Journal of Parallel and Distributed Computing 142 (2020) 90–105

Contents lists available at ScienceDirect

J. Parallel Distrib. Comput.

journal homepage: www.elsevier.com/locate/jpdc

Artificial intelligence inspired energy and spectrum aware cluster

based routing protocol for cognitive radio sensor networks
∗
Thompson Stephan a , , Fadi Al-Turjman b , K. Suresh Joseph c , Balamurugan Balusamy d ,
Sweta Srivastava a
a
Department of Computer Science & Engineering, Amity School of Engineering and Technology, Amity University, Uttar Pradesh, India
b
Department of Artificial Intelligence Engineering, Research Center for AI and IoT, Near East University, Nicosia, Mersin 10, Turkey
c
Department of Computer Science, Pondicherry University, Puducherry, India
d
School of Computer Science and Engineering, Galgotias University, Greater Noida, India

article info a b s t r a c t

Article history: A Cognitive Radio Sensor Network (CRSN) is a distributed network of sensor nodes, which senses
Received 8 October 2019 event signals and collaboratively communicates over dynamically available spectrum bands in a multi-
Received in revised form 8 December 2019 hop mode. All nodes participating in CRSN have to be cognitive of the network environment and
Accepted 14 April 2020
autonomous in decision making for resolving issues related to throughput maximization, delay, and
Available online 24 April 2020
energy minimization. Clustering in CRSN is proven to tackle such issues and enlarges the network’s
Keywords: lifetime. However, the existing clustering algorithms designed for WSNs do not consider the CR
Cognitive radio sensor networks functionalities and challenges, and CR based networks work on the assumption of unlimited energy.
Clustering This paper proposes an energy and spectrum aware unequal cluster based routing (ESUCR) protocol
Energy efficiency intending to resolve the issues of clustering and routing in CRSN. In ESUCR, cluster formation is
Spectrum sensing mainly performed considering the residual energy of the secondary users (SUs) and relative spectrum
Event driven awareness such that the common data channels for clusters are selected based on the appearance
probability of PUs. ESUCR performs energy-efficient channel sensing by deciding the channel state
with the statistic previous channel states. The premature death of cluster heads (CHs) is minimized
by selecting and rotating the CHs based on intra-cluster channel stability, energy, distance, and
neighbor connectivity. During event detection, ESUCR performs energy-efficient data routing towards
the sink node by employing hop by hop forwarding through the CHs and primary/secondary gateways.
The performance of the proposed ESUCR protocol is proved through extensive simulations and
compared to those of the state-of-the-art protocols under a dynamic spectrum-aware data transmission
environment.
© 2020 Elsevier Inc. All rights reserved.

1. Introduction area which leads to the problem of spectrum scarcity and causes
interference among the co-existent sensor networks. Such spec-
1.1. Cognitive radio sensor networks (CRSNs) trum over-utilization problem in the ISM band could be solved
by allowing cognitive radio (CR) assistance to traditional WSNs,
Wireless sensor network (WSN) technologies are playing an forming cognitive radio sensor networks (CRSNs) which allows
important role by closely coupling the physical infrastructure efficient utilization of the radio spectrum and helps minimize in-
with information and communication technologies through the terference and jamming through dynamic spectrum access (DSA).
use of network embedded devices. WSNs are the key technol- In cognitive radio sensor networks (CRSN), secondary users (SUs)
ogy for the Internet of Things (IoT) with its rapid development or sensor nodes are allowed to dynamically vary its transmit-
in sensor technologies which allows intelligent monitoring and ting and receiving parameters, by estimating the adjacent radio
management of the environment. For enabling communication, conditions either independently or in collaboration with other
the nodes deployed ion WSNs utilizes industrial, scientific, and sensor nodes and opportunistically utilize the licensed channels.
medical (ISM) band. However, the same ISM band is utilized However, on the detection of any primary user (PU) activity,
by different sensor networks deployed in the same geographical the SUs must quit the transmission and vacate the occupied
channel. In any CR based network, there are two major spec-
∗ Corresponding author. trum sharing schemes, underlay and overlay [24]. Underlay CR
E-mail address: [email protected] (T. Stephan). networks minimize the interference and noise to PUs by allowing

https://doi.org/10.1016/j.jpdc.2020.04.007
0743-7315/© 2020 Elsevier Inc. All rights reserved.
 T. Stephan, F. Al-Turjman, K. Suresh Joseph et al. / Journal of Parallel and Distributed Computing 142 (2020) 90–105 91

the SUs to share the licensed spectrum through the use of spread size increases the proximity among the nodes within the clus-
spectrum techniques, like CDMA and UWB [8] whereas overlay ter boundary which allows the SUs to discover a similar set
CR networks allow the SUs with opportunistic access of vacant of channel availability. This permits the clusters to utilize and
spectrum. This paper focuses on the overlay spectrum sharing share different channels among them and hence minimizes the
scheme for implementing CRSN. channel contention which minimizes the interference levels via
addressing the challenges of dynamic channel behavior. Further-
1.2. Clustered CRSNs: characteristics and challenges more, it improves the efficiency of channel selection and routing
which eventually minimizes the number of retransmissions and
In CRSN, all SUs work in a peer-to-peer fashion [7] forming a enhances the energy conservation in the CRSN. More importantly,
decentralized network. As the number of SUs increases, the com- the re-clustering overhead in smaller sized clusters is greatly
plexity of managing large-scale CRSNs increases which leads to reduced owing to improved cluster stability, which is achieved
underperformance. For large-scale CRSNs, clustering seems to be as a result of enhanced connectivity among the SUs. For cluster
an effective way to improve the performance of the network via construction, it is important to consider the availability of chan-
improved network scalability, minimizing the cost of network co- nels in each SU to ensure common channels in the cluster which
ordination and maximizing network lifetime [34]. In a distributed helps to minimize network complexity and improve connectivity.
environment, upper-layer protocols such as routing protocol [18, Though, a properly formed cluster may be disintegrated owing to
47] may not work well without clustering techniques. However, it PU occurrence on the channels associated with the cluster and
is important to overcome the design challenges for implementing interrupt the ongoing transmission. This forces the network to
clustering in CRSNs. In typical WSNs, the topology information is rebuild the cluster and may force frequent re-clustering which
exchanged among the nodes (i.e., SUs) through message broad- forces longer delay, higher overhead, and energy loss in the net-
cast on the predefined common control channel (CCC) and as work. Therefore, it is highly necessary to prevent the cluster from
a result, overcomes connectivity issues. Conversely, the SUs in the activity of PUs through proper maintenance and minimize
the chances of re-clustering. Therefore, it is important to tackle
CRSN are available with diverse channels that force the SUs
the dynamicity in channel availability which causes frequent
with no global CCC for sharing information among them. It is
re-clustering, inefficient routing, and energy loss and results in
important to consider the network connectivity for the proper
unstable clusters formed in the network.
implementation of clustering in CRSNs.
Based on the aforementioned limitations, the major objective
In the past years, clustering has been a traditional approach
of this research work is to develop an energy-efficient spectrum
for effective management of the topology of many WSNs where
aware event-driven unequal cluster-based routing strategy with
the entire network is grouped into different clusters in which
a focus on minimizing the energy consumption and maximiz-
certain sensor nodes are selected as cluster heads (CHs) and rest
ing the communication efficiency of CRSN. Thus, to achieve the
of the nodes act as cluster members (CMs) of their respective
above objective, it is important to implement energy-efficient
cluster. The CH forwards the aggregated data collected from its
spectrum sensing with improved accuracy over channel detection
CMs to the next-hop node along the routing path where the data
and perform optimal channel selection to minimize PU interfer-
eventually reaches the destination or sink node. The major prob-
ence during data transmission. Also, it is necessary to minimize
lem in any WSN is the energy efficiency of the deployed nodes.
intra-cluster energy consumption through optimal selection and
Hence, most of the clustering approaches focus on considering
maintenance of CHs for each cluster and perform the best channel
the residual energy [19,36] of the nodes before finalizing the
selection for improved intra-cluster and inter-cluster data com-
decision on cluster formation. Also, CRSN based cluster formation
munication. Furthermore, it is essential to maximize the inter-
mechanisms should ensure the clusters are formed with at least
cluster connectivity through the selection of appropriate gateway
one channel common among CMs residing inside a cluster. To
SUs.
manage CR based operations such as channel sensing and routing,
cluster structure provides an effective means to cooperate with
1.4. Contributions
such activities. A SU in a clustered CRSN can act as a CH, CM, relay,
and gateway. The intra-cluster communication between the CH To overcome the major issues associated with CRSN, this paper
and the CMs is performed on a common channel shared between proposes an energy and spectrum aware unequal cluster based
them known as operating channel, which is common to all the routing (ESUCR) protocol for CRSN with the following contribu-
SUs within the cluster boundary. A CM inside a cluster may act tions:
as a relay SU to relay the message from the CH to the CM of the 1. A channel ranking technique to rank and select the optimal
neighboring cluster which is outside the communication range of channel based on channel stability, which is calculated consid-
the CH. A SU located at the cluster boundary may act as a gateway ering the weighted state transition probabilities, for performing
node to provide two or more hops of inter-cluster connectivity. intra-cluster and inter-cluster communication.
2. A spectrum sensing technique that improves the accuracy
1.3. Problem statements and objective of energy-based sensing by resolving the ambiguous states.
3. A CH selection and rotation technique that intelligently
The performance of any clustered network is mainly affected selects and rotates the CH for balancing the energy within the
by the cluster size. The cluster size denotes the number of sen- cluster.
sor nodes inside the cluster. Large-sized clusters improve the 4. A gateway and relay selection technique for performing
scalability and allow the possibility to minimize the overhead inter-cluster data forwarding.
in routing the messages since the flooding of route control in-
formation only involves the CHs and gateways and the sensing 2. Related work and research gap
error concerning channel availability greatly reduces owing to
improved accuracy in channel decision with a higher number The existing works have used clustering techniques to over-
of SUs inside the cluster. For small-sized clusters, the connec- come bandwidth limitations, routing, and energy-related issues
tivity among the SUs is maximized owing to a higher number to ensure topology maintenance. The existing research on WSNs
of common channels within the cluster since smaller cluster [10] focuses on energy minimization without consideration of
92 T. Stephan, F. Al-Turjman, K. Suresh Joseph et al. / Journal of Parallel and Distributed Computing 142 (2020) 90–105

