This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3065691, IEEE Internet of
Things Journal

Energy Efficient k-hop Clustering in Cognitive

Radio Sensor Network for Internet of Things
Rajendra Prajapat1 , Ram Narayan Yadav2 , and Rajiv Misra3
1 Department
of CSE, Indian Institute of Information Technology (IIIT), Dharwad, India
2 Computer Science & Engineering, Institute of Infrastructure Technology Research And Management (IITRAM), India
3 Department of CSE, Indian Institute of Technology (IIT), Patna, India

Abstract—The design and development of energy and spectrum radios are equipped with spectrum sensing capabilities to
efficient solutions are important in the success of Internet of detect the idle spectrum. Thus, they can detect and utilize
Things (IoT). Due to presence of enormous number of smart de- the available spectrum opportunity by dynamically changing
vices such as sensors, actuators, and different household devices
achieving such scalable and efficient solutions are challenging. A their parameters. The cognitive radio sensor network (CRSN)
wireless sensor network (WSN) with dynamic spectrum access [4]–[6], a combination of cognitive radio network (CRN) and
(DSA) capability, known as cognitive radio sensor network wireless sensor network (WSN) has emerged as a promising
(CRSN) is recently introduced to deal with spectrum scarcity solution for spectrum scarcity. In CRSN, sensor nodes are
problem. Although the spectrum scarcity is reduced with DSA equipped with cognitive radio and can take advantage of DSA.
paradigm, the energy efficient solutions are still required to be
addressed due to involvement of energy constrained devices in Sensor nodes perform periodic spectrum sensing and detect the
CSRN. Clustering is one of the efficient ways to optimize the idle channels and can communicate over licensed bands owned
energy consumption in the networks. Due to combination of both by primary users (PUs). These sensor nodes are unlicensed
WSN and cognitive radio network (CRN), existing solutions of users and also called as secondary users (SUs). If PUs arrive
WSN and of CRNs are not applicable to CRSN. In this paper, on their licensed channels and these channels are being used
we propose a neighbour discovery algorithm and two greedy
k-hop clustering schemes (k-SACB-WEC, and k-SACB-EC) for by SUs then they must have to perform spectrum hand-off to
CRSN with the aim focusing on IoT applications which require avoid the interference with PUs’ communication.
constant intra-cluster and inter-cluster communications. We focus Due to integration of WSN and CRN, CRSN suffers from
on achieving bi-channel connectivity while maximizing network limited energy and low computational resources (as sensor
life. In our clustering different parameters such as nodes’ nodes have limited resources). In CRN, the available channels
residual energy, spectrum awareness, appearance probability of
primary users (PUs) of channels, channel quality, robustness are dynamic and depends on PUs’ activities and location of
on PUs’ arrival, and euclidean distance betweeen nodes are SUs. This spatial and temporal variation in channel availability
taken into consideration to select the hop count and common can cause network partitioning (if channels assigned to a link
channels for clusters. Through simulation, we have highlighted is reclaimed by PU and no channel is available). Therefore,
the performance improvements of our proposed schemes in terms it is important to design a channel allocation scheme which
of the lifetime of the network, number of clusters, stability of
networks, and frequency of re-clustering over recently reported ensures that network remains connected (if possible) whenever
clustering algorithm in CRSN. The simulation results show that PUs’ reclaim their licensed channels, is called robust channel
k-SACB-WEC generates at least 40% less number of clusters connectivity [7]–[9]. We defined a CRSN to be a bi-channel
as compared to k-SACB-EC, NSAC, PSEP, SAC-WCM and connected if the nodes in the network remain connected when
CogLEACH. Also in terms of network stability, the k-SACB- any channel is reclaimed by PU and becomes unavailable to
WEC achieves at least approximately 100% higher number of
rounds before the first node dead than the compared competitive SUs. On the other hand, in WSN, sensor nodes communicate
approaches. collaboratively by sending their sensed data continuously. In
Index Terms—Cognitive radio sensor networks, clustering, CRSN, this cooperation improves the efficiency of spectrum
neighbor discovery algorithm, IoT sensing, spectrum decision and routing mechanisms etc. Fur-
ther, to reduce the energy consumption in different operations
I. I NTRODUCTION such as sensing, routing, and data transmission, clustering is
Enormous growth of IoT devices and different applications applied extensively in WSN [10]–[12]. Since these clustering
demand more spectrum and energy efficient solutions [1]. schemes and other routing schemes such [13] are designed for
According to FCC [2], the assigned spectrum are used spo- WSN and do not consider the spectrum heterogeneity issue
radically. Thus, the current static spectrum allocation policy of CRN, they are not applicable to CRSN. Therefore, it is
is not effective and leads to spectrum scarcity. To mitigate very important to develop a scheme which maintains global
the spectrum scarcity problem, a dynamic spectrum access robust channel connectivity and increases network lifetime of
(DSA) paradigm can be employed. Cognitive radio is a com- CRSN. Due to spectrum heterogeneity in time and space, and
munication technology which enable DSA [3]. The cognitive limited energy of sensor nodes, the problem of achieving bi-

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 00:01:33 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3065691, IEEE Internet of
Things Journal

channel connectivity and network lifetime maximization is • If a k-hop clustering scheme generates more number of
very challenging. As clustering is one of the efficient ways clusters then these larger number of clusters lead to higher
to optimize the energy consumption in the networks [14], we inter-cluster communication overhead.
employed clustering to address the bi-channel connectivity and • If a k-hop clustering scheme generated larger sized
network lifetime maximization in CRSN. Due to spectrum clusters (by increasing the size of cluster to cover more
heterogeneity in CRSN, traditional clustering algorithms such SUs) (i) it may reduce number of commonly available
as maximum node degree, and k-mean clustering etc. cannot channels to all SUs in a cluster. This may lead to frequent
be applied to CRSN. Recently, researchers have proposed 1- re-clustering due to PU appearance. Thus, there is a trade-
hop clustering schemes in CRSN [15]. Since 1-hop clustering off between hop value (parameter k) and number of
schemes generate large number of clusters in the networks, common channels in a cluster, (ii) When a SU is far
this may incur huge overheads in terms of inter-cluster co- from the cluster head, it has to transmit data with more
ordination and communication. Therefore, we propose k- transmission power, which leads to lower network life.
hop spectrum aware clustering for bi-channel connectivity in So, motivated as above, we propose a neighbour discovery
CRSN (called k-SACB) which maximizes the network life algorithm and two k-hop clustering schemes to address the
time. Now, we define the concept of k-hop clustering. above discussed issues in CRSN.
k-hop Clustering: A cluster is a logical group of SUs The remainder of the paper is organized as follows: Section
in which all nodes operate on a common channel. k-hop II presents the related work and Section III presents the system
clustering accommodates nodes at most k-hop distance from model, problem definition and preliminaries and assumptions.
cluster head (CH) node. In Section IV, we describe our scheme for neighbour discov-
ery, k-hop clustering with edge contraction, and another k-
A. Motivation hop clustering scheme without edge contraction. We present
simulation results in Section V and conclude the paper in the
In this section, we discuss our motivation in terms of issues
Section VI.
and possible trade-off during k-hop clustering to address the
k-SACB problem in CRSN. In general, an IoT application II. R ELATED W ORK
based on CRSN (which involves number of SUs) should be Several reported work have considered the characteristics
scalable and distributed in nature. As the number of SUs of CRSN and proposed different clustering algorithms for
increases, the complexity of managing the network increases CRSN [17]–[19]. These schemes mainly focused on bandwidth
which degrades the performance of the system. In distributed limitation, collaborative spectrum sensing, spectrum decision,
environment, various upper layer protocols (such as routing) routing, energy related issues in CSRN [20]. A detailed survey
work well with clustering [16]. So, the clustering is one of on existing clustering schemes in CRSN is discussed in [15],
the effective ways (i) to improve the network scalability, (ii) to [21]. Mustafa Ozger et al. [17] proposed an event driven
reduces the cost of network coordination, and (iii) to maximize clustering algorithm which forms temporal clusters for each
the network lifetime. However, there are many challenges for event. First, they select the set of eligible nodes that are
implementing clustering in CRSN. They are as follows: allowed in clustering process according to the positions of
• In typical network, the topology information such as nodes between event and sink. Then, CHs are selected based
neighbouring nodes and their information such as energy on the nodes’ energy, available channels and distance to the
level are exchange among nodes through broadcasting sink. Shah, Ghalib A. et al. [22] proposed a spectrum aware
on the predefined common control channel. However, clustering scheme for multimedia routing in CRSN. In the
in case of CRSN, SUs located at different locations proposed scheme, clustering is utilized to limit the number of
may have different set of available channels (depends participating nodes in route establishment and improves the
on PUs’ activities). The available channel of a node in quality of services (QoS) and energy efficiency. In the cluster-
CRSN may change over time due to PUs’ activities. ing, channels are clustered based on spectrum sensing results
So, the assumptions of having a global common control in each round and spectrum usage statistics. CH selection is
channel seems impractical. But, it is important to consider based on the spectrum awareness and the energy. They have
the network connectivity for implementing clustering in also determined the optimal number of clusters according to
CRSN. network scenario which minimizes the distortion of SU video
• If nodes in a cluster is operating on a channel and the sources. Tabassum et al. [23] proposed a event driven routing
PU arrives at that channel, all nodes in that cluster will protocol (ERP) by considering parameters such as energy, state
have to switch to another common channel or if there is and distance of neighbours, and available channels to nodes
no common channel then re-clustering is triggered. The etc. To select CH and gateways nodes the proposed clustering
overhead of re-clustering is higher than switching from algorithm considered PUs’ dynamics. However, frequency of
one channel to another channel. So, a k-hop clustering re-clustering increases with increase in occurrences of events.
scheme should generate clusters with more than one In our previous work [24], we have proposed a Energy Aware
common channels that helps in reducing the frequency Cluster based Routing Protocol (EACRP) with the aim of re-
of re-clustering. ducing the frequency of re-clustering by selecting the suitable

