SPECTRUM SENSING BASED ON CAPON POWER SPECTRAL DENSITY
ESTIMATION
by
Ola Ashour Mohammed
A thesis submitted in conformity with the requirementsfor the degree of Master of Applied Science
Faculty of Engineering and Applied Science (Electrical and ComputerEngineering)
University of Ontario Institute of Technology
Supervisor(s): Khalil El-Khatib, Miguel Vargas Martin
Copyright c© 2015 by Ola Ashour Mohammed
Abstract
Spectrum Sensing Based on Capon Power Spectral Density Estimation
Ola Ashour Mohammed
Master of Applied Science
Faculty of Engineering and Applied Science
University of Ontario Institute of Technology
2015
Cognitive radio (CR) technology has evolved to solve the spectrum scarcity problem and im-
prove spectrum utilization. Spectrum sensing is a CR function that allows secondary users to
efficiently utilize the spectrum without interfering with primary users. The performance of this
function depends on the efficiency of the used detection method. In this thesis, we propose a
spectrum sensing based on the Capon Power Spectral Density (PSD) estimation method. The
proposed method estimates the received PSD, and uses it to identify free and busy channels.
A cooperative spectrum sensing approach is also introduced. The goal is to solve the common
hidden node problem and help devices without CR capability to identify free channels. Exper-
imental results show that the proposed method outperforms the spectrum sensing based on the
Periodogram method in detecting both busy and free channels. In addition, simulation results
show that the cooperative approach improves the spectrum sensing function.
ii
Dedication
I would like to dedicate my work to my lovely parents for their endless love and whose words
of encouragement were lightening my road to success.
To my adorable son Yusuf, and my husband Mohamed who has been a constant source of
support during the challenges of graduate school and life. I thank Allah for having you in my
life and cannot imagine the long way we passed together.
To my pretty sister Rehab and brother Ahmed who believe on me and were always there to
support me.
iii
Acknowledgements
Thanks to Almighty ALLAH, the most gracious and the most merciful, who gave me the
strength and patience to complete this thesis.
I sincerely want to thank my supervisors Prof. Khalil El-Khatib and Prof. Miguel Vargas
Martin for their guidance and encouragement. They were always supporting me on the hardest
times.
I want to express my gratitude to Prof. Shahram Shahbazpanahi for his time, advice and
guidance to finish this thesis.
I would like to thank Prof. Michael Bennett for his guidance which helped me to obtain the
prestigious Danny Ho Scholarship.
I would also like to thank the Natural Sciences and Engineering Research Council of
Canada for funding part of my research work.
iv
Contents
1 Introduction 1
1.1 Thesis Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 White Spaces (Spectrum Holes) . . . . . . . . . . . . . . . . . . . . . 3
1.1.2 TV White Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.3 Countries Utilizing White Spaces . . . . . . . . . . . . . . . . . . . . 5
1.2 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Contribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2 Dynamic Spectrum Access and Cognitive Radio Technology 9
2.1 Dynamic Spectrum Access . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.1 Dynamic Exclusive Use Model . . . . . . . . . . . . . . . . . . . . . . 9
2.1.2 Open Sharing Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.3 Hierarchical Access Model . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Cognitive Radio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.1 Cognitive Radio Network Architecture . . . . . . . . . . . . . . . . . 11
2.2.2 Cognitive Radio Network Operation . . . . . . . . . . . . . . . . . . . 12
3 Cognitive Radio Management Functions 15
3.1 White Spaces Determination . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.1.1 Spectrum Sensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
v
Primary Transmitter Detection . . . . . . . . . . . . . . . . . . . . . . 16
Hidden Node Problem . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Cooperative Spectrum Sensing . . . . . . . . . . . . . . . . . . . . . . 18
Primary Receiver Detection . . . . . . . . . . . . . . . . . . . . . . . 20
Interference Temperature Metric . . . . . . . . . . . . . . . . . . . . . 20
3.1.2 Geo-Location Database . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Spectrum Decision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Spectrum Sharing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.3.1 Spectrum Sharing Architectures . . . . . . . . . . . . . . . . . . . . . 23
3.3.2 Spectrum Allocation Behaviour . . . . . . . . . . . . . . . . . . . . . 23
3.3.3 Spectrum Access Techniques . . . . . . . . . . . . . . . . . . . . . . . 24
3.3.4 Spectrum Sharing Scope . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3.5 Spectrum Sharing Models . . . . . . . . . . . . . . . . . . . . . . . . 24
3.4 Spectrum Mobility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4 Spectrum Sensing Based on Capon PSD Estimation 27
4.1 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.2 Choosing Threshold Value . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.3 Energy Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.3.1 Energy Detection in Time Domain . . . . . . . . . . . . . . . . . . . . 31
4.3.2 Energy Detection in Frequency Domain . . . . . . . . . . . . . . . . . 32
4.4 Power Spectral Density Estimation . . . . . . . . . . . . . . . . . . . . . . . . 33
4.4.1 Power Spectral Density . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.4.2 Periodogram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.4.3 Capon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.5 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
vi
5 Simulation and Results 40
5.1 Determining the Operating Frequency of Primary Transmitters . . . . . . . . . 41
5.2 Measuring the Detection Performance . . . . . . . . . . . . . . . . . . . . . . 44
5.2.1 Detecting Primary Transmitters . . . . . . . . . . . . . . . . . . . . . 46
5.2.2 Detecting White Spaces . . . . . . . . . . . . . . . . . . . . . . . . . 48
5.3 Cooperative Spectrum Sensing Based on Capon Method . . . . . . . . . . . . 50
5.3.1 Changing Threshold Value . . . . . . . . . . . . . . . . . . . . . . . . 52
5.3.2 Changing Number of Sensors . . . . . . . . . . . . . . . . . . . . . . 53
5.3.3 Changing Search Distance . . . . . . . . . . . . . . . . . . . . . . . . 55
6 Conclusion and Future Work 58
References 59
vii
List of Tables
5.1 Common parameters for the three scenarios. . . . . . . . . . . . . . . . . . . . 40
5.2 Performance detection parameters. . . . . . . . . . . . . . . . . . . . . . . . . 45
viii
List of Figures
Figure 1.1 The electromagnetic spectrum [54] . . . . . . . . . . . . . . . . . . . . 2
Figure 1.2 Spectrum utilization [7] . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Figure 1.3 Concept of spectrum hole [3] . . . . . . . . . . . . . . . . . . . . . . . 4
Figure 1.4 Canadian TV transmitters coverage . . . . . . . . . . . . . . . . . . . . 5
Figure 2.1 Dynamic spectrum access classification [62] . . . . . . . . . . . . . . . 10
Figure 2.2 Cognitive radio concept [32] . . . . . . . . . . . . . . . . . . . . . . . 12
Figure 2.3 Infrastructure-based CRN [10] . . . . . . . . . . . . . . . . . . . . . . 13
Figure 2.4 Ad hoc CRN [10] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Figure 2.5 CRNs operation inside the coverage area of primary networks [55] . . . 14
Figure 3.1 Cognitive radio functions [42] . . . . . . . . . . . . . . . . . . . . . . 15
Figure 3.2 Methods used for white spaces determination [42] . . . . . . . . . . . . 17
Figure 3.3 Hidden node problem [3] . . . . . . . . . . . . . . . . . . . . . . . . . 19
Figure 3.4 Cooperative sensing methods (a) Centeralized (b) Distributed [1] . . . . 20
Figure 3.5 Spectrum sharing categorization [42] . . . . . . . . . . . . . . . . . . . 23
Figure 4.1 Types of error in hypothesis testing . . . . . . . . . . . . . . . . . . . . 29
Figure 4.2 Time domain representation of energy detection . . . . . . . . . . . . . 31
Figure 4.3 Frequency domain representation of energy detection . . . . . . . . . . 32
Figure 4.4 Spectrum sensing steps . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Figure 5.1 SUs distribution with respect to primary transmitters . . . . . . . . . . . 41
ix
Figure 5.2 PSD at the location of SU1 for high SNR . . . . . . . . . . . . . . . . . 42
Figure 5.3 PSD at the location of SU1 for low SNR . . . . . . . . . . . . . . . . . 43
Figure 5.4 PSD at the location of SU2 for high SNR . . . . . . . . . . . . . . . . . 43
Figure 5.5 PSD at the location of SU2 for low SNR . . . . . . . . . . . . . . . . . 44
Figure 5.6 Probability of detecting primary transmitters versus threshold . . . . . . 46
Figure 5.7 Probability of false alarm versus threshold . . . . . . . . . . . . . . . . 47
Figure 5.8 Average ROC for both Capon and Periodogram methods . . . . . . . . 48
Figure 5.9 Probability of detecting white spaces versus threshold . . . . . . . . . . 49
Figure 5.10 Probability of miss detection versus threshold . . . . . . . . . . . . . . 49
Figure 5.11 Average probability of detecting white spaces versus probability of miss
detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Figure 5.12 Distribution of sensors and nodes . . . . . . . . . . . . . . . . . . . . . 51
Figure 5.13 ROC of cooperative spectrum sensing . . . . . . . . . . . . . . . . . . 52
Figure 5.14 PW versus PMiss for cooperative spectrum sensing with different thresh-
old values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Figure 5.15 PD versus number of sensors for cooperative spectrum sensing . . . . . 54
Figure 5.16 Average PD versus PFA for cooperative spectrum sensing with different
number of sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
Figure 5.17 Average PW versus PMiss for cooperative spectrum sensing with differ-
ent number of sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
Figure 5.18 PD versus different search distances in cooperative spectrum sensing . . 56
Figure 5.19 Average PD versus PFA with different search distances for cooperative
spectrum sensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
Figure 5.20 Average PW versus PMiss with different search distances for coopera-
tive spectrum sensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
x
xi
List of Acronyms
CR Cognitive Radio
SUs Secondary Users
PUs Primary Users
PSD Power Spectrum Density
CM Capon Method
FCC Federal Communications Commission
USA United States of America
Ofcom Office of Communications
UK United Kingdom
ITU International Telecommunications Union
FCA Fixed Channel Allocation
ISM Industrial, Scientific, and Medical
DSA Dynamic Spectrum Access
TV Television
VHF Very High Frequency
UHF Ultra High Frequency
TVWS TV White Spaces
CRD Cognitive Radio Device
UWB Ultra Wide Band
SDR Software Defined Radio
WSDs White Space Devices
xii
WiFi Wireless Fidelity
CRNs Cognitive Radio Networks
BSs Base Stations
APs Access Points
QoS Quality of Service
SNR Signal to Noise Ratio
FC Fusion Center
RF Radio Frequency
LO Local Oscillator
L-R Longley-Rice
WSDB White Space Database
WSDs White Space Devices
SAPs Secondary Access Points
H0 Null hypothesis
H1 Alternative hypothesis
PDF probability density function
PFA Probability of False Alarm
PMiss Probability of Miss Detection
NP Neyman Pearson
PD Probability of Detection
γ Threshold
ROC Receiver Operating Characteristic curve
AWGN Additive white Gaussian noise
BPF Band-Pass Filter
ADC Analog to Digital Converter
FFT Fast Fourier Transform
NBS Nash Bargaining Solution
xiii
MA Moving Average
ARMA Auto Regressive Moving Average
MUSIC Multiple Signal Classification
FIR Finite Impulse Response
CCC Common Control Channel
xiv
Chapter 1
Introduction
Wireless communication is an important technology that changed the way we live. Many
things we use every day depends on wireless communication such as our cellular phones, wire-
less keyboards, and remote car starters. Crisis and emergency situations can be handled very
quickly with the help of wireless communication. The fast growth in wireless communication
services and in the number of wireless users, lead to an increasing demand for more spectrum
bands. However, the radio spectrum that can be used for wireless communication is limited by
nature, and most of its spectrum bands have already been allocated using long term licenses.
