Dynamic Spectrum Allocation for Cognitive Radio Networks: A Comprehensive Optimization Approach by Ayman Sabbah A thesis submitted to the Department of Electrical and Computer Engineering in conformity with the requirements for the degree of Doctor of Philosophy Queen’s University Kingston, Ontario, Canada October 2015 Copyright c Ayman Sabbah
158
Embed
Dynamic Spectrum Allocation for Cognitive Radio Networks ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
P jr,thr Reception sensitivity threshold of node j
ro Normalization distance
Λ(i, j) Connectivity between nodes i and j.
Ro Radius of PER region
xix
ε Guard Region (GR) distance
Ψ2 Variance of shadowing when correlation is not considered
σ2ij Variance of correlated shadowing at link (i, j)
η Pathloss exponent
νik Speed of SU i at waypoint Cik
α Shadowing Correlation coefficient
S(i, j) Mutual distance between nodes i and j
PAj Activity of PU j
SAi Activity of SU i
ψ Set of Time slots
Λ(i, t) Connectivity of SU i at time slot t
γthr Threshold of maximum allowed interference
Ω Cost matrix
s(i, j, t) Normalized Euclidean distance between (i, j) at time slot t
EjIo Aggregated interference from all SUs on channel j
EijIo|S(i, j) Interference from SU i to PU j given the mutual distance
EjKLIo Co-channel interference from SUK to SUL on channel j
γintra Co-channel interference threshold
xx
Pfa Probability of false alarm
Pd Probability of detection
Q(.) Q function
Γ(a, b) Incomplete Gamma function
LoBi The % of battery level in SU i
LoBithr Threshold of minimum allowed LoBi in SU i
κi Local EH controlling parameter for SU i
κ Local EH controlling parameter for all SUs
z(t) Global EH controlling parameter
EjiPUI0 Interference introduced from PU j to SU i
σ2AWGN Variance of the Additive White Gaussian Noise
γconn thr Threshold of SUs connectivity
%,$, ξ,ϕ, ς The Lagrange multipliers
N Optimization round length
xxi
1
Chapter 1
Introduction
Even though spectrum is becoming increasingly scarce, spectrum occupancy rates are
very low [1, 2]. New technologies are currently being adopted to overcome wireless
spectrum shortages. Cognitive Radio (CR) has been proposed in order to leverage
spectrum utilization efficiencies and communication reliability through adaptation of
operating parameters, continuous learning and possessing awareness of surrounding
environments and activities [3].
Within the CR framework, Secondary Users (SUs) are allowed to opportunisti-
cally access the licensed spectrum of the Primary Users (PUs), provided that the
interference level is below an acceptable threshold. If the interference condition is
not satisfied, SUs must evacuate the channel immediately. To ensure that such cri-
teria are met, seamless schemes to dynamically access the spectrum are vital for CR
Networks (CRNs).
SUs can access the spectrum using one of the following schemes: interweave,
underlay, or overlay [4, 5], as depicted in Fig. 1.1. In an interweave scheme, SUs
are not allowed to cause any interference to the PUs and SUs can access the vacant
channels only. Thus, the CRN must keep an eye on the activity of the PUs and
2
immediately vacate the channel once a PU becomes active and move to another
available channel. The concept of moving between the available White Spaces (WSs)
is referred to as spectrum mobility [1].
Figure 1.1: Spectrum access schemes
In an underlay scheme, SUs are allowed to share the channel with an active PU
provided that SU interference levels do not exceed an acceptable threshold. The
maximum acceptable interference level is often referred to as interference temperature
[6]. This will give the CRN access to the spectrum at any time with the cost of
restricting the transmission power in a way that prevents harmful interference to the
PUs.
The overlay scheme allows SUs to simultaneously access the spectrum along with
the PUs provided that the CRN implements an appropriate coding technique to
mitigate the interference caused to the primary network [7]. In this case, the CRN
has to have knowledge about the code books or even messages belonging to the
primary network which may raise security concerns. Each of the three access schemes
1.1. MOTIVATION 3
have their benefits and costs.
1.1 Motivation
The motivation for this thesis arose from the necessity to efficiently utilize the un-
derutilized spectrum. CR research is mainly focused on the physical layer aspects
such as sensing, analyzing the activity of PUs, detecting the available WSs, and in-
terference mitigation [8]. Good progress in this direction has been made and many
techniques and approaches have been proposed in the literature. In order to enable
CR technology, however, other aspects should be addressed, other than the physical
layer issues, such as the issues of Medium Access Control (MAC) layer.
1.1.1 Conventional MAC protocols Vs. CR-MAC Protocols
Resource allocation problems for conventional networks have been thoroughly stud-
ied in the literature [9, 10]. The developed schemes and algorithms are not suitable
for CRNs, however, due to the existence of two different types of users. As such,
certain constraints on the interference levels should be applied while designing the
multiple access algorithms. Spectrum heterogeneity is the main factor that deter-
mines the CRN’s performance, level of protection to the primary network, and the
overall Spectral Efficiency (SE) gain. This heterogeneity imposes the need for new
MAC protocols designed specifically for CRN’s spectrally heterogeneous environment.
A comparison between the conventional MAC protocols and the CR-tailored MAC
protocols is provided in table 1.1.
1.1. MOTIVATION 4
Table 1.1: A comparison between conventional MAC protocols and CR-MAC proto-cols
Conventional MAC CR-MACMain design Efficient allocation of - Efficient utilization of spectrumchallenges and the available channels among SUs & PUs.requirements among the users - Protecting PUs from interference
- Adapt to spectrum mobility- Supervise fairness among SUs
Most of the algorithms so far assume either interweave or underlay access schemes.
This might not be the best approach, however. I use a hybrid interweave/underlay
access scheme throughout the thesis. Also, I compare the performance of the proposed
algorithm using a hybrid access scheme versus either underlay only or interweave only
access schemes. In addition, allocating a channel to more than one SU should be
1.3. RESEARCH OBJECTIVES AND CONTRIBUTIONS 14
designed carefully. To do so, I take into consideration the co-channel interference and
optimize spectrum and power allocations to minimize the interference between SUs
accessing the same channel simultaneously.
1.3.3 Supporting Mobility of SUs
SUs’ mobility changes networks’ topology as well as the interference between PUs and
SUs. Moreover, the connectivity between SUs plays a significant role when assigning
channels to multiple mobile SUs. In this thesis, I propose a DSA algorithm, called
Mobility-Supported DSA (MSDSA) that optimizes spectrum allocation when SUs are
mobile. The goal is to maximize the spectrum utilization and the access success-rate.
To achieve this goal, the MSDSA algorithm allocates the bands dynamically by joint
consideration of different factors such as mobility of the SUs, SUs’ connectivity, PERs,
the geographical locations of the nodes, shadow fading, and the activity of both PUs
and SUs.
1.3.4 Enabling Energy Harvesting in the Context of Dynamic Spectrum
Allocation
Enabling EH for CRNs is of value for extending their lifetime. Since SUs can either
access the spectrum or harvest energy, EH should be integrated within the context
of DSA, in order to increase both the energy efficiency and the spectrum efficiency. I
propose a novel algorithm that enables SUs to harvest energy with minimal impact
on their spectrum access performance. The algorithm allows SUs to participate in
making decisions regarding their operating mode. Moreover, the algorithm ensures
that the level of energy in the CRN cannot be lower than a specific threshold.
1.4. ORGANIZATION OF THE THESIS 15
1.3.5 Adaptive Power Allocation Algorithms
I propose two algorithms that optimize power allocation among SUs that were success-
ful in accessing the spectrum. The objective is to maximize the SE while respecting
the power budget constraints. I address the scenario in which CRN has multiple SUs
that are interfering with several PUs. Consequently, the power budget should be
allocated to SUs subject to different power constraints especially so that the hybrid
interweave/underlay access scheme can be adopted, which means that SUs can access
the active and non-active PU bands. Hence, different SUs will have different power
and interference limits depending on PUs’ activity and on the set of SUs that are
expected to access the same channel simultaneously. Moreover, since the complexity
of the optimal algorithms can be high, I propose a suboptimal Cap-Limited Heuris-
tic (CLH) algorithm. CLH algorithm considers assigning power to the SUs from a
discrete set of power levels, as will be discussed later.
1.4 Organization of the Thesis
The remainder of this thesis is organized as follows: Chapter 2 reviews CR technol-
ogy standardization’s efforts and a literature review of the proposed approaches in
the fields of spectrum allocation, power allocation, and EH for CRNs. Chapter 3
presents the system model and the proposed algorithm for DSA. In Chapter 4, I dis-
cuss supporting the mobility of SUs and study its impact on allocating the spectrum.
Chapter 5 presents the proposed approach to integrate EH in the context of DSA. In
Chapter 6, I propose a number of novel algorithms to optimize the power allocation
while satisfying the available power budget and the maximum transmission power of
SUs. Finally, the conclusion and directions for future work are given in Chapter 7.
16
Chapter 2
Literature Review
Several research and regulation efforts have been undertaken in the field of CRNs.
An overview of CR technology standardization’s efforts is presented in section 2.1
and a literature review of the proposed approaches for spectrum allocation, power
allocation, and EH in CRNs is provided in section 2.2.
2.1 Standardization of CR Technology
Over the past few years, there has been significant progress in the spectrum regulation
domain to address the growing demands of radio communication services. The first
CR-based standard, IEEE 802.22, is a centralized one where a CRBS is acting as
the control unit. Moreover, the CRBS might have information about the wireless
environment, users’ activities, and type of transmitted data. Benefiting from such
information in the resource allocation is expected to improve the coexistence between
the networks; hence, the algorithm efficiency will increase. There have been two
amendments of the standard: IEEE 802.22a and the IEEE 802.22b amendments [22].
The IEEE 802.22a amendment proposes standardization for management and control
interfaces and IEEE 802.22b amendment discusses supporting broadband services and
2.2. LITERATURE REVIEW 17
monitoring applications, such as supporting a large number of low energy units and
different QoS classes.
Another example of wireless standards for CR technology is IEEE 802.11.(af, ac,
n) series of standards which are proposed to support the traditional 802.11 Wireless
Local Area Network (WLAN) services [23,24]. Also, IEEE 1900.x series of standards
aim to define a standardized framework for radio resource management in future
wireless systems [25–27]. Another standardization effort is that of the European
Telecommunications Standards Institute (ETSI) to regulate the licensed shared access
among Long Term Evolution (LTE) operators in the band 2.3 GHz and 2.4 GHz [28].
Moreover, IEEE 802.19 standard is aiming to set the framework for coexistence of
several unlicensed systems such as 802.11af, 802.22, and 802.15.4 [29].
2.2 Literature Review
In the following subsections we provide a literature review of the proposed approaches
in the research areas of spectrum and power allocation, and EH.
2.2.1 Spectrum and Power Allocation
Autonomous spectrum allocation algorithms are proposed in [30–34], where the spec-
trum access is accomplished by achieving individual goals like the QoS requirements
or the energy consumption of a given SU. In [30, 31], the focus is on computing the
minimal SU’s transmission power that satisfies the individual SUs’ QoS goals. The
algorithm proposed in [32] employs Stackelberg’s game theory to calculate the opti-
mal resource allocation, while the algorithm proposed in [33] selects the SU pair with
the highest Signal to Noise Ratio (SNR) to utilize the lowest transmission power. [34]
2.2. LITERATURE REVIEW 18
presents a pricing-based non-cooperative game model for power control by SUs. The
objective is to provide throughput fairness among these users while guaranteeing a
minimum Signal to Interference and Noise Ratio (SINR) at the secondary receiver.