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 00:01:33 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3065691, IEEE Internet of
Things Journal

common channel for intra-cluster and inter-cluster communi-

cations. However, the proposed clustering algorithm does not
address the bi-channel connectivity problem. Mustafa Ozger et
al. [25] have investigated benefits and functionalities clustering
such energy, topology and spectrum managements in multi-
channel channel cognitive radio sensor networks. Liu Sisi et al.
[26] addressed the problem of control channel assignment in
CRN using clustering. They formulated the clustering problem
using bi-partite graph and developed algorithms to guarantee
the desirable number of common channels for controls. Also
the work [27] proposed a clustering scheme for control channel
establishment using one-hop neighbours and available chan- Figure 1: System model
nels. Though these schemes consider spectrum opportunities
to address PUs’ dynamics, they generate large number of
clusters as nodes in the network increases and thus suffer from mizing the network lifetime 1 .
huge inter-cluster overheads. Tianjing Wang et al. [19] have • Our proposed k-hop clustering without edge contraction
proposed a clustering algorithm based on weighted clustering (k-SACB-WEC) finds the optimal value of the parameter
metric. They established clustering metric by using temporal- k (hop parameter) for a given number of lower threshold
spatial correlation of spectrum sensing, sensing confidence and on the number of common channels in a cluster. Thus,
residual energy. Zeng et al. [18] proposed a network stability it reduces the frequency of re-clustering (by achieving
aware clustering (NSAC) by considering energy consumption the desired number of common channels per cluster in
and spectrum dynamics for CRSN. In NSAC protocol, each the network). Our proposed algorithms also increases the
node computes maximum edge bi-clique based on neighbours lifetime of the network by reducing the energy consump-
and channels qualities. Thompson Stephan et al. [28] have tion for intra-cluster and inter-cluster overheads.
proposed an energy and spectrum aware unequal cluster based
Now, we discuss the system model and definitions used
routing (ESUCR) scheme for clustering and routing in CRSN.
throughout the paper.
They optimized the energy consumption by selecting appro-
priate CH and rotating CH within the cluster. They rank and III. S YSTEM M ODEL
select the appropriate channels based on channels stability for
intra-cluster and inter-cluster communications. We consider a distributed CRSN with N SUs that co-exists
with one or more primary radio network. The set of SUs is rep-
Though reported schemes have considered both energy and
resented as S = {s1 , s2 . . . sN }. In this model, we assume
spectrum related issues jointly but still need more efficient
that each SU is equipped with a single radio and and has a
clustering technique to improve the efficiency of CRSN.
unique ID. We also assume that each SU has certain initial
Most of the clustering approach do not consider the limited
energy, denoted as E0 . A primary radio network is composed
resources constraints of nodes such as hardware capacity. Also,
of cellular network with base stations and primary users (PUs),
most of the reported work are based on 1-hop clustering,
as shown in Figure 1. PUs have taken license to operate on
thus they are not applicable in large scale IoT applications.
a fixed spectrum which are divided into M non-overlapping
Thus, energy efficient clustering is required to address the
channels and represent as M = {c1 , c2 , . . . cM }. SUs are
challenges of CRSN such as limited energy and dynamic
equipped with CR capabilities and perform spectrum sensing
spectrum access. In this paper, we propose two multi-hop
using energy detection, or cyclostationary feature extraction
clustering to address the issues involved in CRSN (different
periodically and cooperate among themselves to discover
from the existing 1-hop clustering schemes). The advantages
spectrum opportunities [29]. Each SU  ∈ S has a set of
of multi-hop clustering are shown using simulations in the
available channels, represented by C , which is subset of M.
result section. Our contributions are as follows:
For an SU  ∈ S, C represents the set of idle PUs’ channels
and they can be used by the SU for its own transmissions.
We model a CRSN as a graph G = (V, E), where the set
A. Our contributions V(G) represents the vertices and corresponds to the set of
SUs, and the set E(G) represents the set of links between
• To achieve neighbourhood information for implementing them. We assume there is an edge (, ) ∈ E(G) between
clustering in CRSN, we have proposed a distributed SUs  and  if they are in the transmission range of each
neighbour discovery protocol. other and C ∩ C 6= ∅. Two node2 can only communicate
• We propose two greedy algorithms one with edge contrac-
1 We abbreviated the k-hop clustering with edge contraction as k-SACB-
tion (k-SACB-EC) and another without edge contraction
EC and the k-hop clustering without edge contraction as k-SACB-WEC
k-SACB-WEC for k-hop spectrum aware clustering for respectively.
bi-channel connectivity in CRSN with the aim of maxi- 2 node and SU are used interchangeably in this paper.

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 00:01:33 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3065691, IEEE Internet of
Things Journal

if they are neighbors of each other and tuned on the same Algorithm 1: MWCBG (node,channels,neighbours)
channel. Input: node, channels,neighbors
Definition 3.1 (1-hop neighbours): For an SU , the set N1 Output: channels,CMs,W
of the nodes located within its transmission range and have 1 BG(V,E)=();
at least one common channel with the node  is called 1-hop 2 V(BG) = neighbours ∪ channels // assign
neighbours of the node . vertex set to BG
We assume that nodes are unaware of its neighbouring nodes. 3 for each channel c in channels do
So, we propose a neighbour discovery protocol to achieve 4 E(BG) ← add an edge (, y), ∀ ∈ neighbours
neighbour discovery and neighbour’s control information such which have the channel c;
as residual energy, location and channel availability. Now, we // assign edges to the graph BG
define the concept of edge contraction in graph theory [30] 5 end
that we will use in k-hop clustering in the Section IV. 6 Find a maximum weight complete bipartite
Definition 3.2 (Edge contraction): In a graph G = (V, E), subgraph(MWCBG) from BG;
contraction of an edge e = (, ) ∈ E is the replacement 7 return channels,nodes in MWCBG and the weight 
∗
of the vertices  and  with a single vertex  such that of MWCBG;
edges incident to the vertex  are the edges other than e that
were incident with either  or . Thus, edge contraction of
an edge e results a new graph G0 = (V 0 , E0 ) where V 0 =
V \ {, } ∪ {} and E0 = E \ {e}. of available channels at node , and neghbors is the set
We assume that every user estimates the idle probability of node 0 s neighbours. Node  constructs a bipartite graph
P and idle duration D of each channel . Then the channel G(N ∪C , E ) using its set of available channels C and set of
quality of the channel  can be defined as [18]: neighbors N . There is an edge (, y) ∈ E between vertices
 ∈ N and y ∈ C if y ∈ C . After that, Node  finds
Q = (1 + ogε P )D (1) a complete bipartite subgraph G(N∗ ∪ C∗ , E∗ ) of the graph
  
