www.nature.

com/scientificreports

OPEN A weighted energy consumption

minimization‑based multi‑hop
uneven clustering routing protocol
for cognitive radio sensor networks
Jihong Wang* & Conghui Li
Aiming at solving the effective data delivery and energy hole problem in multi-hop cognitive radio
sensor networks (CRSNs), a weighted energy consumption minimization-based uneven clustering
(ECMUC) routing protocol is proposed in this paper. For the first time, the impact of control overhead
on the network performance is taken into consideration, to be specific, the energy consumption
of control overhead is integrated with that of data communication to model the network energy
consumption. Through effective transformation and theoretical analysis, cluster radius of each ring
is derived by minimizing the network energy consumption and balancing the residual energy among
nodes in different rings. Distributed cluster heads (CHs) selection and cluster formation are carried
out within this range to control the cluster size and the corresponding energy cost. Expected times
for being CHs metric is defined to measure nodes’ energy and spectral potential and help select
powerful CHs. Simulation results show that ECMUC protocol is superior to most clustering protocols
designed for CRSNs in terms of network surveillance capability and network lifetime, and it is also
demonstrated that taking control overhead into consideration is beneficial for improving the network
performance.

Cluster-based cognitive radio sensor networks (CRSNs) are promising solutions to solve the energy efficiency
and spectral efficiency problem faced by legacy wireless sensor networks (WSNs)1. Cognitive radio technology
allows sensors to exploit licensed spectrum in an opportunistic manner and relieves the performance degradation
incurred by spectrum shortage in unlicensed b ­ ands2. Due to the introduction of CR, the definition of neighbors
has changed, i.e., two nodes are neighbors if they can satisfy the Euclidean distance requirement and share at
least one common available channel. However, CR functions such as spectrum sensing and dynamic spectrum
access will consume limited node energy, and the energy constraint problem of CRSNs is further aggravated.
Therefore, minimizing the network energy consumption becomes the primary goal of protocol design for CRSNs.
Clustering logically groups similar nodes in proximity and reduces traffic through data fusion and aggregation,
which can help conserve ­energy3. In large-scale CRSNs, all nodes cannot reach the sink through single-hop com-
munication, and multi-hop inter-cluster routing problem needs to be solved. Therefore, multi-hop clustering
routing protocol design for CRSNs has become a hot topic in academia and industry.
The limitations of current research on multi-hop clustering routing protocol design for CRSNs can be sum-
marized as follows:

• Current literature just focuses on the energy consumed by data communication and ignores the control over-
head incurred during cluster heads (CHs) selection, cluster formation and route selection when theoretically
analyzing the optimal number of clusters. Control information exchange will bring non-negligible energy
consumption to the whole n ­ etwork4. Ignoring this part of energy consumption is beneficial for simplifying
the theoretical analysis, but the obtained results may be inappropriate for achieving the design goal.
• Most existing clustering protocols for CRSNs are uniform clustering protocols which establish clusters with
almost the same size. Due to the towards-the-sink type of traffic in CRSNs, CHs close to the sink need to
perform more inter-cluster relay forwarding tasks which will accelerate their energy exhaustion. The energy
hole problem incurred by uneven distribution of energy consumption among CHs will o ­ ccur5. In addition,

School of Electrical Engineering, Northeast Electric Power University, Jilin 132012, China. *
email:
[email protected]

Scientific Reports | (2022) 12:14039 | https://doi.org/10.1038/s41598-022-18310-9 1

Vol.:(0123456789)
 www.nature.com/scientificreports/

the early death of CHs which are close to the sink and network partition will significantly degrade the sur-
veillance capability of ­CRSNs6.
• Current uneven clustering protocols proposed for CRSNs usually adopt constant coefficients, and the cluster
radius is designed to linearly increase as the Euclidean distance to the sink increases. However, the optimality
of the simple linear relationship has not been proven, and the coefficients should be analyzed and optimized
according to specific network configurations instead of fixing their values. Here, the specific network configu-
rations refer to the network size, node density, maximum node transmission range, data packet size, control
packet size and Euclidean distance to the sink.

In order to take full advantages of uneven clustering, i.e., enabling the CHs close to the sink to conserve more
energy for inter-cluster relay and solving the energy hole problem e­ ffectively7, a weighted energy consumption
minimization-based uneven clustering (ECMUC) routing protocol is proposed in this paper. By minimizing
the network energy consumption and balancing the residual energy among nodes in different rings, the optimal
cluster radius under specific network configurations is derived. ECMUC partitions the whole CRSNs into uni-
form rings and forms uneven clusters based on the obtained cluster radius within these rings. The innovations
of ECMUC protocol are summarized as follows:

• Energy consumption of data communication and control overhead are for the first time integrated to model
the network energy consumption, and minimizing it is the primary design goal of ECMUC protocol.
• The objective of minimizing the network energy consumption and balancing the residual energy among
nodes in different rings to solve the energy hole problem is innovatively transformed into minimizing the
weighted sum of the energy consumption of each ring and the additional energy consumption introduced to
the whole network by it. Through theoretical analysis, cluster radius in each ring is reasonably set to control
cluster size.
• Expected times for being CHs (ETBCHs) metric is defined and leveraged to measure nodes’ energy and
spectral potential. Based on ETBCHs comparison within the cluster radius, the most powerful node in the
neighborhood becomes a final CH. ECMUC protocol can autonomously achieve distributed CHs selection,
cluster construction and multihop inter-cluster route selection to forward the monitored data towards the
sink. Extensive simulations show that ECMUC protocol can significantly improve the network surveillance
capability and extend the network lifespan.

The rest of this paper is organized as follows. “Related work” section reviews clustering protocols designed for
CRSNs, including uniform and uneven clustering protocols. System model is described and ECMUC protocol is
explained in details in “Methods” section. “Results and discussion” section tests and evaluates the performance
of ECMUC protocol. Finally, “Conclusions” section concludes the paper and points out our future research
directions.

Related work
According to the data report model adopted, current clustering protocols for CRSNs can roughly be classified
into event-driven clustering protocols and time-triggered clustering protocols. Event-driven clustering protocols,
such as ­ESAC8, ­mESAC9 and ­ERP10, are triggered by emergent events and form temporary clusters in the cor-
ridor between the event and the sink until the event ends up. They are designed for event-driven applications
and are not suitable for applications which require continuous and periodical information c­ ollection11. On the
contrary, time-triggered clustering protocols form clusters in the whole network and maintain them by periodi-
cal calculation and communication until the end of network lifetime. They can be further divided into uniform
clustering protocols and uneven clustering protocols based on the cluster size. In the following sub-sections, we
will review related work from these two aspects.

Uniform clustering protocols for CRSNs. Uniform clustering protocols for time-triggered CRSNs can
be further divided into 3 sub-categories, i.e., centralized, distributed and hybrid protocols. CogLEACH-C12,
­ABCC13 and ­IMOCRP14 are typical representatives of centralized protocols. Though different factors are con-
sidered when selecting CHs, necessary information should be collected and handled by the sink, which requires
that all CRSNs nodes should reach the sink through single-hop communication. Therefore, they all suffer prob-
lems such as network scalability and single point of failure, which limits their application field.
Most clustering protocols for time-triggered CRSNs are distributed and uniform cluster-based in nature, such
as ­CogLEACH15, ­NSAC16, ­EACRP17, ESUCR​18 and so on, as shown in Table 1. These protocols select CHs and
form clusters in the locality through extensive information exchange. Among them, CogLEACH is a distributed
spectrum-aware extension of LEACH ­protocol19. CogLEACH uses the number of idle available channels as
probability weight for CHs selection, and each CRSNs node can independently judge whether itself can become
a CH or not by comparing its CHs weight with a random number. NSAC builds stable clusters by taking the
spectrum dynamics and energy consumption comprehensively into consideration. Nodes with the highest weight
in the neighborhood win competition and become CHs, and other nodes in the maximum edge biclique become
corresponding cluster members (CMs). These nodes will be excluded from clustering, and cluster formation
continues until all CRSNs nodes are clustered. EACRP and ESUCR iteratively merge neighboring clusters until
the optimal number of clusters is achieved. Except for the residual energy, the number of available channels,
the number of neighbors and the distance to the sink are used to compute CHs weight. Primary and secondary
gateway nodes are leveraged to relay packets between neighboring CHs.

Scientific Reports | (2022) 12:14039 | https://doi.org/10.1038/s41598-022-18310-9 2

Vol:.(1234567890)
www.nature.com/scientificreports/

Control overhead
Protocols Type Scenarios Clustering Routing Inter-cluster routing Objective
CogLEACH-C C SHop-All 3 N-K 0 × ×
ABCC C SHop-All N 0 × MinASDE
IMOCRP C SHop-All N 0 × MinASDE
CogLEACH D SHop-CHs 2N 0 × ×
NSAC D SHop-CHs – 0 × ×
EACRP D MHop 3 N × iteEACRP – √ ×
ESUCR​ D MHop 4 N × iteESUCR​ – √ ×
WCM H SHop-CHs 4N 0 × ×
LEAUCH D MHop 2 N + 2Ncandi – × ×
R-bUCRP D SHop-CHs N + 2Ncandi 0 × ×
IACUCAPTEEN D MHop 2 N + Ncandi – √ ×
ESAUC​ D MHop N + 2Ncandi – √ ×

Table 1.  Characteristics analysis of existing clustering protocols for CRSNs. C: centralized, D: distributed,
H: hybrid; SHop-All: single-hop communication between all nodes and the sink, SHop-CHs: single-hop
communication between CHs and the sink, MHop: multi-hop communication; MinASDE: minimize the
average node energy consumption and the standard deviation of node residual energy; N is the number
of living nodes in current round; K is the optimal number of CHs; iteEACRP and iteESUCR​ are the number of
merging iterations performed by EACRP and ESUCR, respectively; Ncandi is the number of candidate CHs
in uneven clustering protocols; ― represents that the corresponding value is unable to be explicitly
quantified; × denotes the corresponding problem has not been solved while √ represents the opposite situation.

WCM-based spectrum-aware clustering protocol (hereinafter refer to as WCM)20 is a typical case of hybrid
clustering protocols. In WCM, a new weighted metric which simultaneously evaluates the temporal-spatial
correlation, confidence level and residual energy is proposed to select CHs. Then the sink merges neighboring
clusters with the highest temporal-spatial correlation until the objective function cannot be improved any more.
Extensive control information exchange will quickly drain node battery, and then nodes will die.
In all uniform clustering protocols, besides intra-cluster data communication, CHs near the sink should also
relay packets for other parts of the network, which means balanced energy distribution among CHs cannot be
guaranteed. Therefore, early death of CHs near the sink will occur and this will further result in network partition.

Uneven clustering protocols for CRSNs. Different from above uniform clustering protocols, cluster
radius in uneven clustering protocols increases as the distance to the sink increases, which can help balance
the residual energy among CHs. ­LEAUCH21 is a representative uneven clustering protocol for CRSNs, and it
determines final CHs by competition among candidate CHs within the cluster radius. Cluster radius is calculated
according to the Euclidean distance to the sink di,sink, as shown in Eq. (1).
 
dmax − di,sink
Rci = 1 − c Rc0 (1)
dmax − dmin
where Rc0 is the maximum cluster radius of candidate CHs; c is a constant coefficient for uneven clustering;
dmax and dmin are the maximum and minimum Euclidean distance to the sink from CRSNs nodes, respectively.
However, the candidate CHs selection threshold 0.4 has a large impact on network performance, and in most
cases, no candidate CHs can be selected at all. R-bUCRP22 adopts the same manner to calculate cluster radius,
and its random candidate CHs selection cannot guarantee the optimal distribution of candidate CHs. Therefore,
the CHs selection is largely affected. I­ ACUCAPTEEN23 and ESAUC​24 improve the cluster radius calculation of
LEAUCH by taking more factors into consideration, such as the node residual energy, the number of neighbors
and the number of available channels. However, they still adopt fixed weighted coefficients which need to be
analyzed and optimized according to specific network configurations. In addition, the weighted coefficient ω
in ESAUC is not given.
As stated above, current uneven clustering protocols are basically proposed on the basis of LEAUCH, there-
fore, cluster radius is not obtained by theoretical derivation with the purpose of balancing the residual energy
among CHs. Instead, the cluster radius is designed to linearly increase as the Euclidean distance to the sink
increases, which needs to be analyzed and ­optimized25.
From above table and analysis, we can obtain the following observations:

• Current research on theoretical analysis of the optimal number of clusters wholly focuses on the energy
consumption of data communication, while the energy cost of control information exchange is neglected. In
addition, in order to simplify analysis, current literature usually simplifies the calculation process of energy
consumption of inter-cluster communication, that is to say, assumptions are made that each data packet
needs at most once relay to reach the sink. Actually, multi-hop inter-cluster relay is more practical, especially

Scientific Reports | (2022) 12:14039 | https://doi.org/10.1038/s41598-022-18310-9 3

Vol.:(0123456789)
 www.nature.com/scientificreports/

Figure 1.  Uniform ring division with ring width Rt.

in large-scale CRSNs. This may restrict the network scalability and lead to the inapplicability of the derived
results.
• Current uniform clustering protocols cannot guarantee balanced energy distribution among CHs. Therefore,
early death of the CHs which are close to the sink will occur and this will further result in network partition.
• In existing uneven clustering protocols for CRSNs, the cluster radius is usually quantified under given con-
stant coefficients, and it is assumed to be linearly proportional to the Euclidean distance to the sink. However,
in order to take full advantages of the potentials of uneven clustering in balancing the residual energy among
CRSNs nodes, these coefficients and cluster radius should be optimized according to specific network con-
figurations.

All these motivate us to propose ECMUC protocol which integrates energy consumption of data communica-
tion and control information exchange together for the first time and solves the energy hole problem effectively
by balancing the residual energy among nodes in different rings as much as possible.

Methods
System model. N homogeneous CRSNs nodes with initial energy E0 are uniformly and randomly distrib-
uted in the surveillance area (the area is approximated as a circle), coexisting with P randomly-deployed primary
users (PUs). The sink is located at the center to gather useful information about the whole area and provides it to
end ­users26. Once deployed, CRSNs nodes no longer move unless their residual energy is exhausted. Each CRSNs
node is configured with one transceiver, so it cannot send out and receive information at the same time. Addi-
tionally, it leverages received signal strength or certain localization algorithm to obtain its relative distance from
the receiver, according to which it can adapt its transmission power. Competition-based method is applied for
CHs selection, and the most qualified node in the neighborhood becomes final CH who is responsible for intra-
cluster data collection and aggregation. The data aggregation effect of CHs can be measured by aggregation coef-
ficient β. Here, perfect aggregation is assumed, which means packets from CMs can be fused into a single packet
with fixed length, i.e., β is equal to the reciprocal of the number of CRSNs nodes in the whole ­cluster27. Restricted
by limited node transmission range, multi-hop communication is exploited to realize inter-cluster data delivery.
In order to determine cluster membership based on the Euclidean distance to the sink, the surveillance area
is partitioned into different rings around the sink. Uneven ring division usually results in more rings, especially
for large-scale CRSNs. More rings mean more times of inter-cluster data relay which will lead to heavier energy
consumption. Therefore, in order to reduce energy consumption and guarantee the minimum data transmission
delay, we prefer to organize our CRSNs into uniform rings whose width is Rt, as shown in Fig. 1.
Each CRSNs node consumes energy in sending and receiving control and data packets, and the energy con-
sumption of transmitting an l-bit packet over distance dj,k is quantified b ­ y19:
 2
l × (Eelec + εfs dj,k ) if dj,k ≤ d0
ETX = 4 (2)
l × (Eelec + εmp dj,k ) otherwise

where Eelec is the energy consumption of transceiver electronics per bit; εfs and εmp are the energy consumption
of power amplifier per bit in free-space and multi-path loss model, respectively; d0 is the distance threshold, and
d0 = (εfs/εmp)1/2. The energy consumption of receiving corresponding packet is:

Scientific Reports | (2022) 12:14039 | https://doi.org/10.1038/s41598-022-18310-9 4

Vol:.(1234567890)
www.nature.com/scientificreports/

ERX = l × Eelec (3)

Markov renewal process is widely used in literature to imitate PUs activity, and it assumes ON/OFF model
­ ehavior28. In other words, PUs activity on given channel alternates between ON and OFF states whose
for PUs b
duration are geometrically distributed random variables with parameters pc and qc.

Weighted energy consumption minimization‑based multi‑hop uneven clustering routing pro‑

tocol design for CRSNs. As mentioned earlier, our design goal is to minimize the energy consumption of
all rings in CRSNs and balance the residual energy among nodes in different rings to avoid energy holes. The
energy hole problem occurs due to the uneven distribution of energy consumption among nodes in different
rings, i.e., nodes in inner rings drain their battery more quickly than those in outer rings. Therefore, in order to
solve this problem, ECMUC must slow down the node death rate of inner rings to guarantee that it should not
be faster than outer rings. Based on uniform ring division with fixed ring width Rt, our clustering objective can
be transformed into minimizing the weighted sum of the energy consumption of ring i (i ∈ [1, 2, …, Lmax]) and
the additional energy consumption brought to the whole network by it. Here, Lmax is the maximum ring number
of all CRSNs nodes, as shown in Eq. (4).
 � 
max (xj − xsink )2 + (yj − ysink )2
∀ j∈[1,2,··· ,N]
Lmax =  (4)
 
Rt

 
 

where N is the total number of CRSNs nodes; (xj, yj) and (xsink, ysink) are the coordinates of node j and the sink,
respectively. Rt is the maximum transmission range of CRSNs nodes. Each node j determines its ring index
according to Eq. (5) below.
dj,sink
 
L(j) = (5)
Rt
where dj,sink is the Euclidean distance from node j to the sink. L(j) will be used for CHs selection and cluster
formation later on.
In order to achieve our clustering objective, ECMUC enables all CRSNs nodes in ring 1 to act as separate
CHs and share the inter-cluster data forwarding tasks, which can help reduce their energy consumption speed.
In addition, ECMUC balances the residual energy among nodes in different rings by reasonably setting cluster
radiuses, and we will show how to determine their values through theoretical analysis according to specific
network configurations. Of course, ECMUC also tries to balance the residual energy of nodes in the same ring
as much as possible, such as CHs rotation among nodes in ring i (i ≠ 1) based on the ETBCHs metric or ena-
bling nodes in ring i (i ≠ Lmax) to take turns acting as relays. ETBCHs metric is defined by combining the energy
potential and channel availability of CRSNs nodes. Based on this metric, CHs competition in the neighborhood
is carried out to determine final CHs who are responsible for building their own clusters and search for available
inter-cluster routing paths.

Design details of ECMUC protocol. In CRSNs, due to the dynamic channel occupancy behavior of PUs,
the channel availability is time- and position-dependent. At the beginning of each round, each CRSNs node j
leverages CR functions such as spectrum sensing to perceive the channel availability at its position and decides
its available channel list Cj, which determines whether CRSNs nodes can communicate with each other and
whether information transmission can be successful. CHs selection, cluster formation and route selection of
ECMUC are all based on the available channel information. In other words, channel availability obtained from
CR functions will affect the clustering results and the next-hop relay selection, which will further affect node
energy consumption. In ECMUC, CHs competition is conducted among neighboring nodes belonging to the
same ring, and different strategies will be applied for distinct Lmax. To be specific, if all CRSNs nodes can reach
the sink through single-hop communication, i.e., Lmax = 1, energy consumption of the whole network can be
minimized by setting each CRSNs node as a separate CH. In other words, non-clustering is the most energy-
efficient way of network organization in this ­scenario14. For example, if data from n normal nodes needs to be
transmitted to the sink, for simplicity reason, distance from each node to the sink dtosink is assumed to be the
same and dtosink ≤ d0.
Without clustering (as shown in Fig. 2a), the total energy consumption Etotal is:
2
Etotal = n × (Eelec + εfs dtosink )×l (6)

In clustering case, assuming that data from other nodes is aggregated at node A and then the aggregated data
is transmitted to the sink, as shown in Fig. 2b. The total energy consumption Etotal’ is:
n−1

′ 2 2
Etotal = (Eelec + εfs di,A ) × l + (n − 1) × Eelec × l + n × EDA × l + (Eelec + εfs dtosink )×l (7)
i=1

Scientific Reports | (2022) 12:14039 | https://doi.org/10.1038/s41598-022-18310-9 5

Vol.:(0123456789)
 www.nature.com/scientificreports/

Figure 2.  Comparison of energy consumption in non-clustering and clustering scenarios.

where the first item on the right side is the total energy consumption of data transmission from other nodes
to A. The second item is the energy consumed by A in data reception. The third item is the energy cost of data
aggregation at A, and the last item is the energy spent on data transfer from A to the sink.
The difference in the total energy consumption between above two cases is as follows:
n−1
 

′ 2 2
�E = Etotal − Etotal = l × (n − 1) × Eelec + n × EDA + εfs di,A − (n − 1) × εfs dtosink (8)
i=1

For typical transmission range of CRSNs nodes, it is usually the case that ∆E > 0, which means non-clustering
can lead to lower energy consumption. This conclusion will be verified by the simulation results in Performance
analysis in Case 1 section. In this case, each living CRSNs node does not need to consume extra energy to
exchange control information for CHs selection and cluster formation. Instead, it sends its monitored data to
the sink directly to minimize the total energy consumption. In particular, data transmission is performed on
randomly selected channel from available channel list, and the purpose is to reduce the number of competing
CHs who contend for channel access when carrier sense multiple access with collision avoidance (CSMA/CA)
protocol is applied.
If all CRSNs nodes cannot reach the sink through single-hop communication, i.e., Lmax ