sensing-assisted spectrum access strategy and optimization ... kumar_mondal_thesis.pdf ·...
TRANSCRIPT
Sensing-Assisted Spectrum AccessStrategy and Optimization inCognitive Radio Networks
by
Ratan Kumar Mondal
M.S. (Electronics); B. Sc. Eng. (EEE)
A thesis submied in fullment of the requirement for the degree of
Doctor of Philosophy
School of Electrical Engineering and Computer Science
Science and Engineering Facultyeensland University of Technology
2018
© Copyright 2018
Ratan Kumar Mondal
Keywords
Cognitive radio, cognitive radio network, contention access, cross-layer, medium access
control, MAC protocol, multiple access, physical layer, random access protocol, spec-
trum sensing, spectrum sensing optimization, spectrum access, sensing-assisted access,
throughput analysis.
Abstract
e rapid growth of demand for wireless broadband data has emerged as a challenging
task to be accommodated with xed spectrum access policy in the future deployment
of the h generation (5G) mobile communication technology. e increasing demand
of augmenting the bandwidth has actuated the evolutionary technology to use the un-
derutilized spectrum for ecient spectral utilization and best eort service. Despite
advancement in access technology, data rate growth continues its exponential trend
in an already crowded spectrum. Cognitive Radio (CR) technology has been proposed
as a promising solution for future generation networks and aims to improve spectral
eciency by opening underutilized portions of the spectrum to secondary users (SU)
while controlling interference to the primary users (PU).
Spectrum sensing and spectrum access are two key components of CR operations.
Spectrum sensing facilitates the detection of the primary signal in physical (PHY) layer
to nd the spectrum opportunity. Spectrum access enables ecient data transmission
through medium access control (MAC) while multiple secondary users share the trans-
mission medium to improve the throughput performance. Existing studies have focused
on improving the SU’s throughput performance; therefore, conventional cognitive MAC
(C-MAC) protocols cannot provide sucient protection to PU due to the exclusion of
spectrum sensing.
In cognitive radio networks (CRN), SUs strive to utilize the full potential of the spec-
trum opportunity while protection to the legacy users from interference caused by sec-
ondary users must be guaranteed. is conicting but interrelated issue is investigated
over two stages for the purpose of solving with a cross-layer model. e concept of
the cross-layer model is an emerging design trend that dramatically improves on the
performance gains of the single layer approach. In the research conducted for this
vi
thesis, the spectrum sensing of the PHY is integrated with the access strategy of the
MAC layer by a cross-layer approach to improving both the throughput performance
and interference protection.
e rst investigation illustrates the impact of spectrum sensing on the maximization
of the spectrum opportunity. In the current literature, multi-stage spectrum sensing has
gained the reputation for providing signicant protection to primary users. However,
this sensing requires a long sensing period which reduces throughput performance.
Motivated by this fact, a dual-level sensing (DS) based access mechanism is proposed
with a short sensing period to explore higher transmission opportunities and then utilize
the sensing outcome to reduce the collision during the multiple access phase.
e DS mechanism requires optimization of detection sensitivity and the sensing
period such that throughput is maximized under the constraint of PU protection. ere-
fore, a method of solving the sensing-throughput trade-o is developed for the DS-based
access mechanism. rough mathematical derivations, it has proved that by allowing a
portion of the sensing period to be devoted to reducing the probability of false alarm, the
constraint is still met while transmission opportunity is improved. Furthermore, the nu-
merical analysis reveals that proposed solution algorithms can maximize the achievable
secondary throughput signicantly within a limited computational complexity.
e second investigation provides the way to reduce collision during multiple access.
A multiple access protocol is proposed that is associated with DS mechanism. e
proposed DS-based multiple access (DSMA) is formulated analytically using a Markov
chainmodel to obtain the service time and throughput performance. e sensingmethod
in the DSMA is designed with a sensitivity that can only expose maximum opportunity.
As a result, the collision rate is increased during channel access. However, when condi-
tional sensing is used in the contention process during channel access, the eect on the
collision rate is minimized. e target detection performance during contention access
is segmented by the cross-layer formulation of the sensing and contention parameters.
A novel sensing-assisted access (SAA) protocol is nally proposed as a complete
random access mechanism for CRN. e contention-based access is designed based on
vii
the integration of the backo process and spectrum sensing. e sensing-embedded
backo process is modeled by applying the Markov chain analysis in the presence of
sensing error. Exploitation of all sensing aspects during the backo process reveals the
spectrum opportunity extensively, and the consequent possibility of collision is reduced
through the sensing-assisted contention process. Performance analysis and numerical
results consolidate that the sensing-assisted access protocol improves the throughput
signicantlywithin a short access delaywhile ensuring sucient interference protection
to the legacy system.
Contents
Abstract v
List of Figures xiii
List of Tables xvii
Variables and Notations xix
List of Abbreviations xxiii
Statement of Original Authorship xxv
Acknowledgments xxvii
Chapter 1 Introduction 11.1 Scarcity versus Underutilization in Radio Spectrum . . . . . . . . . . 1
1.2 A New Paradigm of Radio Spectrum: Dynamic Spectrum Access . . . 4
1.3 Spectrum Access in Cognitive Radio Networks . . . . . . . . . . . . . 6
1.4 Research Motivation: Sensing-Assisted Access . . . . . . . . . . . . . 8
1.5 Research Goal and Approaches . . . . . . . . . . . . . . . . . . . . . . 10
1.6 Overview of esis Structure . . . . . . . . . . . . . . . . . . . . . . . 12
1.7 List of Publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Chapter 2 Background and Literature Review 152.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Components of Cognitive Radio . . . . . . . . . . . . . . . . . . . . . 16
2.3 Standardization and Implementation of CR . . . . . . . . . . . . . . . 20
2.4 Current Trends and Applications of CR . . . . . . . . . . . . . . . . . 22
2.4.1 CR-based Wireless Sensor Networks . . . . . . . . . . . . . . 22
2.4.2 Cognitive Radio in Cellular Networks . . . . . . . . . . . . . 23
2.5 Spectrum Access rough MAC Protocol . . . . . . . . . . . . . . . . 25
2.6 Cross-Layer Components for Sensing-Assisted Access Protocol . . . . 26
2.6.1 Spectrum Sensing Algorithm . . . . . . . . . . . . . . . . . . 26
2.6.2 Spectrum Occupancy Modeling . . . . . . . . . . . . . . . . . 28
2.6.3 Data Transmission Mechanism . . . . . . . . . . . . . . . . . 30
2.7 Sensing-Transmission Optimization . . . . . . . . . . . . . . . . . . . 32
2.8 Model of Access Protocols Based on Cross-layer Design . . . . . . . . 34
2.9 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
x CONTENTS
Chapter 3 Impact of Spectrum Sensing on the Capacity Measurementof Spectrum Opportunity 393.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2.1 Spectrum Sensing Model . . . . . . . . . . . . . . . . . . . . . 41
3.2.2 PU Activity Model . . . . . . . . . . . . . . . . . . . . . . . . 43
3.2.3 Spectrum Access Decision . . . . . . . . . . . . . . . . . . . . 44
3.3 Conventional Single-level Sensing Mechanism . . . . . . . . . . . . . 45
3.4 Proposed Dual-level Sensing Mechanism . . . . . . . . . . . . . . . . 46
3.5 Performance Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.5.1 Receiver Operating Characteristic . . . . . . . . . . . . . . . . 49
3.5.2 Access Probability . . . . . . . . . . . . . . . . . . . . . . . . 52
3.6 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
Chapter 4 Optimization of Dual-level Sensing for Ecient Utilizationin Cognitive Radio Networks 574.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.1.2 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.2.1 roughput of Dual-Level Sensing Based Access Protocol . . 61
4.2.2 Problem and Strategy Formulation . . . . . . . . . . . . . . . 62
4.3 Minimization of Overall PFA . . . . . . . . . . . . . . . . . . . . . . . 64
4.3.1 Feasibility Analysis of PFA Minimization . . . . . . . . . . . . 64
4.3.2 Discussion on Feasibility Analysis . . . . . . . . . . . . . . . 68
4.3.3 Algorithm of PFA Minimization . . . . . . . . . . . . . . . . . 69
4.4 roughput Maximization . . . . . . . . . . . . . . . . . . . . . . . . 72
4.4.1 Joint Optimization with Numerical Analysis . . . . . . . . . . 73
4.5 Numerical Results and Discussion . . . . . . . . . . . . . . . . . . . . 75
4.5.1 roughput Optimization and Model Validation . . . . . . . . 76
4.5.2 Performance Evaluation of DLS Based Access with
Post-optimization . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.6 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
Chapter 5 Sensing AssistedMultiple Access Strategy in Cognitive RadioNetworks 855.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.1.2 Contribution . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.2.1 Network Entity . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.2.2 Energy Detection Based Spectrum Sensing . . . . . . . . . . . 89
5.3 Proposed Model of the DSMA Protocol . . . . . . . . . . . . . . . . . 90
5.3.1 Underlying Mechanisms of Proposed Protocol . . . . . . . . . 90
5.3.2 Proposed Protocol . . . . . . . . . . . . . . . . . . . . . . . . 92
5.4 Analytical Modeling of Proposed DSMA Mechanism . . . . . . . . . . 93
5.4.1 Operational Time in Spectrum Discovery . . . . . . . . . . . 93
CONTENTS xi
5.4.2 Time Sequence Adaptation Based on Backo Process and
Detection Mechanism . . . . . . . . . . . . . . . . . . . . . . 94
5.4.3 Cross-layer Formulation of Backo and Detection Process in
CCA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.4.4 Packet Transmission Service . . . . . . . . . . . . . . . . . . . 96
5.5 Performance Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.5.1 Transmission Probability . . . . . . . . . . . . . . . . . . . . . 99
5.5.2 Packet Service Process in Multiple Access . . . . . . . . . . . 102
5.5.3 Average Packet Service Time . . . . . . . . . . . . . . . . . . 104
5.5.4 Normalized roughput . . . . . . . . . . . . . . . . . . . . . 105
5.6 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.6.1 Model Validation . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.6.2 roughput Performance Analysis . . . . . . . . . . . . . . . 107
5.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
Chapter 6 Sensing-AssistedAccess Protocol with Imperfect Sensing andPerformance Analysis for Multiple Access 1136.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
6.1.1 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
6.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
6.2.1 Network Entity . . . . . . . . . . . . . . . . . . . . . . . . . . 116
6.2.2 Channel Modeling with Imperfect Sensing . . . . . . . . . . . 117
6.3 Proposed Sensing-Assisted Access Protocol . . . . . . . . . . . . . . . 118
6.3.1 PHY/MAC Cross-layer Based Contention Mechanism . . . . . 118
6.3.2 Packet Transmission Structure of Proposed SAA Protocol . . 120
6.3.3 Analytical Modeling with Markov Chain Analysis . . . . . . 121
6.3.4 Cross-layer Relationship Between Backo Mechanism and
Physical Channel Sensing . . . . . . . . . . . . . . . . . . . . 126
6.4 Performance Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
6.4.1 Packet Service Process and Normalized roughput in
Multiple Access Operation . . . . . . . . . . . . . . . . . . . . 128
6.4.2 Average Access Delay . . . . . . . . . . . . . . . . . . . . . . 130
6.5 Numerical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
6.5.1 roughput and Delay Performance of Proposed SAA Protocol 132
6.5.2 Model Validation and Performance Comparison . . . . . . . . 138
6.6 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
Chapter 7 Conclusion and Recommendations for Future Research 1437.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
7.2 Recommendations for Future Research . . . . . . . . . . . . . . . . . 147
Appendix A Proof of Propositions andeorems 149A.1 Proof of Proposition 4.2 . . . . . . . . . . . . . . . . . . . . . . . . . . 149
A.2 Proof of Proposition 4.4 . . . . . . . . . . . . . . . . . . . . . . . . . . 150
A.3 Proof of Proposition 4.5 . . . . . . . . . . . . . . . . . . . . . . . . . . 151
A.4 Proof of eorem 4.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
A.5 Proof of eorem 4.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
xii CONTENTS
A.6 Proof of eorem 4.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
A.7 Proof of Proposition 4.6 . . . . . . . . . . . . . . . . . . . . . . . . . . 153
A.8 Proof of Proposition 4.7 . . . . . . . . . . . . . . . . . . . . . . . . . . 154
Bibliography 157
List of Figures
2.1 Components of the cognitive cycle [1]. . . . . . . . . . . . . . . . . . 16
2.2 Review of dierent detection techniques based on complexity versus
accuracy [2]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3 Network architecture of the IEEE 802.22 WRAN, where users of TV
bands and wireless microphones are the primary users, and BS and
CPE are the secondary users [3]. . . . . . . . . . . . . . . . . . . . . . 21
2.4 Network model of a proposed CR-WSN system [4]. . . . . . . . . . . 23
2.5 Network model of a proposed CR-LTE system [5]. . . . . . . . . . . . 24
2.6 Review of frame format with sensing-transmission mechanism. . . . 34
3.1 Frame structure for CR operation with spectrum sensing and access
in every frame. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2 Frame structure of conventional single-level sensing mechanism. . . 45
3.3 Frame structure of proposed dual-level sensing mechanism. . . . . . 46
3.4 ROC comparison of the SS and DS mechanism with theoretical and
simulation results at a given SNR value. . . . . . . . . . . . . . . . . . 50
3.5 Probability of false alarm vs. sensing time of DS and SS strategy;
PfDS is less than PfSS for a given Pd = 0.9. . . . . . . . . . . . . . . . 51
3.6 PfDS vs. Pd1 ; PfDS has a minimum value for an optimum value of Pd1 . 53
3.7 Access probability vs. sensing time (sec) for Pd = 0.99, 0.9. For a
given value of Pd, the PaDS is higher than the PaSS . . . . . . . . . . . 53
3.8 Pa vs. PH0 of the DS and SS mechanism for Pd = 0.99, 0.9; For a
given value of Pd, the PaDS is higher than the PaSS . . . . . . . . . . . 54
4.1 MAC frame format and time slot operation of the proposed
DLS-based access protocol. . . . . . . . . . . . . . . . . . . . . . . . . 60
xiv LIST OF FIGURES
4.2 Illustration of the PFA minimization problem with the characterizing
of Pf1 ,(1− Pf1)Pf2 , and Pf corresponding to Pd1 , where the
simulation parameters are, γ = 0 dB, Ns = 2, Pd = 0.95, σ2w = 1. . . 69
4.3 (a) Mesh plot and (b) contour plot of Pf (Pd1 and τs) where the
simulation parameters are, γ = −15 dB, Pi = 0.9, P aMAC = 0.99,
σ2w = 1, fs = 6 MHz, τ ∗s,s = 1.7 ms and Tf = 10 ms. . . . . . . . . . . 77
4.4 (a) Mesh plot and (b) contour plot of R(Pd1 , τs) where the simulation
parameters are, γ = −15 dB, Pi = 0.9, P aMAC = 0.99, σ2
w = 1, fs = 6
MHz, τ ∗s,s = 1.7 ms and Tf = 10 ms. . . . . . . . . . . . . . . . . . . . 78
4.5 Mesh plot of throughput corresponding to Pd1 and τds, where the
simulation parameters are, γ = −15 dB, Pi = 0.9, P aMAC = 0.99,
σ2w = 1, fs = 6 MHz, and Tf = 10 ms. . . . . . . . . . . . . . . . . . . 79
4.6 Characterization of the change of R corresponding to τds(τs) for
Pd = 0.9, 0.95, 0.99 and its optimal sensing period as given by
Table 4.1, where the simulation parameters are, γ = −15 dB,
Pi = 0.9, P aMAC = 0.99, σ2
w = 1, fs = 6 MHz, and Tf = 10 ms. . . . . 80
4.7 Characterization of the change of R corresponding to τds(τs) for
Pd = 0.9, 0.95, 0.99 and its optimal Pd1 , where the simulation
parameters are, γ = −15 dB, Pi = 0.9, P aMAC = 0.99, σ2
w = 1, fs = 6
MHz, and Tf = 10 ms. . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.8 roughput performance comparison of the proposed DLS and the
conventional SLS based access mechanism for Pd = 0.9, 0.95, 0.99,
where the simulation parameters are, γ = −15 dB, Pi = 0.9,
P aMAC = 0.99, σ2
w = 1, fs = 6 MHz, and Tf = 10 ms. . . . . . . . . . 82
4.9 roughput performance comparison of the proposed DLS and the
conventional SLS based access mechanism for γ = −10,−15,−20
dB, where the simulation parameters are, Pd = 0.95, Pi = 0.9,
P aMAC = 0.99, σ2
w = 1, fs = 6 MHz, and Tf = 10 ms. . . . . . . . . . 83
5.1 Network Architecture of a cognitive radio network with multiple
access functionality. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.2 MAC frame format for proposed DSMA mechanism. . . . . . . . . . . 88
LIST OF FIGURES xv
5.3 Block diagram of the Proposed DSMA Mechanism. . . . . . . . . . . 91
5.4 Time slot operation of proposing dual-level sensing based multiple
access protocol. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.5 Markov chain model as the state transition of packet service process. 100
5.6 Normalized throughput versus probability of transmission for
comparing the analytical, simulated, and approximated model of
DSMA scheme for N = 10. . . . . . . . . . . . . . . . . . . . . . . . . 107
5.7 Normalized throughput versus probability of transmission of DSMA
scheme for N = 5, 10, 20, and 50. . . . . . . . . . . . . . . . . . . . . 108
5.8 Comparison of normalized throughput versus sensing time of three
schemes for N = 10 and γ = −15dB. . . . . . . . . . . . . . . . . . . 108
5.9 Normalized throughput versus sensing time of proposed DSMA
scheme for N = 5, 10, 20, 50 and γ = −15dB. . . . . . . . . . . . . 109
5.10 Normalized throughput versus sensing time performance of proposed
DSMA scheme for γ = −10 dB, −15 dB ,−20 dB, and N = 10. . . . . 109
5.11 Comparison of normalized throughput versus SNR of the three
schemes for N = 10 and Ts = 1 ms. . . . . . . . . . . . . . . . . . . . 111
5.12 Normalized throughput versus SNR of proposed DSMA scheme for
N = 5, 10, 20, 50 and Ts = 1 ms. . . . . . . . . . . . . . . . . . . . 111
6.1 Network conguration of cognitive radio network for SA-MAC
protocol. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
6.2 Flowchart of the channel access mechanism. . . . . . . . . . . . . . . 119
6.3 A complete packet transmission service of proposed SAA protocol. . 120
6.4 Markov chain model as the proposed backo process of the proposed
SAA protocol. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
6.5 Characteristic of normalized throughput (S) corresponding to
probability of missed detection (Pm); where the parameters are:
γ = −15 dB, fs = 6 MHz, PH1 = 0.1, u = 5, and N = 20. . . . . . . . 133
6.6 Characteristic of average access delay (E[D]) corresponding to
probability of missed detection (Pm); where the parameters are:
γ = −15 dB, fs = 6 MHz, PH1 = 0.1, u = 5, and N = 20. . . . . . . . 134
xvi LIST OF FIGURES
6.7 Probability of collision (PC) with respect to probability of missed
detection (Pm) of SAA protocol; where the parameters are: where the
parameters are: γ = −15 dB, fs = 6 MHz, PH1 = 0.1, u = 5, and
N = 20. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
6.8 Probability of access (φ) with respect to probability of missed
detection (Pm) of SAA protocol; where the parameters are: γ = −15
dB, fs = 6 MHz, PH1 = 0.1, u = 5, and N = 20. . . . . . . . . . . . . 135
6.9 Normalized throughput (S) versus contention window (ω0) of SAA
protocol; where the parameters are: γ = −15 dB, fs = 6 MHz,
PH1 = 0.1, u = 5, and ω0 = 16. . . . . . . . . . . . . . . . . . . . . . 136
6.10 Variation of normalized throughput (S) corresponding to number of
SU (N ) in analytical and simulation cases; where the parameters are:
γ = −15 dB, fs = 6 MHz, PH1 = 0.1, u = 5, and Pm = 0.1. . . . . . . 137
6.11 Variation of average access delay (E[D]) corresponding to number of
SU (N ) in analytical and simulation cases; where the parameters are:
γ = −15 dB, fs = 6 MHz, PH1 = 0.1, u = 5, and Pm = 0.1. . . . . . . 137
6.12 Normalized throughput (S) versus probability of access (φ) with
approximation, simulation, and analytical results; where the
parameters are: γ = −15 dB, fs = 6 MHz, PH1 = 0.1, u = 5,
ω0 = 16, and Pm = 0.1. . . . . . . . . . . . . . . . . . . . . . . . . . . 140
6.13 Normalized throughput comparison among distributed-MAC [6],
CR-CSMA [7], and our proposed SAA protocol with respect to
number of SU. In this analysis, the using parameters are: γ = −15
dB, fs = 6 MHz, PH1 = 0.1, u = 5, and Pm = 0.1. . . . . . . . . . . . 140
List of Tables
1.1 Mean spectral occupancy for various allocations. . . . . . . . . . . . . 3
4.1 Numerical results about comparing proposed solution of
optimization problem. . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.1 System parameters used in the simulation. . . . . . . . . . . . . . . . 106
6.1 Parameters for Performance Analysis of SAA Protocol. . . . . . . . . 132
Variables and Notations
In approximate order of appearance,
Tf Operational frame length
τs Sensing period
y(m), s(m), w(m) Received signal, PU’s transmied signal, and noise signal
fs Sampling frequency
m,M Sampling index and total number of sampling
H0,H1 Null hypothesis, alternative hypothesis
σ2s , σ
2w Variance of PU signal and noise signal
γ Signal-to-noise radio
Y, Y1, Y2 Test statistic of a detector, test statistic of rst level, test
statistic of second level of DS method
Pd, P d Probability of detection and target probability of detection
Pd1 , Pd2 Probability of detection at rst and second level of DS mech-
anism
Pf , PfSS , PfDS Probability of false alarm for generic, SS, and DS method
Pf1 , Pf2 Probability of false alarm at rst and second level of DS
mechanism
Pm Probability of missed-detection
ε, εSS Generic threshold value for detection and threshold for SS
mechanism
εS1, εS2 reshold of rst and second level of the DS mechanism
PaSS , PaDS Access probability of the SS and DS mechanism
N Normal distribution
Q(.) Gaussian Q function (one minus the standardized normal
distribution function)
xx VARIABLES AND NOTATIONS
ti, tb Sojourn periods (or holding times) in idle and busy states of
the PU
λi, λb PU’s arrival and departure rate
PH0 , PH0 e probabilities that PU is active and inactive in the channel
FY (.), F−1Y (.) Distribution function, inverse distribution function of Y
χ2, χ2
2M χ2(chi-square) distribution, with 2M degrees of freedom
Γ(a, b) Gamma distribution of shape parameter a and scale parameter
b
τ(1)s,max Maximum sensing period at sensing level (1)
Ts Expected sensing period
i, u Index and maximum size of backo stage
k,Wi, ωi Index of contention window and size of the contention win-
dow at i-th backo stage
W0, ω0 Minimum value of the contention window
σslot Length of a mini-slot
Pidle Probability of idle of the channel state
Pbusy Probability of busy of the channel state
NB Number of backo
D Propagation delay
TDIFS Expected time length of the DIFS
φ, φ(1), φ(2)Transmission probability, transmission probability in condi-
tion (1) and (2)
fφ(.) Symbolic function of φ
τds, τs, τc Sensing time for DS, spectrum sensing, and carrier sensing
CH0 , CH1 Capacity of the secondary link atH0 andH1 hypothesis
P aMAC Access probability through MAC protocol
R, R, R, Rmax Overall, aggregated, normalized aggregated, and maximum
normalized aggregated throughput
RDLS, RSLS Normalized aggregated throughput of dual-level sensing and
single level sensing
τs,s, τ∗s,s Sensing time and optimal sensing time of SS mechanism
VARIABLES AND NOTATIONS xxi
Pfmin Minimum probability of false alarm
x, F1, F2 Symbolic notation of Pd1 , Pf1 , and Pf2(1− Pf1)
DF (x) Dierential matrix of F (x)
xl, xu, x∗
Lower, upper, and optimal value of x
P ∗d1 Optimal value of Pd1
Λx,ΛPd1,Λτs Convergence criteria for x, Pd1 , and τs
List of Abbreviations
3GPP 3rd Generation Partnership Project
ACK Acknowledgement
AWGN Additive White Gaussian Noise
BS Base Station
CA Channel Assessment
CCA Clear Channel Assessment
CDMA Code Division Multiple Access
CoI Channel-of-Interest
CPE Customer Premise Equipment
CR Cognitive Radio
CRN Cognitive Radio Network
CRS Coarse Resolution Sensing
CSMA/CA Carrier Sense Multiple Access with Collision Avoidance
CSI Channel State Information
CSS Cooperative Spectrum Sensing
CTMC Continuous Time Markov Chain
CTSMC Continuous Time Semi-Markov Chain
CTS Clear-To-Send
CW Contention Window
DoA Direction-of-Arrival
DIFS Distributed Inter Frame Space
DS Dual-level Sensing
DSA Dynamic Spectrum Access
DSMA Dual-level Sensing based Multiple Access
DTMC Discrete Time Markov Chain
ETSI European Telecommunication Standards Institute
FSA Fixed Spectrum Access
FRS Fine Resolution Sensing
xxiv LIST OF ABBREVIATIONS
ITU International Telecommunication Union
IEEE Institute of Electrical and Electronic Engineers
IMT-A International Mobile Telecommunications-Advanced
ISM Industrial, Scientic, and Medical bands
MAC Medium Access Control
NAV Network Allocation Vector
OSA Opportunistic Spectrum Access
PBS Primary Base Station
PD Probability of Detection
PFA Probability of False Alarm
PHY Physical Layer
LTE Long Term Evolution
LTE-A LTE Advanced
PU Primary User
QoS ality of Service
RTS Ready-To-Send
ROC Receiver Operating Characteristic
RF Radio Frequency
SAA Sensing-Assisted Access
SBS Secondary Base Station
SDR Soware Dened Radio
SIFS Short Inter Frame Space
SU Secondary User
SNR Signal-to-Noise Ratio
TDM Time-Divisional Multiplexing
TDMA Time-Division Multiple Access
WLAN Wireless Local Area Networks
WRAN Wireless Regional Area Networks
WSN Wireless Sensor Network
Statement of Original Authorship
e work contained in this thesis has not been previously submied to meet require-
ments for an award at this or any other higher education institution. To the best of my
knowledge and belief, the thesis contains no material previously published or wrien
by another person except where due reference is made.
Signed:
Date:
QUT Verified Signature
Acknowledgments
I would like to take this opportunity to express my sincere gratitude to my supervisors,
A/P Bouchra Senadji and Dr Dhammika Jayalath, for their guidance and encouragement
throughout my research period. Bouchra has always guided my research to maintain
on the right track and helped me in transforming my visionary idea into a proven and
constructive research contributions. From the start of the research study, Dhammika
kindly guided me to improve my research skills and critical thinking which contributed
immensely to evolving the research contributions into a PhD level thesis.
My sincere acknowledgement goes to the panel members of my PhD nal seminar,
A/P Jonathan Bunker and Dr Jacob Coetzee, for their valuable comments to improve the
thesis. I am indebted to theeensland University of Technology for providing me with
the opportunity and stimulating research facilities over the entire study period. ank
you also to all academic and administrative sta in the School of Electrical Engineering
and Computer Science and in the Science and Engineering Faculty oce for helping me
with their prompt support.
I would like to thank professional editor, Dr Adele Fletcher, for providing copyediting
and proofreading services according to the guidelines laid out in the University-endorsed
national policy guidelines.
I am grateful to all members of my family for their love and generous support. I would
like to acknowledge my heartfelt gratitude to my parents for their faith in me and for
allowingme to be as ambitious enough for doctoral study. Without their encouragement,
the long journey of my PhD study would not come this far.
Finally, I would like to thank all mentioned and unmentioned fellow researchers and
friends for providing a sense of community to support each other. Special thanks to all
of my friends living in Australia, for their friendship and accompanying support which
xxviii ACKNOWLEDGMENTS
helped me to feel so engaged while living and studying abroad.
Ratan Kumar Mondal
eensland University of Technology
June 2018
CHAPTER 1
Introduction
1.1 Scarcity versus Underutilization in Radio Spec-
trum
Over the last decade, signicant developments in mobile devices, such as smartphones,
mobile phones, laptops, tablets, and personal digital assistants, have made those devices
a constant companion of our everyday work [8]. Already we have seen the advent of
a revolutionary technological advance on mobile devices and arguably the most trans-
formative: the marriage of mobile computing and seamless connectivity for widespread
applications and services [9–11]. e underlying communication medium of wireless
technology, the electromagnetic radio spectrum, is a precious natural resource. In the
twenty-rst century, no natural resource is more crucial to human prosperities than the
radio spectrum. Invisible, ubiquitous, and limited in physical extent, it is the transmis-
sion medium by which wireless technologies convey limitless sources of information to
revolutionize our access to the world around us.
Currently, the radio spectrum is regulated by governmental policy in the deployment
of various applications and services. International governing bodies, such as the Interna-
tional Telecommunication Union (ITU) [12], proposed xed spectrum allocation based
on applications, technological aspects, and geographical location to promote rational,
legitimate, and ecient accessibility of the radio spectrum all over the world [13]. Based
on the xed spectrum allocation (FSA), local governmental agencies, such as Australian
Communications and Media Authority (ACMA) in Australia [14], Federal Communica-
tions Commission (FCC) in United States (US) [15] etc., then allocate the accessibility of
2 1.1 Scarcity versus Underutilization in Radio Spectrum
the spectrum bands to the organizations and/or reserve the accessibility for non-prot
services. e accessibility of a particular spectrum band for the allocated applications is
referred to as the licensed spectrum. For instance, the allocated bands for cellular phone
service are 800, 900, 1800, and 2100 MHz bands.
According to Cooper’s Law, the maximum number of voice conversations or equi-
valent data transactions that can be conducted in all of the useful radio spectrum over
a given area doubles every 30 months [16, 17]; the wireless trac is doubling roughly
in every two years [18]. is wireless trac uptake requires huge amount of spectrum
capacity even though the technological advancement in the underlying system is relent-
less. Since the rst launch of a smartphone, the Apple iPhone in 2007, the smartphone
has become an integral part of our lives. A Survey shows that the smartphone ownership
has jumped to 84% in Australia and 81% globally at the end of 2016 [8]. Consequently,
mobile-broadband subscriptions are increasing with an average annual growth rate of
40% [8]. In February 2010, Cisco Systems Inc. predicted that 3.6 billion GB will be
communicated over wireless networks on a monthly basis by 2014, and the prediction
has proved correct [19, 20]. is widespread accessibility and sky-rocketing growth rate
demand high-speed broadband connections to mobile devices, which has had a massive
impact on current trends in wireless communication technologies. e far-reaching
deployment of wireless networks has saturated the spectrum, and thereby, FSA policy
in the eort to accommodate the data-hungry applications faces a spectrum scarcity
problem [17, 21]. Moreover, governing bodies [14, 15] mostly assign non-overlapping
bands to avoid mutual interference in FSA policy, which leads to inecient usage in
both temporal and spatial domains.
It is predicted that by 2020, we will have 50 billion connected devices, mostly wireless,
in the world, and existing spectrum usage policy will be unable to deliver the “end of
spectrum scarcity” [20, 22]. Meanwhile, smartest entrepreneurs in garages continue
to launch killer apps to the airwaves. e mobile bandwidth for excellent voice calls
appears to be underutilized because subscribers are increasingly interested in texting,
Facebook posting, tweeting, and video streaming [18]. A survey [8] shows, as of mid-
1.1 Scarcity versus Underutilization in Radio Spectrum 3
2016, 27% of mobile consumers claimed that they have not made any cellular voice calls
in a week, whereas that gure was about 23% in 2015. Users’ interest in using things
over mobile phones are changing drastically.
Table 1.1: Spectrum occupancy status of dierent applications over 30 - 3000MHz bandwith average 14% overall occupancy inChicago city, USfor 2010 [23].
Application Frequency Band Minimum Average Maximum
(MHz) (%) (%) (%)
PLM, amateur 30 - 54 8 18 60
TV 2-6 54 - 87 30 35 42.5
FM 87 - 108 80 90 92
Fixed, mobile, others 225 - 406 6 10 15
LMR, others 406 - 475 15 15.5 18
SMR 798 - 840 0.5 2 3
Cellular 840 - 902 50.2 55 68
Unlicensed 902 - 928 0.5 2.5 11
Radar, military, GPS 1240 - 1710 0 0 0
PCS cellular 1710 - 2010 17 17.5 20
ISM 2400 - 2500 18 24.5 45
WiMAX 2500 - 2700 17 26 31
Surveillance Radar, others 2700 - 3000 0 0 0
On the other hand, FCC reported that spectrum utilization varies temporally and
geographically between 15-85% with the FSA policy [15, 24]. Another study [25] indic-
ated that in particular applications such as in TV bands, broadcasting services do not
make complete use of the spectrum in regional areas and spectrum utilization is less
than 5% on average. A spectrum occupancy measurement was conducted by McHenry
et al. from 30 MHz to 3 GHz for a few hours in Chicago, US [26]. e instantaneous
spectrum occupancy was measured in terms of duty cycle1. It was found that TV bands
have the highest occupancy with an average duty cycle of 70.9%. e cellular and ISM
bands had occupancy with an average duty cycle of 55% and 29.1%, respectively. On the
other hand, few spectrum bands were entirely underutilized such as satellite bands. is
1Duty cycle is dened as the fraction of one period inwhich the licensed users is active in the spectrum.
It is expressed as percentage or a ratio, e.g., 40% duty cycle means the signal is on 40% of the operational
period but o 60% of the time.
4 1.2 A New Paradigm of Radio Spectrum: Dynamic Spectrum Access
measurement in Chicago city was further observed for the years 2008-2010 and reported
in [23]. For the year of 2010, the measurement data from [23] is presented in Table 1.1;
they found that the average overall occupancy was just 14% for the spectrum band of
30 MHz to 3 GHz. Based on their measurements, it is found that high occupancy was
observed at lower frequencies (less than 1 GHz) such as for cellular phone, broadcasting
radio, and TV, where high power with long range services are provided. On the other
hand, low occupancy occurred at the higher frequency ranges (greater than 1 GHz) such
as for satellite and radar operation [23, 26]. us, in some cases, the FSA policy faces
the spectrum underutilization problem.
1.2 A New Paradigm of Radio Spectrum: Dynamic
Spectrum Access
Studies [20, 21, 23–26] indicate that some of the radio spectrum is overutilized and caus-
ing spectrum scarcity for the deployment of new services and applications. Meanwhile,
several spectrum bands are underutilized owing to the static assignment of the radio
spectrum. To tackle the growing needs [19, 20] and dynamic behaviour of users’ interest
[8], a new spectrum usage paradigm is required that exploits the full potential of the
available spectrum. e only reasonable approach is to use the spectrum dynamically to
improve the spectral eciency with smartest technology [17, 24, 27].
To combat the spectral ineciency of the FSA, the Defence Advanced Research Pro-
jects Agency (DARPA) introduced dynamic spectrum access (DSA) strategy through the
NeXt Generation (xG) program [28]. e aim of the xG program was to propose a
spectrum access mechanism based on intelligent radios for providing high bandwidth
to mobile users [1, 28]. However, the xG proposal imposed enormous challenges to
communication technologies due to the unavailability of compatible heterogeneous net-
works that can support intelligent radios [1, 24, 27]. Owing to expectations of full-scale
deployment of the DSA strategy, a new network model for intelligent radios was in
demand.
1.2 A New Paradigm of Radio Spectrum: Dynamic Spectrum Access 5
To build the underlying framework for implementing the concept of DSA, several
international bodies such as the Institute of Electrical and Electronic Engineers (IEEE),
European Telecommunication Standards Institute (ETSI), and 3rd Generation Partner-
ship Project (3GPP), had taken initiatives to standardize the technological framework for
the governmental agencies and telecommunication industries. With the uptake require-
ment, the idea of DSA then evolved into the usage of radio with learning capabilities,
that is to say, radios able to gain knowledge about surrounding radios and tune up their
operational radio parameters and protocols accordingly. e term cognitive radio (CR)
was coined by Mitola and Maguire for soware-dened radio with learning capabilities
[29]. As dened by Haykin, “CR is an intelligent wireless communication system that
is aware of its environment and uses the methodology of understanding-by-building to
learn from the environment and adapt to statistical variations in the input stimuli, with
two primary objectives: highly reliable communication whenever and wherever needed and
ecient utilization of the radio spectrum”[30].
e enthusiasm behind the initial CR ideas unfolded in various directions, starring
to a variety of visions. However, behind the diverse CR interpretations lie the com-
mon features of awareness about the environment and dynamic accessibility. In the
terminology of communication theory, CR technology formed a network architecture
- cognitive radio network (CRN), whereby license holders of a spectrum, referred to as
primary users (PUs), allow non-licensed secondary users (SUs) to use the spectrum since
PUs’ transmission is not interrupted by the SU’s transmission [1, 30, 31].
To implement the DSA strategy through CRNs, three approaches to spectrum sharing
have been developed: spectrum underlay, overlay, and interweave [1, 32]. In the under-
lay approach, SUs coexist with the PUs subject to SUs’ interference to the PUs remaining
lower than a given threshold. Due to the stringent conditions of transmission power, the
underlay approach works only for short range communications. Although the overlay
approach can support long range communications with optimal data rates, knowledge
of the PUs’ codebook and/or messages is required by the secondary users. On the other
hand, the interweave system coheres with the original motivation of the cognitive radio
6 1.3 Spectrum Access in Cognitive Radio Networks
as it exploits the void spectrum opportunistically to communicate without interrupting
the primary user transmissions which brings the idea of opportunistic spectrum access
(OSA)[31, 33]. is research is focused on the nature of the interweave approach to CR,
where SUs are allowed to transmit only while the PU is sensed to be absent and required
to vacate when the PU reappears in the spectrum [1, 30–33].
