Sensing-Assisted Spectrum Access Strategy and Optimization in Cognitive Radio Networks by Ratan Kumar Mondal M.S. (Electronics); B. Sc. Eng. (EEE) A thesis submied in fullment of the requirement for the degree of Doctor of Philosophy School of Electrical Engineering and Computer Science Science and Engineering Faculty eensland University of Technology 2018
198
Embed
Sensing-Assisted Spectrum Access Strategy and Optimization ... Kumar_Mondal_Thesis.pdf · Sensing-Assisted Spectrum Access Strategy and Optimization in Cognitive Radio Networks by
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.
Transcript
Sensing-Assisted Spectrum AccessStrategy and Optimization inCognitive Radio Networks
by
Ratan Kumar Mondal
M.S. (Electronics); B. Sc. Eng. (EEE)
A thesis submied in fullment of the requirement for the degree of
Doctor of Philosophy
School of Electrical Engineering and Computer Science
Science and Engineering Facultyeensland University of Technology
based detection. On the other hand, interference-temperature measurement considers
the cumulative RF energy from PU transmission and sets a maximum limit that primary
users can tolerate. Secondary users can use the band if their transmissions do not
exceed the interference-temperature limit [75]. Owing to the diculty in distinguishing
the primary signal from noise/interference in interference-temperature measurement,
the primary signal detection method becomes ecient in cognitive radio for spectrum
sensing.
2.2 Components of Cognitive Radio 19
Figure 2.2: Review of dierent detection techniques based on complexityversus accuracy [2].
In the literature on spectrum sensing, energy detection is popular because of its low
complexity and cost eectiveness. However, its detection performance during channel
impairments is poor when compared with the performance of other detection methods
as shown in Fig. 2.2 [2, 46]. If the SUs have prior knowledge regarding the primary
transmission then the matched-lter is the optimal detector, as it has the capability to
maximize the received signal-to-noise ratio during the worst channel conditions. As PUs
do not share their transmission informationwith the CR network, it is, therefore, dicult
to implement the matched-lter in conventional spectrum sensing. e cyclostationary
method can classify the signal hence there is capability to distinguish the co-channel
interference. However, the cyclostationary method requires longer computational time
with higher complexity to achieve a target detection when compared with the energy
detection and the matched-lter method, which may increase the overall sensing time
[31]. Waveform-based sensing and radio identication method are relatively robust
than energy detector owing to the coherent processing and feature extraction capab-
ilities with the help of a prior knowledge regarding PU’s characteristics and paerns.
Each detection method has pros and cons in relation to accuracy, complexity, cost-
eectiveness, system limitations, and assumptions. Even though energy detection has
poor detection performance, it becomes the method for PU detection in CR due to the
20 2.3 Standardization and Implementation of CR
working capability without any prior knowledge.
Considering the importance of sensor parameters for further contributing to spec-
trum access, the parametric hypothesis-testing procedure is used in this research.
Simply, the transmission of the SUs are constrained with the target PD which is not
specically provided with the selection factors of any detector. However, the sensing
eciency of the detectors may vary in dierent channel conditions. ese variations
only signify the aribute of dierent detectors. ere is a large body of literature on the
choice of detection method for RF signal. e study of detector selection as a tool for
spectrum sensing is not an objective of this research, but most importantly, the further
utilization of the sensing parameters with a post-processing algorithm is an objective
to improve the spectral eciency. Detector-independence and post-processing of the
sensing parameters provide greater exibility in designing the universal access protocol,
and they have become extremely signicant in the recent literature on CR technology.
2.3 Standardization and Implementation of CR
Due to the demand of CR technology for ecient utilization of the spectrum, the IEEE
802.22 Working Group on Wireless Regional Area Network (WRAN) has launched a
standard based on CR operation. IEEE 802.22 [78] is the rst standard that allows CR
operation for wireless networks. is standard species the air interface which operates
in the VHF and UHF broadcasting bands in the range of 54 − 862 MHz. e support-
ing network architecture in this standard is point-to-multipoint WRAN consisting of a
professional xed base station (BS) and customer premise equipment (CPE) as shown in
Fig. 2.3. e CPE can be xed and portable user terminals which can tune to the given
TV broadcast bands. e purpose of this standard is to provide alternatives to wire-line
broadband access in diverse geographic areas where population density is low.
e IEEE standard mainly denes the PHY and theMAC layer operations for support-
ing the purposes of WRAN in TV bands. e network coverage under a BS is typically
10 − 30 km depending on the EIRP and antenna’s specication. e coverage of the
WRAN system can be further upgraded to 100 km based on special scheduling in the
2.3 Standardization and Implementation of CR 21
Figure 2.3: Network architecture of the IEEE 802.22 WRAN, where users ofTV bands and wireless microphones are the primary users, andBS and CPE are the secondary users [3].
MAC and exceptional RF signal propagation in the PHY. To meet the requirements of
PU protection and ecient spectrum utilization, the CR capabilities comprise spectrum
sensing, database access services, channel set management, and geolocation services.
All supporting devices, such as BS and CPE, need to be empowered with the given CR
capabilities. e CR capabilities enable the BS and the CPE to produce robust decisions
regarding the characteristic of the using RF. e trac activity of the incumbents of
the TV bands is dynamically updated in the database which information can be used as
a supplement of the sensing to protect the incumbents. e type of detectors are not
regulated by this standard. However, this standard promotes the scheduled spectrum
sensing and sensing information sharing to make a central decision.
As the research activities documented in this thesis is focused on the issue of sensing-
assisted access strategy, the IEEE 802.22 standard is thus reviewed on the context of
research issues. According to the standard, the MAC accommodates all necessary tools
for protecting the incumbent users of TV bands and for the coexisted services. In a cell,
22 2.4 Current Trends and Applications of CR
the BSmanagesmultiple CPEs and themedium access is controlled by theMAC protocol.
In the downstream transmission, when BS transmits and CPE receives, MAC protocol
supports time-divisional multiplexing (TDM). During the upstream transmission, MAC
provides a combination of access strategy depending on the user application including
its QoS. When multiple CPEs aempt to transmit simultaneously by sharing the same
channel within a cell or overlapping cells, then the MAC provides the upstream schedul-
ing based on the following mechanisms: unsolicited bandwidth grants, polling, MAC
header-based contention, and CDMA-based contention. No specic access protocol is
proposed in the standard, it is assumed that the existing access protocols are sucient
to support the CR capabilities. In the implementation, however, the existing access
mechanisms are not directly applicable as the current working group of the IEEE 802.22
have suggested.
2.4 Current Trends and Applications of CR
2.4.1 CR-based Wireless Sensor Networks
econcept of CR technique has received considerable aention for its capacity to enable
opportunistic access in wireless sensor networks (WSN).e cognitive capabilities in the
sensor networks empower the sensor node with the ability to access reachable channels
opportunistically [4, 79]. is feature drastically the transmission reliability and energy-
eciency drastically of the WSN. Currently, WSN works in the industrial, scientic,
and medical (ISM) bands which are typically unlicensed bands. e services provided
through ISM bands are increasing rapidly due to the lack of licensing which apparently
makes this band overcrowded. erefore, current WSNs include CR capabilities to nd
the underutilized spectrum band that can be accessed for best eort service or bursty
trac as shown in Fig. 2.4.
e CR-based wireless sensor network (CR-WSN) is a promising paradigm for sensor
oriented services by upgrading the conventional sensor network with CR capabilities.
Most of the sensor network supports low-power and short range communication in a
2.4 Current Trends and Applications of CR 23
Figure 2.4: Network model of a proposed CR-WSN system [4].
single radio channel. Moreover, sensor nodes are densely deployed, producing large
numbers of packet bursts. In such cases, the chance of packet collision increases lead-
ing to inecient power consumption and large packet delay [4]. With CR capabilities
the sensory nodes can access the reachable radios hence the communication reliability
overcomes the shortcomings of conventional WSNs [4, 79, 80].
2.4.2 Cognitive Radio in Cellular Networks
In the RF spectrum, the 5 GHz band is an unlicensed band with 500 MHz bandwidth and
used for pure WiFi (in accordance with the IEEE 802.11a/ac/ax standard) and weather
radar applications. is unlicensed band is oen used for providing excellent data rates
for short range communication. erefore, LTE-A2operators are progressively con-
sidering the unlicensed bands as a complementary resource for achieving best eort
services in the small cell scenario [82, 83]. As depicted in Fig. 2.5, the BS of a small cell,
i.e., eNB in an LTE network is able to tune up with the carriers simultaneously both the
licensed and the unlicensed bands, where the licensed carriers in the macrocell and small
2Long Term Evolution (LTE) is a technical standard for cellular network proposed by the 3GPP
organization. e LTE-Advanced (LTE-A) is the upgraded specications of the LTE which completely
fulls the requirements set by ITU for IMT-Advanced and 4G [81]. e cellular networks adhering to the
LTE-A standard are oen referred to as LTE-A networks.
24 2.4 Current Trends and Applications of CR
Figure 2.5: Network model of a proposed CR-LTE system [5].
cell coexist. In such cases, the conventional licensed carrier (called the primary carrier)
remains fundamentals for guaranteeing its QoS to the user equipment (UE) [5].
Because of the opportunities oered by CR, the LTE cellular networks are envisioned
to support the coexistence of CR capabilities with unlicensed bands. In the proposed
model, the main data operation streams through a licensed carrier (LC) and additional
data bursts can be oered via unlicensed bands [83]. In coexisted scenario, the UE of the
LTE carrier acts like an SU and incumbent users of the unlicensed band (e.g., stations
of WiFi networks), are the PUs. Such a utilization of the unlicensed bands must be
conducted as a good neighbor of the incumbent users of the unlicensed bands [5, 84].
e 3rd Generation Partnership Project (3GPP) has initiated the standardization of
utilizing the 5 GHz unlicensed band as a secondary component carrier integrated with
the fundamental (licensed) carrier of the LTE-A networks using a novel access tech-
nology referred to as “Licensed-Assisted Access” (LAA) in the Release 13 [84]. e
European Telecommunication Standards Institute (ETSI) [85] has proposed a novel ac-
cess mechanism incorporating channel monitoring to accommodate the LTE-WiFi coex-
2.5 Spectrum Access rough MAC Protocol 25
istence in the unlicensed band.
2.5 Spectrum Access Through MAC Protocol
Medium access control (MAC) protocol provides channel access mechanisms for several
users while sharing the communication medium [66, 86]. In most of the wireless net-
works (e.g., WLAN, WSN, and WPAN) all users must be organized by MAC protocol
for ecient utilization of the shared medium. Traditional MAC protocol, however, can
interfere with the primary network and degrades the overall throughput. Robust opera-
tion of a cognitive cycle can be ensured by the proper control of the sensing and access
mechanism. A MAC protocol can facilitate the control operation on the distribution of
spectrum sensing and access in a cognitive cycle [6].
Due to the spectrum heterogeneity3, CR users require additional features in the
MAC protocol for providing interference protection to the primary network without
any internetwork collaboration. is requirement forces a drastic redenition of the
functionality of the MAC protocol for CRN. e cognitive MAC protocol (C-MAC) must
assist the secondary network to cope with the spectrum heterogeneity through a com-
patible channel access strategy for exploiting the full potential of the opportunity.
e operational eciency of the C-MAC protocol is generally inuenced by several
MAC-independent factors, such as PUs’ trac activity, tolerable interference to PU, and
hardware constraints for multi-band operation. An ecient spectrum access strategy
can aid the C-MAC protocol in improving the operational eciency. us, C-MAC
should also take into account the sensing scheduling task as well as the spectrum access
task because sensing is an obligatory task to protect the primary network. Not only
the sensing scheduling but also the sensing decision can be crucial for improving the
capabilities of the C-MAC. Spectrum access through C-MAC is a complex task due to the
dependency on sensing decisions, and there are many issues requiring research in the
area of sensing integrated MAC protocol for CRN.e following sections review factors
3CR users have their own legacy spectrum and can operate in other spectrum, depending on the
availability of the spectral resources. e phenomenon of the CR users’ operation in several spectrum is
referred to as spectrum heterogeneity.
26 2.6 Cross-Layer Components for Sensing-Assisted Access Protocol
relating to the C-MAC protocol and its potential to oer higher opportunity in operation,
current status, and challenges associated with the issues to justify the contention of this
thesis. e review provides with extensive overview on state-of-the-art of C-MAC, using
techniques of spectrum sensing and accessing with further improvement through cross-
layer operation.
2.6 Cross-Layer Components for Sensing-Assisted Ac-
cess Protocol
2.6.1 Spectrum Sensing Algorithm
is research is focused on the post-processing of the spectrum sensing data regard-
less of working beyond the detection theory. Moreover, the objective is to connect
the advantages of sensing data in the design of the access protocol to improve the CR
capabilities. To improve CR capabilities, numerous sensing algorithms are proposed
in the literature of sensing algorithm. e proposed algorithms are associated with
dierent terms and conditions. In this section, a state-of-the-art of the sensing algorithm
is presented in the context of access protocol design.
Spectrum sensing algorithm allows the post-processing of the signal detection data
from the PHY for further evaluation. In the PHY, PU detection is carried out by using the
following techniques: energy detection, matched lter, cyclostationary, etc. [2, 41]. In
particular, the received signal is synthesized according to the detection techniques, and
by comparing the processed output and pre-dened threshold value, the nal decision
is obtained. e nal decision can be either a so decision (parametric value) or a hard
decision (e.g., binary hard decision is ON or OFF). is detection decision can be further
proceed to increase the controllability of the spectrum sensing, and this is referred to as
post-processing of the detection [2, 31, 41]. In this research, it is assumed that the sensing
algorithm conducts the entire operation of the spectrum discovery for CR operation from
PU signal detection to post-processing.
Although an optimal detector applies for individual sensing, an inappropriate estima-
2.6 Cross-Layer Components for Sensing-Assisted Access Protocol 27
tion can be experienced due to channel impairments [31, 35, 87]. For instance, path loss,
multipath fading, and shadowing eects can degrade the sensing output of multiple SUs
over a certain network. In such conditions, an individual decision varies depending on
local observation which needs to be further processed for making a global and robust
decision about the channel activity. To overcome the deciency of an individual de-
cision, cooperative spectrum sensing (CSS) is proposed [31, 42, 72, 87], by sharing the
local observation over the network to make a combined decision about the PU activity.