channel-related issues, whereas research works on CR networks in two phases, where the first phase determines the eligible
[39] neglect energy issues. Hence, CRSNs need to consider both SUs and the second phase performs cluster formation among
energy and spectrum related issues. Currently, only certain works the SUs according to the channel conditions. Also, the clusters
[28,37] have considered the issue jointly but still need more ef- are dismantled at the end of each event. The authors [12,44]
ficiency for implementation in CRSNs. Joshi et al. [16] have given utilized cooperative spectrum sensing to overcome the issues
a detailed study on existing cluster-based strategies for CRSNs. of signal attenuation which decreases the accuracy of channel
Abolarinwa et al. [1] have used an intelligent learning technique sensing. Ram et al. [40] proposed Energy Aware Cluster based
that works on previous experience considering the rate of energy Routing Protocol (EACRP) which aimed at reducing the frequency
depletion and energy-efficiency achievable in a particular channel
of re-clustering by clustering the SUs in accordance with the
transmission. Xiaoyuan et al. [20] demonstrated an improved
channel conditions and energy status. EACRP primarily focused
energy-efficient cluster-based multi-channel CRSN. The proposed
on channel assignment problem for performing intra-cluster and
technique focuses on varying packet sizes following the state of
the channel and further performs channel assignment according inter-cluster communication.
to the residual energy of the sensor nodes. This approach ex-
tended the lifetime of CRSN as compared to the random channel 3. System model
pairing approach. Furthermore, the issues of dynamic spectrum
access were investigated by modeling the behavior of PUs as
We consider a distributed CRSN consisting of uniformly de-
a two-state Markov chain and importantly maximum-likelihood
ployed SUs (i.e., sensor nodes) with certain initial energy and
(ML) estimator was used to estimate the channel state transi-
PUs. A predefined threshold over residual energy is set on each
tion probabilities. Ghalib et al. [33] proposed a spectrum-aware
deployed SU, beyond which it is denied from operating in CRSN.
cluster-based routing (SCR) protocol which limits the number
of SUs taking part in the route establishment process. For en- Each SU switches among the available channels using its CR func-
suring transmission with minimum interference, SCR protocol tionality to minimize PU interference. All deployed SUs perform
schedules the vacant channels to the SUs considering its energy CR based operations such as channel sensing, channel selection,
and spectrum requirements. Though SCR is not designed for channel sharing, and channel hand-off. The SUs use the CHs and
the event-driven environment, and hence unnecessary cluster gateway SUs to transmit their event data packets in a distributed
formations lead to energy loss and increase the frequency of multi-hop manner to the sink which is placed at the center of
re-clustering. Fadel et al. [13] aimed to address the QoS require- the area. The internal functionality of our proposed protocol falls
ments through honey bee mating optimization based routing into three main layers, namely network layer, data link layer, and
and cooperative channel assignment algorithms. The proposed physical layers and a cross-layer repository. The joint channel de-
algorithms maximized the quality of channel links among the SUs cision and routing decisions are undertaken at the network layer.
which is further achieved by minimizing the packet loss probabil- Channel decision involves channel selection and channel switch-
ity. Tabassum et al. [38] proposed event-driven routing protocol ing which is, in turn, decided based on the underlying channel
(ERP) which performs clustering based on certain parameters sensing module at the data link layer. The channel selection
such as energy, channel availability, state of neighbors and dis- module decides the best available channel from multiple channels
tance. The ERP protocol considers the interference probability of
which is selected based on our proposed channel ranking and se-
PUs for efficient cluster formation and makes use of CHs and
lection algorithm. Channel switching module is triggered to cease
gateways for data delivery to the sink. However, with an increase
the ongoing transmission on detecting any PU activity and select
in the event occurrence rate, the rate of re-clustering increases,
the next best channel based on its availability and priority which
which is an issue for consideration. Huazi et al. [45] proposed
a Distributed Spectrum-Aware Clustering (DSAC) scheme for ef- is further decided at the channel selection module. The channel
ficient intra-cluster aggregation and inter-cluster relaying. The handoff [41] happens at the physical layer and any transmission
DSAC algorithm performs clustering based on groupwise con- must be stopped during such operation. Sensing and sharing of
straint for minimizing the intra-cluster distance in a channel channels are performed at the data link layer. The channel sensing
aware environment. Christian et al. [25] proposed an energy module observes the utilization levels of each channel based on
and CR aware routing (ECR) protocol that performs AODV based which the channel quality and availability can be predicted for
routing in which the node-channel pairing is performed consid- access in a shared wireless environment. Cross-layer repository
ering the channel attributes and current energy status. However, allows the information to be shared among the different layers
the ECR still demands higher efficiency in terms of reducing such as the information of the next channel for operation from
the rerouting frequency and channel awareness. Abbas et al. [4] the channel decision module to the channel handoff module. Each
proposed EMCOS which is an energy-efficient mechanism for SU sensor node pair may experience a different amount of white
multimedia streaming. EMCOS aims to form stable clusters in space but opportunistically share the channels with the same
terms of reduced channel switching among the nodes by consid- amount of link capacity. PUs’ traffic in each channel is generated
ering the geographical positions, channel conditions and energy by the PUs’ activities module according to the PUs traffic model.
utilization. The channel selection technique followed in EMCOS The channels are heterogeneous and hence the channel sensing
is mainly dependent on the prediction results of each available
results are based on the SUs’ experience on the PUs’ behavior
channel. Mustafa et al. [26] proposed event-driven spectrum-
at its location. Using the neighborhood discovery protocol [11],
aware clustering (ESAC) protocol which forms spectrum aware
each SU is aware of the residual energy, location and channel
clusters not in the entire network but between the event and
sink and is maintained until the end of a particular event. ESAC availability of its one-hop neighbors. Also, a SU is aware of its
improves the connectivity among the clusters by maximizing residual energy and the location of itself and all the deployed
the number of vacant channels between clusters. However, gate- SUs in the CRSN. The sensor nodes exchange control information
ways SUs for inter-cluster communication are selected randomly among them and the sink using a fixed common control channel.
which increases the frequency of rerouting owing to PU inter- Each SU measures its distance with other SUs and sink node using
ference. Furthermore, the authors [27] aimed to overcome the the location information and furthermore chooses a SU as its
event-to-sink forwarding issue in mobile CRSNs with a spectrum- neighbor only if it is in its transmission range and availability of
aware clustering protocol. The proposed protocol was designed one common channel between them for packet transmission.
 T. Stephan, F. Al-Turjman, K. Suresh Joseph et al. / Journal of Parallel and Distributed Computing 142 (2020) 90–105 93