1.3 Spectrum Access in Cognitive Radio Networks
e integrity of the CR depends on the ability of the SU to restrict interference to the
PU and maintain a reliable quality of service (QoS) with the spectrum access for its
own operations. To achieve this goal, SU must support the functionality to identify
the spectrum opportunity and to exploit the opportunity to its full potential [34, 35].
Spectrum opportunity is referred to as the specic dimension of wireless communica-
tion that is temporarily unutilized by its licensed users, PU, and which can be accessed
opportunistically by the SU [32, 34]. e conventional and most popular dimensions in
the modeling of spectrum opportunity are time, frequency, and space [30–32]. In this
research, in a single frequency band, which is called a channel, time-divisional spectrum
opportunity is considered for the CR operation.
Spectrum access (SA) is the task of the SU to exploit the spectrum opportunity with
the decision where and how an ecient transmission can take place [36, 37]. Before
exploiting the spectrum opportunity, SUs are responsible for identifying the spectrum
opportunity accurately and intelligently; this task is carried out by spectrum sensing
(SS) [31, 38–41]. SS is employed at the physical (PHY) layer of the SU to monitor a
channel of interest for detecting the PU transmission, and interference protection to the
PU can thereby be controlled. Spectrum sensing techniques are extensively studied in
the literature of CRN [2, 38, 39, 42–47] and relevant signal detection techniques such
as energy detection [42, 43], matched lter [44], and cyclostationary-based detection
[45–47], etc are adopted for cognitive radio networks.
Spectrum access strategy comes with data transmission decisions from the medium
access control (MAC) layer of the SUs [48]. MAC protocol has a crucial role in providing
1.3 Spectrum Access in Cognitive Radio Networks 7
several CR services: spectrum heterogeneity or mobility [36], sensing cycle assignment
[34, 49], resource allocation, and multiple access [1]. Spectrum heterogeneity allows
the SU to nd a best available free channel for the channel of interest and to operate
in multiple radio frequency bands [36]. Spectrum sensing is important to identifying
the spectrum opportunity that is associated with MAC-layer commands regarding how
oen and in which order SU senses the channel [34]. According to QoS request, MAC
protocol allocates the available resources to the SUs opportunistically. Multiple access
operation enables spectrum access among multiple SUs more dynamic which must be
required for real-world implementation of the CRN.
CR operation poses a lot of challenges in designing an ecient MAC protocol. More
importantly, the access strategymust consider the nature of heterogeneity among the PU
and SU for contention access in order to protect PU transmission from SU data transmis-
sion. Interference protection to PUs from the secondary transmission is guaranteed by
the spectrum sensing task. On the other hand, MAC ideally does not aware of the insight
of the spectrum sensing when it has to improve spectral eciency. Investigations [3, 50–
53] show that the maximization of spectral eciency relies on the following aspects:
• How eciently an extensive amount of spectrum opportunity can be discovered
by using a smaller amount of underlying resources.
• By using which techniques, the spectrum opportunities can be utilized to its fullest
extent.
e cognitive MAC protocol strives to achieve the fastest access decision in the chan-
nel as the legacy users of the channel do not share their transmission information with
the CR users. erefore, a resource consuming technique, a spectrum sensing method
is applied to the CRN to observe the PU’s transmission. Moreover, the SU’s channel
observation for a certain period changes abruptly due to the ad hoc nature of the cur-
rent wireless users [1, 31]. In existing ad hoc networks, a dedicated control channel is
typically used for sharing the channel state information among the users. To facilitate
the fastest access decision, few access protocols [54, 55] rely on the control channel
operation in CRN. However, the availability of the control channel, and coordination
8 1.4 Research Motivation: Sensing-Assisted Access
between the control channel and data channel is still a complex task due to the oppor-
tunistic nature of the data channel. Data transmission in the CRN itself depends on the
availability of a vacant channel, so maintaining a xed control channel for opportunistic
data transmission is not cost-eective for the SU [56, 57]. us, spectrum access has to
aware of the dynamic channel state in the ad hoc environment and needs to provide the
fastest access transmission excluding the control channel operation.
1.4 Research Motivation: Sensing-Assisted Access
Cognitive radio networks (CRNs) aim to maximize throughput while avoiding interfer-
ence to the primary network. ewhole time frame consists of sensing and transmission
operations, and the data rate2achieved through the transmission depending on the
sensing decision in that whole frame is referred to as throughput of the secondary
network. Previous research focused on optimizing sensing and transmission techniques
at the PHY [3, 51, 53, 58] and showed that throughput maximization and interference
reduction are conicting criteria. Interference reduction is based on sensing the presence
of PUs during a short sensing period designed to meet a target probability of detection
3(PD) of PUs. roughput is increased when the sensing period is short, allowing for
a longer transmission period. A short sensing period, however, also leads to a higher
probability of false alarm (PFA) (i.e., detecting the presence of a PU where no PU is
present), therefore limiting transmission opportunities and reducing throughput. is
issue is referred to as sensing-throughput trade-o [3, 59–61].
e trade-o issue has been formulated in [3, 51, 58] with the proof the existence
of an optimal sensing period for maximizing throughput under the constraint of target
PD. roughput is optimized by considering the best combination of two variables, the
sensing period and the detection threshold, to meet the constraint, which controls the
2Data rate is dened as the rate of successful bit transmission over a communication channel and is
usually expressed in the unit of bits per second (bit/s or bps).
3It is a constraint level of the spectrum sensing which denes the reliable detection decision for the
CR operation. e sensing operation must achieve the detection performance for which probability of
detection is greater than the target probability of detection.
1.4 Research Motivation: Sensing-Assisted Access 9
PFA (and transmission period). e length of the optimal sensing period may vary
with the impacts of PU trac model, channel degradation, and frame structure of the
sensing operation. For instance, a larger sensing period is required to achieve the highest
throughput (as achieved by [3] for the same channel model) while dynamic PU is con-
sidered as investigated in [61, 62]. e optimum trade-o also requires a longer sensing
period to achieve maximum throughput when fading [60] and noise variance [53] de-
grade the channel signal-to-noise ratio (SNR) compared to that found by Liang et al.
[3].
roughput can also be optimized by improving access techniques at the medium
access control (MAC) layer. Even though throughput optimization techniques at PHY
andMAC layers have evolved independently, someMAC-based protocols for throughput
optimization also rely on sensing [6, 63–65]. e above-mentioned techniques, however,
consider that all PUs and SUs are homogeneous, and do not give preference to PUs for
access purposes. A similar sort of access protocol is carrier-sense multiple access with
collision avoidance (CSMA/CA), which allows channel monitoring through physical sig-
nal detection before any packet transmission. eCSMA/CA is themost ecientmethod
for random channel access in the homogeneous type network. erefore, it is widely
used in wireless local area network (WLAN) [66, 67]. Due to the compatibility of the
CSMA/CA with CR operation, several existing works [6, 7, 49, 63, 65, 68] have adopted
the CSMA/CA protocol directly into the CRN by considering conventional underlying
mechanisms. Even though those models [6, 7, 49, 63, 65, 68] improve the throughput
performance, they cannot guarantee sucient interference protection to the primary
network.
CR technology has emerged as a promising paradigm for ecient spectrum utiliz-
ation, and spectrum access is a key component to improve the utilization of the CR
operation. e research gaps that exist in the designing of sensing assisted access pro-
tocol are potentially threatening the advancement of CR capability and from them also
unsolved questions, such as:
• How does the spectrum sensing impact on the modeling of the access strategy to
10 1.5 Research Goal and Approaches
improve the access capability?
• Is the sensing decision signicant for the enhancement of the access decision?
• If so, how can the sensing be embedded with the access strategy to overcome the
sensing-throughput trade-o issue?
• What is the best way to integrate the sensing with access strategy for the purpose
of overcome the sensing-throughput trade-o issue?
is research is motivated by the above questions and aims to provide satisfactory
solutions, which are presented in this thesis.
1.5 Research Goal and Approaches
e main goal of this research is summarized as follow:
“To investigate the impact of spectrum sensing on the improvement of access capability
and develop access strategies to ensure greater performance in both interference protection
and achieved throughput.”
e core idea behind this research is to integrate the sensing and the access mech-
anism by exploiting the cross-layer concept to overcome the sensing-throughput trade-
o issue. To do that, it must track down the aspects that are the main barrier behind
the sensing-throughput trade-o issue. erefore, this research rst conducts a com-
prehensive investigation into the measurement of capacity variation of the spectrum
opportunity in relation to the sensing parameters. e investigation implies that the
outcome of the spectrum sensing determines the potential of the spectrum opportunity.
In addition, the detection is congured by the target value of the PD which appears to
impact on the variation of spectrum opportunity.
e spectrum opportunity is reduced when a strongest interference protection is
aimed for by seing a large value of the target PD. By contrast, obtaining the highest
possible spectrum opportunity inuences the sensing mechanism in achieving the target
1.5 Research Goal and Approaches 11
PD (i.e., missed detection increases), which leads to packet collision during channel ac-
cess. is interrelated issue is synthesized explicitly to obtain a solution for the purpose
of exposing the larger spectrum opportunity and reducing the collision eect. Both
factors cannot be compensated by the underlying improvement of the sensing method.
erefore, the issues are divided into two stages and solved by a single cross-layer
platform.
e objective of the rst stage is to acquire the fullest possible capacity of the spec-
trum opportunity from the spectrum sensing operation. e detection sensitivity is
compensated to obtain the maximum spectrum opportunity by reducing the target PD.
e probability of collision during the SU transmission is increased as a consequence. In
the second stage, the objective is to reduce the collision rate. If the enforced collision rate
can be limited below a tolerable range, then the ultimate goal of this research, throughput
improvement, can be achieved successfully. is challenging task is accomplished by
using a cross-layer design, where the spectrum sensing in the PHY layer is integrated
with the contention access method in the MAC layer.
Contention-based access is a transmission protocol by which the transmission time
can be scheduled randomly among multiple users. In particular, a backo mechanism
including channel monitoring is used before any data transmission in the contention-
based access method. e backo process is a collision avoidance feature to reduce
the collision among packets being transmied by the SU. e backo process accounts
channel sensing for processing the transmission delay. However, existing studies did
not propose any ecient solutions by which the backo process can contribute to the
overall interference protection in the CR operation. is research aims to integrate the
backo and detection process to guarantee a robust interference protection as well to
achieve greatest throughput performance.
e main challenges in sensing-assisted access strategy is to interrelate the sensing
outcome with the backo process where ideally both are designed separately in conven-
tional access protocol. e integration process is discussed with a wide range of analysis
to reveal the eectiveness of cross-layer design in achieving the research goal. e
12 1.6 Overview of esis Structure
research accomplished in this thesis is the activity of the future trend to raise cognisance
of above-mentioned problem and provide solutions with modeling and analysis which
outperforms the existing studies.
1.6 Overview of Thesis Structure
e research contribution presented in this thesis is organized as follows:
Chapter 1 explains the motivation for the research and draws the signicance of
research that contributes to the advancement of cognitive radio technology. e research
objectives and the technical approaches used are also presented in this chapter.
Chapter 2 provides an overview of cognitive radio operation and a comprehensive lit-
erature review on relevant aspects of spectrum access in current wireless technology and
its deployment scenarios. is chapter focuses on the challenges and solution branches
of the access mechanisms. e required branches of a cognitive radio network, such as
the network architecture, spectrum occupancy modeling, sensing, and access methods,
are thoroughly reviewed along with the current challenges and future trends.
Chapter 3 shows the impacts of spectrum sensing on the design of the CR operation
through a capacity measurement of the spectrum opportunity. e underlying obstacles
of the sensing-throughput trade-o issue is synthesized with the analysis of the receiver
operating characteristic (ROC) and access probability. To achieve the highest possible
spectrum opportunity, a novel sensing mechanism is proposed by dierentiating the
target PD into dual steps.
Chapter 4 provides the design aspects of the dual-level sensing mechanism in order
to achieve the highest possible throughput. e achievable throughput by an SU is for-
mulated based on the proposed DS mechanism. To obtain maximum throughput under
the constraint of PU protection, an optimization is conducted between the sensing period
and the detection probability of the DS mechanism in this chapter. is optimization
shows the signicance of the DS mechanism for designing an ecient access protocol.
Chapter 5 proposes a dual-level sensing-based multiple access (DSMA) protocol. e
cross-layer framework of the sensing and access mechanism is described, with consider-
1.7 List of Publications 13
ation of practical data transmission scenario. e derivation of the detection sensitivity
conguration according to the contention parameters is provided in this chapter. e
achievable throughput is formulated with the cross-layer parameters by using Markov
chain analysis.
Chapter 6 presents a complete sensing-assisted access protocol for a cognitive ra-
dio network. e cross-layer framework developed in chapter 4 is exploited with the
contribution of embedding the sensing in the backo process in this chapter. Besides
the throughput measurement in multiple access, a delay analysis is also developed to
characterize the entire behavior of the proposed access protocol.
Chapter 7 presents the conclusions of this thesis alongwith the essentials of the tasks
accomplished and the technical contributions made. In addition, recommendations are
made for the further improvement of the access strategies in cognitive radio technology.
1.7 List of Publications
e publication works during this research are listed below:
1. R. K. Mondal, B. Senadji, and D. Jayalath, “Dual-Level Sensing Based Multiple
Access Protocol for Cognitive Radio Networks,” in 2017 IEEE 85th Vehicular Tech-
nology Conference (VTC Spring), Jun. 2017.
2. R. K. Mondal, B. Senadji, and D. Jayalath, “A Novel Sensing-Assisted Access Pro-
tocol for Cognitive Radio Networks and its Performance with Imperfect Sensing,”
IEEE Wireless Communications Leers, (under review).
3. R. K. Mondal, B. Senadji, and D. Jayalath, “Sensing-Assisted Access Protocol with
Imperfect Sensing and Performance Analysis for Multiple Access in Cognitive
Radio Networks,” (manuscript to be submied).
4. R. K. Mondal, B. Senadji, and D. Jayalath, “Sensing-roughput Tradeo of Dual-
Level Sensing Based Access Mechanism for Cognitive Radio Networks,” (manu-
script to be submied).
CHAPTER 2
Background and Literature Review
2.1 Introduction
Dynamic spectrum access (DSA) policy brings the idea of cognitive radio for ecient
spectral utilization to challenge outdated spectrum access policy based on xed spectrum
access. CR technology allows the SU to occupy a licensed spectrum that is owned by
the PU, without producing any harmful interference [31, 32, 39]. erefore, SU must
empowers the intrinsic capabilities to aware about its surrounding environments, tune
up onto reachable RF, and adapt its operation to restrict harmful interference to the
legacy users and obtain best eort service from its network [69].
e introduction of new spectrum access policy also brings numerous technical chal-
lenges for real-world implementation. Since CR technology aims to operate in the best
available spectrum, existing RF hardware has to be upgradedwith the capability ofmulti-
band operation [70]. Smart soware models have to be connected with the physical RF
model with learning capabilities [29, 32, 69]. e network model will be more complex
due to the dynamic and high load balancing issue [31, 71]. e transmission protocol
needs to be recongured because the primary and secondary network within a CRN
cannot be coordinated with a dedicated control channel [36, 72]. Overall, the CR tech-
nology has raised many research challenges as well as promoted huge opportunities for
innovation, to bring about a new era of wireless industry.
CR operation promotes the concept of dynamic access in the underutilized spectrum
for improving the spectral eciency. ere are no restrictions on the network archi-
tecture of the cognitive radio networks unlike other wireless networks such as WLAN
16 2.2 Components of Cognitive Radio
Figure 2.1: Components of the cognitive cycle [1].
and cellular network. erefore, many network models are adopted the CR concept
for providing the best eort services in dierent applications. is chapter provides an
overview of the CR operation and shows the application site of the CR concept in current
trends.
2.2 Components of Cognitive Radio
e CR capability of an SU is such that the SU is able to interact with its reachable radio
environment to determine the most ecient and target radio. e SU can choose the
channel of interest and adapt with the dynamic access environment. e entire task
depends on the adaptive operation in the channel of interest which is called a “cognitive
cycle”, as depicted in Fig. 2.1 [1, 29, 30]. ere are three major components of a cognitive
cycle: spectrum sensing, spectrum analysis, and spectrum decision. e key roles of
these components are described below.
1. Spectrum Sensing: is component allows the SU to monitor the scannable
spectrum bands, measuring the radio information for nding the spectrum holes1
1Spectrum hole is a band of frequencies allocated to PU, however, the band is not being utilized fully
by its incumbent user at a particular dimension (time, frequency, and space).
2.2 Components of Cognitive Radio 17
[31, 39].
2. Spectrum Analysis: e SU can characterize the spectrum hole with objective
performance parameters through this component; for instance, channel state in-
formation (CSI) and capacity prediction could be accomplished using the spectrum
analysis [30].
3. Spectrum Decision: rough this component, the SU performs the data trans-
mission in the spectrum hole. Resource management, power control, and access
strategy during transmission are carried out in this section [30, 70].
Receiving section of the SU is responsible for executing the tasks residing into the
rst two components and the transmiing section mainly executes the tasks of the third
component [1, 30]. According to layer-based model (e.g., OSI model), these components
are transformed into two major functionalities in two dierent layers, particularly in the
PHY and the MAC layer [31, 73]. Specically, sensing and analysis are carried out by
the spectrum sensing operation in the PHY layer. e spectrum decision, including data
transmission, are taken into account by the spectrum access functionality in the MAC
layer [70].
Spectrum sensing has emerged as the key enabler of cognitive radio operation [1,
30, 31, 39, 70]. e main task of sensing is to characterize the available spectrum hole
through radio signal measurement in the tunable air interface without interfering the
legacy system. e task of the spectrum sensing is classied into four steps [30]:
1. Spectrum holes detection
2. Dening the resolution of spectrum holes
3. Determining the directions of arriving interference
4. Classication of signals
Spectrum hole detection means nding the sub-band which is only occupied by white
noise. Simply put, the detection of spectrum holes can be performed using existing RF
18 2.2 Components of Cognitive Radio
detection techniques, for example, the energy-detection method. When the sub-bands
are partially occupied by interference and noise, then the detection may further require
power spectrum estimation [2, 39]. ereaer, addressing the resolution of spectrum
and deciding of the interference’s direction-of-interval (DoA), can boost up the spectrum
utilization of the cognitive users. erefore, the sensing outcome is formed through fur-
ther analysis, for example, time-frequency analysis [39] and binary hypothesis-testing
problem [31], to classify the target signal accordingly. e hypothesis-testing problem
is simple and based on parametric spectrum sensing. is procedure can only be desig-
nated for each sub-band as black (blocked space, busy) or white (exploitable) space. For
instance, null hypothesis H0 indicates the absence of the PU’s signal and alternative
hypothesis H1 indicates the presence of the PU’s signal. On the other hand, time-
frequency analysis (non-parametric sensing) can decide that a signal may be useful for
the low-power secondary userwhichmay be tooweak to be of use in a particular location
for the primary user. Due to the simplication and usefulness of the parametric sensor in
performance optimization, the hypothesis-testing procedure becomes mostly applicable
in cognitive radio research [3, 31].
As discussed above, the sensing operation mostly relies on a detection process which
can be done by primary signal detection and interference-temperature measurement
[74–76]. Primary signal detectionworks according to the principle of RF signal detection.
Many RF detection methods have been adopted as primary signal detection method in
cognitive radio networks such as energy detection [3, 42, 43], matched-lter detection
[44], cyclostationary detection [45–47], eigenvalue-based detection [77], covariance-
based detection. On the other hand, interference-temperature measurement considers
the cumulative RF energy from PU transmission and sets a maximum limit that primary
users can tolerate. Secondary users can use the band if their transmissions do not
exceed the interference-temperature limit [75]. Owing to the diculty in distinguishing
the primary signal from noise/interference in interference-temperature measurement,
the primary signal detection method becomes ecient in cognitive radio for spectrum
sensing.
2.2 Components of Cognitive Radio 19
Figure 2.2: Review of dierent detection techniques based on complexityversus accuracy [2].
In the literature on spectrum sensing, energy detection is popular because of its low
complexity and cost eectiveness. However, its detection performance during channel
impairments is poor when compared with the performance of other detection methods
as shown in Fig. 2.2 [2, 46]. If the SUs have prior knowledge regarding the primary
transmission then the matched-lter is the optimal detector, as it has the capability to
maximize the received signal-to-noise ratio during the worst channel conditions. As PUs
do not share their transmission informationwith the CR network, it is, therefore, dicult
to implement the matched-lter in conventional spectrum sensing. e cyclostationary
method can classify the signal hence there is capability to distinguish the co-channel
interference. However, the cyclostationary method requires longer computational time
with higher complexity to achieve a target detection when compared with the energy
detection and the matched-lter method, which may increase the overall sensing time
[31]. Waveform-based sensing and radio identication method are relatively robust
than energy detector owing to the coherent processing and feature extraction capab-
ilities with the help of a prior knowledge regarding PU’s characteristics and paerns.
Each detection method has pros and cons in relation to accuracy, complexity, cost-
eectiveness, system limitations, and assumptions. Even though energy detection has
poor detection performance, it becomes the method for PU detection in CR due to the
20 2.3 Standardization and Implementation of CR
working capability without any prior knowledge.
Considering the importance of sensor parameters for further contributing to spec-
trum access, the parametric hypothesis-testing procedure is used in this research.
Simply, the transmission of the SUs are constrained with the target PD which is not
specically provided with the selection factors of any detector. However, the sensing
eciency of the detectors may vary in dierent channel conditions. ese variations
only signify the aribute of dierent detectors. ere is a large body of literature on the
choice of detection method for RF signal. e study of detector selection as a tool for
spectrum sensing is not an objective of this research, but most importantly, the further
utilization of the sensing parameters with a post-processing algorithm is an objective
to improve the spectral eciency. Detector-independence and post-processing of the
sensing parameters provide greater exibility in designing the universal access protocol,
and they have become extremely signicant in the recent literature on CR technology.
2.3 Standardization and Implementation of CR
Due to the demand of CR technology for ecient utilization of the spectrum, the IEEE
802.22 Working Group on Wireless Regional Area Network (WRAN) has launched a
standard based on CR operation. IEEE 802.22 [78] is the rst standard that allows CR
operation for wireless networks. is standard species the air interface which operates
in the VHF and UHF broadcasting bands in the range of 54 − 862 MHz. e support-
ing network architecture in this standard is point-to-multipoint WRAN consisting of a
professional xed base station (BS) and customer premise equipment (CPE) as shown in
Fig. 2.3. e CPE can be xed and portable user terminals which can tune to the given
TV broadcast bands. e purpose of this standard is to provide alternatives to wire-line
broadband access in diverse geographic areas where population density is low.
e IEEE standard mainly denes the PHY and theMAC layer operations for support-
ing the purposes of WRAN in TV bands. e network coverage under a BS is typically
10 − 30 km depending on the EIRP and antenna’s specication. e coverage of the
WRAN system can be further upgraded to 100 km based on special scheduling in the
2.3 Standardization and Implementation of CR 21
Figure 2.3: Network architecture of the IEEE 802.22 WRAN, where users ofTV bands and wireless microphones are the primary users, andBS and CPE are the secondary users [3].
MAC and exceptional RF signal propagation in the PHY. To meet the requirements of
PU protection and ecient spectrum utilization, the CR capabilities comprise spectrum
sensing, database access services, channel set management, and geolocation services.
All supporting devices, such as BS and CPE, need to be empowered with the given CR
capabilities. e CR capabilities enable the BS and the CPE to produce robust decisions
regarding the characteristic of the using RF. e trac activity of the incumbents of
the TV bands is dynamically updated in the database which information can be used as
a supplement of the sensing to protect the incumbents. e type of detectors are not
regulated by this standard. However, this standard promotes the scheduled spectrum
sensing and sensing information sharing to make a central decision.
As the research activities documented in this thesis is focused on the issue of sensing-
assisted access strategy, the IEEE 802.22 standard is thus reviewed on the context of
research issues. According to the standard, the MAC accommodates all necessary tools
for protecting the incumbent users of TV bands and for the coexisted services. In a cell,
22 2.4 Current Trends and Applications of CR
the BSmanagesmultiple CPEs and themedium access is controlled by theMAC protocol.
In the downstream transmission, when BS transmits and CPE receives, MAC protocol
supports time-divisional multiplexing (TDM). During the upstream transmission, MAC
provides a combination of access strategy depending on the user application including
its QoS. When multiple CPEs aempt to transmit simultaneously by sharing the same
channel within a cell or overlapping cells, then the MAC provides the upstream schedul-
ing based on the following mechanisms: unsolicited bandwidth grants, polling, MAC
header-based contention, and CDMA-based contention. No specic access protocol is
proposed in the standard, it is assumed that the existing access protocols are sucient
to support the CR capabilities. In the implementation, however, the existing access
mechanisms are not directly applicable as the current working group of the IEEE 802.22
have suggested.
2.4 Current Trends and Applications of CR
2.4.1 CR-based Wireless Sensor Networks
econcept of CR technique has received considerable aention for its capacity to enable
opportunistic access in wireless sensor networks (WSN).e cognitive capabilities in the
sensor networks empower the sensor node with the ability to access reachable channels
opportunistically [4, 79]. is feature drastically the transmission reliability and energy-
eciency drastically of the WSN. Currently, WSN works in the industrial, scientic,
and medical (ISM) bands which are typically unlicensed bands. e services provided
through ISM bands are increasing rapidly due to the lack of licensing which apparently
makes this band overcrowded. erefore, current WSNs include CR capabilities to nd
the underutilized spectrum band that can be accessed for best eort service or bursty
trac as shown in Fig. 2.4.
e CR-based wireless sensor network (CR-WSN) is a promising paradigm for sensor
oriented services by upgrading the conventional sensor network with CR capabilities.
Most of the sensor network supports low-power and short range communication in a
2.4 Current Trends and Applications of CR 23
Figure 2.4: Network model of a proposed CR-WSN system [4].
single radio channel. Moreover, sensor nodes are densely deployed, producing large
numbers of packet bursts. In such cases, the chance of packet collision increases lead-
ing to inecient power consumption and large packet delay [4]. With CR capabilities
the sensory nodes can access the reachable radios hence the communication reliability
overcomes the shortcomings of conventional WSNs [4, 79, 80].
2.4.2 Cognitive Radio in Cellular Networks
In the RF spectrum, the 5 GHz band is an unlicensed band with 500 MHz bandwidth and
used for pure WiFi (in accordance with the IEEE 802.11a/ac/ax standard) and weather
radar applications. is unlicensed band is oen used for providing excellent data rates
for short range communication. erefore, LTE-A2operators are progressively con-
sidering the unlicensed bands as a complementary resource for achieving best eort
services in the small cell scenario [82, 83]. As depicted in Fig. 2.5, the BS of a small cell,
i.e., eNB in an LTE network is able to tune up with the carriers simultaneously both the
licensed and the unlicensed bands, where the licensed carriers in the macrocell and small
2Long Term Evolution (LTE) is a technical standard for cellular network proposed by the 3GPP
organization. e LTE-Advanced (LTE-A) is the upgraded specications of the LTE which completely
fulls the requirements set by ITU for IMT-Advanced and 4G [81]. e cellular networks adhering to the
LTE-A standard are oen referred to as LTE-A networks.
24 2.4 Current Trends and Applications of CR
Figure 2.5: Network model of a proposed CR-LTE system [5].
cell coexist. In such cases, the conventional licensed carrier (called the primary carrier)
remains fundamentals for guaranteeing its QoS to the user equipment (UE) [5].
Because of the opportunities oered by CR, the LTE cellular networks are envisioned
to support the coexistence of CR capabilities with unlicensed bands. In the proposed
model, the main data operation streams through a licensed carrier (LC) and additional
data bursts can be oered via unlicensed bands [83]. In coexisted scenario, the UE of the
LTE carrier acts like an SU and incumbent users of the unlicensed band (e.g., stations
of WiFi networks), are the PUs. Such a utilization of the unlicensed bands must be
conducted as a good neighbor of the incumbent users of the unlicensed bands [5, 84].
e 3rd Generation Partnership Project (3GPP) has initiated the standardization of
utilizing the 5 GHz unlicensed band as a secondary component carrier integrated with
the fundamental (licensed) carrier of the LTE-A networks using a novel access tech-
nology referred to as “Licensed-Assisted Access” (LAA) in the Release 13 [84]. e
European Telecommunication Standards Institute (ETSI) [85] has proposed a novel ac-
cess mechanism incorporating channel monitoring to accommodate the LTE-WiFi coex-
2.5 Spectrum Access rough MAC Protocol 25
istence in the unlicensed band.
2.5 Spectrum Access Through MAC Protocol
Medium access control (MAC) protocol provides channel access mechanisms for several
users while sharing the communication medium [66, 86]. In most of the wireless net-
works (e.g., WLAN, WSN, and WPAN) all users must be organized by MAC protocol
for ecient utilization of the shared medium. Traditional MAC protocol, however, can
interfere with the primary network and degrades the overall throughput. Robust opera-
tion of a cognitive cycle can be ensured by the proper control of the sensing and access
mechanism. A MAC protocol can facilitate the control operation on the distribution of
spectrum sensing and access in a cognitive cycle [6].
Due to the spectrum heterogeneity3, CR users require additional features in the
MAC protocol for providing interference protection to the primary network without
any internetwork collaboration. is requirement forces a drastic redenition of the
functionality of the MAC protocol for CRN. e cognitive MAC protocol (C-MAC) must
assist the secondary network to cope with the spectrum heterogeneity through a com-
patible channel access strategy for exploiting the full potential of the opportunity.
e operational eciency of the C-MAC protocol is generally inuenced by several
MAC-independent factors, such as PUs’ trac activity, tolerable interference to PU, and
hardware constraints for multi-band operation. An ecient spectrum access strategy
can aid the C-MAC protocol in improving the operational eciency. us, C-MAC
should also take into account the sensing scheduling task as well as the spectrum access
task because sensing is an obligatory task to protect the primary network. Not only
the sensing scheduling but also the sensing decision can be crucial for improving the
capabilities of the C-MAC. Spectrum access through C-MAC is a complex task due to the
dependency on sensing decisions, and there are many issues requiring research in the
area of sensing integrated MAC protocol for CRN.e following sections review factors
3CR users have their own legacy spectrum and can operate in other spectrum, depending on the
availability of the spectral resources. e phenomenon of the CR users’ operation in several spectrum is
referred to as spectrum heterogeneity.
26 2.6 Cross-Layer Components for Sensing-Assisted Access Protocol
relating to the C-MAC protocol and its potential to oer higher opportunity in operation,
current status, and challenges associated with the issues to justify the contention of this
thesis. e review provides with extensive overview on state-of-the-art of C-MAC, using
techniques of spectrum sensing and accessing with further improvement through cross-
layer operation.
2.6 Cross-Layer Components for Sensing-Assisted Ac-
cess Protocol
2.6.1 Spectrum Sensing Algorithm
is research is focused on the post-processing of the spectrum sensing data regard-
less of working beyond the detection theory. Moreover, the objective is to connect
the advantages of sensing data in the design of the access protocol to improve the CR
capabilities. To improve CR capabilities, numerous sensing algorithms are proposed
in the literature of sensing algorithm. e proposed algorithms are associated with
dierent terms and conditions. In this section, a state-of-the-art of the sensing algorithm
is presented in the context of access protocol design.
Spectrum sensing algorithm allows the post-processing of the signal detection data
from the PHY for further evaluation. In the PHY, PU detection is carried out by using the
following techniques: energy detection, matched lter, cyclostationary, etc. [2, 41]. In
particular, the received signal is synthesized according to the detection techniques, and
by comparing the processed output and pre-dened threshold value, the nal decision
is obtained. e nal decision can be either a so decision (parametric value) or a hard
decision (e.g., binary hard decision is ON or OFF). is detection decision can be further
proceed to increase the controllability of the spectrum sensing, and this is referred to as
post-processing of the detection [2, 31, 41]. In this research, it is assumed that the sensing
algorithm conducts the entire operation of the spectrum discovery for CR operation from
PU signal detection to post-processing.
Although an optimal detector applies for individual sensing, an inappropriate estima-
2.6 Cross-Layer Components for Sensing-Assisted Access Protocol 27
tion can be experienced due to channel impairments [31, 35, 87]. For instance, path loss,
multipath fading, and shadowing eects can degrade the sensing output of multiple SUs
over a certain network. In such conditions, an individual decision varies depending on
local observation which needs to be further processed for making a global and robust
decision about the channel activity. To overcome the deciency of an individual de-
cision, cooperative spectrum sensing (CSS) is proposed [31, 42, 72, 87], by sharing the
local observation over the network to make a combined decision about the PU activity.
In the CSS algorithm, all local decisions, either so or hard decisions, are taken into
the fusion machine. Based on decision rules, either AND or OR rules, a nal decision
comes out regarding the channel status. e diversity gain is imposed to overcome the
deciency of the sensing experienced by low SNR. Challenges in the implementation of
CSS algorithm include increased complexity and the requirement of an additional control
channel [2, 41]. Even though the CSS algorithm can enhance the sensing output under
channel impairments, the large overhead in the CSS algorithm makes it inecient for
access protocol design [52].
A dynamic sensing technique is proposed by [50] with the scheduling of multiple
sensing cycles before data transmission. Higher spectrum utilization is achieved by
the dynamic sensing method [50] when compared with single and static method of the
sensing [3]. e sensing period is dierentiated by lower sensitivity into multiple stages
in multi-stage sensing algorithms. e dynamic sensing or multi-stage sensing has a
great advantage in wideband sensing, as shown by [51], where an optimal sensing round
with dierent sensitivity is allocated tomaximize the throughput. e idea ofmulti-stage
sensing is narrowed down in [52, 88] by exposing the multi-threshold values in dierent
sensing stages under the constraint of primary user protection. A signicant diculty
in multi-stage sensing includes modeling of the multi-threshold value in a discrete time
scale where the channel state may changes dynamically over the given sensing period.
Sensing cycle design is an important task for conguring the transmission period in
the MAC protocol. e capability of sensing cycle design in the MAC layer is referred
to as spectrum discovery and/or spectrum search. Sensing sequence scheduling through
28 2.6 Cross-Layer Components for Sensing-Assisted Access Protocol
the MAC protocol is proposed by [34]. ey found that the overall discovery is maxim-
ized and the delay in nding idle channels is minimized by choosing an optimal sensing
cycle in [34]. e authors in [89] proposed a two-stage sensing, where the multiple
frequency-divided channels are simultaneously observed by using multiple antennas.
In the frequency dimension, sensing starts with coarse resolution sensing (CRS) and
partitions all the channels within a dened bandwidth regardless of obtaining any idle
channels. en, ne resolution sensing (FRS) takes into account that newly dened
bandwidth and continues until obtaining an idle channel. is type of sensing algorithm
is particularly proposed for spread spectrum where frequency hopping is used [89].
To overcome the sensing-throughput trade-o issue, multi-stage sensing algorithm
gains great aention in the literature of CRN [34, 51, 52, 64, 88, 90]. Nonetheless, the
resourceful ordering of the stages, weighting factors in stages, the optimal parameters
designing, and integrating with the access protocol can impact on the performance
improvement.
2.6.2 Spectrum Occupancy Modeling
According to underlying condition of CR operation, there are no cooperation and net-
work association among the PU and SU. erefore, SUs do not have exact networking
knowledge about of trac of PUs. Hence, SUs can only gain knowledge about PU trac
by the spectrum measurement and this has been studied extensively in the current
literature [23, 26, 33, 91–93]. Experimental measurements suggest that the spectrum
occupancy can be modeled using certain statistical and/or mathematical models [33, 91–
93]. Spectrum occupancy modeling is important for determining the full potential of
the spectrum opportunities before accessing the spectrum. Moreover, the accuracy of
the spectrum sensing can be evaluated with the help of knowledge of the spectrum
occupancy; hence, interference protection to PUs can be designed deliberately. For
instance, interference to PU is minimized with the sensing parameters optimizations
achieved by using the dynamic trac model of the PU [94]. Without paying aention
to the approach, the statistical model of the spectrum occupancy is comprehensively
2.6 Cross-Layer Components for Sensing-Assisted Access Protocol 29
equipped to envision the CR operation [48, 49, 57, 64, 65] but of barely sucient accuracy
to characterize the PU activity.
e statistical model is extracted from the measurement data for CR designs obtained
from measurement campaigns. e most popular and natural choice for statistical mod-
eling of the spectrum occupancy is Markov chain model. In the current literature, the
following Markov chain based models are found widely used: continuous time Markov
chain (CTMC), continuous time semi-Markov chain (CTSMC) [33, 91], discrete time
Markov chain (DTMC) [92], and heuristic model [93].
In the Markov chain model, the state of the spectrum is dened as a random process
that switches between several possible states, and eventually, the spectrum occupancy
rate can be characterized by the transition probabilities. Based on the characteristics
of the random process and its post-processing, the given models [33, 91–93] can be
distinguished. In CTMC and CTSMC models, the spectrum state is characterized by the
holding time or sojourn time where the holding time follows an exponential distribution
and an arbitrary distribution, respectively. On the other hand, the DTMC model does
not allow the channel or spectrum state to stay on any of the states; hence, distribution
of the holding time is applicable in the occupancy modeling.
Early works found that the CTMCmodel is widely used in modeling the occupancy of
high-frequency bands. Several measurement campaigns revealed that the CTSMC has
beer accuracy than the CTMC, especially in the modeling of the WLAN trac over
2.4 GHz band with an approximation model. ence, the CTSMC became a popular
choice for occupancy modeling at the early stage of the development of CR technology
[33, 91]. e study in [33] suggested a generalized Pareto distribution was suitable for
dierent frequency bands when the sampling rate was relatively low. e simplifying
Markovian assumption, (i.e., semi-Markov) was, however, insuciently accurate for the
all other radio trac regimes due to the large approximation of the measurement data as
suggested by [92]. Unlike the standardMarkov chain model [33, 91], an empirical DTMC
model was also suitable for occupancy modeling where spectrum states are assumed as
continuously switching. To accelerate the dynamic switching in the discrete-time, the
30 2.6 Cross-Layer Components for Sensing-Assisted Access Protocol
transitions probabilities were expressed as the functions of time. By applying both the
deterministic and stochastic methods, transitions probabilities were characterized from
the measurement data with the perfect agreement between the empirical curves and
ed curves.
e Markov chain-based model is quite simple but largely acceptable statistical ap-
proach for the spectrum occupancy modeling in time dimension. However, in addi-
tion to the time dimension, the spectrum occupancy can also be modeled by space and
frequency dimension. e shortcoming of Markov chain-based modeling is that the
spectrum occupancy in the space and frequency dimension.
2.6.3 Data Transmission Mechanism
Inspired by the success of random access technique, several data transmission protocols
have directly adopted the random access techniques, such as sloed ALOHA, CSMA/CA,
in the CRN [7, 63, 68, 95]. For the multiple access scenario in CRN, two types of users
with dierent prioritized access in the channel have been considered in [7, 63, 68]. For
multiple access in the primary channel among multiple SUs, the existing access protocol
such as sloed ALOHA [7] and CSMA/CA [6, 63, 64] are adopted for the cognitive radio
scenario. In the given literature, two types of users with dierent prioritized access
were considered. e CSMA/CA has an advantage over sloed ALOHA for CRN, as
the CSMA/CA allows channel monitoring functionality before transmission which is
essential for occupying the primary channel. Time-division multiple access (TDMA) is
also used in CR with a cooperative MAC protocol as proposed by [96, 97]. Nevertheless,
without any inter-network collaboration and/or precise synchronization, the TDMA
approach cannot guarantee sucient protection to the primary network.
e data transmission proposed in [7, 68] is based on a two-level access policy, where
the interference protection to the PU and sensing time optimization is done at the rst
level, and packet scheduling based on the MAC protocol is enabled at the second level.
In particular, two dierent MAC protocols, i.e., CR-ALOHA and CR-CSMA mechanisms
are used for the packet scheduling. e main limitation of this model is that if missed
2.6 Cross-Layer Components for Sensing-Assisted Access Protocol 31
detection occurred at the rst level, then the CR-ALOHA and the CR-CSMA cannot
provide a sucient interference protection to the PU as they did not have any collision
avoidance procedure during the access period. On the other hand, this limitation has
been aempted to overcome in their other works [63] by introducing a new control
packet named prepare-to-send (PTS) with ready to send (RTS)/ clear to send (CTS) mech-
anism during the channel access. However, the carrier sensing is performed before the
spectrum sensing so that the detection performance the channel is comparatively poor,
which can impact negatively on the PU’s protection.
Traditionally, MAC access protocols do not take into account spectrum sensing when
designing access strategies, which leaves the PUs potentially open to severe interference.
A decentralized cognitive MAC protocol [48] rst allows for spectrum sensing where
access is enhanced by compensating for a higher probability of false alarmwhile keeping
the sensing period unchanged. An aempt at improving throughput by considering
both sensing and access was made in[64] based on the IEEE 802.11 distributed coordin-
ated function (DCF) [66]. Even though all the proposed access methods based on the
CSMA/CA can improve the throughput, they cannot guarantee sucient interference
protection to the PU. is has occurred because conceptually, the PU is not incorpor-
ated in the backo mechanism with the CRN and the fundamental spectrum sensing is
omied while proposing the access protocol in the existing literature.
In the other networks that consider CR capabilities also rely on the Listen-Before-Talk
(LBT) mechanism. Such an mechanism is proposed for LTE architecture. is proposal
exploited the coexistence of the LTE and WiFi networks in unlicensed 5 GHz band and
has mainly been adopted by a radio access technology called carrier-sense adaptation
transmission (CSAT), as proposed by alcomm [98]. To accommodate the LTE-WiFi
coexistence in the unlicensed band, the ETSI has developed a frame-based equipment
(FBE) scheme which is quite similar to the CSAT scheme [85]. ere has been a fast
uptake of the LTE globally, and the LTE-A is a hugely successful platform in terms of
its widespread adoption for 5G deployment as well as meeting the recent demand. At
the same time, usage of the unlicensed band needs to comply with certain regulations
32 2.7 Sensing-Transmission Optimization
in several regions in the world, for instance, LBT mechanism must follow to use the
unlicensed band in Japan and Europe. us, to enable the CR capabilities in real-world,
the data transmission should be associated with the spectrum sensing.
2.7 Sensing-Transmission Optimization
e sensing-transmission optimization is formulated explicitly in [3, 53] and proved
the existence of optimal sensing period that could maximize the throughput under the
constraint of target PD. e eect of PU trac on the sensing-throughput trade-o
has been investigated in [61, 62]. Moreover, the impacts of the fading and noise vari-
ance in channel propagation on this trade-o problem are demonstrated in [60] and
[53] respectively. e channel degradation could be overcome by using cooperative
spectrum sensing (CSS), however, there is an additional trade-o between cooperative
overhead and gain. e trade-o is conducted in [87] by allocating optimal number of
SUs in cooperative detection to meet the target PD to maximize the throughput within
shorter sensing period; also the maximum throughput obtained in [87] is larger than the
throughput achieved in [3, 53, 60].
By exerting interleaved transmission with periodic sensing, the authors in [99] pro-
posed an opportunistic channel-aware access to reduce the sensing error. Despite lever-
age the sensing purpose by the periodic sensing, the interleaved transmission imposes
large overhead when SU transmits a large data packet. Furthermore, for a stable oper-
ation and small delay tolerance, the access scheme should incorporate with the robust
sensing policy as suggested in [100], which presents a cross-layer (PHY/MAC) approach
to clarify the eect of sensing in access scheme for maximizing the throughput under a
PU’s stability constraints. Moreover, the authors in [59] enhanced the throughput per-
formance with simultaneous sensing and transmission like full-duplex mode by forming
a new frame structure. However, full-duplex adaptation in cognitive radio network is
still a challenging task as stated in [101].
e access strategy in the perspective of sensing-throughput trade-o problem has
been investigated in [6, 102]. In [6], a distributed MAC protocol is designed similarly
2.8 Model of Access Protocols Based on Cross-layer Design 33
with the control channel operation of [54] and the throughput is optimized in terms of
sensing period and contentionwindow. However, their optimization is almost the simple
form of sensing-throughput trade-o problem [3] as physically the contention window
is formulated linearly in the time slot which is nothing but the same scale of sensing
period. Based upon only the resolution of access contention, the authors re-established
the distributedMAC protocol of the [6] by an overlapping channel assignment algorithm
where the secondary users use the interference avoidance approach during the packet
transmission similar with the CSMA/CA approach. However, the improvements of [6,
102] are largely dependent on the control channel and the synchronization which are
still a burden for designing the access protocol in the CRN as suggested by [7, 63].
To ll up the above-mentioned research gap, a new framework for improving
throughput in CSMA/CA by “restructuring” the sensing period to meet the target
probability of detection is proposed in [103]. e proposed protocol was referred to
as dual-level sensing based multiple access (DSMA) where the spectrum sensing is
accompanied with carrier sensing to decide the channel status jointly. We illustrated
that the throughput improvement mostly dependent on how much the overall PFA can
be reducible. On the other hand, interference protection is enhanced by addressing the
contention access method aer the nishing of the CR sensing period. Moreover, the
PU protection can be controlled by choosing the suitable threshold values into two steps
to meet the target PD that may also vary the overall PFA; which also arises the sensing-
throughput trade-o problem. erefore, it is of great importance to nd the impact
of the dual-level sensing on the sensing-throughput optimization problem. Motivated
with this importance, we will study the feasibility analysis of the optimization problem
and propose an algorithm to maximize the throughput of the DSMA protocol under the
constraint of the PU protection.
34 2.8 Model of Access Protocols Based on Cross-layer Design
BackoffF
S
S
F
S
SData
(d)
BackoffF
S
S
C
S
S
Data(b)
BackoffF
S
S
P
T
S
Data(c)
C
S
S
F
S
S
C
S
S
C
S
SData
(a)
Figure 2.6: Review of frame format with sensing-transmission mechanism.
2.8 Model of Access Protocols Based on Cross-layer
Design
According to the fundamental principle of cognitive radio operation, SUs are only al-
lowed to transmit data while the channel is sensed as idle. To adhere to this principle,
SUs have usually employed the LBT [28] mechanism in which spectrum sensing fulls
the listening function at the PHY, and the transmission function refers to the packet
scheduling at the MAC layer. In IEEE 802.22 standard [78], the MAC protocol allows
sensing-transmission combination in an operating frame, where two types of periodic
sensing are proposed: coarse spectrum sensing (CSS) and ne spectrum sensing (FSS) as
shown in Fig. 2.6(a). e objectives of CSS and FSS are to identify vacant spectrum with
a shorter sensing period and to support the previous sensing algorithm with a longer
sensing period, respectively [88]. For a fair comparison, it is assumed that the frame used
in [49, 63, 65, 78, 88] followed a similar format of two-stage physical sensing and backo
period as shown in Fig. 2.6. Note that individuals’ detection outcomes and contention
access impacted on the achievable throughput, which was dierent, even though the
same time duration is reserved for the data transmission.
Unlike [78, 88], a conventional backo mechanism (Fig. 2.6(b)) is applied between
2.8 Model of Access Protocols Based on Cross-layer Design 35
the two stages [49] to enable the CSMA mechanism to cope with two-stage sensing.
Although imperfect sensing was considered for two xed sensing stages, a conventional
backo mechanism [66] with perfect detection was adopted which led to a burden on
the second sensing in making the nal decision. e authors in [63] overcome the
shortcomings of [49] by introducing a conventional backo process [66] at the start
of the frame, as illustrated in Fig. 2.6(c). Despite leveraging the sensing purpose by two-
stage detection, the proposed protocol in [63] imposed an overhead by introducing a new
control packet, prepare-to-sense (PTS), between backo process and the FSS. In contrast,
a relatively robust sensing mechanism is used in [65] compared to [49, 63] by allowing
two FSS operation consecutively before the backo mechanism, as shown in Fig. 2.6(d),
for enhancing the spectrum opportunity (by reducing the probability of false alarm).
However, the access protocol in [65] causes severe interference to the primary users as
perfect detection is also assumed during the backo process. All the MAC protocols
mentioned above have signicant outcomes in conict with the IEEE 802.22 [78]; how-
ever, the access protocol can be more ecient and practical if the sensing aspects can be
exploited during the backo process [64, 100, 104]. To analyze the impact of the sensing
error, the authors in [104] included the sensing error cases in the backo mechanism
of the CSMA/CA protocol which is not thoroughly examined for the cognitive radio
environments.
e main challenge in the sensing-assisted MAC protocol is to reveal the cross-layer
eectiveness towards achieving the goals of the CRNs, such as improving spectral e-
ciency without producing severe interference. To expose the importance of sensing on
access protocol, the sensing parameters is integrated with the contention window for
improving the throughput and delay performance in the proposed model.
Owing to the advantages of sensing-assisted MAC protocols for throughput maxim-
ization in a multiple access scenario, [54] and [65] integrated the spectrum sensing with
the channel assignment on the MAC protocol. In [54], SUs access the channel through
theMAC protocol which uses clear channel assessment (CCA) functionality to detect the
transmission in a channel. In [54], the channel assessment is done through two phases,
36 2.8 Model of Access Protocols Based on Cross-layer Design
the reporting phase and the negotiating phase, in an additional control channel. In the
reporting phase, SUs do the spectrum sensing and report the acquired information. In
the negotiation phase, a p-persistent based access protocol is applied to contend for the
transmission in the next frame. However, a dedicated control channel may not always
be available in practice, and also the consideration of the CSI of an additional channel
may increase the computational complexity. e authors [65] proposed a contention
access strategy by sensing two channels sequentially in a single slot duration. However,
the authors in [65] did not consider the detection operation in the contention duration
(where detection usually occurs by carrier sensing in the contention window of the
CSMA/CA mechanism) as the contention window is assumed to be short compared
with sensing duration. Nevertheless, taking a larger threshold value for the detector
during short contentionwindow [63] can also play the same role as the spectrum sensing
does for the primary user detection in the CRN. In that case, single channel sensing
including the carrier sensing during the contention window can improve the detection
performance, instead of the two channels sensing with the exclusion of carrier sensing
in the contention period.
Multiple access in cognitive radio can be enabled by enhancing the PHY-MAC jointly
[33, 49, 54, 95]. e enhancement requires the conventional access exhibited in a distinct
network where the trac dynamics of the user is identical throughout the transmission
time. Since SU measures the energy level of PU’s transmission before accessing, the
SU does not acquire exact knowledge about the PU’s trac dynamics [49, 95, 105]. A
decentralizedMACprotocol has been proposed based on the partially observableMarkov
decision processes (POMDPs) framework to overcome the absence of any central entity
in [48]. Although the MAC protocol proposed in [33, 48, 54] can increase the SU’s
performance while inhibiting the interference experienced by the PUs, it requires exact
information on the PU trac. Moreover, the proposed model in [33, 48, 54] consumed
most of the resources, such as time duration and energy for that collaboration which
reduced the eciency of the CRN. On the other hand, the proposed PHY/MAC cross-
layer in [54, 64] based opportunistic MAC protocols for provisioning the QoS in CRN,
2.9 Chapter Summary 37
but this is costly as it requires additional control channel operation for the negotiation
based mechanism. So far, all the proposed models related to the cross-layer approach
[33, 49, 54, 64, 95] have not directly addressed the measurement and eectiveness of
spectrum sensing when multiple SU access the licensed spectrum.
In this work, the focus is maintained on the sensing-assisted MAC protocol where,
contrary to [6, 54, 63], neither an additional control channel operation is considered
nor the carrier sensing is omied over the contention access period. Instead of direct
adoption of the CSMA/CA protocol into the CRN [63], we propose the dual-level sensing
within the same sensing period (used in [3, 6, 65]) and exploit it into the CSMA/CA-
based access mechanism. e impact of the entire sensing heterogeneity in improving
the throughput of the DSMA scheme has been illustrated in [103]. However, the optim-
ization of the DSMA protocol is required to accomplish the sensing-throughput trade-o
issue. An investigation is conducted over this optimization problem to nd a solution
framework.
2.9 Chapter Summary
is chapter has given an operational overview of the cognitive radio and some applica-
tions of CR technology. e CR operation is comprised by multiple tasks to conrm pro-
cient communication and incumbent protection. Such multiple tasks are accomplished
by the cognitive cycle. e cognitive cycle is empowered by three building blocks:
spectrum sensing, spectrum analysis, and spectrum decision. e CR technology shows
huge opportunities in ecient spectrum utilization of the future generation network.
erefore, IEEE proposed a new standard for CR capabilities of the wireless devices
that can tune up in the TV band in WRAN environment. In addition, CR technology is
adopted in other networks such as in sensor networks, and cellular networks. e design
aspects of the SU’s transmission protocol associated with spectrum sensing identied
in the above literature review. Also the review illustrates the necessity of cross-layer
design for developing a sensing-assisted access protocol. According to the cognitive
cycle, spectrum sensing is one of the key enablers of the CR operation that should be
38 2.9 Chapter Summary
considered explicitly in the design of a complete access protocol. erefore, the impact of
sensing on the design of the access protocol is thoroughly reviewed in the next chapter,
with their applications and technical challenges.
CHAPTER 3
Impact of Spectrum Sensing on theCapacity Measurement of Spectrum
Opportunity
3.1 Introduction
Spectrum sensing and spectrum access are two key components of a “cognitive cycle”
[1]. In the context of CR operation, sensing and access contribute to the discovery
of spectrum opportunity and the proper utilization of that opportunity, respectively.
rough the spectrum sensing, SUs obtain the occupancy status of the channel which
also aids to measure the capacity of the spectrum opportunity. Before transmiing any
data into the vacant channel, it is also essential to know the oered capacity of the
opportunity for designing an ecient access protocol. By identifying the oered capa-
city, which is determined by spectrum sensing, the access mechanism can congure its
transmission policy to achieve the maximum utilization of the oered capacity. Studies
[3, 34–37] have shown that the utilization of the opportunity without causing harmful
interference to the primary network are related to spectrum sensing. In this chapter,
therefore, a comprehensive analysis is conducted regarding the impact of sensing on the
measurement of the capacity of the spectrum opportunity, before proposing the access
protocol.
Spectrum sensing takes place at the physical layer (PHY) and has been extensively
studied, with numerous sensing techniques being developed for interference reduction
[1, 3, 30]. Spectrum access takes place at the medium access control (MAC) layer and
traditionally the sensing aspect if not taken into account when designing the data trans-
40 3.1 Introduction
mission strategies. Due to the dierence between traditional wireless networks and
cognitive radio network regarding the sensing before data transmission, the SU needs
to take into account the sensing aspect in designing data transmission strategies by
assessing the capacity of the spectrum opportunity [34, 48, 49].
In a single frame, the SU performs spectrum sensing and then decides on the data
transmission based on the sensing decision. e sensing period is designed to meet the
PD requirements set by the interference protection for the PU. Investigation in CRN [3]
indicates that a longer sensing period provides greater interference protection to the PU
and consequently reduces the transmission time of the SU. As a result, the SU cannot
achieve enough throughput to maintain the QoS by using that shorter transmission time.
To obtain a longer transmission period, the SU needs to perform the sensing within a
shorter period. On the other hand, a shorter sensing period causes larger PFA which
eventually reduces the spectrum opportunity. Hence, reduction of sensing time cannot
be a straightforward solution for increasing the throughput. is issue is referred to as
the sensing-throughput trade-o and cannot be overcome adequately by using single-
level sensing where typically a single threshold is used in determining the spectrum
occupancy [3, 7, 52, 57].
In practice, the PU is protected by designing a higher detection probability by which
themissed detection can be kept within a tolerable range. Studies [3, 50] indicate that the
spectral eciency of the single-level sensing (SS) mechanism is relatively inadequate for
a higher target probability of detection. To the best of our knowledge, the SS method is
unable to exploit the sensing period eciently towards improving the spectral eciency.
erefore, a dual-level sensing (DS) is proposed where two sensing levels are employed
conditionally during the sensing operation to determine the channel status jointly. e
contribution of the DS mechanism to the gain of higher spectrum opportunity is em-
phasized by the measurement of access probability. rough mathematical derivations,
it is proven that by allowing a section of the sensing period to be devoted to reducing
the probability of false alarm, the overall probability of detection is still met while the
access probability is improved.
3.2 System Model 41
time
. . . .
sτ
f sT τ−
fT
Spectrum AccessSpectrum Sensing
Frame 2 Frame 3Frame 1. . . .
Figure 3.1: Frame structure for CR operation with spectrum sensing andaccess in every frame.
3.2 System Model
In considering the CRN, SUs are allowed to access a single time-sloed channel only
while no PU is present in the channel. ere is no cooperation between the SU and
PU, and PU transmission is not aware of the SU transmission. e SU follows a frame
structure which consists of a sensing period and an access period as shown in Fig. 3.1.
Since the PU and SU are non-cooperative, SU performs spectrum sensing at the starting
of each frame. en, access operation comes into account followed by the sensing.
During the sensing operation, SUs are allowed to employ a signal detection method
to determine the channel status by comparing the received signal with a predened
threshold value.
3.2.1 Spectrum Sensing Model
Let y(m) denote the received signal to the secondary user for primary user detection
over τs period with sampling frequency fs, where m is the sampling index and total
number of samplingM = bfsτsc. Applying a binary hypothesis-testing problem [1, 3],
the detection process can be modelled as,
H0 : y (m) = w (m) (3.1)
H1 : y (m) = s (m) + w (m) (3.2)
42 3.2 System Model
where s(m) and w(m) denote the transmied signal and the additive white Gaussian
noise (AWGN) respectively. Assume that, s(m) and w(m) are independent and identic-
ally distributed (iid) random process with both having the mean zero, and variance σ2s
and σ2w respectively. HypothesisH0 andH1 describe the absence and presence of the PU
signal, respectively. e measured signal-to-noise ratio (SNR) under the H1 hypothesis
is γ = σ2s/σ
2w.
Aer the post-processing of the received signal through a specic detector, such as
energy detector (ED), matched-lter (MF), the generated outcome is called test stat-
istic which is denoted by Y (y). e test statistic Y (y) is compared with a predened
threshold ε to obtain the nal detection decision about the channel occupancy, under
the hypothesis of H0 and H1. e performance of the detection is evaluated with the
following metrics:
• Probability of detection (Pd): the probability of deciding the PU signal is present
whileH1 is true, which can be determined by Pd = P Y > ε|H1.
• Probability of false alarm (Pf ): the probability of deciding the PU signal is present
when H0 is true, i.e., Pf = P Y > ε|H0. In the context of CRN, false alarm is
treated as a sensing error which means that the spectrum holes is not detected
even though there is a spectrum hole. us, a lower Pf is desirable to obtain large
spectrum opportunity for the SUs.
• Probability of missed-detection (Pm): the probability of deciding the PU signal is
absent whenH1 is true, i.e., Pm = P Y < ε|H1, and thus, Pm = 1− Pd. is is
also a sensing error where lower value ofPm is desirable to reduce the interference
to the PUs.
Energy Detector
Let us apply an energy detector (ED) in spectrum sensing. In the energy detector, the
received signal y(m) is ltered through a bandpass lter with the bandwidth ofBW , then
the ltered output is squared and integrated overM samples to produce a test statistic
3.2 System Model 43
Y (y). us, the test static of the energy detector is given by
Y (y) =1
M
M∑m=1
|y (m)|2 (3.3)
Let us assume that the transmied signal in the channel is a complex-valued PSK
modulated signal and the noise is circularly symmetric and complex Gaussian (CSCG)
signal. For a largeM , the distribution of the test statistic is obtained as follows [1, 3, 43],
H0 : Y ∼ N(σ2w,σ4w
M
)(3.4)
H1 : Y ∼ N(
(γ + 1)σ2w, (2γ + 1)
σ4w
M
)(3.5)
whereN indicates the normal distribution. By using central limit theorem (CLT) for the
large number of samples, the performance metrics are expressed as follows [3, 64, 65],
Pf (ε) = Q
((ε
σ2w
− 1
)√M
)(3.6)
Pd(ε) = Q
((ε
σ2w
− γ − 1
)√M
2γ + 1
)(3.7)
where Q(.) is a complementary distribution of standard Gaussian i.e.,
Q (x) =1√2π
ˆ ∞x
e(−t2/2)dt (3.8)
3.2.2 PU Activity Model
Studies [48, 49, 57, 64, 65] suggest that PU trac modeling is well ed by a two-
state ON-OFF process where ON and OFF states respectively indicate busy and idle
activity of the primary user. is trac model was authenticated with experimental
results [33, 91] for modeling the PU’s activity in wireless local area network (WLAN).
We assume that PU activity consists of idle period interspersed with a busy period in a
frame of a single channel scenario. By following [48, 65], we also assume that the sojourn
periods (or holding times) in idle and busy states are ti and tb. e arrival of PU trac
44 3.3 Conventional Single-level Sensing Mechanism
is independent; thus, both periods ti and tb are independent and identically distributed
(iid). e transition among the alternating states follows Poisson arrival process1where
ti and tb with rate parameters λi (arrival rate) and λb (departure rate), respectively; thus
mean holding periods E[ti] = 1/λi and E[tb] = 1/λb. Applying two-state discrete time
Markov chain process [64, 65, 106], the steady-state probabilities of the PU’s activity can
be expressed as
PH0 =E[ti]
E[ti] + E[tb]=
λbλi + λb
(3.9)
PH1 =E[tb]
E[ti] + E[tb]=
λiλi + λb
(3.10)
where PH1 and PH0 are the probabilities that the PU is active and inactive in the channel,
respectively, and PH1 + PH0 = 1.
3.2.3 Spectrum Access Decision
By assuming the activity model of the PU, the CRN can congure the channel state aer
every detection process as follows,
1. When the PU is inactive, and the detector produces no false alarm then the channel
state is decided as idle with probability PH0(1− Pf ).
2. In contrast, the channel can also be idle with probability PH1Pm, if missed detec-
tion occurred.
e both events are independent in occurrence, thus, SU access the channel with the
probability of
Pa = PH0(1− Pf ) + PH1Pm (3.11)
3.3 Conventional Single-level Sensing Mechanism 45
sτ
f sT τ−
Spectrum Accessssε
Single-level
Sensing with
threshold
Figure 3.2: Frame structure of conventional single-level sensingmechanism.
3.3 Conventional Single-level Sensing Mechanism
e frame structure for the SS mechanism is given by Fig. 3.2, where a single threshold
value is applied for making the nal decision over the sensing period τs. e sensing
operation is congured by the fullling the given constraint Pd ≥ Pd. By considering
the equality constraint as Pd = Pd in (3.7), the detection threshold for SS is determined
by
εSS = σ2w
(√2γ + 1
fsτsQ−1
(Pd)
+ γ + 1
)(3.12)
By using (3.12) and (3.6), for a given target probability of detection Pd, the probability of
false alarm of the SS can be obtained as follows,
PfSS(Pd, τs) = Q(√
2γ + 1Q−1(Pd)
+ γ√τsfs
)(3.13)
e corresponding access probability for which SU can access the channel is [3]
PaSS = PH0(1− PfSS) + PH1(1− Pd) (3.14)
To achieve higher access probability, the SU needs to reduce thePfSS asmuch as possible.
e authors in [3, 50] has already proved that the PfSS is a convex function with respect
to sensing period τs, thus, PfSS can be reduced monotonically by increasing of τs. As
the tolerable interference imposed on PU is determined by the Pd, and the τs is xed to
correspond that Pd, so a lot of transmission opportunities are wasted with the higher
1With the experimental measurements [33, 91], it is proved that the transition can be described by
using Poisson distributions. e core idea of using this Poisson distribution is to track down the arrival
and departure rate over the observation period by which the SUs can employ spectrum sensing on the
detected ON periods and wisely occupy the channel during OFF periods.
46 3.4 Proposed Dual-level Sensing Mechanism
sτ
f sT τ−
Spectrum Access1s
ε2s
ε
Dual-level Sensing
with threshold
1sτ
2sτ
Figure 3.3: Frame structure of proposed dual-level sensing mechanism.
PfSS in this SS mechanism.
3.4 Proposed Dual-level Sensing Mechanism
e dual-level sensing (DS) policy operates over two steps in the time frame and de-
termines the channel state jointly. It consists of rst sensing (S1) over τs1 period and
conditional second sensing (S2) over τs2 period as shown in Fig. 3.3. e proposed
detection process computes the rst test statistic Y1 and compares to the rst threshold
εs1 . If Y1 > εs1 then the channel is declared to be busy. Otherwise the second sensing
comes into operation for the next sensing period τs2 . e channel status is assessed as
busy or idle, similarly to the rst sensing process, by comparing second test statistic Y2
with εs2 .
e distribution of the test statistics is formulated now to obtain the detection per-
formance of the DS. To do that, the rst and the second sensing are revised under H0
andH1 hypothesis. Similar with the Y , the two test statistics of the S1 and S2 steps can
be expressed as Y1 and Y2 where Y1 + Y2 is equivalent with Y under the compensation
of τs1 + τs2 = τs. If YC denote the distribution of S1 ∪ S2 then YC ⊂ Y2 given that
Y1 + Y2 < εs1 (which means that S1 is failed). However, in general, the distribution of
the YC can be expressed numerically [107] as follows,
P (YC ≤ x) =P (Y2 ≤ x ∩ Y1 + Y2 ≤ εs1)
P (Y1 + Y2 ≤ εs1)
=
´ x0P (Y1 ≤ εs1 − Y2)P (Y2) dY2
P (Y1 + Y2 ≤ εs1)(3.15)
Let YC,0 and YC,1 be the distribution of YC under H0 and H1 hypothesis, respectively.
3.4 Proposed Dual-level Sensing Mechanism 47
Hence, the distribution of the test statistic for S2 is estimated as follows,
P (YC,0 ≤ x) =
´ x0P (Y1,0 ≤ εs1 − Y2,0)P (Y2,0) dY2,0
P (Y1,0 + Y2,0 ≤ εs1)(3.16)
P (YC,1 ≤ x) =
´ x0P (Y1,1 ≤ εs1 − Y2,1)P (Y2,1) dY2,1
P (Y1,1 + Y2,1 ≤ εs1)(3.17)
From the above distributions, the probability of detection and probability of false alarm
for the S2 can be determined by considering εs2 in equation (3.16) and (3.17) as, Pd2 =
P (YC,1 ≥ εs2) and Pf2 = P (YC,0 ≥ εs2), where FYC,0 = P (YC,0 ≤ εs2) and FYC,1 =
P (YC,1 ≤ εs2) are the cumulative distribution functions (CDF) of YC,0 and YC,1 respect-
ively. By using the distribution of YC , Pf2 and Pd2 can be expressed as follows,
Pf2 = 1− P (YC,0 ≤ εs2) = 1− FYC,0 (εs2) (3.18)
Pd2 = 1− P (YC,1 ≤ εs2) = 1− FYC,1 (εs2) (3.19)
For a given probability of detection at the rst sensing (Pd1) to achieve the overall
detection probability Pd, the Pd2 is obtained as
Pd2 =Pd − Pd11− Pd1
(3.20)
Hence the threshold εs2 and the probability of false alarm Pf2 of S2 step is estimated by
as follows,
εs2 = F−1YC,1
(1− Pd2) = F−1YC,1
(1−
(Pd−Pd11−Pd1
))(3.21)
Pf2 = 1− FYC,0(F−1YC,1
(1−
(Pd−Pd11−Pd1
)))(3.22)
Similarly the Pf1 can be expressed as a function of Pd1 as follows,
Pf1 = 1− FYC,0(F−1YC,1
(1− Pd1))
(3.23)
48 3.5 Performance Analysis
Finally the overall probability of false alarm of the DS mechanism can be expressed by
PfDS = Pf1 + (1− Pf1)Pf2 (3.24)
According to the decision criteria, the access probability by which SU can access the
channel is expressed as
PaDS = PH0(1− PfDS) + PH1(1− Pd) (3.25)
From (3.22), (3.23), and (3.24), we see that the overall probability of false alarm when
dual-level sensing is used is a function of both the overall probability of detection (Pd)
and the probability of detection achieving through the rst sensing. erefore, when Pd
is xed, for example Pd = 0.9, the overall probability of false alarm can still be controlled
by the appropriate choice of Pd1 whereas in the single sensing case, once Pd is xed, Pf
is also xed. e aim is to select an appropriate target Pd1 , so that the PfDS is less than
PfSS .
e target PD is aliated exponentially with the threshold values. For the sake
of simplicity, the PD of the rst detection is chosen as the controlling parameters to
correspond with the target PD instead of the threshold values in the analysis of system
performance. As the PU is protected with the target PD which is set as constraint of the
optimization, so it is a linear transformation of the optimization problem with the PD at
any sensing level.
3.5 Performance Analysis
is section presents the performance analysis of proposed DS mechanism. Firstly, the
detection capability of the DS and SSmechanism is comparedwith the receiver operating
characteristic (ROC) curve. Secondly, the achievable opportunity by the SS and DS is
evaluated with the access probability.
3.5 Performance Analysis 49
3.5.1 Receiver Operating Characteristic
e ROC curve determines the sensitivity performance of a detector which is congured
with binary hypothesis testing problem [43]. In conventional detection theory, the ROC
curve is a graphical plot of probability of detection versus the probability of false alarm as
its decremental threshold value changes. Here, the complementary ROC curve, graphical
plot of Pf versus Pd, is used to illustrate the overall detection capability of any particular
detection method, where any point of the curve describes a set of (Pd, Pf ) for a given
threshold value. Since the comparison between the proposed DS and conventional SS
mechanism is carried out to obtain the reducible amount of Pf , hence, the using ROC
curve can be expressed as
Pf = f (Pd) (3.26)
In the conguration of a CRN, the PU is protected by seing a target Pd. For instance,
the IEEE 802.22 standard has recommended to use Pd = 0.9 for−20 dB SNR value in the
sensing of white space in TV band [108]. In this condition, the detection must employ
a detection threshold ε by satisfying Pd ≥ Pd. On the other hand, the target detection
performance depends on the number of sample M and channel SNR γ as depicted in
(3.7).
Apart from protecting the PU with higher Pd, it is also preferred to obtain the higher
spectrum opportunity with lower Pf . e preferred set of (Pd, Pf ) for a given SNR value
is examined from the ROC curve analysis. e overall detection capability is said to be
high when the ROC curve exhibits the lower value of Pf at the higher value of Pd. e
detection capability of the DLS and SLS mechanism is compared with the ROC curve
and analyzed the capability in dierent SNR values.
By providing the equal weight in the both sensing levels, the individual probability of
detection is obtained as Pd1 = Pd2 while the overall probability of detection Pd has to be
xed for a given value ofM . To meet this condition, let us assume that Pd1 = Pd2 = Pd
and Pf1 = Pf2 = Pf , and then obtain Pd as follows,
Pd = 1−√
1− Pd (3.27)
50 3.5 Performance Analysis
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Probability of missed detection
Pro
babili
ty o
f fa
lse a
larm
SS: Theoretical, SNR = −15 dB
SS: Simulation, SNR = −15 dB
DS: Theoretical, SNR = −15 dB
DS: Simulation, SNR = −15 dB
SS: Theoretical, SNR = −20 dB
SS: Simulation, SNR = −20 dB
DS: Theoretical, SNR = −20 dB
DS: Simulation, SNR = −20 dB
Increasing of SNR
Figure 3.4: ROC comparison of the SS and DS mechanism with theoreticaland simulation results at a given SNR value.
Now, employing the Pf and Pd in overall PFA as expressed in (3.24), the Pf can be
expressed as,
Pf (Pd) = Pf (Pd)(
2− Pf (Pd))
(3.28)
where Pf (Pd) is obtained as follows,
Pf (Pd) = Q
(√2γ + 1Q−1
(1−
√1− Pd
)+ γ
√⌈M
2
⌉)(3.29)
Now the required expressions for ROC curve of the SS and DSmechanism are as follows,
PfSS (Pd) = Q(√
2γ + 1Q−1 (Pd) + γ√M)
(3.30)
PfDS (Pd) = Pf (Pd)(
2− Pf (Pd))
(3.31)
Fig. 3.4 gives the ROC curve for SS and DS mechanism with the theoretical and
simulation results where Pm = 1−Pd relationship is used to present the ROC curve. e
simulation parameters are: Tf = 10 ms, γ = −15,−20 dB, Pd ∈ (0, 1), fs = 6 MHz.
Fig. 3.4 shows that there is a close agreement between the simulation and theoretical
results for both of the comparing mechanisms. From the ROC curve, the operating point
3.5 Performance Analysis 51
0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009 0.010
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Sensing time (τs)(sec)
Probabilityoffalsealarm
Pf2
Pf1
PfDS
PfSS
Figure 3.5: Probability of false alarm vs. sensing time of DS and SS strategy;PfDS is less than PfSS for a given Pd = 0.9.
of the detector for an SNR value can be determined. For example, if the PU interference
protection is guaranteed with the constraint of Pm ≤ 0.1 at γ = −20 dB, then, the
operating point of the detector can be determined from this curve which provide with
the coordinate of (Pm, Pf ) corresponding at Pm = 0.1. Ideally, the operating point is
eective when both the Pf and Pm have the lowest value. In some detection analysis,
the operating point of the detector is chosen when the summation of Pf and Pm has the
minimum value. From this gure, it can be noticed that Pf and Pm cannot be reduced
jointly. In the context of CR technology, the detection mechanism is chosen based on
having an ecient ROC curve at given SNR value. Achieving the lowest Pf decides
an ecient ROC for a given value of Pm (or Pd = 1 − Pm) at an SNR value. Fig.
3.4 indicates that the Pf of the proposed DS mechanism is always lower than the SS
mechanism for any given value of Pm. Hence, the proposed DS mechanism shows beer
ROC characteristic than the SS mechanism regarding ecient signal detection at lower
SNR value.
52 3.5 Performance Analysis
3.5.2 Access Probability
Based on (3.23), it can be said that Pf declines monotonically with the increasing of the
number of samplesM or sensing period τs. Lower value of Pf is desired to obtain higher
spectrum opportunity. Higher spectrum opportunity is determined with the higher
access probability which is achieved in a shorter sensing period. erefore, the reduction
of Pf regarding sensing period τs is analyzed at rst between the DS and SS mechanism
to estimate the ecient detection mechanism in terms of spectrum opportunity.
Fig. 3.5 illustrates the characteristic of Pf against the variation of τs in the range of
0 < τs < Tf . For a given Pd = 0.9, the corresponding PfSS and PfDS are computed
by using (3.13) and (3.24), respectively. To compute the PfDS , it is also required to
calculate Pf1 and Pf2 which is measured by using (3.29). In overall, Pf1 , Pf2 , PfDS , and
PfSS decrease monotonically with the increasing of τs as depicted in Fig. 3.5. It is also
observed that the PfDS is much lower than the PfSS . When τs increases towards Tf
(sensing takes place in the full time length of a frame), only then PfDS and PfSS are
reduced closely in the range of 0 < Pf < 0.05. However, lower sensing period reveals
higher transmission period, (Tf − τs) which is also constrained the design to obtain a
possible lowest value of Pf in the shorter τs. In these circumstances, the proposed DS
mechanism is beer than the SS mechanism as PfDS < PfSS when τs is relatively low.
e eectiveness of the DS mechanism over the SS mechanism is emphasized by Fig.
3.4 and Fig. 3.5. It is also important to intensify the insight of the DSmechanism towards
Pf reduction at a given Pd. Fig. 3.6 provides the explicit description regarding- how the
PfDS deceases with the aid of its underlying Pf1 and Pf2 . Based on equations (3.23) and
(3.22), the PfDS is computed by using (3.24) as a function of Pd1 when the simulation
parameters are Pd = 0.99, γ = −15 dB, τs = 2 ms, Tf = 10 ms, and fs = 6 MHz.
Fig. 3.6 shows that with the increasing of Pd1 , Pf1 and Pf2(1−Pf1) have the exponential
increasing and exponential decreasing property, respectively, with an intersecting point.
Due to an intersection among two underlying functions, the PfDS exhibits a minimum
value in its curvature correspond to the variation of Pd1 while the Pd is xed at 0.99. By
choosing dierent PD into two levels conditionally with dierent sensing period, thus,
3.5 Performance Analysis 53
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Probability of detection at first sensing (Pd1)
Probabilityoffalsealarm
Pf1
Pf2 (1 − Pf1 )
PfDS
Figure 3.6: PfDS vs. Pd1 ; PfDS has a minimum value for an optimum valueof Pd1 .
0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009 0.010
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Sensing time (τs) (sec)
Accessprobability(P
a)
PaDSfor PD = 0.9
PaDSfor PD = 0.99
PaSSfor PD = 0.9
PaSSfor PD = 0.99
Figure 3.7: Access probability vs. sensing time (sec) for Pd = 0.99, 0.9.For a given value of Pd, the PaDS is higher than the PaSS .
the overall PFA is reduced signicantly as well as the target interference protection is
guaranteed.
In Fig. 3.7, the access probability versus sensing period performance of the DS and
SS mechanism is presented for two dierent target PD. e simulation parameters for
this analysis are: Pd = 0.99, 0.9, PH0 = 0.2, γ = −15 dB, Tf = 10 ms, and fs = 6
MHz. It is observed that PaDS and PaSS increase monotonically with the increasing of
54 3.5 Performance Analysis
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
PH0
AccessProbability(P
a)
PaDS, Pd = 0.9
PaDS, Pd = 0.99
PaSS, Pd = 0.9
PaSS, Pd = 0.99
Figure 3.8: Pa vs. PH0 of the DS and SS mechanism for Pd = 0.99, 0.9;For a given value of Pd, the PaDS is higher than the PaSS .
τs. However, when Pd is increased from 0.9 to 0.99, then both PaDS and PaSS achieve
lower value. Nevertheless, for the both cases of Pd, the PaDS is outperformed the PaSS
when the sensing period is relatively short. us, the spectrum opportunity is enhanced
by reducing the Pf within a shorter τs for a higher Pd in the proposed DS mechanism
as shown in Fig. 3.7.
In Fig. 3.8, the access probability of the DS and SS mechanism is evaluated in re-
spect of PH0 to compare the capability of the detection mechanisms. Regardless of
taking into account the PU trac parameters, only the numeric probability range of
the equation (3.9) is considered to demonstrate Pa versus PH0 . e higher value of PH0
means the higher opportunity to the SU for the CR operation. erefore, the capability
is characterized when any detection mechanism achieves higher access probability in
a higher value of PH0 . Fig. 3.8 shows that PaDS is larger than the PaSS in the higher
value of PH0 . Most importantly, when the target PD is increased from 0.9 to 0.99,
then DS mechanism outperforms the SS mechanism in a wide margin in terms of access
probability as depicted in Fig. 3.8.
3.6 Chapter Summary 55
3.6 Chapter Summary
is chapter presents a comprehensive analysis of how the dual-level sensing mech-
anism impacts on improving the spectrum opportunity for CR operation. Before the SU
transmission, the capacity measurement of the spectrum opportunity is essential for set-
ting an ecient access protocol. erefore, spectrum opportunity is expressed using the
underlying parameters of the spectrum sensing to gain insights into sensing in order to
increase the spectrum opportunity. On the other hand, interference protection to the PU
is provided by seing a high target PD value in the detectionmechanismwhich increases
the PFA. e detection mechanism cannot produce greater spectrum opportunity when
the detection has a large PFA. is dilemma cannot be handled eciently when an SS
mechanism is applied. To overcome this issue, a DS mechanism is proposed based on
two conditional sensing steps during the sensing period to jointly decide the channel
occupancy status.
In the proposed DS mechanism, two dierent target PDs are set over the sensing
period where those two steps also achieve the overall PD. When the target PD is relat-
ively low, then the detection mechanism produces a lower PFA. As a result, the spectrum
opportunity is increased for SU transmission with the lower PFA. e detection capab-
ility of the DS and SS mechanisms is assessed with the ROC curve for dierent SNR
values. Also, the achieved opportunity is examined by measuring the access probability
in relation to the sensing period and the channel idleness. e ROC curve analysis
implies that the proposed DS mechanism has beer signal detection capability than
the SS mechanism at any SNR value. Access probability analysis shows that the DS
mechanism also achieves higher spectrum opportunity than the SS mechanism, even
when the PU is greatly protected with a high value of the target PD. Overall, this chapter
concludes with the statement that the proposed DS mechanism discovers a large spec-
trum opportunity which can be capitalized on through an advanced access protocol to
signicantly improve the throughput and interference protection.
CHAPTER 4
Optimization of Dual-level Sensingfor Eicient Utilization in Cognitive
Radio Networks
4.1 Introduction
In cognitive radio networks, secondary users struggle to utilize the spectrum opportun-
ity to its fullest extent while guaranteeing the legacy users protection from secondary
users’ interference. erefore, multi-stage spectrum sensing gained a reputation for
largely protecting the primary users; however, this sensing may require a longer sensing
period which impacts on the reduction of throughput performance. Motivated by this
fact, we develop a dual-level sensing (DLS) based access mechanism whereby the multi-
stage detection sensitivity within a limited sensing period explores higher spectrum op-
portunities, and then utilization of the regarding sensing outcome reduces the collision
rate during the spectrum access. However, an appropriate optimization is required in the
DLS mechanism for selecting the detection sensitivity and the sensing period to achieve
maximum throughput under the constraint of primary user protection. erefore, we
provide an ecient solution for the sensing-throughput trade-o for the DLS-based
access mechanism in this chapter. We prove, through mathematical derivations, that
by allowing a portion of the sensing period to be devoted to reducing the probability
of false alarm, the constraint is still met while transmission opportunity is improved.
Furthermore, numerical analysis reveals that the proposed solution algorithms can max-
imize the achievable secondary throughput signicantly within a limited computational
complexity.
58 4.1 Introduction
4.1.1 Motivation
Cognitive radio networks (CRNs) aim atmaximizing throughput while avoiding interfer-
ence to the primary network. ewhole time frame consists of sensing and transmission
operations, and the data rate achieves through the transmission depending on sensing
decision in that whole frame is referred to as throughput of the secondary network.
Previous research focused on optimizing sensing and transmission techniques at the
physical layer (PHY) [3, 51, 53, 58] and showed that throughput maximization and in-
terference reduction are conicting criteria. Interference reduction is based on sensing
the presence of PUs during a short sensing period designed to meet a target probability
of detection (PD) of PUs. roughput is increased when the sensing period is short,
allowing for a longer transmission period. A short sensing period, however, also leads
to a higher probability of false alarm (PFA) (i.e. detecting the presence of a PU where no
PU is present), therefore limiting transmission opportunities and reducing throughput.
is issue is referred to as sensing-throughput trade-o [3].
roughput can also be optimized by improving access techniques at medium access
control (MAC) layer. Even though throughput optimization techniques at PHY and
MAC layers have evolved independently, some MAC-based protocols for throughput
optimization also rely on sensing and can be considered cross-layer based protocols
[6, 63–65]. e above-mentioned techniques, however, consider that all PUs and SUs are
homogeneous, and do not give preference to PUs for access purposes. At the start of each
time frame, they use conventional backo mechanism which improve SU throughput
performance but do not guarantee interference protection to primary network. is
chapter provides a new framework for improving throughput in carrier sense multiple
access (CSMA) protocol by restructuring the sensing period to meet the target probabil-
ity of detection [103]. e proposed protocol was referred to as dual-level sensing based
multiple access (DSMA)where the spectrum sensing is accompaniedwith carrier sensing
to decide the channel status jointly. We illustrated that the throughput improvement
mostly dependent on howmuch the overall PFA can be reducible. e PU protection can
be controlled by choosing the suitable threshold values into two sensing steps to meet
4.1 Introduction 59
the overall target PD which can also control the overall PFA; this also arises sensing-
throughput trade-o problem. erefore, it is of great importance to nd the impact
of the dual-level sensing on the sensing-throughput optimization problem. Motivated
by this importance, we study the feasibility analysis of the optimization problem and
propose an algorithm to maximize the throughput of the dual-level sensing (DLS) based
access protocol under the constraint of the PU protection.
4.1.2 Contributions
In this optimization problem, the main goal is to maximize the throughput of the DLS-
based access protocol under the constraint of overall target PD. As the internal opera-
tion of DLS mechanism is conditioned with each other by the sensing decision [103],
therefore, any one of the sensing steps’ PD and sensing period can be relevant to use
as optimizer to meet the target PD. To emphasize the reduction of PFA of DLS-based
access protocol [103] over single-level sensing (SLS) method, we set the constraint of
the total sensing period equal to the optimal sensing period of SLS mechanism [3, 51]
for a given target PD. In this condition, we check the feasibility of minimum PFA by
convex analysis with respect to the rst detector’s PD for the constraints of target PD
and total sensing period. rough the comprehensive feasibility analysis, we proved that
DLS-based access method reduced the overall PFA compared to SLS-based method at the
same target PD regardless of extending the overall sensing period.
e feasibility analysis also determines the impact of minimum PFA on through-
put minimization problem. However, it is dicult to nd a closed-form mathematical
equation of minimum PFA due to the mathematical complexity for multi-level sensing
case [65, 109]. erefore, we proposed a semi-analytical method to nd the boundaries
for PFA minimization and applied the boundaries into the steepest descent method to
calculate the optimal solution. en we maximize the throughput by optimizing the
sensing period and the PD of rst sensing level to meet the overall target PD. For the
performance comparison and validation, we additionally develop a line search algorithm
with considerable complexity to solve the joint optimization problem. We showed that
60 4.2 System Model
Spectrum
sensingSU data transmission
c
τ f ds
T τ−
f
T
Carrier
sensing
PU Busy PU Idle PU Busy . . .PU Idle
s
τ
ds
τ
Figure 4.1: MAC frame format and time slot operation of the proposed DLS-based access protocol.
the shorter boundary reduces the computation complexity signicantly than the direct
usage of the numerical method in solving the optimization problem jointly. Finally,
the numerical analysis validates the results of developed algorithms and indicates that
proposed algorithm can achieve ecient throughout which outperforms the state of the
arts of the CRN.
4.2 System Model
e SUs are organized in a wireless local area network (WLAN) with multiple access
functionality as shown in Fig. 4.1. e SUs use clear channel assessment (CCA)1function
to detect the presence of ongoing transmission in the channel. In our proposed sensing
model, the CCA module is composed of spectrum sensing (SS) and carrier sensing (CS)
[103]. e SS refers to the ability of the SU to detect the energy level present on the
channel based on the noise oor. e CS refers the ability to detect and decode an
incoming signal on the channel. e CCA reports the channel as busy when a signal
is detected through a combination of spectrum sensing and carrier sensing.
e MAC frame length Tf is designed as shown in Fig. 4.1, where τs period is used
for SS and τc period is allocated for contention-based access. e dierence between the
conventional CR frame [5]-[7] and our proposed frame [103] is that a sensing period τs
designed to specically protect the PU is added to conventional carrier sensing τc; thus
1e CCA is a carrier sensing functionality in WLAN system. e term carrier sensing is oen
considered as equivalent to CCA. e CCA is physical carrier sensing which listens the received signal
on the radio interface. Here, carrier sensing is specically used to only imply a certain type of CCA in the
context of wideband spectrum sensing.
4.2 System Model 61
total sensing period is τds = τs + τc.
4.2.1 Throughput of Dual-Level Sensing Based Access Protocol
e overall probability of detection and false alarm in the dual-level sensing mechanism
in [103] was derived as,
Pd(ε1, ε2, τs, τc) = Pd1(ε1, τs) + (1− Pd1(ε1, τs))Pd2(ε2, τc) (4.1)
Pf (ε1, ε2, τs, τc) = Pf1(ε1, τs) + (1− Pf1(ε1, τs))Pf2(ε2, τc) (4.2)
Using equation (3.23) and (3.7), Pf1 and Pf2 are expressed as follows,
Pf1 (Pd1 , τs) = Q(√
2γ + 1Q−1(Pd1) + γ√τsfs
)(4.3)
Pf2 (Pd, Pd1 , τc) = Q(Q−1
(Pd − Pd11− Pd1
)+√τcfsγ
)(4.4)
For the system parameter of γ, fs, and by considering the relationship between τs and
τc, the overall PFA can be derived as,
Pf (Pd, Pd1 , τs, τds) = Q(√
2γ + 1Q−1(Pd1) + γ√τsfs
)+(
1−Q(√
2γ + 1Q−1(Pd1) + γ√τsfs
))×Q
(Q−1
(Pd − Pd11− Pd1
)+√
(τds − τs)fsγ)
(4.5)
According to DLS-based access protocol, the overall throughput of the secondary net-
work is obtained [103] as,
R = (Pi (1− Pf )CH0 + Pb (1− Pd)CH1) ×Kφ (1− φ)K−1 · Tf − τdsTf
(4.6)
where CH0 and CH1 are the capacity of the secondary link at H0 and H1 cases respect-
ively, and φ is the transmission probability during the contention access period. Ac-
cording to [64, 66, 103], φ depends on the parameters of backo mechanism (contention
62 4.2 System Model
window size and backo stage which are not directly related to sensing mechanism
according to [66]) and collision probability. For cognitive radio network, the channel
collision probability is addressed with the resultant of missed detection which is Pb(1−
Pd). At a given K and Pd = Pd, thus, φ and Pb(1 − Pd) is quite constant and do not
impact on the change of achievable throughput. Only Pf and τds are then impact onR.
Note that overall PFA is a function of Pd1 , τs, and τds according to (4.5) when Pd = Pd.
In this circumstances,R is addressed by aggregated throughput [64] with the following
simplied form,
R (Pd1 , τs, τds) = Pi · (1− Pf (Pd1 , τs, τds)) · CH0 · P aMAC ·
Tf − τdsTf
(4.7)
where we assume that P aMAC = Kφ (1− φ)K−1
. We adopt the normalized aggregated
throughput (R = R/CH0) as follows,
R (Pd1 , τs, τds) = Pi · P aMAC · (1− Pf (Pd1 , τs, τds)) ·
Tf − τdsTf
(4.8)
4.2.2 Problem and Strategy Formulation
Cognitive radio network is congured by the target probability of detection Pd which
restricts the interference to the primary network. In single-level sensing case [3, 53, 58],
for a given Pd, longer the sensing time τs,s, the shorter the available data transmission
time (Tf − τs,s) but higher the spectrum opportunity. Under the constraint of Pd ≥ Pd,
the sensing-throughput trade-o can be solved with the optimal sensing duration τ ∗s,s to
maximize the throughput as shown in [3]. Let us recall single-level sensing optimization
[3], the optimal sensing period in the minimization of probability of false alarm for given
target probability of detection (Pd) can be obtained from
τ ∗s,s = argmin
Pd
Pf,s(τs,s) (4.9)
where Pf,s(τs,s) = Q(√
2γ + 1Q−1(Pd)− γ√τs,sfs
). One of the objectives of DLS
mechanism is to enhance the spectrum opportunity by reducing the PFA than the single
4.2 System Model 63
level mechanism for the same target probability of detection (Pd). erefore, we will
set the upper bound as τds ≤ τ ∗s,s. Considering the equality constraint of τds = τ ∗s,s, R
is then as a function of Pd1 and τs based on (4.8). Maximization of R requires only the
optimization of a function with two variables, Pd1 and τs. e solution of the throughput
maximization problem for dual-level sensing based access protocol [103] is the signic-
ant improvement where the optimization problem is
OP1 : Maximize R (Pd1 , τs) (4.10)
s.t. Pd1 ∈ (0, Pd), Pd ≥ Pd (4.11)
s.t. 0 < τs < Tf , τds ≤ τ ∗s,s (4.12)
e optimization of R for dual-level sensing case is more complicated than the single-
level case as there are a lot of mathematical complexity in deriving the Hessian matrix of
the R (Pd1 , τs). erefore, we solve the optimization problem (OP1) with the following
strategies:
Step 1
In the rst step, we investigate the impact ofPf minimization in throughput optimization
for constant sensing period (thereby Pf is only function of Pd1). Our rst objective is to
analyze the feasibility of minimum PFA Pfmin with respect to Pd1 . us, the optimization
problem is given by (4.13),
OP2 : Minimize Pf (Pd1) (4.13)
s.t. Pd ≥ Pd; Pd1 ∈ (0, Pd) (4.14)
e solution of OP2 is described as Solver 1 that produces a Pfmin at optimal Pd1 .
Step 2
According to equation (4.8), the normalized throughput is also as a function of sensing
period τs when τds = τ ∗s,s. erefore, we also check the feasibility of maximum R with
64 4.3 Minimization of Overall PFA
respect to τs for any constant Pd1 which is less than Pd. Once the feasibility analysis will
be done for both of the controlling parameters Pd1 and τs, we compute the maximum R
with Solver 2.
Step 3
As there are huge mathematical complexity in deriving the Hessian matrix of the
R(Pd1 , τs), we deploy a semi-analytical method to solve the OP1. Step 1 and Step
2 will characterize the R maximization analytically in terms of optimal Pd1 and τs. In
particular, both the above steps set the boundary of optimalPd1 and τs under the equality
constraint of Pd = Pd and τds = τ ∗s,s. Aer obtaining the boundaries, we compute and
analyze the joint optimization of OP1 by a numerical method.
4.3 Minimization of Overall PFA
In this section, we check the feasibility of minimum PFA in order to solve OP2 as
instructed through equation (4.13). e analysis is done under the range of Pd1 ∈ (0, Pd)
where the sensing time τs is assumed to be unchanged. Under this assumption, Pf can
be expressed as,
Pf (Pd1) = Pf1(Pd1) + (1− Pf1(Pd1))Pf2(Pd1) (4.15)
where the overall PFA is entirely expressed as a function of Pd1 . Equation (4.15) implies
that Pf is the summation of two terms of Pf1 and (1− Pf1)Pf2 . erefore, rst of all,
we analyse the convexity and/or concavity property of both of the terms individually.
Once the convexity and/or the concavity property of Pf is completely dened with its
boundary values then we will conduct the algorithm to compute Pfmin .
4.3.1 Feasibility Analysis of PFA Minimization
In this subsection, we estimate the optimal range of Pd1 for which there exists at least
one feasible value of Pfmin . To do that, we examine the property of every elements of
4.3 Minimization of Overall PFA 65
Pf (Pd1) as described below:
Proposition 4.1: For a given constant value τs, fs and γ, Pf1 is a convex function for
the range of Pd1 ∈ (0, Pd1(θ1)), where Pd1(θ1) = Q(−√τsfs(2γ + 1)/2).
Proof: For a given value of τs, fs and γ, we take the rst dierentiation of Pd1(ε1)
and Pf1(ε1) with respect to ε1, and obtain as follows,
dPd1dε1
= − 2√Ns
σ2w
√π(2γ + 1)
exp
[− Ns
2γ + 1
(ε1σ2w
− γ − 1
)2]
(4.16)
dPf1dε1
= −2√Ns
σ2w
√πexp
[−Ns
(ε1σ2w
− 1
)2]
(4.17)
Now assuming wd1 =√Ns/(2γ + 1) (ε1/σ
2w − γ − 1) and wf1 =
√Ns (ε1/σ
2w − 1),
(3.7) and (3.23) become Pd1 = Q(wd1) and Pf1 = Q(wf1). Let us divide (4.17) with (4.16)
to obtaindPf1dPd1
and taking the twice dierentiation of Pf1 with respect to Pd1 , it becomes
as follows,
d2Pf1dP 2
d1
= −2√Ns
σ2w
(√2γ + 1wf1 − wd1
)× exp
[w2d1− w2
f1
] dε1dPd1
(4.18)
Puing the value ofdPd1dε1
from equation (4.16) into above equation, the (4.18) becomes,
d2Pf1dP 2
d1
=√π(2γ + 1)
(√2γ + 1wf1 − wd1
)× exp
[2w2
d1− w2
f1
](4.19)
Ifd2Pf1dP 2
d1
> 0 then Pf1 would be convex function of Pd1 . To obtain that, we need to satisfy
the following property:
Property 4.1: For any given non-negative value of Ns, σ2w, and γ, wd1 is always less
than wf1 by which the inequalityd2Pf1dP 2
d1
> 0 is being satised.
Explanation 4.1: Let us assume that Φ1 = wf1 − wd1 and take the rst partial
dierentiation of Φ1 with respect to ε1,
∂Φ1
∂ε1=
1
σ2w
√Ns
2γ + 1
(√2γ + 1− 1
)> 0 (4.20)
66 4.3 Minimization of Overall PFA
Furthermore,
limε1→0
Φ1 =
√Ns
2γ + 1
(√γ2 + 2γ + 1−
√2γ + 1
)> 0 (4.21)
Equations (4.20) and (4.21) imply that for any given non-negative values of Ns, σ2w, and
γ, the Φ1 is always non-negative. Moreover, Q(wf1) < Q(wd1) in our system design.
erefore, wf1 > wd1 from the fact thatQ(x) is a decreasing function of x; hence wf1 −
wd1 > 0 for any given non-negative value of Ns, σ2w, and γ.
Now assume that Φ2 =√
2γ + 1wf1 −wd1 and take the rst partial dierentiation of
Φ2 with respect to Pd1 ,
∂Φ2
∂Pd1= −γ
√π exp
[w2d1
](4.22)
Furthermore Φ2 is bounded with
limPd1→0
Φ2 = +∞ and, limPd1→1
Φ2 = −∞ (4.23)
Equations (4.22) and (4.23) imply that Φ2 is decreasing function with respect to Pd1 . For
the given range of Pd1 ∈ (0, 1), Φ2 is not always non-negative. Hence, the range of Pd1
for whichΦ2 residing non-negative can be found from√
2γ + 1wf1−wd1 > 0 inequality.
From this inequality, we can say that for Pd1 < Q(−√Ns(2γ + 1)/2), the Φ2 is always
non-negative; hence,d2Pf1dP 2
d1
> 0. So, the upper limit of Pd1 is
Pd1(θ1) = Q
(−√τsfs(2γ + 1)
2
)(4.24)
According to Property 4.1, for the range of Pd1 ∈ (0, Pd1(θ1)), the second derivative
of Pf1 with respect to Pd1 is always non-negative. us, Pf1 is strictly convex for all
Pd1 ∈ (0, Pd1(θ1)) which proves Proposition 4.1.
Remark 4.1: Proposition 4.1 implies that Pf1 is convex function in terms of Pd1 .
Likewise, we can say that (1 − Pf1) is strictly concave function for the range of Pd1 ∈
(0, Pd1(θ1)) (illustration in page 67 at [110]). Nevertheless, if we want to show that the
4.3 Minimization of Overall PFA 67
product of (1− Pf1) and Pf2 is convex/concave we must show that (1− Pf1) and Pf2 is
log-concave or log-convex. e product of two concave/convex function is not always
a concave/convex function but the product of two log-concave/log-convex function is
always log-concave/log-convex respectively (see Section 3.5 of [110] for further explan-
ation). erefore, we now prove that (1 − Pf1) and Pf2 both exhibit the log-concave
property with regards to Pd1 throughout the following propositions and then we extract
the boundary value for the concavity property of their product.
Proposition 4.2: For the same assumption of proposition 1, (1 − Pf1) is a strictly
log-concave function for the range of Pd1 ∈ (0, Pd1(θ2)) where
Pd1(θ2) = Q(
(1/2γ√
2π)− (√τsfs(2γ + 1)/2)
)Proof : e proof is provided in Appendix A.1.
As the matched-lter [44] can be used in the second step of the proposed DLS strategy
so that the property of Pf2 exploiting the parameters of the MF is characterized as
well. Furthermore, we assess the property of Pf2 with respect to Pd1 while the ED is
employed in the second sensing for making a comparison between ED-ED and ED-MF
combination. e following propositions characterize the property of Pf2 with respect
to Pd1 .
Proposition 4.3: When the ED is applied in the second sensing then for a given con-
stant value ofNs, σ2w, and γ, Pf2 is a concave function for the range of Pd1 ∈ (0, Pd1(θ3))
where Pd1(θ3) =(Pd − Pd1(θ1)
)/ (1− Pd1(θ1)).
Proof : e proof is omied due to the similarity to that of proposition 4.1.
Proposition 4.4: When the ED is applied in the second sensing then for a given con-
stant value ofNs, σ2w, and γ, Pf2 is a log-concave function as well for Pd1 ∈ (0, Pd1(θ5)),
where
Pd1(θ5) =(Pd − Pd1(θ4)
)/ (1− Pd1(θ4))
Pd1(θ4) = Q(−(√
2γ + 1/2γ)(γ√Ns + (1/
√2π))
)
68 4.3 Minimization of Overall PFA
Proof : e proof is provided in Appendix A.2.
Proposition 4.5: When ED-MF combination is applied then for the same assumption
of Proposition 4.4, Pf2 is a log-concave function with respect to Pd1 .
Proof : e proof is provided in Appendix A.3.
4.3.2 Discussion on Feasibility Analysis
Based upon the proved propositions, we found thatPf2 and (1−Pf1) both are strictly log-
concave function for 0 < Pd1 < Pd1(θ5); thus, their product also maintains the concavity
property up to that limit. However, the log-concavity of (1− Pf1) is bit extended up to
Pd1(θ2) than Pf2 . So, the overall Pf (Pd1) is now as a function of summation of a convex
(Pf1) and a concave (Pf2(1 − Pf1)) function for 0 < Pd1 < Pd1(θ5). According to [111],
there exists a minimum value of Pf in the span of 0 < Pd1 < Pd1(θ5). Beside that, from
Pd1(θ5) to Pd1(θ2), the functionality of the product of Pf2 and (1−Pf1) could not be well
dened whether it is concave or convex. Moreover, it could be an ane function and an
ane has the both possibility to be a concave and convex function. If Pf2(1 − Pf1) is
convex duringPd1(θ5) < Pd1 < Pd1(θ2) span then simplywe can evaluate that minimum
value of the overall PFA lies in this interval as all the terms in this interval is convex. As
we could not dene robustly the characteristic of PFA up to the upper bound of Pd1 so
by considering the worst case we assume that the term Pf2(1 − Pf1) remains concave
until the upper bound of Pd1 .
Fig. 4.2 illustrates the PFAminimization problemwhere we can see that the minimum
value of Pf is situated between Pd1(θ5) and Pd1(θ2). However, the feasibility analysis of
the product of Pf2 and (1− Pf1) cannot decide exactly whether it is convex or concave
or ane from Pd1(θ5) to Pd1(θ2). As we described above, we assume the product of Pf2
and (1−Pf1) remains unchanged in their concavity property until Pd1(θ2) for the worst
case situation. Nowwe apply the proposed PFAminimization algorithm in the following
subsection to obtain the optimal value of Pd1 .
4.3 Minimization of Overall PFA 69
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Probabability of detection (Pd1)
Probabilityoffalsealarm
Pf1
(1 −Pf1)Pf2
PF
Pd1(θ1)
Pd1(θ2)
Pd1(θ5)
Figure 4.2: Illustration of the PFAminimization problemwith the character-izing of Pf1 ,(1 − Pf1)Pf2 , and Pf corresponding to Pd1 , wherethe simulation parameters are, γ = 0 dB, Ns = 2, Pd = 0.95,σ2w = 1.
4.3.3 Algorithm of PFA Minimization
Recall equation (4.15) and transform into as follows,
F (x) = F1(x) + F2(x) (4.25)
where x, F1, F2, and F represent Pd1 , Pf1 , Pf2(1−Pf1), and Pf , respectively. For a given
value of Ns, both F1 and F2 are nite function and hold their convexity and concavity
properties on the space of En = [xl, xu] where xl and xu are the lower limit and upper
limit of x. Hence, F (x) is a continuous and quasi-dierentiable onEn, and its dierential
matrix can be obtained as the following functions,
DF (x) =
[∂F1(x)
∂x,∂F2(x)
∂x
](4.26)
Now the OP2 can be rewrien as,
OP2 : minx∈En
F (x) (4.27)
70 4.3 Minimization of Overall PFA
According to the quasi-dierential calculus, x∗ ∈ En to be optimum point which satises
OP2, while
− ∂F2(x)
∂x
∣∣∣∣x=x∗
⊂ ∂F1(x)
∂x
∣∣∣∣x=x∗
(4.28)
Now, the problem of minimizing F (x) on the spaceEn can be formulated to minimizing
a concave function on a convex set. As the properties of epigraph dened in [110] (see
page 75 in [110] for detail), a function is convex if and only if its epigraph is a convex
set; conversely, this property holds true. e epigraph of convex function F1 is issued
as,
Ω = epi F1 = (x, µ) : F1(x) ≤ µ (4.29)
Let represent the set of epi F1 in the space of z = [x, µ] where z ∈ En × E1; then we
found the following relationship:
Ω = z ∈ En × E1 : h(z) ≤ 0
h(z) ≡ F1(x)− µ ≤ 0
Ψ(z) = F2(x) + µ
Now set Ω is closed with convexity property and new objective function Ψ is quasi-
dierentiable at any z onEn×E1 space. en theOP2 can be re-formulated as following
sub-optimization problem,
SOP2 : minz∈Ω
Ψ(z) (4.30)
Note that if a concave function achieves its minimum value on a convex set, this value
is achieved on the boundary of the set [112]. For example, the minimum value of (1 −
Pf1)Pf2 can be achieve on the epigraph of Pf1 which can be an convex set as depicted in
Fig. 4.2. Moreover, the convex set is constrained with boundary value Pd1(θ5), therefore,
the minimum value can be obtained easily at Pd1(θ5). Using a numerical algorithm, we
can now compute the maximum of the convex function or the minimum of the concave
function. To nd the solution of OP2 it is necessary to nd the condition as driven in
the following theorem.
Theorem 4.1: For an optimal point x∗ to be a solution of OP2, it is necessary that
4.3 Minimization of Overall PFA 71
point c be a solution to SOP2, where µ∗ = F1(x).
Proof : See Appendix A.4.
Simply, we can obtain the optimal x∗ at the crossing point of F1(x) and F2(x) for
minimizing the objective function F (x). However, this crossing point is not the exact
minimum point of the F (x). To avoid the higher-order mathematical approximation,
we adopt the following numerical algorithm to nd an optimal point of x that provides
a closest minimum point of F (x). But before that numerical algorithm, we must check
the feasibility test of the minimization of the F (x) analytically. In that case, we consider
the sub-gradient of F1(x) and F2(x).
Considering that, a point xo ∈ En is a ξ-minimum critical point of the objective
function F on En space if
− ∂
∂xF2(x)
∣∣∣∣x=xo
⊂(∂
∂x
)ξ
F1(x)
∣∣∣∣∣x=xo
(4.31)
where (∂
∂x
)ξ
F1(x) =F1(z)− F1(xo)− ξ
(z − xo)≥ v, ∀x ∈ En, v ∈ En
Now generate g ∈ En and set
(∂
∂g
)ξ
F (x) = maxv∈ ∂
∂xF1(xo)
(v, g) + minw∈ ∂
∂xF2(xo)
(w, g) (4.32)
Theorem 4.2: For a point xo to be a ξ-minimum critical point of the function F on
xo ∈ [xl, xu], it is necessary that
min‖g‖=1
(∂
∂g
)ξ
F (xo) ≥ 0
Proof : See Appendix A.5.
Theorem 4.3: If ξ > 0 then the function max(v, g), where v ∈ (∂/∂x)ξF1(x), is a
continuous in x ∈ [xl, xu] for any xed g ∈ En.
Proof : See Appendix A.6.
e following algorithm Solver 1 nds the optimal Pd1 for PFA minimization.
72 4.4 roughput Maximization
1. Given parameters: σ2w, Ns, γ, Pd
2. Calculate the upper and lower bound as xu = Pd1(θ2) and xl = Pd1(θ5)
3. Set iteration number i = 1, initial start point x0 = xl, and convergence criteria Λx
4. Compute the search direction gξi = − (voξ + wo) / (‖voξ + wo‖)
5. Estimate the optimal step size α∗i = arg minα≥0 F (xi + αgξi)
6. en set xi+1 = xi + α∗i gξi
7. If
∣∣∣F (xi+1)−F (xi)F (xi)
∣∣∣ ≤ Λx, then stop the iteration
8. Otherwise, take i = i+ 1 and go to step 4
4.4 Throughput Maximization
In the previous section, the optimal Pd1 has been obtained for minimizing the PFA under
the equality constraint of Pd = Pd and τds = τ ∗s,s. For a constant sensing period which
is nothing but the xed number of sampling, we have proved and estimated that there
exists a feasible and global minimum value of Pf in the range of Pd1 ∈ (0, Pd). For the
same assumption, it can be stated that obviously R will show the converse properties
as PFA shows with respect to Pd1 . In this section, we will rst check feasibility of
the optimal sensing period to maximize the normalized throughput as instructed in
Step 2 and Step 3 of the optimization strategy. We assume that the secondary user
transmits over τx period where τds + τx = Tf . As we want to optimize the throughput
corresponding to rst sensing operation so sensing period τs of the rst sensing will be
led as the variable in this analysis. Let us recall equation (4.8) and reform as follows,
R = PiPaMAC (1− Pf )
Tf − τdsTf
=PiP
aMAC (1− Pf ) τx
τx + τs + (τds − τs)Pf1(4.33)
Above equation indicates that R can be maximized while the denominator of the equa-
tion is minimized. By turning R maximization into the convex problem at given value
4.4 roughput Maximization 73
of Pd1 and Pd, the new optimization problem is given below
OP3 : min R0(τs) = τs + (τds − τs)Pf1(τs) (4.34)
st. 0 < τs < Tf , τds ≤ τ ∗s,s
We conduct the following propositions to solveOP3 towards proposing semi-analytical
algorithm for OP1.
Proposition 6: For a given Pd1 , Pd, and τ ∗s,s, R0 is a continuous and convex function
with respect to τs.
Proof : See Appendix A.7.
Proposition 7 R0 is a convex function of τ under Pf1 ≤ Q(√π/2 + Q−1(P ∗d1)),
where τ =√
2γ + 1Q−1(P ∗d1) + γ√τsfs.
Proof : Proof is given in Appendix A.8
Remark 2: Proposition 6 indicates that R0 has a feasible minimum in the range of
0 < τs < Tf which ultimately promotes Rmaximization. In addition, Proposition 7 set
the tone about convergence criteria in developing Solver 2 to nd the optimal sensing
period toward throughput maximization. e proposed algorithm Solver 2 is described
below:
1. Given parameters: σ2w, fs, γ, Pd
2. Calculate the τ ∗s,s for single level sensing at Pd
3. Take the lower and upper limit of τs(= y) as yl = 0 and yu = τ ∗s,s
4. Solve the equation R′0(y) = 0 using Bisection method and obtain the optimal
value yopt = τ ∗s
4.4.1 Joint Optimization with Numerical Analysis
Since it is dicult to obtain the joint optimal sensing time and probability of detection
at rst sensing (τ ∗s , P∗d1
) of OP1 by a complete analytical solver. erefore, we develop
a semi-analytical model to solve the optimization problem by several steps. Firstly,
74 4.4 roughput Maximization
we check the feasibility of an optimal Pd1 analytically by which we can ensure the
existence of an unique minimum value of Pf . Later, the optimal P ∗d1 is obtained by
Solver 1. Secondly, we show that over the entire range of τs, the optimal point Pd1
remain same where the other parameters remain unchanged. However, there might be
several minimum operating curve of Pf corresponding to τs at the P∗d1. erefore, lastly,
we check the feasibility of maximum R in terms of the sensing period and obtain the
optimal solution through Solver 2.
Algorithm 1 Line Search Algorithm For Joint Optimization
Require: Step size of Pd1 = tp, step size of τs = tτ ,X =⌈Pdtp
⌉, Y =
⌈Tftτ
⌉, α ∈ (0, 0.5),
β ∈ (0, 1), i = 0Ensure:1: while i ≤ X do2: i = i+ 1; t
(i)p = 1; R(i) = R(P
(i)d1, τ ∗s,s);
3: ∆R = α R′(P (i)d1
; ∆Pd1);
4: obtain R(i+1) = R(P(i)d1
+ t(i)∆Pd1 , τ∗s,s);
5: if R(i+1) < R(i) + t(i)p ∆R then
6: t(i+1)p = β t
(i)p ;
7: R(i+1) = R(P(i)d1
+ t(i+1)∆Pd1 , τ∗s,s);
8: else9: R(i+1) = R(P
(i)d1
+ t(i)∆Pd1 , τ∗s,s);
10: end if11: end while12: while i ≤ X + Y do13: τ
(i)s = (i−X)Tf ; t
(i)τ = 1; R(i) = R(Pd, τ
(i)s );
14: ∆R = α R′(τ (i)s ; ∆τs);
15: obtain R(i+1) = R(τ(i)s + t
(i)τ ∆τs, Pd);
16: if R(i+1) < R(i) + t(i)τ ∆R then
17: t(i+1)τ = β t
(i)τ ;
18: R(i+1) = R(τ(i)s + t
(i+1)τ ∆τs, Pd);
19: else20: R(i+1) = R(τ
(i)s + t
(i)τ ∆τs, Pd);
21: end if22: i = i+ 1;23: end while24: return (P ∗d1 , τ
∗s ), R(P ∗d1 , τ
∗s )
In this section, we develop a line search based numerical method to obtain (P ∗d1 , τ∗s )
jointly as given in Algorithm 1. Firstly, we generate the feasible Pd1 into X discrete
data, i.e., Pd1(1), . . . , Pd1(X), and the feasible τs in Y discrete data, i.e., τs(1), . . . , τs(Y ).
4.5 Numerical Results and Discussion 75
Secondly, we consider all Pd1(1 ≤ i ≤ X) by satisfying Pd1 ≤ Pd into the search sub-
algorithm where the backtracking line search method2is applied. In this sub-algorithm,
the step size t (where t ≥ 0) is updated with convergence criterion R(P(i+1)d1
) >
R(P(i)d1
) + tiαR′(P (i)d1
; ∆Pd1) where ∆Pd1 is the search direction. is backtracking line
search will be stop while the updated step size is less than 10−3. In this way, we nd a
suboptimal root of Pd1 at τ∗s,s(Pd) which results RPd1 = max1≤i≤XR (i). Likewise, we
can obtain the Rτs = max1≤i≤Y R (i) for suboptimal τs. irdly, we obtain the maximum
utilization R = max
RPd1 , Rτs
which corresponds to the optimal set of (P ∗d1 , τ
∗s ).
e proposed line search method is much simpler than the exhaustive search method,
a short enough step size is required to reach the optimal set of controlling para-
meters. However, our proposed semi-analytical algorithm (combination of Solver
1 and Solver 2) has less computational complexity compared with the line search
algorithm. e computational complexity of the proposed line search algorithm is
O(X ln
PdΛPd1
)+O
(Y ln
TfΛτs
). On the other hand, the proposed semi-analytical method
requires O(X ln
∣∣∣Pd1 (θ2)−Pd1 (θ5)
Λτs
∣∣∣) + O(Y ln
τ∗s,sΛτs
)computational complexity which is
much less than line search method as Pd1(θ2) − Pd1(θ5) < Pd and τ ∗s,s < Tf . e
proposed both methods have less computational complexity than the exhaustive search
method as O(XY ) computational complexity is required for nding the optimal roots
in the exhaustive search method.
4.5 Numerical Results and Discussion
In this section, the model validation of the optimization algorithm is presented with
the numerical results. Moreover, the performance of the DLS-based access protocol
is evaluated incorporating with the post-optimization data of the proposed solution.
e proposed algorithm is implemented in the MATLAB simulation environment and
acquainted data is presented herein including their graphical representation. Firstly, the
2Backtracking line search is the most eective and simple line search method where the step size is
incorporated with two constants α, β with 0 < α < 0.5, 0 < β < 1. is algorithm starts with initial
step t0 and stops based on the given tolerable range particularly that satises t ∈ (βt0, t0].
76 4.5 Numerical Results and Discussion
optimal set of the design parameters, i.e., Pd1 and Ts, has been identied with the mesh
plot of the developed analytical model. Secondly, the optimal Pd1 and τs is measured
by using Solver 1 and Solver 2, and compared with the outcome of Algorithm 1.
And lastly, the normalized throughput based on the optimization solution is analyzed
corresponding to several system parameters.
4.5.1 Throughput Optimization and Model Validation
Towards the maximization of the throughput, the overall PFA is minimized at rst ac-
cording to Step 1 of the proposed solution strategy. e requirements of feasibility
analysis and the description ofOP2 is illustrated by Fig. 4.3, where the entire variation of
Pf for all the simulation range ofPd1 and τs is captured. In Fig. 4.3, it is clearly noticeable
that the objective function of OP2, Pf (Pd1 , τs), has denitely a minimum value for an
optimal set of Pd1 and τs which consolidates the rst step of the optimization strategy
towards solving the OP1. e trade-o can be further investigated by explaining the
(b) contour plot of Fig. 4.3, whereby the minimum value of PFA resides in the region
of 0.78 < Pd1 < Pd for the operating range of 0 < τs < 0.85 ms where τ ∗s,s = 1.7 ms
for P = 0.95. is interesting behaviour of the PFA validated the assumption created
(xed the number of sample) during the feasibility analysis of PFAwith respect to Pd1 . In
addition, the optimal sensing time cannot be drawn from this gure as larger τs always
reduce the Pf value. erefore, we conduct the Step 2 and Step 3 to capture the all
variations of the main objective function, R, with respect to τs at optimal Pd1 .
In Fig. 4.4, the entire variation of the R(Pd1 , τs) is captured, where the maximum
throughput can also be noticeable for an optimal set of (Pd1 , τs). e total sensing period
required for DLS based technique, τds, be extracted as a function of τs for a given Pd. For a
fair comparisonwith the benchmark [3, 50, 64, 65], we also describe the R corresponding
toPd1 and τds as shown in Fig. 4.5, where the feasible range of maximum R can be drawn
within a smaller boundary of optimal set of Pd1 and τds.
e model validation of the proposed optimization solutions is carried out through
the comparison between semi-analytical method (Solver 1 and Solver 2) and numerical
4.5 Numerical Results and Discussion 77
0.1
0.1
0.2
0.2
0.3
0.3
0.4
0.4
0.5
0.5
0.6
0.6
0.7
0.7
0.8
0.8
0.9
Pd1
τs
(b) Contour plot
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
x 10−3
0
0.5
1
0
1
2
x 10−3
0
0.2
0.4
0.6
0.8
1
Pd1
(a) Mesh plot
τs
Pf(P
d1,τs)
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Figure 4.3: (a) Mesh plot and (b) contour plot of Pf (Pd1 and τs) where thesimulation parameters are, γ = −15 dB, Pi = 0.9, P aMAC =0.99, σ2
w = 1, fs = 6 MHz, τ∗s,s = 1.7 ms and Tf = 10 ms.
joint optimization method (Algorithm 1). For this validation, we set that simulation
parameters as follows: γ = −15 dB, σ2w = 1, fs = 6 MHz, Tf = 10 ms, Pi = 0.9,
N = 1024,M = 1024. For the three dierent values of Pd, we run the simulation for all
the methods and obtain the relevant outcomes as given by Table 4.1. For a given Pd =
78 4.5 Numerical Results and Discussion
0.10.2
0.2
0.3
0.3
0.4
0.4
0.5
0.5
0.6
0.6
0.7
0.8
Pd1
τs
(b) Contour plot
0 0.2 0.4 0.6 0.8 10
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
x 10−3
0
0.5
1
0
1
2
x 10−3
0
0.2
0.4
0.6
0.8
1
Pd1
(a) Mesh plot
τs
R(P
d1,τs)
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Figure 4.4: (a) Mesh plot and (b) contour plot of R(Pd1 , τs) where thesimulation parameters are, γ = −15 dB, Pi = 0.9, P aMAC =0.99, σ2
w = 1, fs = 6 MHz, τ∗s,s = 1.7 ms and Tf = 10 ms.
0.9, the optimal set of (P ∗d1 , τ∗s ) is obtained as (0.68, 0.45ms) to get maximum throughput
Rmax = 0.7942. For the same Pd, the optimal data is (0.6751, 0.4391ms) to lead Rmax =
0.7836 which is almost similar to the semi-anlytical method. e extracted optimal sets
of Pd1 and τs for rest of the two experiments are also shown closest performance which
4.5 Numerical Results and Discussion 79
0
0.5
1
00.0020.0040.0060.0080.01
0
0.2
0.4
0.6
0.8
τds(τs)(sec)Pd1
R
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Figure 4.5: Mesh plot of throughput corresponding to Pd1 and τds, wherethe simulation parameters are, γ = −15 dB, Pi = 0.9, P aMAC =0.99, σ2
w = 1, fs = 6 MHz, and Tf = 10 ms.
Table 4.1: Numerical results about comparing proposed solution of optimiz-ation problem.
Target PD Semi-analytical Method Joint Optimization
Pd P ∗d1 τ ∗s (ms) Rmax P ∗d1 τ ∗s (ms) Rmax
0.9 0.68 0.45 0.7942 0.6751 0.4391 0.7836
0.95 0.78 0.55 0.7628 0.7881 0.5414 0.7683
0.99 0.9 0.7 0.7223 0.9102 0.6883 0.7169
are denitely validated those models for solving the optimization problem. Furthermore,
we can see that by increasing the Pd up to 0.99 the maximum achievable R is 0.75
which is theoretically true as proven in [3, 50, 59]. From this outcomes, the signicance
of DLS-based access protocol can be drawn. Our proposed DLS-based access protocol
[103] with above conducted optimization is achieved greater throughput performance
than the performance achieved by [3, 50, 64, 65] for relatively higher target probability
of detection which can also provide with stronger interference protection to primary
network.
80 4.5 Numerical Results and Discussion
0 0.2 0.4 0.6 0.8 10.55
0.6
0.65
0.7
0.75
0.8
Probabability of detection at first sensing (Pd1)
Norm
alizedaggregatedthroughput(R
)
For Pd = 0.9
For Pd = 0.95
For Pd = 0.99
Figure 4.6: Characterization of the change of R corresponding to τds(τs) forPd = 0.9, 0.95, 0.99 and its optimal sensing period as givenby Table 4.1, where the simulation parameters are, γ = −15 dB,Pi = 0.9, P aMAC = 0.99, σ2
w = 1, fs = 6 MHz, and Tf = 10 ms.
4.5.2 Performance Evaluation of DLS Based Access with Post-
optimization
Fig. 4.2 already showed that the change of overall PFA with respect to Pd1 which implied
that a minimum value of Pf could be found for the optimal value of Pd1 . Likewise, the
variation of the normalized throughput is contained a certain maximum region with
respect to Pd1 as depicted in Fig. 4.6. For a given Pd, the normalized throughput starts
to increase with the increment of Pd1 and starts to decline aer reaching its maximum
region. As Pd1 was the subset in the space of (0, Pd) so that R cannot be measured for
Pd1 ≥ Pd. Most importantly, this gure shows the potential of the DLS based access
protocol, i.e., the maximum value of throughput laid on the value of Pd1 that is less than
the target Pd. By segmenting the sensing process and seing the Pd1 less than the Pd, we
reduced mainly the sensitivity of the rst detection process which conversely reduced
the probability of false alarm and ultimately maximized the normalized throughput.
Fig. 4.7 illustrates the normalized aggregated throughput versus the total sensing
4.5 Numerical Results and Discussion 81
0 0.5 1 1.5 2 2.5 3
x 10−3
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Total sensing period (τds(τs)) (sec)
Norm
alizedaggregatedthroughput(R
)
For Pd = 0.90, P ∗
d1= 0.68
For Pd = 0.95, P ∗
d1= 0.78
For Pd = 0.99, P ∗
d1= 0.90
Figure 4.7: Characterization of the change of R corresponding to τds(τs) forPd = 0.9, 0.95, 0.99 and its optimal Pd1 , where the simulationparameters are, γ = −15 dB, Pi = 0.9, P aMAC = 0.99, σ2
w = 1,fs = 6 MHz, and Tf = 10 ms.
time required in each frame of the secondary network. e results are extracted from
the analytical model of the DLS based access strategy using equation (4.8) aer obtaining
the optimal Pd1 . is gure reveals that the maximum throughput can be achieved for
a certain optimal sensing period and the throughput will be decreased linearly with the
sensing period aer a while of that optimal sensing time. Moreover, it is seen that the
higher throughput regime is obtained in lower sensing period for the case of lower target
Pd. For example, the normalized throughput closed to 0.8 at the sensing period of 1 ms
for Pd = 0.9 which is larger than the Pd = 0.99 case. is nding follows the actual
character of the throughput versus sensing period as proven by [3, 58, 59] for the CRN
and consolidates the Proposition 6. However, lower value of Pd e.g., Pd = 0.9 found by
seing Pd1 = 0.68 which leads severe collision during channel accessing. On the other
hand, by seing Pd1 = 0.9 for Pd = 0.99, the interference protection can be improved as
well as optimal throughput can be achieved similar with the throughput of the Pd = 0.9
case. In overall, the signicance of the DLS based access protocol can be exposed as
the SU can achieved near about the maximum throughput without producing too much
82 4.5 Numerical Results and Discussion
0 0.5 1 1.5 2 2.5 3
x 10−3
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Total sensing period (τds(τs)) (sec)
Norm
alizedaggregatedthroughput(R
)
RDLS for Pd = 0.9
RDLS for Pd = 0.95
RDLS for Pd = 0.99
RSLS for Pd = 0.9
RSLS for Pd = 0.95
RSLS for Pd = 0.99
Figure 4.8: Throughput performance comparison of the proposed DLSand the conventional SLS based access mechanism for Pd =0.9, 0.95, 0.99, where the simulation parameters are, γ = −15dB, Pi = 0.9, P aMAC = 0.99, σ2
w = 1, fs = 6 MHz, and Tf = 10ms.
interference to primary network within a shorter sensing period.
Fig. 4.8 shows that proposed DLS based access mechanism achieves higher through-
put than the conventional SLS based access mechanism in the lower sensing period at
a certain target Pd. For example, the DLS system is maintained near about 0.8 of the
normalized throughput which cannot be achieved by the SLS system at 1 ms period
while the Pd are 0.9 and 0.95. Also, the dierence of the throughput between the DLS
and the SLS system keeps increasing while the value of the Pd increases at the region of
lower sensing period. Moreover, RDLS for Pd = 0.99 is quite same with the RSLS for
Pd = 0.9. is indicates that the proposed mechanism achieved higher throughput at a
given target Pd and outperformed the conventional SLS mechanism with a large margin
while the target Pd is increased for limiting interference to primary network.
At low SNR value, the proposedDLS based access scheme achieved higher throughput
than the conventional SLS method for any given Pd as illustrated in Fig. 4.9. R in
comparing both systems are maintained quite same characteristic when γ = −10 dB
4.6 Chapter Summary 83
0 0.5 1 1.5 2 2.5 3
x 10−3
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Total sensing period (sec)
Norm
alizedaggregatedthroughput
RDLS for γ = −10 dB
RDLS for γ = −15 dB
RDLS for γ = −20 dB
RSLS for γ = −10 dB
RSLS for γ = −15 dB
RSLS for γ = −20 dB
Figure 4.9: Throughput performance comparison of the proposed DLSand the conventional SLS based access mechanism for γ =−10,−15,−20 dB, where the simulation parameters are,Pd = 0.95, Pi = 0.9, P aMAC = 0.99, σ2
w = 1, fs = 6 MHz,and Tf = 10 ms.
with the increasing of sensing period. Nevertheless, the DLS system showed beer
throughput performance while the SNR value keeps decreasing. For γ = −15 dB, the
maximum R achieved by the DLS system is more than 0.75 which is apparently not
achievable by the SLS system. is analysis depicted that proposed DLS based access
protocol showed excellent throughput performance than the SLS systemwhile spectrum
sensing is impaired by channel’s low SNR value.
4.6 Chapter Summary
is chapter provides the solution of the sensing-throughput optimization problem of
the DLS-based access mechanism. e main goal of the optimization is throughput
maximization of the SU when the DLS mechanism is employed in spectrum access under
the constraint of interference protection to the primary network. e main hurdle of
sensing-throughput trade-o is to detect the active user in the channel with a single
84 4.6 Chapter Summary
step including a single-target PD. e proposed DLS mechanism overcomes this issue.
However, the detection sensitivity such as a target PD in each level with required sensing
time is a crucial design aspect for the deployment of the DLS mechanism. e conducted
optimization provides the solution for this design aspect of the DLS mechanism.
e solution approach to the optimization problem is formulated over three steps.
In the rst step, the PFA is minimized regarding the PD at the rst sensing stage while
other system parameters remain unchanged. Before this minimization, the feasibility of
minimum PFA is analyzed by convex analysis. From the feasibility analysis, the optimal
boundary of the PD at rst sensing is obtained. is PFA minimization ensures the
signicance of the DLS mechanism as the throughput improvement is hugely dependent
on theminimized amount of PFA from the sensing. In addition, the required sensing time
to obtain a decreased PFA also impacts on the throughput performance. erefore, in
the second step, the optimal boundary of the sensing period for a unique value of the
maximum throughput is identied with the analytical model. Due to the mathematical
complexity in computing the Hessian matrix of the objective function regarding the
optimizer, the solution is accomplished with a proposed semi-analytical algorithm. At
the last step, the feasible boundary of the optimizer is employed in the proposed al-
gorithm to obtain the maximum throughput. For fair comparison and model validation,
the proposed algorithm is comparedwith a purely numerical approach. e performance
analysis indicates that the solution algorithms can enhance the throughput performance
optimally under the constraint of PU protectionwithin a limited computational complex-
ity.
CHAPTER 5
Sensing Assisted Multiple AccessStrategy in Cognitive Radio Networks
5.1 Introduction
Multiple access is an essential functionality to enable secondary users to access un-
used spectrum while improving the overall throughput performance of a cognitive radio
network. However, traditional medium access control (MAC) protocols can interfere
with primary users’ transmission and degrades the secondary users’ throughput. In this
chapter, a multiple access protocol is proposed by a PHY/MAC cross-layer design. e
proposed protocol is referred to as a dual level sensing based multiple access (DSMA),
where the sensing is integrated with the MAC-based transmission. An analytical model
of the proposed DSMA mechanism is developed by Markov chain analysis to estimate
the average service time and the normalized throughput. Performance analysis shows
that the proposed scheme improves throughput signicantly when multiple access takes
place under the large sensing error and low signal-to-noise (SNR) conditions.
5.1.1 Motivation
A MAC protocol facilitates the control operation of the spectrum access to execute
the transmission timing. Few studies [34, 49] showed that the MAC can govern the
distribution of sensing task over a certain operational period. Besides, MAC protocol
allows multiple secondary users to access the channel with a higher rate of utilization
through a multiple access functionality [6, 57, 63]. On the other hand, its deployment
consumes the spectrum opportunity for its control operation. e full capacity of the
86 5.1 Introduction
spectrum opportunity is not eciently utilized [36]. As a result, SU cannot achieve
higher throughput. Moreover, the throughput can be reduced due to the waste of spec-
trum opportunity and the packet collision [102]. When the false alarm occurs during
the spectrum sensing, the secondary users lose the access opportunities even though the
spectrum is vacant. On the other hand, missed detection leads to collision eect during
the channel access. If missed detection occurs, not only the throughput is reduced but
also the activity of the PU is interrupted.
In cognitive radio, the access protocol denes its operational mechanism associating
with the spectrum sensing [54, 64]. Without the additional sensing policy, only the MAC
protocol cannot reduce the collision eect occurring due to the missed detection. e
periodical sensing before any data transmission during the access period can reduce the
likelihood of collisions [6, 50, 57, 63]. e purposes of the spectrum sensing are not
only to nd the access opportunity but also to scaling the interference protection to the
PU [3, 37]. Motivated by the above facts, a novel MAC protocol is aimed to develop
in this chapter to increase the throughput concurrently with guaranteeing a strongest
interference protection.
5.1.2 Contribution
To overcome the challenges as mentioned above, the access decision is incorporated
with a dual-level sensing by the PHY/MAC cross-layer design. At the start of the MAC
operation, the full capacity of the spectrum opportunity is determined explicitly by using
energy detection based spectrum sensing. Aer the rst sensing, SUs can access the
channel following the second sensing step which is conditional on the rst sensing
decision. e second sensing is referred to as clear channel assessment (CCA) due to
the compatibility with the existing MAC protocols.
Spectrum access comes to the operation depending on the outcomes of the cross-
layer based detection decision and the backo process. e CCA is somewhat related
and included into the backo process [57, 63, 66, 67]. However, a separated design is
proposed here to change the sensitivity of the target detection according to backo delay.
5.2 System Model 87
In particular, the advantages of backo mechanism towards collision reduction during
multiple access has contributed in scaling the detector. As a result, the detection process
achieves larger opportunity with larger collision probability. By using the contention
process and the RTS/CTS based packet transmission mechanism, the overall collision
eect is reduced during the channel access. A comprehensive analysis is presented to
demonstrate the enhanced detection and the throughput performance at various channel
conditions.
e main contributions of this chapter are as follows:
• Firstly, the DSMA mechanism is proposed by a PHY/MAC cross-layer design
which utilizes the spectrum opportunity and reduces the collision eect concur-
rently.
• Secondly, an analytical model of the DSMA is developed by exploiting Markov
chain analysis, and further extended the packet service process in amultiple access
operation to compute the normalized throughput.
• irdly, a comprehensive assessment is carried out by the model validation and
performance comparison which implies that the proposed mechanism provides
stable throughput regime with a short sensing time and low SNR values.
5.2 System Model
5.2.1 Network Entity
Let us considerN number of SUs are randomly distributed in a cognitive radio network
with the capability of multiple access as shown in Fig. 5.1. SUs use a single time-
sloed channel for data transmission where PU has prioritized to access the channel,
thereby SU can access the channel only while PU is detected as idle. rough spec-
trum sensing mechanism, SUs decide whether PU is idle or busy in the channel. In
our model, upstream transmission from multiple SU to a secondary base station, is
considered through single-hop communication link. To accommodate multiple access
88 5.2 System Model
Listening + multiple access
PU BUSY PU BUSYPU IDLE
PU
SU # 1 SU # 2 SU # K
Figure 5.1: Network Architecture of a cognitive radio network with multipleaccess functionality.
SU data transmission
S
Tf S
T T−
f
T
PU Busy PU Idle PU Busy . . .PU Idle
Spectrum discovery
Figure 5.2: MAC frame format for proposed DSMA mechanism.
during upstream transmission, proposed MAC protocol supports the random schedul-
ing among multiple SU. As PU transmission is not coordinated by the cognitive radio
network, therefore, MAC protocol takes into account the spectrum sensing in upstream
scheduling for detecting the PU transmission. e operational MAC frame format is
given in Fig. 5.2, where Ts period is allocated for the spectrum discovery over the frame
duration Tf . In the remaining Tf − Ts period, the proposed MAC adopts dynamic time
sequence for the packet transmission protocol. e seing of the operational time for
MAC operation is accomplished dynamically by a cross-layer design, where sensing
parameters are integrated with the random scheduling mechanism. Before any packet
transmission, a conditional channel sensing takes place aer the spectrum discovery
in the proposed model. erefore, the access mechanism is referred to as a dual-leve
sensing based multiple access (DSMA) protocol.
5.2 System Model 89
5.2.2 Energy Detection Based Spectrum Sensing
In the proposed model, channel monitoring plays a key role in the design of packet
transmission protocol. When the channel monitoring is required, the spectrum sensing
operation is carried out by using energy detection method. A sensing model is presented
below with its underlying performance metrics.
Let y(m) denotes the received signal to the secondary user for primary user detection
over τs period with sampling frequency fs, wherem is the sampling index; thus, the total
number of sampling isM = bfsτsc. Applying a binary hypothesis-testing problem [1, 3],
the detection process is modeled as,
H0 : y (m) = w (m) (5.1)
H1 : y (m) = h .s (m) + w (m) (5.2)
where s(m) is the transmied signal,w(m) is the additivewhite Gaussian noise (AWGN),
and h is complex channel gain. It is assumed that s(m) and w(m) are independent and
identically distributed (iid) random process with both having the mean zero, and vari-
ance σ2s and σ
2w respectively. Hypothesis H0 and H1 describe the absence and presence
of the PU signal, respectively. e measured signal-to-noise ratio (SNR) under the H1
hypothesis is γ = |h|2 σ2s/σ
2w. e test statistics of the detector is [43]
Y ∼
χ2
2M H0 : PU is absent
χ22M (2γ) H1 : PU is present
(5.3)
where test statistic Y follows a central chi-square (χ22M) distribution with 2M degrees
of freedom for H0, and a non-central chi-square distribution (χ22M (2γ)) with 2M de-
grees of freedom and a non centrality parameter 2γ for H1. e performance of the
detection is evaluated with probability of detection (Pd), probability of false alarm (Pf ),
90 5.3 Proposed Model of the DSMA Protocol
and probability of missed detection (Pm), which are expressed as follows [43],
Pd = P Y1 > ε|H1 = Q(√
2Mγ,
√ε
σw
)(5.4)
Pf = P Y1 > ε|H0 =Γ(M, ε
2σ2w
)Γ (M)
(5.5)
where ε is the threshold value of the energy detector,Q(., .) is a generalized MarcumQ-
function, and Γ(a, b) is an incomplete gamma function given by Γ (a, b) =´∞bta−1e−tdt,
and Γ(a) is a gamma function. Consequently, the probability of missed detection is
measured by Pm = 1 − Pd. For a large number of sampling, it can be shown that the
distribution of the test statistic is normal distribution [3, 43]. By using central limit
theorem, the performance metrics can be expressed in terms of Gaussian Q(.) function
as derived in previous chapter.
e detector must satisfy a given constraint, Pd ≥ P d, to provide the interference
protection to the primary network from the secondary transmission. For instance, if
P d = 0.9, then it indicates that the primary network can tolerate the maximum 10% of
interference (Pm = 1 − Pd = 0.1) from the SU transmission which occurs depending
on the detection. us, the detector should design based on the given constraint. For a
given target probability of detection, P d, the Pf is obtained by
Pf(P d, τs
)= Q
(√2γ + 1Q−1(P d) + γ
√τsfs
)(5.6)
5.3 Proposed Model of the DSMA Protocol
5.3.1 Underlying Mechanisms of Proposed Protocol
e proposed protocol relies on the following mechanisms and their underlying para-
meters:
• Spectrum discovery: It initiates the CR operation by applying spectrum sensing
to nd the vacant channel from the channel of interest (CoI) as shown in Fig.
5.3 Proposed Model of the DSMA Protocol 91
Spectrum
Discovery
Clear
Channel
Assessment
Backoff
Process
Start CR
Operation
Packet
Transmission
Figure 5.3: Block diagram of the Proposed DSMA Mechanism.
5.3. e MAC layer requests the PHY to perform this operation. erefore, a
certain operational period takes into account in theMAC frame in time dimension.
MAC only concerns the time length of this operation and its decision regarding the
channel occupancy status. e eective time length of its operation and detection
outcome rely on the interference protection to the legacy system in a given channel
condition. Also, this operation can estimate the full capacity of the CoI before
utilizing the channel which is required to design an ecient transmision protocol.
By considering all these issues, the eective operational time is designed in Section
5.4.1.
• Backo Process: e protocol initiates a backo process when the channel is
sensed as busy either by spectrum discovery or by CCA. e backo process is
designed to produce a random delay for scheduling the packet transmission among
multiple contenders asynchronously. is process is accumulated with the backo
counter and the backo stage. In the proposed model, backo counter is a decre-
mental mechanism when SU neither senses the channel nor transmits any packet.
At the start of backo process, a random number is chosen for counter from the
contention window (CW) and the CCA is performed when the counter reaches
to zero. Since this process is executed before the CCA, it can be integrated into
the model to reformulate the detection objectives in the CCA.is integration has
been done by a PHY/MAC cross-layer design in the proposed model. e details
formulation of the using backo mechanism including its depending parameters
is explained in 5.4.2.
92 5.3 Proposed Model of the DSMA Protocol
• Clear Channel Assessment: is function is normally generated in MAC layer
and executed in PHY layer through signal detection. Only the decision is taken
into account in making the decision regarding packet transmission. ere are two
possible ways in running the CCA as shown in Fig. 5.3. e CCA performs just
aer the spectrum discovery if the channel is obtained as idle by the spectrum
discovery. e CCA also comes into operation through the backo process. e
command for initiating the CCA comes when the CW reaches to 0 in each backo
stage. When the CCA is evolvedwith the backo process, the detection parameters
in the CCA is updating according to the backo parameters which is one of the
main contributions of this model. e formulation is depicted in Section 5.4.3.
• Packet Transmission: It allows the SU to transmit the data packet in the chan-
nel. e RTS/CTS based mechanism [57, 66] is adopted in the proposed model
for reducing the collision period during the channel access. Packet transmission
protocol starts when CCA declares that the channel is idle, with the RTS packet
transmission instead of the main data packet transmission. e details of the
packet transmission including the measurement of its service time are presented
in Section 5.4.4.
5.3.2 Proposed Protocol
eprotocol structure of the DSMAprotocol is shown in Fig. 5.3. Each SU in the network
starts with the spectrum discovery operation. MAC layer requests PHY to perform this
spectrum discovery through signal detection algorithm. is is a mandatory task of
the proposed protocol to nd the spectrum opportunity from the CoI. If the channel is
assessed to be idle, the MAC (sets the CW as 0) does not allow backo delay and requests
the PHY to perform the CCA operation by using signal detection method. However, the
operational time and target detection in the spectrum discovery andCCA are followed by
the proposed PHY/MAC cross-layer designing which is explained later. If the channel is
obtained as busy in the contrary, then MAC initiates the backo process. Aer nishing
the backo process (when the CW reaches to 0), MAC requests the PHY to perform CCA.
5.4 Analytical Modeling of Proposed DSMA Mechanism 93
In overall, the CCA operation comes into operation either by directly from the spectrum
discovery or through the backo process.
When the channel is sensed as idle by the CCA operation, then SU goes for the imme-
diate packet transmission. Otherwise, when the channel is sensed as busy, SU has to wait
again a random backo delay according the backomechanism. e backomechanism
relies on two parameters: minimumvalue of the contentionwindow andmaximumvalue
of the backo stage. e detail working principle of the backo process for delaying the
channel access is revolved with the CWmin and maxBS which is explained in the Section
5.4.2.
5.4 Analytical Modeling of Proposed DSMA Mechan-
ism
5.4.1 Operational Time in Spectrum Discovery
Even though the signal detection method is applied both to spectrum discovery and
CCA but the aliation is separated due to their physical aributions. In particular,
expected operational period for the spectrumdiscovery is quite adaptive and relies on the
occupancy history of the PU. InMAC frame format, the period of the spectrum discovery
is acquired based on a dynamic decisional process. An optimization is conducted to
allocate the dynamic decisional period for spectrum discovery in every frame. e
operational period of the spectrum discovery is allocated by the following method:
1. For a given constraint, Pd ≥ P d, the detector threshold is designed where it
exhibits lowest PFA. is can be done through the ROC curve for a given SNR
value.
2. As two sensing steps are taken into account before any packet transmission, there-
fore, the overall target PD, P d, is achieved by distributing the PD into two steps.
is operation can be accomplished by applyingPd1 = Pd2 = 1−√
1− P d, where
Pd1 andPd2 tune the detector in the spectrum discovery and the CCA, respectively.
94 5.4 Analytical Modeling of Proposed DSMA Mechanism
3. Aer seing the detector’s sensitivity as explained above, the maximum opera-
tional time, τ(1)s,max, is obtained based on the following optimization:
τ (1)s,max = argmax
Pd≥P dPH0 (1− Pf (Pd, τs))
(1− τs
Tf
)(5.7)
By applying the equality constraint as Pd = P d, the objective function of (5.7) can be
derived as a function of τs and it is proved that this objective function is then a log-
concave function with respect to sensing period τs [3, 50]. us, there is a feasible
optimal value of the τs existing over the Tf period for which the objective function has
a maximum value [110].
5.4.2 Time Sequence Adaptation Based on Backo Process and
Detection Mechanism
Let us assume that i and k denote the backo stage and backo counter, respectively,
where i ∈ (0, u) and k ∈ (0,Wi − 1). Here, u is the maximum size of the backo
stage and the corresponding contention window isWu−1. Backo counter is described
with the minimum value of the contention window (W0) and the contention window is
described in the unit of slots.
At the start of backo process, MAC layer initializes the following variables: max-
imum value of the backo stage and minimum value of the contention window. en, a
random value is chosen for the backo counter from the contentionwindow (0, 1, 2, · · · ,
W0 − 1). e counter decrements its value uniformly and initiates the CCA while it
reaches to 0. If the channel is sensed as busy then the counter value is incremented based
on the binary exponential method [66] and performs the decremented counting for the
next backo stage. is process continues until the backo stage reaches its maximum
value. During this process, if the channel is sensed as idle in any stage, then SU transmits
the packet into the channel and comebacks to the initial spectrum discovery task. On the
contrary, if the channel is not found as idle for packet transmission within the maximum
contention, then the packet is discarded from the transmission aempt. e maximum
5.4 Analytical Modeling of Proposed DSMA Mechanism 95
size of the contention window is determined by 2iW0 and mathematically, it is referred
to as a function ofW0 when the maximum value of backo stage is given.
For the sake of simplicity, the complex backo process with innite re-transmission
aempt [67] is not considered herein as it can only consolidate the eectiveness of
backo process by means of packet service protocol. Since this research deals with
the cross-layer design of the sensing-access method, thus, the only situation where the
sensing has taken place in the backo process is sucient to consider for formulation.
As described above, channel sensing is performed at every backo stage aer n-
ishing the counter. In this proposed model, the target PD in channel sensing at every
backo stage is then reformed according to the backo parameters. e objective of this
target PD reformulation is to adaptively change the sensitivity of the detection process
to characterize the channel state. As a result, the detection output become an integrated
function of the backo parameters and detection sensitivity. By examining the merged
function, the overall objectives, throughput improvement and collision reduction, can
be accomplished signicantly.
5.4.3 Cross-layer Formulation of Backo and Detection Process
in CCA
Let Pidle be the idle probability through the channel aer nishing the backo counter
in a backo stage. According to the proposed cross-layer design, the Pidle is hence
expressed as a function of P d,W0, i. Assuming that x = Pd in backo stage i, the target
PD in this contention window is set by,
x(i,W0, P d) = 1− Wi
√1− P d = 1−
(1− P d
) 12iW0 (5.8)
us, the corresponding probability of false alarm is given by,
Pf(i,W0, P d
)= Q
(AQ−1
(x(i,W0, P d
))+B
)(5.9)
96 5.4 Analytical Modeling of Proposed DSMA Mechanism
RTS
CTS
DATA
ACK
NAV (RTS)
DIFS SIFS SIFS SIFS
DS phase Packet transmission phase
Sensing
PU Active
Sensing
Sensing
RTS/CTS based access
PU
SU1
SU2
SU3
Figure 5.4: Time slot operation of proposing dual-level sensing based mul-tiple access protocol.
where, A =√
2γ + 1 and B = γ√σslotfs are assumed in (5.6). e channel idle
probability can be formulated as,
Pidle(i,W0, P d) = PH0
(1− Pf (i,W0, P d)
)+ PH1
(1− x(i,W0, P d)
)(5.10)
5.4.4 Packet Transmission Service
is section presents a complete packet transmission protocol based on the proposed
model. Fig. 5.4 shows a packet transmission process when a cognitive user S1 wants
to transmit a data packet to another user S2 via a single-hop communication link. A
single radio channel is considered where PU occupies the channel for a particular period
over the full frame. e vertical dierentiation of this single-radio channel describes the
dierent activities of each user in the the same time dimension.
As shown in Fig. 5.4, all cognitive users have to wait for a certain period to complete
the spectrum discovery according to Section 5.4.1. Once the channel is found as idle, SU1
initiates the second sensing steps through CCA function. If SU1 nds the channel as idle
for a predened distributed inter frame space (DIFS), then SU1 immediately transmits
the RTS frame to the channel instead of the data packet transmission. is is a frame
based sensing method, where the reception of the RTS packet to the adjacent contenders
initiates another function, the network allocation vector (NAV). As shown in Fig. 5.4, SU3
which is not the desired user of the data packet receiver, initiates the NAV by encoding
the RTS frame. Furthermore, technically, the RTS frame contains the information about
5.5 Performance Analysis 97
the length of the data packet with the other relevant information about the ongoing
transmission aempt. us, SU2 replies with the clear-to-send (CTS) just following the
short inter frame spacing (SIFS) period by reading the RTS frame, and other contenders
keep following the backo process. If SU1 receives the CTS successfully, then it goes for
data transmission aer a SIFS. In contrast, if the sender does not receive any CTS aer
SIFS interval, then it is assumed that the packet collision occurs in the channel. Aer
receiving the DATA frame, SU2 acknowledges with an ACK frame to the sender SU1
aer a SIFS interval which makes a completion of a successful data transmission.
5.5 Performance Analysis
e performance of this proposed model is evaluated by the achievable throughput in a
single transmission aempt. e probabilistic performances from every step involve in
the measurement of the throughput. Step-by-step analysis (SA) is presented according
to the following sequences:
SA1 :e spectrum discovery is taken place at the start of the frame with the model
as given in Section 5.4.1. From that design, the operational time of the spectrum
discovery τ(1)s,max → Ts and the corresponding probabilities take into account in
the computation of the throughput.
P(1)idle (Ts, Pd1) = PH0 (1− Pf (Ts, Pd1)) + PH1 (1− Pd1) (5.11)
P(1)busy (Ts, Pd1) = 1− P (1)
idle (Ts, Pd1) (5.12)
SA2 :When channel is decided as idle with probability P(1)idle, then the total frame is
divided into mini slots with the per-slot length of σslot and the CCA is performed.