So, it is hard to find vacant bands.
A chart of the electromagnetic spectrum which contains all types of electromagnetic ra-
diation is shown in Figure 1.1 [54]. Radio spectrum is a part of electromagnetic spectrum
corresponding to radio frequencies lower than 300 GHz. Electromagnetic waves in this fre-
quency range are called radio waves and can be used for wireless communication and other
different applications.
The usage of the radio spectrum is typically regulated by government agencies around
the world, such as the Federal Communications Commission (FCC) in the United States of
America (USA), Industry Canada in Canada, and Office of Communications (Ofcom) in the
United Kingdom (UK). These agencies are supervised by an international standards body called
1
CHAPTER 1. INTRODUCTION 2
Figure 1.1: The electromagnetic spectrum [54]
the International Telecommunications Union (ITU). They divide the radio spectrum range into
non-overlapping frequency bands. The assignment of these frequency bands is made based on
a Fixed Channel Allocation (FCA) policy, in which the usage of each band is exclusively given
to a single provider or to one wireless application. Some bands like the Industrial, Scientific,
and Medical (ISM) bands are unlicensed. They are allocated free of charge and can be used
by several applications like cordless phones, microwave ovens, medical diathermy machines,
and wireless computer networks. There are also licensed bands that can be allocated to various
operators such as cellular telephone operators or broadcast television stations using long term
licenses.
CHAPTER 1. INTRODUCTION 3
Figure 1.2: Spectrum utilization [7]
1.1 Thesis Motivation
1.1.1 White Spaces (Spectrum Holes)
The need for spectrum bands that can facilitate the implementation of emerging wireless ser-
vices, increase on a daily basis. However, most of the suitable spectrum bands are either sold
or leased using long term licenses. Recent studies show that the allocated spectrum is not effi-
ciently used [21, 5, 27, 38]. Figure 1.2 [7] shows the distribution of the signal strength versus
frequency for a part of the radio spectrum measured at downtown Berkeley. The distribution
indicates that some frequency bands are intensely used, while others are barely used.
Moreover, measurements done by FCC [21] in Atlanta, New Orleans, and San Diego, show
that the FCA policy led to huge temporal and spatial disparities ranging from 15% to 85% in
the allocated spectrum usage, which make the spectrum inefficiently utilized. Recent measure-
ments done in different places like Vietnam [5], Singapore [27], and HULL-UK [38] also show
similar results. Figure 1.3 [3] shows the spectrum parts that are unused at a specific time or
CHAPTER 1. INTRODUCTION 4
Figure 1.3: Concept of spectrum hole [3]
location, which are called white spaces or spectrum holes. The blocks represent frequencies
in use, while the empty regions represent temporally unused frequencies. Dynamic Spectrum
Access (DSA) techniques that will be covered in Chapter 2, allow the unlicensed use of these
spectrum holes.
1.1.2 TV White Spaces
After switching the TV transmission from analog to digital, big parts of frequencies in the Very
High Frequency (VHF) and Ultra High Frequency (UHF) bands have been freed up. These un-
used frequency portions on the TV broadcasting are referred to as TV White Spaces (TVWS).
Frequencies in the TV broadcast bands benefit from high bandwidth, long transmission ranges
and better building penetration. Figure 1.4 shows an example for the TV transmitters coverage
in Canada. Each circle represents the coverage area of one transmitter located at the center of
the circle. After switching to digital TV a lot of these transmitters are no longer used, resulting
in a large number of TVWS.
CHAPTER 1. INTRODUCTION 5
Figure 1.4: Canadian TV transmitters coverage
1.1.3 Countries Utilizing White Spaces
Although the USA was not the first country to switch to digital TV, it became a leader in making
the decision to utilize white spaces. The FCC allowed the usage of TVWS by unlicensed radio
transmitters [16]. These unlicensed transmitters have to avoid interfering with licensed users
[19]. The FCC rules state that the unlicensed devices must both consult a database and perform
spectrum sensing to determine white spaces in their locations. The database is queried by pro-
viding a certain location, then it returns a list of the available channels at that location. While,
the spectrum sensing is used to ensure that licensed devices such as wireless microphones, that
may not be registered in the database, are also protected.
The FCC issued a Memorandum Opinion and Order on September 23, 2010 [17], which
states the final regulations that the unlicensed devices should follow for utilizing white spaces.
The mandatory requirement for using spectrum sensing was eliminated. The reason for remov-
ing spectrum sensing is that the FCC assigned two channels for the exclusive use of wireless
microphones that are not guaranteed to be registered on the database. Meanwhile the FCC
encourage the development in spectrum sensing as it believes that it can enhance the spectrum
efficiency.
The UK was the second country to utilize white spaces , when Ofcom published a docu-
CHAPTER 1. INTRODUCTION 6
ment in September 2011 expressing the intention to support commercial utilization of white
spaces [44]. In that document, Ofcom mentioned that their approach is based on geo-location
database rather than alternative approaches like sensing or beacons. Ofcom indicated that the
UK supports the development of a harmonized European approach for White Space Devices
(WSDs), as it believes that the consumers can obtain the largest benefits from that approach.
In August 2011, Industry Canada started a consultation to address the challenges and bene-
fits of using TVWS [9]. Depending on the results of this consultation process Industry Canada
made the decision to allow the usage of TVWS devices in Canada in April 2013 [8].
1.2 Problem Statement
A spectrum scarcity problem appeared as a result of the gap between the available and required
radio spectrum bands. As it is not possible to find new spectrum bands, the only way to
overcome the spectrum scarcity problem is to find better ways to use the available spectrum.
A closer look to the allocated spectrum showed that, the FCA scheme causes the spectrum
to be inefficiently utilized over time and space. Hence, it is not suitable to accommodate the
increasing requirements for wireless communication services. Innovative techniques that can
make efficient use of the allocated spectrum need to be developed to facilitate the deployment
of emerging wireless application.
Dynamic spectrum access techniques have arisen to overcome the drawbacks of the FCA
scheme, by making efficient use of the allocated spectrum. DSA techniques enable unlicensed
Secondary Users (SUs) to make use of the licensed spectrum in an opportunist manner, pro-
vided that they do not interfere with the licensed Primary Users (PUs). Since the SUs are
allowed to use the spectrum when PUs are not using it, so they need a special kind of devices
that enable them to identify the used and unused spectrum bands in their location.
Cognitive radio technology is the enabling technology for DSA, which can help SUs to
identify spectrum holes in their location. Thus for a SU to be able to utilize the spectrum holes
CHAPTER 1. INTRODUCTION 7
in the allocated spectrum, it has to be equipped with a Cognitive Radio Device (CRD), which
is a device that has CR capability. The key function for any CRD is called spectrum sensing,
which allow SUs to determine the presence or absence of PUs signals.
The spectrum sensing function is a very critical and challenging function and can greatly
affect the spectrum utilization. When performing spectrum sensing we have to compromise
between protecting PUs from interference and achieving efficient utilization of the spectrum.
The performance of the spectrum sensing function is measured by the performance of the
detection technique used to detect the spectrum status. A miss detection for a PU signal results
in a SU transmission on the same channel which will cause interference to the PU. Falsely
detecting the presence of a PU signal in an empty channel would result in underutilization
of the spectrum. Thus, it is extremely important to select a reliable detection technique that
can maximize the spectrum utilization and at the same time provide interference protection for
PUs.
1.3 Contribution
The contribution of this thesis are twofold:
1. First, we propose a spectrum sensing approach based on Capon PSD estimation method.
The proposed method efficiently estimates the spectrum from a finite length of received
data samples. The estimated spectral is then used for detecting spectrum holes. The
effect of changing the signal power threshold value on the detection performance of
the proposed method is investigated. A performance comparison is made between the
proposed spectrum sensing based on the CM and that based on the conventional Peri-
odogram method.
2. Second, we propose a cooperative spectrum sensing approach that can be used to over-
come the hidden node problem. It is also used for helping devices with no CR capability
to accurately determine white spaces in their location. Those devices depend on the PSD
CHAPTER 1. INTRODUCTION 8
estimation of neighboring sensors to detect spectrum holes in their location. The effect
of different key parameters such as threshold value, number of sensors, search distance
on the performance of the proposed approach was examined.
1.4 Thesis Organization
The remainder of this thesis is organized as follows:
Chapter 2, discusses the concept of DSA, which is used to overcome the spectrum scarcity
problem, with a detailed classification of its access models. An overview of the CR technology,
its definition, and the characteristics of the CRD is also provided. Followed by a description of
the CR network architecture, its operation, and priority of users operating on that network.
Chapter 3, covers the four CR management functions with a detailed description of each
function. Followed by the challenges that we may face and the processes or the methods that
can be used to implement these function. The contents of Chapter 3 were published in the
fourth international conference on advances in cognitive radio, COCORA 2014.
Chapter 4, formulates the spectrum sensing problem as a binary hypothesis test problem
and discusses the errors that can happen when we are deciding which hypothesis has occurred.
The performance metrics for any detection technique is also discussed. Then a description
for the conventional energy detection technique and its implementation in time and frequency
domain was covered. After that we give an overview about the PSD estimation techniques
which is the main part of our thesis. A special focus is given to the Periodogram and Capon
PSD estimation methods, which will be used on the simulation.
Chapter 5, contains the implementation of a spectrum sensing function based on Capon
and Periodogram methods using Matlab simulations. Different scenarios are discussed and
illustrated by simulation results. After that a cooperative spectrum sensing scenario based on
the Capon PSD estimation method is also implemented.
Chapter 6, concludes the thesis and provides recommendations for future work.
Chapter 2
Dynamic Spectrum Access and Cognitive
Radio Technology
2.1 Dynamic Spectrum Access
The limited spectrum availability and inefficient spectrum utilization, in addition to the in-
creasing demand for radio spectrum bands causes a spectrum scarcity problem [23, 3]. DSA
techniques have appeared to overcome this problem, by making efficient use of the allocated
spectrum. DSA techniques allow the spectrum to be shared and reused among different wire-
less applications on a negotiable or opportunistic basis. DSA techniques can be classified into
three categories which are shown in Figure 2.1 [62].
2.1.1 Dynamic Exclusive Use Model
The dynamic exclusive use model allows the licensees to exclusively use certain spectrum
bands which is similar to the concept of FCA. However more flexibility is given for the li-
censees, which can use one of the two approaches called spectrum property rights and dynamic
spectrum allocation to efficiently utilize the spectrum. In spectrum property rights, the license
holders can use the spectrum for different kind of services. They can even lease or trade their
9
CHAPTER 2. DYNAMIC SPECTRUM ACCESS AND COGNITIVE RADIO TECHNOLOGY 10
Figure 2.1: Dynamic spectrum access classification [62]
assigned spectrum with various operators. In dynamic spectrum allocation, the spectrum can
be shared among different services based on the observed traffic statistics variation of these
services in time and location.
2.1.2 Open Sharing Model
In this model, the spectrum is shared among users who have equal priorities in accessing the
spectrum. The model is adopted by wireless systems sharing the ISM band like WiFi networks,
microwave ovens and Blue-tooth devices.