3.1. Energy consumption model incurs energy loss. Though, some of the above discussed sensing
schemes calculated double threshold with consideration of noise
In CRSN, a SU depletes energy mostly due to channel sensing, uncertainty, the problem of noise uncertainty is not properly
channel switching, data transmission, and reception. For SUa , the identified in the realistic wireless atmosphere. Furthermore, a
energy depleted for channel sensing is fixed and the same for all major of the above schemes working on double-threshold tech-
licensed channels and denoted as es . Also, ew denotes the energy niques are computationally complex due to cooperative decision
consumed to configure the radio and switch to a new channel. making and importantly, increases the duration of sensing which
For SUa , the energy utilization for data transmission
( Ea,t is based has a direct negative impact over energy efficiency. Also, the
on the classic energy model [30], i.e., Ea,t = Pa + Pa,cr .ta,x ,
)
impact of sensing duration is not properly studied by the above
where the data transmission time is denoted as ta,x , transmis- schemes which raise questions over the realistic implementation
sion power as Pa and circuit power as( Pa,cr . )According to [14], of CR functionalities.
Pa,cr is determined as Pa,cr = αa + 1
η
− 1 .Pa , where αa is With respect to spectrum sensing, this research work pro-
a transmission-power-independent component that accounts for poses a two-stage sensing technique for minimizing the impact of
the power consumed by the circuit, and η is the power amplifier noise uncertainty and further improving the accuracy of channel
efficiency which is physically determined by the drain efficiency detection (idle or busy). The first stage defines thresholds, θl
of the RF power amplifier and the modulation scheme [9,46]. and θu , where θl denotes lower threshold and θu denotes upper
Thus, energy consumed for transmitting data at SUa is according threshold, which is used to maintain the channel sensing robust
to Eq. (1). to noise uncertainty. The second stage improves the detection
accuracy by resolving the ambiguous states generated during
1 1(
.Pa .ta,x + αa .ta,x = Pa + αcr ,a .ta,x spectrum sensing, which is further achieved by jointly consid-
)
Ea,t = (1)
η η ering the weighted state transition probabilities and the last
where αcr ,a = η.αa is defined as the equivalent circuit power generated state (i.e. the binary state recorded at the previous
consumption for data transmission. The energy consumption for sensing instant) for a particular channel.
data reception is related to the data that a SU receives [30]. If SUa
receives l bits data, energy consumed is Ea,rc v = ecr .l, where the 3.2.1. Conventional energy sensing technique
circuit power for data reception as ecr . Generally, the wireless spectrum environment is built over the
wideband frequency range, which constitutes of C sub-channels
3.2. Spectrum sensing with the same bandwidth and each sub-channel is operated at a
particular frequency, fc . The SUs are allowed to detect the occu-
Spectrum sensing [17] is an important operation in any CRSN pancy of PU over a particular sub-channel using the traditional
for efficient spectrum utilization. Numerous techniques have energy-sensing technique [5]. In this paper, the channel and
been proposed and implanted for improving the quality of spec- sub-channel are used interchangeably and refers to a particular
trum sensing such as autocorrelation sensing, cyclostationarity- frequency.
based sensing, energy sensing and other techniques The main function of the energy detector is to determine the
(time–frequency analysis and wavelet transforms) [42]. Impor- presence or absence of primary signal in at fc without any prior
tantly, the majority of the sensing techniques is based on energy knowledge of PU signal parameters. The detection performance is
sensing [17] owing to its simplicity, energy efficiency and neglect- determined by important signal parameters namely, the number
ing the need for past PU information. However, under the cases of of samples, estimated noise power and detection threshold.
lower SNR, the performance of energy sensing techniques dras-
tically reduces. In the past, most of the existing signal detection 3.2.2. Binary hypothesis testing problem
techniques based on energy sensing focus on a single threshold. Based on the PU activity (idle/busy) with noise occurrence, it
Recently, certain works have focused on improving the quality of is possible to model the binary hypothesis testing problem for the
energy-based detection through a double threshold [22]. The au- signal detected at the SU as below:
thors [2], proposed a cooperative spectrum sensing scheme based
on two-stage detection which performs detection of channel Hypothesis 0 (H0 ) : absence of PU (channel vacant)
occupancy based on a single threshold in the first phase, followed
by application of double threshold in the second phase. Similarly, Hypothesis 1 (H1 ) : presence of PU (channel occupied)
the performance of channel detection was improved through a
two-step spectrum sensing method [23], which constitutes of The PU transmitted signal, sk = sr+ jsi , where sr denotes the
double-threshold sensing in the first phase, followed by cyclo- real component and si denotes the imaginary component. Let y
stationarity spectrum sensing in the second phase. Though, both denote the signal received, where the nth sample, y(g) is sampled
the schemes [2,23], are highly complex in computation and incur as follows:
long duration of spectrum sensing which drastically increases the w (g)
{
H0
energy consumption. A double threshold technique [6] was used y (g) = (2)
x (g) + w (g) H1
to implement spectrum sensing where the results of the energy
detection are split into a hard and soft decision. However, the where x(g) = hsk (g), the channel gain, h = hr +jhi is constant
implementation is complicated as the decision on signal detection within each spectrum sensing period (Tsp ), sk (g) denotes the sig-
is finally taken by the fusion center. The fusion rule technique is nal licensed to PU and gth (g = {1, 2, . . . , G} sample, where G is the
implemented in [15], where common k-out-of N rule (a decision total number of samples. The noise sample, w(g) = wr (g)+jwi (g),
made based on the number of positive responses not lower is assumed to be circularly symmetric complex Gaussian (CSCG)
than k) is used. A multi-stage sensing process is implemented random variable with mean zero (E[w(g)] = 0) and variance
in [21], which allows decision making by the CR based on the 2σ (var[w(g) = 2σ ]), where w(g)∼G(0, 2σ ), the expectation and
collected samples and based the number of samples, the energy variance operations are denoted by E[.] and Var[.], respectively,
level is varied between two thresholds. However, CR is triggered and complex Gaussian distribution is denoted by G(.,.). The null
to move to several stages for sample collection until a proper hypothesis, H0 denotes the channel to be idle and the alternative
decision is reached. This increases the sensing time and thus hypothesis, H1 denotes the presence of PU over the channel.
94 T. Stephan, F. Al-Turjman, K. Suresh Joseph et al. / Journal of Parallel and Distributed Computing 142 (2020) 90–105

The SU compares the observed energy levels with a prefixed the binary state may be forced to move into an ambiguous state
threshold, θ and finalizes its decision for a sub-channel to be in due to inappropriate energy levels. According to Eq. (11), the
H0 or H1 . The energy of the PU signal, Es is represented according occupancy state of a channel, Co can be classified into the binary
to Eq. (3). states, 0 and 1, and an ambiguous state, z.
Es < θl
G
0 H0
{
1∑
Es = |y(g)|2 (3) C0 = z θl < Es < θu H (11)
G
g=1 1 Es > θu H1
Here, Es is approximated as a normal variable, with G being large From Eq. (11), it is observed that it is hard to derive an optimal
and y(g) being independent. Therefore, Es is normally distributed decision on whether a signal is detected with PU activity or
under H0 and H1 as follows:
not with a single threshold. Previous research works focused
2σw
⎧ ( 4
)
⎪ G σw2 , H0 on resolving the issue of the ambiguous state through imple-
⎪ G
⎨
( )2 ) menting techniques such as exploiting cyclostationary behav-
Es ∼ (4)
(
2 σw
2 +σ 2
s ior of the communication signals [23] and cooperative decision
⎩G σw + σs ,
2 2
⎪ H1
making [3]. The authors [22] implemented an adaptive double-
⎪ G

threshold energy-sensing method based on Markov model

where the noise variance and signal variance is represented as, (ADEMM), which resolves the ambiguous state by considering
σw2 and σs2 , respectively. the time-varying channel behavior. The ADEMM technique uses
Based on the noise power, the value of the threshold is defined the state transition probabilities of the channel to finalize the
to consider the false alarm probability (PF ), detection probability decision on the channel’s binary state in case the measured
(PD ) and missing probability (PM ) which are derived in Eqs. (5), energy levels fall in ‘z’. ADEMM uses the weight count of the past
(6) and (7), respectively.
( ) occurrences of channel state to calculate the corresponding state
θ − Gσw2 transition probabilities of a channel, based on which the final
PF = P {ES > θ | H0 } = Q decision on whether a channel is idle or busy is taken by the SU
2Gσw4
√
(√ in case of ambiguity.
))
θ
(
G
=Q −1 (5)
2 Gσw2 4. ESUCR: Our proposed protocol
(√ )
G(
θ̃ − 1
)
=Q 4.1. ESUCR: Energy efficient spectrum sensing
2
⎛ )⎞
θ − G σw2 + σs2
(
This research work focuses on estimating the channel state
PD = P {ES > θ | H1 } = Q ⎝ √ ⎠
)2 transition probabilities based on the number of continuous or
2G σw2 + σs2
(
uninterrupted idle/busy states observed during a time interval, tk .
(√ )
G θ / Gσw2 − (1 + γ ) Here, the state is considered to be in transition when there is an
( )
=Q (6) interruption in the ongoing channel state, i.e., idle (0) → busy (1)
2 1+γ
(√ ) or busy (0) → idle (1). The period between two such transitions
G θ̃ − (1 + γ ) is considered as the transition period. We focus on measuring
=Q the number of states (idle or busy) involved during the transition
2 (1 + γ )
period. Here, at a particular sensing instant t, the probability of
PM = P {ES < θ | H1 } = 1 − PD (7) a channel being idle and the probability of a channel being busy
are denoted as Pidle and Pbusy , respectively.
where the Gaussian-Q function is denoted by Q(.) ( and) SNR is
denoted as γ where γ = σs2 /σw2 . Also, θ̃ = θ/ Gσw2 , then as
observed from Eqs. (4) and (5), for the given γ and G, PD is a 4.1.1. PU occupancy model
function of θ̃ and PF is independent of θ̃ . The continuous-time, alternating busy/idle Markov Renewal
As investigated in [43], the noise uncertainty can be repre- Process (MRP) [29,32] is used to model the channel usage pattern
sented according to Eq. (8). of each PU, i.e. the binary state of the PU signal. Each state
σ˜w2 [ −V/10 transition follows the Poisson arrival process since both the idle
ρ= , 10V/10 , V ≥ 0
]
∈ 10 (8) and busy periods of a channel are assumed to be independent
σw2
and identically distributed and the length of idle and busy periods
where the maximum noise uncertainty and real noise variance is follows exponential distribution [29] with p.d.f. fidle (t) = λidle ×
denoted by V and σ˜w2 respectively. e−λidle t for idle state and fbusy (t) = λbusy × e−λbusy t for busy state.
Consequently, θl and θu are defined by noise uncertainty ac- The PU channel model as illustrated in Fig. 1, describes the traffic
cording to Eqs. (9) and (10). pattern of channel k and the corresponding binary states.
Generally, the probabilities, Pidle and Pbusy are computed based
(√ )
2 1
θl = Q −1
(PF ) + 1 σw2 (9) on the number of channel state measurements [31,32], Ncsm =
G ρ
(√ ) Nidle + Nbusy , calculated over tk . Let, the number of idle and
2 busy state measurements over tk is denoted as Nidle and Nbusy ,
θh = Q −1
(PF ) + 1 ρσw2 (10) respectively, the mean idle and busy rate of a channel is denoted
G
as λidle and λbusy , respectively. For example, if Ncsm = 44, where
Generally, the output of an energy-sensing technique decides the (Nidle = 16, Nbusy = 28). Then, (λidle = 0.363 per second,
state of a signal to be vacant (0) or PU occupied (1). However, λbusy = 0.636 per second). Now, Pidle = 63.6% and Pbusy = 36.4%.
 T. Stephan, F. Al-Turjman, K. Suresh Joseph et al. / Journal of Parallel and Distributed Computing 142 (2020) 90–105 95

Fig. 1. PU occupancy model.