Total number of slots are estimated in the given frame by (2NB(i) − 1)W0, where
NB(i) = i + 1, i + 2, · · · , i + u − 1. Apart from the spectrum discovery, the
remaining time period Tf−Ts−D is divided into mini-slots, whereD denotes the
total time required for signal propagation and turnaround time which is assumed
98 5.5 Performance Analysis
as constant parameter. us, the σslot can be expressed as follows,
σslot =Tf − Ts −D
(2NB(i) − 1)W0
(5.13)
If the channel is detected as idle for DIFS period, the packet is transmied in that
condition. In the proposed model, the period of DIFS and the idle probability in
CCA are derived as function ofPd2 , Ts, i, andW0. e constraint of DIFS is adopted
herein from IEEE 802.11 [66] which is as follows,
TDIFS ≥ 2σslot (5.14)
Considering the equality constraint, TDIFS can be obtained as a function of Ts, i,
andW0 based on equation (5.13). us the corresponding channel idle probability
over the TDIFS period is
P(2)idle (TDIFS, Pd2) = PH0 (1− Pf (TDIFS, Pd2)) + PH1 (1− Pd2) (5.15)
As Pd1 , Pd2 related to P d, thus the above equation can be simply transformed as
below:
P(2)idle (TDIFS, Pd2)⇒ P
(2)idle
(P d, Ts, i,W0
)(5.16)
SA3 :e conditional backo process is formulated by a Markov chain model. In the
Markov chain model, whatever the reason behind the starting of backo process,
it comes with the formulation of a transmission aempt in the channel which
does not related with the outcome of the spectrum discovery. e transmission
probability is, however, depended on the cross-layer model as proposed in Section
5.4.2 and expressed in terms of u, P d,W0.
SA4 :e transmission probability is further formulated in a multiple access scenario
with N contenders. By considering all possible situations in a transmission at-
tempt, nally, the throughput is derived as a function of transmission probability
5.5 Performance Analysis 99
and the number of contenders.
5.5.1 Transmission Probability
ere are two possible ways to occurring at least a single transmission into the channel.
In the rst way, the channel is obtained as idle consecutively both in the spectrum
discovery and the CCA operation as explained in SA1 and SA2. By using (5.11) and
(5.16), the transmission probability is expressed as follows,
φ(1) = P(1)idleP
(2)idle (5.17)
In second possible way, SA1 nds the channel as busy with probability 1 − P(1)busy and
initiates the backo process. As explained before, the backo process follows a two-
dimensional Markov chain similarly to [63, 66]. erefore, the generic transmission
probability achieved at the end of backo process is φ(2)which does not rectify the
initial starting probability 1 − P(2)busy according to the characteristics of Markov chain
1. Formulating the backo process into Markov chain as given in Section 5.5.1, the
transmission probability for SA3 is expressed as follows,
φ(2) = f(2)φ (P d,W0, i) (5.18)
us, the total transmission probability is
φ = φ(1) +(1− φ(1)
)φ(2)
(5.19)
Derivation of the transmission probability φ(2)
Let us consider a 2-dimensional Markov process of the proposed DSMA/CA scheme
as given in Fig. 5.5, where the states s (t) , r (t) are denoted as the value of i, k,
i ∈ (0, u), k ∈ (0,Wi − 1). us, P i1, k1|i0, k0 denotes the transition probability
1In this Markov chain, any state transition probability only depends on its adjacent previous state and
does not aware about the transition probability from where it starts (initial state transition probability).
100 5.5 Performance Analysis
Figure 5.5: Markov chain model as the state transition of packet serviceprocess.
from r(t) = i0, s(t) = k0 to r(t+ 1) = i1, s(t+ 1) = k1.
• e process starts from the state 0, k, where k ∈ (0,Wi − 1), and then forwards
to 0, k − 1 direction until i, 0 state with the probability of
P i, k|i, k + 1 = 1; i ∈ (0, u) ; k ∈ (0,Wi − 2) (5.20)
• At i, 0, the CCA is performed. If CCA nds the channel is idle with probability
P(2)idle, then the RTS packet transmits. Aer completion of a successful RTS trans-
mission, a new packet transmission aempt starts with the seing of the backo
stage i = 0 and the contention window k = W0 − 1. In the Markov chain, this
5.5 Performance Analysis 101
transformation occurs with the probability of
P 0, k|i, 0 =P
(2)idle
W0
; i ∈ (0, u) ; k ∈ (0,Wi − 1) (5.21)
• If CCA nds the channel is busy at i, 0, the value of backo stage is increased
by 1, and a random contention window is chosen in the range of (0,Wi − 1) with
the probability of
P i, k|i− 1, 0 =1− P (2)
idle
Wi
; i ∈ (1, u) ; k ∈ (0,Wi − 1) (5.22)
• Once i reaches the maximum value u, it does not increase the value further for
packet transmission. As a result, the transition probability is
P u, k|u, 0 =1− P (2)
idle
Wu
; k ∈ (0,Wu − 1) (5.23)
For observing the long run behavior of the proposed model, the stationary probability
of the Markov chain is estimated by taking πi,k = limt→∞ P s(t) = i, r(t) = k . In the
Markov chain, the transition probabilities can be simplied due their regularities; which
is obtained as follows,
πi,k =
(1− k
Wi
).
∑ul=0,l 6=i P
(2)idle πl,0 if i = 0
(1− P (2)idle) πi−1,0 if 0 < i < u
(1− P (2)idle) (πu−1,0 + πu,0) if i = u
(5.24)
Since, the summation of these transition probabilities is 1,
1 =u∑i=0
Wi−1∑k=0
πi,k (5.25)
By using (5.24), and (5.25), π0,0 can be expressed as a function of P(2)idle, u, andW0 where
P(2)idle is also as a function of u,W0, and P d. According to proposed model, a packet
102 5.5 Performance Analysis
transmission occurs only while the backo counter reaches to zero at any backo stage.
erefore, a generic transmission probability in a random slot is dened as,
φ(2) =u∑i=0
πi,0 =π0,0
P(2)idle
⇒ f (2)(u,W0, P d) (5.26)
5.5.2 Packet Service Process in Multiple Access
e multiple access operation is modeled by using node-state model and channel-state
model. From the node state model, a generic transmission probability, φ, in a random
slot is obtained to correspond with a backo process. By exploiting the per-node trans-
mission probability into the multiple access scenario among N number of contenders,
the packet service is constructed. e packet service in multiple access operation is
presented with channel-state model. In packet service, a transmission in a given slot
assumes that the next channel state C(t+ 1) is idle given that the current channel state
C(t) is also idle. Without loss of generality, the channel remains in the idle state at
(t + 1) slot when it is idle in the current slot t with a probability of (1− φ)N , only if
none of the (N − 1) SUs start to sense (second sensing) in the current slot t; where φ is
the probability for which an SU transmits a packet in a randomly chosen time slot. Let
us dene the probability Ptx that there at least a single transmission can take place in a
given time slot, where
Ptx = 1− (1− φ)N (5.27)
Similarly, a successful transmission occurs with the probability that an SU transmits on
the channel given that at least a single transmission takes place in a slot, i.e.,
Psc =Nφ (1− φ)N
Ptx(5.28)
According to the proposed protocol, all possible scenarios are described below:
5.5 Performance Analysis 103
Successful Transmission
In this scenario, the state of PU’s is idle, no false alarm is produced, and in this case,
a successful packet transmission has been accomplished. e probability for which
successful transmission occurs is
PS = PH0 (1− Pf )PtxPsc
= PH0 (1− Pf )×Nφ (1− φ)N (5.29)
Collision
ere are two possible collision scenarios can occur. Firstly, SU can collide with the PU
due to the imperfect sensing during spectrum discovery. e probability for which SU
can collide with PU is given by
Pc1 = PH1
(1− P d
)(5.30)
Collision can also occur between two SU’s packets due to simultaneous transmission.
is collision not only depends on the detector but also related with the failure of the
backo process. When the packet is not successfully transmied, then it is dened as
the collision with the probability of
Pc2 = PH0 (1− Pf )Ptx (1− Psc)
= PH0 (1− Pf )×(
1− (1− φ)N−1 (1− φ+Nφ))
(5.31)
us, the total collision probability is given by
PC = Pc1 + Pc2 (5.32)
Empty Slot
In this scenario, even though the channel is idle, access mechanism produces a false
alarm during sensing and therefore, no transmission occurs in the slot. us, the prob-
104 5.5 Performance Analysis
ability for which the slot can be empty is
PE = PH0Pf (1− Ptx) = PH0Pf × (1− φ)N (5.33)
5.5.3 Average Packet Service Time
Now the length of the time slot required for the completion of a packet service is cal-
culated in this section. e average service time Tser is dened as the average duration
from the instant a frame becomes the head-of-line at the MAC buer to the end of its
successful transmission [63, 66]. Note that the length of a state in aMarkov process is not
a xed period of real time. Each state might be occupied by a successful transmission,
a collision, or be empty. erefore, the calculation of the expected time spent in those
three considered scenarios are converted the state into the real time. e variables which
are needed to be expressed for evaluating the expected service time are as follows,
• TS is the duration of the time slot in which a successful transmission is completed
with the probability PS . Let T S be the average time taken to complete a transmis-
sion successfully, which can be calculated as follows,
T S = TSD + TDIFS + TRTS + 3δ + 3TSIFS + TCTS + TDATA + TACK (5.34)
us, the time required for a successful transmission is given by
TS = T SPS (5.35)
• TC is the duration of time slot where collision has occurred with the probability
PC . If TC is the average time for collision then
TC = TCPC (5.36)
5.5 Performance Analysis 105
where TC is measured by,
TC = TSD + TDIFS + TRTS (5.37)
• TE is the duration of the empty time slot where no transmissions have occurred
with the probability of PE . As each slot duration is assumed to be σslot, the
expected time for the empty slot would be the same as the slot time σslot, thus
TE = σslotPE (5.38)
Without loss of generality, it is assumed that all SUs use an identical length of the data
frame. us, the average service time, Tser, can be computed by using (5.35),(5.36), and
(5.38), as follows,
Tser = TS + TC + TE (5.39)
5.5.4 Normalized Throughput
Normalized throughput N can be dened as the ratio of time the channel is used for
transmiing the payload successfully to the average service time. A complete successful
transmission can occur in a random slot with the probability of PS as shown in the
previous subsection. IfE[P ] is the average packet payload size, then the average amount
of payload information successfully transmied in a slot time isPSE[P ]. us, according
to the denition of normalized throughput N can be expressed as
N =PSE[P ]
Tser(5.40)
where PS and Tser can be evaluated by using equations (5.29) and (5.39), respectively.
106 5.6 Simulation Results
Table 5.1: System parameters used in the simulation.
Parameters Value
Packet payload 8512 Bytes
MAC header 728 bits
PHY header 512 bits
RTS 450 bits+ PHY header
CTS 320 bits+ PHY header
ACK 320 bits+ PHY header
Channel bit rate 1 Mbps
Propagation delay(δ) 1 µs
SIFS 28 µs
Size of CW(W0) 16
Maximum backo stage(u) 4
5.6 Simulation Results
In this section, the performance of the proposed DSMA protocol is evaluated with the
performance matrices and the validation of the analytical model. In simulation, it is
considered that primary channel has 6 MHz bandwidth with PU activity as follows:
PH1 = 0.1 over Tf = 10 ms period. To protect the PU’s transmission, the target
probability of detection is set as P d = 0.9 as dened in IEEE 802.22 dra standard [78].
5.6.1 Model Validation
To validate the analytical model, we analysed and compared the analytical result with
the simulation result. Additionally, an approximatedmodel of φ(2)from [66] is compared
with the proposedDSMA.e parameters used in this analysis are given in Table 5.1. e
analytical model of N in equation (5.40), is very convenient to determine the maximum
level of the achievable throughput. Let us rearrange equation (5.40) as follows,
N =E[P ]
Tser/PS=E[P ]
Tden(5.41)
5.6 Simulation Results 107
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.10.4
0.45
0.5
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9
Norm
aliz
ed T
hro
ughput
Probability of Transmission
Approximation of λ
Analytical
Simulated
Figure 5.6: Normalized throughput versus probability of transmission forcomparing the analytical, simulated, and approximated modelof DSMA scheme for N = 10.
whereas N can be maximized while the denominator of the above equation (5.41) is
minimized. With the help of [66], the following approximated solution of φ(2)is obtained
as,
φ(2) =1
N√T ∗C/2
(5.42)
where T ∗C = TC/σslot. By applying this approximation of φ(2)with respect toN , we can
also evaluate N . In Fig 5.6, we evaluated the normalized throughput N with respect to
probability of transmission φ and observed that the simulation result closely matched
with the analytical result. Moreover, the normalized throughput N also increased with
the same rate due to the approximation, and it was even smaller when φ has small value.
Nevertheless, with the increasing of φ, the normalized throughput N maintains closer
value to the analytical result.
5.6.2 Throughput Performance Analysis
Fig. 5.7 shows the variation of normalized throughput with respect to transmission
probability φ for N = 5, 10, 20, 50, which implies that throughput decreases rapidly
when a large number of SUs are intended to access the channel. When there are 5
108 5.6 Simulation Results
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9N
orm
aliz
ed
Th
rou
gh
pu
t
Probability of Transmission
N = 5
N = 10
N = 20
N = 50
Figure 5.7: Normalized throughput versus probability of transmission ofDSMA scheme for N = 5, 10, 20, and 50.
0 0.005 0.01 0.015 0.02 0.025 0.030
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Sensing time (sec)
Norm
aliz
ed thro
ughput
SS−OSA
CSS−OSA
DSMA
Figure 5.8: Comparison of normalized throughput versus sensing time ofthree schemes for N = 10 and γ = −15dB.
contenders, the throughput reaches its maximum level quite slowly and requires a large
value of φ compared with 10, 20, and 50 user cases, but it can be sustained for a long
range of φ. On the other hand, the throughput reaches its maximum level with a very
small value of φ but it decreases more rapidly while the probability of transmission
increases among 50 users.
5.6 Simulation Results 109
0.5 1.5 2.5 3.5 4.5 5
x 10−3
0.8105
0.8145
0.8185
0.8225
0.8265
0.8305
Sensing time (sec)
Norm
aliz
ed thro
ug
hput
N = 5
N = 10
N = 20
N = 50
Figure 5.9: Normalized throughput versus sensing time of proposed DSMAscheme for N = 5, 10, 20, 50 and γ = −15dB.
0.5 1 1.5 2 2.5 3 3.5 4
x 10−3
0.808
0.812
0.816
0.82
0.824
Sensing time (sec)
Norm
aliz
ed thro
ughput
γ = −10 dB
γ = −15 dB
γ = −20 dB
Figure 5.10: Normalized throughput versus sensing time performance ofproposed DSMA scheme for γ = −10 dB, −15 dB ,−20 dB,and N = 10.
Now we analyse the normalized throughput N achieved by the DSMA scheme with
respect to the sensing time Ts and the SNR γ. Fig. 5.8 shows the comparison of the DSMA
schemewith the other two schemes, where γ = 15 dB andN = 10. When single sensing
110 5.6 Simulation Results
with opportunistic spectrum access (OSA) (SS-OSA) scheme is deployed, the throughput
has increased w.r.t sensing time and reaches its maximum level at 2.5 ms. Aer that the
throughput starts to decrease corresponding to the increases of sensing time. If the CSS
is used in sensing with the OSA (CSS-OSA) then N can reach a maximum level at quite
a low value of sensing time, e.g. less than 1 ms. is is the advantage of the CSS, that
it can detect the PU’s activity strongly within a short sensing duration even though
the sensing channel experiences a low SNR. Nevertheless, the CSS-OSA scheme goes
downward aer 2.5 ms, similar to the SS-OSA scheme. On the other hand, our proposed
scheme provides a quite stable throughput, over 80% of the oered load.
e exact variation of throughput of the DSMA scheme can be seen in Fig. 5.9 and
Fig. 5.10. In Fig.5.10, the variation of average normalized throughput is depicted w.r.t
sensing time for dierent sets of N at very low SNR, γ = −15 dB. In that case, the
variation is less than 1.5% between N = 5 and N = 50 cases. Another change of the
normalized throughput can be found in Fig. 5.10, where for three very low SNR, −10
dB, −15 dB, and −20 dB, the proposed DSMA scheme also exhibits a stable throughput
regime within a very short sensing duration. At the very worst channel condition, such
as −20 dB, the DSMA scheme only takes a longer sensing time to reach its maximum
and stable conditions. Otherwise at −10 dB and −15 dB, N reaches the stable region
within a very short sensing time and eventually aer 3 ms are sustained at the same
maximum throughput rate. is is one of the main advantages of the DSMA scheme,
that at very low SNR within a very short sensing time, it can reduce the eect of false
alarms and missed detection which nally leads to increased throughput.
Usually the physical layer data rate is directly related to the channel’s SNR. Here
we also show the throughput performance of the proposed cross-layer based random
access scheme when subject to low SNR value. As we are interested in worst channel
conditions, we performed our simulation from 0 dB to −20 dB comparing the three
dierent schemes. Fig. 5.11 shows that the normalized throughput of the DSMA scheme
is robust and consistent over very low SNR conditions. In contrast, the SS-OSA and the
CSS-OSA schemes have had low normalized throughput, such as, N below 0.5 at −20
5.6 Simulation Results 111
−20 −18 −16 −14 −12 −10 −8 −6 −4 −2 00
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
SNR (dB)
Norm
aliz
ed thro
ug
hput
SS−OSA
CSS−OSA
DSMA
−20 −15 −100.823
0.824
0.825
0.826
SNR (dB)
Norm
aliz
ed thro
ughput
Figure 5.11: Comparison of normalized throughput versus SNR of the threeschemes for N = 10 and Ts = 1 ms.
−20 −16 −12 −8 −50.8105
0.8145
0.8185
0.8225
0.8265
0.8305
SNR (dB)
Norm
aliz
ed thro
ughput
N = 5
N = 10
N = 20
N = 50
Figure 5.12: Normalized throughput versus SNR of proposed DSMA schemefor N = 5, 10, 20, 50 and Ts = 1 ms.
dB. e CSS-OSA technique can mitigate this inability during low SNR values as we see
in Fig. 5.11, but it is sustained at the same level of normalized throughput while the
SNR value is increased. In contrast, in our proposed scheme, we use the DS technique to
reduce the loss of transmission opportunities. e exact variation of the throughput of
112 5.7 Conclusion
DSMA w.r.t SNR is shown in Fig. 5.12. As the number of SUs increases, the average nor-
malized throughput increases, but the required SNR to achieve themaximum normalized
throughput decreases. erefore, the above performance analysis has characterized the
advantages of the proposed DSMA scheme and has validated the protocol.
5.7 Conclusion
In this chapter, an ecient spectrum access mechanism is proposed for cognitive radio
networks where the DS technique is integrated with the access mechanism to improve
the throughput performance. e DSMA is developed by combining the DS and the
backo process to make a rm decision about channel activity before data transmission.
An analytical model of the proposed protocol is developed by using a Markov chain
model to compute the normalized throughput. e performance evaluations are demon-
strated with model validation and performance comparison, which implies that the pro-
posed scheme achieves signicant improvements both in protecting the PU’s activity and
in increasing throughput. Compared with other existing schemes, the DSMA scheme
provides a stable throughput regime when the SNR value is comparatively low.
CHAPTER 6
Sensing-Assisted Access Protocol withImperfect Sensing and Performance
Analysis for Multiple Access
Spectrum access is an essential functionality to enable secondary users to occupy the
under-utilized spectrum while improving the throughput performance of the cognitive
radio networks (CRN). Existing cognitive medium access control (C-MAC) protocols
have the limitations in providing sucient protection to primary users due to the ex-
clusion of spectrum sensing and due to the aggressive contention-based access policy
for improving the secondary users’ throughput. To overcome these issues, we propose
a novel sensing-assisted access (SAA) protocol by designing a cross-layer based random
access mechanism including imperfect sensing. In particular, we model a backo mech-
anism of the SAA protocol by capturing the sensing error from the spectrum sensing
in physical layer (PHY). Exploitation of all sensing aspects during backo process re-
veals the spectrum opportunity extensively, and the consequent possibility of collision
is reduced through the sensing-assisted contention process. Here, an analytical model
of the proposed SAA protocol is acquired with Markov chain analysis for evaluating the
throughput and delay performance. Performance analysis and numerical results show
that the SAA protocol improves the throughput and delay performance signicantly in
a large multiple-access scenario alongside ensuring sucient interference protection to
the legacy system.
114 6.1 Introduction
6.1 Introduction
Several MAC-based access protocols [6, 7, 48, 49, 57, 63, 65] have been proposed nom-
inally in association with spectrum sensing. Among all access protocols, a carrier-sense
multiple access (CSMA) protocol has the compatibility with cognitive radio operation
as it also does channel monitoring before any data transmission. To deploy the CSMA
protocol, a backo mechanism is the collision avoidance feature to reduce the collision
among packets being transmied by the other users, with a predened enumeration
process [66]. In particular, the backo mechanism not only depends on the MAC layer
parameters, i.e., contention window (CW), but also depends on the decision of physical
channel monitoring [6, 7]. For the equal access priority among homogeneous type users,
the authors in [66] assumed perfect sensing during the backo mechanism and captured
the sensing error regarding collision probability while the simultaneous transmission
occurred in a given slot. Following this strategy [66], the authors in [6, 7] applied a
similar backo mechanism without considering the sensing error which can result in
intolerable collision among secondary and primary users. Although those strategies
may improve the throughput performance, they cannot guarantee enough interference
protection to primary users. To facilitate sucient interference protection to the PU
under similar backomechanisms requires a higher sensing period [49, 57, 63, 65] which
brings more challenges to sensing-throughput trade-o problem.
In CRN, two types of sensing errors can be generated [3]: false alarm (detecting PU
where no PU is present) and missed detection (detecting no PU where PU is present).
Missed detection leads collision eects in accessing the channel, and a false alarm pro-
duces wastage of transmission opportunity. e impacts of imperfect sensing are not
explicitly synthesized in [6, 7, 48, 63] due to only considering of collision eect in channel
accessing. Also, the throughput improvement cannot be maximized to the fullest extent
as the loss of spectrum opportunity due to false alarm was ignored [6, 7, 48, 63]. Even
though the authors in [49, 65] considered imperfect sensing, they still failed to improve
the throughput as an additional (dedicated or xed) time slot longer than the backo
slot was required for spectrum sensing to reduce the collision eect. Instead of an
6.1 Introduction 115
additional and dedicated sensing slot, a conditional channel sensing aer the backo
counter can be an eective method for further collision reduction [113]. In our study, we
consider imperfect sensing aspects during the design of the access protocol to explore
throughput improvement by the cross-layer sensing parameters as well as to provide
higher interference protection to primary users.
6.1.1 Contributions
In this chapter, we propose a sensing-assisted access (SAA) protocol through a cross-
layer based backo mechanism in the presence of sensing error. e SAA protocol
exploits all aspects of sensing from the PHY while integrating with the access strategy
at MAC. e major contributions of this chapter are pointed below:
• We propose a sensing-assisted access protocol with PHY/MAC cross-layer design
to improve the throughput performance and to guarantee higher interference pro-
tection to primary user. All the regarding aspects of physical channel sensing such
as missed detection and false alarm are explicitly synthesized in deciding the state
transitions in the backo process, which outperforms the conventional backo
process [63, 66] in the deployment of random access in cognitive radio networks.
• To reduce the collision among inter-network (among primary and secondary net-
work), a conditional channel assessment is allowed aer the backo process for a
unit slot period of the backomechanism instead of providingwith larger and ded-
icated spectrum sensing period. is improves the performance of our proposed
SAA protocol over [49, 65].
• We develop an analytical model of the SAA protocol by using Markov chain ana-
lysis to measure the average packet service time, normalized throughput, and av-
erage delay. rough this analysis, the throughput is derived as an integrated func-
tion of MAC layer parameters i.e., contention window, number of backo stage,
and PHY layer parameters, i.e., signal-to-noise ratio (SNR), detector’s threshold,
and noise variance.
116 6.2 System Model
PU
PBS
SBS
Spectrum
opportunity
SU
SU
1
N
Figure 6.1: Network configuration of cognitive radio network for SA-MACprotocol.
• We validate our proposed SAA protocol alongside with CSMAmechanism [63, 66]
through simulation and analytical results. Also, the numerical results provide the
guideline with tackling the sensing error to acquire an ecient throughput and
delay performance under a large multiple access scenario.
6.2 System Model
6.2.1 Network Entity
We consider a cognitive radio network consisting of a single time-sloed channel
whereby the primary network is the legacy system in accessing the channel and
secondary network can occupy the channel only while no PU is using the channel.
Primary network is structured with a primary base station (PBS) and a PU, and the
secondary network is organizedwith a secondary base station (SBS) andN (index byn =
1, · · · , N ) numbers of SU as shown in Fig. 6.1. We consider the upstream transmissions
of the secondary network, where SUs transmit by sharing the channel and SBS receives.
To accomplish this upstream scheduling, the contention-based access mechanism build
6.2 System Model 117
the communication link between SUs and SBS based on channel monitoring. Our
proposed SAA mechanism allows MAC protocol to control the contention between
SUs in upstream scheduling, by taking into account the physical layer based spectrum
sensing.
According to the principle of CR operation, secondary and primary networks are
non-cooperative. In the absence of cooperation and synchronization amongst PU and
SUs, SUs do not know the exact communication mechanisms of the primary channel.
erefore, each SU employs channel sensing to detect the PU’s transmission through
channel assessment (CA) function in MAC. In our proposed model, we use double chan-
nel assessment according to the backo process. e rst CA (denoted by CA1) refers
the ability to detect the energy level based on noise oor, and the second CA (denoted
by CA2) refers the ability to detect and decode an incoming signal. Each SU initiates the
contention access including CA1 at the starting of each frame. Once the contention is
nished based on CA1’s decision, the CA2 is invoked to enhance the detection perform-
ance. Unlike [49, 63, 65], we use CA1 and backo process jointly and CA2 aer that in
order to ecient utilization of the backo mechanism.
6.2.2 Channel Modeling with Imperfect Sensing
By considering the imperfect sensing [3, 63], the sensing errors are evaluated as the
probability of false alarm (Pf ) and the probability of missed detection (Pm) and are
related to the threshold value, noise power, signal-to-noise ratio (SNR), and type of the
detector as described by equations (3.6) and (3.7). By assuming the activity paern of
primary network, the secondary network can congure the channel state aer every
detection process as follows:
1. When the PU is inactive, and the detector produces no false alarm then the channel
state is decided as idle with probability PH0(1− Pf ). In contrast, the channel can
also be idle with probability PH1Pm, if missed detection occurred. us, channel
idle probability is
Pi = PH0(1− Pf ) + PH1Pm (6.1)
118 6.3 Proposed Sensing-Assisted Access Protocol
2. When the PU is also inactive but false alarm is produced, then the channel state is
decided as busy with probability PH0Pf . e channel status can also be decided as
busy with probability PH1(1−Pm), if no missed detection occurred (i.e. perfectly
detected with probability Pd where Pd = 1−Pm). us, channel busy probability
is
Pb = PH0Pf + PH1(1− Pm) (6.2)
6.3 Proposed Sensing-Assisted Access Protocol
6.3.1 PHY/MAC Cross-layer Based Contention Mechanism
To accomplish multiple access, the contention procedure allows mapping between con-
vergence process at SUs and SBS [78, 114]. During this process, the secondary network
needs to concern about not to cause harmful interference with incumbents. erefore,
our proposed SAA mechanism allows a MAC protocol to control the contention in mul-
tiple access, by taking into account the PHY sensing.
eworking principle of the proposed contention accessmechanism is describedwith
the owchart in Fig. 6.2, where the backo process is initiated when a packet is intended
to transmit in the channel. Before transmiing a packet, each SU monitors the channel
to avoid collision with packets being already transmied into the channel. e backo
process is modeled with the parameters of backo slot-counter and backo stage (BS).
At the starting of each frame, a discrete backo slot-counter is chosen in the range of
contention window (0, 1, · · · , ω0− 1), where ω0 is the minimum contention window in
the unit of the slot, and aer that, CA1 is executed by the SU. If CA1 declares that the
channel is busy, then the counter value is unchanged, and SU keeps doing CA1 until the
channel is found as idle. Note that ongoing execution of the CA1 is an obligatory task
that satises the network association in cognitive radio network [28, 78, 114]. On the
other hand, if CA1 declares that the channel is idle then SU reduces its counter value
by one and continues this procedure until the slot-counter value reaches to zero. Aer
reaching to zero backo counter, SU performs CA2. If CA2 declares that the channel
6.3 Proposed Sensing-Assisted Access Protocol 119
Start: CR Access with
new packet
BS = 0, CW = Wo
Maximum BS = u
Maximum CW = W
Perform CA1 over unit
backoff period
Channel
idle?
CW = CW - 1
CW = 0?
Perform CA2 over unit
backoff period
Channel
idle?
Transmit packet
BS = BS + 1
BS > u
Discard packet
YES
YES
YES YES
NO
NO
NONO
u
0ω
02u
ω
Updating CW
= 2*CW2
Figure 6.2: Flowchart of the channel access mechanism.
is busy, then SU increments it contention window and moves to the next backo stage.
An exponential incremental method [49, 63, 66] is adopted in proposed backo process,
where aer each unsuccessful backo stage, the contention window is doubled and
continued to do that until reaching maximum contention window ωmax = 2uω0, where
u is the maximum number of backo stage. On the other hand, SU goes for immediate
data transmission if the CA2 indicates that the channel is idle. SU can discard the packet
from contention access if the channel is still not available for packet transmission aer
backo stage value reaches to its maximum value.
120 6.3 Proposed Sensing-Assisted Access Protocol
RTS
CTS
DATA
ACK
NAV (RTS)
SIFS SIFS SIFS
CA1+Contention+CA2
PU Active
CA1+Contention
PU
SU1
SBS
SU2
Proposed
contention RTS/CTS based packet transmission
DIFS
Figure 6.3: A complete packet transmission service of proposed SAA pro-tocol.
e above mechanism implies that the MAC contention relies on the declaration of
CA1 and CA2 which can be induced by sensing error. erefore, we conduct the cross-
layer analysis for evaluating the performance of the proposed protocol.
6.3.2 Packet Transmission Structure of Proposed SAA Protocol
e protocol structure of proposed SAA is described with a complete packet transmis-
sion in the upstream of an SU accommodated with an SBS’s coverage. As shown in
Fig. 6.3, SU1 wants to transmit a data packet via a single-hop communication link to
SBS. According to the proposed contention mechanism, SU1 initiates CA1 with the
initialization of given backo parameters. Aer completion of CA2 with channel idle
probability given that the channel was also idle in CA1, SU1 defers for a distributed
inter-frame spacing (DIFS) period and then, transmits the request-to-send (RTS) packet
into the channel instead of the data packet for mitigating the hidden node problem.
Once the SBS receives the RTS, a clear-to-send (CTS) packet is transmied following
a short inter-frame spacing (SIFS) period. When SU1 receives the CTS successfully,
then SU1 nally goes for data transmission with DATA packet. Aer receiving the
DATA packet, SBS acknowledges with an ACK packet to the sender aer a SIFS interval
which makes the completion of a packet service procedure in our SAA protocol. e
information about the size of the data packet and the mapping between SU1 and SBS
have been addressed in the RTS/CTS control packet which can be decoded by other SUs
for updating the network allocation vector (NAV). If any hidden SUs can read any one of
6.3 Proposed Sensing-Assisted Access Protocol 121
the RTS and CTS packets, then it defers any transmission in the channel. In worst case
scenario, collisions can occur, among SUs’ packets or SU and PU packet, only for the
duration of RTS packet which is determined by the no reception of CTS. As the length of
RTS packet is much shorter than the length of the data packet, therefore, collision period
due to SAA protocol is shorter compared with other cognitive radio access mechanism
[48, 100]. In particular, when a longer data packet is considered then RTS/CTS scheme
increases the access eciency even though it utilizes two control packets without the
payload.
For analyzing the impact of sensing on spectrum access, we assume the perfect packet
reception in a communication link; therefore, re-transmission criteria is not considered
in the packet service. In that case, if ACK packet is not received by the sender then it
will be treated as the eect of the collision. In current IEEE 802.11 based CSMA/CA
protocol for the innite re-transmissions case (u =∞), aer a successful transmission,
station defers DIFS periods and then in the next slot the station can go for direct packet
transmission with the probability of 1/(ωi+1) [67, 115, 116]. is model is quite ecient
for homogeneous contenders in WLAN but may produce severe interference to primary
networks as the consecutive packet transmission by SU does not rely on the current
spectrum sensing. erefore, unlike [67], we consider only a single packet transmission
based on its aempt through proposed backo mechanism.
6.3.3 Analytical Modeling with Markov Chain Analysis
Proposed SAA protocol is formulated with a two-dimensional Markov chain process as
shown in Fig. 6.4, where each state s(t), r(t) is dened by the stochastic process of the
backo stage s(t) and the backo counter r(t), respectively. e value of s(t) and r(t)
is described by i, k given that i ∈ (0, u), k ∈ (−1, ωi − 1) where u is the maximum
number of backo stages and ωi is its corresponding backo counter value. To analyze
the service process of proposed strategy, let consider P i1, k1 | i0, k0 represents the
transition probability from s(t) = i0, r(t) = k0 to s(t+ 1) = i1, r(t+ 1) = k1 state.
Let us dene the transition probabilities according to the proposed contention-based
122 6.3 Proposed Sensing-Assisted Access Protocol
00 2,ω −0 1,
2p
4p
1p
3p
3 0/p ω
2p
1p
2p
1p
2p
1p
2p
11 1,ω −
11 2,ω −
2p
1 0,1 1,−
2p2
p4
p
1p
1p
1p
1p
1p
3p
0 0,0 1,−
1p 0
0 1,ω −
2p
2p
2p
1p
1p
1p
1p
1p
2p
4p
3p
1,
i
i ω −
2p
2p
2p
1p
1p
1p
1p
1p
2p
4p
3p
2,
i
i ω −1,i0,i
1 1,
1,i −
1,
u
u ω −2,
u
u ω −1,u0,u1,u −
Figure 6.4: Markov chain model as the proposed backo process of theproposed SAA protocol.
mechanism and sensing-assisted packet transmission protocol as explained below:
1. Backo process starts with (0, k) where k ∈ (0, ω0−1) and forwards in the (0, k−
1) direction until it reaches the (i, 0) state, while CA1 declares the channel is idle
with probability P1 → Pi, which yields
P i, k | i, k + 1 = P1 (6.3)
2. Backo state loops in the same state with holding the same value of k when CA1
declares the channel is busy with probability P2 → Pb, which yields,
P i, k | i, k = P2 = 1− P1 (6.4)
3. When k reaches to zero with P1, then in-state looping is terminated, and backo
state transits to either negative backo counter (i,−1) with probability P1 or next
6.3 Proposed Sensing-Assisted Access Protocol 123
backo stage with transition probability
P i, k | i− 1, 0 =1− P1
ωi, i ∈ (1, u) , k ∈ (0, ωi − 1) (6.5)
and a new k is chosen uniformly in the range of (0, ωi). Once i reaches its max-
imum value u, then it is not increased in subsequent packet transmissions which
yields,
P 0, k | u, 0 =1− P1
ωu, k ∈ (0, ωu − 1) (6.6)
4. When channel is sensed idle at (i,−1) with probability P3 given that it was also
idle at the previous state with probability P1, then SU transmits the RTS packet
into the channel. In contrast, if the channel is sensed busy at (i,−1) then backo
state transits to next backo stage (i+ 1, k) with the transition probability of
P i, k | i− 1,−1 =1− P3
ω0
, i ∈ (1, u) , k ∈ (0, ωi − 1) (6.7)
As a consequence, at initial backo stage, a new packet following a RTS transmis-
sion starts with the probability of
P 0, k | i,−1 =P3
ω0
, i ∈ (0, u) , k ∈ (0, ωi − 1) (6.8)
5. According to above steps, no. (3) and (4), the states i− 1, 0 and i− 1,−1
both move to next backo stage while the channel is sensed as busy, however, in
dierent scale of k. In overall, the total probability of this state transitions, for
i ∈ (1, u) and k ∈ (0, ωi − 1), can be expressed as
P i, k | i− 1, k − 1 =P2 + P1P4
ωi(6.9)
is Markov chain model is composed with the following state transition probability
124 6.3 Proposed Sensing-Assisted Access Protocol
matrix in a block form,
P =
Π0,0 Π0,1 · · · Π0,u
Π1,0 Π1,1 · · · Π1,u
.
.
.
.
.
....
.
.
.
Πu,0 Πu,1 Πu,u
(6.10)
where the sub-matrix Πi,k can be expressed as,
Πi,k =
P u, 0|i, 0 · · · P u, ωu − 1|i, 0
.
.
....
.
.
.
P u, 0|i, ωu − 1 · · · P u, ωu − 1|i, ωu − 1
(6.11)
Let the stationary probability of this chain be π(i, k) = limt→∞ P s(t) = i, r(t) = k
with i ∈ (0, u), k ∈ (−1, ωi − 1), then the row vector of Πi,k would be
π = [π(0, 0), · · · , π(i, ωi − 1), · · · , π(u, ωu − 1)]
and the stationary distribution of s(t), r(t) can be computed from the following con-
ditions,
π = πP (6.12)∑π(i, k) = 1, i ∈ (0, u) , k ∈ (−1, ωi − 1) (6.13)
which indicates that π is the le eigenvector of P corresponding to the eigen-value
1. Hence, for a closed-form solution of this Markov process, we obtain the following
balance equations,
π(i− 1, 0). (P2 + P1P4) = π(i, 0) ; 0 < i ≤ u (6.14)
π(i, 0) = P i.π(0, 0) ; 0 < i ≤ u (6.15)
where we assume that P = P2 + P1P4. For the chain regularities, the stationary
6.3 Proposed Sensing-Assisted Access Protocol 125
probabilities are derived as follows,
π(i, k) =ωi − kωi
.P1P3
u∑j=0
π(j, 0), for i = 0 (6.16)
π(i, k) =ωi − kωi
π(i, 0), for i > 0 (6.17)
Since, the summation of the probabilities would be 1,
1 =u∑i=0
ωi−1∑k=0
π(i, k) +u∑i=0
π(i,−1)
=u∑i=0
π(i, 0)
ωi−1∑k=0
ωi − kωi
+u∑i=0
P1π(i, 0)
=u∑i=0
π(i, 0)
(ωi + 1
2+ P1
)
=π(0, 0)
2
[ω0
u∑i=0
(2P )i + (1 + 2P1)u∑i=0
P i
]
=π(0, 0)
2
ω0
(1− (2P )u+1
)1− 2P
+(1 + 2P1)
(1− P u+1
)1− P
(6.18)
For the compactness of this analytical model, SU starts to transmit with the probability
of P1P3φ (comparing with transmission probability τ in [66], P1P3φ = τ ) where the
access probability φ is dened as follows,
φ =
∑ui=0 π(i, 0)∑u
i=0
∑ωi−1k=0 π(i, k) +
∑ui=0 π(i,−1)
=π(0, 0)
(1− P u+1
)1− P
(6.19)
By using equation (6.15), (6.16), and (6.17), φ is derived as a function of π(0, 0) in equation
126 6.3 Proposed Sensing-Assisted Access Protocol
(6.19). Now, by using equation (6.18), the access probability φ can be expressed as,
φ(P1, P , ω0, u
)=
2(
1− 2P)(
1− P u+1)
(1 + 2P1)(
1− 2P)(
1− P u+1)
+ ω0
(1− P
)(1− (2P )u+1
) (6.20)
where φ is solely depended on the sensing probabilities (P and P1) and backo paramet-
ers (ω0 and u).