2.1.3 Hierarchical Access Model
Unlike the open sharing model, the users in the hierarchical access model have different pri-
orities in accessing the spectrum. There are two types of users PUs and SUs. The PUs or
the licenses owners, have the absolute privilege to use the spectrum. On the other hand, the
SUs who do not have licenses, can only use the spectrum in a non-intrusive manner. Two ap-
proaches can be used by SUs to access the spectrum. The spectrum underlay approach, which
permits PUs and SUs to transmit concurrently with certain restrictions on the amount of power
transmitted by SUs. The SUs transmitted power must be kept below the noise floor allowed
by PUs. This can be achieved by using ultra wide band (UWB) technology, which maintains
CHAPTER 2. DYNAMIC SPECTRUM ACCESS AND COGNITIVE RADIO TECHNOLOGY 11
low transmission power by spreading the transmitted signal over a wide frequency band. This
approach is suitable for short range, high bandwidth communications. The spectrum overlay
approach does not have restrictions on the amount of transmitted power. However, there is a
limitation on when and where you are allowed to use the spectrum. SUs can utilize the spec-
trum in an opportunistic manner, so they have to determine the presence or absence of PUs and
try to exploit the white spaces for their transmission. As the white spaces differ based on the
location, frequency and sensing time, SUs need devices that have the ability to detect whether
the spectrum is being utilized at a specific location, frequency and at a certain time [3].
2.2 Cognitive Radio
Cognitive radio, first proposed by mitola [41], is the enabling technology for DSA. It is a
wireless communication technology based on Software Defined Radio (SDR), which helps
SUs identify spectrum holes in their locations.
A device that uses the CR technology to determine white spaces on its location is called a
CRD. This device has two main characteristics [3, 23] illustrated in Figure 2.2 [32]. The cog-
nitive capability allows the device to sense (observe) the medium, determine (learn) available
spectrum bands and then make a decision on which channel to use. Re-configurability enables
the CRD to adjust its software (communication protocol) and hardware (radio frequency front
end and antenna) to operate on the selected channel.
2.2.1 Cognitive Radio Network Architecture
Cognitive Radio Networks (CRNs) are unlicensed networks formed only from SUs that exploit
spectrum holes for their transmission. These networks can be classified based on their archi-
tecture [10, 37] as infrastructure-based and ad hoc networks. Figure 2.3 [10] shows an example
of infrastructure-based networks. These kind of networks have secondary Base Stations (BSs)
or secondary Access Points (APs) that can coordinate the communication between SUs in their
CHAPTER 2. DYNAMIC SPECTRUM ACCESS AND COGNITIVE RADIO TECHNOLOGY 12
Figure 2.2: Cognitive radio concept [32]
coverage areas. Secondary APs can be connected together through a wired network (core net-
work) to facilitate the communication between SUs in different coverage areas. Figure 2.4 [10]
displays an example for the ad-hoc or distributed CRNs. In this type of networks there is no
infrastructure, SUs communicate directly with each other using available white spaces in their
locations.
2.2.2 Cognitive Radio Network Operation
CRNs either infrastructure-based or ad hoc networks usually work inside the transmission
range of primary networks as shown in Figure 2.5 [55]. Primary networks or licensed net-
works refer to already existing infrastructure-based wireless networks like mobile networks
CHAPTER 2. DYNAMIC SPECTRUM ACCESS AND COGNITIVE RADIO TECHNOLOGY 13
Figure 2.3: Infrastructure-based CRN [10]
Figure 2.4: Ad hoc CRN [10]
CHAPTER 2. DYNAMIC SPECTRUM ACCESS AND COGNITIVE RADIO TECHNOLOGY 14
Figure 2.5: CRNs operation inside the coverage area of primary networks [55]
that are allocated certain frequency bands for their operation. When the CRN operates in the
licensed band, the priority is for PUs operations. SUs that utilize spectrum holes, have to vacate
the channel whenever a PU appears.
Chapter 3
Cognitive Radio Management Functions
The CRD has to perform four CR management functions to manage spectrum holes in its loca-
tion, avoid interference with PUs, coexist with other devices, and efficiently use the spectrum
[3, 55]. These functions are shown in Figure 3.1 [42], which are white space determination,
spectrum decision, spectrum sharing, and spectrum mobility.
Figure 3.1: Cognitive radio functions [42]
The main function for any CRD is to determine white spaces at a certain place and specific
time. After that, the CRD has to select the best channel for its operation, which is called spec-
15
CHAPTER 3. COGNITIVE RADIO MANAGEMENT FUNCTIONS 16
trum decision. The channel selection is usually done based on specific criteria such as avoiding
interference with other CRDs or requiring a certain Quality of Service (QoS). As the spectrum
is shared among multiple SUs, spectrum sharing is required to coordinate how SUs can coexist
and access the same spectrum without interfering or colliding with each other. Spectrum mo-
bility plays an important role in preserving the PUs priority to access the spectrum. A SU has
to stop its transmission and switch to another available channel if a PU appears on the same
channel.
3.1 White Spaces Determination
The key function for any CRD is to determine spectrum holes in its location which vary in time
and space. Figure 3.2 [42] displays the two approaches used for identifying spectrum holes.
3.1.1 Spectrum Sensing
The first approach is called spectrum sensing, which allows the CRD to sense the spectrum and
determine its availability for use. In spectrum sensing, the CRD depends on its own sensing
measurement, to find the spectrum holes in its current location. To overcome the time varying
spectrum holes the device should periodically sense the medium and adapt to any variations in
the sensed spectrum usage. Figure 3.2 displays the different detection techniques that can be
used to perform the spectrum sensing function.
Primary Transmitter Detection
In primary transmitter detection, the medium is considered available if the CRD cannot hear the
signal sent from any primary transmitter. Several methods can be used for primary transmitter
detection such as matched filter detection [36, 51, 49], feature detection [22, 40, 47], and energy
detection [60, 4, 45, 58].
CHAPTER 3. COGNITIVE RADIO MANAGEMENT FUNCTIONS 17
Figure 3.2: Methods used for white spaces determination [42]
• Matched filter detection is the optimal detection technique in the presence of stationary
Gaussian noise. The matched filter operates by correlating the pattern that needs to be
detected (known information about the signal) with the received signal. If the magnitude
of the resulting signal is above a certain threshold; the medium is considered busy other-
wise, the medium is free. The matched filter is a fast detection technique, but it requires
previous knowledge about the signal to be detected.
• Feature detection or cyclostationary feature detection depends on detecting the cyclosta-
tionary (built-in periodicity) feature of the modulated signal for detecting the presence of
a signal. The cyclostationary feature detector can distinguish the PU signal from noise.
CHAPTER 3. COGNITIVE RADIO MANAGEMENT FUNCTIONS 18
This kind of detector is more robust against the uncertainty in the noise power, but re-
quires more observation time and is computationally complex.
• Energy detection or radiometer does not require prior information about the primary
transmitter signal. In this detection technique, the CRD measures the energy of the
received signal and compares it with a threshold. The medium is considered busy if the
measured energy is greater than the threshold , otherwise the medium is considered free
and can be used by SUs. The energy detector is the most popular detection technique
due to its simple implementation and because it does not require prior information about
the PU signal like the matched filter or the feature detection techniques. However, one of
the concerns of using energy detection is that it cannot differentiate whether the detected
signal is from a primary transmitter or from a secondary transmitter. Another concern is
adjusting the threshold value of the detector as this value is affected by the noise level.
Energy detector will be discussed in details in Chapter 4.
Hidden Node Problem
The main problem of the primary transmitter detection techniques is the hidden node problem
[46]. In this problem, the CRD may be shadowed from detecting the signal of a primary
transmitter, due to the presence of an object that block the transmitter signal, as shown in
Figure 3.3 [3]. As a result of the hidden node problem, the CRD will fail to identify the
presence of a the PU in the channel and hence, will interfere with that user. To overcome the
hidden node problem cooperative spectrum sensing or cooperative detection can be used.
Cooperative Spectrum Sensing
When performing spectrum sensing, SUs may suffer from unexpected multipath fading and/or
shadowing. These multipath fading and shadowing issues can significantly affect the detection
performance. Cooperative spectrum sensing is an effective way to mitigate the effect of these
issues, improve the detection performance and can be also used as a solution for the hidden
CHAPTER 3. COGNITIVE RADIO MANAGEMENT FUNCTIONS 19
Figure 3.3: Hidden node problem [3]
node problem. The idea behind cooperative spectrum sensing is to exploit the observations of
CRDs in different locations to obtain better information about the spectrum. Each CRD will
share the sensed information on its location with other CRDs to make a combined decision
which will be more accurate than individual decisions. Cooperative spectrum sensing can be
done in a centralized or in a distributed manner as shown in Figure 3.4 [1]
• In centralized method, CRDs send the sensed information to a central CRD called fusion
center (FC). The FC is responsible for analyzing the collected information and determin-
ing white spaces that can be used by SUs. This method requires a point to point link
between each CRD and the FC.
• In distributed method, each CRD sends the sensed information to other CRDs. Each
device is responsible for combining its own measurement with the information received
from other devices to determine spectrum holes in its location.
Cooperative spectrum sensing, either centralized or distributed, can enhance the detection ca-
pability, but comes with the cost of additional overhead.
CHAPTER 3. COGNITIVE RADIO MANAGEMENT FUNCTIONS 20
Figure 3.4: Cooperative sensing methods (a) Centeralized (b) Distributed [1]
Primary Receiver Detection
As the interference always happens at the receiver, so the most efficient technique in deter-
mining spectrum holes is to detect the presence of the primary receivers. The CRD needs to
detect primary receivers in its communication range and avoid interfering with them. Primary
receiver detection is not an easy process. Usually primary receivers such as televisions or cellu-
lar phones are passive, which makes it harder for the CRD to detect their presence or determine
their location. One way to allow the CRD identify the existence of a primary receiver [59], is
by detecting the leakage power emitted by the RF front end of the the primary receiver’s Local
Oscillator (LO). This method is currently feasible only in TV receiver detection and still has
some problems like the long sensing time and short detection range [25].
Interference Temperature Metric
Interference temperature is a model introduced by the FCC [15] to accurately measure and limit
the amount of interference at the receiver. Interference temperature dictates the cumulative
amount of interference from all the undesired RF energy sources that exist at a receiver at any
point of time. Interference temperature provides a higher protection for the receiver against
CHAPTER 3. COGNITIVE RADIO MANAGEMENT FUNCTIONS 21
harmful interference. A CR user can use the channel at any place and time as long as it does
not surpass the limit on interference temperature which is the concept of underlay spectrum
sharing that we will discuss in Section 4.3.3 .
3.1.2 Geo-Location Database
In the Geo-location database approach, the CRD does not use spectrum sensing to determine
white spaces; instead it depends on an up-to-data database of incumbents. The database stores
information about all primary transmitters and their locations. It also stores terrain information.
The database uses these information to calculate white spaces at the CRD location.
According to FCC regulation [19], the CRD should have the capability to determine its lo-
cation, and a way to connect to the Internet to access the database. The process of determining
white spaces is done through several steps. First, the CRD has to provide identifying infor-
mation to register with the database. Then, the CRD determines its location and send it to the
database, which uses some propagation models to calculate white spaces at that location. After
that, the database will send a list of the available spectrum holes to the user. The database may
also inform the device with the maximum allowable transmit power for its operation. In this
case, the device can use its adaptable power control to ensure that the transmitting power does
not exceed the maximum allowed value. Using a database to calculate spectrum holes over-
comes the problem of false alarm that can happen with spectrum sensing and thus provides
more efficient use of the spectrum.
Microsoft presented an approach for a geo-location database called SenseLess: A Database
Driven White Spaces Network [43]. The SenseLess architecture consisted of a logically cen-
tralized entity called SenseLess Service. Base stations and CRDs are connected to this central
entity, which is responsible for determining white spaces for any given location. Two compo-
nents mainly constitute the SenseLess service, the back-end store and the SenseLess engine.
The back-end store consisted of a terrain server and a database of incumbents, such as TV trans-
mitters (their location, antenna height, transmission power) and wireless microphones. The
CHAPTER 3. COGNITIVE RADIO MANAGEMENT FUNCTIONS 22
database can also cache the computed white spaces for different locations. The terrain server
store high resolution terrain elevation data, which can be obtained from one of the publicly
available terrain maps. Sophisticated signal propagation modeling is used by the SenseLess
engine to compute spectrum holes at any given location. Results showed that the Longley-Rice
(L-R) with terrain propagation model [14] provides accurate results when determining white
spaces at any given location.
3.2 Spectrum Decision
Spectrum decision is the capability of a CRD to choose the most appropriate channel for its
operation. Channel selection should satisfy the QoS requirements of SUs and at the same time
ensures that they do not interference with PUs. The spectrum decision function consists of
three processes called spectrum characterization, spectrum selection and CR re-configurability
[37, 2].
• Spectrum characterization is the first step to be performed after determining spectrum
holes. These spectrum holes are characterized based on the PUs activities and the condi-
tions of the radio environment. Modeling PUs activities based on the historical informa-
tion of previous spectrum usage, can help in predicting the future usage of the spectrum.
The condition of the radio environment is another factor that can be used to character-
ize the channel based on the number of users utilizing the same channel, the amount of
interference, and the strength of the received signal.
• Spectrum selection is used to select the best characterized channel that satisfies the QoS
requirements of the SU.
• CR re-configurability adjusts the transceiver parameters of the CRD to communicate on
the selected channel.
CHAPTER 3. COGNITIVE RADIO MANAGEMENT FUNCTIONS 23
3.3 Spectrum Sharing
The spectrum sharing function is used to coordinate the coexistence of CRDs in a shared
medium without interfering with each other. Figure 3.5 [42] displays a classification for spec-
trum sharing according to architecture, scope, spectrum allocation behaviour, spectrum sharing
models and spectrum access techniques [3] .
Figure 3.5: Spectrum sharing categorization [42]
3.3.1 Spectrum Sharing Architectures
The spectrum sharing architecture can be centralized or distributed. In centralized spectrum
sharing, a central unit is responsible for allocating the spectrum and controlling access to it. In
distributed spectrum sharing, spectrum allocation and access is done by each SU independently.
3.3.2 Spectrum Allocation Behaviour
The spectrum allocation behaviour can be cooperative or non-cooperative spectrum sharing. In
the cooperative spectrum sharing, the CRDs cooperate together to avoid interference with each
other. Each CRD adjusts its transmission power taking into account other devices transmission.
In the non-cooperative spectrum sharing, each CRD behaves in a selfish manner. The CRD will
transmit without considering if its transmission will affect other devices transmission. Thus,
in the non-cooperative case there will be high interference between CRDs, which in turn will
reduce the spectrum utilization.
CHAPTER 3. COGNITIVE RADIO MANAGEMENT FUNCTIONS 24
3.3.3 Spectrum Access Techniques
The spectrum access techniques can be classified as overlay spectrum sharing and underlay
spectrum sharing. In the overlay spectrum sharing, SUs can opportunistically use the spectrum
when it is not used by PUs. In the underlay spectrum sharing, SUs can transmit at the same
time with PUs as long as their transmission is below the noise floor of PUs. In this case, SUs
use spread spectrum techniques and can only transmit over short range.
3.3.4 Spectrum Sharing Scope
The spectrum sharing scope in infrastructure-based CRNs can be classified to intra-cell spec-
trum sharing and inter-cell spectrum sharing [33]. The intra-cell spectrum sharing is concerned
with spectrum sharing between SUs in the same cell. The inter-cell spectrum sharing is con-
cerned with spectrum sharing between different cells.
3.3.5 Spectrum Sharing Models
There are two spectrum sharing models called, exclusive allocation model and common use
model [33]. In the exclusive allocation model, each SU is allocated a different channel to
mitigate the interference between SUs. This model is optimum in maximizing the capacity of
the network, but it provides unfair resource allocation in networks that have limited spectrum
availability. Although the focus of this approach is on spectrum allocation, power allocation
has to be considered to avoid interfering with PUs. In the common use model, several SUs
can simultaneously use the same channel by adjusting their transmission power to minimize
interference. This model is preferred in networks with limited spectrum holes; as it can provide
fairness in the allocation of the available spectrum. However, the achieved capacity is lower
than that of the exclusive model.
Several research papers have been proposed to address the spectrum sharing problem in
CRNs [28, 26]. These papers usually have two main differences which are the objective of the
CHAPTER 3. COGNITIVE RADIO MANAGEMENT FUNCTIONS 25
network and the used spectrum sharing model.
In [28], the authors presented a combined power/channel allocation method in a WiFi-like
spectrum sharing scenario in TVWS. Three types of SUs were defined which are protected,
interfered and out of range SUs. The network objective is to increase the number of supported
SUs and at the same time minimize the interference between the different WiFi CRNs. The
Nash Bargaining Solution (NBS) was used to allocate transmit power for each Secondary Ac-
cess Point (SAP). SAPs cooperate by exchanging information through relaying nodes. These
relaying nodes are the interfered SUs that can hear from more than one SAP. SAPs compete on
their transmission powers to maximize the number of supported SUs. SAPs have to decrease
their power until there is no overlapping between their coverage areas. Two stage of cooper-
ation were made. In the first stage, only neighboring SAPs cooperate and compete for power
control, while in the second stage, all the next hop neighboring SAPs can cooperate. When the
SAPs are highly overlapped, the algorithm can switch to channel allocation instead of power
allocation to improve the performance of the network. The switch can be done if the number
of interfered SUs is above a certain number. Simulation results showed that the number of
iteration required to reach optimality is decreased by the cooperation of SAPs. It also showed
that the proposed sharing algorithm succeded in increasing the number of supported SUs, but
it comes with the cost of decreasing the average throughput per SAP.
A downlink channel assignment and power control for an infrastructure-based CRN was
implemented in [26]. The opportunistic spectrum access problem was formulated as a non-
cooperative game in which the game players are the base stations. Each base station bargain
to increase the number of supported SUs. Channel allocation was done by the base stations,
which randomly assign channels to users. A distributed power allocation is then applied us-
ing the Iterative Water Filling algorithm. Results showed that the pure non cooperative game
might have multiple Nash equilibrium points [20] that are undesirable and may lead to non-
convergence. To obtain better results, the NBS which requires the cooperation of base stations
was applied. Simulation results showed that a unique optimal solution was achieved by using
CHAPTER 3. COGNITIVE RADIO MANAGEMENT FUNCTIONS 26
the NBS.
3.4 Spectrum Mobility
Spectrum mobility is the process of performing seamless transition from one channel to another
available channel. After a SU selects the channel and starts transmitting on it, a PU may appear
on the same channel; in this case the SU has to move to another empty channel and vacate the
channel to avoid causing interference to the PU. The SU may also change its channel to access
another spectrum hole with better QoS. Spectrum mobility consists of two processes [12],
spectrum handoff and connection management.
• In the spectrum handoff process, the SU transfers its transmission from the current chan-
nel to another empty channel. Three events can trigger the spectrum handoff process.
The first event is the arrival of a PU in a channel occupied by a SU. The second is the
spatial movement of SUs to a place where their coverage overlap with PUs already uti-
lizing the channel. The third is the degradation of the link quality.
The spectrum handoff process consists of two phases, evaluation phase and link mainte-
nance phase. In the evaluation phase, the SU keep monitoring the environment to deter-
mine if an event that trigger spectrum handoff occurred, then the SU moves to the next
phase (link maintenance). In the Link maintenance phase, the SU pauses its transmission
on the current channel and continues the transmission on another available channel.
• Connection management process is used to compensate for the unavoidable handoff de-
lay, which happens when the SU transmission is transferred from a channel to another,
by adjusting the parameters of the protocol stack according to the existing situation.
Chapter 4
Spectrum Sensing Based on Capon PSD
Estimation
In this chapter we describe the proposed Capon PSD estimation method. We start by explaining
the spectrum sensing problem and the metrics used for measuring its performance. Then we
cover the popular energy detection technique. After that we explain how we can use the CM to
find an estimate for the PSD of the received signal, which is used for detecting spectrum holes.
We also cover the advantage of using the CM over the conventional energy detection method.
4.1 Problem Formulation
Spectrum sensing is the fundamental task of any CRD, which detect the presence or absence
of a radio frequency signal. The spectrum sensing problem can be considered as a binary
hypotheses testing problem with the following hypotheses:
H0 : Y (n) = W (n)
H1 : Y (n) = X(n) +W (n)
27
CHAPTER 4. SPECTRUM SENSING BASED ON CAPON PSD ESTIMATION 28
where the null hypothesis H0 indicates that there is no signal transmitted, only noise exist, and
the alternative hypothesis H1 indicates that there is a signal transmitted. Y (n) are the received
signal samples, W (n) are the noise samples, X(n) are the transmitted signal samples, and
n = 0, 1, 2, .., N − 1 is the sample index.
The goal of the binary hypothesis test is to decide which of the two hypotheses has occurred
by comparing the received data with a certain threshold. If we have only one received sample,
the decision can be easily taken by comparing the received sample with the threshold and
declare H1 if the received sample is greater than the threshold and H0 otherwise. However, we
usually have a sequence of data samples, and in this case we have to find the most appropriate
way to process these samples. Several detection techniques can be used to solve the binary
hypothesis testing problem. Each technique has a different strategy in processing the data
samples to obtain a test statistics. This test statistics is then compared with a threshold to
decide whether a signal exist or not. The performance of any detection technique depends
on the amount of information known about the signal and noise characteristics. If we have
less information about the signal and noise, the performance of the detection technique will
decrease. The optimal detection technique can be obtained if we have a perfect knowledge
about the signal and noise probability density functions (PDFs).
Figure 4.1 shows the two types of errors that can happen when we are deciding which
hypothesis has occurred [30]. If we decide H1 while H0 is true this is called TypeI error or
probability of false alarm PFA, and if we decideH0 whileH1 is true this is called TypeII error
or probability of miss detection PMiss. The PFA is the probability of detecting the presence of a
PU while it is actually absent. A high PFA value leads to inefficient utilization of the spectrum,
because even though the spectrum is free in reality, the SU assumes it is busy and hence will
not use it. The PMiss is the probability of declaring that the channel is free while it is actually
occupied by a PU. A high PMiss value leads to higher interference, as the SU assumes that
the channel is free, and once he starts using the channel, he will interfere with the PU already
transmitting on that channel.
CHAPTER 4. SPECTRUM SENSING BASED ON CAPON PSD ESTIMATION 29
Figure 4.1: Types of error in hypothesis testing
The performance of any detection technique is measured by its probability of detection PD
and PFA. The PD is the probability of detecting a PU while it is actually present. A high PD
value leads to less interference, as the SU will detect the presence of a PU transmission on
the channel, so he will not transmit on that channel. In mathematics the PD and PFA can be
expressed as:
PD = P (H1;H1)
= Pr{T > γ;H1}
= 1− PMiss
PFA = P (H1;H0)
= Pr{T > γ;H0}
where P (Hi;Hj) is the probability of deciding Hi when Hj is true, γ is the threshold, and
the plot of PD versus PFA is called the Receiver Operating Characteristic (ROC) curve, which
is the most important performance measure for the hypothesis testing problem. An efficient
detector should achieve a low PFA and high PD. As an example the IEEE 802.22 WRAN
standard requires a PFA of 0.1 and a PD of 0.9 [52].