In [35], the authors propose a scalable MAC protocol for heterogeneous machine-
to-machine networks. The proposed protocol achieves hierarchical performance by
using different contending priorities and incorporates both the persistent Carrier
Sense Multiple Access (CSMA) and Time Division Multiple Access (TDMA) schemes.
Moreover, an incremental contention priority scheme is used to guarantee fair access
among multiple heterogeneous devices.
A DAB MAC protocol for CRNs (CMAC) is proposed in [36] where each SU is
equipped with a single transceiver to detect PUs’ activities in its vicinity and then
SU shares its sensing information with other SUs. CMAC divides the time frame
into two parts: the beacon period and the data transfer period. Each SU periodically
visits a common control channel to obtain information about PUs’ activities.
In [37], authors introduce a DAB Opportunistic Spectrum MAC (OS-MAC), in
which SUs that want to communicate with each other are grouped together to form
a cluster. Cluster heads are responsible for acquiring the traffic load information
of a channel and for propagating this information within their respective clusters.
OS-MAC uses a probabilistic channel selection scheme to reduce the inter-cluster
interference. However, interference caused by PUs, which is a key role of CR, has not
been implemented in OS-MAC.
In contrast to the autonomous non-cooperative techniques, which aim to optimize
the performance of a single SU through local decisions, the goal of cooperative spec-
trum sharing techniques is to maximize the entire CRN performance by introducing
2.2. LITERATURE REVIEW 19
cooperation among all active SUs. This is done by solving complex system-level op-
timization problems that commonly focus on the overall system performance [38].
In [39], the resource allocation problem is considered as a binary integer optimal
programming in a quasi-static spectrum environment.
In [40], a mixed integer nonlinear programming optimization model for spectrum
sharing is developed where the interference avoidance is considered. Another approach
is presented in [41], where an objective of maximizing the throughput of the whole
CRN is achieved by a suboptimal strategy for channel assignment.
Queuing theory is used as well to model the performance of single-channel CRNs.
For example, In [42], a simple network consisting of one SU and one PU is modeled
as an M/D/1 queuing model where both users share a single channel [43]. Another
scheme is introduced in [44], where accessing management is considered for a single
hop CR. The average transmission rate of a stationary SU, that is providing video
services, is studied using M/M/1 queuing model. In [45], the network with a single
PU and multiple SUs, sharing a single channel, is modeled with an M/G/1 queuing
model. The considered networks are simple and do not reflect real-life deployment
scenarios.
Also, a distributed QoS-Aware MAC (QC-MAC) protocol for multi-channel CRNs
is proposed in [46]. It deals with QoS-aware transmissions with the objective of
minimizing the PU-to-SU collision rate. In [47], the authors propose a distributed
PU Activity-Aware Distributed MAC (PAD-MAC) protocol for heterogeneous multi-
channel CRNs that selects the best channel for each SU to enhance its throughput.
PAD-MAC controls SUs’ activities by allowing them to exploit the licensed channels
2.2. LITERATURE REVIEW 20
only for the duration of estimated idle slots. However, the distributed MAC archi-
tecture still has a long path to go in order to be considered as a reliable solution for
a minimal-interference cognitive communication holding to its strict requirements.
Two schemes that are based on game theoretic approaches are discussed in [48,49].
In these models, SUs behave selfishly where each node will use all of its available
power to transmit data. Recent advances in employing bargaining game theoretic
approaches for DSA introduce the Nash bargaining approach as an efficient solution
[50]. The objective is to perform a joint channel and power allocation which maximizes
the SU throughput while taking into consideration the PU protection requirements.
The bargaining game allows SUs to reach a mutually beneficial agreement. However,
SUs have conflicts of interest; hence no agreement may be imposed on any individual
without its consent [51].
In [52], the authors propose a Quality-of-Experience (QoE)-driven channel alloca-
tion scheme for SUs and CRN base station. The historical QoE data under different
primary channels is collected by the SUs and delivered to the base station which
will then allocate the available channel resources to the SUs based on their QoE ex-
pectations and maintain a priority service queue. The modified ON/OFF models of
channels and service queue models of SUs are jointly investigated for this channel
allocation scheme.
An algorithm called Double hopping is proposed in [53] where a hopping pattern is
generated in a way that minimizes the time of using any channel. This will maximize
the number of channels used for a transmission and might also reduce the interference
to the PUs. This algorithm may lead to channel over-assignment, however, and the
system would not be able to guarantee a good QoS for SUs.
2.2. LITERATURE REVIEW 21
A biologically inspired spectrum sharing algorithm is proposed in [54]. It is based
on an insect colony’s adaptive task allocation model. A genetic algorithm is devel-
oped in [55] to jointly solve the channel allocation and power control problems. A
localized knowledge of the network is assumed which reduces the signaling overhead.
The algorithm needs a large number of iterations to converge, however.
An opportunistic Multi-Radio MAC (OMMAC) is proposed in [56], where a multichannel-
based packet scheduling algorithm is employed and packets are sent using the channel
with the highest achievable bit rate. Another algorithm that is based on Carrier Sense
Multiple Access with Collision Avoidance (CSMA/CA) is proposed in [57].
In [58], a spectrum access strategy for SUs is presented. This strategy is based on
the α-retry policy in queueing theory, where a pre-empted SU joins the transmission
queue with probability α for retrial. A two-dimensional discrete-time Markov chain
model is used to analyze performance of the proposed channel access strategy.
Employing an overlay access scheme, in which the SU transmitter employs part of
its resources to help the communication between the PUs, to enable SUs to access the
spectrum is proposed in [59–61]. Using Orthogonal Frequency Division Multiplexing
(OFDM) as the transmission technology for CRNs is studied in [62, 63] where the
resource allocation process is formulated as an optimization problem that looks for
the optimal power and subcarrier allocation. Employing an underlay access scheme
is addressed in [64,65]. The focus in [64] is on the routing process in ad-hoc underlay
based CRNs, while the authors of [65] study using a multi-antenna scenario to increase
the spatial diversity, where the resource allocation process is defined in such a way
that it maximizes SUs’ performance and diminishes their impact on the PUs.
2.2. LITERATURE REVIEW 22
In [66], a spectrum scheduling scheme is proposed for mobile CRN by formulat-
ing a throughput maximization problem and solving it using a bipartite graph. The
mobility of the nodes is assumed to be deterministic and an interweave access scheme
is employed. A one-dimensional mobility model is studied in [67] where the SUs
are assumed to be moving only in the x-direction of the cartesian coordinates. The
studied network consists of one PU and four SUs where SUs can access the channel
using an interweave access scheme. In [68], a handoff management scheme is proposed
for mobile SUs. It allows SUs that are moving according to a linear mobility model
to switch from one primary network to another. This scheme adopts a multi-agent
based solution that uses a trading and pricing system between SUs and the primary
networks. Also, in [69], a mobility management scheme for multi-cell CRN is pro-
posed. SUs are allowed to access the spectrum using an interweave scheme under this
management scheme.
2.2.2 Energy Harvesting
As with DSA algorithms, in order to enable CRNs to harvest energy, the dynamistic
nature of the available EH opportunities in CRNs should be taken into considera-
tion. EH has been studied and reported in the literature for regular Wireless Sensor
Networks (WSNs). For instance, a game-theoretic sleep and wake-up strategy [70],
queuing theoretic transmission policies [71], and modified back-pressure-based algo-
rithms with energy queues [72] are proposed to model a WSN that consists of one
EH transmitter and one receiver.
Some studies that investigate enabling EH in CRNs have been conducted as well.
For example, the authors of [73] analyze the achievable throughput of a SU, which
2.2. LITERATURE REVIEW 23
harvests energy from ambient sources while opportunistically accessing the spectrum.
The primary traffic is modeled as a time-homogeneous discrete Markov process.
In [74], the performance of a SU with EH capability is studied, where the goal
is to determine an optimal spectrum sensing policy that maximizes the expected to-
tal throughput subject to an energy causality constraint and a collision constraint.
The energy causality constraint keeps the consumed energy lower than the harvested
energy, and the collision constraint mandates that the probability of accessing the
spectrum while a PU is active is equal to or less than a predefined maximum proba-
bility of collision.
In [75], the maximum stable throughput region for a simple CRN, with one SU
pair and one PU pair, is analyzed. The SU transmitter harvests ambient energy while
the PU transmitter is assumed to be plugged into a reliable power supply.
Authors of [76] analyse the optimal random access for a SU with EH capabilities.
The access probabilities are obtained under the constraints of primary queue stability
and primary queueing delay being kept below a specified value.
In [77], an EH and information transfer protocol in a cognitive two-way relay
network is developed. In this protocol, a SU harvests energy from a neighbouring
PU while assisting the primary’s transmission in an overlay setting. Specifically, the
network is assumed to have two PUs that exchange information through a SU which
will first harvest energy from these two PUs and then use the harvested energy to
forward the remaining primary signals along with the secondary signals in the second
part of the operation cycle.
The work in [78] proposes a probabilistic access strategy by a SU based on the
number of energy units at its energy queue. The system is assumed to have one SU
2.2. LITERATURE REVIEW 24
and one PU. The authors investigate the effect of the energy arrival rate, and the
capacity of the energy queue on the SU’s performance.
25
Chapter 3
Efficient Spectrum Allocation Schemes for
Cognitive Radio Networks
CRNs have the ability to reconfigure and adapt their software components and archi-
tectures, thus enabling flexible delivery of broad services, as well as sustaining robust
operation under highly dynamic conditions [5]. The geometry of the SUs and PUs
has an effect on the performance of the spectrum sharing algorithms. Since PUs have
the right of claiming the spectrum whenever needed, it is crucial to ensure that PUs’
satisfaction and QoS are not impacted by SUs’ activity on the shared bands.
A review of the literature shows that several factors have not been comprehensively
studied. For instance, the impact of the correlated shadow fading maps on the DSA
algorithm was not investigated. Also, fairness between the SUs is an important issue
that needs to be addressed during the design of the DSA algorithm. Moreover, the
practical implementation of complex algorithms should be considered.
In order to increase the utilization of the spectrum bands, I propose a DSA al-
gorithm that integrates both interweave and underlay spectrum access schemes. The
proposed algorithm will jointly take into account the geographical locations of the
26
nodes, the shadow fading, the interference between the primary and the secondary
networks, the co-channel interference between SUs that are transmitting on the same
channel, and the communications activity of the users. Moreover, a heuristic DSA
algorithm that jointly takes into consideration all of the aforementioned issues, while
requiring lower computational- and time-costs, is developed.
The contributions in this Chapter are illustrated in the following steps. First, the
problem of dynamic spectrum sharing is formulated as a binary integer optimization
problem. Second, due to the complexity of this optimization problem, a PAH-DSA
algorithm is developed. Third, the surrounding environment is taken into considera-
tion by employing the model of a correlated shadow fading map. Moreover, fairness
between SUs is considered where any SU cannot access more than one channel at any
given time slot. Furthermore, an extra measure of protecting PUs’ QoS is enforced
by not allowing SUs that are close to a specific active PU to access the channel that
is assigned to this PU. Also, multiuser access to the same channel is enabled provided
that the interference between the SUs is acceptable. Lastly, to increase the chances
of successful access to the spectrum, a hybrid interweave/underlay access scheme is
adopted. To the best of my knowledge, a study of these issues jointly has not been
reported in the literature previously.
The remainder of this Chapter is organized as follows: In the next section, I de-
scribe the network model. The communication traffic model is described in Section
3.2, and the correlated shadow fading maps are explained in Section 3.3. The extra
measure in protecting PUs’ communication is provided in Section 3.4. The formula-
tion of the Optimization problem is given in Section 3.5. The proposed algorithm is
presented in 3.6. Simulation results and interpretations are presented in Section 3.7.
