 128, 131–133
CR Cognitive Radio. 3–5, 7, 11, 16, 18, 20, 23–25,
107
CSI Channel State Information. 99

DoS Denial-of-Service. 20
DSL Digital Subscriber Line. 5
DTR Data Transmission Rate. 29–31, 36, 55, 60, 76

ECC Electronic Communications Committee. 1

153
 Appendix B Acronym List

FBS Femtocell Base Station. 5, 88–100, 102, 103,

106, 107
FCC Federal Communications Commission. 1, 2
FSMC Finite-State Markov Channel. 112
FU Femtocell User. 5, 11, 100

ISM Industrial, Scientific and Medical. 1, 2

JAPC-MKP Joint Admission and Power Control with Mul-

tidimensional Knapsack Problem modeling.
27, 53, 55, 57–61, 75, 131, 134
JAPC-MRER Joint Admission and Power Control scheme
using a Minimal Revenue Efficiency Removal
algorithm. 10, 27, 32, 36–40, 42, 55, 57–60,
131, 134
JAPC-MSRA Joint Admission and Power Control scheme
using a Minimal SINR Removal Algorithm.
10, 27, 32, 34, 36–40, 42, 55, 57–60
JAPC-Rand Joint Admission and Power Control scheme
using a Random removal algorithm. 10, 27,
32, 35–40, 60

MCKP Multiple-Choice Knapsack Problem. 12, 132,

133
MINLP Mixed Integer Non-linear Programming. 65,
94, 107, 132, 133
MKP Multidimentional Knapsack Problem. 10, 61,
69, 70, 84, 131, 133, 134
MMKP Multidimentional Multiple Knapsack Prob-
lem. 11, 61, 67, 69, 84, 132, 133
MRS Maximization of Revenue from SUs. 64

NPT The Norwegian Post and Telecommunications

Authority. 1

OFDM Orthogonal Frequency-Division Multiplexing.

18
OFDMA Orthogonal Frequency-Division Multiple Ac-
cess. 24, 62, 88

154
Appendix B Acronym List

POMDP Partially observable Markov Decision Pro-

cesse. 17
PR Primary Receiver. 4, 5, 10, 28–30
PT Primary Transmitter. 28, 29
PU Primary User. 3–5, 7, 8, 16–25, 27–29, 31–33,
36–40, 42, 55, 57, 58, 60–65, 75, 76, 111, 115,
117

QoS Quality of Service. 5, 10, 17, 21, 22, 27–31,

60, 109, 110, 131

SIMO-MAC Single Input Multiple Output Multiple Access

Channel. 23, 24
SINR Signal to Interference and Noise Ratio. 10, 12,
19, 23, 30–32, 34, 36, 38, 40, 42, 55, 56, 59–61,
75, 92, 109, 112, 124, 133
SMG Secondary Mesh Gateway. 7, 110, 111, 115
SMR Secondary Mesh Router. 7, 109–115, 128, 132
SMU Secondary Mesh User. 7, 8, 110
SU Secondary User. 3–5, 8, 10, 11, 16–25, 27–40,
45, 46, 53, 55–65, 75, 76, 78, 85

WLAN Wireless Local Area Network. 1

WPAN Wireless Personal Area Network. 1
WRAN Wireless Regional Area Network. 2

155