CHAPTER 4. SPECTRUM SENSING BASED ON CAPON PSD ESTIMATION 30
4.2 Choosing Threshold Value
In the binary hypothesis test problem, the test statistics is compared with a certain threshold
to decide which hypothesis has occurred. Choosing the threshold value is a very critical and
challenging issue, and it can greatly affect the PFA and PMiss values. The threshold value that
is suitable for one transmitter may not be suitable for another transmitter. Since each trans-
mitter has a different signal power and transmission range. Increasing the threshold value will
decrease the PFA, but on the other hand will increase the PMiss and vice versa, so we cannot
minimize both errors simultaneously. Different criteria can be used to determine the threshold
value such as Bayesian, NeymanPearson (NP) and Minimax [56]. Most of the detectors depend
on the NP lemma for determining threshold value as it tries to maximize the PD while keeping
the PFA fixed at a constant value.
4.3 Energy Detection
Energy detection is the simplest and most popular signal detection technique used for spectrum
sensing. It can be easily implemented even if there is no information available about the pri-
mary user’s signal. The idea of energy detection is to measure the energy of the received signal
within a certain bandwidth and time period and compare it with a predefined threshold that
depends on the noise floor. If the energy exceeds the threshold the detector decides that there
is a signal transmission on the channel, otherwise there is no signal transmission and the chan-
nel can be used for SUs operation. As the signal energy is maintained in time and frequency
domain, the implementation of energy detector can be made in time and frequency domain as
well. Although the energy detector is simple and has low implementation complexity there are
some challenges associated with energy detection. These challenges include determining the
threshold value, the poor performance under low Signal to Noise Ration (SNR), and inability
to differentiate between the PU and SU signals.
CHAPTER 4. SPECTRUM SENSING BASED ON CAPON PSD ESTIMATION 31
Figure 4.2: Time domain representation of energy detection
4.3.1 Energy Detection in Time Domain
The first model for a time domain energy detector was presented by Urkowitz in [57]. Urkowitz
focused on the detection of a deterministic signal in the presence of band-limited additive white
Gaussian noise (AWGN). Urkowitz assumed that the structure of the signal is unknown, but
the noise variance is known. The designed energy detector consisted of three main blocks as
shown in Figure 4.2. The received signal is first pre-filtered by an ideal Band-Pass Filter (BPF)
to limit the noise bandwidth. Then in order to calculate the energy content of the signal, the
signal is passed through a squaring device followed by a finite time integrator. The integrator
integrates the squared signal over the observation period T to produce a test statistic. The
produced test statistics, which is actually the energy of the signal, is compared with a threshold
and if it exceeds the threshold the detector decides that a signal has been detected.
Mills and Prescott [39], presented six different detection models for the common wide
band energy detector. They tried to find a way to calculate the required SNR that will achieve
certain detection performance. Comparing results obtained from the derived models with ex-
act results showed that results from the models agreed with the exact results for very large
time-bandwidth product. In [13], Ciftci and Torlak analyzed the performance of the detec-
tion models presented in [39] in both AWGN and Rayleigh channels and compare the result
obtained with exact solution results. Simulation results showed that the performance of the
models is nearly equal in Rayleigh channel, and any of the models can be used as a suitable
approximation for energy detection instead of complicated exact solution. Lee and Akyildiz
[34], used the maximum a posteriori (MAP) based energy detection model to develop an op-
timal spectrum sensing framework. The developed framework works on optimizing sensing
CHAPTER 4. SPECTRUM SENSING BASED ON CAPON PSD ESTIMATION 32
parameters to maximize the spectrum efficiency and on the same time avoid causing interfer-
ence to PUs. In [31] Kostylev investigated the detection of a signal with random amplitude
over different fading channels such as Rayleigh, Rice, and Nakagami. The decision statistics
of the energy detector and the probability of detection expression for each fading channel has
been derived. Chen [11], also investigated the detection of a random signal in Gaussian noise,
but he made a modification to the conventional energy detector. In conventional energy de-
tector the amplitude of the signal is squared to obtain the energy. Chen replaces the squaring
operation with a positive power operation, then he made a derivation for the performance of
the improved energy detector.
4.3.2 Energy Detection in Frequency Domain
The implementation of energy detection in time domain usually suppose that the frequency
and bandwidth of the PU signal are known. However this is not the case in reality, which
encourage investigating the frequency domain implementation of energy detection. A block
diagram for measuring the energy of the signal in frequency domain is shown in Figure 4.3
[1]. First the received signal is passed through a BPF to select the frequency band of interest,
then an analog to digital converter (ADC) is used to sample the filtered signal to discrete time
samples. After that the fast Fourier transform (FFT) is taken to convert the sampled signal
from time to frequency domain. Once the signal is converted to frequency domain the energy
is calculated by squaring the coefficients and taking the average. The calculated energy is
used as the decision statistics that will be compared with a predefined threshold to decide the
presence or absence of the PU signal.
Figure 4.3: Frequency domain representation of energy detection
CHAPTER 4. SPECTRUM SENSING BASED ON CAPON PSD ESTIMATION 33
In the frequency domain implementation we are not just measuring energy, we are measur-
ing the energy spectral density, which shows the distribution of energy versus frequency, so we
can easily determine the frequency location of the PU signal [61]. We can calculate the energy
spectral density of a finite energy signal from the Fourier transform of its autocorrelation se-
quence [29]. However, most of the signals used in communication systems are random signals
that can be modeled by probabilistic terms. Such signals do not have finite energy, but they
have finite average power, so they can be described by their average PSD [53]. The problem is
that we cannot find the true PSD of the signal when all what we have about the signal is just
a finite set of measurements. However we can depend on estimating the PSD of the random
signal using one of the PSD estimation techniques.
4.4 Power Spectral Density Estimation
PSD estimation techniques are used to estimate the distribution of the total power over fre-
quency from a finite length of stationary data sequence. Two major methods can be used for
spectral estimation: [53] parametric and nonparametric methods. The parametric methods as-
sume that the data follows a certain structure or model such as Moving Average (MA) models
and AutoRegressive Moving Average (ARMA) models, which can be described by a number
of parameters. Then the spectral estimation problem is reduced to estimation of the parameters
in the assumed model. One example of the parametric methods is called MUltiple SIgnal Clas-
sification (MUSIC) method. The nonparametric methods do not need to have prior information
or make any assumption about the signal. The Periodogram and Capon methods are examples
of the nonparametric methods.
Parametric methods can provide more accurate PSD estimation than the nonparametric
methods, if the assumed model correctly reflects the behaviour of the data. Otherwise, which
is the more likely case the estimated PSD might not be reliable and the nonparametric methods
outperforms the parametric ones. Nonparametric methods are more suitable for the implemen-
CHAPTER 4. SPECTRUM SENSING BASED ON CAPON PSD ESTIMATION 34
tation of energy detector, as they do not require any information about the transmitted signal,
which matches the concept of energy detection.
4.4.1 Power Spectral Density
Consider a discrete time signal {y(t); t = 0,±1,±2, ......} consists of a stationary sequence of
random variables with zero mean E{y(t)} = 0 for all values of t. The auto-covariance function
r(k) of y(t) can be determined by:
r(k) = E{y(t)y∗(t− k)} (4.1)
where k is the time lag between two samples and * represents complex conjugate. The PSD
of the signal y(t) can be defined in two ways:
• First it can be defined as the DTFT of the auto-covariance sequence of the signal:
ϕ(ω) =∞∑
k=−∞
r(k)e−iωk (4.2)
• Or it can be defined according to the equation:
ϕ(ω) = limN→∞
E
1
N
∣∣∣∣∣N∑t=1
y(t)e−iωt
∣∣∣∣∣2 (4.3)
If we assume that the auto-covariance sequence of y(t) decays so fast, The two PSD definition
will be equivalent.
4.4.2 Periodogram
The Periodogram is the most widely used method for energy detection. The PSD estimate
using Periodogram method depends on equation (4.3) of the PSD definition. The difference is
CHAPTER 4. SPECTRUM SENSING BASED ON CAPON PSD ESTIMATION 35
that the expectation and limit operation in equation (4.3) were neglected because they cannot
be computed when the only known information about the signal is its samples.
∧ϕp(ω) =
1
N
∣∣∣∣∣N∑t=1
y(t)e−iωt
∣∣∣∣∣2
(4.4)
The performance or the quality of an estimator is usually measured based on two parameters
named, the bias and variance of the estimator [48]. The difference between the expected
value and the true value of the estimated parameter represents the bias of the estimator. The
variance of the estimator is a measure of how the estimated values varies around their mean.
A good estimator should have the value of its bias and variance fade when the number of data
sequences increase to be infinite.
The Periodogram is considered as a bad estimator. Since it is unbiased estimator and its
variance does not decrease even if we increase the data length. Moreover, it has a poor reso-
lution, which means that its ability to identify frequency components that are closely spaced
is not good [29]. However the resolution of the Periodogram method can be improved by
increasing the length of the observed data samples.
4.4.3 Capon
The drawbacks of the Periodogram method encourage the investigation of other spectral esti-
mation methods that can overcome the high variance and poor resolution such as the Capon
Method (CM). The CM, also known as minimum variance method [29] uses a Filter Bank
Approach (FBA) based on a data dependent BPF to estimate the PSD of the signal. The idea
behind the FBA is that if the input signal is applied to an array of BPFs, each filter is tuned to
pass certain frequency band without distortion while minimizing the power of other frequency
components in the signal. Thus, the signal can be divided into multiple components with each
component representing a single frequency sub-band of the input signal.
In the CM the signal of interest is introduced to only one BPF that has a narrow bandwidth
CHAPTER 4. SPECTRUM SENSING BASED ON CAPON PSD ESTIMATION 36
[53, 24]. The filter band-width can be swept over the frequency band of interest to calculate the
value of the estimated spectrum. The filter should be designed to pass the required frequency
band undistorted while minimize the power outside that band. In order to do that consider a
Finite Impulse Response (FIR) filter whose impulse response h is given by:
h = [h0 h1 ...... hm]∗ (4.5)
where m is a positive integer representing the length of the impulse response of the filter.
If the data sequence y(t) is introduced as an input to the filter, then the filter output in
response to the input data at time t can be given by:
yF (t) =∑k=0
hky(t− k)
= h∗
y(t)
y(t− 1)
...
y(t−m)
(4.6)
The power of the filtered signal yF (t) can be written as:
E{|yF (t)|2
}= h∗Rh (4.7)
where R is the covariance matrix of the input data vector and is equivalent to:
R = E
y(t)
y(t− 1)
...
y(t−m)
[y∗(t).....y∗(t−m)]
(4.8)
Our goal is to determine how the output power calculated above is distributed over frequency.
CHAPTER 4. SPECTRUM SENSING BASED ON CAPON PSD ESTIMATION 37
In order to do that, we need to look at the frequency response of the filter, which shows how
the filter will react to a sinusoidal input of the form e−jω, where ω represents the frequency.