3.1. NETWORK MODEL 27
Finally, the Chapter’s concluding remarks are given in Section 3.8.
3.1 Network Model
The CRN is assumed to have M SUs and one CRBS which is assumed to be aware of
PUs’ activity. This is made possible by employing some sensing techniques [79]. The
CRBS is located in the middle of the operating area and its main duty is to coordinate
the assignment of the channels and optimize the spectrum allocation. Both the CRN
and the primary network are located in close proximity and the topology of each of
the networks follows a homogeneous Poisson Point Process (PPP) with nodes’ den-
sity of θPU and θSU for the primary and secondary networks, respectively [80]. The
transmission power of SUs and PUs are referred to as PSU and PPU respectively. The
system is assumed to have Υ PUs where each PU is assigned one channel with a band-
width B. Hence, the number of channels is equal to the number of PUs. The CRN
physical layer is assumed to be Filter Bank Multi-Carrier (FBMC). The interference
from other PUs on the channels that they are not transmitting on is considered neg-
ligible. This can be justified by the fact that FBMC has very small sidelobe which
significantly reduces the interference [81]. Also, Channel State Information (CSI)
and sensing results are assumed to be sent from SUs to the CRBS using a dedicated
Control Channel (CC) that is not affected by the activity of PUs.
3.2 Traffic Flow Model
In order to increase the efficiency of the DSA algorithm and reduce the collision rate
between SUs, the operating time is divided into time slots where SUs are allowed to
begin transmitting only at the start of any time slot. The communications traffic flows
3.2. TRAFFIC FLOW MODEL 28
of the PUs and SUs are modelled as Bernoulli arrival processes with parameters λPU
and λSU , respectively [82]. Bernoulli process is the discrete-time analog of the Poisson
arrival process where the arrivals of the packets can take place at some time slot k.
Mathematically, the traffic of a SU can be described as a point process consisting of
a sequence of arrival instants T1, T2, ..., Tm, ... measured from the origin To = 0.
The number of arrivals for slot k follows a binomial distribution:
PNk = n =
(k
n
)λnSU(1− λSU)k−n (3.1)
Also, the number of time slots between two arrivals is geometrically distributed with
parameter λSU and the probability of having O packets arriving at the same timeslot
is given by:
PAn = O = λSU(1− λSU)O , O ∈ N0 (3.2)
Under this model, communication sessions are allowed to begin at the start of any
time slot and if a SU generates a packet before the start of a time slot, this packet has
to be stored in a local buffer and the SU would wait till the next time slot to contend
for a channel. The SU then will initiate a communication session if it is granted
access. It is important to note that the change in the PUs’ communication behaviour
does not happen frequently and can be assumed to be fixed for a minimum duration
of one time slot. Hence, the sensing information is valid for at least one time slot.
Fig. 3.1 shows the activity of a node that has traffic intensity of 0.5, which means
that this node will be producing packets with a probability of 0.5 at each time-slot.
This node is producing one packet at time slots 1, 3, 5 and 8, and two packets at time
slot 2.
3.3. CORRELATED SHADOW FADING MAP 29
Figure 3.1: Packet arrival from a node that has traffic intensity of 0.5
3.3 Correlated Shadow Fading Map
When the transmitter and the receiver are located in free space with a Line of Sight
(LOS) link, we can use (3.3) to calculate the level of the received power as follows:
Pr = PtGtGr
(λ
4πS(i,j)
)2
(3.3)
where Pt is the transmitted power, Gt is the gain of the transmitting antenna, Gr
is the gain of the receiving antenna, λ is the wavelength of the transmitted signal,
and S(i, j) is the Euclidean distance between the transmitter i and the receiver j.
The strength of the received signal depends also on the surrounding environment and
its topography, however. This can be modeled by shadow fading [83]. The standard
3.3. CORRELATED SHADOW FADING MAP 30
deviation of the shadow fading (σ) is in the range of 3 to 12 dB [84]. Shadow fading
affects the connectivity of the nodes where some nodes that are far away from each
other might be connected while some neighboring nodes might be disconnected due
to the level of fading between them. The impact of shadow fading on the connectivity
of the network is shown in Fig. 3.2.
Figure 3.2: Impact of shadow fading on nodes’ connectivity
The transmission range of nodes is considered as a stochastic process and in or-
der to find the probability of connectivity between the pair of nodes i and j let us
define the event of having a direct communication link between them as Λ(i, j). The
conditional probability of having a link given the Euclidian distance is defined as [85]:
P (Λ(i, j)|S(i, j)) = P (β(i, j) ≤ βthr|S(i, j))
= Q
(10η
σlog10
S(i, j)
r0
) (3.4)
where η is the pathloss exponent and σ is the standard deviation of the shadow
3.3. CORRELATED SHADOW FADING MAP 31
fading. Q(.) is the Q-function defined as:
Q(x) =1√2
∫ ∞x
exp
(−u
2
2
)du (3.5)
βthr is the attenuation threshold that considers the reception sensitivity of the
nodes. When the attenuation, experienced by any channel between (i, j), is less than
βthr, a direct communication will be available through this channel. βthr is defined
as:
βthr = 10 log10
P it
P jr,thr
(3.6)
P it is the transmission power of node i and P j
r,thr is the reception sensitivity of node
j (i.e. if the received power at node j from node i is larger than P jr,thr, then a
direct communication is available through this channel. ro is a normalization term
which represents the maximum range that a node can reach directly under the purely
geometric link model (as shown in Fig. 3.2(a)) and is defined as:
ro = 10βthr10η (3.7)
Figure 3.3 shows the link probability over s/ro for η = 2 and 4 with different
values of σ. For instance, when η = 4 and σ = 6dB, there is still a link probability
of more than 12% at a distance S = 1.5ro.
To accurately model the statistical nature of the channels, the correlation between
the shadowing that is affecting different links should be considered due to the fact that
the links within the same vicinity are impacted by the same nearby large objects. Also,
the correlation of shadow fading is of high importance when studying CRNs due to
their coexistence with the primary network [84]. As such, the proposed DSA algorithm
3.3. CORRELATED SHADOW FADING MAP 32
Figure 3.3: Effect of fading and pathloss exponent on nodes connectivity
takes the correlation and variance of the shadowing into account when assigning
channels to SUs in such a way that guarantees a minimal level of shadowing to be
introduced to SUs, and the correlated shadow fading can be modeled as an exponential
correlation model [86]. The normalized correlation function can be written as [87]:
r(x) = e−αx, x ≥ 0 (3.8)
where α is the correlation coefficient. The variance of the correlated shadow fading
for link (i, j) can be calculated using:
σ2ij = Ψ2
[1 +
1
‖d(i, j)‖exp(−‖d(i, j)‖α)− 1
‖d(i, j)‖
](3.9)
where Ψ2 is the variance of the shadow fading when the correlation is not taken into
consideration, and ‖d(i, j)‖ is the distance between nodes i and j. For an intermediate
environment between suburban and urban areas, a value of α = 1/20 was suggested
in [88]. Fig. 3.4 shows the correlated shadow fading versus the distance between the
3.3. CORRELATED SHADOW FADING MAP 33
transmitter and the receiver for different values of α. As shown, σ2ij will change slowly
when the value of α is relatively low, and vice versa.
0 20 40 60 80 1003.6
3.7
3.8
3.9
4
||dij||
σ ij2
α=0.025α=0.05α=0.67α=0.1
Figure 3.4: Correlated shadow fading versus dij for different values of α
Fig. 3.5 shows an example of both of the networks being under the effect of a
correlated shadow fading map. The density of the nodes is set to 10 for each of the
networks. The connectivity between the nodes is shown in the figure as well. Each
color represents a different shadowing level and the contour lines show the correlation
in the standard deviation of shadow fading within the simulation region.
3.4. PROTECTING PUS’ COMMUNICATIONS 34
Figure 3.5: Network topology under the correlated shadow fading model
3.4 Protecting PUs’ Communications
In order to protect the PUs from harmful interference, while facilitating the hybrid
access scheme, the concept of PER is used, where SUs are not allowed to transmit on
a band that is allocated for a PU if they are located inside the PER of this PU [89].
As such, the network geometry is employed to protect the communication of PUs.
Let the radius of the PER be Ro, as shown in Fig. 3.6. All SUs that are transmit-
ting on the same channel of a PU must be at least ε Guard Region (GR) distance away
from any primary receiver. Practically, the exact location of the primary receivers is
unknown to the SUs. Hence, as shown in Fig. 3.6, SUs that lie inside a circle centered
at the primary transmitter with a radius of (Ro + ε) cannot transmit on the same
3.4. PROTECTING PUS’ COMMUNICATIONS 35
Figure 3.6: Employing the geometry of the networks to protect PUs’ QoS
channel as the nearby PU. While this can be considered to have a negative impact
on the performance of the DSA algorithm, it can be viewed as a great opportunity
to increase the lifetime of the CRN by directing the affected SUs to harvest energy
from the nearby active PUs, instead of contending to access the spectrum, as will be
explained in chapter 5.
Practically speaking, the impact of SUs on PUs’ communications is captured by
the expected amount of interference introduced to the PUs due to the SUs’ activity.
Let Ijo be the aggregated interference power from all SUs to PU j. Assuming that all
SUs are located at the border of PU j PER region, the expected interference power
experienced by PU j is given by [90]:
3.4. PROTECTING PUS’ COMMUNICATIONS 36
EjIo = Gjo(Ro + ε)−η
M∑i=1
exp
(1
2
(σij
log 10
10
)2)P iSU (3.10)
where Gjo = (c/4πfj)
2, fj is the centre frequency of the channel under investiga-
tion, c is the speed of light, P iSU is the transmission power of SU i, η is the pathloss
exponent, and σij is the standard deviation of the correlated shadowing at link (i, j).
As expected, when the radius of the PER and ε increases, the aggregated interference
will decrease. A more realistic scenario is to consider the actual locations of the SUs
when calculating the aggregated interference. This can be done as follows:
EjIo =M∑i=1
GjoS(i, j)−η exp
(1
2
(σij
log 10
10
)2)P iSU
, s.t. S(i, j) ≥ Ro + ε, ∀j ∈ Υ
(3.11)
where S(i, j) is the mutual Euclidean distance between SU i and PU j. Moreover,
the activity of SUs affects the value of EjIo since all of the SUs will neither be active
at a specific time slot nor will they be assigned the same channel. Let xij(t) ∈ 0, 1
be a binary integer variable where 1 means that PU j’s channel is allocated to SU i
to transmit at time slot t, and 0 means that channel j is not allocated to SU i. The
actual aggregated interference AEjIo is calculated as follows:
AEjIo =M∑i=1
xij(t)GjoS(i, j)−η exp
(1
2
(σij
log 10
10
)2)P iSU ,
s.t. S(i, j) ≥ Ro + ε, ∀j ∈ Υ
(3.12)
3.5. PROBLEM FORMULATION 37
3.5 Problem Formulation
In order to satisfy the protection needs of the PUs, let us introduce the maximum
interference threshold as following:
AEjIo < γthr,∀j ∈ Υ (3.13)
where γthr is the maximum allowed interference level to be introduced to the PUs
from SUs. Equation (3.13) ensures that PUs are experiencing an interference that is
less than the allowed level all the time. This constraint holds for all of the PUs. The
optimization problem can then be formulated as follows:
Minimize
M∑i=1
xij(t)GjoS(i, j)−η exp
(1
2
(σij
log 10
10
)2)P iSU ,∀j ∈ Υ (3.14)
such that:
S(i, j) ≥ (Ro + ε), ∀ pairs (i, j) (3.15)
AEjIo < γthr, ∀j ∈ Υ (3.16)
1 ≤M∑i=1
xij + PAj, ∀ i ∈M, j ∈ Υ (3.17)
Υ∑j=1
xij ≤ 1, ∀ i ∈M (3.18)
M∑K=1
EjKLIo < γintra, ∀L ∈M,K 6= L,∀j ∈ Υ (3.19)
where PAj is the activity of PU j. Condition (3.15) prevents allocating channel j
3.6. THE PROPOSED PAH-DSA ALGORITHM 38
to any SU that is located in the vicinity of PU j. Also, condition (3.16) guarantees
that the level of the interference is less than the acceptable threshold. Condition
(3.17) assures that any channel is assigned to as many users as possible and that the
channels are allocated to at least one active user. Moreover, condition (3.18) restricts
the maximum number of channels that any SU can access at any time slot to one.