Where ε > 1 is a parameter indicating the preference to P G(N ∪C , E ) with the maximum weight  . The weight o
∗

and D . To give more preference to P SU could choose bigger for a complete bipartite subgraph G(No ∪ Co , Eo ) is defined
ε and smaller ε to give more preference to D . as follows
Let the set of nodes upto k-hop neighbors (including all the
nodes from 1-hop) at th SU be Nk and the remaining energy
of an SU  is E . Each node maintains an ordered channel
X X dj
o = ν(|No |· Qc )+(1−ν)(|Co |· (1− P ))
sequence with ranking based on channel quality. Nodes will k∈No dk
c∈Co j∈No
access the channels based on the channel ranking. The nodes (2)
with no channel available can not participate in clustering Where:
(since we are talking about bi-channel connectivity, each nodes
must have at least two available channels in the network). We • ν is preference factor between network stability and
assume that each node have a clock but it doesn’t required to residual energy.
be synchronise with other nodes and there is no centralised • |No | is the number of CRSN nodes in the complete
clock. bipartite subgraph G(No ∪ Co , Eo ).
Next, we describe a procedure named as • |Co | is the number of commonly available channels to
MWCBG(node, chnnes, neghbors) that will all nodes in the complete bipartite subgraph G(No ∪
used as a subroutine in our clustering algorithms. The Co , Eo ).
procedure MWCBG(node, chnnes, neghbors) • Qc = channel quality for cth common channel
takes a node, its available channels and and its neighbouring • Co is the set of commonly available channels to all the
nodes and construct a a bipartite graph. The procedure finds nodes in the complete bipartite subgraph G(No ∪Co , Eo )
the maximum weight complete bipartite subgraph from the • No is the set of CRSN nodes in complete bipartite
constructed bipartite graph. subgraph G(No ∪ Co , Eo )
A. Procedure for finding the cluster members with maximum • dj = euclidean distance from th node to its jth neighbor
number of common channels and with the maximum weight in the complete bipartite subgraph G(No ∪ Co , Eo )
In this section, we describe a procedure that is used Finding the maximum weight complete bipartite subgraph is
to construct a bipartite graph from the given inputs (a hard problem [31]. We used the approximation algorithms in
node, its available channels and and its neighbouring nodes) [32] to find the maximum weight complete bipartite subgraph
and find the maximum weight complete bipartite sub- in our simulation.
graph of the constructed bipartite graph. We call this pro- Now, we discuss the neighbour discovery protocol for
cedure as MWCBG(node , chnnes , neghbors ), CRSN followed by the algorithms called k-SACB-EC and k-
where node is the ID of the node , chnnes is the set SACB-WEC in the next section.

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 00:01:33 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3065691, IEEE Internet of
Things Journal

(1,3,4) where DIFS is Distributed Coordination Function (DCF) Inter-

frame Space, SIFS is Short Inter-frame Spacing, and TR is
D
Transmission Time on one channel and LS is Listening Period
at one channel.
After completing each round the leader waits for DIFS time.
If any another node declares itself as a leader in this period
then the current leader will back-off and switch to listen mode.
(1,2,3) (2,3,4) (3,4) If a node is listening on one of the common channels, in that
case the node will listen the beacons and declares itself as the
A B C
neighbour of the leader. In this case the node will definitely
Figure 2: Illustration of neighbour discovery in CRSN discovers the neighbour.
Let the set of common channels between the node A and B
is C = CA ∩ CB . So the probability of finding B as a neighbor
IV. P ROPOSED N EIGHBOUR DISCOVERY AND k- HOP when A is the leader will be (i.e. node A is in transmitting
CLUSTERING ALGORITHMS beacons on all the channels in CA and node B is listening on
some channel in C):
In this section, first we discuss a neighbour discovery
algorithm. Next, we discuss our two proposed k-hop clustering p1 = NC / NB (4)
algorithms named k-SACB-EC and k-SACB-WEC.
where NC is the number of channels in C and NB is the
A. Neighbour discovery algorithm number of channels in CB .
Now, we compute the probability that node A will be
One of the major challenges in clustering schemes is how successfully elect local leader. Suppose there are  nodes on
control messages will be exchanged among SUs because the channels available in CA . Then the probability that node
initially all the SUs are unaware about their neighbors infor- A senses all channels in CA and there are no beacons on
mation. For an example (see Figure 2), let A wants to send any of the channel in CA is given by 1 . So, for node A,
its available channels and other details to its neighbours for the probability of successfully discover to B as neighbour is
clustering. But the node A is not aware on which channel it p1 · 1 . Thus, the expected number of attempts required for
should broadcast such that all the neighbors get these infor- first occurrence of node A to successfully transmit its beacon
mation (because initially the SUs don’t have any information is given by p . So, the expected time for neighbour discovery
about the neighbours). Suppose we say that it broadcasts to 1

all its available channels even in this case the following issue is n · p · T. In the worst case, the probability p1 can be M1
1
may arise. Let A broadcasts control beacons on channels 1,2 when NC = 1 and NB = M (when there is only one common
and 3 and the nodes B and D are listening on the channel 4, channel available between neighbouring nodes and node CB
so the nodes B and D will not get control messages sent from contains all the PUs channels).
the node A. Theorem 4.1: The expected time of the neighbour discovery
protocol is N ·  · M · T, where  is the maximum number of
As an alternative, we can think of control beacons broad- nodes having a commonly available channel, M is number of
casting over the unlicensed bands but as we know that these PUs’ channels, T is the length of time slot of each round, and
unlicensed bands are already overcrowded, so it cannot guar- N is the number SUs in the CRSN.
antee the reliability. To address the above issue, we present Now, we discuss the k-hop clustering algorithms. First we
a probabilistic neighbour discovery and control message ex- describe the k-hop clustering with edge contraction called k-
change protocol in CRSN. SACB-EC, and then k-hop clustering without edge contraction
Every node will sense its medium by listening on all its called k-SACB-WEC.
available channels. If a node finds that there are no bea-
B. k-hop clustering for CRSN with edge contraction (k-
con messages, it will declare itself a local-leader and start
SACB-EC)
transmitting the beacons which will contains its coordinates
and available channels and preferences on all its available In this section, we discuss the high level idea of our
channels. It will transmit on all its channels at a time and then proposed k-hop clustering algorithm which will increase net-
listen on each channel one by one according to its preferences. work lifetime by reducing the inter-cluster communication
When a node senses beacons on a channel, it will wait to and by reducing number of cluster in the network. In k-
receive beacons. Once the node receives the beacons on a hop clustering, cluster members can be k hop away from the
channel, it will wait for the leader to switch to listen mode on CH and also the all members in the cluster operate on the
that channel(as beacons contain channel preferences) to send same channel. In k-hop clustering with edge contraction (k-
reply to the leader. The time for one round can be given as T SACB-EC), initially each node executes neighbour discovery
algorithm and becomes aware of its 1-hop neighbours and
T = DFS + (N × TR) + SFS + (N × LS) (3) their associated information such as their available channels,

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 00:01:33 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3065691, IEEE Internet of
Things Journal

Algorithm 2: k-hop clustering with edge contraction (k-SACB-EC)

Input: set of SUs S, C , ∀ ∈ S
Output: Clusters
1 Clusters = ϕ /* set of final clusters in CRSN */
2 for each node  ∈ S do
3 set state(i) = initial
4 be chnne = C
5 end
6 while ∃ ∈ S, stte() ∈ / {clustered_CM, clustered_CH} do
7 initialize prtcpnt = ϕ /* set of nodes in bipartite graph for node  */
8 initialize ntermedte cster = ϕ /* set of nodes in intermediate cluster formed by
the node  */
9 initialize  = 0
10 set Chnne = be chnne /* current set of available channels at the node 
*/
11 for each node j ∈ N1 in the updated graph G(V, E) do
12 if state(j) ∈/ {clustered_CM, clustered_CH} and {be chnne ∩
be chnnej } 6= ϕ then
13 prtcpnt ← prtcpnt ∪ j
14 end
15 end
16 [cmn , PCM ,  ] = MWCBG(, Chnne , prtcpnt ) /* call the function MWCBG() at
node  */
17 if cmn ≥ 2 and  > j , ∀j ∈ PCM then
18 if state(i) == initial then
19 state(i) ← intermediate_CH, assign CM = PCM
20 send join message to nodes in CM /* node  becomes intermediate CH */
21 end
22 if state(i) == intermediate_CH then
23 find the subset of nodes CM ⊆ PCM as discussed in Step-I (2b)
24 end
25 for each node j ∈ CM do
26 if node j has received join message only from the node  then
27 ntermedte cster ← ntermedte cster ∪ j /* node j joins the
intermediate cluster formed by the node  */
28 state(j) ← clustered_CM
29 end
30 else
31 node j joins the intermediate cluster of the node k as discussed in text Step-I (4)
32 state(j) ← clustered_CM /* when a node j receives join from more than one
node */
33 end
34 end
35 end
36 apply edge − contrcton(ntermedte cster ∪ ) and update the graph G(V, E) /* replace all
nodes in ntermedte_cster by the node  and update edges as discussed in
text. */
37 be chnne = {set of channels commonly available to all node in cluster with CH  } ⊆ cmn
38 if cmn < 2 and  > j , ∀j ∈ PCM then
39 node  ignores the current call of MWCBG() and forms a cluster C based on the results of the last
MWCBG() call
40 C ← ntermedte cster
41 state(i) ← clustered_CH
42 Clusters = Clusters ∪ C
43 else
44 node  becomes CM of the intermediate CH j having the maximum weight among its neighbours
45 state(i) ← clustered_CM
46 end
47 end
48 return Clusters
2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 00:01:33 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3065691, IEEE Internet of
Things Journal