The frequency response of the filter is given by:
H(ω) =m∑k=0
hke−iωk = h∗a(ω) (4.9)
where
a(ω) = [1 e−iω ..... e−imω]T (4.10)
We want to design the filter to be as selective as possible for a certain frequency band centered
around frequency ω. So, we can think of minimize the total power at the filter output sub-
ject to the constraint of passing the frequency band centered around ω without distortion and
suppressing other frequencies. This is equivalent to the optimization problem [53]:
minhh∗Rh subject to h∗a(ω) = 1 (4.11)
The solution to this optimization problem is given by [53]:
h = R−1a(ω)/a∗(ω)R−1a(ω) (4.12)
If we substitute for the value of h in equation (4.7) we can find that the PSD of the input
sequence y(t) at the output of the BPF with center frequency ω can be written as:
∧ϕc(ω) ' 1/a∗(ω)R−1a(ω) (4.13)
4.5 System Model
Consider a number of primary transmitters (tx) with fixed locations operating inside an area
(A). Each transmitter can transmit on a number of channels (ch), and each channel operates
CHAPTER 4. SPECTRUM SENSING BASED ON CAPON PSD ESTIMATION 38
on a different frequency (fc). The transmitters can randomly transmit in all or some of the
channels. In order to efficiently utilize the spectrum, a number of SUs with CR capability are
distributed in the area. Our goal is to allow SUs to perform spectrum sensing based on the
Capon and Periodogram PSD estimation methods and see which method has a better detection
performance.
We assume that the transmitters are transmitting a continuous time signal in the form:
s(t) = A(t)eiωt + n(t) (4.14)
where A(t) represents the signal amplitude, which is a complex Gaussian random variable
with zero mean and variance σ2s , n(t) is the noise which is also a complex Gaussian random
variables with zero mean and variance σ2n. The received signal strength at any location can
be calculated using the free space path loss model. According to this model, the path loss
is proportional to the separation distance between the transmitter and receiver raised to some
power. The relation between the transmitted and received power can be given by Friis free
space equation [50]:
Pr(d) =PtGtGrλ
2
(4π)2d2L(4.15)
where Pt is the transmitted power, Pr(d) is the received power as a function of distance, Gt
is the transmitter antenna gain, Gr is the receiver antenna gain, Gt and Gr are dimensionless
quantities, d is the separation distance in meters between the transmitter and receiver, L is the
system loss factor not related to propagation (L >= 1), and λ is the wavelength in meters
which is related to the carrier frequency according to the equation:
λ =c
f=
2πc
ωc
(4.16)
where c is the speed of light given in m/s, f is the carrier frequency in Hz, ωc is the carrier
frequency in rad/s.
CHAPTER 4. SPECTRUM SENSING BASED ON CAPON PSD ESTIMATION 39
Figure 4.4: Spectrum sensing steps
In order to perform spectrum sensing the CRD has to perform some steps illustrated by
Figure 4.4. First the device senses the medium to determine the strength of the received signal
on its location, which can be calculated by making use of equation 4.15 as follow:
Sr(d) = St
√GtGrλ2
(4π)2d2L(4.17)
where St is the transmitted signal, Sr(d) is the received signal, and d is the distance between
the transmitter and receiver. The distance can be calculated using the equation of the euclidean
distance between two points [18], which is given by:
d =√
(xt − xr)2 + (yt − yr)2 (4.18)
where (xt, yt) are the transmitter coordinates and (xr, yr) are the receiver coordinates.
After sensing the medium the device needs to process the received signal samples (sensed
signal) to obtain a test statistics. In our model we use the Capon and Periodogram methods to
process the received signal samples. These methods provide an estimation for the received PSD
at the SU location, calculated using equation 4.13 for Capon and equation 4.4 for Periodogram.
The estimated PSD is used as the test statistics for the hypothesis test, which is compared with
a predefined threshold to identify busy channels and white spaces at the SU location.
Chapter 5
Simulation and Results
In this chapter, the performance of the Capon and Periodogram methods introduced in Chapter
4 are evaluated through simulations. Three different scenarios are considered: 1) determining
the operating frequency of primary transmitters, 2) measuring the performance of the Capon
and Periodogram methods, 3) cooperative spectrum sensing based on the Capon method. All
the simulations are conducted using the MATLAB software. For all the three scenarios, the
values of the parameters listed in the table below are set to the following default values:
Parameter name Parameter valueSimulation area A = 1 km2
Number of transmitters tx = 3Transmitter antenna gain Gt = 1Receiver antenna gain Gr = 1system loss L = 1propagation speed c = 3 ∗ 108m/sNumber of samples N = 1000
Table 5.1: Common parameters for the three scenarios.
40
CHAPTER 5. SIMULATION AND RESULTS 41
5.1 Determining the Operating Frequency of Primary Trans-
mitters
In the first scenario, we investigate the ability of the Capon and Periodogram methods to de-
termine the operating frequency of primary transmitters. Assume that we have three primary
TV transmitters distributed as shown in Figure 5.1, and each transmitter transmits only on one
channel. The TV transmitters are transmitting on frequencies fc = 57, 69, 79MHz.
Figure 5.1: SUs distribution with respect to primary transmitters
To be able to estimate the PSD at the locations of the two SUs, we start by calculating the
strength of the received signal from each transmitter using Equation 4.17. The total received
signal at the SU location is actually the summation of the received signals from the three
transmitters given by:
Sr(SU) =3∑
tx=1
Sr(tx) (5.1)
where Sr(SU) is the total received signal at the SU location and Sr(tx) is the received signal
CHAPTER 5. SIMULATION AND RESULTS 42
Figure 5.2: PSD at the location of SU1 for high SNR
from the tx transmitter calculated using Equation 4.17. The total received signal is the signal
sensed by the CRD which is used as the input for both the Capon and Periodogram methods.
The PSD estimation using the Periodogram method is calculated using Equation 4.4 which
relies on the Fast Fourier Transform (FFT) of the input data sequence. On the other hand, the
CM estimates the PSD using Equation 4.13.
Figure 5.2 plots the estimated PSD versus frequency for both the Capon and Periodogram
methods measured at SU1. The figure illustrates the ability of the Capon method to determine
the operating frequency of the three primary transmitters, while it is hard to identify them in
case of Periodogram due to the excessive amount of noise. Figure 5.3 displays the effect of
decreasing the SNR on the estimated PSD at SU1. It is clear from the figure that as the noise
level increase, the CM is still able to determine the operating frequency of the three primary
transmitters, while it becomes harder in case of Periodogram. Similar results are obtained for
SU2 which are shown in Figures 5.4 and 5.5.
CHAPTER 5. SIMULATION AND RESULTS 43
Figure 5.3: PSD at the location of SU1 for low SNR
Figure 5.4: PSD at the location of SU2 for high SNR
CHAPTER 5. SIMULATION AND RESULTS 44
Figure 5.5: PSD at the location of SU2 for low SNR
5.2 Measuring the Detection Performance
In this scenario, we measure the ability of the Capon and Periodogram methods to detect busy
channels and white spaces at the SUs locations. In order to have white spaces in the simula-
tion area, we assume that each transmitter has a fixed number of channels, and each channel
operates on a different frequency. The transmitters can randomly transmit in some or all of
the channels. When the transmitter is only transmitting on some of its channels, the remaining
channels are available for use by SUs. If the transmitter transmits in all of the channels, this
means that no white spaces are available for use at that time. The SU in this case have to wait
a period of time and then repeat the sensing process to see if a channel becomes available.
In order to detect the status of the channel, the CRD starts by sensing the medium to mea-
sure the received signal on its location, which is calculated based on Equation 4.17. The sam-
ples of the received signal is then used as an input for our detection methods. The capon method
estimates the PSD at the SU location using Equation 4.13, while the Periodogram method uses
CHAPTER 5. SIMULATION AND RESULTS 45
Equation 4.4. After that, the estimated PSD is compared with a threshold and if it is greater
than the threshold, the channel is considered busy, otherwise the channel is considered free.
In our simulation we use the following parameters:
Parameter name Parameter value
No of channels in each transmitterchtx1 = 3chtx2 = 3chtx3 = 1
Operating frequencies of tx1fch1 = 57MHzfch2 = 85MHzfch3 = 69MHz
Operating frequencies of tx2fch1 = 79MHzfch2 = 183MHzfch3 = 177MHz
Operating frequency of tx3 fch1 = 63MHzThreshold range γ = 0 : 10−13 W
Table 5.2: Performance detection parameters.
In order to calculate the detection probability of each method we made the following steps.
1. We generate a random vector of 0′s and 1′s indicating the original state of the channels
in the three primary transmitters.
2. We perform a hypothesis test based on our detection methods to determine the status of
the transmitter channels.
3. We compare the result obtained from the hypothesis test with the original state of each
channel to determine correctly and incorrectly detected channels.
4. We count the number of correctly detected busy channels and divide it by the total num-
ber of busy channels to calculate the probability of detecting primary transmitters PD for
each iteration.
5. We count the number of correctly detected free channels and divide it by the total num-
ber of free channels to calculate the probability of detecting white spaces PW for each
iteration.
CHAPTER 5. SIMULATION AND RESULTS 46
6. We count the number of incorrectly detected busy channels and divide it by the total
number of busy channels to calculate the probability of miss detection PMiss for each
iteration.
7. We count the number of incorrectly detected free channels and divide it by the total
number of free channels to calculate the probability of false alarm PFA for each iteration.
8. The PD of each method is calculated as the sum of PD of each iteration divided by the
number of iterations, we did the same for PW , PMiss, and PFA.
5.2.1 Detecting Primary Transmitters
Figure 5.6 plots the PD versus different threshold values. The figure illustrates that the PD can
be greatly affected by changing the threshold value. As we mentioned earlier in Chapter 4, the
performance of any detection technique is measured by the value of its PD and PFA. An ideal
detector should maximize the PD and minimize PFA, which cannot happen in reality at the
Figure 5.6: Probability of detecting primary transmitters versus threshold
CHAPTER 5. SIMULATION AND RESULTS 47
Figure 5.7: Probability of false alarm versus threshold
same time. Selecting a zero threshold value will achieve a PD equal to one, but on the other
hand will increase the PFA to one. If we start increasing the threshold value we can see that
the PD will decrease.
Figure 5.7 displays the PFA versus different threshold values. The figure illustrates that
when the threshold value equal to zero any noise in the channel no matter how small it is,
cause the detector to falsely assume that there is a signal transmitted in the channel. That is
the reason why when the threshold value is zero, the PFA equals one. Increasing the threshold
value will cause the PFA to decrease. Thus, choosing the threshold value is a very critical issue
when designing a detector.
The performance of any detector can be measured from its ROC curve, which is a plot for
the PD versus PFA for different threshold values. The ROC curves for both the Capon and
Periodogram methods are shown in Figure 5.8. The figures shows that for a given PFA the PD
of the Capon method is better than that of the Periodogram method, which means that the CM
has a better performance.
CHAPTER 5. SIMULATION AND RESULTS 48
Figure 5.8: Average ROC for both Capon and Periodogram methods
5.2.2 Detecting White Spaces
Till now, we saw the effect of changing threshold on PD, as we want to have a good PD to
protect PUs from interference. Meanwhile, our major goal is to efficiently utilize the spectrum
by making use of the white spaces. Figure 5.9 plots the PW versus threshold. From the figure
we can see that having a zero threshold value will cause PW to be zero, which means that we
are not efficiently utilizing the spectrum. Increasing the threshold value will achieve a better
PW , but on the other hand will increase the PMiss as we can see in Figure 5.10. A high PMiss
value means that the detector fails in detecting most of the primary transmitters, hence those
primary transmitter are susceptible to be interfered by SUs. Thus, we should carefully select a
threshold value that can achieve a good PW with low PMiss value. Figure 5.11 displays a plot
for the PW versus PMiss. The figure illustrates that at any given PMiss value, the PW for the
CM is better than that of the Periodogram method. This indicates that the CM outperforms the
Periodogram method in detecting white spaces.
CHAPTER 5. SIMULATION AND RESULTS 49
Figure 5.9: Probability of detecting white spaces versus threshold
Figure 5.10: Probability of miss detection versus threshold
CHAPTER 5. SIMULATION AND RESULTS 50
Figure 5.11: Average probability of detecting white spaces versus probability of miss detection
5.3 Cooperative Spectrum Sensing Based on Capon Method
As we saw from previous simulation results that the CM outperforms the Periodogram method
in detecting both primary transmitters and white spaces. In this scenario, we are implementing
a cooperative spectrum sensing based on the CM. The goal of the cooperative spectrum sensing
is to help SUs that do not have CRDs to identify white spaces in their location. The cooperative
spectrum sensing can be also used to overcome the hidden node problem that usually happen
in primary transmitter detection, as discussed in Section 3.1.1.
We consider the same simulation area and primary transmitters used in the second scenario.
In addition, we assume two types of SUs, sensors and nodes, which are shown in Figure 5.12.
Sensors are SUs equipped with CRDs and are uniformly distributed in the simulation area.
Nodes are SUs that do not have CRDs, so they cannot determine white spaces on their own.
Those nodes use cooperative spectrum sensing to determine white spaces on their location.
Each sensor use Equation 4.13 of the CM to estimate the PSD on its location. In order for
the node to find white spaces on its location, it depends on the estimated PSD of the sensors
CHAPTER 5. SIMULATION AND RESULTS 51
Figure 5.12: Distribution of sensors and nodes
located within a certain search distance. We assume that the node can establish a common
control channel (CCC) [6, 35] with its neighboring sensors to collect their estimated PSD.
Then, the node processes the collected PSD of the different sensors by dividing the PSD of
each sensor by the square of the separation distance between the node and that sensor. After
performing the division the node will add up all the resultant PSD values and consider the result
as the PSD on its location. The node will compare the calculated PSD value with a predefined
threshold to determine the availability of the channel.
The PSD at the node location can be calculated by:
ϕi=
N(i)∑j=1
∧ϕj
d2(i, j)
(5.2)
where, ϕi
represents the calculated PSD at the node i location,∧ϕj
is the estimated PSD of
neighboring sensor j,N(i) is the number of neighboring sensors located within a certain search
distance from node i, and d(i, j) is the distance between the sensor j and node i.
CHAPTER 5. SIMULATION AND RESULTS 52
The performance of the proposed cooperative spectrum sensing is investigated under dif-
ferent key parameters such as threshold, number of sensors, and search distance.
5.3.1 Changing Threshold Value
First we start by investigating the effect of changing threshold value on the performance of the
cooperative spectrum sensing method. We assume a number of sensors S = 1000 and they
Figure 5.13: ROC of cooperative spectrum sensing
are uniformly distributed in the simulation area A = 1 Km2 and a search distance d = 500m.
The performance of the cooperative spectrum sensing is measured by its ROC, which is shown
in Figure 5.13. The figure indicates that the cooperative spectrum sensing achieves higher PD
than the spectrum sensing performed by a single SU shown in Figure 5.8. Figure 5.14 plots PW
versus PMiss. The figure illustrates that the cooperative spectrum sensing approach achieves
higher PW values for lower values of PMiss compared to the result obtained from spectrum
sensing performed by a single SU displayed in Figure 5.11.
CHAPTER 5. SIMULATION AND RESULTS 53
Figure 5.14: PW versus PMiss for cooperative spectrum sensing with different threshold values
5.3.2 Changing Number of Sensors
Now we want to see the effect of changing the number of sensors in the simulation area S =
0 : 700 on the performance of the cooperative spectrum sensing, while keeping the value
of the threshold and search distance fixed. Figure 5.15 discusses the effect of increasing the
number of sensors on the PD. Increasing the number of sensors means that the sensors will be
densely distributed on the simulation area, hence larger number of sensors will be available for
cooperation within the same search distance, which achieves a better PD.
The plot for PD versus PFA for different number of sensors is shown in Figure 5.16. This
figure is similar to the ROC figure, but here the PFA value does not reach 1 as we have a fixed
threshold value that is greater than zero. Figure 5.17 displays a plot for PW versus PMiss for
different number of sensors. When the number of sensors equal to zero, the value of PMiss
equals 1 . Having S = 0 means that the simulation area does not contain any sensors that
could participate in the cooperative spectrum sensing. So, we can not get information about the
primary transmitters. Increasing the number of sensors will decrease the PMiss to an acceptable
values, while keeping good PW values.
CHAPTER 5. SIMULATION AND RESULTS 54
Figure 5.15: PD versus number of sensors for cooperative spectrum sensing
Figure 5.16: Average PD versus PFA for cooperative spectrum sensing with different numberof sensors
CHAPTER 5. SIMULATION AND RESULTS 55
Figure 5.17: Average PW versus PMiss for cooperative spectrum sensing with different numberof sensors
5.3.3 Changing Search Distance
In this section, we investigate the effect of changing the search distance d = 0 : 700 on the
performance of the cooperative spectrum sensing, while keeping a fixed threshold value and
fixed number of sensors S = 700. Figure 5.18 indicates that increasing the search distance
achieves higher PD values. Figure 5.19 shows a plot for PD versus PFA for different search
distances. This figure is different from the ROC, as it is an increasing curve that starts from
PD = 0 when d = 0. Increasing the value of d will cause the PD value to grow. On the other
hand, Figure 5.20 is a decreasing curve representing the plot of PW versus PMiss for different
values of search distance. When the search distance d = 0 this is similar to having S = 0. We
do not have any information about primary transmitters, which makes PMiss = 1. Increasing
the search distance means that we will search for neighboring sensors in a larger distance.
Thus, it is expected that more sensors will cooperate and the PMiss will decrease. At the same
time, we will notice a slight decrease in the PW .
CHAPTER 5. SIMULATION AND RESULTS 56
Figure 5.18: PD versus different search distances in cooperative spectrum sensing
Figure 5.19: Average PD versus PFA with different search distances for cooperative spectrumsensing
CHAPTER 5. SIMULATION AND RESULTS 57
Figure 5.20: Average PW versus PMiss with different search distances for cooperative spec-trum sensing
Chapter 6
Conclusion and Future Work
In this thesis, we have discussed the spectrum scarcity problem that appeared due to the natu-
ral limitation of the radio spectrum, and the underutilization of the allocated spectrum bands.
In Chapter 1, we showed that the current FCA is not suitable to accommodate the increasing
demand for high data rate services and the increasing number of wireless users. DSA tech-
niques that can be used to overcome the limitation of the FCA scheme and the underutilization
of the allocated spectrum were covered in Chapter 2. DSA techniques allow SUs to oppor-
tunistically use the licensed spectrum when it is not used by PUs. For the SUs to access the
licensed spectrum, they have to be equipped with devices that have CR capability to determine
white spaces in the allocated spectrum. Chapter 3, discussed the four CR management func-
tions that must be performed by any CRD. In Chapter 4, we have proposed a spectrum sens-
ing based on the Capon PSD estimation method. In addition, we have discussed our system
model and covered the popular Periodogram spectral estimation method. Chapter 5 covered
a simulation implemented in Matlab for the Periodogram and Capon PSD estimation meth-
ods. The chapter presented the experimental results for three studied scenarios. The results
showed that the proposed spectrum sensing based on the capon method outperforms the one
based on the Periodogram method. The first scenario showed that the Capon method is more
accurate in identifying the frequency locations of primary transmitters. The second scenario
58
CHAPTER 6. CONCLUSION AND FUTURE WORK 59
illustrates that the PD value obtained from the Capon method is higher than that obtained from
the Periodogram method for any given PFa value. Similar results were obtained for PW . The
third scenario demonstrated that the performance of the cooperative scenario is affected by the
search distance and the number of sensors included in the cooperation. Results also showed
that the cooperative spectrum sensing is more efficient in detecting busy and free channels than
the spectrum sensing performed by a single SU.
The work done in this thesis can be extended as follow:
• In the thesis, the performance of the Capon method is investigated and compared to the
performance of the Periodogram method. However, the effect of using different fading
channels on the performance of the Capon method is not covered.
• In the thesis, the threshold value is set based on simulation results. Further investigation
is required to drive a mathematical expression for selecting the threshold value based on
a given PFA or using any of the criteria mentioned in Chapter 4.
• Although simulation results prove that, the Capon method outperforms the Periodogram
method, it needs more computational time than the Periodogram method to calculate the
estimated PSD. The thesis used TV white spaces which do not change so fast, hence the
Capon method can be an effective method, but in other applications time can be more
critical than accuracy.
• In the cooperative spectrum sensing, it is assumed that, the node can establish a Common
Control Channel (CCC) to collect the PSD from its neighboring sensors. The design of
the CCC is a challenging issue in cognitive radio networks which needs more investiga-
tion.
Bibliography
[1] Mahmood A Abdulsattar and Zahir A Hussein. Energy detection technique for spectrum
sensing in cognitive radio: a survey. International Journal of Computer Networks &
Communications, 4(5), 2012.
[2] Ian F Akyildiz, Won-Yeol Lee, Mehmet C Vuran, and Shantidev Mohanty. Next gener-
ation/dynamic spectrum access/cognitive radio wireless networks: a survey. Computer
Networks, 50(13):2127–2159, 2006.
[3] Ian F Akyildiz, Won-Yeol Lee, Mehmet C Vuran, and Shantidev Mohanty. A survey on
spectrum management in cognitive radio networks. Communications Magazine, IEEE,
46(4):40–48, 2008.
[4] Saman Atapattu, Chintha Tellambura, and Hai Jiang. Energy detection based cooperative
spectrum sensing in cognitive radio networks. Wireless Communications, IEEE Transac-
tions on, 10(4):1232–1241, 2011.
[5] Vo Nguyen Quoc Bao, Tran Dinh Thuan, Nguyen Thien Quy, Lam Minh Trung, et al.
Vietnam spectrum occupancy measurements and analysis for cognitive radio applications.
In Advanced Technologies for Communications (ATC), 2011 International Conference on,
pages 135–143. IEEE, 2011.
[6] Kaigui Bian, Jung-Min Park, and Ruiliang Chen. Control channel establishment in cog-
nitive radio networks using channel hopping. Selected Areas in Communications, IEEE
Journal on, 29(4):689–703, 2011.
60
BIBLIOGRAPHY 61
[7] Danijela Cabric, Shridhar Mubaraq Mishra, and Robert W Brodersen. Implementation
issues in spectrum sensing for cognitive radios. In Signals, systems and computers, 2004.
Conference record of the thirty-eighth Asilomar conference on, volume 1, pages 772–776.
IEEE, 2004.
[8] Industry Canada. Framework for the use of certain non-broadcasting applications in
the television broadcasting bands below 698 mhz. http://www.ic.gc.ca/eic/
site/smt-gst.nsf/eng/sf10493.html, April 2013. Accessed Feb. 6, 2015.
[9] Industry Canada. Consultation on a policy and technical framework for the use of non-
broadcasting applications in the television broadcasting bands below 698 mhz. http://
www.ic.gc.ca/eic/site/smt-gst.nsf/eng/sf10058.html#a5.3, Au-
gust 2011. Accessed Feb. 6, 2015.