Finally, condition (3.19) ensures that the interference between SUs K and L that
are assigned channel j simultaneously, EjKLIo , does not exceed a prescribed limit,
γintra. EjKLIo is defined as follows:
EjKLIo = Gj
oS(K,L)−η exp
(1
2
(σKL
log 10
10
)2)PSU (3.20)
3.6 The Proposed PAH-DSA Algorithm
The problem formulated in the previous section is a Binary Integer Non-Linear Pro-
gramming (BINLP) which is difficult to solve. Alternatively, the binary optimization
variable can be relaxed and then the new problem can be solved using the primal-dual
interior-point method. Such an approach still requires huge computational resources
to find the solution for six sets of Lagrangian multipliers, however.
To overcome this obstacle, I propose a PAH-DSA algorithm that provides a sat-
isfactory protection level to the PUs while quickly assigning the available bands to
the SUs. The algorithm will employ both interweave and underlay access schemes.
First, let us define the expected interference caused by SU i to PU j given the mutual
Euclidean distance between them as follows:
EijIo|S(i, j) = GjoS(i, j)−η exp
(1
2
(σij
log 10
10
)2)P iSU (3.21)
3.6. THE PROPOSED PAH-DSA ALGORITHM 39
Next, let Ω be the cost function where Ωij is the cost of assigning channel j to
SU i. Ωij is defined as follows:
Ωij =(LN |S(i, j) < Ro + ε)(√
2− s(i, j))EijIo|S(i, j)
+ 0.6PAj + (SAi + 1)−1 ∀ i, j(3.22)
where LN is any large number greater than 1, s(i, j) is the normalized Euclid-
ian distance between SU i and PU j, and SAi is the activity of SU i. The term
(LN |S(i, j) < Ro + ε) increases the cost of assigning channel j to SU i considerably,
if it is close to PU j. Also, the term (√
2− s(i, j)) ensures maximizing the Euclidian
distance between SUs and PUs that will be transmitting on the same channel. For
example, when s(i, j) has a high value, this means that PU j and SU i are far away
from each other, and hence the cost will be reduced. On the other hand, as s(i, j)
decreases, the cost will increase. The second term in (3.22), 0.6 PAj, ensures use of
an interweave scheme first by adding an extra cost for invoking the underlay scheme.
The last term represents the activity of SUs; if SU i is very active, it will get a higher
priority to access the spectrum. The flow chart of the proposed algorithm is shown
in Fig. 3.7.
The mechanism of the PAH-DSA algorithm is described as follows: the algorithm
will start by checking the needs of all SUs for spectrum. Once the CRBS receives this
information, it will calculate the cost function for all of the SUs. Next, the CRBS
will define two new sets of channels, Υ1 =set of the channels that do not have active
PUs and Υ2 =set of the channels that have active PUs. After that, the channels
that are in Υ1 will be allocated to the SUs that have the lowest cost and introduce
the least amount of noise to the system. Then the algorithm will check if the demand
3.6. THE PROPOSED PAH-DSA ALGORITHM 40
Figure 3.7: Flowchart of the proposed PAH-DSA algorithm
3.7. RESULTS AND INTERPRETATION 41
from all SUs is satisfied. If not, the underlay access scheme will be invoked to assign
channels in Υ2 to the demanding SUs provided that the interference level is less than
the threshold and SUs’ locations are outside the PER and GR of the active PUs.
A maximum of one channel is assigned to any SU at a given time slot. This will
guarantee a level of fairness between SUs to some extent, where the demanding SUs
will not be affecting the requests from other SUs. Finally, when the number of SUs
is large as compared to the number of available channels, the algorithm will double
check whether the conditions of PER, interference introduced to the primary network,
and co-channel interference conditions are met or not. If these conditions are met,
then the algorithm will declare the allocation results and exit. If not, SUs that do not
fulfill any of these conditions will be prohibited from transmission. However, when
SUs number is comparable to the number of channels, the double check step is not
necessary since the algorithm will efferently allocate the channels to the SUs that are
faraway from the PUs.
The computational complexity of the PAH-DSA algorithm is lower than or equal
to O(max(MΥ1,MΥ2)) + O(logM), which is much lower than the complexity of
finding the optimal solution (O((MΥ)3 )).
3.7 Results and Interpretation
The performance of the proposed algorithm is evaluated using the probability of suc-
cess (Prsuccess) parameter, which is the probability that SUs with data to transmit
succeed in getting channels assigned to them. The correlation of the shadow fading
map is set to 1/20 and the pathloss exponent (η) is set to 4. Also, the commu-
nication activity of the PUs (λPU) is set to 0.5. The performance of the proposed
3.7. RESULTS AND INTERPRETATION 42
algorithm is compared to two different algorithms: a CSMA/CA-based algorithm and
an interweave-only algorithm.
Fig. 3.8 shows the effect of SUs’ communication traffic (λSU) on Prsuccess
when the number of available channels is 30. The PAH-DSA algorithm always out-
performs the CSMA/CA-based algorithm for all values of λSU . When PER is taken
into consideration, the performance will slightly degrade, especially when λSU > 0.3
for θSU = 50 and when λSU > 0.7 for θSU = 30. This is due to the increased demand
from SUs, which in return leads to assigning channels to some SUs even if they are
located near the active PUs. Moreover, the PAH-DSA algorithm outperforms the
interweave-only algorithm when λSU is less than 0.5. When λSU is larger than 0.5,
the interweave-only algorithm will be a better option. This is due to the fact that
the PAH-DSA algorithm puts the protection of PUs’ communications first regardless
of how this may affect SUs’ QoS due to rejecting their access to the spectrum.
A plot of Prsuccess versus the θSU is given in Fig. 3.9. The PAH-DSA algo-
rithm outperforms both CSMA/CA-based and interweave-only algorithms when λSU
is small. However, the interweave-only algorithm outperforms the PAH-DSA algo-
rithm when SUs are acting on saturated communication mode (λSU = 1) and the
θSU is larger than 30. This is due to the fact that PAH-DSA algorithm puts the
protection of PUs’ communications as the first and foremost priority that comes even
before satisfying the needs of SUs to access the spectrum.
The impact of the number of available channels on Prsuccess is shown in Fig.
3.10. As expected, when the number of channels increases, the performance of all of
the three algorithms improves. Also, the PAH-DSA algorithm outperforms both of
the other algorithms and can reach a success rate of 95% in cases where the number
3.7. RESULTS AND INTERPRETATION 43
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
λSU
Pr
succ
ess
PAH, No PER, θSU
= 30
PAH, with PER, θSU
= 30
Interweave only, θSU
= 30
CSMA/CA, No PER, θSU
= 30
CSMA/CA, with PER, θSU
= 30
PAH, No PER, θSU
= 50
PAH, with PER, θSU
= 50
Interweave only, θSU
= 50
CSMA/CA, No PER, θSU
= 50
CSMA/CA, with PER, θSU
= 50
Figure 3.8: Success probability vs. λSU for different θSU values. Number of channels= 30
of channels is greater than 30 and θSU is equal to 50.
Fig. 3.11 studies the impact of (Ro+ε) radius on the performance of the algorithms
when the θSU = 30 and the number of channels = 50. In the case of a small λSU ,
increasing Ro + ε radius does not affect the performance of the PAH-DSA algorithm
3.7. RESULTS AND INTERPRETATION 44
10 15 20 25 30 35 40 45 500.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
θSU
Pr
succ
ess
PAH, No PER, λSU
= 0.3
PAH, with PER, λSU
=0.3
Interweave only, λSU
=0.3
CSMA/CA, No PER, λSU
=0.3
CSMA/CA, with PER, λSU
=0.3
PAH, No PER, Saturated modePAH, with PER, Saturated modeInterweave only, Saturated modeCSMA/CA, No PER, Saturated modeCSMA/CA, with PER, Saturated mode
Figure 3.9: Success probability vs. θSU for different λSU values. Number of channels= 30
since the channels are already assigned to the furthest SU from the active PU. When
λSU = 1 and Ro + ε = 100 m, however, the performance will degrade by 7% , as
compared to when Ro + ε ≤ 70 m, due to the increase in the SUs’ demand to access
the spectrum, which will lead to an increase in the failure of assigning appropriate
3.7. RESULTS AND INTERPRETATION 45
channels to a portion of the SUs. For the CSMA/CA-based algorithm, as (Ro + ε)
radius increases, the success rate will drop dramatically.
10 20 30 40 50 60 700
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Number of Channels
Pr
succ
ess
PAH, No PER, λSU
= 0.3
PAH, with PER, λSU
=0.3
Interweave only, λSU
=0.3
CSMA/CA, No PER, λSU
=0.3
CSMA/CA, with PER, λSU
=0.3
PAH, No PER, Saturated modePAH, with PER, Saturated modeInterweave only, Saturated modeCSMA/CA, No PER, Saturated modeCSMA/CA, with PER, Saturated mode
Figure 3.10: Success probability vs. number of available channels for different λSUvalues. θSU = 50
3.8. CONCLUDING REMARKS 46
10 20 30 40 50 60 70 80 90 1000
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
PER+ε (m)
Pr
succ
ess
PAH, No PER, λSU
= 0.3
PAH, with PER, λSU
=0.3
Interweave only, λSU
=0.3
CSMA/CA, No PER, λSU
=0.3
CSMA/CA, with PER, λSU
=0.3
PAH, No PER, Saturated modePAH, with PER, Saturated modeInterweave only, Saturated modeCSMA/CA, No PER, Saturated modeCSMA/CA, with PER, Saturated mode
Figure 3.11: Success probability vs. PER+ε for different λSU values. θSU = 30,Number of channels = 50
3.8 Concluding Remarks
In this Chapter, a PAH-DSA algorithm was proposed to dynamically allocate spec-
trum in CRNs. The proposed algorithm employs a hybrid interweave/underlay ap-
proach and supports the mobility of SUs. The results demonstrate a high success
level in fulfilling CRNs’ demands for spectrum access while protecting the QoS of the
3.8. CONCLUDING REMARKS 47
primary network. Simulation results have also shown that the proposed algorithm
outperforms traditional spectrum allocation algorithms.
48
Chapter 4
Mobility-Supported Dynamic Spectrum Allocation
for Cognitive Radio Networks
4.1 Introduction
Since nodes’ connectivity is coupled with their locations and the Euclidean distances,
it will be dynamically changing when nodes are moving. Moreover, mobility will
change the network topology and hence will cause losses of some links between nodes.
In addition, it can change the geographical distribution of the traffic and the interfer-
ence pattern will be changing accordingly. Taking mobility into consideration while
designing DSA algorithms is required for an effective design. A Mobility-Supported
Dynamic Spectrum Allocation (MSDSA) is essential for enabling the future mobile
CRNs.
The schemes proposed in the literature are mainly focused on the spectrum al-
location without jointly considering several important factors such as a multiuser
multichannel scenario, the geographical locations of the nodes, the correlated shadow
fading maps, the PER, and the hybrid allocation scheme. Moreover, only a few algo-
rithms took users’ mobility into consideration. The impact of mobility on DSA and
4.2. MOBILITY MODEL 49
how to mitigate this effect was not addressed thoroughly, however.
The contributions in this chapter are as follows:
• I propose a DSA algorithm that is designed to support mobility by considering
SUs’ behaviour during the design phase.
• The algorithm employs a PER region concept to provide an extra measure
of protection for PUs’ communications. It also jointly considers users’ activ-
ity, correlation in shadow fading, and the geographical locations of the nodes.
Moreover, a hybrid access scheme is adopted to increase the success rate, as
described in the previous chapter.
• The proposed algorithm considers a multichannel multiuser scenario where the
system will have multiple channels and more than one SU will be allowed to
access the same channel simultaneously.
• A suboptimal MSDSA algorithm is proposed in order to reduce the computa-
tional costs of solving the optimization problem.
The remainder of this chapter is organized as follows: Section 4.2 explains the
mobility Model. The problem formulation is given in Section 4.3, while the subopti-
mal MSDSA algorithm is explained in Section 4.4. Extensive simulation results and
interpretations are given in Section 4.5. Conclusions to the chapter appear in Section
4.6.
4.2 Mobility Model
Modeling the mobility can be performed by extracting a pattern from mobility traces
[91]. SUs are moving according to a Random Waypoint Model (RWPM). The RWPM
4.2. MOBILITY MODEL 50
is very useful in evaluating the performance of mobile wireless networks. It is a random
model for the movement of mobile users where it considers the changes over time in
the nodes’ locations, velocities and accelerations [92]. This model is flexible for a wide
range of wireless networks and can be used in simulating the movement of SUs [93].
Fig. 4.1 illustrates the mobility of a SU following the RWPM. The mobility of this
SU is described as follows:
1. The node starts from an initial location Ci0 = (x0, y0), and a variable H is set
to 1.
2. A destination location CiH = (xH , yH) is chosen, using normal distribution,
within the network area.
3. The movement speed, νH , is chosen uniformly from an interval [υmin, υmax].
After that, the node moves along the line segment between the two locations
toward CiH at speed νH .
4. Node i will stay at waypoint CiH for T ip,H time.
5. Set H = H + 1, then go to step 2.
Figure 4.1: Mobility of a SU following the random Waypoint mobility model
The impact of mobility on the connectivity of the nodes is derived in appendix A.
The probability of connectivity of SU i to the network can be obtained as follows:
The optimal allocation can be found by solving the minimization problem of the
Lagrangian using Newton’s method to sequence of equality constrained problems.
The more Lagrangian multipliers the problem has, however, the more time and
computational resources the problem will be consuming [96, Chapter 11]. In order to
provide a low complexity algorithm that can be implemented practically, I propose
the MSDSA suboptimal algorithm. Suboptimal MSDSA will eliminate the number
of required Lagrangian multipliers by carrying out the steps explained next.
4.4 Description of MSDSA Algorithm
The Pseudo code for the algorithm is given in Algorithm 1. The MSDSA algorithm
starts with ensuring that SUs have the capability of transmitting their data to the
intended receiving station, i.e. being connected to the network. As such, the MSDSA
4.4. DESCRIPTION OF MSDSA ALGORITHM 58
algorithm will start by checking the connectivity of each SU prior to performing the
optimization of the resources. Let M ′ = SU1, SU2, ..., SUm′ represent the set of the
connected SUs. Next, the cost matrix Ω(t) is modified accordingly, by omitting the
data relevant to the disconnected nodes (in order to prevent considering the discon-
nected SUs as contending users). This will introduce two advantages to the algorithm:
First, the size of the problem is reduced, and second the constraint defined in (4.10)
and the accompanying Lagrangian multipliers (ϕ) will be removed.
The next step is to give priority to the interweave allocation scheme by enforcing
the assignment of the vacant channels before invoking the underlay scheme, without
needing to consider condition (4.8) in the optimization problem. This is done by
modifying the definition of Ω′ij to include the PU activity as follows:
Ω′
ij =0.6(PAj(t)) + (√
2− s(i, j, t))EijIo|PER
+ (SAi(t) + 1)−1 ∀i ∈M ′, ∀ j ∈ Υ, ∀t ∈ ψ
(4.23)
where j represents all of the available channels and i iterates over all of the con-
nected SUs. Another advantage for defining Ω this way is that the cost will be
dependent on the mutual Euclidian distance. Hence, the channels will be assigned
to the SUs that are far away from the PER. Based on this observation, condition
(4.20) and its Lagrangian multipliers will be omitted from the Lagrangian problem.
To guarantee that the omission of this condition will not impact the PUs’ QoS, the
CRBS will double check the total interferences introduced to the PUs after perform-
ing the DSA. If the level of the interference is larger than the threshold, then the SUs
will not be allowed to transmit simultaneously on that timeslot. Based on the above
4.4. DESCRIPTION OF MSDSA ALGORITHM 59
simplification, the modified Lagrangian is given in (4.24).
G(%,$) =M∑i=1
Υ∑j=1
Ωijx∗ij +
M∑i=1
Υ∑j=1
%i(log(−x∗ij(t) + 1))
−M∑i=1
Υ∑j=1
$j
(x∗ij(t) + PAj(t)− 1
) (4.24)
This will leave only two sets of Lagrangian multipliers: % and $ . Moreover,
since the optimization problem has to be solved in real time, a new parameter is
introduced to reflect the frequency of updating the optimization parameters. Let N
be a real positive integer and call it the round length. N defines the time spacing
between the consecutive runs of the allocation algorithm. For instance, if N = 100
then this means that the algorithm will solve for the optimal allocation once every 100
timeslots. To keep the focus of the algorithm on protecting the PUs’ communications,
however, SUs that were successful in getting access to the spectrum will be obliged
to frequently check their compliance with the PER-transmission free region and the
maximum allowed interference level. At this stage, the primal-dual interior-point
method can be used to solve the simplified optimization problem.
The computational complexity of the MSDSA algorithm is lower than or equal to
O((2M ′Υ/N)3)+O(log N). On the other hand, the complexity of solving the original
optimization problem is (O(5MΥ)3 ) where M could potentially be much higher than
M ′. As previously stated, the complexity of the proposed solution is lower than the
optimal solution.
4.4. DESCRIPTION OF MSDSA ALGORITHM 60
Algorithm 1 Suboptimal MSDSA Algorithm
1: procedure MSDSA F,∀i ∈M, j ∈ Υ2: F = N, γthr, γconn thr, PAj, SAi,Ω, EijIo|PER, S3: Define ζ := timeslots counter4: Initialize ζ := 05: Optimization cycle:6: Update PAj, SAi,Ω, EijIo|PER, S(i, j),∀i ∈M, j ∈ Υ7: Calculate Pr(Λi) using (4.1), ∀i ∈M8: Define M ′ := Set of the active & connected SUs9: Initialize M ′ := 10: for i := 1 to M do Eliminate disconnected SUs11: begin12: if Pr(Λi) > γconn thr then13: M ′ = M ′, SUi14: end for loop15: Calculate Ω′ ∀i ∈M ′, j ∈ Υ16: Solve (4.24) and find x∗ij17: for L := 1 to M’ do Check co-channel interference18: begin19: if
∑M ′
K=1EjKLI0 > γintra, K 6= L, ∀j ∈ Υ then
20: xLj = 021: end for loop22: Declare MSDSA results and inform SUs23: while ζ ≤ ζ + N do24: begin25: for T := 2 to N do Perform local monitoring26: for all (i, j) pairs:27: begin28: if EijIo|PER < γthr & S(i, j) > Ro + ε &
∑MK=1E
jKLI0 < γintra then
29: xij(ζ)← xij(ζ − 1)30: else xij(ζ)← 031: end for loop32: end while loop33: Wait until N timeslots pass then goto 5.
4.5. RESULTS AND INTERPRETATION 61
4.5 Results and Interpretation
The performance of the algorithm is evaluated by the outage percentage (Poutage) of
the SU nodes. Poutage is the percentage of the time slots when any SU that requests
a channel fails to be assigned one. Success rate is defined as the percentage of time
slots that the SUs that have data to transmit succeed in getting a channel assigned
for them. Success rate is equivalent to 100 − Poutage, hence we discuss the perfor-
mance using the Poutage parameter only. SUs are assumed to be working in saturated
communication mode. As such, all SUs will have data to transmit all the time. The
communication traffic parameter, (λPU), of the PUs is set to 0.5.
Fig. 4.2 represents the outage percentage when the speed of the nodes is varying,
the round length N = 50, and number of channels = 10. It can be seen that as
the speed of mobile nodes increases, the outage percentage increases. Also, the MS-
DSA algorithm outperforms both underlay-only and random (CSMA/CA) allocation
algorithms.
Fig. 4.3 shows the impact of the number of SUs on Poutage when the speed of
the nodes is fixed at 3.6 km/hr or 100 km/hr. N is set to 50 and the number of
available channels is set to 10. When the SUs’ number is much larger than the
number of available channels in the system, the speed will have a small impact on the
performance of the MSDSA algorithm since the outage percentage is already high,
as illustrated in Fig. 4.3. Poutage can reach 70% at a speed of 3.6 km/hr and can
increase to 85% at a speed of 100 km/hr. This is due to the fact that the system is
trying to accommodate many nodes in this scenario, which are working in saturated
communication mode. This will lead to over-assigning the channels to many SUs
which will result in higher collision rates and more scrambled signals.
4.5. RESULTS AND INTERPRETATION 62
Figure 4.2: Outage percentage, Pout%, vs. speed of SUs for different number of SUswhen the number of available channels = 10
On the other hand, when the number of SUs is comparable to the number of
available channels (e.g. number of SUs≤ 3 x number of channels), then the algorithm
will perform much better and the success rate can reach more than 60% in fulfilling
the access requests from SUs that have data to transmit all the time (saturated
communication mode).
The impact of N on the performance of the algorithm for different numbers of SUs
and different speeds is illustrated in Figs. 4.4 and 4.5, respectively. It can be seen in
Fig. 4.4 that as N grows, the performance degrades faster when the number of SUs
4.5. RESULTS AND INTERPRETATION 63
Figure 4.3: Outage percentage, Pout%, vs. number of SUs for different speeds whenthe number of available channels = 10
is large. This is due to the fact that the duration between two consecutive rounds of
solving the optimization problem is long. While this may save a lot of computational
cost, the outage percentage will increase considerably especially when there are many
SUs contending to access the spectrum.
As illustrated in Fig. 4.5, the performance will degrade by 7.8%, at a speed of
3.6 km/hr, when N is 500 timeslots as compared to the case when N is set to 10.
This degradation in performance is acceptable since the computational cost will be
50 times less. When the speed is 100 km/hr, however, the performance will degrade
4.5. RESULTS AND INTERPRETATION 64
Figure 4.4: Outage percentage, Pout%, vs. N for different number of SUs whenspeed=15km/hr and the number of available channels = 10
by 52% and the outage percentage will jump from 26% to 78%. In other words, the
cost of increasing N from 10 to 500 timeslots is very high when the nodes are moving
at high speed since the geographical locations, the network topology, and the inter-
ference levels are changing quickly in this case.
In the following, the algorithm is improved by letting N be dependent on the
speed. Specifically, N is calculated as follows: N = min(⌊
500Speed
⌋, 500
). As such,
when nodes are moving on low speed, the algorithm will be run less often (N will
4.5. RESULTS AND INTERPRETATION 65
Figure 4.5: Outage percentage, Pout%, vs. N for different speeds when number ofSUs= 10 and the number of available channels = 10
have a high value, e.g. 500 at a speed of 1 km/hr) as compared to the case of moving
on high speed (e.g. 5 at a speed of 100 km/hr). Fig. 4.6, represents the outage
percentage versus speed when the number of channels is 10 and the number of SUs
is set to 10 or 30. Also, Fig. 4.7 shows the outage percentage versus speed when the
number of SUs is 50 and for a number of channels equal to 10 or 30. As seen in both
figures, the impact of mobility on the performance of the algorithm is reduced and
we can say that the performance is stable with minimal effect of the SUs’ mobility
speed.
4.5. RESULTS AND INTERPRETATION 66
Figure 4.6: Outage percentage, Pout%, vs. speed for different number of SUs whenthe number of channels = 10 and N is dependent on the speed of SUs
Moreover, Fig. 4.8 shows the performance when the SUs’ number is varying from
10 to 100, when N is dependent on the speed, the number of channels is equal to 10
and at speeds of either 3.6 km/hr or 100 km/hr. As the number of SUs grows, the
performance degrades since the number of channels (and hence access opportunities)
is fixed, while the demand is increasing. In order to comment on the effect of the
number of available channels on the performance of the algorithm, Fig. 4.9 shows
the outage percentage when the number of channels is varying from 10 to 100 when
4.5. RESULTS AND INTERPRETATION 67
Figure 4.7: Outage percentage, Pout%, vs. speed of SUs for different number of avail-able channels when the number of SUs = 30 and N is dependent on thespeed of SUs
N is dependent on the speed,. The number of SUs is set to 30 and mobility speed
is set to either 3.6 km/hr or 100 km/hr. As expected, the performance gets better
as the number of channels increases. When the number of channels is more than 50,
however, the performance will not be enhanced. This is due to the fact that most of
the SUs will be spread over different channels and there will be very little room for
improvement when the number of channels is significantly larger than SUs’ number.
4.6. CONCLUDING REMARKS 68
Figure 4.8: Outage percentage vs. number of SUs for different speeds when the num-ber of available channels = 10 and N is dependent on the speed
4.6 Concluding Remarks
In this chapter, I proposed a suboptimal DSA with a mobility support algorithm.
The algorithm considers a multichannel multiuser scenario. The proposed algorithm
is different from the traditional resource allocation algorithms in that it supports the
mobility of SUs and jointly takes into consideration nodes’ activity, the interference
levels, the connectivity, the PER regions and the correlated shadowing. It also adopts
an interweave/underlay hybrid approach as an access scheme while favouring the
4.6. CONCLUDING REMARKS 69
Figure 4.9: Outage percentage, Pout%, vs. the number of available channels for dif-ferent speeds when number of SUs = 30 and N is dependent on the speed
interweave scheme. The algorithm outperforms other classical DSA protocols while
being suitable to be implemented practically by simplifying the optimization problem
and reducing the computational cost.
70
Chapter 5
Integrating Energy Harvesting and Dynamic
Spectrum Allocation in Cognitive Radio Networks
Harvesting energy from ambient energy sources is an efficient method for extending
the lifetime of energy-constrained networks without the need for external power sup-
ply or periodic battery replacements [97]. Moreover, EH is promising to increase the
Energy Efficiency (EE) of CRNs, which is important due to the limited capacity of
SUs’ batteries. Although EH is an attractive technique, it introduces new challenges
due to the dynamic and discontinuous characteristics of EH opportunities.
In order to enable EH, it is crucial to discover the EH opportunities, which can be
performed using energy detection. Fig. 5.1 shows the components that an EH-enabled
SU should have to perform EH [98]. Namely, these components are:
• A software-defined radio-based wireless transceiver.
• A Radio Frequency (RF) energy harvester.
• A spectrum analyzer which observes and analyzes the activity of spectrum us-
age.
71
• A memory-equipped processing unit which can maintain a database, extract
information and make intelligent decisions about sensing and the available EH
opportunities.
• A decision making unit.
• An energy storage device, which could be a rechargeable battery to store the
harvested energy for future use.
• A power management unit, which dispatches the energy from the RF energy
harvester.
Figure 5.1: Components of an EH-enabled SU
SUs that have only one transceiver can either harvest energy or access the spec-
trum at any given time. This means that SUs that harvest energy will miss opportuni-
ties to access the spectrum. EH should not severely impact the performance of DSA,
however. This can be ensured by employing two strategies: first, by implementing
EH within the context of the DSA algorithm and second, by introducing some local
and global parameters to be used to control the operating mode preferences of the
72
individual SUs and the overall CRN. The general platform for integrating EH and
DSA is provided in Fig. 5.2.
Figure 5.2: General Platform for integrating EH with DSA
In previous studies, reviewed in Chapter 2, PUs’ activity is the main factor in
making decisions regarding EH. The average level of energy in the CRN and the SUs’
operating-mode preferences had no control over the EH process. This approach is
not the best since it may severely impact the performance of the DSA and the QoS
of SUs that have delay-sensitive data.
Moreover, all of the previous efforts assumed very simple networks that consist
of two SUs at most. In practical scenarios, however, CRNs can have multiple SUs
that contend to access the spectrum and/or harvest energy. Furthermore, the work
that has been done in this field focused on the theoretical aspects of the EH and its
5.1. RANGE OF HARVESTING 73
impact on the achieved throughput for simple systems. The practical implementation
of this technology in CRN and how to integrate it with the DSA algorithms was
not investigated. Motivated by this and the recent advances in designing efficient
circuits and devices for RF energy harvesting that are suitable for low-power wireless
applications [98], I propose a flexible and effective operating framework to enable EH
in CRNs.
Specifically, I develop a novel mode-selection strategy that allows every SU to
have some control over its preferred mode of operation: either transmitting data,
harvesting energy, or staying silent. The goal is to improve both energy efficiency
and spectral efficiency while being able to control the impact of EH on the DSA
algorithm.
The remainder of this chapter is organized as follows: Range of Harvesting (RoH)
concept is described in Section 5.1. Discovering the opportunities for EH is discussed
in section 5.2. Section 5.3 describes the proposed EH-enabling platform for CRNs.
The results and interpretations are given in Section 5.4 and concluding remarks are
provided in Section 5.5.
5.1 Range of Harvesting
The CRBS executes optimizing the DSA and EH. It also supervises the level of energy
in the overall CRN and has the power to force SUs to harvest energy in case the level
of energy is lower than a specific threshold, as will be discussed in Section 5.3.
The considered system is similar to the one described in Chapter 3. In addition,
SUs that are within a distance of RoH of an active PU will be identified as SUs that
potentially can harvest one unit of energy every time slot, as shown in Fig. 5.3. Given
5.2. ENERGY DETECTION 74
the current advancement is EH circuits, RoH can be in the range of 10 meters from
the active nodes. The energy detector is also used to make decisions concerning the
EH cycle, as explained in the following section.
Figure 5.3: RoH Concept
5.2 Energy Detection
Energy detection is a widely used approach to detect the strength of a signal [99],
and based on this, a decision on the availability of EH opportunities can be made.
The block diagram of an Energy Detector (ED) is given in Figure 5.4.
Figure 5.4: Block diagram of an energy detector
5.2. ENERGY DETECTION 75
The Band-Pass Filter (BPS) removes the out-of-band signals and keeps the signal
in the band under consideration. An Analog to Digital Converter (ADC) digitizes
the filtered signal Rs(t) at the Nyquist rate (fs = 2B samples/sec) where fs is the
sampling frequency. This results in a set of Sset = 2BT samples which are then fed
to a squaring device followed by an accumulator to get the energy content,Ec, in Sset.
After that, Ec is compared to a threshold, EDthr , to decide whether there exists an
EH opportunity or not.
The performance of the ED is linked to its specificity (measured by the false
alarm probability, Pfa, which represents the probability that the detection algorithm
falsely decides that a PU is present in the considered frequency band when it is not)
and sensitivity (measured by the probability of detection, Pd, which represents the
probability of correctly detecting the PU signal in the scanned frequency band) [100].
Mathematically speaking, Pfa and Pd can be defined as:
Pfa = Pr(signal is detected|H0) = Pr(Ec > EDthr |H0)
=
∫ ∞EDthr
f(Ec|H0)du(5.1)
Pd = Pr(signal is detected|H1) = Pr(Ec > EDthr |H1)
=
∫ ∞EDthr
f(Ec|H1)du(5.2)
where f(Ec|Hl) represents the probability Density Function (PDF) of Ec when
the investigated PU is active (H1) or inactive (H0). In order to maximize the amount
of harvested energy units, Pd should be maximized while minimizing Pfa. There is a
trade-off, however, between Pd and Pfa where choosing higher Pd will result in having
high Pfa and vice versa. Moreover, the EDthr has full control over the performance
5.3. FRAMEWORK OF ENABLING EH IN CRNS 76
of the ED and it is crucial to have a suitable EDthr [101].
Practically speaking, if the goal of the EH algorithm is to have a low probability
of mis-detecting EH opportunities, Pfa is set to a reasonably small value and EDthr
is chosen such that Pd is maximized. This requirement is met by evaluating EDthr
based on (5.1). In this case, Pfa is given by [102]:
Pfa =Γ(Ec2, EDthr
2σ2ij
)Γ(Ec)
∆=FEc
2
(EDthr
2σ2ij
)(5.3)
where Γ(a) is the Gamma function, Γ(a, b)∆=∫∞bxa−1e−xdx is the incomplete
Gamma function and σ2ij is the variance of the correlated shadow fading map that
SU i is experiencing at channel j. Next, EDthr can be found based on (5.3) as follows:
EDthr = 2σ2ijF−1Ec2
(Pfa) (5.4)
To this end, the formulation of detecting the available EH opportunities was pro-
vided. The next step is to study the implication of these discovered opportunities on
the performance of both EH and DSA. The proposed approach is described in the
next section.
5.3 Framework of Enabling EH in CRNs
Due to the specific nature of EH (such as the dynamic availability of the harvesting
opportunities and the communication requirements of the SU units), traditional DSA
access protocols need to be modified in order to integrate EH along with efficient
spectrum allocation. In particular: QoS, the spectrum efficiency, and the available
EH opportunities need to be jointly optimized.
5.3. FRAMEWORK OF ENABLING EH IN CRNS 77
In order to give CRN the control over deciding whether to choose EH or DSA,
depending on the level of energy in the network and on the type of data to be transmit-
ted (whether it is a delay-sensitive data or not), two new parameters are introduced:
κi and z(t). κi is a controlling parameter that is set individually by each SU i in order
to have the flexibility to decide which mode of operation they prefer while z(t) is a
global parameter set by the CRBS to take the energy level in the overall CRN into
consideration while making the final spectrum allocation and EH decisions.
Choosing a suitable value for z(t) depends on the type of the CRN and the char-
acteristics of the supported applications. For example, if the supported applications
need to send delay-sensitive data (e.g. the readings of smart meters in Smart Grids
(SGs)), EH will not be a priority and the CRBS will set z(t) to have a high value
(e.g. z(t) = 0.9) which means that the overall aim of the network is to focus on
accessing the spectrum and transmitting as much data as possible. On the other
hand, if the type of supported applications can tolerate delay and the CRN prefers
to have their batteries to last longer (i.e. have a longer lifetime) rather than having
a high transmission rate, then EH mode will be favoured globally in the network by
setting z(t) to have a small value. A real-life scenario for this case is the CRNs that
are used for forest monitoring where sensors are deployed in the forest and transmit
information about the surrounding environment until their batteries are depleted.
After that, there will be a need to replace nodes that are not functioning anymore
due their depleted battery. The replacement cost can be avoided by enabling EH such
that nodes’ lifetime is increased.
Also, SUs will have some control over the decision making by selecting values for
κi,∀i ∈ M , that reflect their preference. For example, if SU i prefers to access the
5.3. FRAMEWORK OF ENABLING EH IN CRNS 78
spectrum, it will choose a small value for κi (e.g. κi = 0.1) and if it is interested
more in harvesting energy, whenever possible, it will choose a large value for κi (e.g.
κi = 0.9).
Let us also define a parameter called the level of battery threshold for SU i (LoBithr)
which represents the percentage of the available energy in SU i’s battery when it is
ordered by the CRBS to harvest energy instead of accessing the spectrum. LoBithr
depends on z(t) and κi and can be calculated by employing the following equation:
LoBithr% =
(1− 2z(t)(1− κi)) ∗ 100 z(t) < 0.5
100κi z(t) = 0.5
100(2κi(1− z(t))), z(t) > 0.5
(5.5)
Fig. 5.5 shows the relationship between z(t) and LoBithr for two SUs with different
κi values. SU1 sets κ1 to 0.1 which means that it prefers to transmit data over
harvesting energy and hence it will not be forced to switch to EH mode until its
battery level drops to 55% when z(t) = 0.25 or when the battery level is below 5%
when z(t) = 0.75. Additionally, SU2 sets κ2 to 0.9 which means that it prefers to
harvest energy and hence it will be forced to switch to EH mode once its battery
level drops to 95% when z(t) = 0.25 or when the battery level is below 45% when
z(t) = 0.75.
Based on the aforementioned parameters, the proposed algorithm that jointly op-
timizes the EH and DSA is provided in Algorithm 2. The algorithm starts with col-
lecting information about the activity of PUs (PAj ∀j), the activity of SUs (SAi ∀i),
the mutual distances between the SUs and the PUs (S(i, j) ∀i, j), and RoH. The
decision parameter of the algorithm is xij ∈ 0, 1, EH where 1 means that SU i is
5.3. FRAMEWORK OF ENABLING EH IN CRNS 79
0 0.2 0.4 0.6 0.8 10%
20%
40%
60%
80%
100%
z( t )
LoB
thr
i %
SU1,κ1=0.1
SU2,κ2=0.9
Figure 5.5: CRBS’s decision parameter vs. the energy level in the SUs’ batteries fordifferent values of κi
assigned channel j to transmit data in the next time slot, EH means that SU i will
harvest energy from channel j, and 0 means that SU i will not have any activity on
channel j.
The CRBS chooses a suitable z(t) depending on the type of the supported appli-
cations. Also, a value is set to Pfa and based on this value, EDthr is calculated. After
that, each SU i selects its κi and sends this value along with the sensing information
and its battery level to the CRBS. Next, if the average level of energy in the network
is lower than a threshold, the CRBS will direct SUs that are in the vicinity of an
active PU to harvest energy, provided that the level of energy in the network is lower
than LoBithr. The other active SUs that indicated interest in accessing the spectrum
instead of harvesting energy will be allocated the available channels. The channels
will be allocated by solving an optimization problem that minimizes the interference
5.3. FRAMEWORK OF ENABLING EH IN CRNS 80
Algorithm 2 Enabling-EH in DSA platform
1: procedure EH(S(i, j), PAj, SAi, RoH, ∀i ∈M, j ∈ Υ)2: CRBS chooses z(t)3: Define LoC1:=List of contending SUs to access the spectrum4: Initialize LoC1:=5: Set Pf and calculate EDthr using (5.4)6: for i := 1 to M do Select a suitable κ7: begin8: SU i chooses κi9: SU i senses 1 channel and sends sensing results, LoBi and κi to CRBS10: end For loop11: Define Υ1 := set of active channels12: Define Υ2 := set of inactive channels13: Define (LoB) := average level of Energy in the CRN14: (LoB) =
∑Mi=1 LoBi/M
15: for i := 1 to M do Decide the best operating mode16: begin17: Find LoBi
thr based on (5.5)
18: if S(i, j) < RoH & (LoB) ≤ LoBithr & j ∈ Υ1 then
19: [ Force SU i to switch to ch. j and operate in EH mode20: xij ← EH ]
21: Else If S(i, j) > RoH & (LoB) ≥ LoBithr & SAi 6= 0,∀i
22: LoC1 = LoC1, SUi
23: end For loop24: Assign channels ∈ Υ2 to the SUs ∈ LoC1 by solving interference minimization
optimization problem25: xij ← 1 for successful contending SUs, phase 1 (Interweave)26: if all SUs ∈ LoC1 got a channel then27: goto 32.28: else [29: Define LoC2 := LoC1− SUs granted access in phase 130: Underlay assignment of channels ∈ Υ1 for the SUs ∈ LoC2 by solving interference
minimization optimization problem31: xij ← 1 for successful contending SUs, phase 2 ]32: Declaration of results :33: if SU iwas unsuccessful in accessing spectrum or HE then34: mode of operation: Silent35: Ensure: EijIo < γthr∀ allocated pairs(i, j)36: Declare results and exit
5.4. RESULTS AND INTERPRETATIONS 81
introduced from the SUs to the PUs. A hybrid interweave/underlay access scheme is
adopted and the DSA will be performed in two phases: in the first phase, channels
with non-active PUs will be assigned to SUs (interweave access scheme). The sec-
ond phase involves allocating the channels that have active PUs using an underlay
scheme. SUs that were neither directed to harvest energy nor allowed to access the
spectrum will remain silent in the next time slot. Finally the CRBS decisions will be
broadcasted to the SUs and the EH/DSA cycle will start again.
5.4 Results and Interpretations
SUs are considered to have limited storage capabilities and can store up to 1000 units
of energy. They require one unit of energy to transmit one packet of data. The
activity of the PUs is considered to follow a Bernoulli arrival process with parameter
λPU = 0.5. To study the worst case scenario, the SUs are assumed to be working in
the saturated mode, i.e. λSU = 1, so that all of the SUs will have a packet to transmit
in each time slot. This means that if a SU is harvesting energy in a time slot, a packet
will be prevented from transmission. By assuming the worst case scenario, we will
be able to capture the impact of EH on the performance of accessing the spectrum.
Without loss of generality, z(t) is set to 0.6 and Pfa is set to 5%. SUs are considered
to choose a similar value for κi, i.e., κi = κ ∀ SU i ∈M . RoH is set to 50 meters and
the simulation area is set to 1000 x 1000 m. The levels of SUs’ batteries are assumed
to be at 50% of the maximum storage capacity at the start of the simulations.
Figs. 5.6 and 5.7 show the average energy level in the SUs and the cumulative
dropped packets over time, respectively, for different values of κ. θSU in this simulation
is set to 50 and the number of available channels is set to 10. As these figures illustrate,
5.4. RESULTS AND INTERPRETATIONS 82
opting for EH mode, by setting a high value for κ, may increase the lifetime of the
network but it does not mean a better performance in terms of spectrum utilization
and data transmission. This is due to the fact that when SUs switch to EH mode,
they miss transmission opportunities. For instance, in Fig. 5.6, when κ is equal to
0.9, the CRN will maintain a 55% and 58% level of energy for hybrid and interweave
access schemes, respectively, after 900 time slots of simulation. However, this will
come with the cost of increased dropped packets which will be 58% and 60% after
900 time slots of simulation for hybrid and interweave access schemes, respectively,
as shown in Fig. 5.7. On the other hand, setting κ to 0.3 will substantially improve
the spectrum access performance and the cumulative dropped packets will decrease
to 42% and 39% in the case of hybrid and interweave access schemes, respectively.
Also, as shown in Fig. 5.6, the energy level in the network will be maintained at
15% and the hybrid scheme outperforms the interweave scheme in the first 500 time
slots, in terms of success rate. After that, the energy level in the CRN will be very
low and the CRBS will start forcing SUs to harvest energy or stay silent, to comply
with the minimum allowed level of energy in the network, and the system will fail in
maintaining a good spectrum efficiency that is comparable to the performance in the
first 500 time slots. Moreover, when the EH operating mode is not supported, the
energy in the network will be fully depleted after 900 time slots and Fig. 5.7 shows
that the packet drop rate will be 31% and 37% in the case of hybrid and interweave
access schemes, respectively.
In order to study the effect of the number of available channels on the performance
of the proposed algorithm, Figs. 5.8 and 5.9 provide the average energy level in
the SUs and the cumulative dropped packets, respectively, for different numbers of
5.4. RESULTS AND INTERPRETATIONS 83
0 100 200 300 400 500 600 700 800 9000%
10%
20%
30%
40%
50%
60%
Time−slots
Ene
rgy
leve
l %
EH using Hybrid access scheme, κ=0.3
EH using Interweave access scheme, κ=0.3No EHEH using Hybrid access scheme, κ=0.9
EH using Interweave access scheme, κ=0.9
Figure 5.6: The average energy level in the SUs over time for different values of κwhen θSU is 50 and number of available channels is 10 and z(t) = 0.6
channels and different values of κ. These results are taken after 100 time slots of
simulations and with θSU set to 50.
As the number of channels increases, both the opportunities to harvest energy
and to access the spectrum increase substantially. In Fig. 5.8, when κ = 0.9, the
energy level will be 59% for both of the studied access schemes when the number of
available channels is 80. Also, the cumulative dropped packets will be 9% in the same
case and this percentage represents the time slots in which the SUs chose to harvest
energy instead of transmitting data. On the other hand, when SUs prefer to access
the spectrum over harvesting energy (by setting κ to 0.3), the packet drop rate will
5.4. RESULTS AND INTERPRETATIONS 84
0 100 200 300 400 500 600 700 800 9000%
10%
20%
30%
40%
50%
60%
Time−slots
Cum
ulat
ive
Dro
pped
pac
kets
%
No EH, Hybrid scheme
No EH, Interweave scheme
EH using Hybrid access scheme, κ=0.3
EH using Interweave access scheme, κ=0.3
EH using Hybrid access scheme, κ=0.9
EH using Interweave access scheme, κ=0.9
Figure 5.7: The cumulative % of dropped packets over time for different values of κwhen θSU is 50 and number of available channels is 10 and z(t) = 0.6
be as low as 1% and 3.3%, when the number of channels is 80, in the case of hybrid
and interweave access schemes, respectively. The energy level will drop to 42% and
47% in the case of hybrid and interweave access schemes, respectively, as depicted in
Fig. 5.9. The performance of the CRN in the case of κ = 0.3 and the case of no EH is
the same since the system has many available spectrum access opportunities and all
of the focus will be on accessing the spectrum. Hence, no energy will be harvested.
While this might be useful in the short term, since it will increase SUs’ chances to
access the spectrum and increase the number of successfully transmitted packets, the
energy level in the CRN will drop and the network will eventually die if SUs do not
start harvesting energy before the occurrence of this event.
5.4. RESULTS AND INTERPRETATIONS 85
10 20 30 40 50 60 70 8042
44
46
48
50
52
54
56
58
60
Number of Channels
Ene
rgy
Leve
l %
No EH, Hybrid schemeNo EH, Interweave schemeEH using Hybrid access scheme, κ=0.3EH using Interweave access scheme, κ=0.3EH using Hybrid access scheme, κ=0.9EH using Interweave access scheme, κ=0.9
Figure 5.8: The average energy level in the SUs versus the number of channels in thesystem for different values of κ at time slot 100 when number of SUs is50 and z(t) = 0.6
Fig. 5.10 presents the cumulative harvested energy units versus θSU in the CRN
after 200 time slots of simulations when the number of available channels is set to 30.