y ∈ C . Then the node  finds a complete bipartite

clustered CM subgraph of G (N1 ∪ C , E1 ) with the maximum weight
 using the Algorithm 1. Let PCM denotes the set of
initial nodes involved in maximum weight complete bipartite
subgraph.
2) For a nodes , if  > j , ∀j ∈ PCM then node  elects
itself as a leader and does the following steps:
intermediate CH clustered CH a) If the node  is in theinitial state, it changes to
intermediate_CH state and forms an intermediate
cluster C() by sending a join message to the nodes
Figure 3: State diagram of in nodes for k-hop clustering with involved in maximum weight complete bipartite sub-
edge contraction (k-SACB-EC) graph for the graph G (N1 ∪ C , E1 ) (i.e. to all the
nodes in the set PCM ). Nodes in C() change their
status as clustered_CM.
channel’s statistics such as quality of channels (considering b) If the node  is in the state intermediate_CH then
the probability of being idle and average time of being idle), it may stay in the same state and form bigger sized
ID, location etc. Using 1-hop neighbours information, each cluster, or it may change to clustered_CH state and
node  constructs a bipartite graph G (N1 ∪ C , E1 ), where terminates the execution of the algorithm. The node 
N1 is the set of 1-hop neighbours of node , and C is the set finds the subset of nodes CM ⊆ PCM involved in
of available channels at node . There is an edge (, y) ∈ E1 the maximum weight complete bipartite subgraph for
between vertices  ∈ N1 and y ∈ C if y ∈ C . the graph G (N1 ∪ C , E1 ) such that the number of
Then, each node  finds the complete bipartite subgraph of common channels of newly formed cluster C() at least
G (N1 ∪ C , E1 ) with maximum weight using the Algorithm the minimum of the numbers of common channels of
1. The weight of a complete bipartite graph is defined in the clusters in CM .
equation (2). i) If the subset CM is non-empty, then the node 
Each node starts clustering process in initial state. stays in the state intermediate_CH and forms
A node changes its state from initial to either an intermediate cluster C() by sending a join
intermediate_CH or clustered_CM. The state diagram message to the nodes in the subset CM . All the
is shown in the Figure 3. intermediate CHs of clusters in CM change their
State: initial state to clustered_CM and become member of
Each CRSN node starts clustering in this state. the cluster with CH , and stop execution of the
State: intermediate_CH algorithm.
When a node in state initial becomes intermediate CH, ii) If the subset of nodes CM in the step 2b is empty
it changes to intermediate_CH state. then the node  changes to state clustered_CH.
State: clustered_CM 3) For a node , if there are more nodes having same as 
A node changes its state to clustered_CM in any of the then the node with smaller ID elects itself CH and forms
following cases: cluster as discussed above.
• case 1: when a node in state initial becomes cluster 4) When a node j receives join messages from more than
member. one node then it joins the one of them based on following
• case 2: when a node in state intermediate_CH criteria that aims to distribute the load fairly among
becomes cluster member. clusters:
State: clustered_CH a) Node j joins the cluster having smaller cluster size
When a node in the state intermediate_CH becomes among all from which it has received the jon mes-
final CH it changes to the state clustered_CH. sages.
The k-SACB-EC executes in phases. Each phase involves b) If there are more than one clusters with the same size
three steps which are as follows: then it prefers to join the one having more number of
common channels, higher remaining energy level, and
less distance with it. Further, a tie is broken with node
Step I - Initial clustering and intermediate CH formation
ID.
1) Each node  ∈ V(G) in state initial or
intermediate_CH constructs a bipartite graph Step II - Edge contraction
G (N1 ∪ C , E1 ), where N1 is the set of 1-hop
neighbours of node  in the updated graph, and C is In this step, for each intermediate cluster C , its intermediate
the set of available channels at node . There is an edge CH C(intermediate_CH) behaves as follows: It contracts
(, y) ∈ E1 between vertices  ∈ N1 and y ∈ C if the edges incident to its cluster members and replace cluster

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 00:01:33 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3065691, IEEE Internet of
Things Journal

X1
X2
b a b a

6 x
g x
c g
y c
z y
4 z
f
5 f
d e
d e
C1 F1 F2
(a) (b)
Figure 4: Illustration of k-hop clustering with edge contraction. (i) Figure. 4 (a) shows the nodes , y, z, , b, c, d, e, ƒ , and h with their available
channel set {2, 3, 4, 5}, {1, 2, 3, 4, 5}, {2, 3, 5, 6}, {2, 3, 5, 6}, {1, 2, 3, 4, 5}, {2, 3, 5, 7}, {2, 3, 5, 8}, {1, 2, 3, 5, 6}, {1, 2, 3, 6}, and
{1, 2, 3, 8} respectively. Each node computes its weight. To understand the mechanism, assume that the weight of the nodes , c, and ƒ are 6, 4, and 5
respectively using 1-hop information. Assume that these nodes have the maximum weight in their 1-hop neighbourhood (ii) If the node with maximum weight
becomes CH then the nodes , c and ƒ elect themselves intermediate_CH and form the intermediate clusters X1 , C1 and F1 with their 1-hop neighbours
(represented in Figure 4(a)). If we use the concept of edge contraction then the nodes in each cluster X1 , C1 , and F1 will be replaced with a single node and
the topology is updated accordingly as shown in Figure 4(a). Set the available channels to the node X1 , C1 and F1 as {2, 3, 5}, {2, 3, 5} and {1, 2, 3}.
Then, the nodes as a intermediate_CH tries to form cluster with its neighbouring nodes (it may be normal node or intermediate_CH) and this
process repeats till clustering is possible. (iii) Figure 4(b) shows the clusters where node  is CH and cluster members are at distance at most 2-hop from
CH. The advantages of k-hop clustering with edge contraction are as follows (i) k-SACB-EC generates more robust clusters as compared to k-SACB-WEC,
as the clusters may have more number of common channels which are more robust to PUs’ activities and reduced the chances of re-clustering events on PU
arrivals, (ii) nodes consume less energy, since only CHs require knowledge of its k-hop neighbours. Thus, it is energy efficient during clustering process.

C with a single node and set its available channels as the set available channels then this phase will not take place and
of channel commonly available to all the nodes in cluster C , the output of the previous phase will be the final cluster of
denoted as CC . By the end of the Step II, the network graph that node i.e. the intermediate CH becomes clustered CH and
G is transformed to a new graph G0 = (V 0 , E0 ), where the stops the execution of the algorithm. The final set of clusters
vertex set V 0 (G0 ) is the set of intermediate CHs genereted in of the k-hop clustering algorithm with edge contraction will
the current network graph G. There is an edge between two be returned when all the nodes changes to the state either
intermediate cluster nodes C and Cj if both of the following clustered_CM or clustered_CH.
conditions are satisfied The pseudocode of k-hop clustering algorithm with edge
• There is a pair of nodes (p, q) in V(G) such that p ∈ C contraction is presented in the Algorithm 2. The stages and
and q ∈ Cj and (p, q) ∈ E(G). clustering process of k-SACB-EC is illustrated in the Figure
• There is at least one common channel between CC and 4.
CCj i.e. CC ∩Cj 6= ∅. Lemma 4.2: After the execution of the algorithm 2,
Finally, the graph G ← G0 becomes new network graph each CRSN node will be in either clustered_CM or
(from the view of intermediate CHs). clustered_CH.
Lemma 4.3: Each nodes will belong to exactly one cluster
after the execution of algorithm 2.
Step III - cluster merging
Theorem 4.4: The algorithm 2 partitions the CRSN into
In this step, each node 0 ∈ V 0 repeats Steps I and II until the disjoint clusters and each cluster has at least two common
constraint that each cluster C(0 ) has at least two commonly channels.
available channels to all of its cluster members is violated. In In k-SACB-EC, only intermediate CHs explore higher hops
the current phase if any cluster has less than two commonly to extend the cluster size so it may results smaller sized clus-

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 00:01:33 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3065691, IEEE Internet of
Things Journal