6.3.4 Cross-layer Relationship Between Backo Mechanism and
Physical Channel Sensing
Existing CR access protocols [49, 57, 63, 65] were adopted the conventional mechanism
[66, 115, 116] similar to IEEE 802.11 protocol where a discrete and integer time scale
applied which was not directly connected to the system time. In our system, we inter-
connect the backo time scale with PHY’s operational time to complete the cross-layer
design of proposed SAA protocol. In our system, the total length of a single frame Tf is
divided into multiple slots with length τu.
According to proposed protocol, PHY executes channel sensing over the period of
the slot. Hence, sensing period of the detector τs is assumed to be τu. e optimal
sensing period in cognitive radio network can be designed by sensing-throughput trade-
o for either a target probability of detection [3] or a constant false alarm rate (CFAR)
[94]. In our model, the length of sensing period is chosen with the exploitation of
receiver operating characteristic (ROC) of the detector where the sensing error (Pf +
Pm) is minimum as depicted in [87, 117]. On the other hand, MAC adopts this slot
in designing contention window which is decided by the unit of the number of slots.
During operation, t and (t+ 1) correspond to the nishing of two consecutive slots and
the backo slot-counter either decrements or holds its value according to the detectors
(CA1 and CA2) evaluations.
Based on this connection between backo slot-counter and detectors, the state trans-
6.3 Proposed Sensing-Assisted Access Protocol 127
ition probabilities of the Markov chain can be estimated with the detector’s parameters.
Let us consider the same threshold value using for channel sensing in CA1 and CA2. At
same threshold value ε in CA1 and CA2, we obtain the regarding transition probabilities
as P1 = P3 = Pi(ε) and P = 1− (Pi(ε))2. Let rearrange equation (6.20) as follows,
φ =1
φd(6.21)
where φd can be obtained as,
φd (Pi, u, ω0) =1 + 2Pi
2+
ω0P2i
(1− (2 (1− P 2
i ))u+1)
2 (2P 2i − 1)
(1− (1− P 2
i )u+1) (6.22)
Let applying the approximation of 1 − (1 − P 2i )u+1 ≈ (u + 1)P 2
i (2 − uP 2i )/2 and
1−(2(1−P 2i ))u+1 ≈ 1−2u(2−(u+1)P 2
i (2−uP 2i )), and considering channel modeling
with imperfect sensing, an approximated φad is given by (6.23),
φad (Pf , Pm, u, ω0) ≈1+2(PH0
(1−Pf )+PH1Pm)
2+
2uω0(PH0(1−Pf )+PH1
Pm)2
2(PH0(1−Pf )+PH1
Pm)2−1
+ω0(1−2u+1)
(u+1)(
2(PH0(1−Pf )+PH1
Pm)2−1
)(2−u(PH0
(1−Pf )+PH1Pm)
2) (6.23)
By using equation (6.21) and (6.23), the approximated φ for the proposed SAA protocol
can be obtained. Comparing with [49, 65, 66], we exposed the sensing error such as
Pf and Pm instead of the collision probability (p in [66]) for analyzing the performance
in the consequence of imperfect sensing. Note that imperfect sensing does not always
result in the collision, it could be wastage of the spectrum opportunity for secondary
network for larger Pf [3, 63]. By extracting the spectrum opportunities, the throughput
of the random access can be enhanced as depicted in [103, 104].
128 6.4 Performance Analysis
6.4 Performance Analysis
6.4.1 Packet Service Process andNormalized Throughput inMul-
tiple Access Operation
Let us consider multiple access scenario with distinct access probability by indexing the
station’s label n = 1, 2, · · · , N . Also, we assume that any transmission in a given slot
needs a sensing outcome that the channel state (C) remains idle at (t+1) slot given that
the current state is also idle at t slot. Without loss of generality, the channel remains idle
during (t + 1) slot when the current slot was slot idle with probability
∏Nn=1 (1− φn)
only if none of the (N − 1) SUs start to second sensing in the current slot. If a packet
transmission occurs in an arbitrarily time slot with the probability φn, then let us dene
the probability that at least a single transmission (Tx) can take place in a given slot as,
P C(t) = Tx = 1−N∏n=1
(1− φn) (6.24)
A transmission is said to be successful if the packet is received successfully given that
at least a single transmission occurred in a given slot with probability,
P Success | C(t) = Tx =P Success, C(t) = Tx
P C(t) = Tx(6.25)
where P Success, C(t) = Tx =∑N
n=1 φn∏
n6=l (1− φl). By using (6.24) and (6.25),
the probability of successful transmission in a slot can be measured as,
PS = P C(t) = TxP Success | C(t) = Tx
=N∑n=1
φn∏n6=l
(1− φl) (6.26)
e author in [66] introduced the collision probability p considering the per-station
collision probability, which is the probability that more than one station is transmiing
simultaneously in a given slot. Exploiting this per-station collision probability [66], the
authors in [64] distinguished the collision in primary and secondary networks where
6.4 Performance Analysis 129
the outcome of the imperfect sensing is not rectied with a sensing-resultant collision
relationship. erefore, the generic collision probability is dened here as the probability
that SU experiences collision by at least on aempting transmission and derived as when
a SU transmied but it was not successful, i.e.,
PC = P C(t) = Tx (1− P Success | C(t) = Tx)
= 1−N∏n=1
(1− φn)−N∑n=1
φn∏n6=l
(1− φl) (6.27)
Apart from the collision and successful transmission, the channel can be idle with no
transmission from the cognitive users with the probability of
PI = 1− P C(t) = Tx =N∏n=1
(1− φn) (6.28)
Now we can compute the length of the time slot required for the completion of a
packet service where each slot may contain a successful transmission or a collision, or
be empty, as described above. e average service time TSer is dened as the average
duration from the instant a frame becomes the head-of-line at the MAC buer to the
end of its successful transmission [63, 65]. We calculate the expected time spent in those
three considered scenarios to convert the state into the real time, as follows,
Tser = TSPS + TCPC + TIPI (6.29)
where TS , TC , and TI are the total period required of successful transmission, collision,
empty period, respectively. e length this periods can be obtained as follows,
TS = TH + TDIFS + TRTS + 3TSIFS + TCTS + TDATA
+TACK + 3δ (6.30)
TC = TH + TDIFS + TRTS (6.31)
TI = τu (6.32)
130 6.4 Performance Analysis
where TH , TDIFS , TSIFS , TRTS , TCTS , and TDATA are the time length of payload header,
DIFS, SIFS, RTS, CTS, and DATA packet, respectively. Moreover, δ denotes the signal
propagation delay and TH is the summation of the time length of MAC and PHY headers.
Normalized throughput S can be estimated as the ratio of the time that SU used the
channel for a successful packet transmission to the average service time [66, 103, 115].
Let us assume E[P ] is the expected size of a data payload of the secondary user, then
the expected amount of payload successfully transmied in a slot time is PSE[P ]. us,
the normalized throughput S can be expressed as,
S =PSE[P ]
Tser(6.33)
6.4.2 Average Access Delay
e average access delay of our model is dened as the average time interval between
the moment that the packet is in service and the time that the packet is successfully
transmied. In particular, the required average time is computed from a packet enters
into MAC operation for transmission to the instant of reception of the acknowledgment
regarding the successful transmission. us, the average access delay E[D] is given by
E[D] = E[X]τu + E[NH ] (PSTS + (1− PS)TC) + TS (6.34)
where E[D] is the average number of slots required for doing a successful packet trans-
mission into the channel. According to SAA protocol, E[X] can also be the average
backo delay that the SU waits in generic before accessing the channel as our back-
o mechanism consists of all the relevant cases such as slot-counter decremented and
holding value.
e mean period of collisions as accounted in [49, 115] for delay analysis is not
relevant in our model. Since the sensing error has already been synthesized directly into
the backo process of the SAA protocol; thus, the entire eect of collisions is included
in dening the channel status with Pi and Pb. Furthermore, we assume that the delay
due to packet dropping is not relevant to this calculation as the packet is not successfully
6.5 Numerical Results 131
received. Based on our proposed protocol and its interpretation into the Markov chain,
E[X] can be measured by considering that the slot-counter requires k number of slots
to reach i,−1 state for transmission from i, k state and the time interval between
these transitions is quite random whose average is given by
E[X] =u∑i=0
ωi−1∑k=0
kπ(i, k) (6.35)
By using equations (6.15), (6.16), and (6.17), we obtain as follows,
E [X] =π(0, 0)
6
[ω2
0(1− (4P )u+1)
1− 4P− 1− P u+1
1− P
](6.36)
By puing the value of π(0, 0) from equation (6.18) into (6.36), we can obtain the E[X].
In (6.34), E[NH ] is the average number of times that the SU holds on its slot-counter
value due to the detection of transmission which is given by,
E[NH ] =E[X]− E[kH0 ]
E[kH0 ](6.37)
E[kH0 ] is the average number of idle slots before a transmission occurs which can be
obtained by
E[kH0 ] =1− P C(t) = TxP C(t) = Tx
(6.38)
Exploiting (6.24) and (6.36), E[NH ] can also be derived as a function of n and φn. By put-
ting the value of E[X] and E[NH ] into (6.34), the average access delay of our proposed
SAA protocol can be obtained.
6.5 Numerical Results
In this section, we evaluate the performance of SAA protocol in respect of physical-
layer sensing parameters Pf and Pm, and also for MAC contention parameters. We
consider that primary network operates with 6 MHz bandwidth in an AWGN channel
and secondary network follows the frame length as Tf = 10 ms. Numerical parameters
132 6.5 Numerical Results
Table 6.1: Parameters for Performance Analysis of SAA Protocol.
Parameters Value
Packet payload 8512 Bytes
MAC header 728 bits
PHY header 512 bits
RTS 450 bits+ PHY header
CTS 320 bits+ PHY header
ACK 320 bits+ PHY header
Channel bit rate 1 Mb/s
Propagation delay(δ) 1 µs
Slot time(τu) 50 µs
TDIFS 136 µs
TSIFS 28 µs
Size of CW(ω0) 4 ∼ 128
Maximum backo stage(u) 5
Number of SU (N ) 5 ∼ 50
used in this analysis are outlined in Table 6.1. e size of the regarding packets is given
in the unit of the bit which can be converted into time scale based on channel bit rate. In
this analysis, we assume that SU follows same bit rate both in control packet (RTS and
CTS) and data packet transmission.
6.5.1 Throughput and Delay Performance of Proposed SAA Pro-
tocol
e performance of throughput is analyzed based on the equation (6.33) where E[P ]
can be obtained in slot time using the numerical data provided in Table 6.1. Also, by
using the numerical data, the average slot time for successful transmission, collision,
and empty slot can be estimated based on the equations of TS , TC , and TI . We consider
that φ1 = · · · = φn = φ for a given cognitive radio network which only rely on cross-
layer based backomechanism. According to (6.29), Tser is then as a function of φ andN
for the estimated value of TS , TC , and TI . In the detector, the sensing error is related to
6.5 Numerical Results 133
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Probability of missed detection
No
rma
lize
d t
hro
ug
hp
ut
ω0 = 4 (A)
ω0 = 4 (S)
ω0 = 8 (A)
ω0 = 8 (S)
ω0 = 16 (A)
ω0 = 16 (S)
ω0 = 32 (A)
ω0 = 32 (S)
Figure 6.5: Characteristic of normalized throughput (S) corresponding toprobability of missed detection (Pm); where the parameters are:γ = −15 dB, fs = 6 MHz, PH1 = 0.1, u = 5, and N = 20.
Pf = Q(√
2γ + 1Q−1(1− Pm) + γ√τsfs
)according to (3.6) and (3.7). By employing
this relationship into (6.21) and (6.22), the access probabilityφ becomes a function ofPH1 ,
Pm, ω0, and u. For a given value of PH1 , ω0, and u, however, the normalized throughput
S also depends on the number of contenders in multiple access scenario.
In a multiple access scenario with a xed number of contenders N = 10, the vari-
ation of S with respect to Pm is illustrated in Fig. 6.5 for several values of minimum
contention window ω0. In this gure, there is a close consent between the analytical (A)
and simulated (S) results. When ω0 = 4, S decreasing exponentially with the increasing
of Pm which is an essential property of SAA protocol. Nevertheless, the variation of S is
advanced into the steady condition by increasing the size of contention window ω0 from
4 to 32, which implies that proposed SAA protocol achieves higher throughput even the
detector’s Pm is large. Note that larger the Pm value can cause severe interference to
the primary network. erefore, we should choose the operating characteristic of SAA
protocol when Pm is low. Fig. 6.5 also indicates that proposed SAA protocol achieves
higher throughput performance when the sensing error is small for a given ω0 which
can able to reduce interference to primary transmission.
134 6.5 Numerical Results
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
1
2
3
4
5
6
Probability of missed detection
Ave
rag
e a
cce
ss d
ela
y (
ms)
ω0 = 4 (A)
ω0 = 4 (S)
ω0 = 8 (A)
ω0 = 8 (S)
ω0 = 16 (A)
ω0 = 16 (S)
ω0 = 32 (A)
ω0 = 32 (S)
Figure 6.6: Characteristic of average access delay (E[D]) corresponding toprobability of missed detection (Pm); where the parameters are:γ = −15 dB, fs = 6 MHz, PH1 = 0.1, u = 5, and N = 20.
Fig. 6.6 shows the behavior of average access delay E[D] as a function of Pm based
on (6.34) for the same simulation conditions applied in measuring the S in Fig. 6.5. With
the increasing of Pm, E[D] also increases for a given ω0 but the rate of increment is
relatively steep for lower size of the contention window. For instance, E[D] increases
abruptly from 1 ms to 5 ms with the increasing of Pm when ω0 = 4. In contrast, for
larger contention window i.e., ω0 = 32, E[D] increases abruptly only when Pm ≤ 0.1
and sustains almost in the same time range even Pm increases. is behavior implies
that proposed SAA protocol is used the contention mechanism eciently with relatively
higher value of contention window to achieve less access delay for the secondary users.
e performance of SAA protocol is further evaluated with the behavior of collision
and access probability corresponding to the probability of missed detection as shown in
Fig. 6.7 and Fig. 6.8, respectively. In Fig. 6.7, the probability of collision (PC) during the
channel access is very low at the lower value of Pm which is desirable for cognitive radio
network; and PC increases traditionally with the increasing of missed detection. us,
the rate of collision due to the missed detection in physical sensing can be overcome by
6.5 Numerical Results 135
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Probability of missed detection
Pro
ba
bili
ty o
f co
llisio
n
For ω0 = 4 (A)
For ω0 = 8 (A)
For ω0 = 16 (A)
For ω0 = 32 (A)
For ω0 = 4 (S)
For ω0 = 8 (S)
For ω0 = 16 (S)
For ω0 = 32 (S)
Figure 6.7: Probability of collision (PC ) with respect to probability of misseddetection (Pm) of SAA protocol; where the parameters are: wherethe parameters are: γ = −15 dB, fs = 6MHz,PH1 = 0.1, u = 5,and N = 20.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.05
0.1
0.15
0.2
0.25
0.3
0.35
Probability of missed detection
Pro
ba
bili
ty o
f a
cce
ss
For ω0 = 4 (A)
For ω0 = 8 (A)
For ω0 = 16 (A)
For ω0 = 32 (A)
For ω0 = 4 (S)
For ω0 = 8 (S)
For ω0 = 16 (S)
For ω0 = 32 (S)
Figure 6.8: Probability of access (φ) with respect to probability of misseddetection (Pm) of SAA protocol; where the parameters are: γ =−15 dB, fs = 6 MHz, PH1 = 0.1, u = 5, and N = 20.
increasing the contention window as shown in Fig. 6.7. For instance, when Pm is 0.2 or
20% then there is an about 25% of chance of collision for ω0 = 4. is probability is
dropped down to below 10% when the size of contention window ω0 is increased which
136 6.5 Numerical Results
0 20 40 60 80 100 1200
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Contention window
No
rma
lize
d t
hro
ug
hp
ut
For Pm = 0.1 (A)
For Pm = 0.1 (S)
For Pm = 0.5 (A)
For Pm = 0.5 (S)
For Pm = 0.9 (A)
For Pm = 0.9 (S)
Figure 6.9: Normalized throughput (S) versus contention window (ω0) ofSAA protocol; where the parameters are: γ = −15 dB, fs = 6MHz, PH1 = 0.1, u = 5, and ω0 = 16.
is a signicant contribution of the proposed SAA protocol. Fig. 6.8 shows that access
probability for lower value of ω0 is higher than for higher value of ω0 in respect of Pm.
e small value of Pm means the lower chance of inter-network collision between the
primary and secondary network [3, 49, 65]. On the other hand, higher access probability
among a large number of contenders can reduce the normalized throughput as depicted
in contention based access [63, 66]. In this circumstances, we can say that proposed SAA
protocol can achieve stable access condition for a relatively large number of contention
window even though the Pm increases.
Fig. 6.9 describes the impact of the size of the contention window on the throughput
performance. At a target Pm, the normalized throughput S is increased by extending the
value of ω0. In particular, when the target Pm is set as 0.1, S reaches its maximum value
with the extending of ω0 and starts to decline slowly with the further extension of ω0.
us, there is an optimal value of ω0 to achieve the maximum throughput performance.
On the other hand, SAA protocol requires a comparatively larger value of ω0, when Pm
increases, to stable the throughput performance onto a maximum condition as depicted
in both Fig. 6.5 and Fig. 6.9.
6.5 Numerical Results 137
5 10 15 20 25 30 35 40 45 500
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Number of SU
No
rma
lize
d t
hro
ug
hp
ut
ω0 = 4 (A)
ω0 = 4 (S)
ω0 = 8 (A)
ω0 = 8 (S)
ω0 = 16 (A)
ω0 = 16 (S)
ω0 = 32 (A)
ω0 = 32 (S)
Figure 6.10: Variation of normalized throughput (S) corresponding to num-ber of SU (N ) in analytical and simulation cases; where theparameters are: γ = −15 dB, fs = 6 MHz, PH1 = 0.1, u = 5,and Pm = 0.1.
5 10 15 20 25 30 35 40 45 500
1
2
3
4
5
6
Number of SU
Ave
rag
e a
cce
ss d
ela
y (
ms)
ω0 = 8 (A)
ω0 = 8 (S)
ω0 = 16 (A)
ω0 = 16 (S)
ω0 = 32 (A)
ω0 = 32 (S)
Figure 6.11: Variation of average access delay (E[D]) corresponding tonumber of SU (N ) in analytical and simulation cases; wherethe parameters are: γ = −15 dB, fs = 6 MHz, PH1 = 0.1,u = 5, and Pm = 0.1.
evariation of performancemetrics duringmultiple access scenario is also depended
on the network size. Here the number of SUs dene the network size. Fig. 6.10 and Fig.
138 6.5 Numerical Results
6.11 illustrated the changing behavior of S and E[D] regarding N for several values of
ω0. It is observed that the simulation results are rmly agreed with the analytical results.
With the increasing of the number of contenders for access, the normalized throughput
decreases as shown in Fig. 6.10. e decrement is occurred largely for ω0 = 4 compared
to the case for ω0 = 32. Likewise, SAA protocol suers less E[D] by taking the larger
value of contentionwindow as shown in Fig. 6.11. In overall, SAA protocol requires large
contention window for improving both the throughput and delay performance when a
large number of users needed to be accommodated in multiple access.
6.5.2 Model Validation and Performance Comparison
To validate the analytical model of the SAA protocol, we analyzed and compared the
analytical result with the simulation result. Additionally, two approximations of the
access probability based on equation (6.23) and [66] are adopted in our throughput
calculation to examine the characteristic of S in respect of φ for a given number of
contenders.
Fig. 6.12 shows the variation of normalized throughput in respect of probability of
access φ for N = 20 and 50, which merely indicates that throughput decreases rapidly
when a large number of SUs are intended to access the channel. When there are 20
contenders, the throughput reaches its maximum level quite gradually and requires a
signicant value of φ compared with the case of N = 50. Also, S maintains relatively
higher throughput performance when N = 20, even through φ increases. On the other
hand, S reaches its maximum level with a very small value of φ, but it decreases more
rapidly when φ increases among 50 users.
For any given number of contenders, the simulated results are closely matched with
the analytical results in Fig. 6.12. e throughput achieved based on our proposed
approximation of φ is always lower than our exact throughput measurement for both
the size ofN which indicates that the exact measurement reveals the maximum range of
throughput that the secondary network can achieve during the multiple access. Another
approximation of φ follows the seminal work of [66]. e analytical model of S in
6.5 Numerical Results 139
equation (6.33), is very convenient to determine the maximum level of the achievable
throughput. Let us rearrange equation (6.33) as follows,
S =E[P ]
Tser/PS=E[P ]
Tden(6.39)
whereas S can be maximized while Tden of the above equation (6.39) is minimized. With
the help of [66], the approximated solution of φ ≈ 1/N√T ∗c /2 is obtained where
T ∗c = TC/τu. By applying this approximation of φ regarding N , we also evaluate S
without considering the sensing aspects in Fig. 6.12. By comparing the approximation
of φ among our proposed model (equation (6.23)) and [66], it is found that our proposed
approximation still outperforms the approximation made by [66] with a signicant mar-
gin always in the increasing range of φ. is comparison also implies that the proposed
SAA protocol follows the similar but improved characteristic of CSMA/CA protocol
corresponding to access probability for the cognitive radio network. is S versus φ
characteristic for several values of N exhibits the similar operational characteristic of
CSMA/CA protocol [66] which validate our proposed SAA protocol. Furthermore, the
higher value of S compared with [66] indicates the impact of sensing-assisted mechan-
ism in throughput improvement.
Finally, the throughput performance of proposed SAA protocol is compared with
other C-MAC protocols in Fig. 6.13. Distributed MAC protocol [6] and CR-CSMA [7]
are the most relevant MAC protocols for performance comparison due to their seminal
contributions in contention-based access for CR users which outperformed over other
protocols [48, 49, 57, 65]. e performance of distributed-MAC and CR-CSMA proto-
cols are computed under our simulation conditions to conduct a fair comparison with
our proposed SAA protocol. e normalized throughput of all protocols is compared
regarding the number of SUs for the contention window of 8 and 32. For both values of
the contention window, all the three protocols show a similar decreasing characteristic
of throughput corresponding to the increasing of contenders for channel accessing as
previously described by Fig. 6.10. However, proposed SAA protocol maintains com-
paratively higher throughput performance than the distributed-MAC and CR-CSMA
140 6.5 Numerical Results
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.090.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Probability of access
No
rma
lize
d t
hro
ug
hp
ut
Analytical, N = 20
Simulated, N = 20
Approximated (proposed), N = 20
Approximated ([12]), N = 20
Analytical, N = 50
Simulated, N = 50
Approximated (proposed), N = 50
Approximated ([12]), N = 50
Figure 6.12: Normalized throughput (S) versus probability of access (φ) withapproximation, simulation, and analytical results; where theparameters are: γ = −15 dB, fs = 6 MHz, PH1 = 0.1, u = 5,ω0 = 16, and Pm = 0.1.
5 10 15 20 25 30 35 40 45 500
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Number of SU
No
rma
lize
d t
hro
ug
hp
ut
Distributed MAC (ω = 8)
CR-CSMA/CA (ω = 8)
Proposed SAA (ω = 8)
Distributed MAC (ω = 32)
CR-CSMA/CA (ω = 32)
Proposed SAA (ω = 32)
Figure 6.13: Normalized throughput comparison among distributed-MAC[6], CR-CSMA [7], and our proposed SAA protocol with respectto number of SU. In this analysis, the using parameters are:γ = −15 dB, fs = 6 MHz, PH1 = 0.1, u = 5, and Pm = 0.1.
.
protocol with the increment of the number of users for both cases of ω0. When ω0 = 32,
6.6 Chapter Summary 141
SAA protocol achieves quite same S as achieved by distributed-MAC and CR-CSMA, but
SAA protocol outperforms the other protocols when the contender increases. In overall,
when a large number of contenders are required to be accommodated by using limited
contention window, then our proposed SAA protocols performs extremely well than
the other comparing protocols. Based on this thorough investigation, it can be outlined
that the proposed SAA protocol can achieve ecient throughput and delay performance
by synthesizing the PHY/MAC cross-layer parameters correctly under the presence of
sensing error as well as during the multiple access environment.
6.6 Chapter Summary
In this chapter, we proposed the SAA protocol by PHY/MAC cross-layer operation to
enhance the performance of CRN during multiple access. We developed a novel sensing-
assisted contention access including all the sensing with regard to the random backo
process. Furthermore, we derived and analyzed the SAA protocol where the aspect of
imperfect sensing is captured to investigate the consequence of physical-layer sensing
for the proper measurement of throughput and delay. Performance evaluation of numer-
ical results indicate that SAA protocol improves the throughput and delay performance
when the sensing error is tolerable; and by selecting the proper contention window,
the performance can be enhanced when sensing error and the number of contenders
are increasing in the network. By considering the imperfect sensing in the backo
process, the SU overcame the waste of spectrum opportunity by reducing the false alarm
in detection and the consequent increasing collision probability is compensated by the
wider contention window in our proposed SAA protocol. e model validation through
simulation results and performance comparison conrms the signicance of the pro-
posed SAA protocol for the improvement of the throughput and delay performance for
cognitive radio networks.
CHAPTER 7
Conclusion and Recommendationsfor Future Research
Mobile devices and connections are geing smarter with the capabilities of seamless
connectivity and mobile computing. e explosion of mobile connectivities and advance
multimedia applications demand fast and intelligent networks in wireless technologies.
e existing xed spectrum access policy strives to handle the ever-growing end-user
demand which leads to the spectrum scarcity problem in current wireless technology.
CR technology has emerged as a promising solution to spectrum scarcity based on the
idea of dynamic spectrum access. On the other hand, some of the allocated RF bands are
not utilized to their full potential. CR technology allows the unlicensed users to use the
underutilized portion of a licensed band while ensuring the necessary protection and
ecient utilization.
Spectrum sensing and access are two crucial components of the CR operation. With
the combined operation of these two components, SU can monitor the channel activity
and apply an appropriate transmission strategy to meet the goal of CR operation. To
increase the SU throughput performance, existing access protocols apply aggressive
transmission strategies leading to harmful interference to the legacy users. e ob-
jective of the research documented in this thesis was to propose the access strategy
incorporatedwith spectrum sensing to overcome the sensing-throughput trade-o issue.
e conducted research conrms, through comprehensive analysis and validation, that
the proposed methods and strategies outperform the state-of-the-art of the cognitive
radio network. e ndings and technical contributions accomplished throughput this
research are summarized in this chapter.
144 7.1 Conclusion
7.1 Conclusion
Cognitive radio networkswork according to a hierarchical accessmodelwhereby unused
portions of a licensed spectrum are open to SUs, provided PUs are protected by limiting
the interference from SUs’ transmission. Traditionally, the interference protection is
guaranteed by spectrum sensing designed to meet a target detection probability. SUs
can increase the throughput when the sensing period is short enough allowing for a
more prolonged access period. A short sensing period leads to a more substantial prob-
ability of false alarm, hence limiting spectrum opportunity and decreasing throughput.
is sensing-throughput trade-o issue cannot be overcome with advancement in the
underlying sensing and access operation when both of these operations are independ-
ently designed in the dierent layers. In addition, the sensing performance reects the
capacity and interference of the CR access. Hence, it was necessary to determine the
impact of sensing on the access protocol design’s capacity to satisfy the target of sensing-
throughput trade-o.
A statistical model of the spectrum sensing was established to analyze the impact
of detection performance in the realistic channel condition. e research focused on
the design of access strategy by exploiting the post-processing data of the sensing. e
sensing in this research is referred to as detector-independent sensing algorithm. How-
ever, energy detector and matched-lter are applied for signal detection to model the
sensing system. e performance parameters of the spectrum sensing are formulated
by applying the binary hypothesis testing problem. Dening the channel state based
on the detection is important to measure the capacity SU achieved by the spectrum
sensing. Apart from PU detection, it is also required to model the PU trac to measure
the probabilistic channel state. PU trac is modeled as a two-state random process with
Poisson distribution in state transitions of ON and OFF states. e steady-state probab-
ilities of the channel state are formulated by using a discrete time Markov chain process.
Finally, the spectrum opportunity achieved by spectrum sensing is formulated with the
probabilistic relationship between the PU occupancy status and detection performance.
Investigation into the spectrum opportunity reveals that the PFA has greater potential
7.1 Conclusion 145
than the PD in enhancing the opportunity. It is also assessed that single-level sensing
failed to produce the greatest spectrum opportunity as its detection experienced high
PFA. A dual-level sensing mechanism is proposed over the same sensing period used in
an SS mechanism by segmenting the target PD conditionally over that two sensing-level.
e overall PFA of the DS mechanism obtained is lower than the SS mechanism. e
access capability is characterized by receiver operating characteristic curve and access
probability. e ROC curve denes the maximum bounding of PFA and PD relation for
a given SNR. e ROC curve analysis consolidates that the proposed DS mechanism has
greater detection capability than conventional SS mechanism at a given SNR value. e
access probability analysis proved that the DS mechanism outperforms the SS mech-
anism by a wide margin with the fastest growing rate towards maximum capacity and
greater utilizing capability.
e eectiveness of theDSmechanism is capitalized on access operation by proposing
a dual-level sensing based multiple access (DSMA) protocol. In DSMA, the SU can access
the channel following a conditional second sensing once the channel is obtained as idle in
the rst sensing. e SU can defer the transmission aempt when the channel is sensed
as busy in any sensing steps and proceeds with a backo process. e backo process
is devised with a cross-layer integration of the physical detection and the contention
method. In the contention method, the backo process has the advantages in collision
reduction by deferring the transmission aempt with random delay. In the conventional
backo process, the predened distribution, such as a uniform and exponential distri-
bution, in the backo process characterizes the transmission aempt and its suitable
rate. Unlike the conventional backo process, the transmission aempt with random
delay is recongured by using detector parameters and distribution of the transmis-
sion aempt in the proposed model. e detector parameters impose controllability on
the random delay. Consequently, the backo process reduces the compulsion of the
spectrum sensing in collision reduction. is adaptive design of the backo process
and detection sensitivity eventually contributes to both throughput improvement and
collision reduction.
146 7.1 Conclusion
It is imperative to nd the impact of DS mechanism on the sensing-throughput op-
timization. In conventional SS mechanism, there is an optimal sensing period to obtain
maximum throughput for a given target PD. In the DS mechanism, on the contrary,
two target PDs in two conditional sensing levels need to be set in order to meet the
overall target PD. With the advantages the DS mechanism brings additional challenges
in nding the optimal sensing period by which maximum throughput can be achieved.
As the internal operation of DS mechanism is conditioned by the sensing decision, it is
relevant to use the entire sensing period and the PD at any one of the sensing steps, in
the optimization.
For a fair comparison between theDS and SSmechanism, the constraint of the sensing
period in DS mechanism is set as equivalent to the optimal sensing period of SS mech-
anism to obtain maximum throughput for a given PD. By applying convex analysis, the
feasibility of the minimum PFA is examined regarding the target PD of the rst sensing
step and the total sensing period. With a thorough convex analysis, it is proved that
there is a global minimum of the PFA regarding the operational range of the sensing
period and target PD of the rst sensing step. However, it is hard to obtain a closed-form
mathematical equation for minimum PFA due to the mathematical complexity. A semi-
analytical algorithm is then proposed to solve the optimization where the boundaries
of the optimizers are provided by the feasibility analysis. Furthermore, a numerical
method, i.e., backtracking line search algorithm, is applied with considerable complexity
for joint optimization and model validation. By employing the post-optimization data
into the system, the DS mechanism achieves higher throughput than the SS mechanism
in a given channel condition.
A novel sensing-assisted access (SAA) protocol is proposed as a complete random
access mechanism for the secondary users. e sensing feature is integrated inside of
the backo process to enhance the capabilities of CR operation in reducing the packet
collision. e access contention, i.e., the sensing-embedded backo process, is modeled
byMarkov chain in the presence of sensing error. Conventional contention-based access
with backo process does not reect the original cause behind the packet collision and
7.2 Recommendations for Future Research 147
relies only upon the acknowledgment to determine the collision. e state character-
ization in the conventional backo process considers the perfect sensing, and hence
cannot accurately reect the interference to PU. A novel backo process is developed by
integrating the backo and sensing parameters for state characterization. e obtained
state characteristic reects the sensing error rendering for the packet collision. With
a proper choice of backo parameters, i.e., by increasing the contention window, the
collision probability is reduced signicantly leading to throughput improvement. In
essence, SAA protocol maximizes the throughput performance of the secondary users
and simultaneously ensures sucient interference protection to the primary user.
7.2 Recommendations for Future Research
is research concentrated on developing solutions for the access strategy to overcome
the sensing-throughput trade-o issue and was less focused on the issue of the energy
eciency. As such, proposed access strategies have mostly relied on the dual steps of the
sensing operation which may consume relatively higher power than the conventional
case. e proposed SAA protocol applied continuous sensing operation during the con-
tention access period for a transmission aempt. It is shown that the SAA protocol
required a much smaller access delay for any transmission aempt when compared
to the existing methods. For a shorter access delay, the access protocol may consume
energy for a shorter period. However, the performance of the proposed access strategies
can be further evaluated in the context of energy eciency.
Energy eciency of a protocol is also related to the transmission power of the SU.
Transmission power control is an important issue for improving not only the energy
eciency but also the CR capability by limiting the interference power to the PU. For
example, the interference protection in the systemmodel used in this research is demon-
strated through the target PD (equivalent to missed detection) which is the maximum
bound of the interference. If transmission power of the SU is related to the detection er-
ror, then the expected interference limit can be obtained that is lower than the maximum
interference. By considering the maximum level of the interference, the conducted re-
148 7.2 Recommendations for Future Research
search exposed the normalized capacity of the access protocol, but there is an additional
dimension in the resource block, i.e., power, to emphasize the achievable capacity limit
in the context of energy eciency.
e underlying detection method in the spectrum sensing algorithm was based on
energy detection due to its lower complexity and compatibility along the CR operations.
In the proposed DS mechanism, two thresholds are chosen based on their target PD
of the sensing steps for making the nal decision. When double threshold values are
applied to a binary hypothesis testing problem, then traditionally there is a region of
confusion between two threshold values in the probability distribution of the perform-
ance function. Hence, decision uncertainty for the samples laid down in the confusion
region could be taken into account for further research of the DS mechanism.
e solution approaches to optimization in Chapter 4 were based on semi-analytical
and pure numerical methods. In the proposed semi-analytical algorithm, the boundaries
of the feasible region were determined with the help of precise mathematical derivation.
e optimization was accomplished iteratively within analytical boundaries by using a
numerical method. ere is potential to enhance the computational eciency of this
algorithm. rough applying certain approximations, it may be possible to develop an
entirely analytical approach with closed-form mathematical solution.
e multi-channel scenarios can be recommended for further enhancement of the
capability of the proposed access protocol. is multi-channel network can provide fur-
ther exibility and access reliability with higher throughput and lower delay. In addition,
channel assignment algorithms with multi-channel sensing features have to be taken
into account for developing the multi-channel capability. Overall, there are countless
research challenges relating to spectrum access in the cognitive radio networks. e
above recommendations are only a few possible candidates, and the research presented
in this thesis can be expanded in relevant directions to construct ecient cognitive radio
networks for the deployment of future generation networks.
APPENDIX A
Proof of Propositions and Theorems
A.1 Proof of Proposition 4.2
For this proof, we assume that Pd1 = x and 1− Pf1(Pd1) = f1(x). Note that a function
f1(x) is said to be log-concave while h′(x) is monotonically decreasing function with
respect to the dened range of x, so as h′(x) < 0 where h (x) = f1′(x)
f1(x). Now taking the
rst dierentiation of h(x) yields,
h′(x) =f1(x)f ′′1 (x)− (f ′1(x))2
(f1(x))2 (A.1)
Using equation (4.16), (4.17), and (4.19), h′(x) is derived as,
h′(x) = −Φ3
√2γ + 1
(1− Pf1)2 exp
[2w2
d1− w2
f1
](A.2)
where we assume that Φ3 =√π(1 − Pf1)(
√2γ + 1wf1 − wd1) + exp[−w2
f1]. Based on
equation (A.2), it can be stated that h′(x) < 0 when Φ3 > 0. Simply, it can be said that
Φ3 > 0 as previously we proved that
√2γ + 1wf1 − wd1 > 0 for Pd1 ∈ [0, Pd1(θ1)].