4.1.2. Estimation of weighted state transition probabilities 4.1.3. Spectrum sensing

In our approach, the probability weight of a channel increases To ensure minimal interference to PUs, the accuracy of spec-
or decreases based on the past channel behavior observed with trum sensing should be improved. Thus, cooperative spectrum
uninterrupted idle/busy states. It is important to note that, our sensing [44] allows the SUs to collaborate in finalizing the de-
research does not focus on measuring the number of indepen- cision on channel availability. The SUs are grouped in clusters
dently observed channel states or the number of channel state to achieve cooperative spectrum sensing. The SUs are initially
transitions over tk , rather we focus on measuring the number of allowed with independent sensing and the clusters are formed
continuously observed channel states (0 or 1) between two chan- based on the sensing results. The CH applies AND or OR rule
nel state transitions, i.e. during a transition period. To achieve to decide the occupancy state of PU. The number of SUs to
this, the parameter Nidle and Nbusy are modified to measure tran- participate in cooperative spectrum sensing is bounded according
sition periods with weights. It means, each transition period, tp is to [40]. To improve the accuracy of sensing and discover trans-
assigned a weight, wtp , which is based on the number of idle/busy mission opportunities, the sensing time in EACRP [40] scheme
states measured during tp . Thus, for channel k, N′idle and N′busy is for independent sensing of each channel is more which results in
calculated according to Eqs. (12) and (13). increased energy loss. Hence, ESUCR focuses on minimizing the
pTk sensing time to improve energy conservation while improving
′k ∑ the quality of spectrum sensing, i.e. enhancing the efficiency
Nidle = iktp .wktp ; wktp = ikt
p /η (12)
to resolve the ambiguous states. Thus, in cases of ambiguity in
tkp =1 deciding the channel state for a particular sensing instant, the
pTk measured state at the last sensing instant, te − 1 and the probabil-
′k ities, P′idle and P′busy , which denote the past channel statistics are
∑
Nbusy = bktp .wktp ; wktp = bkt
p/η (13)
considered. Thus, the state, co , at time te is represented according
tkp =1
to Eq. (19).
where, Tp denotes the total number of transition periods, and the
0 c0te −1 < θl
⎧
H0
number of idle and busy states measured in tp is denoted by itp ⎪
⎪
⎪
⎪
and btp , respectively. Here, Tp is calculated over the measurement
⎪
⎪
⎪
⎪ ) ( )
period tm, and itp and btp is calculated considering each sensing > P′idlet −1
(
c0te −1 = 1 ∩ P′busy
⎪
⎪
⎪ H1
⎪ te −1 e
period, ts over tp and sensing time, te . For weightwktp , controlling
⎪
⎪
⎪ ( )
< P′idlet −1
⎪ ( )
c0te −1 = 1 ∩ P′busy H0
factor η = 10. Thus, the mean idle rate considering the weighted
⎪
⎪
te −1
⎨ e
cote = z θl ≤ c0te −1 ≤ θu (
ON states measurements, N′idle , and mean busy rate consider- ) ( ′ )
⎪ c0te −1 = 0 ∩ Pbusy > Pidlet −1
′
H1
ing the weighted OFF states measurements, N′busy , are calculated
⎪
⎪ te −1 e
⎪
⎪ ( )
⎪
according to Eqs. (14) and (15). Furthermore, the channel state < P′idlet −1
( )
c0te −1 = 0 ∩ P′busy H0
⎪
⎪
⎪ te − 1
⎪ e
transition probabilities are calculated based on Eqs. (16) and (17).
⎪
⎪
⎪
⎪
⎪
N′idle
⎪
⎪
⎪
λ′idle = 1 c0te −1 > θu H1
⎩
(14)
Nidle + N′busy
′
(19)
N′busy
λbusy =
′
(15)
N′idle + N′busy 4.2. ESUCR: Channel measurement
λidle ′ λ′busy
( )
− λ′idle +λ′busy t
P′idle (t) = + ′ e (16) ESUCR jointly considers the probability, P′idle and the total
λ′idle + λ′busy λidle + λ′busy number of transitions observed over a channel, for performing
λ′busy λ′idle
( ) channel selection. The transition count, TCk for channel k is con-
− λ′idle +λ′busy t
P′busy (t) = ′ + e (17) sidered to determine the behavior of a channel with respect
λidle + λ′busy λ′idle + λ′busy to channel handoff. A channel observed with high TCki,b is less
Here, the addition of Eqs. (16) and (17) yields Eq. (18). stable, and a channel is considered with more stability with lesser
frequency of channel state transitions. Here, TCki,b denotes the
P′idle + P′busy = 1 (18) total number of idle state to busy state transitions observed over
96 T. Stephan, F. Al-Turjman, K. Suresh Joseph et al. / Journal of Parallel and Distributed Computing 142 (2020) 90–105

Fig. 2. Idle–busy alternating channels.

the kth channel. Thus, ESUCR performs channel selection based on With K number of channels, where K = list_channel and k ∈
channel stability, CS, which is calculated for channel k according {1,[. . . , K}, and channel] vacancy matrix cvmSU (t)
to Eq. (20). = cSU 0 (t) , c0 (t) , . . . , c0 (t) . Here, CS
SU SU SU
denotes the channel
1 2 K

′k
stability matrix for the list of channels sensed at the SU, and
k
CS = Pidle + TCki,b (20) c0k takes the binary value 1 if the kth channel is sensed idle,
otherwise 0. Higher ranked channels are subject to higher con-
The list of channels are ranked according to the CS metric and tention among SUs which could be resolved by operating them
the channel with the maximum CS value is given the highest in CSMA-CA mode.
rank and considered as the optimal channel for data transmission
subject to its availability at a particular transmission instance. 4.3. ESUCR: Spectrum aware clustering
Fig. 2 illustrates 10 channels with different traffic patterns which
are considered to identify and rank the channels according to the ESUCR aims to balance the energy consumed during intra-
CS metric which is calculated considering 30 sampling periods of cluster and inter-cluster communication. A higher number of
past PU traffic on each channel. clusters in the CRSN increase the inter-cluster energy consump-
Table 1 shows the efficiency of ESUCR’s channel ranking tech- tion, and a higher number of SUs within a cluster increases the
nique, which uses weighted state transition probabilities over the intra-cluster energy consumption. In ESUCR, the SUs exchange
traditional channel ranking technique, which uses unweighted their information with the neighbors to compute the minimum
state transition probabilities. A channel’s state (idle/busy) is de- distance between any two SUs and between any two clusters,
cided based on its availability status at the channel availabil- which allows merging of two neighboring clusters. In a cluster, it
ity matrix (CAM), which is updated through periodic sensing is important to ensure the CHs are rotated since the SUs selected
according to Eq. (21). as CHs consume the maximum energy and it is necessary not to
restrict any SU as CH for the entire rounds of cluster operation.
CAMSU (t) = cvmSU × CSSU
[ ] [ ]
(21) Hence, at the start of each round in a cluster, ESUCR intelligently
 T. Stephan, F. Al-Turjman, K. Suresh Joseph et al. / Journal of Parallel and Distributed Computing 142 (2020) 90–105 97

Table 1
Channel ranking based on CS.
Ch. Channel ranking based on Channel ranking based on
No unweightedstate transition weighted state transition
probabilities probabilities
Nidle Nbusy Pidle Pbusy CU Ch. N′idle N′busy P′idle P′busy TCi,b CS Ch.
Rk Rk
1 20 10 0.67 0.33 0.223 4 16.6 3.6 0.822 0.178 3 0.274 2
2 20 10 0.759 0.241 0.189 5 11.6 2.6 0.817 0.183 4 0.204 4
3 24 6 0.408 0.592 0.689 1 19.2 1.2 0.941 0.059 3 0.314 1
4 9 21 0.806 0.194 0.268 3 3.3 11.3 0.226 0.774 3 0.075 8
5 6 11 0.811 0.189 0.27 2 12.1 3.3 0.786 0.214 4 0.262 3
6 19 23.6 0.314 0.686 0.078 8 11.6 3.4 0.773 0.227 4 0.193 5
7 10 20 0.649 0.351 0.162 5 3.3 9.3 0.262 0.738 4 0.066 9
8 12 18 0.545 0.455 0.182 6 4.8 10.8 0.308 0.692 3 0.103 7
9 15 15 0.21 0.79 0.07 9 7.5 7.5 0.5 0.5 3 0.167 6
10 15 15 0.208 0.792 0.026 10 3.1 3.1 0.5 0.5 8 0.063 10

rotates the CH by assigning ranks to each SU within the cluster To minimize the network-wide energy depletion, the optimal
boundary and selecting the SU with the highest rank. Thus, for a number of clusters, CLRK can be analytically determined for an
cluster, CLRl , the rank of a SU a inside CLRl is calculated according evenly distributed CRSN. For analyzing the simulation, the opti-
to Eq. (22). mal number of clusters [45], CLRK can be derived according to
(
1
) Eq. (26).
RSUa = RESUa × ICCSkSUa × NCLSSUa × (22) ⏐ ⏐
d(SUa ,sn ) opt NSU
+ 0.5⏐⏐
⏐ ⏐
CLRL = ⏐⏐ √ (26)
The residual energy of SUa can be calculated according to TRmax 3ρ
Eq. (23).
4.4. ESUCR: Energy aware clustering
REcre
SUa
RESUa = (23)
max To efficiently avoid interference to nearby PUs, ESUCR im-
REcre
∀i ∈ NSU
CLR
i
l
plements cooperative spectrum sensing among the SUs. Each
SU independently performs energy efficient spectrum sensing
where, REcre
SUa denotes current residual energy of SUa and NCLRl
SU
(Section 4) on all the channels in list_channel and maintains
denotes the number of SUs inside CLRl .
the list of available channels. Following the clustering process,
From Eq. (24), the intra-cluster channel stability, ICCS, for
each member SU of a cluster sends the avl_list_channel to their
channel k with respect to SUa inside cluster CLRl , is calculated,
respective CH for finalizing the decision on channel availability.
only considering the channels which are vacant and common to
To minimize re-clustering frequency and energy depletion, CHs
the other SUs of CLRl .
appropriately selects the channel for intra-cluster communica-
SU
NBCLRa tion and gateway SUs to reach the neighboring clusters, and
l
Σ
( )
CSkSUa × CSki thus forms steady routing paths. The selection of intra-cluster
i=1
ICCSkSUa = ( ) (24) communication channel is as follows,
NSU
CLR − 1
1. Each SU performs independent sensing on list_channel ac-
l
cording to Section 4.1.3.
where, CSkSUa denotes the channel stability of channel k computed 2. The SUs in CRSN are grouped into clusters according to
SU Algorithm 1. The required communication for grouping the SUs
at SUa . NBCLRa denotes the neighbor count of SUa with regard to
l into clusters is performed using the common control channel.
CLRl .
3. Using the common control channel, each member SU of the
From Eq. (25), the neighbor cluster strength, NCLS for SUa
cluster sends the independently sensed results to their respective
inside cluster CLRl , is calculated considering the neighbor connec-
CH for collaboratively finalizing the decision on the current state
tivity of SUa with the members of the other clusters. Thus, NCLS
of the channels.
denotes the ability of a SU inside a cluster to connect with the
4. The CH selects the optimal channel, copt , according to Eq.
adjacent clusters.
(27), and uses the common control channel to declare it as
NNBCLR
SUa the communication channel for carrying out intra-cluster data
Σ CCCLR
j=1 j transmission, i.e., from CMs to CHs.
NCLSSUa = (25) Each SU joins a disjoint cluster based on Algorithm 1. At the
max
CCiCLR start of every round, the CHs of the respective clusters send con-
∀i ∈ NSU
CLR
l
l trol information which contains the size of the cluster and unoc-
where, NNBCLR denotes the total number of neighboring clusters cupied channels common to the cluster. Furthermore, through the
SU
and CCCLRa denotes the cluster connectivity, i.e., number of neigh- neighboring SUs, the distance between the clusters is also deter-
j
bors in the neighboring cluster, CLRj . d(SUa ,sn ) denotes the distance mined. Based on the number of commonly unoccupied channels
between SUa and sink. and measured distance between the clusters, a merge invitation is
The SU with the highest Ra is selected as the CH and the other sent by each CH to the optimal neighbor cluster. On acceptance
SUs within CLRl report to the SU selected as CH in CLRl . Further- of merge request, two clusters are merged into a single cluster
more, the CH transmits the aggregated data from the SU, which and further trigger the selection of a new CH for the newly
detected the event to the sink through the vacant channels shared formed cluster. Also, CH rotation is performed to ensure energy
with the gateway SU, which belongs to an upstream cluster. The balance among the SUs inside the cluster. Moreover, each SU
network-wide intra-cluster and inter-cluster energy consumption performs periodic sensing on list_channel and on discovering the
in communication can be calculated according to [40]. unavailability of common channels with other CMs, it isolates
98 T. Stephan, F. Al-Turjman, K. Suresh Joseph et al. / Journal of Parallel and Distributed Computing 142 (2020) 90–105

opt
from the current cluster and further based on the value of CLRL ,
it decides to form a new cluster or remain isolated.

4.5. ESUCR: Channel selection

For each cluster, CLRl , the respective CH, CHa ranks the chan-
nels contained in avl_list_channel based on the CS metric (Sec-
tion 4.2), accessibility to maximum number of SUs in CLRl , and
connectivity to maximum number of SUs residing in neighboring
clusters. Thus, a channel k in avl_list_channel is ranked according
to Eq. (27), where a higher value of Rk denotes higher rank.

Rk = CSk × NSUk × N′SUk (27)

where NSUk denotes the number of SUs that can use k in CLRl
for carrying intra-cluster communication and N′SU denotes the
k
number of SUs that can be reached in the neighboring cluster of
CLRl via k. Algorithm 2 presents the CS based channel ranking
technique of ESUCR.

4.6. ESUCR: Routing

Following cluster formation, CHs employ TDMA scheme to

allow channel access for intra-cluster communication and CSMA
scheme for inter-cluster communication. The routes are estab-
lished by the CHs and data is transmitted in transmission rounds.
In each TDMA frame, the channels are scheduled by the CH over
the time slots. Since ESUCR works on event-driven applications, Fig. 3. Illustration of forwarder selection.
proactive routing is not performed and thus the routes are es-
tablished and stored on-demand, i.e., only on the occurrence of
an event. Following route establishment, the CHs and gateways where, SUGW denotes the weight of candidate gateway SU, P,
P
are used to route the packets from the event area to the sink. In and KSUP,A denotes the number of commonly available channels
ESUCR, the CM list is not exchanged among the CHs since it works between P and CH, A. CSkSU denotes the measured channel sta-
P
on the assumption that each SU knows its neighbors which are
bility of the common available channel at P and CSkSU denotes
a one-hop distance away. The primary and secondary gateways A
the measured channel stability of the common available channel
for data forwarding are decided by the CHs using the one-hop
at A. RESUP (based on Eq. (23)) denotes the residual energy of P
neighbor information. To achieve this, neighbors of the CH receive
and d(SUP ,sn ) denotes the distance between P and sink. Thus, SU
beacon signals from the respective CH, for which it replies back
with the highest weight is selected as the gateway SU. In case, no
with the one-hop neighbor information. Furthermore, the poten-
primary gateway is discovered, then a relay SU is selected from
tial candidate primary and secondary gateway SUs are decided
the list candidate relay SUs. The selection of relay SU is performed
based on its location and used for inter-cluster data transmis-
according to Eq. (29).
sion. Thus, intra-cluster communication is enabled by the CHs
and inter-communication among adjacent clusters is enabled by KSU
R
the gateway SUs. Two scenarios are possible concerning inter- Σ CSkSUR SU
NBCCLRRm
k=1
cluster transmission. In the first scenario, primary gateway SU is SURL
R = RESUR × × (29)
max KSU
R max SU
selected from a list of candidate SUs, which are members of the Σ CSkSU NBCCLRi
destination CH and reachable to both the adjacent CHs. In the ∀i ∈ NSU k=1 i ∀i ∈ NSU
CLRl
l
CLR l
second scenario, if a CH cannot discover any candidate primary
On detecting an event, the event detectors perform data transmis-
gateway SU to reach the neighboring cluster, then it can use a
sion to their respective CHs. Subsequently, CHs discover the list
relay SU, which is a member of the current CH to relay the data to
of potential gateway SUs that are within its communication range
the secondary gateway SU, which is a member of the destination
and a member to the adjacent cluster. In case candidate primary
CH, and finally forward the packet to the neighboring CH. For
gateway SU set is empty, data forwarding is performed through
each data packet, this process of inter-cluster communication
relay SUs. Eventually, messages are delivered to the destination
continues until it reaches the sink. In Fig. 3, consider {R}, {P, S,
via this forwarding process. Algorithm 3 presents the routing
Q} and {U, T} to be the CMs of respective CHs A, B and C in
procedure of ESUCR.
respective clusters CLRl , CLRm , and CLRn . Here, A can use either
P or Q as the primary gateway SU to reach B. A cannot reach C
since there is no possibility of direct transmission or availability 5. Performance comparison and simulation results
of primary gateway SUs. Hence, A uses R as the relay SU to reach
the secondary gateway SU, U and enables communication with C. To verify the efficiency of the proposed ESUCR protocol for
Once the candidate gateway SUs is decided, the optimal gate- CRSNs, the performances of our algorithms are compared to
way SU for inter-cluster communication is selected based on EACRP [40], ERP [37], and ESAC [26] schemes. In our simulation
Eq. (28). environment, different CR layers are implemented such as physi-
cal, MAC and network layers. The physical layer performs sensing
KSU ( )
P, A 1 on the list of channels. The MAC layer ensures multiple channel
SUGW = Σ CSkSUP × CSkSUA × RESUP ×
( )
P (28)
k=1 d(SUP ,sn ) access and tracks PU traffic and packet collisions. The network
 T. Stephan, F. Al-Turjman, K. Suresh Joseph et al. / Journal of Parallel and Distributed Computing 142 (2020) 90–105 99