N number of SUs
M number of PU channels
a cluster member. If it does not receive any join
S set of SUs message, then it forms single node cluster.
M set of PUs’ channel b) Else the node  stores the values of hop parameter
C set of available channels at the node  and weight value in H [1] = 1 and  [1] = 1
C() cluster with CH 
C th cluster
respectively (H and W are data structures to store the
C C set of common channels available to all nodes in cluster C hop values and weight values).
E0 SU’s initial energy 3) Set hop parameter k = 2.
Pd detection probability
Pƒ false alarm probability 4) Further the node  constructs bipartite graph G (Nk ∪
T length of time slot of each round C , Ek ) using k hop neighbours (using current value of
o weight of complete bipartite subgraph for node 
k). The edge set Ek is defined as follows:
Qc channel quality of the channel c
dj euclidean distance between the node  and the node j a) All the edges in Ek−1
 will also be part of Ek .
α departure rate of PU
β arrival rate of PU
b) Also, there will be an additional edge (, y) between
vertices  ∈ Nk \ Nk−1
 and y ∈ C if there is a vertex
Table I: Symbol table
∗ ∈ Nk−1
 such that (∗ , y) ∈ Ek−1
 and ∗ and 
∗
are neighbours (( , ) ∈ E(G)) and y ∈ C .
ters. Thus, it may results many clusters in the networks which 5) Node  computes a complete bipartite subgraph
∗ ∗
incur higher inter-cluster coordination and communication CG (Nk ∪ C∗ 
, Ek ) of the graph G (Nk ∪ C , Ek )
overheads. To explore optimal cluster size, we discuss our next having the maximum weight k such that |C∗ 
| ≥ 2 (e.g.
algorithm called k-hop clustering without edge contraction (k- the set C∗

has at least two elements, this guarantees that
∗
SACB-WEC) in which every node explores the possibility of nodes in Nk have at least two channels in common).
constructing optimal clusters. a) If there is no CG (Nk ∪ C∗
∗
, Ek ) with |C∗ |≥2
∗
 
then the node  stops this process and decides whether
C. k-hop clustering without edge contraction (k-SACB-WEC) (k−1)
or not to become CH based on its N neighbours
In this section, we discuss the k-hop clustering without edge computed weights. Node  chooses the hop parameter
contraction (k-SACB-WEC). The advantages of (k-SACB- H [k 0 ] that has the maximum weight  [k 0 ] for all
WEC) are that it generates larger size clusters, see Figure k 0 ≤ (k − 1).
(k 0 )∗
5 . Thus, it reduces inter-cluster communication coordination i) If node  has the highest weight among its N
and communication overheads. It increases the network life nodes then it becomes CH and notifies about it to
(k 0 )∗
time by forming optimal clusters and distributed the load the nodes in N send a join message to them.
fairly among the CHs. This has been demonstrated in the Else the node  waits for receiving a join as cluster
results section, we have shown that in our proposed scheme member message from a CH.
the number of rounds before first node dead is higher than b) If the computed weight  [k] =  [k − 1] then
the compared state of the art schemes. In k-SACB-WEC, return the value of hop parameter (k − 1) as an
each node requires the knowledge of its k-hop neighbours for saturated hop count for the node  and decides whether
clustering . Along with creating bipartite graph with only one (k−1)
or not to become CH based on its N neighbours
hop neighbours, each node also creates bipartite graphs using computed weights. The node  chooses the hop param-
k-hop neighbours and its available channel C . The detailed eter H [k 0 ] that has the maximum weight  [k 0 ] for
description of the algorithm is given as follows. all k 0 ≤ (k − 1) (the text described in step 5(a)i).
1) Each node  starts with 1-hop neighbours knowledge c) Else the node  stores the values of hop parameter in
(parameter k = 1) and construct the bipartite graph H [k] = k and computed corresponding weight in
G (N1 ∪C , E1 ), where C is the set of available channels  [k] = k .
at node  and N1 is the set of 1-hop neighbours of node d) Node  increases its hop parameter as k = k + 1 and
. There is an edge (, y) ∈ E1 between vertices  ∈ N1 go to the step 4.
and y ∈ C if y ∈ C . The clustering process and advantages of k-SACB-WEC
2) Node  computes a complete bipartite subgraph graph is illustrated in the Figure 5. It can be notices from the
∗ ∗
CG (N1 ∪ C∗ 
, E1 ) of the graph G (N1 ∪ C , E1 ) Figures 4 and 5 that for a smaller value of a predefined
having the maximum weight 1 such that |C∗ | ≥ 2 (e.g. required number of common channel per cluster, k-SACB-

the set C∗ has at least two elements, this guarantees that WEC generates larger size cluster as compared to k-SACB-

1 ∗ EC whereas k-SACB-EC may generates clusters having more
nodes in N have at least two channels in common).
∗ ∗
number of common channels per clusters i.e. more robust
a) If there is no CG (N1 ∪ C∗
, E1 ) with |C∗

|≥2 to PUs’ activities. To increase the robustness against PUs’
then the node  stops this process and waits for re- activities, we need to set higher value of predefined required
ceiving a join message from some CH to become

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 00:01:33 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3065691, IEEE Internet of
Things Journal

X0
b a b a

x x
g g
c c
y y
z z
4
f f
d e d e

(a) (b)
Figure 5: Illustration of k-hop clustering without edge contraction (k-SACB-WEC). We consider the same configuration (number of nodes and their available
channel sets) as in 5. This shows the drawbacks of k-hop clustering with edge contraction. Though k-hop clustering with edge contraction is energy efficient
during clustering process, it may generates more number of clusters. It can be noticed from Figure 4(b) that there are two clusters (each of them has three
common channels). (c) In k-SACB-WEC, to minimize the number of clusters with desired properties (for pre-defined number of common channels per cluster
here we set this value to 2 i.e. each cluster must has at least two common channels), each node computes clustering metrics k-hop counts (k = 1, 2, . . . )
using its k-hop neighbouring information. The advantage of k-SACB-WEC that it reduces the energy consumption required for inter-cluster communications
among the clusters through gateways since it generates less number of clusters. For a given number of required common channel per cluster, k-SACB-WEC
generates optimal clusters in terms of energy efficient and balances the nodes fairly among the CHs. In this example, we set the number of required common
channel per cluster to 2. The algorithm k-SACB-WEC generates only one cluster X 0 with the node  as CH and having two common channels {2, 3} as
shown in the Figure 5(b).

number of common channel per cluster. The k-SACB-WEC level and residual energy. In SAC-WEM, after clustering only
is more complex than k-SACB-EC whereas one of the main CHs sense spectrum periodically instead of all the nodes
advantages of k-SACB-WEC is that it can find optimal hop which alone may not detect PU arrival accurately and hence
counts in clustering for a predefined required number of may interfere with PUs’ transmissions. The third algorithm is
common channel per cluster. On the other hand, k-SACB-EC modified Stable Election Protocol named Prolong-SEP (PSEP)
is relatively simpler as compared to k-SACB-WEC but may proposed for Fog supported WSN [33]. The aim of the PSEP
generate smaller size clusters in the network. Next, we discuss is to prolong the stable time period of the network (mainly
simulation results. before the failure of the first node) by maintaining balanced
energy consumption. For comparison, we modified PSEP to
V. R ESULTS
have opportunistic spectrum access environment by integrating
In this section we discuss simulation results of our pro- dynamic spectrum opportunities in the network. The fourth
posed algorithms (one with edge contraction k-SACB-EC algorithm is Cognitive LEACH (CogLEACH) [34], which is
and another without edge contraction k-SACB-WEC) for k- an extension of the Low Energy Adaptive Clustering Hierarchy
hop spectrum aware clustering for bi-channel connectivity in (LEACH) protocol which consider spectrum dynamics in
CRSN (k-SACB) with the aim of maximizing the network clustering process.
lifetime. We compare the results of our proposed schemes
with four state of the art clustering algorithms. First algorithm The performance of our proposed approaches is evaluated
is network stability-aware clustering (NSAC) protocol for by varying various network parameters such as network size
CRSNs proposed in [18]. This work aims to form stable in comparison with considered state of the art clustering algo-
clusters by integrating the energy consumption and spectrum rithms. We consider the following parameters for performance
dynamics while clustering. Second algorithm is spectrum comparison: (i) Number of clusters, (ii) Energy efficiency
aware clustering based on weighted clustering metric [19]. We in terms of network remaining energy and number of alive
abbreviated it as SAC-WEM in this paper. In this scheme each nodes, (iii) Stability of networks: we consider two metrics
node computes a weight using temporal-spatial correlation to evaluate the stability of networks. First metric is in terms
between itself and its one-hop neighbours, sensing confidence of the first node dead (i.e. the time interval since beginning