In the CSS algorithm, all local decisions, either so or hard decisions, are taken into
the fusion machine. Based on decision rules, either AND or OR rules, a nal decision
comes out regarding the channel status. e diversity gain is imposed to overcome the
deciency of the sensing experienced by low SNR. Challenges in the implementation of
CSS algorithm include increased complexity and the requirement of an additional control
channel [2, 41]. Even though the CSS algorithm can enhance the sensing output under
channel impairments, the large overhead in the CSS algorithm makes it inecient for
access protocol design [52].
A dynamic sensing technique is proposed by [50] with the scheduling of multiple
sensing cycles before data transmission. Higher spectrum utilization is achieved by
the dynamic sensing method [50] when compared with single and static method of the
sensing [3]. e sensing period is dierentiated by lower sensitivity into multiple stages
in multi-stage sensing algorithms. e dynamic sensing or multi-stage sensing has a
great advantage in wideband sensing, as shown by [51], where an optimal sensing round
with dierent sensitivity is allocated tomaximize the throughput. e idea ofmulti-stage
sensing is narrowed down in [52, 88] by exposing the multi-threshold values in dierent
sensing stages under the constraint of primary user protection. A signicant diculty
in multi-stage sensing includes modeling of the multi-threshold value in a discrete time
scale where the channel state may changes dynamically over the given sensing period.
Sensing cycle design is an important task for conguring the transmission period in
the MAC protocol. e capability of sensing cycle design in the MAC layer is referred
to as spectrum discovery and/or spectrum search. Sensing sequence scheduling through
28 2.6 Cross-Layer Components for Sensing-Assisted Access Protocol
the MAC protocol is proposed by [34]. ey found that the overall discovery is maxim-
ized and the delay in nding idle channels is minimized by choosing an optimal sensing
cycle in [34]. e authors in [89] proposed a two-stage sensing, where the multiple
frequency-divided channels are simultaneously observed by using multiple antennas.
In the frequency dimension, sensing starts with coarse resolution sensing (CRS) and
partitions all the channels within a dened bandwidth regardless of obtaining any idle
channels. en, ne resolution sensing (FRS) takes into account that newly dened
bandwidth and continues until obtaining an idle channel. is type of sensing algorithm
is particularly proposed for spread spectrum where frequency hopping is used [89].
To overcome the sensing-throughput trade-o issue, multi-stage sensing algorithm
gains great aention in the literature of CRN [34, 51, 52, 64, 88, 90]. Nonetheless, the
resourceful ordering of the stages, weighting factors in stages, the optimal parameters
designing, and integrating with the access protocol can impact on the performance
improvement.
2.6.2 Spectrum Occupancy Modeling
According to underlying condition of CR operation, there are no cooperation and net-
work association among the PU and SU. erefore, SUs do not have exact networking
knowledge about of trac of PUs. Hence, SUs can only gain knowledge about PU trac
by the spectrum measurement and this has been studied extensively in the current
literature [23, 26, 33, 91–93]. Experimental measurements suggest that the spectrum
occupancy can be modeled using certain statistical and/or mathematical models [33, 91–
93]. Spectrum occupancy modeling is important for determining the full potential of
the spectrum opportunities before accessing the spectrum. Moreover, the accuracy of
the spectrum sensing can be evaluated with the help of knowledge of the spectrum
occupancy; hence, interference protection to PUs can be designed deliberately. For
instance, interference to PU is minimized with the sensing parameters optimizations
achieved by using the dynamic trac model of the PU [94]. Without paying aention
to the approach, the statistical model of the spectrum occupancy is comprehensively
2.6 Cross-Layer Components for Sensing-Assisted Access Protocol 29
equipped to envision the CR operation [48, 49, 57, 64, 65] but of barely sucient accuracy
to characterize the PU activity.
e statistical model is extracted from the measurement data for CR designs obtained
from measurement campaigns. e most popular and natural choice for statistical mod-
eling of the spectrum occupancy is Markov chain model. In the current literature, the
following Markov chain based models are found widely used: continuous time Markov
chain (CTMC), continuous time semi-Markov chain (CTSMC) [33, 91], discrete time
Markov chain (DTMC) [92], and heuristic model [93].
In the Markov chain model, the state of the spectrum is dened as a random process
that switches between several possible states, and eventually, the spectrum occupancy
rate can be characterized by the transition probabilities. Based on the characteristics
of the random process and its post-processing, the given models [33, 91–93] can be
distinguished. In CTMC and CTSMC models, the spectrum state is characterized by the
holding time or sojourn time where the holding time follows an exponential distribution
and an arbitrary distribution, respectively. On the other hand, the DTMC model does
not allow the channel or spectrum state to stay on any of the states; hence, distribution
of the holding time is applicable in the occupancy modeling.
Early works found that the CTMCmodel is widely used in modeling the occupancy of
high-frequency bands. Several measurement campaigns revealed that the CTSMC has
beer accuracy than the CTMC, especially in the modeling of the WLAN trac over
2.4 GHz band with an approximation model. ence, the CTSMC became a popular
choice for occupancy modeling at the early stage of the development of CR technology
[33, 91]. e study in [33] suggested a generalized Pareto distribution was suitable for
dierent frequency bands when the sampling rate was relatively low. e simplifying
Markovian assumption, (i.e., semi-Markov) was, however, insuciently accurate for the
all other radio trac regimes due to the large approximation of the measurement data as
suggested by [92]. Unlike the standardMarkov chain model [33, 91], an empirical DTMC
model was also suitable for occupancy modeling where spectrum states are assumed as
continuously switching. To accelerate the dynamic switching in the discrete-time, the
30 2.6 Cross-Layer Components for Sensing-Assisted Access Protocol
transitions probabilities were expressed as the functions of time. By applying both the
deterministic and stochastic methods, transitions probabilities were characterized from
the measurement data with the perfect agreement between the empirical curves and
ed curves.
e Markov chain-based model is quite simple but largely acceptable statistical ap-
proach for the spectrum occupancy modeling in time dimension. However, in addi-
tion to the time dimension, the spectrum occupancy can also be modeled by space and
frequency dimension. e shortcoming of Markov chain-based modeling is that the
spectrum occupancy in the space and frequency dimension.
2.6.3 Data Transmission Mechanism
Inspired by the success of random access technique, several data transmission protocols
have directly adopted the random access techniques, such as sloed ALOHA, CSMA/CA,
in the CRN [7, 63, 68, 95]. For the multiple access scenario in CRN, two types of users
with dierent prioritized access in the channel have been considered in [7, 63, 68]. For
multiple access in the primary channel among multiple SUs, the existing access protocol
such as sloed ALOHA [7] and CSMA/CA [6, 63, 64] are adopted for the cognitive radio
scenario. In the given literature, two types of users with dierent prioritized access
were considered. e CSMA/CA has an advantage over sloed ALOHA for CRN, as
the CSMA/CA allows channel monitoring functionality before transmission which is
essential for occupying the primary channel. Time-division multiple access (TDMA) is
also used in CR with a cooperative MAC protocol as proposed by [96, 97]. Nevertheless,
without any inter-network collaboration and/or precise synchronization, the TDMA
approach cannot guarantee sucient protection to the primary network.
e data transmission proposed in [7, 68] is based on a two-level access policy, where
the interference protection to the PU and sensing time optimization is done at the rst
level, and packet scheduling based on the MAC protocol is enabled at the second level.
In particular, two dierent MAC protocols, i.e., CR-ALOHA and CR-CSMA mechanisms
are used for the packet scheduling. e main limitation of this model is that if missed
2.6 Cross-Layer Components for Sensing-Assisted Access Protocol 31
detection occurred at the rst level, then the CR-ALOHA and the CR-CSMA cannot
provide a sucient interference protection to the PU as they did not have any collision
avoidance procedure during the access period. On the other hand, this limitation has
been aempted to overcome in their other works [63] by introducing a new control
packet named prepare-to-send (PTS) with ready to send (RTS)/ clear to send (CTS) mech-
anism during the channel access. However, the carrier sensing is performed before the
spectrum sensing so that the detection performance the channel is comparatively poor,
which can impact negatively on the PU’s protection.
Traditionally, MAC access protocols do not take into account spectrum sensing when
designing access strategies, which leaves the PUs potentially open to severe interference.
A decentralized cognitive MAC protocol [48] rst allows for spectrum sensing where
access is enhanced by compensating for a higher probability of false alarmwhile keeping
the sensing period unchanged. An aempt at improving throughput by considering
both sensing and access was made in[64] based on the IEEE 802.11 distributed coordin-
ated function (DCF) [66]. Even though all the proposed access methods based on the
CSMA/CA can improve the throughput, they cannot guarantee sucient interference
protection to the PU. is has occurred because conceptually, the PU is not incorpor-
ated in the backo mechanism with the CRN and the fundamental spectrum sensing is
omied while proposing the access protocol in the existing literature.
In the other networks that consider CR capabilities also rely on the Listen-Before-Talk
(LBT) mechanism. Such an mechanism is proposed for LTE architecture. is proposal
exploited the coexistence of the LTE and WiFi networks in unlicensed 5 GHz band and
has mainly been adopted by a radio access technology called carrier-sense adaptation
transmission (CSAT), as proposed by alcomm [98]. To accommodate the LTE-WiFi
coexistence in the unlicensed band, the ETSI has developed a frame-based equipment
(FBE) scheme which is quite similar to the CSAT scheme [85]. ere has been a fast
uptake of the LTE globally, and the LTE-A is a hugely successful platform in terms of
its widespread adoption for 5G deployment as well as meeting the recent demand. At
the same time, usage of the unlicensed band needs to comply with certain regulations
32 2.7 Sensing-Transmission Optimization
in several regions in the world, for instance, LBT mechanism must follow to use the
unlicensed band in Japan and Europe. us, to enable the CR capabilities in real-world,
the data transmission should be associated with the spectrum sensing.
2.7 Sensing-Transmission Optimization
e sensing-transmission optimization is formulated explicitly in [3, 53] and proved
the existence of optimal sensing period that could maximize the throughput under the
constraint of target PD. e eect of PU trac on the sensing-throughput trade-o
has been investigated in [61, 62]. Moreover, the impacts of the fading and noise vari-
ance in channel propagation on this trade-o problem are demonstrated in [60] and
[53] respectively. e channel degradation could be overcome by using cooperative
spectrum sensing (CSS), however, there is an additional trade-o between cooperative
overhead and gain. e trade-o is conducted in [87] by allocating optimal number of
SUs in cooperative detection to meet the target PD to maximize the throughput within
shorter sensing period; also the maximum throughput obtained in [87] is larger than the
throughput achieved in [3, 53, 60].
By exerting interleaved transmission with periodic sensing, the authors in [99] pro-
posed an opportunistic channel-aware access to reduce the sensing error. Despite lever-
age the sensing purpose by the periodic sensing, the interleaved transmission imposes
large overhead when SU transmits a large data packet. Furthermore, for a stable oper-
ation and small delay tolerance, the access scheme should incorporate with the robust
sensing policy as suggested in [100], which presents a cross-layer (PHY/MAC) approach
to clarify the eect of sensing in access scheme for maximizing the throughput under a
PU’s stability constraints. Moreover, the authors in [59] enhanced the throughput per-
formance with simultaneous sensing and transmission like full-duplex mode by forming
a new frame structure. However, full-duplex adaptation in cognitive radio network is
still a challenging task as stated in [101].
e access strategy in the perspective of sensing-throughput trade-o problem has
been investigated in [6, 102]. In [6], a distributed MAC protocol is designed similarly
2.8 Model of Access Protocols Based on Cross-layer Design 33
with the control channel operation of [54] and the throughput is optimized in terms of
sensing period and contentionwindow. However, their optimization is almost the simple
form of sensing-throughput trade-o problem [3] as physically the contention window
is formulated linearly in the time slot which is nothing but the same scale of sensing
period. Based upon only the resolution of access contention, the authors re-established
the distributedMAC protocol of the [6] by an overlapping channel assignment algorithm
where the secondary users use the interference avoidance approach during the packet
transmission similar with the CSMA/CA approach. However, the improvements of [6,
102] are largely dependent on the control channel and the synchronization which are
still a burden for designing the access protocol in the CRN as suggested by [7, 63].
To ll up the above-mentioned research gap, a new framework for improving
throughput in CSMA/CA by “restructuring” the sensing period to meet the target
probability of detection is proposed in [103]. e proposed protocol was referred to
as dual-level sensing based multiple access (DSMA) where the spectrum sensing is
accompanied with carrier sensing to decide the channel status jointly. We illustrated
that the throughput improvement mostly dependent on how much the overall PFA can
be reducible. On the other hand, interference protection is enhanced by addressing the
contention access method aer the nishing of the CR sensing period. Moreover, the
PU protection can be controlled by choosing the suitable threshold values into two steps
to meet the target PD that may also vary the overall PFA; which also arises the sensing-
throughput trade-o problem. erefore, it is of great importance to nd the impact
of the dual-level sensing on the sensing-throughput optimization problem. Motivated
with this importance, we will study the feasibility analysis of the optimization problem
and propose an algorithm to maximize the throughput of the DSMA protocol under the
constraint of the PU protection.
34 2.8 Model of Access Protocols Based on Cross-layer Design
BackoffF
S
S
F
S
SData
(d)
BackoffF
S
S
C
S
S
Data(b)
BackoffF
S
S
P
T
S
Data(c)
C
S
S
F
S
S
C
S
S
C
S
SData
(a)
Figure 2.6: Review of frame format with sensing-transmission mechanism.
2.8 Model of Access Protocols Based on Cross-layer
Design
According to the fundamental principle of cognitive radio operation, SUs are only al-
lowed to transmit data while the channel is sensed as idle. To adhere to this principle,
SUs have usually employed the LBT [28] mechanism in which spectrum sensing fulls
the listening function at the PHY, and the transmission function refers to the packet
scheduling at the MAC layer. In IEEE 802.22 standard [78], the MAC protocol allows
sensing-transmission combination in an operating frame, where two types of periodic
sensing are proposed: coarse spectrum sensing (CSS) and ne spectrum sensing (FSS) as
shown in Fig. 2.6(a). e objectives of CSS and FSS are to identify vacant spectrum with
a shorter sensing period and to support the previous sensing algorithm with a longer
sensing period, respectively [88]. For a fair comparison, it is assumed that the frame used
in [49, 63, 65, 78, 88] followed a similar format of two-stage physical sensing and backo
period as shown in Fig. 2.6. Note that individuals’ detection outcomes and contention
access impacted on the achievable throughput, which was dierent, even though the
same time duration is reserved for the data transmission.
Unlike [78, 88], a conventional backo mechanism (Fig. 2.6(b)) is applied between
2.8 Model of Access Protocols Based on Cross-layer Design 35
the two stages [49] to enable the CSMA mechanism to cope with two-stage sensing.
Although imperfect sensing was considered for two xed sensing stages, a conventional
backo mechanism [66] with perfect detection was adopted which led to a burden on
the second sensing in making the nal decision. e authors in [63] overcome the
shortcomings of [49] by introducing a conventional backo process [66] at the start
of the frame, as illustrated in Fig. 2.6(c). Despite leveraging the sensing purpose by two-
stage detection, the proposed protocol in [63] imposed an overhead by introducing a new
control packet, prepare-to-sense (PTS), between backo process and the FSS. In contrast,
a relatively robust sensing mechanism is used in [65] compared to [49, 63] by allowing
two FSS operation consecutively before the backo mechanism, as shown in Fig. 2.6(d),
for enhancing the spectrum opportunity (by reducing the probability of false alarm).
However, the access protocol in [65] causes severe interference to the primary users as
perfect detection is also assumed during the backo process. All the MAC protocols
mentioned above have signicant outcomes in conict with the IEEE 802.22 [78]; how-
ever, the access protocol can be more ecient and practical if the sensing aspects can be
exploited during the backo process [64, 100, 104]. To analyze the impact of the sensing
error, the authors in [104] included the sensing error cases in the backo mechanism
of the CSMA/CA protocol which is not thoroughly examined for the cognitive radio
environments.
e main challenge in the sensing-assisted MAC protocol is to reveal the cross-layer
eectiveness towards achieving the goals of the CRNs, such as improving spectral e-
ciency without producing severe interference. To expose the importance of sensing on
access protocol, the sensing parameters is integrated with the contention window for
improving the throughput and delay performance in the proposed model.
Owing to the advantages of sensing-assisted MAC protocols for throughput maxim-
ization in a multiple access scenario, [54] and [65] integrated the spectrum sensing with
the channel assignment on the MAC protocol. In [54], SUs access the channel through
theMAC protocol which uses clear channel assessment (CCA) functionality to detect the
transmission in a channel. In [54], the channel assessment is done through two phases,
36 2.8 Model of Access Protocols Based on Cross-layer Design
the reporting phase and the negotiating phase, in an additional control channel. In the
reporting phase, SUs do the spectrum sensing and report the acquired information. In
the negotiation phase, a p-persistent based access protocol is applied to contend for the
transmission in the next frame. However, a dedicated control channel may not always
be available in practice, and also the consideration of the CSI of an additional channel
may increase the computational complexity. e authors [65] proposed a contention
access strategy by sensing two channels sequentially in a single slot duration. However,
the authors in [65] did not consider the detection operation in the contention duration
(where detection usually occurs by carrier sensing in the contention window of the
CSMA/CA mechanism) as the contention window is assumed to be short compared
with sensing duration. Nevertheless, taking a larger threshold value for the detector
during short contentionwindow [63] can also play the same role as the spectrum sensing
does for the primary user detection in the CRN. In that case, single channel sensing
including the carrier sensing during the contention window can improve the detection
performance, instead of the two channels sensing with the exclusion of carrier sensing
in the contention period.
Multiple access in cognitive radio can be enabled by enhancing the PHY-MAC jointly
[33, 49, 54, 95]. e enhancement requires the conventional access exhibited in a distinct
network where the trac dynamics of the user is identical throughout the transmission
time. Since SU measures the energy level of PU’s transmission before accessing, the
SU does not acquire exact knowledge about the PU’s trac dynamics [49, 95, 105]. A
decentralizedMACprotocol has been proposed based on the partially observableMarkov
decision processes (POMDPs) framework to overcome the absence of any central entity
in [48]. Although the MAC protocol proposed in [33, 48, 54] can increase the SU’s
performance while inhibiting the interference experienced by the PUs, it requires exact
information on the PU trac. Moreover, the proposed model in [33, 48, 54] consumed
most of the resources, such as time duration and energy for that collaboration which
reduced the eciency of the CRN. On the other hand, the proposed PHY/MAC cross-
layer in [54, 64] based opportunistic MAC protocols for provisioning the QoS in CRN,
2.9 Chapter Summary 37
but this is costly as it requires additional control channel operation for the negotiation
based mechanism. So far, all the proposed models related to the cross-layer approach
[33, 49, 54, 64, 95] have not directly addressed the measurement and eectiveness of
spectrum sensing when multiple SU access the licensed spectrum.
In this work, the focus is maintained on the sensing-assisted MAC protocol where,
contrary to [6, 54, 63], neither an additional control channel operation is considered
nor the carrier sensing is omied over the contention access period. Instead of direct
adoption of the CSMA/CA protocol into the CRN [63], we propose the dual-level sensing
within the same sensing period (used in [3, 6, 65]) and exploit it into the CSMA/CA-
based access mechanism. e impact of the entire sensing heterogeneity in improving
the throughput of the DSMA scheme has been illustrated in [103]. However, the optim-
ization of the DSMA protocol is required to accomplish the sensing-throughput trade-o
issue. An investigation is conducted over this optimization problem to nd a solution
framework.
2.9 Chapter Summary
is chapter has given an operational overview of the cognitive radio and some applica-
tions of CR technology. e CR operation is comprised by multiple tasks to conrm pro-
cient communication and incumbent protection. Such multiple tasks are accomplished
by the cognitive cycle. e cognitive cycle is empowered by three building blocks:
spectrum sensing, spectrum analysis, and spectrum decision. e CR technology shows
huge opportunities in ecient spectrum utilization of the future generation network.
erefore, IEEE proposed a new standard for CR capabilities of the wireless devices
that can tune up in the TV band in WRAN environment. In addition, CR technology is
adopted in other networks such as in sensor networks, and cellular networks. e design
aspects of the SU’s transmission protocol associated with spectrum sensing identied
in the above literature review. Also the review illustrates the necessity of cross-layer
design for developing a sensing-assisted access protocol. According to the cognitive
cycle, spectrum sensing is one of the key enablers of the CR operation that should be
38 2.9 Chapter Summary
considered explicitly in the design of a complete access protocol. erefore, the impact of
sensing on the design of the access protocol is thoroughly reviewed in the next chapter,
with their applications and technical challenges.
CHAPTER 3
Impact of Spectrum Sensing on theCapacity Measurement of Spectrum
Opportunity
3.1 Introduction
Spectrum sensing and spectrum access are two key components of a “cognitive cycle”
[1]. In the context of CR operation, sensing and access contribute to the discovery
of spectrum opportunity and the proper utilization of that opportunity, respectively.
rough the spectrum sensing, SUs obtain the occupancy status of the channel which
also aids to measure the capacity of the spectrum opportunity. Before transmiing any
data into the vacant channel, it is also essential to know the oered capacity of the
opportunity for designing an ecient access protocol. By identifying the oered capa-
city, which is determined by spectrum sensing, the access mechanism can congure its
transmission policy to achieve the maximum utilization of the oered capacity. Studies
[3, 34–37] have shown that the utilization of the opportunity without causing harmful
interference to the primary network are related to spectrum sensing. In this chapter,
therefore, a comprehensive analysis is conducted regarding the impact of sensing on the
measurement of the capacity of the spectrum opportunity, before proposing the access
protocol.
Spectrum sensing takes place at the physical layer (PHY) and has been extensively
studied, with numerous sensing techniques being developed for interference reduction
[1, 3, 30]. Spectrum access takes place at the medium access control (MAC) layer and
traditionally the sensing aspect if not taken into account when designing the data trans-
40 3.1 Introduction
mission strategies. Due to the dierence between traditional wireless networks and
cognitive radio network regarding the sensing before data transmission, the SU needs
to take into account the sensing aspect in designing data transmission strategies by
assessing the capacity of the spectrum opportunity [34, 48, 49].
In a single frame, the SU performs spectrum sensing and then decides on the data
transmission based on the sensing decision. e sensing period is designed to meet the
PD requirements set by the interference protection for the PU. Investigation in CRN [3]
indicates that a longer sensing period provides greater interference protection to the PU
and consequently reduces the transmission time of the SU. As a result, the SU cannot
achieve enough throughput to maintain the QoS by using that shorter transmission time.
To obtain a longer transmission period, the SU needs to perform the sensing within a
shorter period. On the other hand, a shorter sensing period causes larger PFA which
eventually reduces the spectrum opportunity. Hence, reduction of sensing time cannot
be a straightforward solution for increasing the throughput. is issue is referred to as
the sensing-throughput trade-o and cannot be overcome adequately by using single-
level sensing where typically a single threshold is used in determining the spectrum
occupancy [3, 7, 52, 57].
In practice, the PU is protected by designing a higher detection probability by which
themissed detection can be kept within a tolerable range. Studies [3, 50] indicate that the
spectral eciency of the single-level sensing (SS) mechanism is relatively inadequate for
a higher target probability of detection. To the best of our knowledge, the SS method is
unable to exploit the sensing period eciently towards improving the spectral eciency.
erefore, a dual-level sensing (DS) is proposed where two sensing levels are employed
conditionally during the sensing operation to determine the channel status jointly. e
contribution of the DS mechanism to the gain of higher spectrum opportunity is em-
phasized by the measurement of access probability. rough mathematical derivations,
it is proven that by allowing a section of the sensing period to be devoted to reducing
the probability of false alarm, the overall probability of detection is still met while the
access probability is improved.
3.2 System Model 41
time
. . . .
sτ
f sT τ−
fT
Spectrum AccessSpectrum Sensing
Frame 2 Frame 3Frame 1. . . .
Figure 3.1: Frame structure for CR operation with spectrum sensing andaccess in every frame.
3.2 System Model
In considering the CRN, SUs are allowed to access a single time-sloed channel only
while no PU is present in the channel. ere is no cooperation between the SU and
PU, and PU transmission is not aware of the SU transmission. e SU follows a frame
structure which consists of a sensing period and an access period as shown in Fig. 3.1.
Since the PU and SU are non-cooperative, SU performs spectrum sensing at the starting
of each frame. en, access operation comes into account followed by the sensing.
During the sensing operation, SUs are allowed to employ a signal detection method
to determine the channel status by comparing the received signal with a predened
threshold value.
3.2.1 Spectrum Sensing Model
Let y(m) denote the received signal to the secondary user for primary user detection
over τs period with sampling frequency fs, where m is the sampling index and total
number of samplingM = bfsτsc. Applying a binary hypothesis-testing problem [1, 3],
the detection process can be modelled as,
H0 : y (m) = w (m) (3.1)
H1 : y (m) = s (m) + w (m) (3.2)
42 3.2 System Model
where s(m) and w(m) denote the transmied signal and the additive white Gaussian
noise (AWGN) respectively. Assume that, s(m) and w(m) are independent and identic-
ally distributed (iid) random process with both having the mean zero, and variance σ2s
and σ2w respectively. HypothesisH0 andH1 describe the absence and presence of the PU
signal, respectively. e measured signal-to-noise ratio (SNR) under the H1 hypothesis
is γ = σ2s/σ
2w.
Aer the post-processing of the received signal through a specic detector, such as
energy detector (ED), matched-lter (MF), the generated outcome is called test stat-
istic which is denoted by Y (y). e test statistic Y (y) is compared with a predened
threshold ε to obtain the nal detection decision about the channel occupancy, under
the hypothesis of H0 and H1. e performance of the detection is evaluated with the
following metrics:
• Probability of detection (Pd): the probability of deciding the PU signal is present
whileH1 is true, which can be determined by Pd = P Y > ε|H1.
• Probability of false alarm (Pf ): the probability of deciding the PU signal is present
when H0 is true, i.e., Pf = P Y > ε|H0. In the context of CRN, false alarm is
treated as a sensing error which means that the spectrum holes is not detected
even though there is a spectrum hole. us, a lower Pf is desirable to obtain large
spectrum opportunity for the SUs.
• Probability of missed-detection (Pm): the probability of deciding the PU signal is
absent whenH1 is true, i.e., Pm = P Y < ε|H1, and thus, Pm = 1− Pd. is is
also a sensing error where lower value ofPm is desirable to reduce the interference
to the PUs.
Energy Detector
Let us apply an energy detector (ED) in spectrum sensing. In the energy detector, the
received signal y(m) is ltered through a bandpass lter with the bandwidth ofBW , then
the ltered output is squared and integrated overM samples to produce a test statistic
3.2 System Model 43
Y (y). us, the test static of the energy detector is given by
Y (y) =1
M
M∑m=1
|y (m)|2 (3.3)
Let us assume that the transmied signal in the channel is a complex-valued PSK
modulated signal and the noise is circularly symmetric and complex Gaussian (CSCG)
signal. For a largeM , the distribution of the test statistic is obtained as follows [1, 3, 43],
H0 : Y ∼ N(σ2w,σ4w
M
)(3.4)
H1 : Y ∼ N(
(γ + 1)σ2w, (2γ + 1)
σ4w
M
)(3.5)
whereN indicates the normal distribution. By using central limit theorem (CLT) for the
large number of samples, the performance metrics are expressed as follows [3, 64, 65],
Pf (ε) = Q
((ε
σ2w
− 1
)√M
)(3.6)
Pd(ε) = Q
((ε
σ2w
− γ − 1
)√M
2γ + 1
)(3.7)
where Q(.) is a complementary distribution of standard Gaussian i.e.,
Q (x) =1√2π
ˆ ∞x
e(−t2/2)dt (3.8)
3.2.2 PU Activity Model
Studies [48, 49, 57, 64, 65] suggest that PU trac modeling is well ed by a two-
state ON-OFF process where ON and OFF states respectively indicate busy and idle
activity of the primary user. is trac model was authenticated with experimental
results [33, 91] for modeling the PU’s activity in wireless local area network (WLAN).
We assume that PU activity consists of idle period interspersed with a busy period in a
frame of a single channel scenario. By following [48, 65], we also assume that the sojourn
periods (or holding times) in idle and busy states are ti and tb. e arrival of PU trac
are denitely validated those models for solving the optimization problem. Furthermore,
we can see that by increasing the Pd up to 0.99 the maximum achievable R is 0.75
which is theoretically true as proven in [3, 50, 59]. From this outcomes, the signicance
of DLS-based access protocol can be drawn. Our proposed DLS-based access protocol
[103] with above conducted optimization is achieved greater throughput performance
than the performance achieved by [3, 50, 64, 65] for relatively higher target probability
of detection which can also provide with stronger interference protection to primary
network.
80 4.5 Numerical Results and Discussion
0 0.2 0.4 0.6 0.8 10.55
0.6
0.65
0.7
0.75
0.8
Probabability of detection at first sensing (Pd1)
Norm
alizedaggregatedthroughput(R
)
For Pd = 0.9
For Pd = 0.95
For Pd = 0.99
Figure 4.6: Characterization of the change of R corresponding to τds(τs) forPd = 0.9, 0.95, 0.99 and its optimal sensing period as givenby Table 4.1, where the simulation parameters are, γ = −15 dB,Pi = 0.9, P aMAC = 0.99, σ2
w = 1, fs = 6 MHz, and Tf = 10 ms.
4.5.2 Performance Evaluation of DLS Based Access with Post-
optimization
Fig. 4.2 already showed that the change of overall PFA with respect to Pd1 which implied
that a minimum value of Pf could be found for the optimal value of Pd1 . Likewise, the
variation of the normalized throughput is contained a certain maximum region with
respect to Pd1 as depicted in Fig. 4.6. For a given Pd, the normalized throughput starts
to increase with the increment of Pd1 and starts to decline aer reaching its maximum
region. As Pd1 was the subset in the space of (0, Pd) so that R cannot be measured for
Pd1 ≥ Pd. Most importantly, this gure shows the potential of the DLS based access
protocol, i.e., the maximum value of throughput laid on the value of Pd1 that is less than
the target Pd. By segmenting the sensing process and seing the Pd1 less than the Pd, we
reduced mainly the sensitivity of the rst detection process which conversely reduced
the probability of false alarm and ultimately maximized the normalized throughput.
Fig. 4.7 illustrates the normalized aggregated throughput versus the total sensing
4.5 Numerical Results and Discussion 81
0 0.5 1 1.5 2 2.5 3
x 10−3
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Total sensing period (τds(τs)) (sec)
Norm
alizedaggregatedthroughput(R
)
For Pd = 0.90, P ∗
d1= 0.68
For Pd = 0.95, P ∗
d1= 0.78
For Pd = 0.99, P ∗
d1= 0.90
Figure 4.7: Characterization of the change of R corresponding to τds(τs) forPd = 0.9, 0.95, 0.99 and its optimal Pd1 , where the simulationparameters are, γ = −15 dB, Pi = 0.9, P aMAC = 0.99, σ2
w = 1,fs = 6 MHz, and Tf = 10 ms.
time required in each frame of the secondary network. e results are extracted from
the analytical model of the DLS based access strategy using equation (4.8) aer obtaining
the optimal Pd1 . is gure reveals that the maximum throughput can be achieved for
a certain optimal sensing period and the throughput will be decreased linearly with the
sensing period aer a while of that optimal sensing time. Moreover, it is seen that the
higher throughput regime is obtained in lower sensing period for the case of lower target
Pd. For example, the normalized throughput closed to 0.8 at the sensing period of 1 ms
for Pd = 0.9 which is larger than the Pd = 0.99 case. is nding follows the actual
character of the throughput versus sensing period as proven by [3, 58, 59] for the CRN
and consolidates the Proposition 6. However, lower value of Pd e.g., Pd = 0.9 found by
seing Pd1 = 0.68 which leads severe collision during channel accessing. On the other
hand, by seing Pd1 = 0.9 for Pd = 0.99, the interference protection can be improved as
well as optimal throughput can be achieved similar with the throughput of the Pd = 0.9
case. In overall, the signicance of the DLS based access protocol can be exposed as
the SU can achieved near about the maximum throughput without producing too much
82 4.5 Numerical Results and Discussion
0 0.5 1 1.5 2 2.5 3
x 10−3
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Total sensing period (τds(τs)) (sec)
Norm
alizedaggregatedthroughput(R
)
RDLS for Pd = 0.9
RDLS for Pd = 0.95
RDLS for Pd = 0.99
RSLS for Pd = 0.9
RSLS for Pd = 0.95
RSLS for Pd = 0.99
Figure 4.8: Throughput performance comparison of the proposed DLSand the conventional SLS based access mechanism for Pd =0.9, 0.95, 0.99, where the simulation parameters are, γ = −15dB, Pi = 0.9, P aMAC = 0.99, σ2
w = 1, fs = 6 MHz, and Tf = 10ms.
interference to primary network within a shorter sensing period.
Fig. 4.8 shows that proposed DLS based access mechanism achieves higher through-
put than the conventional SLS based access mechanism in the lower sensing period at
a certain target Pd. For example, the DLS system is maintained near about 0.8 of the
normalized throughput which cannot be achieved by the SLS system at 1 ms period
while the Pd are 0.9 and 0.95. Also, the dierence of the throughput between the DLS
and the SLS system keeps increasing while the value of the Pd increases at the region of
lower sensing period. Moreover, RDLS for Pd = 0.99 is quite same with the RSLS for
Pd = 0.9. is indicates that the proposed mechanism achieved higher throughput at a
given target Pd and outperformed the conventional SLS mechanism with a large margin
while the target Pd is increased for limiting interference to primary network.
At low SNR value, the proposedDLS based access scheme achieved higher throughput
than the conventional SLS method for any given Pd as illustrated in Fig. 4.9. R in
comparing both systems are maintained quite same characteristic when γ = −10 dB
4.6 Chapter Summary 83
0 0.5 1 1.5 2 2.5 3
x 10−3
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Total sensing period (sec)
Norm
alizedaggregatedthroughput
RDLS for γ = −10 dB
RDLS for γ = −15 dB
RDLS for γ = −20 dB
RSLS for γ = −10 dB
RSLS for γ = −15 dB
RSLS for γ = −20 dB
Figure 4.9: Throughput performance comparison of the proposed DLSand the conventional SLS based access mechanism for γ =−10,−15,−20 dB, where the simulation parameters are,Pd = 0.95, Pi = 0.9, P aMAC = 0.99, σ2
w = 1, fs = 6 MHz,and Tf = 10 ms.
with the increasing of sensing period. Nevertheless, the DLS system showed beer
throughput performance while the SNR value keeps decreasing. For γ = −15 dB, the
maximum R achieved by the DLS system is more than 0.75 which is apparently not
achievable by the SLS system. is analysis depicted that proposed DLS based access
protocol showed excellent throughput performance than the SLS systemwhile spectrum
sensing is impaired by channel’s low SNR value.
4.6 Chapter Summary
is chapter provides the solution of the sensing-throughput optimization problem of
the DLS-based access mechanism. e main goal of the optimization is throughput
maximization of the SU when the DLS mechanism is employed in spectrum access under
the constraint of interference protection to the primary network. e main hurdle of
sensing-throughput trade-o is to detect the active user in the channel with a single
84 4.6 Chapter Summary
step including a single-target PD. e proposed DLS mechanism overcomes this issue.
However, the detection sensitivity such as a target PD in each level with required sensing
time is a crucial design aspect for the deployment of the DLS mechanism. e conducted
optimization provides the solution for this design aspect of the DLS mechanism.
e solution approach to the optimization problem is formulated over three steps.
In the rst step, the PFA is minimized regarding the PD at the rst sensing stage while
other system parameters remain unchanged. Before this minimization, the feasibility of
minimum PFA is analyzed by convex analysis. From the feasibility analysis, the optimal
boundary of the PD at rst sensing is obtained. is PFA minimization ensures the
signicance of the DLS mechanism as the throughput improvement is hugely dependent
on theminimized amount of PFA from the sensing. In addition, the required sensing time
to obtain a decreased PFA also impacts on the throughput performance. erefore, in
the second step, the optimal boundary of the sensing period for a unique value of the
maximum throughput is identied with the analytical model. Due to the mathematical
complexity in computing the Hessian matrix of the objective function regarding the
optimizer, the solution is accomplished with a proposed semi-analytical algorithm. At
the last step, the feasible boundary of the optimizer is employed in the proposed al-
gorithm to obtain the maximum throughput. For fair comparison and model validation,
the proposed algorithm is comparedwith a purely numerical approach. e performance
analysis indicates that the solution algorithms can enhance the throughput performance
optimally under the constraint of PU protectionwithin a limited computational complex-
ity.
CHAPTER 5
Sensing Assisted Multiple AccessStrategy in Cognitive Radio Networks
5.1 Introduction
Multiple access is an essential functionality to enable secondary users to access un-
used spectrum while improving the overall throughput performance of a cognitive radio
network. However, traditional medium access control (MAC) protocols can interfere
with primary users’ transmission and degrades the secondary users’ throughput. In this
chapter, a multiple access protocol is proposed by a PHY/MAC cross-layer design. e
proposed protocol is referred to as a dual level sensing based multiple access (DSMA),
where the sensing is integrated with the MAC-based transmission. An analytical model
of the proposed DSMA mechanism is developed by Markov chain analysis to estimate
the average service time and the normalized throughput. Performance analysis shows
that the proposed scheme improves throughput signicantly when multiple access takes
place under the large sensing error and low signal-to-noise (SNR) conditions.
5.1.1 Motivation
A MAC protocol facilitates the control operation of the spectrum access to execute
the transmission timing. Few studies [34, 49] showed that the MAC can govern the
distribution of sensing task over a certain operational period. Besides, MAC protocol
allows multiple secondary users to access the channel with a higher rate of utilization
through a multiple access functionality [6, 57, 63]. On the other hand, its deployment
consumes the spectrum opportunity for its control operation. e full capacity of the
86 5.1 Introduction
spectrum opportunity is not eciently utilized [36]. As a result, SU cannot achieve
higher throughput. Moreover, the throughput can be reduced due to the waste of spec-
trum opportunity and the packet collision [102]. When the false alarm occurs during
the spectrum sensing, the secondary users lose the access opportunities even though the
spectrum is vacant. On the other hand, missed detection leads to collision eect during
the channel access. If missed detection occurs, not only the throughput is reduced but
also the activity of the PU is interrupted.
In cognitive radio, the access protocol denes its operational mechanism associating
with the spectrum sensing [54, 64]. Without the additional sensing policy, only the MAC
protocol cannot reduce the collision eect occurring due to the missed detection. e
periodical sensing before any data transmission during the access period can reduce the
likelihood of collisions [6, 50, 57, 63]. e purposes of the spectrum sensing are not
only to nd the access opportunity but also to scaling the interference protection to the
PU [3, 37]. Motivated by the above facts, a novel MAC protocol is aimed to develop
in this chapter to increase the throughput concurrently with guaranteeing a strongest
interference protection.
5.1.2 Contribution
To overcome the challenges as mentioned above, the access decision is incorporated
with a dual-level sensing by the PHY/MAC cross-layer design. At the start of the MAC
operation, the full capacity of the spectrum opportunity is determined explicitly by using
energy detection based spectrum sensing. Aer the rst sensing, SUs can access the
channel following the second sensing step which is conditional on the rst sensing
decision. e second sensing is referred to as clear channel assessment (CCA) due to
the compatibility with the existing MAC protocols.
Spectrum access comes to the operation depending on the outcomes of the cross-
layer based detection decision and the backo process. e CCA is somewhat related
and included into the backo process [57, 63, 66, 67]. However, a separated design is
proposed here to change the sensitivity of the target detection according to backo delay.
5.2 System Model 87
In particular, the advantages of backo mechanism towards collision reduction during
multiple access has contributed in scaling the detector. As a result, the detection process
achieves larger opportunity with larger collision probability. By using the contention
process and the RTS/CTS based packet transmission mechanism, the overall collision
eect is reduced during the channel access. A comprehensive analysis is presented to
demonstrate the enhanced detection and the throughput performance at various channel
conditions.
e main contributions of this chapter are as follows:
• Firstly, the DSMA mechanism is proposed by a PHY/MAC cross-layer design
which utilizes the spectrum opportunity and reduces the collision eect concur-
rently.
• Secondly, an analytical model of the DSMA is developed by exploiting Markov
chain analysis, and further extended the packet service process in amultiple access
operation to compute the normalized throughput.
• irdly, a comprehensive assessment is carried out by the model validation and
performance comparison which implies that the proposed mechanism provides
stable throughput regime with a short sensing time and low SNR values.
5.2 System Model
5.2.1 Network Entity
Let us considerN number of SUs are randomly distributed in a cognitive radio network
with the capability of multiple access as shown in Fig. 5.1. SUs use a single time-
sloed channel for data transmission where PU has prioritized to access the channel,
thereby SU can access the channel only while PU is detected as idle. rough spec-
trum sensing mechanism, SUs decide whether PU is idle or busy in the channel. In
our model, upstream transmission from multiple SU to a secondary base station, is
considered through single-hop communication link. To accommodate multiple access
88 5.2 System Model
Listening + multiple access
PU BUSY PU BUSYPU IDLE
PU
SU # 1 SU # 2 SU # K
Figure 5.1: Network Architecture of a cognitive radio network with multipleaccess functionality.
SU data transmission
S
Tf S
T T−
f
T
PU Busy PU Idle PU Busy . . .PU Idle
Spectrum discovery
Figure 5.2: MAC frame format for proposed DSMA mechanism.
during upstream transmission, proposed MAC protocol supports the random schedul-
ing among multiple SU. As PU transmission is not coordinated by the cognitive radio
network, therefore, MAC protocol takes into account the spectrum sensing in upstream
scheduling for detecting the PU transmission. e operational MAC frame format is
given in Fig. 5.2, where Ts period is allocated for the spectrum discovery over the frame
duration Tf . In the remaining Tf − Ts period, the proposed MAC adopts dynamic time
sequence for the packet transmission protocol. e seing of the operational time for
MAC operation is accomplished dynamically by a cross-layer design, where sensing
parameters are integrated with the random scheduling mechanism. Before any packet
transmission, a conditional channel sensing takes place aer the spectrum discovery
in the proposed model. erefore, the access mechanism is referred to as a dual-leve
sensing based multiple access (DSMA) protocol.
5.2 System Model 89
5.2.2 Energy Detection Based Spectrum Sensing
In the proposed model, channel monitoring plays a key role in the design of packet
transmission protocol. When the channel monitoring is required, the spectrum sensing
operation is carried out by using energy detection method. A sensing model is presented
below with its underlying performance metrics.
Let y(m) denotes the received signal to the secondary user for primary user detection
over τs period with sampling frequency fs, wherem is the sampling index; thus, the total
number of sampling isM = bfsτsc. Applying a binary hypothesis-testing problem [1, 3],
the detection process is modeled as,
H0 : y (m) = w (m) (5.1)
H1 : y (m) = h .s (m) + w (m) (5.2)
where s(m) is the transmied signal,w(m) is the additivewhite Gaussian noise (AWGN),
and h is complex channel gain. It is assumed that s(m) and w(m) are independent and
identically distributed (iid) random process with both having the mean zero, and vari-
ance σ2s and σ
2w respectively. Hypothesis H0 and H1 describe the absence and presence
of the PU signal, respectively. e measured signal-to-noise ratio (SNR) under the H1
hypothesis is γ = |h|2 σ2s/σ
2w. e test statistics of the detector is [43]
Y ∼
χ2
2M H0 : PU is absent
χ22M (2γ) H1 : PU is present
(5.3)
where test statistic Y follows a central chi-square (χ22M) distribution with 2M degrees
of freedom for H0, and a non-central chi-square distribution (χ22M (2γ)) with 2M de-
grees of freedom and a non centrality parameter 2γ for H1. e performance of the
detection is evaluated with probability of detection (Pd), probability of false alarm (Pf ),
90 5.3 Proposed Model of the DSMA Protocol
and probability of missed detection (Pm), which are expressed as follows [43],
Pd = P Y1 > ε|H1 = Q(√
2Mγ,
√ε
σw
)(5.4)
Pf = P Y1 > ε|H0 =Γ(M, ε
2σ2w
)Γ (M)
(5.5)
where ε is the threshold value of the energy detector,Q(., .) is a generalized MarcumQ-
function, and Γ(a, b) is an incomplete gamma function given by Γ (a, b) =´∞bta−1e−tdt,
and Γ(a) is a gamma function. Consequently, the probability of missed detection is
measured by Pm = 1 − Pd. For a large number of sampling, it can be shown that the
distribution of the test statistic is normal distribution [3, 43]. By using central limit
theorem, the performance metrics can be expressed in terms of Gaussian Q(.) function
as derived in previous chapter.
e detector must satisfy a given constraint, Pd ≥ P d, to provide the interference
protection to the primary network from the secondary transmission. For instance, if
P d = 0.9, then it indicates that the primary network can tolerate the maximum 10% of
interference (Pm = 1 − Pd = 0.1) from the SU transmission which occurs depending
on the detection. us, the detector should design based on the given constraint. For a
given target probability of detection, P d, the Pf is obtained by
Pf(P d, τs
)= Q
(√2γ + 1Q−1(P d) + γ
√τsfs
)(5.6)
5.3 Proposed Model of the DSMA Protocol
5.3.1 Underlying Mechanisms of Proposed Protocol
e proposed protocol relies on the following mechanisms and their underlying para-
meters:
• Spectrum discovery: It initiates the CR operation by applying spectrum sensing
to nd the vacant channel from the channel of interest (CoI) as shown in Fig.
5.3 Proposed Model of the DSMA Protocol 91
Spectrum
Discovery
Clear
Channel
Assessment
Backoff
Process
Start CR
Operation
Packet
Transmission
Figure 5.3: Block diagram of the Proposed DSMA Mechanism.
5.3. e MAC layer requests the PHY to perform this operation. erefore, a
certain operational period takes into account in theMAC frame in time dimension.
MAC only concerns the time length of this operation and its decision regarding the
channel occupancy status. e eective time length of its operation and detection
outcome rely on the interference protection to the legacy system in a given channel
condition. Also, this operation can estimate the full capacity of the CoI before
utilizing the channel which is required to design an ecient transmision protocol.
By considering all these issues, the eective operational time is designed in Section
5.4.1.
• Backo Process: e protocol initiates a backo process when the channel is
sensed as busy either by spectrum discovery or by CCA. e backo process is
designed to produce a random delay for scheduling the packet transmission among
multiple contenders asynchronously. is process is accumulated with the backo
counter and the backo stage. In the proposed model, backo counter is a decre-
mental mechanism when SU neither senses the channel nor transmits any packet.
At the start of backo process, a random number is chosen for counter from the
contention window (CW) and the CCA is performed when the counter reaches
to zero. Since this process is executed before the CCA, it can be integrated into
the model to reformulate the detection objectives in the CCA.is integration has
been done by a PHY/MAC cross-layer design in the proposed model. e details
formulation of the using backo mechanism including its depending parameters
is explained in 5.4.2.
92 5.3 Proposed Model of the DSMA Protocol
• Clear Channel Assessment: is function is normally generated in MAC layer
and executed in PHY layer through signal detection. Only the decision is taken
into account in making the decision regarding packet transmission. ere are two
possible ways in running the CCA as shown in Fig. 5.3. e CCA performs just
aer the spectrum discovery if the channel is obtained as idle by the spectrum
discovery. e CCA also comes into operation through the backo process. e
command for initiating the CCA comes when the CW reaches to 0 in each backo
stage. When the CCA is evolvedwith the backo process, the detection parameters
in the CCA is updating according to the backo parameters which is one of the
main contributions of this model. e formulation is depicted in Section 5.4.3.
• Packet Transmission: It allows the SU to transmit the data packet in the chan-
nel. e RTS/CTS based mechanism [57, 66] is adopted in the proposed model
for reducing the collision period during the channel access. Packet transmission
protocol starts when CCA declares that the channel is idle, with the RTS packet
transmission instead of the main data packet transmission. e details of the
packet transmission including the measurement of its service time are presented
in Section 5.4.4.
5.3.2 Proposed Protocol
eprotocol structure of the DSMAprotocol is shown in Fig. 5.3. Each SU in the network
starts with the spectrum discovery operation. MAC layer requests PHY to perform this
spectrum discovery through signal detection algorithm. is is a mandatory task of
the proposed protocol to nd the spectrum opportunity from the CoI. If the channel is
assessed to be idle, the MAC (sets the CW as 0) does not allow backo delay and requests
the PHY to perform the CCA operation by using signal detection method. However, the
operational time and target detection in the spectrum discovery andCCA are followed by
the proposed PHY/MAC cross-layer designing which is explained later. If the channel is
obtained as busy in the contrary, then MAC initiates the backo process. Aer nishing
the backo process (when the CW reaches to 0), MAC requests the PHY to perform CCA.
5.4 Analytical Modeling of Proposed DSMA Mechanism 93
In overall, the CCA operation comes into operation either by directly from the spectrum
discovery or through the backo process.
When the channel is sensed as idle by the CCA operation, then SU goes for the imme-
diate packet transmission. Otherwise, when the channel is sensed as busy, SU has to wait
again a random backo delay according the backomechanism. e backomechanism
relies on two parameters: minimumvalue of the contentionwindow andmaximumvalue
of the backo stage. e detail working principle of the backo process for delaying the
channel access is revolved with the CWmin and maxBS which is explained in the Section
5.4.2.
5.4 Analytical Modeling of Proposed DSMA Mechan-
ism
5.4.1 Operational Time in Spectrum Discovery
Even though the signal detection method is applied both to spectrum discovery and
CCA but the aliation is separated due to their physical aributions. In particular,
expected operational period for the spectrumdiscovery is quite adaptive and relies on the
occupancy history of the PU. InMAC frame format, the period of the spectrum discovery
is acquired based on a dynamic decisional process. An optimization is conducted to
allocate the dynamic decisional period for spectrum discovery in every frame. e
operational period of the spectrum discovery is allocated by the following method:
1. For a given constraint, Pd ≥ P d, the detector threshold is designed where it
exhibits lowest PFA. is can be done through the ROC curve for a given SNR
value.
2. As two sensing steps are taken into account before any packet transmission, there-
fore, the overall target PD, P d, is achieved by distributing the PD into two steps.
is operation can be accomplished by applyingPd1 = Pd2 = 1−√
1− P d, where
Pd1 andPd2 tune the detector in the spectrum discovery and the CCA, respectively.
94 5.4 Analytical Modeling of Proposed DSMA Mechanism
3. Aer seing the detector’s sensitivity as explained above, the maximum opera-
tional time, τ(1)s,max, is obtained based on the following optimization:
τ (1)s,max = argmax
Pd≥P dPH0 (1− Pf (Pd, τs))
(1− τs
Tf
)(5.7)
By applying the equality constraint as Pd = P d, the objective function of (5.7) can be
derived as a function of τs and it is proved that this objective function is then a log-
concave function with respect to sensing period τs [3, 50]. us, there is a feasible
optimal value of the τs existing over the Tf period for which the objective function has
a maximum value [110].
5.4.2 Time Sequence Adaptation Based on Backo Process and
Detection Mechanism
Let us assume that i and k denote the backo stage and backo counter, respectively,
where i ∈ (0, u) and k ∈ (0,Wi − 1). Here, u is the maximum size of the backo
stage and the corresponding contention window isWu−1. Backo counter is described
with the minimum value of the contention window (W0) and the contention window is
described in the unit of slots.
At the start of backo process, MAC layer initializes the following variables: max-
imum value of the backo stage and minimum value of the contention window. en, a
random value is chosen for the backo counter from the contentionwindow (0, 1, 2, · · · ,
W0 − 1). e counter decrements its value uniformly and initiates the CCA while it
reaches to 0. If the channel is sensed as busy then the counter value is incremented based
on the binary exponential method [66] and performs the decremented counting for the
next backo stage. is process continues until the backo stage reaches its maximum
value. During this process, if the channel is sensed as idle in any stage, then SU transmits
the packet into the channel and comebacks to the initial spectrum discovery task. On the
contrary, if the channel is not found as idle for packet transmission within the maximum
contention, then the packet is discarded from the transmission aempt. e maximum
5.4 Analytical Modeling of Proposed DSMA Mechanism 95
size of the contention window is determined by 2iW0 and mathematically, it is referred
to as a function ofW0 when the maximum value of backo stage is given.
For the sake of simplicity, the complex backo process with innite re-transmission
aempt [67] is not considered herein as it can only consolidate the eectiveness of
backo process by means of packet service protocol. Since this research deals with
the cross-layer design of the sensing-access method, thus, the only situation where the
sensing has taken place in the backo process is sucient to consider for formulation.
As described above, channel sensing is performed at every backo stage aer n-
ishing the counter. In this proposed model, the target PD in channel sensing at every
backo stage is then reformed according to the backo parameters. e objective of this
target PD reformulation is to adaptively change the sensitivity of the detection process
to characterize the channel state. As a result, the detection output become an integrated
function of the backo parameters and detection sensitivity. By examining the merged
function, the overall objectives, throughput improvement and collision reduction, can
be accomplished signicantly.
5.4.3 Cross-layer Formulation of Backo and Detection Process
in CCA
Let Pidle be the idle probability through the channel aer nishing the backo counter
in a backo stage. According to the proposed cross-layer design, the Pidle is hence
expressed as a function of P d,W0, i. Assuming that x = Pd in backo stage i, the target
PD in this contention window is set by,
x(i,W0, P d) = 1− Wi
√1− P d = 1−
(1− P d
) 12iW0 (5.8)
us, the corresponding probability of false alarm is given by,
Pf(i,W0, P d
)= Q
(AQ−1
(x(i,W0, P d
))+B
)(5.9)
96 5.4 Analytical Modeling of Proposed DSMA Mechanism
RTS
CTS
DATA
ACK
NAV (RTS)
DIFS SIFS SIFS SIFS
DS phase Packet transmission phase
Sensing
PU Active
Sensing
Sensing
RTS/CTS based access
PU
SU1
SU2
SU3
Figure 5.4: Time slot operation of proposing dual-level sensing based mul-tiple access protocol.
where, A =√
2γ + 1 and B = γ√σslotfs are assumed in (5.6). e channel idle
probability can be formulated as,
Pidle(i,W0, P d) = PH0
(1− Pf (i,W0, P d)
)+ PH1
(1− x(i,W0, P d)
)(5.10)
5.4.4 Packet Transmission Service
is section presents a complete packet transmission protocol based on the proposed
model. Fig. 5.4 shows a packet transmission process when a cognitive user S1 wants
to transmit a data packet to another user S2 via a single-hop communication link. A
single radio channel is considered where PU occupies the channel for a particular period
over the full frame. e vertical dierentiation of this single-radio channel describes the
dierent activities of each user in the the same time dimension.
As shown in Fig. 5.4, all cognitive users have to wait for a certain period to complete
the spectrum discovery according to Section 5.4.1. Once the channel is found as idle, SU1
initiates the second sensing steps through CCA function. If SU1 nds the channel as idle
for a predened distributed inter frame space (DIFS), then SU1 immediately transmits
the RTS frame to the channel instead of the data packet transmission. is is a frame
based sensing method, where the reception of the RTS packet to the adjacent contenders
initiates another function, the network allocation vector (NAV). As shown in Fig. 5.4, SU3
which is not the desired user of the data packet receiver, initiates the NAV by encoding
the RTS frame. Furthermore, technically, the RTS frame contains the information about
5.5 Performance Analysis 97
the length of the data packet with the other relevant information about the ongoing
transmission aempt. us, SU2 replies with the clear-to-send (CTS) just following the
short inter frame spacing (SIFS) period by reading the RTS frame, and other contenders
keep following the backo process. If SU1 receives the CTS successfully, then it goes for
data transmission aer a SIFS. In contrast, if the sender does not receive any CTS aer
SIFS interval, then it is assumed that the packet collision occurs in the channel. Aer
receiving the DATA frame, SU2 acknowledges with an ACK frame to the sender SU1
aer a SIFS interval which makes a completion of a successful data transmission.
5.5 Performance Analysis
e performance of this proposed model is evaluated by the achievable throughput in a
single transmission aempt. e probabilistic performances from every step involve in
the measurement of the throughput. Step-by-step analysis (SA) is presented according
to the following sequences:
SA1 :e spectrum discovery is taken place at the start of the frame with the model
as given in Section 5.4.1. From that design, the operational time of the spectrum
discovery τ(1)s,max → Ts and the corresponding probabilities take into account in
Figure 5.6: Normalized throughput versus probability of transmission forcomparing the analytical, simulated, and approximated modelof DSMA scheme for N = 10.
whereas N can be maximized while the denominator of the above equation (5.41) is
minimized. With the help of [66], the following approximated solution of φ(2)is obtained
as,
φ(2) =1
N√T ∗C/2
(5.42)
where T ∗C = TC/σslot. By applying this approximation of φ(2)with respect toN , we can
also evaluate N . In Fig 5.6, we evaluated the normalized throughput N with respect to
probability of transmission φ and observed that the simulation result closely matched
with the analytical result. Moreover, the normalized throughput N also increased with
the same rate due to the approximation, and it was even smaller when φ has small value.
Nevertheless, with the increasing of φ, the normalized throughput N maintains closer
value to the analytical result.
5.6.2 Throughput Performance Analysis
Fig. 5.7 shows the variation of normalized throughput with respect to transmission
probability φ for N = 5, 10, 20, 50, which implies that throughput decreases rapidly
when a large number of SUs are intended to access the channel. When there are 5
where E[D] is the average number of slots required for doing a successful packet trans-
mission into the channel. According to SAA protocol, E[X] can also be the average
backo delay that the SU waits in generic before accessing the channel as our back-
o mechanism consists of all the relevant cases such as slot-counter decremented and
holding value.
e mean period of collisions as accounted in [49, 115] for delay analysis is not
relevant in our model. Since the sensing error has already been synthesized directly into
the backo process of the SAA protocol; thus, the entire eect of collisions is included
in dening the channel status with Pi and Pb. Furthermore, we assume that the delay
due to packet dropping is not relevant to this calculation as the packet is not successfully
6.5 Numerical Results 131
received. Based on our proposed protocol and its interpretation into the Markov chain,
E[X] can be measured by considering that the slot-counter requires k number of slots
to reach i,−1 state for transmission from i, k state and the time interval between
these transitions is quite random whose average is given by
E[X] =u∑i=0
ωi−1∑k=0
kπ(i, k) (6.35)
By using equations (6.15), (6.16), and (6.17), we obtain as follows,
E [X] =π(0, 0)
6
[ω2
0(1− (4P )u+1)
1− 4P− 1− P u+1
1− P
](6.36)
By puing the value of π(0, 0) from equation (6.18) into (6.36), we can obtain the E[X].
In (6.34), E[NH ] is the average number of times that the SU holds on its slot-counter
value due to the detection of transmission which is given by,
E[NH ] =E[X]− E[kH0 ]
E[kH0 ](6.37)
E[kH0 ] is the average number of idle slots before a transmission occurs which can be
obtained by
E[kH0 ] =1− P C(t) = TxP C(t) = Tx
(6.38)
Exploiting (6.24) and (6.36), E[NH ] can also be derived as a function of n and φn. By put-
ting the value of E[X] and E[NH ] into (6.34), the average access delay of our proposed
SAA protocol can be obtained.
6.5 Numerical Results
In this section, we evaluate the performance of SAA protocol in respect of physical-
layer sensing parameters Pf and Pm, and also for MAC contention parameters. We
consider that primary network operates with 6 MHz bandwidth in an AWGN channel
and secondary network follows the frame length as Tf = 10 ms. Numerical parameters
132 6.5 Numerical Results
Table 6.1: Parameters for Performance Analysis of SAA Protocol.
Parameters Value
Packet payload 8512 Bytes
MAC header 728 bits
PHY header 512 bits
RTS 450 bits+ PHY header
CTS 320 bits+ PHY header
ACK 320 bits+ PHY header
Channel bit rate 1 Mb/s
Propagation delay(δ) 1 µs
Slot time(τu) 50 µs
TDIFS 136 µs
TSIFS 28 µs
Size of CW(ω0) 4 ∼ 128
Maximum backo stage(u) 5
Number of SU (N ) 5 ∼ 50
used in this analysis are outlined in Table 6.1. e size of the regarding packets is given
in the unit of the bit which can be converted into time scale based on channel bit rate. In
this analysis, we assume that SU follows same bit rate both in control packet (RTS and
CTS) and data packet transmission.
6.5.1 Throughput and Delay Performance of Proposed SAA Pro-
tocol
e performance of throughput is analyzed based on the equation (6.33) where E[P ]
can be obtained in slot time using the numerical data provided in Table 6.1. Also, by
using the numerical data, the average slot time for successful transmission, collision,
and empty slot can be estimated based on the equations of TS , TC , and TI . We consider
that φ1 = · · · = φn = φ for a given cognitive radio network which only rely on cross-
layer based backomechanism. According to (6.29), Tser is then as a function of φ andN
for the estimated value of TS , TC , and TI . In the detector, the sensing error is related to
6.5 Numerical Results 133
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Probability of missed detection
No
rma
lize
d t
hro
ug
hp
ut
ω0 = 4 (A)
ω0 = 4 (S)
ω0 = 8 (A)
ω0 = 8 (S)
ω0 = 16 (A)
ω0 = 16 (S)
ω0 = 32 (A)
ω0 = 32 (S)
Figure 6.5: Characteristic of normalized throughput (S) corresponding toprobability of missed detection (Pm); where the parameters are:γ = −15 dB, fs = 6 MHz, PH1 = 0.1, u = 5, and N = 20.
Pf = Q(√
2γ + 1Q−1(1− Pm) + γ√τsfs
)according to (3.6) and (3.7). By employing
this relationship into (6.21) and (6.22), the access probabilityφ becomes a function ofPH1 ,
Pm, ω0, and u. For a given value of PH1 , ω0, and u, however, the normalized throughput
S also depends on the number of contenders in multiple access scenario.
In a multiple access scenario with a xed number of contenders N = 10, the vari-
ation of S with respect to Pm is illustrated in Fig. 6.5 for several values of minimum
contention window ω0. In this gure, there is a close consent between the analytical (A)
and simulated (S) results. When ω0 = 4, S decreasing exponentially with the increasing
of Pm which is an essential property of SAA protocol. Nevertheless, the variation of S is
advanced into the steady condition by increasing the size of contention window ω0 from
4 to 32, which implies that proposed SAA protocol achieves higher throughput even the
detector’s Pm is large. Note that larger the Pm value can cause severe interference to
the primary network. erefore, we should choose the operating characteristic of SAA
protocol when Pm is low. Fig. 6.5 also indicates that proposed SAA protocol achieves
higher throughput performance when the sensing error is small for a given ω0 which
can able to reduce interference to primary transmission.
134 6.5 Numerical Results
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
1
2
3
4
5
6
Probability of missed detection
Ave
rag
e a
cce
ss d
ela
y (
ms)
ω0 = 4 (A)
ω0 = 4 (S)
ω0 = 8 (A)
ω0 = 8 (S)
ω0 = 16 (A)
ω0 = 16 (S)
ω0 = 32 (A)
ω0 = 32 (S)
Figure 6.6: Characteristic of average access delay (E[D]) corresponding toprobability of missed detection (Pm); where the parameters are:γ = −15 dB, fs = 6 MHz, PH1 = 0.1, u = 5, and N = 20.
Fig. 6.6 shows the behavior of average access delay E[D] as a function of Pm based
on (6.34) for the same simulation conditions applied in measuring the S in Fig. 6.5. With
the increasing of Pm, E[D] also increases for a given ω0 but the rate of increment is
relatively steep for lower size of the contention window. For instance, E[D] increases
abruptly from 1 ms to 5 ms with the increasing of Pm when ω0 = 4. In contrast, for
larger contention window i.e., ω0 = 32, E[D] increases abruptly only when Pm ≤ 0.1
and sustains almost in the same time range even Pm increases. is behavior implies
that proposed SAA protocol is used the contention mechanism eciently with relatively
higher value of contention window to achieve less access delay for the secondary users.
e performance of SAA protocol is further evaluated with the behavior of collision
and access probability corresponding to the probability of missed detection as shown in
Fig. 6.7 and Fig. 6.8, respectively. In Fig. 6.7, the probability of collision (PC) during the
channel access is very low at the lower value of Pm which is desirable for cognitive radio
network; and PC increases traditionally with the increasing of missed detection. us,
the rate of collision due to the missed detection in physical sensing can be overcome by
6.5 Numerical Results 135
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Probability of missed detection
Pro
ba
bili
ty o
f co
llisio
n
For ω0 = 4 (A)
For ω0 = 8 (A)
For ω0 = 16 (A)
For ω0 = 32 (A)
For ω0 = 4 (S)
For ω0 = 8 (S)
For ω0 = 16 (S)
For ω0 = 32 (S)
Figure 6.7: Probability of collision (PC ) with respect to probability of misseddetection (Pm) of SAA protocol; where the parameters are: wherethe parameters are: γ = −15 dB, fs = 6MHz,PH1 = 0.1, u = 5,and N = 20.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.05
0.1
0.15
0.2
0.25
0.3
0.35
Probability of missed detection
Pro
ba
bili
ty o
f a
cce
ss
For ω0 = 4 (A)
For ω0 = 8 (A)
For ω0 = 16 (A)
For ω0 = 32 (A)
For ω0 = 4 (S)
For ω0 = 8 (S)
For ω0 = 16 (S)
For ω0 = 32 (S)
Figure 6.8: Probability of access (φ) with respect to probability of misseddetection (Pm) of SAA protocol; where the parameters are: γ =−15 dB, fs = 6 MHz, PH1 = 0.1, u = 5, and N = 20.
increasing the contention window as shown in Fig. 6.7. For instance, when Pm is 0.2 or
20% then there is an about 25% of chance of collision for ω0 = 4. is probability is
dropped down to below 10% when the size of contention window ω0 is increased which
136 6.5 Numerical Results
0 20 40 60 80 100 1200
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Contention window
No
rma
lize
d t
hro
ug
hp
ut
For Pm = 0.1 (A)
For Pm = 0.1 (S)
For Pm = 0.5 (A)
For Pm = 0.5 (S)
For Pm = 0.9 (A)
For Pm = 0.9 (S)
Figure 6.9: Normalized throughput (S) versus contention window (ω0) ofSAA protocol; where the parameters are: γ = −15 dB, fs = 6MHz, PH1 = 0.1, u = 5, and ω0 = 16.
is a signicant contribution of the proposed SAA protocol. Fig. 6.8 shows that access
probability for lower value of ω0 is higher than for higher value of ω0 in respect of Pm.
e small value of Pm means the lower chance of inter-network collision between the
primary and secondary network [3, 49, 65]. On the other hand, higher access probability
among a large number of contenders can reduce the normalized throughput as depicted
in contention based access [63, 66]. In this circumstances, we can say that proposed SAA
protocol can achieve stable access condition for a relatively large number of contention
window even though the Pm increases.
Fig. 6.9 describes the impact of the size of the contention window on the throughput
performance. At a target Pm, the normalized throughput S is increased by extending the
value of ω0. In particular, when the target Pm is set as 0.1, S reaches its maximum value
with the extending of ω0 and starts to decline slowly with the further extension of ω0.
us, there is an optimal value of ω0 to achieve the maximum throughput performance.
On the other hand, SAA protocol requires a comparatively larger value of ω0, when Pm
increases, to stable the throughput performance onto a maximum condition as depicted
in both Fig. 6.5 and Fig. 6.9.
6.5 Numerical Results 137
5 10 15 20 25 30 35 40 45 500
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Number of SU
No
rma
lize
d t
hro
ug
hp
ut
ω0 = 4 (A)
ω0 = 4 (S)
ω0 = 8 (A)
ω0 = 8 (S)
ω0 = 16 (A)
ω0 = 16 (S)
ω0 = 32 (A)
ω0 = 32 (S)
Figure 6.10: Variation of normalized throughput (S) corresponding to num-ber of SU (N ) in analytical and simulation cases; where theparameters are: γ = −15 dB, fs = 6 MHz, PH1 = 0.1, u = 5,and Pm = 0.1.
5 10 15 20 25 30 35 40 45 500
1
2
3
4
5
6
Number of SU
Ave
rag
e a
cce
ss d
ela
y (
ms)
ω0 = 8 (A)
ω0 = 8 (S)
ω0 = 16 (A)
ω0 = 16 (S)
ω0 = 32 (A)
ω0 = 32 (S)
Figure 6.11: Variation of average access delay (E[D]) corresponding tonumber of SU (N ) in analytical and simulation cases; wherethe parameters are: γ = −15 dB, fs = 6 MHz, PH1 = 0.1,u = 5, and Pm = 0.1.
evariation of performancemetrics duringmultiple access scenario is also depended
on the network size. Here the number of SUs dene the network size. Fig. 6.10 and Fig.
138 6.5 Numerical Results
6.11 illustrated the changing behavior of S and E[D] regarding N for several values of
ω0. It is observed that the simulation results are rmly agreed with the analytical results.
With the increasing of the number of contenders for access, the normalized throughput
decreases as shown in Fig. 6.10. e decrement is occurred largely for ω0 = 4 compared
to the case for ω0 = 32. Likewise, SAA protocol suers less E[D] by taking the larger
value of contentionwindow as shown in Fig. 6.11. In overall, SAA protocol requires large
contention window for improving both the throughput and delay performance when a
large number of users needed to be accommodated in multiple access.
6.5.2 Model Validation and Performance Comparison
To validate the analytical model of the SAA protocol, we analyzed and compared the
analytical result with the simulation result. Additionally, two approximations of the
access probability based on equation (6.23) and [66] are adopted in our throughput
calculation to examine the characteristic of S in respect of φ for a given number of
contenders.
Fig. 6.12 shows the variation of normalized throughput in respect of probability of
access φ for N = 20 and 50, which merely indicates that throughput decreases rapidly
when a large number of SUs are intended to access the channel. When there are 20
contenders, the throughput reaches its maximum level quite gradually and requires a
signicant value of φ compared with the case of N = 50. Also, S maintains relatively
higher throughput performance when N = 20, even through φ increases. On the other
hand, S reaches its maximum level with a very small value of φ, but it decreases more
rapidly when φ increases among 50 users.
For any given number of contenders, the simulated results are closely matched with
the analytical results in Fig. 6.12. e throughput achieved based on our proposed
approximation of φ is always lower than our exact throughput measurement for both
the size ofN which indicates that the exact measurement reveals the maximum range of
throughput that the secondary network can achieve during the multiple access. Another
approximation of φ follows the seminal work of [66]. e analytical model of S in
6.5 Numerical Results 139
equation (6.33), is very convenient to determine the maximum level of the achievable
throughput. Let us rearrange equation (6.33) as follows,
S =E[P ]
Tser/PS=E[P ]
Tden(6.39)
whereas S can be maximized while Tden of the above equation (6.39) is minimized. With
the help of [66], the approximated solution of φ ≈ 1/N√T ∗c /2 is obtained where
T ∗c = TC/τu. By applying this approximation of φ regarding N , we also evaluate S
without considering the sensing aspects in Fig. 6.12. By comparing the approximation
of φ among our proposed model (equation (6.23)) and [66], it is found that our proposed
approximation still outperforms the approximation made by [66] with a signicant mar-
gin always in the increasing range of φ. is comparison also implies that the proposed
SAA protocol follows the similar but improved characteristic of CSMA/CA protocol
corresponding to access probability for the cognitive radio network. is S versus φ
characteristic for several values of N exhibits the similar operational characteristic of
CSMA/CA protocol [66] which validate our proposed SAA protocol. Furthermore, the
higher value of S compared with [66] indicates the impact of sensing-assisted mechan-
ism in throughput improvement.
Finally, the throughput performance of proposed SAA protocol is compared with
other C-MAC protocols in Fig. 6.13. Distributed MAC protocol [6] and CR-CSMA [7]
are the most relevant MAC protocols for performance comparison due to their seminal
contributions in contention-based access for CR users which outperformed over other
protocols [48, 49, 57, 65]. e performance of distributed-MAC and CR-CSMA proto-
cols are computed under our simulation conditions to conduct a fair comparison with
our proposed SAA protocol. e normalized throughput of all protocols is compared
regarding the number of SUs for the contention window of 8 and 32. For both values of
the contention window, all the three protocols show a similar decreasing characteristic
of throughput corresponding to the increasing of contenders for channel accessing as
previously described by Fig. 6.10. However, proposed SAA protocol maintains com-
paratively higher throughput performance than the distributed-MAC and CR-CSMA
140 6.5 Numerical Results
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.090.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Probability of access
No
rma
lize
d t
hro
ug
hp
ut
Analytical, N = 20
Simulated, N = 20
Approximated (proposed), N = 20
Approximated ([12]), N = 20
Analytical, N = 50
Simulated, N = 50
Approximated (proposed), N = 50
Approximated ([12]), N = 50
Figure 6.12: Normalized throughput (S) versus probability of access (φ) withapproximation, simulation, and analytical results; where theparameters are: γ = −15 dB, fs = 6 MHz, PH1 = 0.1, u = 5,ω0 = 16, and Pm = 0.1.
5 10 15 20 25 30 35 40 45 500
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Number of SU
No
rma
lize
d t
hro
ug
hp
ut
Distributed MAC (ω = 8)
CR-CSMA/CA (ω = 8)
Proposed SAA (ω = 8)
Distributed MAC (ω = 32)
CR-CSMA/CA (ω = 32)
Proposed SAA (ω = 32)
Figure 6.13: Normalized throughput comparison among distributed-MAC[6], CR-CSMA [7], and our proposed SAA protocol with respectto number of SU. In this analysis, the using parameters are:γ = −15 dB, fs = 6 MHz, PH1 = 0.1, u = 5, and Pm = 0.1.
.
protocol with the increment of the number of users for both cases of ω0. When ω0 = 32,
6.6 Chapter Summary 141
SAA protocol achieves quite same S as achieved by distributed-MAC and CR-CSMA, but
SAA protocol outperforms the other protocols when the contender increases. In overall,
when a large number of contenders are required to be accommodated by using limited
contention window, then our proposed SAA protocols performs extremely well than
the other comparing protocols. Based on this thorough investigation, it can be outlined
that the proposed SAA protocol can achieve ecient throughput and delay performance
by synthesizing the PHY/MAC cross-layer parameters correctly under the presence of
sensing error as well as during the multiple access environment.
6.6 Chapter Summary
In this chapter, we proposed the SAA protocol by PHY/MAC cross-layer operation to
enhance the performance of CRN during multiple access. We developed a novel sensing-
assisted contention access including all the sensing with regard to the random backo
process. Furthermore, we derived and analyzed the SAA protocol where the aspect of
imperfect sensing is captured to investigate the consequence of physical-layer sensing
for the proper measurement of throughput and delay. Performance evaluation of numer-
ical results indicate that SAA protocol improves the throughput and delay performance
when the sensing error is tolerable; and by selecting the proper contention window,
the performance can be enhanced when sensing error and the number of contenders
are increasing in the network. By considering the imperfect sensing in the backo
process, the SU overcame the waste of spectrum opportunity by reducing the false alarm
in detection and the consequent increasing collision probability is compensated by the
wider contention window in our proposed SAA protocol. e model validation through
simulation results and performance comparison conrms the signicance of the pro-
posed SAA protocol for the improvement of the throughput and delay performance for
cognitive radio networks.
CHAPTER 7
Conclusion and Recommendationsfor Future Research
Mobile devices and connections are geing smarter with the capabilities of seamless
connectivity and mobile computing. e explosion of mobile connectivities and advance
multimedia applications demand fast and intelligent networks in wireless technologies.
e existing xed spectrum access policy strives to handle the ever-growing end-user
demand which leads to the spectrum scarcity problem in current wireless technology.
CR technology has emerged as a promising solution to spectrum scarcity based on the
idea of dynamic spectrum access. On the other hand, some of the allocated RF bands are
not utilized to their full potential. CR technology allows the unlicensed users to use the
underutilized portion of a licensed band while ensuring the necessary protection and
ecient utilization.
Spectrum sensing and access are two crucial components of the CR operation. With
the combined operation of these two components, SU can monitor the channel activity
and apply an appropriate transmission strategy to meet the goal of CR operation. To
increase the SU throughput performance, existing access protocols apply aggressive
transmission strategies leading to harmful interference to the legacy users. e ob-
jective of the research documented in this thesis was to propose the access strategy
incorporatedwith spectrum sensing to overcome the sensing-throughput trade-o issue.
e conducted research conrms, through comprehensive analysis and validation, that
the proposed methods and strategies outperform the state-of-the-art of the cognitive
radio network. e ndings and technical contributions accomplished throughput this
research are summarized in this chapter.
144 7.1 Conclusion
7.1 Conclusion
Cognitive radio networkswork according to a hierarchical accessmodelwhereby unused
portions of a licensed spectrum are open to SUs, provided PUs are protected by limiting
the interference from SUs’ transmission. Traditionally, the interference protection is
guaranteed by spectrum sensing designed to meet a target detection probability. SUs
can increase the throughput when the sensing period is short enough allowing for a
more prolonged access period. A short sensing period leads to a more substantial prob-
ability of false alarm, hence limiting spectrum opportunity and decreasing throughput.
is sensing-throughput trade-o issue cannot be overcome with advancement in the
underlying sensing and access operation when both of these operations are independ-
ently designed in the dierent layers. In addition, the sensing performance reects the
capacity and interference of the CR access. Hence, it was necessary to determine the
impact of sensing on the access protocol design’s capacity to satisfy the target of sensing-
throughput trade-o.
A statistical model of the spectrum sensing was established to analyze the impact
of detection performance in the realistic channel condition. e research focused on
the design of access strategy by exploiting the post-processing data of the sensing. e
sensing in this research is referred to as detector-independent sensing algorithm. How-
ever, energy detector and matched-lter are applied for signal detection to model the
sensing system. e performance parameters of the spectrum sensing are formulated
by applying the binary hypothesis testing problem. Dening the channel state based
on the detection is important to measure the capacity SU achieved by the spectrum
sensing. Apart from PU detection, it is also required to model the PU trac to measure
the probabilistic channel state. PU trac is modeled as a two-state random process with
Poisson distribution in state transitions of ON and OFF states. e steady-state probab-
ilities of the channel state are formulated by using a discrete time Markov chain process.
Finally, the spectrum opportunity achieved by spectrum sensing is formulated with the
probabilistic relationship between the PU occupancy status and detection performance.
Investigation into the spectrum opportunity reveals that the PFA has greater potential
7.1 Conclusion 145
than the PD in enhancing the opportunity. It is also assessed that single-level sensing
failed to produce the greatest spectrum opportunity as its detection experienced high
PFA. A dual-level sensing mechanism is proposed over the same sensing period used in
an SS mechanism by segmenting the target PD conditionally over that two sensing-level.
e overall PFA of the DS mechanism obtained is lower than the SS mechanism. e
access capability is characterized by receiver operating characteristic curve and access
probability. e ROC curve denes the maximum bounding of PFA and PD relation for
a given SNR. e ROC curve analysis consolidates that the proposed DS mechanism has
greater detection capability than conventional SS mechanism at a given SNR value. e
access probability analysis proved that the DS mechanism outperforms the SS mech-
anism by a wide margin with the fastest growing rate towards maximum capacity and
greater utilizing capability.
e eectiveness of theDSmechanism is capitalized on access operation by proposing
a dual-level sensing based multiple access (DSMA) protocol. In DSMA, the SU can access
the channel following a conditional second sensing once the channel is obtained as idle in
the rst sensing. e SU can defer the transmission aempt when the channel is sensed
as busy in any sensing steps and proceeds with a backo process. e backo process
is devised with a cross-layer integration of the physical detection and the contention
method. In the contention method, the backo process has the advantages in collision
reduction by deferring the transmission aempt with random delay. In the conventional
backo process, the predened distribution, such as a uniform and exponential distri-
bution, in the backo process characterizes the transmission aempt and its suitable
rate. Unlike the conventional backo process, the transmission aempt with random
delay is recongured by using detector parameters and distribution of the transmis-
sion aempt in the proposed model. e detector parameters impose controllability on
the random delay. Consequently, the backo process reduces the compulsion of the
spectrum sensing in collision reduction. is adaptive design of the backo process
and detection sensitivity eventually contributes to both throughput improvement and
collision reduction.
146 7.1 Conclusion
It is imperative to nd the impact of DS mechanism on the sensing-throughput op-
timization. In conventional SS mechanism, there is an optimal sensing period to obtain
maximum throughput for a given target PD. In the DS mechanism, on the contrary,
two target PDs in two conditional sensing levels need to be set in order to meet the
overall target PD. With the advantages the DS mechanism brings additional challenges
in nding the optimal sensing period by which maximum throughput can be achieved.
As the internal operation of DS mechanism is conditioned by the sensing decision, it is
relevant to use the entire sensing period and the PD at any one of the sensing steps, in
the optimization.
For a fair comparison between theDS and SSmechanism, the constraint of the sensing
period in DS mechanism is set as equivalent to the optimal sensing period of SS mech-
anism to obtain maximum throughput for a given PD. By applying convex analysis, the
feasibility of the minimum PFA is examined regarding the target PD of the rst sensing
step and the total sensing period. With a thorough convex analysis, it is proved that
there is a global minimum of the PFA regarding the operational range of the sensing
period and target PD of the rst sensing step. However, it is hard to obtain a closed-form
mathematical equation for minimum PFA due to the mathematical complexity. A semi-
analytical algorithm is then proposed to solve the optimization where the boundaries
of the optimizers are provided by the feasibility analysis. Furthermore, a numerical
method, i.e., backtracking line search algorithm, is applied with considerable complexity
for joint optimization and model validation. By employing the post-optimization data
into the system, the DS mechanism achieves higher throughput than the SS mechanism
in a given channel condition.
A novel sensing-assisted access (SAA) protocol is proposed as a complete random
access mechanism for the secondary users. e sensing feature is integrated inside of
the backo process to enhance the capabilities of CR operation in reducing the packet
collision. e access contention, i.e., the sensing-embedded backo process, is modeled
byMarkov chain in the presence of sensing error. Conventional contention-based access
with backo process does not reect the original cause behind the packet collision and
7.2 Recommendations for Future Research 147
relies only upon the acknowledgment to determine the collision. e state character-
ization in the conventional backo process considers the perfect sensing, and hence
cannot accurately reect the interference to PU. A novel backo process is developed by
integrating the backo and sensing parameters for state characterization. e obtained
state characteristic reects the sensing error rendering for the packet collision. With
a proper choice of backo parameters, i.e., by increasing the contention window, the
collision probability is reduced signicantly leading to throughput improvement. In
essence, SAA protocol maximizes the throughput performance of the secondary users
and simultaneously ensures sucient interference protection to the primary user.
7.2 Recommendations for Future Research
is research concentrated on developing solutions for the access strategy to overcome
the sensing-throughput trade-o issue and was less focused on the issue of the energy
eciency. As such, proposed access strategies have mostly relied on the dual steps of the
sensing operation which may consume relatively higher power than the conventional
case. e proposed SAA protocol applied continuous sensing operation during the con-
tention access period for a transmission aempt. It is shown that the SAA protocol
required a much smaller access delay for any transmission aempt when compared
to the existing methods. For a shorter access delay, the access protocol may consume
energy for a shorter period. However, the performance of the proposed access strategies
can be further evaluated in the context of energy eciency.
Energy eciency of a protocol is also related to the transmission power of the SU.
Transmission power control is an important issue for improving not only the energy
eciency but also the CR capability by limiting the interference power to the PU. For
example, the interference protection in the systemmodel used in this research is demon-
strated through the target PD (equivalent to missed detection) which is the maximum
bound of the interference. If transmission power of the SU is related to the detection er-
ror, then the expected interference limit can be obtained that is lower than the maximum
interference. By considering the maximum level of the interference, the conducted re-
148 7.2 Recommendations for Future Research
search exposed the normalized capacity of the access protocol, but there is an additional
dimension in the resource block, i.e., power, to emphasize the achievable capacity limit
in the context of energy eciency.
e underlying detection method in the spectrum sensing algorithm was based on
energy detection due to its lower complexity and compatibility along the CR operations.
In the proposed DS mechanism, two thresholds are chosen based on their target PD
of the sensing steps for making the nal decision. When double threshold values are
applied to a binary hypothesis testing problem, then traditionally there is a region of
confusion between two threshold values in the probability distribution of the perform-
ance function. Hence, decision uncertainty for the samples laid down in the confusion
region could be taken into account for further research of the DS mechanism.
e solution approaches to optimization in Chapter 4 were based on semi-analytical
and pure numerical methods. In the proposed semi-analytical algorithm, the boundaries
of the feasible region were determined with the help of precise mathematical derivation.
e optimization was accomplished iteratively within analytical boundaries by using a
numerical method. ere is potential to enhance the computational eciency of this
algorithm. rough applying certain approximations, it may be possible to develop an
entirely analytical approach with closed-form mathematical solution.
e multi-channel scenarios can be recommended for further enhancement of the
capability of the proposed access protocol. is multi-channel network can provide fur-
ther exibility and access reliability with higher throughput and lower delay. In addition,
channel assignment algorithms with multi-channel sensing features have to be taken
into account for developing the multi-channel capability. Overall, there are countless
research challenges relating to spectrum access in the cognitive radio networks. e
above recommendations are only a few possible candidates, and the research presented
in this thesis can be expanded in relevant directions to construct ecient cognitive radio
networks for the deployment of future generation networks.
APPENDIX A
Proof of Propositions and Theorems
A.1 Proof of Proposition 4.2
For this proof, we assume that Pd1 = x and 1− Pf1(Pd1) = f1(x). Note that a function
f1(x) is said to be log-concave while h′(x) is monotonically decreasing function with
respect to the dened range of x, so as h′(x) < 0 where h (x) = f1′(x)
f1(x). Now taking the
rst dierentiation of h(x) yields,
h′(x) =f1(x)f ′′1 (x)− (f ′1(x))2
(f1(x))2 (A.1)
Using equation (4.16), (4.17), and (4.19), h′(x) is derived as,
h′(x) = −Φ3
√2γ + 1
(1− Pf1)2 exp
[2w2
d1− w2
f1
](A.2)
where we assume that Φ3 =√π(1 − Pf1)(
√2γ + 1wf1 − wd1) + exp[−w2
f1]. Based on
equation (A.2), it can be stated that h′(x) < 0 when Φ3 > 0. Simply, it can be said that
Φ3 > 0 as previously we proved that
√2γ + 1wf1 − wd1 > 0 for Pd1 ∈ [0, Pd1(θ1)].
However, when Pd1 → 0 then Φ3 is undened. erefore, we compute the bounding of
Pd1 for which Φ3 has denite and non-negative values, from the following inequality,
√π
2γ + 1(1− Pf1)
(2γwf1 + γ
√Ns
)+ exp
[−w2
f1
]> 0 (A.3)
Note thatddtQ(t) = −2e−t
2/√π is considered which is equivalent to −e−t2/2/
√2π for
t > 0; thus, the upper bound of Q(t) will be√
2e−t2. Likewise, puing the maximum
150 A.2 Proof of Proposition 4.4
value of Pf1 = Q(wf1) into equation (A.3) and obtain as below,
√π
2γ + 1
(2γwf1 + γ
√Ns
)+ exp
[−w2
f1
](1−
√2π
2γ + 1
(2γwf1 + γ
√Ns
))> 0 (A.4)
where
√π/(2γ + 1)(2γwf1 + γ
√Ns) always non-negative for any given value ofNs, γ.
us, by checking the remaining terms of above equation, we obtain
exp[−w2
f1
]> 0 or, 1−
√2π
2γ + 1
(2γwf1 + γ
√Ns
)> 0 (A.5)
ere is no real solution of this exp[−w2f1
] > 0 inequality, so the remaining term of
equation (A.5) becomes
Pd1 > Q
(1
2γ√
2π−√Ns(2γ + 1)
2
)(A.6)
Previously we found that Pd1(θ1) < Q(θ1) where θ1 = −√Ns(2γ + 1)/2. Similarly,
assume that Pd1(θ2) > Q(θ2) where θ2 = 1/(2γ√
2π) −√Ns(2γ + 1)/2. Comparing
these two assumptions, we found that θ2−θ1 = 1/(2γ√
2π) > 0 therebyQ(θ2) < Q(θ1)
and Pd1(θ2) is the upper bound for which Φ3 > 0. So, h′(x) < 0 and monotonically
decreasing function. Hence, (1 − Pf1) is a log-concave function of Pd1 for the range of
Pd1 ∈ [0, Pd1(θ2)]. us Proposition 4.2 is proved.
A.2 Proof of Proposition 4.4
We follow the similar conditions and assumptions dened in Proposition 2 to prove the
log-concavity of Pf2 . Aer taking the rst dierentiation of h2(x) =f ′2(x)
f2(x)when the
f2(x) = Pf2(Pd1), we obtain as follows,
h′2(x) = −Φ5
√2γ + 1
(1− PD
)P 2f2
(1− Pd1)3 exp
[w2d2− w2
f2
](A.7)
A.3 Proof of Proposition 4.5 151
by assuming that
Φ4 = 2 +
√π(1−PD)(1−Pd1)
(wd2 − wf2
√2γ + 1
)exp
[w2d2
]Φ5 = Pf2Φ4 −
√2γ+1(1−PD)(1−Pd1)
exp[w2d2− w2
f2
] (A.8)
By applying similar approach of the proof of proposition 2 and taking the approximated
maximum limit of Pf2 = Q (wf2), we obtain the following condition for which Φ5 > 0,
exp[−w2
f2
]> 0 or,
√2π(wd2 − wf2
√2γ + 1
)−√
2γ + 1 > 0 (A.9)
e rst term is undened. erefore, nally we get the bounding of Pd1 for which
Φ5 > 0, as follows
Pd1 <PD − Pd1(θ4)
1− Pd1(θ4)(A.10)
where,
Pd1(θ4) = Q(−√
2γ+12γ
(γ√Nc + 1√
2π
))In the range of Pd1 ∈ [0, Pd1(θ5)], h′2(x) < 0 so Pf2 is a log-concave function for that
range where Pd1(θ5 =(PD − Pd1(θ4)
)/ (1− Pd1(θ4)). us, Proposition 4.4 is proved.
A.3 Proof of Proposition 4.5
Similar to the solution approach of Proposition 4.1, we obtain dPf2dPd1
for ED-MF combin-
ation by using MF [44], as follows
dPf2dPd1
= −(1− PD
)(1− Pd1)
2 exp[w2d2− w2
f2
](A.11)
152 A.4 Proof of eorem 4.1
Taking the another dierentiation of (A.11) with respect to Pd1 ,
d2Pf2dP 2
d1
= −(1− PD
)(1− Pd1)
3 exp[w2d2− w2
f2
]×
[2−√π(1− PD
)(1− Pd1)
(wf2 − wd2) exp[w2d2
]](A.12)
As we know thatQ(wd2) > Q(wf2) so wd2 < wf2 . Also, for 0 < Pd1 < 1, the last term is
not greater than 1. erefore, the last term of the above equation is non-negative which
implies thatd2Pf2dP 2
d1
< 0. us, Pf2 is a concave function of Pd1 .
A.4 Proof of Theorem 4.1
Let us consider x∗ be the optimal solution of (4.27). en we obtain µ+F2(x) ≥ F1(x)+
F2(x) ≥ F1(x∗) + F2(x∗), ∀µ ≥ F1(x). However, above estimation implies that Ψ(z) ≥
F1(x∗) + F2(x∗) = µ∗ + F2(x∗), where µ∗ = F1(x). erefore, there exists an optimal
point z∗ = µ∗ = F1(x) ∈ Ω such that Ψ(z) ≥ Ψ(z∗) ∀z ∈ Ω. is proves the necessary
condition as stated ineorem 4.1.
A.5 Proof of Theorem 4.2
Let xo be a ξ-minimum critical point of the function F on En, then from (4.31) it follows
that 0 ∈ w +(∂∂x
)ξF1(xo), ∀w ∈ F2(xo). Hence,
min‖g‖=1
maxz∈w+(∂/∂x)ξF1(xo)
(z, g) ≥ 0, ∀w ∈ F2(xo)
and thus for every g ∈ En, ‖g‖ = 1, we have
minw∈(∂/∂x)F1(xo)
maxv∈(∂/∂x)ξF1(xo)
(z, g) ≥ 0
However, this means that
min‖g‖=1
(∂
∂x
)ξ
F (xo) ≥ 0 (A.13)
A.6 Proof of eorem 4.3 153
proving that the condition is necessary. at it is also sucient can be demonstrated in
an analogous way, arguing backwards from the inequality (A.13).
A.6 Proof of Theorem 4.3
Assume that xo is not a ξ-minimum critical point. en, we can describe the vector
gξ(xo) = arg min‖g‖=1
(∂
∂x
)ξ
F (xo)
as a direction of ξ-steepest-descent of function F at point xo and numerically that dir-
ection is
gξ = −(
voξ + wo‖voξ + wo‖
)(A.14)
where voξ ∈(∂∂x
)ξF (xo), wo ∈
(∂∂x
)F (xo) and
− maxw∈
(∂∂x
)F1(xo)
minv∈
(∂∂x
)ξ
F1(xo)
‖v + w‖
= −‖voξ + wo‖
= aξ(xo)
is a direction of ξ-steepest-descent of function F at point xo. is satises the condition
given in the theorem 4.3.
A.7 Proof of Proposition 4.6
Let consider that probability of false alarm for energy detector is a convex function with
respect to sensing period as benchmark for proving this proposition which is proved by
Liang et al. in [3]. Let us take required partial derivative of R0 with respect to τs and
obtain as follows ∂R0
∂τs= 1 + (τds − τs)
∂Pf1∂τs− Pf1
∂2R0
∂τ2s= −2
∂Pf1∂τs
+ (τds − τs)∂2Pf1∂τ2s
(A.15)
154 A.8 Proof of Proposition 4.7
Liang et al. proved that Pf1(τs) is a convex function while Pf1 ≤ 0.5 [3], so obviously
∂Pf1∂τs≤ 0. Hence, from the above equations, we can say that
∂2R0
∂τ2s≥ 0. us, R0 is a
convex function of τs. Conversely, we can say that there has feasible maximum value of
R0 corresponding to τs. e converse property will be also true for remaining sensing
period in τds even though we check the convexity with respect to rst sensing period.
A.8 Proof of Proposition 4.7
Let τ =√
2γ + 1Q−1(P ∗d1) + γ√τsfs then Pf1 = Q(τ), thus R0 is changed to
R0(τ) =(τb− c)2
+
(τds −
(τb− c)2)Q (τ) (A.16)
where we assume that, a =√
2γ + 1, b = γ√fs, and c = aQ−1(P ∗d1)/b. As τds > τs
therefore, (τ/b− c)2 > 0. Hence, the rst term of equation (A.16 ) is obviously a convex
function of τ . Nowwe need to prove that second term of equation (A.16) is also a convex
function of τ . Let the second term is expressed as,
Φ(τ) = τsQ(τ)−(τb− c)2
Q(τ) (A.17)
Taking the second derivative of Φ(τ),
∂2Φ
∂τ 2=
(τs −
(τb− c)2)∂2Q(τ)
∂τ 2
+
(−4
b
(τb− c)) ∂Q(τ)
∂τ+
(− 2
b2
)Q(τ) (A.18)
Based on equation (31) and τds > τs, if we can show that (−4/b (τ/b− c)) (∂/∂τ)Q(τ)
+(−2/b2)Q(τ) ≥ 0 then (∂2/∂τ 2)Φ ≥ 0. According to Cherno boundsQ(τ) ≤ e−τ2/2
and (∂/∂τ)Q(τ) = −e−τ2/2/√
2π, we obtain the following inequality for (∂2/∂τ 2)Φ ≥
0 as (−4
b
(τb− c)) ∂Q(τ)
∂τ+
(− 2
b2
)Q(τ) ≥ 0
A.8 Proof of Proposition 4.7 155
Here, Pd1 at the rst sensing is generally bounded with 0.5 ≤ Pd1 < PD, therefore
Q−1(Pd1) ≤ 0. By recalling the values of a, b, and c into the above inequality, we obtain
the following condition as,
τ ≥√π
2+√
2γ + 1Q−1(Pd1) (A.19)
As τ = Q−1(Pf1), and γ > 0, equation (A.19) implies that if
Pf1 ≤ Q(√
π/2 +Q−1(P ∗d1))
(A.20)
then Φ(τ) is convex. us, the proposition is proved.
Bibliography
[1] I. F. Akyildiz, W.-Y. Lee, M. C. Vuran, and S. Mohanty, “NeXt generation/dynamic
spectrum access/cognitive radiowireless networks: A survey,” Computer networks,
vol. 50, no. 13, pp. 2127–2159, 2006.
[2] T. Yucek and H. Arslan, “A survey of spectrum sensing algorithms for cognitive
radio applications,” IEEE Communications Surveys Tutorials, vol. 11, no. 1, pp. 116–
130, First 2009.
[3] Y. C. Liang, Y. Zeng, E. C. Y. Peh, andA. T. Hoang, “Sensing-throughput tradeo for
cognitive radio networks,” IEEE Transactions on Wireless Communications, vol. 7,
no. 4, pp. 1326–1337, April 2008.
[4] G. P. Joshi, S. Y. Nam, and S. W. Kim, “Cognitive radio wireless sensor networks:
Applications, challenges and research trends,” Sensors (Basel), vol. 13, no. 9, pp.
11 196 – 228, September 2013.
[5] S. Y. Lien, C. C. Chien, H. L. Tsai, Y. C. Liang, and D. I. Kim, “Congurable 3GPP
licensed assisted access to unlicensed spectrum,” IEEE Wireless Communications,
vol. 23, no. 6, pp. 32–39, December 2016.
[6] L. T. Tan and L. B. Le, “Distributed MAC protocol for cognitive radio networks:
Design, analysis, and optimization,” IEEE Transactions on Vehicular Technology,
vol. 60, no. 8, pp. 3990–4003, Oct 2011.
[7] Q. Chen, Y. C. Liang, M. Motani, and W. C. Wong, “A two-level MAC protocol
strategy for opportunistic spectrum access in cognitive radio networks,” IEEE
Transactions on Vehicular Technology, vol. 60, no. 5, pp. 2164–2180, Jun 2011.
158 BIBLIOGRAPHY
[8] J. Drumm, N. White, and M. Swiegers, “Mobile consumer survey 2016, the aus-