As shown in Fig. 5.10, when the number of SUs increases, the cumulative har-
vested energy will grow due to the increased opportunities of SUs to be in the vicinity
of an active PU. The performance of both hybrid and interweave-only access schemes
is similar in the case of κ = 0.9 since all of the efforts are directed towards harvesting
energy. When κ = 0.3, however, the hybrid system will not harvest any energy units
since all SUs are in the mode of accessing the spectrum. On the other hand, using an
interweave access scheme will introduce some balance between energy harvesting and
5.5. CONCLUDING REMARKS 86
10 20 30 40 50 60 70 801
2
3
4
5
6
7
8
9
10
Number of Channels
Cum
ulat
ive
Dro
pped
pac
kets
%
No EH, Hybrid schemeNo EH, Interweave schemeEH using Hybrid access scheme, κ=0.3EH using Interweave access scheme, κ=0.3EH using Hybrid access scheme, κ=0.9EH using Interweave access scheme, κ=0.9
Figure 5.9: The cumulative % of lost packets versus the number of channels in thesystem for different values of κ at time slot 100 when number of SUs is50 and z(t) = 0.6
spectrum access since the SUs will be directed to harvest energy instead of accessing
the spectrum in underlay mode in cases where there are many active PUs and the
number of vacant channels is small.
5.5 Concluding Remarks
Enabling EH for CRNs in the context of DSA has the potential advantage of a pro-
longed lifetime without requiring external power cables or periodic battery replace-
ments. This chapter studied enabling EH for CRNs in the context of DSA, and a
novel algorithm that jointly considers the spectrum efficiency and energy efficiency
5.5. CONCLUDING REMARKS 87
Figure 5.10: The Cumulative harvested energy units for different number of SUs attime slot 200 and number of available channels is 30 and z(t) = 0.6
is proposed. This algorithm provides SUs with a degree of freedom in selecting their
operating mode while keeping the CRBS in charge of making the final decisions about
the operating modes of the SUs such that a certain level of energy in the whole net-
work is maintained. The impact of EH on the performance of the DSA algorithm has
been studied for both the hybrid and the interweave access schemes. CRNs that adopt
the proposed algorithm will have the potential advantage of a prolonged lifetime while
being able to dynamically access the underutilized spectrum.
88
Chapter 6
Power Allocation for Cognitive Radio Networks
As discussed in the previous chapter, power is an important resource and to increase
the lifetime of CRNs, EH should be enabled. In order to simplify the problem of
integrating EH in the context of DSA, it was assumed that the transmission power
budget in the CRN is not constrained. However, power allocation among SUs needs
to be optimized especially when the allowed transmission power level is limited.
Moreover, in order to reduce the high complexity of obtaining an optimal solution
for the multi-objective problem of jointly optimizing spectrum allocation and power
allocation in a multichannel multiuser scenario, the problem is split into two stages:
first, allocate the spectrum (based on the different constraints that were discussed
in the previous chapters) and then allocate the power to SUs that were successful in
reserving a channel. This approach is proven to provide a near optimal solution while
avoiding the complexity of solving combinatorial optimization problems which have
complexity that grows exponentially with the input size [103].
In this chapter, two optimal algorithms are proposed to optimize the power allo-
cation among SUs that were successful in accessing the spectrum. The objective is
to maximize the Spectral Efficiency (SE) while respecting the power budget, along
6.1. PROBLEM FORMULATION 89
with other constraints. The scenario in which the CRN has multiple SUs that are
interfering with several PUs is addressed. Consequently, the power budget should
be allocated to the SUs subject to different power constraints so that the hybrid in-
terweave/underlay access scheme is adopted, which means that SUs can access the
active and non-active PU bands. Hence, different SUs will have different power and
interference limits depending on PUs’ activity and on which SUs will be allocated to
transmit on the same channel simultaneously. Moreover, since the complexity of the
optimal algorithms can be high, a suboptimal discrete Cap-Limited Heuristic (CLH)
algorithm is proposed. The CLH algorithm considers assigning power to the SUs from
a discrete set of power levels as will be discussed later in this chapter.
The remainder of this chapter is organized into the following sections. Section
6.1 provides the problem formulation, while the proposed suboptimal algorithm is
discussed in 6.2. Simulation results and interpretations are given in section 6.3 and
the conclusions are presented in Section 6.4.
6.1 Problem Formulation
The outcome of the spectrum allocation (xij) is assumed to be available at the CRBS
and will be used as an input to the power allocation algorithms. The maximum
achievable throughput can be evaluated using Shannon’s formula:
Throughput = ∆f log2
(1 +
P iSU
σ2AWGN + Eji
PUI0
)(6.1)
where σAWGN is the variance of the Additive White Gaussian Noise (AWGN), and
EjiPUI0 is the interference introduced by PU j that is occupying channel j to SU i,∀i
6.1. PROBLEM FORMULATION 90
transmitting on channel j. EjiPUI0 is defined as follows:
EjiPUI0 =
GjoS(i, j)−η exp
(12
(σij
log 1010
)2)P jPU , PAj = 1
0, PAj = 0(6.2)
After calculating the achievable throughput of each SU, the total achievable through-
put can be defined as the sum of the transmission rates of all active SUs [104]. The
objective is to optimize the power allocation in such a way that maximizes the total
achievable throughput. The algorithm is supposed to protect PUs’ communications
and the total allocated power should be equal to or less than the total available power
budget. Moreover, the allocated power to any SU cannot be higher than the max-
imum allowed transmission power. The optimization problem can be formulated as
SU 6: Find ZA := List of SUs gained access @ TS (TSk+1w + 1)7: ZA := A1
SU , A2SU , A
3SU , ..., A
zaSU
8: Initialize PowiSU =⌊PT ∗100za
⌋∗ TP, ∀ i ∈ ZA
9: for i := 1 to za do limit powSU10: begin11: powiSU = minpowiSU , Pmax
SU 12: end13: Initialize ind := zeros(za, 1), decision := zeros(za, 1)14: for i := 1 to za do Effect of γthr on powSU15: begin16: while EijIo < γthr & powiSU < Pmax
Figure 6.2: The achieved spectral efficiency versus γthr when the total available powerbudget is 4W and σ2
AWGN = 10−7
the power profiles of the three algorithms with each other, it can be noted that they
behave similarly when γthr is high. Also, under this scenario, SUs that are assigned to
a channel with good quality (e.g. channel 8) will be allocated less power as compared
to the scenario where γthr is set to 0.4mW . This is due to the fact that γthr is high
and hence some SUs will be able to transmit higher power (as compared to the case
of γthr = 0.4mW ) even if they produce some interference to the PUs. The power
6.3. RESULTS AND INTERPRETATION 100
1 2 3 4 5 6 7 8 9 100
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
Channel Number
Ass
igne
d P
ower
(W
)
total Power Budget = 0.1Wtotal Power Budget = 0.3Wtotal Power Budget = 0.5Wtotal Power Budget = 0.6Wtotal Power Budget = 0.7Wtotal Power Budget = 0.9Wtotal Power Budget = 1.1W
Figure 6.3: The assigned power for each channel in the case of the CLH algorithmwhen γthr is set to 0.4mW
deducted from the SUs with good channels will be assigned to the other SUs with
bad channels and this will increase the level of fairness among SUs. In addition,
the constraints of total power budget and maximum transmission power will be the
dominant factors in making the decisions about power allocation.
Figs. 6.9 and 6.10 illustrate the total transmitted power and the spectral efficiency,
respectively, versus the total available power budget when γthr is set to 0.4mW . As
6.3. RESULTS AND INTERPRETATION 101
1 2 3 4 5 6 7 8 9 100
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
Channel Number
Ass
igne
d P
ower
(W
)
total Power Budget = 0.1Wtotal Power Budget = 0.3Wtotal Power Budget = 0.5Wtotal Power Budget = 0.6Wtotal Power Budget = 0.7Wtotal Power Budget = 0.9Wtotal Power Budget = 1.1W
Figure 6.4: The assigned power for each channel in the case of the ET algorithm whenγthr is set to 0.4mW
depicted, when the available power budget is higher than 0.8W , the total transmitted
power does not increase since SUs cannot cause high interference levels to the PUs
in order to fulfil the tight interference condition and hence γthr will be the major
decision maker regarding the power allocation. In terms of spectral efficiency, the
CLH algorithm will slightly outperform the other two algorithms since it allocates
higher power to SUs accessing the spectrum using an interweave scheme.
As γthr is increased, the transmitted power will increase linearly with the increase
of the available budget up to 1.2W where the maximum transmission power will be
6.3. RESULTS AND INTERPRETATION 102
1 2 3 4 5 6 7 8 9 100
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
Channel Number
Ass
igne
d P
ower
(W
)
total Power Budget = 0.1Wtotal Power Budget = 0.3Wtotal Power Budget = 0.5Wtotal Power Budget = 0.6Wtotal Power Budget = 0.7Wtotal Power Budget = 0.9Wtotal Power Budget = 1.1W
Figure 6.5: The assigned power for each channel in the case of the ECM algorithmwhen γthr is set to 0.4mW
reached and no higher power can be allocated to the SUs.
To summarize, when the acceptable level of interference (γthr) is high, the domi-
nant parameter that affects the power allocation will be the available power budget.
On the other hand, assuming that the CRN has high power budget, then the power
allocation will depend only on the maximum allowed transmission power and on γthr.
If the constraints on the power budget and the interference levels are loose, then the
only parameter that will play a role in making decisions about power allocation will
be the maximum allowed transmission power.
6.3. RESULTS AND INTERPRETATION 103
1 2 3 4 5 6 7 8 9 100
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
Channel Number
Ass
igne
d P
ower
(W
)
total Power Budget = 0.1Wtotal Power Budget = 0.3Wtotal Power Budget = 0.5Wtotal Power Budget = 0.6Wtotal Power Budget = 0.7Wtotal Power Budget = 0.9Wtotal Power Budget = 1.1W
Figure 6.6: The assigned power for each channel in the case of the CLH algorithmwhen γthr is set to 1.5mW
Finally, to study the impact of AWGN on the SE, Figs. 6.11, 6.12, and 6.13
illustrate the achieved SE for the CLH, ET, and ECM algorithms, respectively, versus
γthr for different values of σAWGN . As depicted, when σAWGN is high, the achieved
SE is low as compared to the cases when σAWGN is low. This is due to the fact that
when AWGN is high, SUs will not have enough room to add much interference to
the PUs and the interference constraint will be tighter. In addition, higher AWGN
means lower SINR and consequently a lower SE is achieved.
6.4. CONCLUDING REMARKS 104
1 2 3 4 5 6 7 8 9 100
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
Channel Number
Ass
igne
d P
ower
(W
)
total Power Budget = 0.1Wtotal Power Budget = 0.3Wtotal Power Budget = 0.5Wtotal Power Budget = 0.6Wtotal Power Budget = 0.7Wtotal Power Budget = 0.9Wtotal Power Budget = 1.1W
Figure 6.7: The assigned power for each channel in the case of the ET algorithm whenγthr is set to 1.5mW
6.4 Concluding Remarks
In this chapter, different optimization algorithms were proposed to distribute the
available power budget among the active SUs in such a way that the spectral effi-
ciency is maximized. As demonstrated by extensive simulations, the proposed subop-
timal CLH algorithm provides a near optimal performance and does not require high
computational cost as compared to the cost of solving the Lagrangian dual problem.
6.4. CONCLUDING REMARKS 105
1 2 3 4 5 6 7 8 9 100
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
Channel Number
Ass
igne
d P
ower
(W
)
total Power Budget = 0.1Wtotal Power Budget = 0.3Wtotal Power Budget = 0.5Wtotal Power Budget = 0.6Wtotal Power Budget = 0.7Wtotal Power Budget = 0.9Wtotal Power Budget = 1.1W
Figure 6.8: The assigned power for each channel in the case of the ECM algorithmwhen γthr is set to 1.5mW