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 00:01:33 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3065691, IEEE Internet of
Things Journal

before the first node exhausts its energy and dies). Second α
is in terms of frequency of re-clustering (re-clustering is an
event that is triggered when there is no common channels
Busy Idle
available for any cluster in the network), (iv) Effect of number
of SUs on the number of clusters (to analyse the performance β
behaviour of schemes during scaling up the network). We
perform simulations by using MATLAB. In our simulation, Figure 6: PU activity model
radio model proposed in [35], [36] is adopted for energy
consumption parameters. The value of Eƒ s = 10pJ / bt / m2 Simulation area 300 × 300m2
Number of SUs N 30-100
and Eeec = 50nJ / bt. The simulation parameters are given Number of channels M 10
in Table II. Number of PUs 10
PU’s transmission range 60m
A. PU activity model SU’s transmission range 40m
Initial energy of SUs (E0 ) 0.2 J
In order to simulate the behaviour of PUs, most of the re- Per-node energy consumption for a channel sensing 1.31 × 10−4 J
ported works have considered the use of exponential ON-OFF Per-node energy consumption for a channel switching 10−5 J
model [37]–[39]. We have also considered the exponential ON- Node deployment type Uniform random
Eƒ s 10pJ / bt / m2
OFF model in this paper. Each PU can be in either of the two
Eeec 50nJ / bt
states ON state and OFF state. The ON state represents that PU’s ON period mean value (α) [0.5, 2.0]
PU is using the channel and the OFF state represents that the PU’s OFF period mean values (β) [0.5, 2.0]
channel is idle. The arrival of each PU is independent and Table II: Simulation parameters
follows Poisson arrival process. The length of ON and OFF
period of PUs on a channel  is modeled using the exponential
distribution with the mean value of 1 /α and 1 / β respectively. the first scenario, we set α = 2, β = 2. In the second
So, alternating ON /OFF model for PUs on a channel  is scenario, we set α = 2, β = 0.5. We set the number of
regulated with the rate α and β as shown in the Figure 6. The channels to M = 10 in both the scenarios. To show the
ON and OFF probabilities are formulated as follows. comparison in terms of energy efficiency, We executed the
considered approaches and evaluated the performance of these
Pon · α = Poƒ ƒ · β (5) approaches in terms of network remaining energy and the
number of alive nodes in the network. We measure time in
Pon + Poƒ ƒ = 1 (6) number of rounds. We set the initial energy of each node to
0.2 J and set the number of SUs to 100. Thus, total energy
of the network is 20 J. Figures 7 and 8 and show that k-
β
Pon = (7) SACB-WEC and k-SACB-EC outperform other schemes in
α+β terms of both network remaining energy and number of alive
α nodes in the network, also see Figures 9 and 10. Specifically,
Poƒ ƒ = (8) Figure 7 demonstrates that for α = 2, β = 2 (this case
α+β considers the intermittent activity of PUs), the energy of the
We use different values of the parameters α and β for PU nodes depletes at faster rate as compared to the case when
activities. There are mainly four cases: α = 2, β = 0.5 (short ON followed by long OFF period),
• Case 1: α ≤ 1, β ≤ 1 long ON periods followed by see Figure 8. In the other considered schemes, the nodes in
long OFF periods the network die at a faster rate because of these schemes
• Case 2: α ≤ 1, β > 1 long ON periods followed by result large number of clusters in network which may require
short OFF periods inter-cluster coordination and communication overheads. Also,
• Case 3: α > 1, β ≤ 1 short ON periods followed by these clustering schemes are not robust to PUs’ activities
long OFF periods and suffer from high frequency of re-clustering events thus
• Case 4: α > 1, β > 1 short ON periods followed by nodes consume more energy. However, our proposed schemes
short OFF periods reduce energy consumption and extend the lifetime of the
network by achieving bi-channel connectivity and forming
B. Network lifetime less number of clusters in the network. Further, the k-SACB-
In this subsection, we conduct simulation to show the energy WEC outperforms the k-SACB-EC in terms of both the
efficiency of our proposed schemes. There are four possible remaining network energy and the number of alive nodes in
cases: (i) α = 2 and β = 0.5 (long OFF followed by the network, see Figures 7, 8, 9 and 10 . It can be noticed that
short ON), (ii) α = 0.5 and β = 2 (long ON followed although k-SACB-WEC and k-SACB-EC require nodes to
by short OFF), (iii) α = 2, β = 2 (short OFF followed by have their k-hop neighbours and consumes energy for getting
short ON), and (iv) α = 0.5, β = 0.5 (long OFF followed this information, the overall network lifetime under k-SACB-
by long ON). We conducted simulation in two scenarios. In WEC and k-SACB-EC is longer than NSAC, PSEP, SAC-

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 00:01:33 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3065691, IEEE Internet of
Things Journal

WCM and CogLEACH individually, see Figures 7 and 8. The the case α = 2, β = 0.5, the re-clustering frequency of the
main reason is that re-clustering and inter-cluster coordination considered schemes is very low because PUs remain active
overheads are higher in the case of NSAC, PSEP, SAC- for shorter period and then remain idle for longer period as
WCM and CogLEACH schemes and thus nodes consume more shown in Figure 14. The performance of k-SACB-WEC is
energy. comparable to NSAC, PSEP, CogLEACH and SAC-WCM.
The simulation results clearly shows improvement our pro- Since we aim to minimize the number of clusters in the
posed schemes in terms of network lifetime because of reduced network, this may cause the less number of common channels
frequency of re-clustering events. per cluster. This may leads to frequent re-clustering in the
k-SACB-WEC algorithm. However, k-SACB-EC outperforms
C. Stability of networks all the considered schemes by maintaining higher number of
In this section, we perform the simulation to analyze the common channels per cluster, see Figures 13 and 14. This
stability of network under the considered schemes. To assess is because clusters merging conditions in the algorithm k-
the stability of a network, We measure two metrics: (i) the SACB-EC ( i.e. cluster merging will happen only if the number
number of rounds before the death of the first node in the of common channels of newly formed cluster is at least the
network, (ii) frequency of re-clustering events. minimum of the numbers of common channels of each cluster
1) The number of rounds before the death of the first node: involved in this merging) ensures more common channels in
Stability of networks can be measured as the number of rounds a cluster.
since beginning before the first node exhausts its energy. To Further, the re-clustering interval of the considered schemes
show the comparison, we executed considered approaches for are shown in the Figures 15 and 16. We consider intermittent
two cases: In first case α = 2, β = 2 and in the second (α = 2, β = 2) and long term (α = 0.5, β = 0.5)
scenario α = 2, β = 0.5. Figures 11 and 12 show that the PU activities for simulation. We consider 50 runs average
k-SACB-WEC outperforms all the considered approaches in the interval (rounds). The average re-clustering interval is the
terms of number of rounds before the first node dies in the average time interval when re-clustering is triggered under
network. For, k-SACB-WEC, the number of rounds before the considered schemes. The Figures 15 and 16 show the
the first node dies is at least approximately 100% higher performance of schemes for intermittent and long term PUs’
than compared schemes in both the scenarios. This shows that activities respectively. It can be observed that the average re-
k-SACB-WEC generates more stable clusters as compared to clustering interval for the scheme k-SACB-EC is higher than
other approaches. For the case α = 2, β = 2, the performance other considered approaches. This is can be justified from the
of k-SACB-EC is comparable to PSEP, NSAC, SAC-WCM re-clustering frequencies of the approaches as discussed above.
and CogLEACH, see the Figure 11. Further, it can also be
noticed from the Figure 11 that for intermittent PUs’ activities D. Number of clusters
(α = 2, β = 2) the number of rounds before the death of the To show the effectiveness and improvement of the proposed
first node is lower as compared to the case when short ON is schemes in term of inter-cluster communication and intra-
followed by long OFF interval (α = 0.5, β = 2), see Figure cluster communication overheads, we use the number of clus-
12. ters generated under different clustering schemes in the CRSN.
2) Re-clustering frequency of clustering schemes: To We are assuming that each inter cluster communication leads
demonstrate the improvement in channel connectivity and to channel switching and the energy consumption for inter-
stability of our proposed schemes, we use the frequency of cluster communication is proportional to number of clusters
re-clustering as a metric which is related to the stability of in the network. As we have discussed in text that if the
the clustering schemes. In this subsection, we compare the cluster size is too large then there will be higher intra-cluster
performance of clustering schemes in terms of re-clustering communication overheads and on the other hand if cluster
frequency and re-clustering interval under various PU activi- size is too small (large number of clusters) then there will
ties. We consider the values of the parameters α and β for many clusters in the network which leads to higher inter-
different PU activities. We consider two cases: (i) α = 2, cluster overheads. In this subsection, we discuss the number
β = 2 (short OFF followed by short ON), and α = 2 and of clusters in CRSN under different number of channels. We
β = 0.5 (long OFF followed by short ON). We consider 10 set the values of SUs to 100 in this experiment. We set the
channels and 100 SUs in the networks for this simulation. number of channels M = 10 in the network. We consider
For re-clustering frequency, the results of considered schemes intermittent (α = 2, β = 2) and long term (α = 0.5,
are presented in the Figures 13 and 14. It can be noticed β = 0.5) PU activities for simulation. We consider 50
that our proposed algorithms form stable clusters as compared runs and show the minimum, average and maximum number
to other schemes. It can also be noticed that the clustering of clusters formed under all the considered schemes. The
frequencies are higher under all the considered schemes in maximum number of cluster is the maximum number among
the case when α = 2, β = 2 (short OFF followed by short all the measurements and minimum is the minimum number
ON) as shown in Figure 13 as compared to the case when of clusters among all the measurement. The main intention
α = 2 and β = 0.5. This is because the case α = 2, β = 2 behind showing the maximum and the minimum values is
considers the intermittent activity of PUs. On the hand, for to demonstrate the variation in number of clusters formed