layer is responsible for the selection of optimal channel by our second phase routes the data packets to the sink using an optimal
channel selection technique. EACRP is a cluster-based routing path. EACRP considers PU interference during intra-cluster and
protocol aimed for an energy-efficient CRSN. The first phase of inter-cluster communication. ECR is a cognitive aware cluster-
EACRP groups the SUs into different clusters based on certain pa- based event-driven routing protocol which considers energy and
rameters and selects an intra-cluster communication channel. The channel parameters during channel allocation among the sensor
100 T. Stephan, F. Al-Turjman, K. Suresh Joseph et al. / Journal of Parallel and Distributed Computing 142 (2020) 90–105

nodes. ESAC maximizes the number of accessible and stable clusters. For conducting this simulation, SUs are varied in the
channels during cluster construction and performs optimized range [140, 160] and PUs in the range [8, 12] for different sim-
routing between the clusters. ESAC is an event-driven protocol ulation runs with uniform distribution of SUs and PUs. From the
which discovers the common channels between clusters for data simulation results, it is observed that minimum energy consump-
routing. ERP is a reactive event-driven cluster-based protocol tion is achieved when the cluster count is in the range [1,39].
which constructs clusters only after detection of an event and When cluster count in the range [24,47], the boundary of the
performs routing between the event detectors and the sink. cluster increase which in turn increases the energy utilization
The ESUCR’s performance is compared among EACRP, ERP, and for intra-cluster communication. This is because of the increased
ESAC in terms of average end-to-end packet delay, average SU en- distance between the CH and the SUs within the cluster. Also,
ergy consumption, the strength of the gateway SUs and delivery the energy consumption increases when cluster count is in the
ratio of packets. Table 2 represents the parameters considered for range [4,13] which is because of increased energy spent on
simulation. performing inter-cluster communication. On the average, it is
realized that the ESUCR always outperforms EACRP, ERP, and
5.1. ESUCR: Energy consumption ESAC related to minimizing average energy consumption at dif-
ferent number of clusters by 10.8%, 18.6%, and 29.3% compared
The efficiency of ESUCR in terms of energy utilization is to EACRP, ERP, and ESAC respectively.
demonstrated against EACRP, ERP, and ESAC under a varying Fig. 5 represents the average per node energy consumption
number of events, clusters, PUs, and SUs. Fig. 4 represents the for varying number of PUs. For this simulation, SUs are varied
average per node energy consumption for varying number of in the range [140, 160] for different simulation runs. From the
 T. Stephan, F. Al-Turjman, K. Suresh Joseph et al. / Journal of Parallel and Distributed Computing 142 (2020) 90–105 101

Table 2
Simulation parameters.

Parameter Value

Simulation area 1000 ×1000

Simulation time 1000 s
Node deployment Uniform random
Application type Event-driven
Traffic type CBR
Size of packet 0.001 KB
Transmission rate of packet 10–20 packets/s
Number of event occurrence 2–20
Location of sink (500, 500)
Number of PUs 2–20
Transmission range of PUs 200 m
Number of channels 10 Fig. 5. Average energy consumption versus varying number of PUs.
Channel occupancy probability of PUs 0.25
Channel selection strategy of PUs Random selection
Number of SUs 20–200
Transmission range of SUs 100 m
SU initial energy 1000 J
Back off mechanism CSMA/CA
Channel sensing time 0.005 s
Energy utilization for sensing a channel 0.131 mJ
Energy utilization for switching a channel 0.01 mJ
Energy utilization for data reception 5×10−6 mJ/bit
Channel bandwidth 1.024 Mbps
Number of simulation runs 100

Fig. 6. Average energy consumption versus varying number of SUs.

higher energy due to high routing load since the number of SUs
participating in the routing process is more. ESUCR performs
better in terms of minimizing energy consumption as compared
to other event-driven protocols namely, EACRP and ERP which
is mainly because of ESUCR’s ability to optimally cluster the
deployed SUs. On average, it is observed that ESUCR achieves
significantly lower average energy consumption of up to 21.7%
as compared to EACRP and 49.9% and 68.2% as compared to ERP
and ESAC, respectively.
Fig. 4. Average energy consumption versus varying number of clusters. Fig. 7 represents average per node energy consumption for
varying number of events. For this simulation, SUs are varied
in the range [140, 160] and PUs in the range [8, 12] for differ-
simulation results, it is observed that with low PU count, the ent simulation runs. With an increased number of events, the
number of clusters formed in the network is less which is due amount of energy consumed for data transmission increases since
to low levels of PU interference, and with high PU count, SUs more number of packets is forwarded to the sink. Owing to
are grouped into higher number of clusters due to increased PU the optimal number of cluster generation and implementing an
activity which further maximizes the energy depletion in CRSN. energy-efficient intra-cluster and inter-cluster communication,
ESUCR outperforms the other related schemes under a vary- ESUCR consumes lesser energy as compared to EACRP, ERP, and
ing number of PUs and on the average achieves significantly ESAC. Moreover, ESUCR senses interference from PUs in differ-
lower average energy consumption of up to 14.9% as compared ent channels with minimum energy consumption which is due
to EACRP and 31.3% and 45.1% as compared to ERP and ESAC, to its ability to perform energy-efficient spectrum sensing with
respectively. reduced sensing time. Also, intelligent channel selection policy of
Fig. 6 represents the average per node energy consumption for ESUCR reduces the frequency of channel switching since a certain
varying number of SUs. For this simulation, the number of PUs amount of energy is consumed in a single channel switch. Com-
is varied in the range [8, 12] and the number of events in the pared to EACRP, ESUCR attains lower average per node energy
range [1,38] for different simulation runs. Here, ESAC consumes consumption of up to 20.5% and of up to 36.2% and 55.6% as
102 T. Stephan, F. Al-Turjman, K. Suresh Joseph et al. / Journal of Parallel and Distributed Computing 142 (2020) 90–105

number of clusters, PUs and SUs. Fig. 9 shows the comparison

of ESUCR, EACRP, ERP, and ESAC with respect to average end-to-
end delay under a varying number of PUs. For conducting this
simulation, SUs are varied in the range [140, 160] for different
simulation runs with uniform distribution of SUs and PUs in
CRSN. End-to-end delay refers to the time taken for a packet to be
transmitted across CRSN from source to sink. All the four schemes
taken for comparison are cluster-based schemes and when com-
pared to non-cluster based techniques [25,35], incurs a bit higher
delay. This is because in cluster-based routing protocols, a CM
performs data forwarding to its respective CHs, following which
the CH forwards the data packet to the selected gateway SU,
which further forwards the packet to its neighboring cluster.
From the simulation results, it is observed that the average end to
end delay is less for all the schemes with lesser number of PUs in
Fig. 7. Average energy consumption versus varying number of events.
the CRSN. This because when PU activity level is low, i.e. the num-
ber of PUs in the range [0, 10], all schemes achieve significantly
lower average end-to-end delay because no route is affected by
PUs’ activities and all schemes can transmit their packets without
many disruptions. As the PU count increases, the interference due
to PUs increases which in turn maximizes the delay of packets
reaching the sink for EACRP, ERP, and ESAC. These schemes,
despite being spectrum-aware, the selected routes may constitute
channels with lower idle probability and hence there is a high
possibility of PUs appearance on the selected channels in the next
time slot and results in packet collision. Whenever a SU’s packet
interferes with a PU’s packet, both PU’s and SU’s packets are lost
which increases the SU–PU interference ratio. ESUCR is not highly
affected by SU–PU interference since it selects stable routes which
comprise of channels predicted with higher idle probability for
the next time slot. Moreover, the sensing accuracy of ESUCR
minimizes the chances of PU interference. SU–PU interference
ratio is defined as the ratio of the total number of SUs’ packets
interfered with PUs’ activities to the total number of packets sent
Fig. 8. Average stability of gateways versus varying number of PUs. by a SU source node. Though ERP scheme exploits the availabil-
ity of multiple channels, its unstable route selection technique
increases the rerouting frequency which is mainly affected due
compared to ERP and ESAC, respectively, at different number of to the inefficient selection of channels over the wireless links.
events. EACRP ensures forming an optimal number of clusters and routes
the event data to the sink. Also, it performs gateway selection
5.2. ESUCR: Gateway strength considering the number of common channels available between
the nodes. However, its channel selection policy requires more
Fig. 8 represents the strength of chosen gateways for vary- intelligence to alleviate the effects of PU interference and link fail-
ing number of PUs. For this simulation, SUs are varied in the ures which affect the established routes with more vulnerability
range [140, 160] for different simulation runs. The strength of of being broken. The route selection policy of ESAC is inefficient
the gateway is measured using Eq. (28), where the gateway is a since it performs route selection without consideration of PU
CH or a CM selected for forwarding the packets. It is observed behavior during route formation. Thus, the SUs may choose non-
that due to the random gateway selection policy of ESAC, the stable routes that have channels with lower idle probability and
selected gateway SUs underperform in forwarding the packets to hence drops a higher number of packets which triggers frequent
the sink. ERP protocol does not take into account the number of retransmission and increases the delay of packets reaching the
common channels which results in unstable gateways selected sink. According to the observed simulation results, it is realized
as forwarders. Though EACRP protocol considers the number of that ESUCR achieves a lower average end to end delay of up to
common channels, it does not consider the stability of the chan- 9.2% than EACRP and of up to 18.4% and 26.5% as compared to
nels which in turn minimizes the lifetime of the links connecting ERP and ESAC, respectively, at varying number of PUs.
the gateway SUs. ESUCR outperforms EACRP, ERP, and ESAC by Fig. 10 shows the network performance achieved by ESUCR
considering the channel stability of all the common available with EACRP, ERP, and ESAC under a varying number of SUs.
channels and helps SUs to make right decisions on SU next-hop Importantly, when the number of SUs increases in the network,
selection in routing. On varying the PU count, it is observed that the opposition for the spectrum holes becomes additionally com-
on average, ESUCR ensures more gateway stability than EACRP of pelling. As a result of this opposition, the average interference
up to 7.1% and of up to 23.4% and 93.5% as compared to ERP and per SU increases in all schemes. Based on the analysis of the
ESAC, respectively. performance of the protocols on average, it is realized that the
ESUCR always outperforms the other schemes related to mini-
5.3. ESUCR: Average end-to-end delay mizing average end to end delay at different number of SU nodes
by 6.93%, 26.9%, and 41.2% compared to EACRP, ERP, and ESAC
The efficiency of ESUCR in terms of minimizing the average respectively. The main reason for that is that EACRP, ERP, and
end-to-end delay of packets is demonstrated under a varying ESAC, do not assurance that the best channel is unexploited for
 T. Stephan, F. Al-Turjman, K. Suresh Joseph et al. / Journal of Parallel and Distributed Computing 142 (2020) 90–105 103

Fig. 9. Average end to end delay versus varying number of PUs. Fig. 11. Average end to end delay versus varying number of clusters.

EACRP, ERP, and ESAC. Fig. 12 confirms that ESUCR can provide
good network reach, suitable for increasing the forwarding relia-
bility of event data in multi-hop CRSNs. As Fig. 12 illustrates we
can realize that the ESUCR outperforms the considered schemes
related to maximizing packet delivery ratio at different event
count by 4.8%, 12.7%, and 18.5% compared to EACRP, ERP, and
ESAC respectively. This is mainly because the ESUCR highlighted
on selecting the channel in a method which guarantees that the
channel is unexploited by PU through channel stability prediction.
Significantly, the packet delivery ratio is mapped to SU–PU inter-
ference ratio and spectrum utilization metrics. Low interference
ratio means higher spectrum utilization which further allows a
higher number of SUs to successfully transmit the data packets
leading to higher delivery ratio. On the contrast, when the SU–PU
interference ratio is high, the missed and interrupted packets will
increase which leads to lower spectrum utilization. Consequently,
Fig. 10. Average end to end delay versus varying number of SUs.
there will be a lower packet delivery ratio. The rate of successful
packet transmission in ESUCR is mainly due to the accurate
selection of optimal channels using channel stability metric. Also,
transmission. Moreover, ESUCR is able to tune both sender and the average throughput is more related to the total number of
receiver to the right channel with high probability for effective successful packet received in the CRSN (i.e., packet delivery ratio).
and reliable data dissemination in multi-hop context. When the packet delivery ratio is increased, the throughput is
Fig. 11 shows the simulation and analytical results for the av- also increased and vice versa. Also note that, when the number of
erage end-to-end packet delay achieved by ESUCR by comparing events increased, the delivery ratio is decreased in all strategies.
it with the other three related approaches under a varying num- The main reason for that is that, with the increase in the number
ber of clusters. As observed in Fig. 6, the good agreement between of the CRs in the network the competition for the vacant channels
the simulation and analytical results, and the confidence interval becomes more compelling. As a result of this competition, the
is again wider at a large number of clusters. As the number of average throughput per SU decreases in all schemes.
clusters increases, the number of hops increases which results in
an increased end-to-end delay. Both ESUCR and EACRP determine
the optimum number of clusters that maximizes the lifetime 6. Conclusion
and ensures achieving the specified delay objective. Though, the
efficiency of ESUCR to select optimal channels and gateways This paper aims to minimize the energy consumption of SUs
for route formation minimizes the packet delay as compared and interference to PUs, through ESUCR which is an interfer-
to EACRP, ERP, and ESAC. On the average, we can realize that ence aware energy-efficient clustered routing protocol for event-
the ESUCR always outperforms EACRP, ERP, and ESAC related driven applications. ESUCR considers the energy efficiency of the
to minimizing end-to-end delay at different number of clusters SUs and allows the most stable available channels to be selected
by 7.2%, 14.8%, and 20.1% compared to EACRP, ERP, and ESAC as common data channels for carrying out intra-cluster commu-
respectively. nication. The primary/secondary gateway SUs for inter-cluster
data forwarding are optimally selected based on the stability of
5.4. ESUCR: Packet delivery ratio common channels, residual energy, and distance to sink node
which makes the routing path formed by ESUCR from the event
This metric is defined to effectively measure the data dissem- detectors to the sink to be more reliable. Furthermore, ESUCR
ination process. It is the ratio of packets received by a particular achieves energy efficiency in spectrum sensing by making use of
CR node over total packets sent in the network. Fig. 12 shows statistic previous channel states for decision making with regard
the results of different simulation experiments to measure the to the channel state. Also, ESUCR performs optimal CH rotation
packet delivery ratio at different numbers of events for the ESUCR, in which the role of the CH is rotated among the associated
104 T. Stephan, F. Al-Turjman, K. Suresh Joseph et al. / Journal of Parallel and Distributed Computing 142 (2020) 90–105

[14] D. Feng, C. Jiang, G. Lim, L.J. Cimini, G. Feng, G.Y. Li, A survey of
energy-efficient wireless communications, IEEE Commun. Surv. Tutor.
(2013).
[15] D. Horgan, C.C. Murphy, Voting rule optimisation for double threshold
energy detector-based cognitive radio networks, in: 4th International
Conference on Signal Processing and Communication Systems, ICSPCS,
2010.
[16] G.P. Joshi, S.W. Kim, A survey on node clustering in cognitive radio wireless
sensor networks, Sensors (Switzerland) (2016).
[17] I. Kakalou, D. Papadopoulou, T. Xifilidis, K.E. Psannis, K. Siakavara, Y.
Ishibashi, A survey on spectrum sensing algorithms for cognitive radio
networks, in: 7th International Conference on Modern Circuits and Systems
Technologies, MOCAST, 2018.
[18] H. Khalifé, N. Malouch, S. Fdida, To route or not to route, IEEE Netw. (2009).
[19] X. Li, D. Wang, J. McNair, J. Chen, Residual energy aware channel as-
signment in cognitive radio sensor networks, in: 2011 IEEE Wireless
Communications and Networking Conference, WCNC 2011, 2011.
[20] X. Li, D. Wang, J. McNair, J. Chen, Dynamic spectrum access with packet
size adaptation and residual energy balancing for energy-constrained
cognitive radio sensor networks, J. Netw. Comput. Appl. (2014).
Fig. 12. Packet delivery ratio versus varying number of events. [21] F. Liu, J. Wang, Y. Han, An adaptive double thresholds scheme for spectrum
sensing in cognitive radio networks, in: 2013 IEEE International Conference
on Signal Processing, Communications and Computing, ICSPCC 2013, 2013.
[22] Y. Liu, et al., Adaptive double threshold energy detection based on Markov
CMs according to energy, channel, distance and neighbor param- model for cognitive radio, PLoS One (2017).
eters. ESUCR achieves minimized average energy consumption, [23] S. Maleki, A. Pandharipande, G. Leus, Two-stage spectrum sensing for
end to end delay, higher average packet delivery ratio, and more cognitive radios, in: Proc. IEEEICASSP, New York, USA, 2010.
[24] K.B.S. Manosha, N. Rajatheva, M. Latva-Aho, Overlay/underlay spectrum
gateway stability than EACRP, ERP and ESAC protocols, which sharing for multi-operator environment in cognitive radio networks, in:
has been evaluated through an extensive simulation study in IEEE Vehicular Technology Conference, 2011.
NS-2. The future work will focus on testing and improving our [25] C.H.W. Oey, I. Christian, S. Moh, Energy- and cognitive-radio-aware routing
algorithms to acquire more robust performance with adaptations in cognitive radio sensor networks, Int. J. Distrib. Sens. Netw. (2012).
to various PU spectrum occupancy models. [26] M. Ozger, O.B. Akan, Event-driven spectrum-aware clustering in cognitive
radio sensor networks, in: Proceedings - IEEE INFOCOM, 2013.
[27] M. Ozger, E. Fadel, O.B. Akan, Event-to-sink spectrum-aware clustering in
Declaration of competing interest mobile cognitive radio sensor networks, IEEE Trans. Mob. Comput. (2016).
[28] E. Pei, H. Han, Z. Sun, B. Shen, T. Zhang, LEAUCH: low-energy adaptive
uneven clustering hierarchy for cognitive radio sensor network, Eurasip J.
The authors declare that they have no known competing finan- Wirel. Commun. Netw. (2015).
cial interests or personal relationships that could have appeared [29] M.H. Rehmani, A.C. Viana, H. Khalife, S. Fdida, SURF: A distributed channel
to influence the work reported in this paper. selection strategy for data dissemination in multi-hop cognitive radio
networks, Comput. Commun. (2013).
[30] J. Ren, Y. Zhang, N. Zhang, D. Zhang, X. Shen, Dynamic channel access to
References improve energy efficiency in cognitive radio sensor networks, IEEE Trans.
Wirel. Commun. (2016).
[1] J. Abolarinwa, N.M.A. Latiff, S.K.S. Yusof, N. Fisal, Energy-efficient, learning- [31] Y. Saleem, K.L.A. Yau, H. Mohamad, N. Ramli, M.H. Rehmani, SMART: A
inspired channel decision and access technique for cognitive radio-based SpectruM-Aware ClusteR-based rouTing scheme for distributed cognitive
wireless sensor networks, Int. J. Multimedia Ubiquitous Eng. (2015). radio networks, Comput. Networks (2015).
[2] A. Bagwari, G.S. Tomar, S. Verma, Cooperative spectrum sensing based on [32] T.M. Salem, S. Abdel-Mageid, S.M. Abdel-Kader, M. Zaki, ICSSSS: An in-
two-stage detectors with multiple energy detectors and adaptive double telligent channel selection scheme for cognitive radio ad hoc networks
threshold in cognitive radio networks, Can. J. Electr. Comput. Eng. (2013). using a self organized map followed by simple segregation, Pervasive Mob.
[3] W. Bin Chien, C.K. Yang, Y.H. Huang, Energy-saving cooperative spectrum Comput. (2017).
sensing processor for cognitive radio system, IEEE Trans. Circuits Syst. I. [33] G.A. Shah, O.B. Akan, Spectrum-aware cluster-based routing for cog-
Regul. Pap. (2011). nitive radio sensor networks, in: IEEE International Conference on
[4] A. Bradai, K. Singh, A. Rachedi, T. Ahmed, EMCOS: Energy-efficient mech- Communications, 2013.
anism for multimedia streaming over cognitive radio sensor networks, [34] G.A. Shah, F. Alagoz, E.A. Fadel, O.B. Akan, A spectrum-aware clustering
Pervasive Mob. Comput. (2015). for efficient multimedia routing in cognitive radio sensor networks, IEEE
Trans. Veh. Technol. (2014).
[5] D. Cabric, A. Tkachenko, R.W. Brodersen, Experimental study of spectrum
[35] T. Stephan, K.S. Joseph, Cognitive radio assisted OLSR routing for vehicular
sensing based on energy detection and network cooperation, 2007.
sensor networks, Procedia Comput. Sci. 89 (2016) 271–282.
[6] C.X. Chen, H. Fu, Z.D. Niu, Cooperative spectrum sensing algorithm based
[36] T. Stephan, K.S. Joseph, Particle swarm optimization-based energy effi-
on double-threshlod energy detection, Syst. Eng. Electron. 35 (8) (2013)
cient channel assignment technique for clustered cognitive radio sensor
1742–1746.
networks, Comput. J. 61 (6) (2018) 926–936.
[7] T. Chen, H. Zhang, G.M. Maggio, I. Chlamtac, CogMesh: A cluster-based
[37] M. Tabassum, M.A. Razzaque, M.N.S. Miazi, M.M. Hassan, A. Alelaiwi, A.
cognitive radio network, in: 2007 2nd IEEE International Symposium on
Alamri, An energy aware event-driven routing protocol for cognitive radio
New Frontiers in Dynamic Spectrum Access Networks, 2007.
sensor networks, Wirel. Netw. (2016).
[8] Z. Chkirbene, N. Hamdi, A survey on spectrum management in cognitive [38] M. Tabassum, M.A. Razzaque, M.N.S. Miazi, M.M. Hassan, A. Alelaiwi, A.
radio networks, Int. J. Wirel. Mob. Comput. 8 (2) (2015) 153–165. Alamri, An energy aware event-driven routing protocol for cognitive radio
[9] S. Cui, A.J. Goldsmith, A. Bahai, Energy-constrained modulation optimiza- sensor networks, Wirel. Netw. (2016).
tion, IEEE Trans. Wirel. Commun. (2005). [39] G.I. Tsiropoulos, O.A. Dobre, M.H. Ahmed, K.E. Baddour, Radio resource
[10] S. Dhiviya, A. Sariga, P. Sujatha, Survey on WSN using clustering, in: allocation techniques for efficient spectrum access in cognitive radio
Proceedings - 2017 2nd International Conference on Recent Trends and networks, IEEE Commun. Surv. Tutor. (2016).
Challenges in Computational Models, ICRTCCM 2017, 2017. [40] R. Yadav, R. Misra, D. Saini, Energy aware cluster based routing protocol
[11] P. Dutta, D. Culler, Practical asynchronous neighbor discovery and over distributed cognitive radio sensor network, Comput. Commun. (2018).
rendezvous for mobile sensing applications, 2008. [41] P. Yawada, M. Dong, Intelligent process of spectrum handoff/mobility in
[12] O. Ergul, O.B. Akan, Energy-efficient cooperative spectrum sensing for cog- cognitive radio networks, J. Electr. Comput. Eng. (2019).
nitive radio sensor networks, in: Proceedings - International Symposium [42] T. Yücek, H. Arslan, A survey of spectrum sensing algorithms for cognitive
on Computers and Communications, 2013. radio applications, IEEE Commun. Surv. Tutor. (2009).
[13] E. Fadel, et al., Spectrum-aware bio-inspired routing in cognitive radio [43] Y. Zeng, Y.C. Liang, Spectrum-sensing algorithms for cognitive radio based
sensor networks for smart grid applications, Comput. Commun. (2017). on statistical covariances, IEEE Trans. Veh. Technol. (2009).
 T. Stephan, F. Al-Turjman, K. Suresh Joseph et al. / Journal of Parallel and Distributed Computing 142 (2020) 90–105 105

[44] X. Zhang, X. Liu, H. Samani, B. Jalaian, Cooperative spectrum sensing in Journal for the period 2015–2018, in addition to the Top Researcher Award for
cognitive wireless sensor networks, Int. J. Distrib. Sens. Netw. (2015). 2018 at Antalya Bilim University, Turkey. Prof. Al-Turjman has led a number
[45] H. Zhang, Z. Zhang, H. Dai, R. Yin, X. Chen, Distributed spectrum-aware of international symposia and workshops in flagship communication society
clustering in cognitive radio sensor networks, in: GLOBECOM - IEEE Global conferences.
Telecommunications Conference, 2011.
[46] D. Zhang, et al., Energy-harvesting-aided spectrum sensing and data
transmission in heterogeneous cognitive radio sensor network, IEEE Trans.
Dr. K. Suresh Joseph is currently working as an
Veh. Technol. (2017).
Associate Professor in the Department of Computer
[47] S. Zubair, N. Fisal, Y.S. Baguda, K. Saleem, Assessing routing strategies for
Science, Pondicherry University. He completed his B.E.
cognitive radio sensor networks, Sensors (Switzerland) (2013).
in Computer Science & Engineering from the University
of Madras and M.E. from Bharathiyar University, Tamil
Nadu, India. He received the Ph.D. degree in Infor-
Dr. Thompson Stephan is currently holding the po-
mation and Communication Engineering from Anna
sition of Assistant Professor in the Department of
University, India. His research interests include Soft
Computer Science and Engineering from Amity Uni-
Computing, Mobile and Wireless Communications and
versity, Noida, India. He received his B.E in Computer
sensor networks.
Science and Engineering and M.E in Computer Sci-
ence and Engineering from Anna University, India. He
received his Ph.D. degree in Computer Science and
Engineering from Pondicherry University, India. His re- Dr. Balamurugan Balusamy is currently holding the
search interests include Nature-Inspired Metaheuristics, position of Professor in the School of Computer Sci-
Cognitive Radio communications, Ad Hoc and Sen- ence and Engineering, Galgotias University, Greater
sor Wireless Networks, Vehicular communications, and Noida, India. He had completed his B.E. in Computer
Internet of Things. Science from Bharathidasan University and M.E. in
Computer Science from Anna University. He did his
Ph.D. in Computer Science and Engineering from VIT
Prof. Dr. Fadi Al-Turjman received his Ph.D. in University, India. His research interests include Mobile
computer science from Queen’s University, Kingston, and Wireless communications, Cloud computing and
Ontario, Canada, in 2011. He is a full professor and Security.
a research center director at Near East University,
Nicosia, Cyprus. Prof. Al-Turjman is a leading authority
in the areas of smart/intelligent, wireless, and mobile
networks’ architectures, protocols, deployments, and Dr. Sweta Srivastava received the B.E. degree in
performance evaluation. His publication history spans Computer Technology from Nagpur University. She
over 250 publications in journals, conferences, patents, Graduated M.E. and Ph.D. from B.I.T Mesra, Ranchi,
books, and book chapters, in addition to numerous India. Her major field of study is in Computer Science.
keynotes and plenary talks at flagship venues. He has She is working in ASET, AMITY University Noida as
authored and edited more than 25 books about cognition, security, and wireless an assistant professor in the department of Computer
sensor networks’ deployments in smart environments, published by Taylor Science. Her research interests include soft computing,
and Francis, Elsevier, and Springer. He has received several recognitions and computational intelligence, machine learning and ar-
best papers’ awards at top international conferences. He also received the tificial intelligence. She has authored several research
prestigious Best Research Paper Award from Elsevier Computer Communications papers in the field of Computer Science.