[10] K-C Chen, Y-J Peng, Neeli Prasad, Y-C Liang, and Sumei Sun. Cognitive radio network
architecture: part i–general structure. In Proceedings of the 2nd international confer-
ence on Ubiquitous information management and communication, pages 114–119. ACM,
2008.
[11] Yunfei Chen. Improved energy detector for random signals in gaussian noise. Wireless
Communications, IEEE Transactions on, 9(2):558–563, 2010.
[12] Ivan Christian, Sangman Moh, Ilyong Chung, and Jinyi Lee. Spectrum mobility in cog-
nitive radio networks. Communications Magazine, IEEE, 50(6):114–121, 2012.
[13] Selami Ciftci and Murat Torlak. A comparison of energy detectability models for spec-
trum sensing. In Global Telecommunications Conference, 2008. IEEE GLOBECOM
2008. IEEE, pages 1–5. IEEE, 2008.
[14] Federal Communications Commission. Longley-rice methodology for evaluating
tv coverage and interference. http://transition.fcc.gov/Bureaus/
BIBLIOGRAPHY 62
Engineering_Technology/Documents/bulletins/oet69/oet69.pdf,
Feb. 2004. Accessed Jan. 8, 2015.
[15] Federal Communications Commission et al. Establishment of interference temperature
metric to quantify and manage interference and to expand available unlicensed operation
in certain fixed mobile and satellite frequency bands. Et Docket, (03-237), 2003.
[16] Federal Communications Commission et al. Unlicensed operation in the tv broadcast
bands. ET Docket, (04-186), 2004.
[17] Federal Communications Commission et al. In the matter of: Unlicensed operation in the
tv broadcast bands (et docket no. 04-186) and additional spectrum for unlicensed devices
below 900 mhz and in the 3 ghz band (et docket no. 02-380). FCC 10-174: Second
Memorandum Opinion and Order, 2010.
[18] Michael Corral. Vector Calculus. Michael Corral, 2008.
[19] U.S FCC. Second report and order and memorandum opinion and order. Act, Enforcement
and ET Docket No, (08-260), 2008.
[20] Mark Felegyhazi and Jean-Pierre Hubaux. Game theory in wireless networks: A tutorial.
Technical report, 2006.
[21] FCC Spectrum Policy Task Force. Report of the spectrum efficiency working group,
2002.
[22] A William Gardner. Exploitation of spectral redundancy in cyclostationary signals. Signal
Processing Magazine, IEEE, 8(2):14–36, 1991.
[23] Anita Garhwal and Partha Pratim Bhattacharya. A survey on dynamic spectrum access
techniques for cognitive radio. arXiv preprint arXiv:1201.1964, 2012.
[24] Simon Haykin and K. J. Ray Liu. Handbook on array processing and sensor networks,
volume 63. John Wiley & Sons, 2010.
BIBLIOGRAPHY 63
[25] Ekram Hossain and Vijay K Bhargava. Cognitive wireless communication networks.
Springer, 2007.
[26] Ghazale Hosseinabadi. Downlink channel assignment and power control for cognitive
radio networks using game theory. http://ufdcimages.uflib.ufl.edu/UF/
E0/04/60/86/00001/OJOWU_O.pdf. Accessed Feb. 6, 2015.
[27] Md Habibul Islam, Choo Leng Koh, Ser Wah Oh, Xianming Qing, Yoke Yong Lai, Cavin
Wang, Ying-Chang Liang, Bee Eng Toh, Francois Chin, Geok Leng Tan, et al. Spec-
trum survey in singapore: Occupancy measurements and analyses. In Cognitive Radio
Oriented Wireless Networks and Communications, 2008. CrownCom 2008. 3rd Interna-
tional Conference on, pages 1–7. IEEE, 2008.
[28] Bisera Jankuloska, Vladimir Atanasovski, and Liljana Gavrilovska. Combined
power/channel allocation method for efficient spectrum sharing in tv white space sce-
nario. In Proceedings of the 4th International Conference on Cognitive Radio and Ad-
vanced Spectrum Management, page 59. ACM, 2011.
[29] Ode Ojowu Jr. Data-adaptive spectral estimation algorithm and their sensing appli-
cations. http://ufdcimages.uflib.ufl.edu/UF/E0/04/60/86/00001/
OJOWU_O.pdf, 2013. Accessed Feb. 6, 2015.
[30] S.M. Kay. Fundamentals of Statistical Signal Processing: Detection theory. Prentice
Hall Signal Processing Series. Prentice-Hall PTR, 1998.
[31] Vladimir I Kostylev. Energy detection of a signal with random amplitude. In Communica-
tions, 2002. ICC 2002. IEEE International Conference on, volume 3, pages 1606–1610.
IEEE, 2002.
[32] Won Yeol Lee. Spectrum management in cognitive radio wireless networks. Thesis,
(PHD) Georgia Institute of Technology, 2009.
BIBLIOGRAPHY 64
[33] Won-Yeol Lee and Ian F Akyildiz. Joint spectrum and power allocation for inter-cell spec-
trum sharing in cognitive radio networks. In New Frontiers in Dynamic Spectrum Access
Networks, 2008. DySPAN 2008. 3rd IEEE Symposium on, pages 1–12. IEEE, 2008.
[34] Won-Yeol Lee and Ian F Akyildiz. Optimal spectrum sensing framework for cognitive ra-
dio networks. Wireless Communications, IEEE Transactions on, 7(10):3845–3857, 2008.
[35] Brandon F. Lo. A survey of common control channel design in cognitive radio networks.
Physical Communication, 4(1):26–39, 2011.
[36] Liangping Ma, Yingxue Li, and Alpaslan Demir. Matched filtering assisted energy detec-
tion for sensing weak primary user signals. In Acoustics, Speech and Signal Processing
(ICASSP), 2012 IEEE International Conference on, pages 3149–3152. IEEE, 2012.
[37] Moshe Timothy Masonta, Mjumo Mzyece, and Ntsibane Ntlatlapa. Spectrum decision
in cognitive radio networks: A survey. Communications Surveys & Tutorials, IEEE,
15(3):1088–1107, 2013.
[38] Meftah Mehdawi, N Riley, K Paulson, A Fanan, and M Ammar. Spectrum occupancy
survey in hull-uk for cognitive radio applications: Measurement & analysis. INTERNA-
TIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH, 2(4), 2013.
[39] Robert F Mills and Glenn E Prescott. A comparison of various radiometer detection
models. Aerospace and Electronic Systems, IEEE Transactions on, 32(1):467–473, 1996.
[40] Jayanta Mishra, Deepak Kumar Barik, and Ch Manoj Kumar Swain. Cyclostationary
based spectrum sensing in cognitive radio: Windowing approach. International Journal
of Recent Technology and Engineering (IJRTE), 3(1), 2014.
[41] Joseph Mitola and Gerald Q Maguire Jr. Cognitive radio: making software radios more
personal. Personal Communications, IEEE, 6(4):13–18, 1999.
BIBLIOGRAPHY 65
[42] Ola A. Mohammed, Khalil El-Khatib, and Miguel V. Martin. A survey of cognitive
radio management functions. In COCORA 2014, The Fourth International Conference
on Advances in Cognitive Radio, pages 6–13, 2014.
[43] Rohan Murty, Ranveer Chandra, Thomas Moscibroda, and Paramvir Bahl. Senseless:
A database-driven white spaces network. Mobile Computing, IEEE Transactions on,
11(2):189–203, 2012.
[44] Ofcom. Implementing geolocation. http://stakeholders.ofcom.org.uk/
binaries/consultations/geolocation/statement/statement.pdf.
Accessed Jan. 8, 2015.
[45] Dong-Chan Oh and Yong-Hwan Lee. Energy detection based spectrum sensing for sens-
ing error minimization in cognitive radio networks. International Journal of Communi-
cation Networks and Information Security (IJCNIS), 1(1), 2009.
[46] KP Pavithra. Optimization of energy detection in cognitive radio networks. International
Journal of Computer Trends and Technology (IJCTT), 12(1), 2014.
[47] V Prithiviraj, B Sarankumar, A Kalaiyarasan, P Praveen Chandru, and N Nandakumar
Singh. Cyclostationary analysis method of spectrum sensing for cognitive radio. In
Wireless Communication, Vehicular Technology, Information Theory and Aerospace &
Electronic Systems Technology (Wireless VITAE), 2011 2nd International Conference on,
pages 1–5. IEEE, 2011.
[48] A. Quinquis. Digital Signal Processing Using Matlab. ISTE. Wiley, 2010.
[49] V Vijayakumari R Vadivelu, K Sankaranarayanan. Matched filter based spectrum sens-
ing for cognitive radio at low signal to noise ratio. Journal of Theoretical and Applied
Information Technology, 62(1), 2014.
BIBLIOGRAPHY 66
[50] Theodore S Rappaport et al. Wireless communications: principles and practice, volume 2.
prentice hall PTR New Jersey, 1996.
[51] S Shobana, R Saravanan, and R Muthaiah. Matched filter based spectrum sensing on
cognitive radio for ofdm wlans. International Journal of Engineering and Technology
(IJET), 5(1), 2013.
[52] Carl R. Stevenson, Gerald Chouinard, Zhongding Lei, Wendong Hu, Stephen J. Shell-
hammer, and Winston Caldwell. Ieee 802.22: The first cognitive radio wireless regional
area network standard. IEEE Communications Magazine, 47(1):130–138, 2009.
[53] Petre Stoica and Randolph L. Moses. Spectral Analysis of Signals. Prentice Hall, Upper
Saddle River, NJ, 2005.
[54] SURA. Chart of the electromagnetic spectrum. https://commons.wikimedia.
org/wiki/File%3aLambdaPlaques.jpg, 2006. Accessed Jan. 8, 2015.
[55] Elias Z Tragos, Sherali Zeadally, Alexandros G Fragkiadakis, and Vasilios A Siris. Spec-
trum assignment in cognitive radio networks: A comprehensive survey. IEEE Communi-
cations Surveys and Tutorials, 15(3):1108–1135, 2013.
[56] H.L. Van Trees. Detection, Estimation, and Modulation Theory. Number pt. 1 in Detec-
tion, Estimation, and Modulation Theory. Wiley, 2004.
[57] Harry Urkowitz. Energy detection of unknown deterministic signals. Proceedings of the
IEEE, 55(4):523–531, 1967.
[58] Hano Wang, Gosan Noh, Dongkyu Kim, Sungtae Kim, and Daesik Hong. Advanced sens-
ing techniques of energy detection in cognitive radios. Communications and Networks,
Journal of, 12(1):19–29, 2010.
BIBLIOGRAPHY 67
[59] Ben Wild and Kannan Ramchandran. Detecting primary receivers for cognitive radio
applications. In New Frontiers in Dynamic Spectrum Access Networks, 2005. DySPAN
2005. 2005 First IEEE International Symposium on, pages 124–130. IEEE, 2005.
[60] Jwo-Yuh Wu, Pei-Hsin Huang, Tsang-Yi Wang, and Vincent WS Wong. Energy detection
based spectrum sensing with random arrival and departure of primary user’s signal. In
Globecom Workshops (GC Wkshps), 2013 IEEE, pages 380–384. IEEE, 2013.
[61] Alexander M Wyglinski and Di Pu. Digital Communication Systems Engineering with
Software-defined Radio. Artech House, 2013.
[62] Qing Zhao and Brian M Sadler. A survey of dynamic spectrum access. Signal Processing
Magazine, IEEE, 24(3):79–89, 2007.