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 00:01:33 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3065691, IEEE Internet of
Things Journal

20 20
k-SACB-EC k-SACB-EC
k-SACB-WEC k-SACB-WEC
NSAC NSAC
15 PSEP 15 PSEP
CogLEACH CogLEACH
Network remaining energy

Network remaining energy

SAC-WCM SAC-WCM

10 10

5 5

0 0

-5 -5
0 100 200 300 400 500 600 700 800 900 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Number of rounds Number of rounds

Figure 7: Network remaining energy under different clus- Figure 8: Network remaining energy under different clus-
tering schemes (number of channels = 10 in the network, tering schemes (number of channels = 10 in the network,
α = 2, β = 2)) α = 2, β = 0.5))

100 100
k-SACB-EC k-SACB-EC
90 k-SACB-WEC 90 k-SACB-WEC
NSAC NSAC
80 PSEP 80 PSEP
CogLEACH CogLEACH
Number of alive nodes

Number of alive nodes

70 SAC-WCM 70 SAC-WCM

60 60

50 50

40 40

30 30

20 20

10 10

0 0
0 100 200 300 400 500 600 700 800 900 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Number of rounds Number of rounds

Figure 9: Number of alive nodes under different clustering Figure 10: Number of alive nodes under different clustering
schemes (number of channels = 10, α = 2, β = 2) schemes (number of channels = 10, α = 0.5, β = 2)

250 400

350
200
300
Number of rounds

Number of rounds

250
150

200

100
150

100
50
50

0 0
-EC EC AC EP CH CM C EC A C EP CH M
B B-W NS PS LE
A W B-E B-W NS PS LE
A WC
AC g C- AC g C-
k-S AC Co SA k-S AC Co SA
k-S k-S
Algorithms Algorithms

Figure 11: Number of rounds before first node dead under Figure 12: Number of rounds before first node dead under
different clustering schemes (number of channels = 10, α = different clustering schemes (number of channels = 10, α =
2, β = 2) 0.5, β = 2)

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 00:01:33 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3065691, IEEE Internet of
Things Journal

0.5 0.18

0.45 0.16

0.4
0.14
Reclustering frequency

Reclustering frequency
0.35
0.12
0.3
0.1
0.25
0.08
0.2
0.06
0.15
0.04
0.1

0.05 0.02

0 0
-EC EC AC EP H CM C EC A C EP CH M
B -W NS PS EAC W B-E B-W NS PS LE
A WC
AC CB gL C- AC g C-
k-S A Co SA k-S AC Co SA
k-S k-S
Algorithms Algorithms

Figure 13: Re-clustering frequency of different clustering Figure 14: Re-clustering frequency of different clustering
schemes for α = 2, β = 2 (short OFF followed by short schemes for α = 2, β = 0.5 (long OFF followed by short
ON) ON)

3.5 14

3 12

2.5 10
Number of rounds

Number of rounds

2 8

1.5 6

1 4

0.5 2

0 0
-E C EC AC EP CH CM C EC AC EP H M
CB B-W NS PS LE
A W B-E B-W NS PS LE
AC WC
A g C- AC g C-
k-S AC Co SA k-S AC Co SA
k-S k-S
Algorithms Algorithms

Figure 15: Re-clustering interval of different clustering Figure 16: Re-clustering interval of different clustering
schemes (α = 2, β = 2) schemes (α = 2, β = 0.5)

25 20
k-SACB-EC k-SACB-EC
k-SACB-WEC 18 k-SACB-WEC
NSAC NSAC
20 PSEP 16 PSEP
CogLEACH CogLEACH
SAC-WCM 14 SAC-WCM
Number of clusters

Number of clusters

15 12

10

10 8

5 4

0 0
Minimum Average Maximum Minimum Average Maximum

Figure 17: Number of clusters under different clustering Figure 18: Number of clusters under different clustering
schemes (number of channels = 10, α = 2, β = 2) schemes (number of channels = 10, α = 0.5, β = 2)

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 00:01:33 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3065691, IEEE Internet of
Things Journal

16
k-SACB-EC
applications. To demonstrate the performance improvement of
14
k-SACB-WEC
NSAC
our schemes, we perform simulation and present the results
PSEP
CogLEACH
in terms of frequency of re-clustering under different PUs’
12 SAC-WCM
activities, re-clustering interval, network energy, number of
Avg number of clusters

10
rounds before first node dies, and number of clusters. Perfor-
mance evaluation shows that k-SACB-WEC generates at least
8 40% less number of clusters as compared to k-SACB-EC,
6
NSAC, PSEP, SAC-WCM and CogLEACH. Further, with the
k-SACB-WEC, the number of rounds before the first node
4 dies is at least approximately 100% higher than the com-
pared competitive approaches. Thus, our proposed schemes
2
30 40 50 60 70 80 90 100 are energy efficient and have increased channel connectivity.
Number of SUs
The future work will focus on to analyze the complexity
Figure 19: Average number of clusters vs. number of SUs of algorithms and improve them to achieve more robustness.
We will also explore the applicability of our schemes under
different PU spectrum occupancy models.
in the networks under various schemes. The average number
of clusters is the average over all the 50 measurements. It ACKNOWLEDGEMENT
is clear from Figure 17 and 18 that on an average the k- The work is supported by the SERB-DST, Government of
SACB-WEC outperforms all the considered schemes. It can India under Grant No.: SRG/2020/000575
be notices that k-SACB-WEC generates at least 40% less
number of clusters as compared to k-SACB-EC, NSAC, PSEP, R EFERENCES
SAC-WCM and CogLEACH. Since in the k-SACB-WEC each [1] F. Li, K.-Y. Lam, X. Li, Z. Sheng, J. Hua, and L. Wang, “Advances and
nodes explores possibility of becoming CH and/or after getting emerging challenges in cognitive internet-of-things,” IEEE Transactions
k-hop neighbour knowledge, it generates larger size clusters on Industrial Informatics, vol. 16, no. 8, pp. 5489–5496, 2019.
[2] E. Fcc, “Docket no 03-222 notice of proposed rule making and order,”
than other competitive schemes. 2003.
Further to see the behaviour of clustering schemes as num- [3] S. Haykin, “Cognitive radio: brain-empowered wireless communica-
ber of SUs increases in a network, we perform an experiment tions,” IEEE journal on selected areas in communications, vol. 23, no. 2,
pp. 201–220, 2005.
and set the number of channels M = 10. We vary the [4] O. B. Akan, O. B. Karli, and O. Ergul, “Cognitive radio sensor
number of SUs from 30 − 100. The Figure 19 shows the networks,” IEEE network, vol. 23, no. 4, pp. 34–40, 2009.
result of this experiment. It can observed that under all the [5] A. Ahmad, S. Ahmad, M. H. Rehmani, and N. U. Hassan, “A survey
on radio resource allocation in cognitive radio sensor networks,” IEEE
considered clustering schemes the number of clusters increases Communications Surveys & Tutorials, vol. 17, no. 2, pp. 888–917, 2015.
as number of nodes increases in the network whereas in k- [6] J. Ren, Y. Zhang, R. Deng, N. Zhang, D. Zhang, and X. S. Shen, “Joint
SACB-WEC the number of cluster is nearly same at different channel access and sampling rate control in energy harvesting cognitive
radio sensor networks,” IEEE Transactions on Emerging Topics in
node density in the network, see Figure 19.This demonstrates Computing, vol. 7, no. 1, pp. 149–161, 2016.
the main advantage of using k-hop clustering. k-SACB-WEC [7] X. Li, J. Zhao, Y. Yao, T. Zhou, Y. Gong, and L. Xiong, “Fault-
outperforms all the competitive clustering schemes in terms tolerant topology control towards k-channel-connectivity in cognitive
radio networks,” IEEE Access, vol. 6, pp. 65 308–65 320, 2018.
of the number of clusters that they generate. With multi-hop [8] R. N. Yadav and R. Misra, “On k-channel connectivity in cognitive radio
clustering, the number of clusters are reduced in the networks networks through channel assignment,” AEU-International Journal of
and we can increase the lifetime by generating more stable Electronics and Communications, vol. 77, pp. 118–129, 2017.
[9] Y. Shi, H. Sun, M. Sheng, J. Li, and X. Li, “Constructing a robust
cluster considering the possibility of every node of becoming topology for reliable communications in multi-channel cognitive radio
CH and/or CM. ad hoc networks,” IEEE Communications Magazine, vol. 56, no. 4, pp.
172–179, 2018.
VI. C ONCLUSIONS [10] P. Ding, J. Holliday, and A. Celik, “Distributed energy-efficient hi-
erarchical clustering for wireless sensor networks,” in International
In this paper, we present two k-hop spectrum aware cluster- conference on distributed computing in sensor systems. Springer, 2005,
pp. 322–339.
ing schemes for cognitive radio sensor networks (CRSN) with [11] P. Nayak and A. Devulapalli, “A fuzzy logic-based clustering algorithm
the aims of achieving bi-channel connectivity and improving for wsn to extend the network lifetime,” IEEE sensors journal, vol. 16,
network lifetime. One important feature of our proposed no. 1, pp. 137–144, 2015.
[12] G. Shanthi and M. Sundarambal, “Fso–pso based multihop clustering
schemes is that they reduce intra and inter-cluster overheads by in wsn for efficient medical building management system,” Cluster
finding the optimal value of hop parameter (value of k) during Computing, vol. 22, no. 5, pp. 12 157–12 168, 2019.
clustering (k-SACB-WEC) which is responsible for cluster [13] P. G. V. Naranjo, M. Shojafar, A. Abraham, and E. Baccarelli, “A new
stable election-based routing algorithm to preserve aliveness and energy
size and number of clusters in CRSN. Further, our scheme is in fog-supported wireless sensor networks,” in 2016 IEEE International
based on k-hop neighbour knowledge it also helps in reducing Conference on Systems, Man, and Cybernetics (SMC). IEEE, 2016,
the frequency of re-clustering by selecting appropriate clusters. pp. 002 413–002 418.
[14] A. Ghosal, S. Halder, and S. K. Das, “Distributed on-demand clustering
Thus, our schemes are energy efficient that maximize network algorithm for lifetime optimization in wireless sensor networks,” Journal
lifetime and applicable to various cognitive radio enabled IoT of Parallel and Distributed Computing, 2020.