However, when Pd1 → 0 then Φ3 is undened. erefore, we compute the bounding of
Pd1 for which Φ3 has denite and non-negative values, from the following inequality,
√π
2γ + 1(1− Pf1)
(2γwf1 + γ
√Ns
)+ exp
[−w2
f1
]> 0 (A.3)
Note thatddtQ(t) = −2e−t
2/√π is considered which is equivalent to −e−t2/2/
√2π for
t > 0; thus, the upper bound of Q(t) will be√
2e−t2. Likewise, puing the maximum
150 A.2 Proof of Proposition 4.4
value of Pf1 = Q(wf1) into equation (A.3) and obtain as below,
√π
2γ + 1
(2γwf1 + γ
√Ns
)+ exp
[−w2
f1
](1−
√2π
2γ + 1
(2γwf1 + γ
√Ns
))> 0 (A.4)
where
√π/(2γ + 1)(2γwf1 + γ
√Ns) always non-negative for any given value ofNs, γ.
us, by checking the remaining terms of above equation, we obtain
exp[−w2
f1
]> 0 or, 1−
√2π
2γ + 1
(2γwf1 + γ
√Ns
)> 0 (A.5)
ere is no real solution of this exp[−w2f1
] > 0 inequality, so the remaining term of
equation (A.5) becomes
Pd1 > Q
(1
2γ√
2π−√Ns(2γ + 1)
2
)(A.6)
Previously we found that Pd1(θ1) < Q(θ1) where θ1 = −√Ns(2γ + 1)/2. Similarly,
assume that Pd1(θ2) > Q(θ2) where θ2 = 1/(2γ√
2π) −√Ns(2γ + 1)/2. Comparing
these two assumptions, we found that θ2−θ1 = 1/(2γ√
2π) > 0 therebyQ(θ2) < Q(θ1)
and Pd1(θ2) is the upper bound for which Φ3 > 0. So, h′(x) < 0 and monotonically
decreasing function. Hence, (1 − Pf1) is a log-concave function of Pd1 for the range of
Pd1 ∈ [0, Pd1(θ2)]. us Proposition 4.2 is proved.
A.2 Proof of Proposition 4.4
We follow the similar conditions and assumptions dened in Proposition 2 to prove the
log-concavity of Pf2 . Aer taking the rst dierentiation of h2(x) =f ′2(x)
f2(x)when the
f2(x) = Pf2(Pd1), we obtain as follows,
h′2(x) = −Φ5
√2γ + 1
(1− PD
)P 2f2
(1− Pd1)3 exp
[w2d2− w2
f2
](A.7)
A.3 Proof of Proposition 4.5 151
by assuming that
Φ4 = 2 +
√π(1−PD)(1−Pd1)
(wd2 − wf2
√2γ + 1
)exp
[w2d2
]Φ5 = Pf2Φ4 −
√2γ+1(1−PD)(1−Pd1)
exp[w2d2− w2
f2
] (A.8)
By applying similar approach of the proof of proposition 2 and taking the approximated
maximum limit of Pf2 = Q (wf2), we obtain the following condition for which Φ5 > 0,
exp[−w2
f2
]> 0 or,
√2π(wd2 − wf2
√2γ + 1
)−√
2γ + 1 > 0 (A.9)
e rst term is undened. erefore, nally we get the bounding of Pd1 for which
Φ5 > 0, as follows
Pd1 <PD − Pd1(θ4)
1− Pd1(θ4)(A.10)
where,
Pd1(θ4) = Q(−√
2γ+12γ
(γ√Nc + 1√
2π
))In the range of Pd1 ∈ [0, Pd1(θ5)], h′2(x) < 0 so Pf2 is a log-concave function for that
range where Pd1(θ5 =(PD − Pd1(θ4)
)/ (1− Pd1(θ4)). us, Proposition 4.4 is proved.
A.3 Proof of Proposition 4.5
Similar to the solution approach of Proposition 4.1, we obtain dPf2dPd1
for ED-MF combin-
ation by using MF [44], as follows
dPf2dPd1
= −(1− PD
)(1− Pd1)
2 exp[w2d2− w2
f2
](A.11)
152 A.4 Proof of eorem 4.1
Taking the another dierentiation of (A.11) with respect to Pd1 ,
d2Pf2dP 2
d1
= −(1− PD
)(1− Pd1)
3 exp[w2d2− w2
f2
]×
[2−√π(1− PD
)(1− Pd1)
(wf2 − wd2) exp[w2d2
]](A.12)
As we know thatQ(wd2) > Q(wf2) so wd2 < wf2 . Also, for 0 < Pd1 < 1, the last term is
not greater than 1. erefore, the last term of the above equation is non-negative which
implies thatd2Pf2dP 2
d1
< 0. us, Pf2 is a concave function of Pd1 .
A.4 Proof of Theorem 4.1
Let us consider x∗ be the optimal solution of (4.27). en we obtain µ+F2(x) ≥ F1(x)+
F2(x) ≥ F1(x∗) + F2(x∗), ∀µ ≥ F1(x). However, above estimation implies that Ψ(z) ≥
F1(x∗) + F2(x∗) = µ∗ + F2(x∗), where µ∗ = F1(x). erefore, there exists an optimal
point z∗ = µ∗ = F1(x) ∈ Ω such that Ψ(z) ≥ Ψ(z∗) ∀z ∈ Ω. is proves the necessary
condition as stated ineorem 4.1.
A.5 Proof of Theorem 4.2
Let xo be a ξ-minimum critical point of the function F on En, then from (4.31) it follows
that 0 ∈ w +(∂∂x
)ξF1(xo), ∀w ∈ F2(xo). Hence,
min‖g‖=1
maxz∈w+(∂/∂x)ξF1(xo)
(z, g) ≥ 0, ∀w ∈ F2(xo)
and thus for every g ∈ En, ‖g‖ = 1, we have
minw∈(∂/∂x)F1(xo)
maxv∈(∂/∂x)ξF1(xo)
(z, g) ≥ 0
However, this means that
min‖g‖=1
(∂
∂x
)ξ
F (xo) ≥ 0 (A.13)
A.6 Proof of eorem 4.3 153
proving that the condition is necessary. at it is also sucient can be demonstrated in
an analogous way, arguing backwards from the inequality (A.13).
A.6 Proof of Theorem 4.3
Assume that xo is not a ξ-minimum critical point. en, we can describe the vector
gξ(xo) = arg min‖g‖=1
(∂
∂x
)ξ
F (xo)
as a direction of ξ-steepest-descent of function F at point xo and numerically that dir-
ection is
gξ = −(
voξ + wo‖voξ + wo‖
)(A.14)
where voξ ∈(∂∂x
)ξF (xo), wo ∈
(∂∂x
)F (xo) and
− maxw∈
(∂∂x
)F1(xo)
minv∈
(∂∂x
)ξ
F1(xo)
‖v + w‖
= −‖voξ + wo‖
= aξ(xo)
is a direction of ξ-steepest-descent of function F at point xo. is satises the condition
given in the theorem 4.3.
A.7 Proof of Proposition 4.6
Let consider that probability of false alarm for energy detector is a convex function with
respect to sensing period as benchmark for proving this proposition which is proved by
Liang et al. in [3]. Let us take required partial derivative of R0 with respect to τs and
obtain as follows ∂R0
∂τs= 1 + (τds − τs)
∂Pf1∂τs− Pf1
∂2R0
∂τ2s= −2
∂Pf1∂τs
+ (τds − τs)∂2Pf1∂τ2s
(A.15)
154 A.8 Proof of Proposition 4.7
Liang et al. proved that Pf1(τs) is a convex function while Pf1 ≤ 0.5 [3], so obviously
∂Pf1∂τs≤ 0. Hence, from the above equations, we can say that
∂2R0
∂τ2s≥ 0. us, R0 is a
convex function of τs. Conversely, we can say that there has feasible maximum value of
R0 corresponding to τs. e converse property will be also true for remaining sensing
period in τds even though we check the convexity with respect to rst sensing period.
A.8 Proof of Proposition 4.7
Let τ =√
2γ + 1Q−1(P ∗d1) + γ√τsfs then Pf1 = Q(τ), thus R0 is changed to
R0(τ) =(τb− c)2
+
(τds −
(τb− c)2)Q (τ) (A.16)
where we assume that, a =√
2γ + 1, b = γ√fs, and c = aQ−1(P ∗d1)/b. As τds > τs
therefore, (τ/b− c)2 > 0. Hence, the rst term of equation (A.16 ) is obviously a convex
function of τ . Nowwe need to prove that second term of equation (A.16) is also a convex
function of τ . Let the second term is expressed as,
Φ(τ) = τsQ(τ)−(τb− c)2
Q(τ) (A.17)
Taking the second derivative of Φ(τ),
∂2Φ
∂τ 2=
(τs −
(τb− c)2)∂2Q(τ)
∂τ 2
+
(−4
b
(τb− c)) ∂Q(τ)
∂τ+
(− 2
b2
)Q(τ) (A.18)
Based on equation (31) and τds > τs, if we can show that (−4/b (τ/b− c)) (∂/∂τ)Q(τ)
+(−2/b2)Q(τ) ≥ 0 then (∂2/∂τ 2)Φ ≥ 0. According to Cherno boundsQ(τ) ≤ e−τ2/2
and (∂/∂τ)Q(τ) = −e−τ2/2/√
2π, we obtain the following inequality for (∂2/∂τ 2)Φ ≥
0 as (−4
b
(τb− c)) ∂Q(τ)
∂τ+
(− 2
b2
)Q(τ) ≥ 0
A.8 Proof of Proposition 4.7 155
Here, Pd1 at the rst sensing is generally bounded with 0.5 ≤ Pd1 < PD, therefore
Q−1(Pd1) ≤ 0. By recalling the values of a, b, and c into the above inequality, we obtain
the following condition as,
τ ≥√π
2+√
2γ + 1Q−1(Pd1) (A.19)
As τ = Q−1(Pf1), and γ > 0, equation (A.19) implies that if
Pf1 ≤ Q(√
π/2 +Q−1(P ∗d1))
(A.20)
then Φ(τ) is convex. us, the proposition is proved.
Bibliography
[1] I. F. Akyildiz, W.-Y. Lee, M. C. Vuran, and S. Mohanty, “NeXt generation/dynamic
spectrum access/cognitive radiowireless networks: A survey,” Computer networks,
vol. 50, no. 13, pp. 2127–2159, 2006.
[2] T. Yucek and H. Arslan, “A survey of spectrum sensing algorithms for cognitive
radio applications,” IEEE Communications Surveys Tutorials, vol. 11, no. 1, pp. 116–
130, First 2009.
[3] Y. C. Liang, Y. Zeng, E. C. Y. Peh, andA. T. Hoang, “Sensing-throughput tradeo for
cognitive radio networks,” IEEE Transactions on Wireless Communications, vol. 7,
no. 4, pp. 1326–1337, April 2008.
[4] G. P. Joshi, S. Y. Nam, and S. W. Kim, “Cognitive radio wireless sensor networks:
Applications, challenges and research trends,” Sensors (Basel), vol. 13, no. 9, pp.
11 196 – 228, September 2013.
[5] S. Y. Lien, C. C. Chien, H. L. Tsai, Y. C. Liang, and D. I. Kim, “Congurable 3GPP
licensed assisted access to unlicensed spectrum,” IEEE Wireless Communications,
vol. 23, no. 6, pp. 32–39, December 2016.
[6] L. T. Tan and L. B. Le, “Distributed MAC protocol for cognitive radio networks:
Design, analysis, and optimization,” IEEE Transactions on Vehicular Technology,
vol. 60, no. 8, pp. 3990–4003, Oct 2011.
[7] Q. Chen, Y. C. Liang, M. Motani, and W. C. Wong, “A two-level MAC protocol
strategy for opportunistic spectrum access in cognitive radio networks,” IEEE
Transactions on Vehicular Technology, vol. 60, no. 5, pp. 2164–2180, Jun 2011.
158 BIBLIOGRAPHY
[8] J. Drumm, N. White, and M. Swiegers, “Mobile consumer survey 2016, the aus-
tralian cut - hyper connectivity: Clever consumption,” Deloie Touche Tohmatsu
Limited, Survey, 2016.
[9] J. C. Hawkins, J. C. Tu, R. Y. Haitani, C. L. Cadwell, and K. A. Townsend, “Mobile
computer system designed for wireless communication expansion,” US Patent
US6 516 202 B1, 2003.
[10] R. Robertson, E. Williams, S. Maes, T. Twerdahl, and C. Stone, “System and
method for managing wireless communications utilizing a mobile device,” US
Patent US7 812 817 B2, 2010.
[11] “Wireless technology evolution towards 5G: 3GPP release 13 to resealse 15 and
beyond,” 5G Americas, February 2017. [Online]. Available: www.5gamericas.org
[12] Iternational Telecommunication Union (ITU). United Nations Specialized Agency.
[Online]. Available: hp://www.itu.int
[13] Guidelines for the preparation of a national table of frequency allocations
(NTFA). Telecommunication Development Sector, International Telecommunic-
ation Union (ITU).
[14] Australian Communications and Media Authority (ACMA). Australian
Government. [Online]. Available: hps://www.acma.gov.au/
[15] Federal Communications Commission (FCC). Government agency, United States.
[Online]. Available: hps://www.fcc.gov/
[16] B. Scan, “Father of the cell phone,”e Economist, Technologyarterly, no. 2, June
2009.
[17] M. Cooper, “e myth of spectrum scarcity,” A Martin Cooper Position Paper,
March 2010. [Online]. Available: hps://ecfsapi.fcc.gov/le/7020396128.pdf
[18] T. W. Hazle, e Political Spectrum: e Tumultuous Liberation of Wireless
Technology, from Herbert Hoover to the Smartphone. Yale University Press, 2017.
BIBLIOGRAPHY 159
[19] Cisco, “Cisco visual networking index: Global mobile data trac forecast update,
2016 - 2021,” February 2017. [Online]. Available: hps://www.cisco.com
[20] ——, “Cisco visual networking index: forecast and methodology, 2016 - 2021,”
2017. [Online]. Available: hps://www.cisco.com
[21] G. Staple and K. Werbach, “e end of spectrum scarcity,” IEEE Spectrum, vol. 41,
no. 3, pp. 45–52, Mar 2004.
[22] E. Odiodu and M. Giles, “e 5G era: Age of boudless connectivity and intelligent
automation,” GSMA Intelligence, 2017. [Online]. Available: www.gsma.com
[23] T. M. Taher, R. B. Bacchus, K. J. Zdunek, and D. A. Roberson, “Long-term spectral
occupancy ndings in Chicago,” in 2011 IEEE International Symposium on Dynamic
Spectrum Access Networks (DySPAN), May 2011, pp. 100–107.
[24] “Federal communications commission strategic plan 2015 - 2018,” Federal Com-
munications Commission (FCC), US, Tech. Rep.
[25] V. Kone, L. Yang, X. Yang, B. Y. Zhao, and H. Zheng, “On the feasibility
of eective opportunistic spectrum access,” in Proceedings of the 10th ACM
SIGCOMM Conference on Internet Measurement, ser. IMC ’10. New York, NY,
USA: ACM, 2010, pp. 151–164. [Online]. Available: hp://doi.acm.org/10.1145/
1879141.1879160
[26] M. A. McHenry, P. A. Tenhula, D. McCloskey, D. A. Roberson, and C. S. Hood,
“Chicago spectrum occupancy measurements, analysis and a long-term studies
proposal,” in Proceedings of the First International Workshop on Technology and
Policy for Accessing Spectrum, ser. TAPAS ’06. New York, NY, USA: ACM, 2006.
[Online]. Available: hp://doi.acm.org/10.1145/1234388.1234389
[27] “Five-year spectrum outlook 2016-20,” Australian Communications and Media
Authority (ACMA), Tech. Rep., October 2016.
160 BIBLIOGRAPHY
[28] “e next generation (XG) program,” e Defense Advanced Research Projects
Agency (DARPA), Tech. Rep., 2005. [Online]. Available: hp://www.darpa/mil/
ato/programs/xg/index.htm
[29] J. Mitola and G. Q. Maguire, “Cognitive radio: making soware radios more
personal,” IEEE Personal Communications, vol. 6, no. 4, pp. 13–18, Aug 1999.
[30] S. Haykin, “Cognitive radio: brain-empowered wireless communications,” IEEE
Journal on Selected Areas in Communications, vol. 23, no. 2, pp. 201–220, Feb 2005.
[31] I. F. Akyildiz, W.-Y. Lee, and K. R. Chowdhury, “CRAHNs: Cognitive radio ad hoc
networks,” Ad Hoc Networks, vol. 7, no. 5, pp. 810 – 836, 2009. [Online]. Available:
hp://www.sciencedirect.com/science/article/pii/S157087050900002X
[32] A. Goldsmith, S. A. Jafar, I. Maric, and S. Srinivasa, “Breaking spectrum gridlock
with cognitive radios: An information theoretic perspective,” Proceedings of the
IEEE, vol. 97, no. 5, pp. 894–914, May 2009.
[33] S. Geirhofer, L. Tong, and B. M. Sadler, “Cognitive radios for dynamic spectrum
access - dynamic spectrum access in the time domain: Modeling and exploiting
white space,” IEEE Communications Magazine, vol. 45, no. 5, pp. 66–72, May 2007.
[34] H. Kim and K. G. Shin, “Ecient discovery of spectrum opportunities with MAC-
layer sensing in cognitive radio networks,” IEEE Transactions onMobile Computing,
vol. 7, no. 5, pp. 533–545, May 2008.
[35] Z. Wang, D., T. Jiang, and T. Jin, “Ecient discovery of spectrum opportunities
via adaptive collaborative spectrum sensing in cognitive radio networks,” in 2011
IEEE International Conference on Communications (ICC), June 2011, pp. 1–5.
[36] G. Haab andM. Ibnkahla, “Multiband spectrum access: Great promises for future
cognitive radio networks,” Proceedings of the IEEE, vol. 102, no. 3, pp. 282–306,
March 2014.
BIBLIOGRAPHY 161
[37] O. Altrad, S. Muhaidat, A. Al-Dweik, A. Shami, and P. D. Yoo, “Opportunistic
spectrum access in cognitive radio networks under imperfect spectrum sensing,”
IEEE Transactions on Vehicular Technology, vol. 63, no. 2, pp. 920–925, Feb 2014.
[38] X. Liu and S. S. N., “Sensing-based opportunistic channel access,”Mobile Networks
and Applications, vol. 11, no. 4, pp. 577–591, Aug 2006. [Online]. Available:
hps://doi.org/10.1007/s11036-006-7323-x
[39] S. Haykin, D. J. omson, and J. H. Reed, “Spectrum sensing for cognitive radio,”
Proceedings of the IEEE, vol. 97, no. 5, pp. 849–877, May 2009.
[40] S. Stotas and A. Nallanathan, “On the throughput and spectrum sensing en-
hancement of opportunistic spectrum access cognitive radio networks,” IEEE
Transactions on Wireless Communications, vol. 11, no. 1, pp. 97–107, January 2012.
[41] E. Axell, G. Leus, E. G. Larsson, and H. V. Poor, “Spectrum sensing for cognitive
radio : State-of-the-art and recent advances,” IEEE Signal Processing Magazine,
vol. 29, no. 3, pp. 101–116, May 2012.
[42] S. Atapau, C. Tellambura, and H. Jiang, “Energy detection based cooperative
spectrum sensing in cognitive radio networks,” IEEE Transactions on Wireless
Communications, vol. 10, no. 4, pp. 1232–1241, April 2011.
[43] ——, Conventional Energy Detector. New York, NY: Springer New York, 2014, pp.
11–26. [Online]. Available: hps://doi.org/10.1007/978-1-4939-0494-5 2
[44] S. Kapoor, S. Rao, and G. Singh, “Opportunistic spectrum sensing by employing
matched lter in cognitive radio network,” in 2011 International Conference on
Communication Systems and Network Technologies, June 2011, pp. 580–583.
[45] A. M. Mossaa and V. Jeoti, “Cognitive radio: Cyclostationarity-based classication
approach for analog TV and wireless microphone signals,” in 2009 Innovative
Technologies in Intelligent Systems and Industrial Applications, July 2009, pp. 107–
111.
162 BIBLIOGRAPHY
[46] D. Bhargavi and C. R. Murthy, “Performance comparison of energy, matched-lter
and cyclostationarity-based spectrum sensing,” in 2010 IEEE 11th International
Workshop on Signal Processing Advances in Wireless Communications (SPAWC),
June 2010, pp. 1–5.
[47] K. L. Du andW. H. Mow, “Aordable cyclostationarity-based spectrum sensing for
cognitive radio with smart antennas,” IEEE Transactions on Vehicular Technology,
vol. 59, no. 4, pp. 1877–1886, May 2010.
[48] Q. Zhao, L. Tong, A. Swami, and Y. Chen, “Decentralized cognitive MAC for
opportunistic spectrum access in ad hoc networks: A POMDP framework,” IEEE
Journal on Selected Areas in Communications, vol. 25, no. 3, pp. 589–600, April 2007.
[49] W. Zhang, C. K. Yeo, and Y. Li, “A MAC sensing protocol design for data
transmission with more protection to primary users,” IEEE Transactions on Mobile
Computing, vol. 12, no. 4, pp. 621–632, April 2013.
[50] W. Han, J. Li, Z. Tian, and Y. Zhang, “Dynamic sensing strategies for ecient
spectrum utilization in cognitive radio networks,” IEEE Transactions on Wireless
Communications, vol. 10, no. 11, pp. 3644–3655, November 2011.
[51] C. L. Wang, H. W. Chen, and Z. Y. Tsai, “roughput maximization for cognitive
radio networks with wideband spectrum sensing,” in 2012 IEEE Wireless Commu-
nications and Networking Conference (WCNC), April 2012, pp. 1293–1298.
[52] J. Jafarian and K. A. Hamdi, “Non-cooperative double-threshold sensing scheme:
A sensing-throughput tradeo,” in 2013 IEEE Wireless Communications and Net-
working Conference (WCNC), April 2013, pp. 3376–3381.
[53] T. E. Bogale, L. Vandendorpe, and L. B. Le, “Wide-band sensing and optimization
for cognitive radio networks with noise variance uncertainty,” IEEE Transactions
on Communications, vol. 63, no. 4, pp. 1091–1105, April 2015.
BIBLIOGRAPHY 163
[54] 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.
[55] X. Zhang and H. Su, “CREAM-MAC: Cognitive radio-enabled multi-channel MAC
protocol over dynamic spectrum access networks,” IEEE Journal of Selected Topics
in Signal Processing, vol. 5, no. 1, pp. 110–123, Feb 2011.
[56] X. Hong, J. Wang, C. X. Wang, and J. Shi, “Cognitive radio in 5G: a perspective
on energy-spectral eciency trade-o,” IEEE Communications Magazine, vol. 52,
no. 7, pp. 46–53, July 2014.
[57] T. Z. Oo, N. H. Tran, D. N. M. Dang, Z. Han, L. B. Le, and C. S. Hong, “OMF-MAC:
An opportunistic matched lter-based MAC in cognitive radio networks,” IEEE
Transactions on Vehicular Technology, vol. 65, no. 4, pp. 2544–2559, April 2016.
[58] Z. Wu and Q. Zhao, “Sensing-throughput tradeo of relay-assisted random
broadcast based cognitive radio networks,” in 2013 IEEE 77th Vehicular Technology
Conference (VTC Spring), June 2013, pp. 1–5.
[59] S. Stotas and A. Nallanathan, “Overcoming the sensing-throughput tradeo in
cognitive radio networks,” in 2010 IEEE International Conference on Communica-
tions, May 2010, pp. 1–5.
[60] M. Cardenas-Juarez and M. Ghogho, “Spectrum sensing and throughput trade-
o in cognitive radio under outage constraints over Nakagami fading,” IEEE
Communications Leers, vol. 15, no. 10, pp. 1110–1113, October 2011.
[61] H. Pradhan, S. S. Kalamkar, and A. Banerjee, “Sensing-throughput tradeo in
cognitive radio with random arrivals and departures of multiple primary users,”
IEEE Communications Leers, vol. 19, no. 3, pp. 415–418, March 2015.
[62] L. Tang, Y. Chen, E. L. Hines, and M. S. Alouini, “Eect of primary user trac on
sensing-throughput tradeo for cognitive radios,” IEEE Transactions on Wireless
Communications, vol. 10, no. 4, pp. 1063–1068, April 2011.
164 BIBLIOGRAPHY
[63] Q. Chen, W. C. Wong, M. Motani, and Y. C. Liang, “MAC protocol design and
performance analysis for random access cognitive radio networks,” IEEE Journal
on Selected Areas in Communications, vol. 31, no. 11, pp. 2289–2300, November
2013.
[64] S. Zhang, H. Zhao, S. Wang, and J. Wei, “A cross-layer rethink on the sensing-
throughput tradeo for cognitive radio networks,” IEEE Communications Leers,
vol. 18, no. 7, pp. 1226–1229, July 2014.
[65] J. Lai, E. Dutkiewicz, R. P. Liu, and R. Vesilo, “Opportunistic spectrum access with
two channel sensing in cognitive radio networks,” IEEE Transactions on Mobile
Computing, vol. 14, no. 1, pp. 126–138, Jan 2015.
[66] G. Bianchi, “Performance analysis of the IEEE 802.11 distributed coordination
function,” IEEE Journal on Selected Areas in Communications, vol. 18, no. 3, pp.
535–547, March 2000.
[67] G. Bianchi and I. Tinnirello, “Remarks on IEEE 802.11 DCF performance analysis,”
IEEE Communications Leers, vol. 9, no. 8, pp. 765–767, Aug 2005.
[68] Q. Chen, Y. C. Liang, M. Motani, and W. C. Wong, “CR-CSMA: A random access
MAC protocol for cognitive radio networks,” in 2009 IEEE 20th International
Symposium on Personal, Indoor and Mobile Radio Communications, Sept 2009, pp.
486–490.
[69] R. W. omas, D. H. Friend, L. A. Dasilva, and A. B. Mackenzie, “Cognitive
networks: adaptation and learning to achieve end-to-end performance objectives,”
IEEE Communications Magazine, vol. 44, no. 12, pp. 51–57, Dec 2006.
[70] D. Cabric, S. M. Mishra, and R. W. Brodersen, “Implementation issues in spectrum
sensing for cognitive radios,” in Conference Record of the irty-Eighth Asilomar
Conference on Signals, Systems and Computers, 2004., vol. 1, Nov 2004, pp. 772–776
Vol.1.
BIBLIOGRAPHY 165
[71] J.-P. Bonin, C. Evci, and A. L. Sanders, “Securing spectrum through the
ITU to fuel the growth of next-generation wireless technologies,” Bell Labs
Technical Journal, vol. 18, no. 2, pp. 99–115, September 2013. [Online]. Available:
hp://dx.doi.org/10.1002/bltj.21607
[72] D. J. Lee, “Adaptive random access for cooperative spectrum sensing in cognitive
radio networks,” IEEE Transactions on Wireless Communications, vol. 14, no. 2, pp.
831–840, Feb 2015.
[73] B. Wang and K. J. R. Liu, “Advances in cognitive radio networks: A survey,” IEEE
Journal of Selected Topics in Signal Processing, vol. 5, no. 1, pp. 5–23, Feb 2011.
[74] D. Gzpek and F. Alagz, “roughput and delay optimal scheduling in cognitive
radio networks under interference temperature constraints,” Journal of Commu-
nications and Networks, vol. 11, no. 2, pp. 148–156, April 2009.
[75] J. S. Pang, G. Scutari, D. P. Palomar, and F. Facchinei, “Design of cognitive radio
systems under temperature-interference constraints: A variational inequality
approach,” IEEE Transactions on Signal Processing, vol. 58, no. 6, pp. 3251–3271,
June 2010.
[76] F. Benedeo, G. Giunta, E. Guzzon, M. Renfors, and M. Arcangeli, “Improving
the interference temperature estimation for dynamic spectrum access in cognitive
radios,” in 2013 IEEE Global Conference on Signal and Information Processing, Dec
2013, pp. 1154–1157.
[77] Y. Zeng and Y. C. Liang, “Maximum-minimum eigenvalue detection for cognitive
radio,” in 2007 IEEE 18th International Symposium on Personal, Indoor and Mobile
Radio Communications, Sept 2007, pp. 1–5.
[78] Part 22: Cognitive Wireless RAN Medium Access Control (MAC) and Physical Layer
(PHY) Specications: Policies and Procedures for Operation in the TV Bands, IEEE
Computer Society Std., July 2011.
166 BIBLIOGRAPHY
[79] T. M. Chiwewe, C. F. Mbuya, and G. P. Hancke, “Using cognitive radio for
interference-resistant industrial wireless sensor networks: An overview,” IEEE
Transactions on Industrial Informatics, vol. 11, no. 6, pp. 1466–1481, Dec 2015.
[80] J. Ren, Y. Zhang, N. Zhang, D. Zhang, and X. Shen, “Dynamic channel access to
improve energy eciency in cognitive radio sensor networks,” IEEE Transactions
on Wireless Communications, vol. 15, no. 5, pp. 3143–3156, May 2016.
[81] J. Wannstrom. (June) LTE-Advanced. 3GPP. [Online]. Available: hp://www.
3gpp.org/technologies/keywords-acronyms/97-lte-advanced
[82] (2015, January) 3GPP and unlicesed spectrum. 3GPP. [Online]. Available:
hp://www.3gpp.org/news-events/3gpp-news/1660-laa ieee
[83] 3GPP. (2015) Overview of 3GPP release 12. 3rd Generation Partnership Project.
[Online]. Available: hp://www.3gpp.org/specications/releases/68-release-12
[84] ——. (2015) Overview of 3GPP release 13. 3rd Generation Partnership Project.
[Online]. Available: hp://www.3gpp.org/release-13
[85] Broadband Radio Access Networks (BRAN); 5 GHz high performance RLAN;
Harmonized EN covering the essential requirements of article 3.2 of the R&TTE
Directive, European Telecommunications Standards Institute Std. ETSI EN 301
893 (Version 1.7.2) (Dra), 2014. [Online]. Available: hp://www.etsi.org/
[86] S. Geirhofer, L. Tong, and B. Sadler, “Cognitive medium access: Constraining
interference based on experimental models,” Selected Areas in Communications,
IEEE Journal on, vol. 26, no. 1, pp. 95 –105, Jan 2008.
[87] E. C. Y. Peh, Y. C. Liang, Y. L. Guan, and Y. Zeng, “Optimization of cooperative
sensing in cognitive radio networks: A sensing-throughput tradeo view,” IEEE
Transactions on Vehicular Technology, vol. 58, no. 9, pp. 5294–5299, Nov 2009.
[88] S. Maleki, A. Pandharipande, and G. Leus, “Two-stage spectrum sensing for
BIBLIOGRAPHY 167
cognitive radios,” in 2010 IEEE International Conference on Acoustics, Speech and
Signal Processing, March 2010, pp. 2946–2949.
[89] L. Luo, N. M. Neihart, S. Roy, and D. J. Allstot, “A two-stage sensing technique for
dynamic spectrum access,” IEEE Transactions on Wireless Communications, vol. 8,
no. 6, pp. 3028–3037, June 2009.
[90] J. So, “Cooperative spectrum sensing with two-stage reporting for cognitive radio
networks,” Electronics Leers, vol. 52, no. 1, pp. 83–85, 2016.
[91] M. Lpez-Bentez and F. Casadevall, “Time-dimension models of spectrum usage for
the analysis, design, and simulation of cognitive radio networks,” IEEE Transac-
tions on Vehicular Technology, vol. 62, no. 5, pp. 2091–2104, Jun 2013.
[92] M. Lopez-Benitez and F. Casadevall, “Empirical time-dimension model of spec-
trum use based on a discrete-time Markov chain with deterministic and stochastic
duty cycle models,” IEEE Transactions on Vehicular Technology, vol. 60, no. 6, pp.
2519–2533, July 2011.
[93] C. Ghosh, S. Pagadarai, D. P. Agrawal, and A. M. Wyglinski, “A framework for
statistical wireless spectrum occupancy modeling,” IEEE Transactions on Wireless
Communications, vol. 9, no. 1, pp. 38–44, January 2010.
[94] K. Chang and B. Senadji, “Spectrum sensing optimisation for dynamic primary
user signal,” IEEE Transactions on Communications, vol. 60, no. 12, pp. 3632–3640,
December 2012.
[95] Y. Zhao, M. Song, and C. Xin, “FMAC: A fairMAC protocol for coexisting cognitive
radio networks,” in 2013 Proceedings IEEE INFOCOM, April 2013, pp. 1474–1482.
[96] M. Sami, N. K. Noordin, and M. Khabazian, “A TDMA-based cooperative MAC
protocol for cognitive networks with opportunistic energy harvesting,” IEEE
Communications Leers, vol. 20, no. 4, pp. 808–811, April 2016.
168 BIBLIOGRAPHY
[97] J. K. Lee, H. J. Noh, and J. Lim, “TDMA-based cooperative MAC protocol for multi-
hop relaying networks,” IEEE Communications Leers, vol. 18, no. 3, pp. 435–438,
March 2014.
[98] “LTE in unlicensed spectrum: Harmonious coexistence with Wi-Fi,” alcomm
Research, alcomm Technologies, Inc., June 2014.
[99] S. S. Tan, J. Zeidler, and B. Rao, “Opportunistic channel-aware spectrum access
for cognitive radio networks with interleaved transmission and sensing,” IEEE
Transactions on Wireless Communications, vol. 12, no. 5, pp. 2376–2388, May 2013.
[100] A. Fanous and A. Ephremides, “Access schemes for mitigating the eects of
sensing errors in cognitive wireless networks,” IEEE Transactions on Wireless
Communications, vol. 13, no. 6, pp. 3343–3352, June 2014.
[101] A. Sabharwal, P. Schniter, D. Guo, D.W. Bliss, S. Rangarajan, and R.Wichman, “In-
band full-duplex wireless: Challenges and opportunities,” IEEE Journal on Selected
Areas in Communications, vol. 32, no. 9, pp. 1637–1652, Sept 2014.
[102] L. T. Tan and L. B. Le, “Channel assignment with access contention resolution
for cognitive radio networks,” IEEE Transactions on Vehicular Technology, vol. 61,
no. 6, pp. 2808–2823, July 2012.
[103] R. K.Mondal, B. Senadji, andD. Jayalath, “Dual-level sensing basedmultiple access
protocol for cognitive radio networks,” in 2017 IEEE 85th Vehicular Technology
Conference (VTC Spring), June 2017, pp. 1–5.
[104] J. W. Chong, D. K. Sung, and Y. Sung, “Cross-layer performance analysis for
CSMA/CA protocols: Impact of imperfect sensing,” IEEE Transactions on Vehicular
Technology, vol. 59, no. 3, pp. 1100–1108, March 2010.
[105] Y. Chen and H. S. Oh, “A survey of measurement-based spectrum occupancy
modeling for cognitive radios,” IEEE Communications Surveys Tutorials, vol. 18,
no. 1, pp. 848–859, Firstquarter 2016.
BIBLIOGRAPHY 169
[106] T. V. J. G. B. Jeremiah F. Hayes, Modeling and Analysis of Telecommunications
Networks. A John Wiley and Sons, Inc., 2004.
[107] J. A. Gubner, Probability and Random Process for Electrical and Computer Engineers.
Cambridge University Press, 2006.
[108] (Dra) Part 22: Cognitive Wireless RANMedium Access Control (MAC) and Physical
Layer (PHY) Specications: Policies and Procedures for Operation in the TV Bands,
IEEE Computer Society Std., March 2011.
[109] Y. Xu, Q. Wu, J. Wang, L. Shen, and A. Anpalagan, “Robust multiuser sequential
channel sensing and access in dynamic cognitive radio networks: Potential games
and stochastic learning,” IEEE Transactions on Vehicular Technology, vol. 64, no. 8,
pp. 3594–3607, Aug 2015.
[110] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge University Press,
2009.
[111] L. N. Polyakova, On minimizing the sum of a convex function and a concave
function. Berlin, Heidelberg: Springer Berlin Heidelberg, 1986, pp. 69–73.
[Online]. Available: hp://dx.doi.org/10.1007/BFb0121137
[112] R. T. Rockafellar, Convex Analysis. Princeton University Press, Princeton, New
Jersey, 1997.
[113] S. Pollin, M. Ergen, S. C. Ergen, B. Bougard, L. V. D. Perre, I. Moerman, A. Bahai,
P. Varaiya, and F. Cahoor, “Performance analysis of sloed carrier sense IEEE
802.15.4 medium access layer,” IEEE Transactions on Wireless Communications,
vol. 7, no. 9, pp. 3359–3371, September 2008.
[114] L. Gavrilovska, D. Denkovski, V. Rakovic, and M. Angjelichinoski, “Medium
access control protocols in cognitive radio networks: Overview and general
classication,” IEEE Communications Surveys Tutorials, vol. 16, no. 4, pp. 2092–
2124, Fourthquarter 2014.
170 BIBLIOGRAPHY
[115] E. Ziouva and T. Antonakopoulos, “CSMA/CA performance under high trac
conditions: throughput and delay analysis,” Computer Communications, vol. 25,
no. 3, pp. 313 – 321, 2002. [Online]. Available: hp://www.sciencedirect.com/
science/article/pii/S0140366401003693
[116] P. Chatzimisios, A. C. Boucouvalas, and V. Vitsas, “IEEE 802.11 packet delay-
a nite retry limit analysis,” in Global Telecommunications Conference, 2003.
GLOBECOM ’03. IEEE, vol. 2, Dec 2003, pp. 950–954 Vol.2.
[117] W. y. Lee and I. F. Akyildiz, “Optimal spectrum sensing framework for cognitive
radio networks,” IEEE Transactions on Wireless Communications, vol. 7, no. 10, pp.
3845–3857, October 2008.