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 00:01:33 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2021.3065691, IEEE Internet of
Things Journal

[15] K. Ntshabele, B. Isong, N. Dladlu, and A. M. Abu-Mahfouz, “Energy [38] L. Sun, W. Zheng, N. Rawat, V. Sawant, and D. Koutsonikolas, “Perfor-
consumption challenges in clustered cognitive radio sensor networks: mance comparison of routing protocols for cognitive radio networks,”
A review,” in 2019 IEEE 28th International Symposium on Industrial IEEE transactions on mobile computing, vol. 14, no. 6, pp. 1272–1286,
Electronics (ISIE). IEEE, 2019, pp. 1294–1299. 2014.
[16] S. Zubair, N. Fisal, Y. S. Baguda, and K. Saleem, “Assessing routing [39] S. H. R. Bukhari, S. Siraj, and M. H. Rehmani, “Ns-2 based simulation
strategies for cognitive radio sensor networks,” Sensors, vol. 13, no. 10, framework for cognitive radio sensor networks,” Wireless Networks,
pp. 13 005–13 038, 2013. vol. 24, no. 5, pp. 1543–1559, 2018.
[17] M. Ozger and O. B. Akan, “Event-driven spectrum-aware clustering in
cognitive radio sensor networks,” in 2013 Proceedings IEEE INFOCOM.
IEEE, 2013, pp. 1483–1491.
[18] M. Zheng, S. Chen, W. Liang, and M. Song, “Nsac: A novel clustering
protocol in cognitive radio sensor networks for internet of things,” IEEE
Internet of Things Journal, vol. 6, no. 3, pp. 5864–5865, 2019.
[19] T. Wang, X. Guan, X. Wan, H. Shen, and X. Zhu, “A spectrum-aware
clustering algorithm based on weighted clustering metric in cognitive
radio sensor networks,” IEEE Access, vol. 7, pp. 109 555–109 565, 2019.
[20] H. Wu, F. Yao, Y. Chen, Y. Liu, and T. Liang, “Cluster-based energy
efficient collaborative spectrum sensing for cognitive sensor network,”
IEEE Communications Letters, vol. 21, no. 12, pp. 2722–2725, 2017.
[21] G. P. Joshi and S. W. Kim, “A survey on node clustering in cognitive
radio wireless sensor networks,” Sensors, vol. 16, no. 9, p. 1465, 2016.
[22] G. A. Shah, F. Alagoz, E. A. Fadel, and O. B. Akan, “A spectrum-aware
clustering for efficient multimedia routing in cognitive radio sensor
networks,” IEEE Transactions on Vehicular Technology, vol. 63, no. 7,
pp. 3369–3380, 2014.
[23] M. Tabassum, M. A. Razzaque, M. N. S. Miazi, M. M. Hassan, A. Ale-
laiwi, and A. Alamri, “An energy aware event-driven routing protocol
for cognitive radio sensor networks,” Wireless Networks, vol. 22, no. 5,
pp. 1523–1536, 2016.
[24] R. N. Yadav, R. Misra, and D. Saini, “Energy aware cluster based routing
protocol over distributed cognitive radio sensor network,” Computer
Communications, vol. 129, pp. 54–66, 2018.
[25] M. Ozger, F. Alagoz, and O. B. Akan, “Clustering in multi-channel
cognitive radio ad hoc and sensor networks,” IEEE Communications
Magazine, vol. 56, no. 4, pp. 156–162, 2018.
[26] S. Liu, L. Lazos, and M. Krunz, “Cluster-based control channel alloca-
tion in opportunistic cognitive radio networks,” IEEE Transactions on
Mobile Computing, vol. 11, no. 10, pp. 1436–1449, 2012.
[27] M. Bradonjić and L. Lazos, “Graph-based criteria for spectrum-aware
clustering in cognitive radio networks,” Ad Hoc Networks, vol. 10, no. 1,
pp. 75–94, 2012.
[28] T. Stephan, F. Al-Turjman, K. S. Joseph, B. Balusamy, and S. Srivastava,
“Artificial intelligence inspired energy and spectrum aware cluster based
routing protocol for cognitive radio sensor networks,” Journal of Parallel
and Distributed Computing, 2020.
[29] I. F. Akyildiz, W.-Y. Lee, M. C. Vuran, and S. Mohanty, “Next
generation/dynamic spectrum access/cognitive radio wireless networks:
A survey,” Computer networks, vol. 50, no. 13, pp. 2127–2159, 2006.
[30] D. West, “Introduction to graph theory , prntice-hall,” Englewood Cliffs,
NJ, 2001.
[31] R. Peeters, “The maximum edge biclique problem is np-complete,”
Discrete Applied Mathematics, vol. 131, no. 3, pp. 651–654, 2003.
[32] D. S. Hochbaum, “Approximating clique and biclique problems,” Jour-
nal of Algorithms, vol. 29, no. 1, pp. 174–200, 1998.
[33] P. G. V. Naranjo, M. Shojafar, H. Mostafaei, Z. Pooranian, and E. Bac-
carelli, “P-sep: A prolong stable election routing algorithm for energy-
limited heterogeneous fog-supported wireless sensor networks,” The
Journal of Supercomputing, vol. 73, no. 2, pp. 733–755, 2017.
[34] R. M. Eletreby, H. M. Elsayed, and M. M. Khairy, “Cogleach: A
spectrum aware clustering protocol for cognitive radio sensor networks,”
in 2014 9th international conference on cognitive radio oriented wireless
networks and communications (CROWNCOM). IEEE, 2014, pp. 179–
184.
[35] W. R. Heinzelman, A. Chandrakasan, and H. Balakrishnan, “Energy-
efficient communication protocol for wireless microsensor networks,”
in Proceedings of the 33rd annual Hawaii international conference on
system sciences. IEEE, 2000, pp. 10–pp.
[36] A. S. Kozal, M. Merabti, and F. Bouhafs, “Spectrum sensing-energy
tradeoff in multi-hop cluster based cooperative cognitive radio net-
works,” in 2014 IEEE Conference on Computer Communications Work-
shops (INFOCOM WKSHPS). IEEE, 2014, pp. 765–770.
[37] M. Di Felice, K. R. Chowdhury, W. Kim, A. Kassler, and L. Bononi,
“End-to-end protocols for cognitive radio ad hoc networks: An evalua-
tion study,” Performance Evaluation, vol. 68, no. 9, pp. 859–875, 2011.

2327-4662 (c) 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Carleton University. Downloaded on May 28,2021 at 00:01:33 UTC from IEEE Xplore. Restrictions apply.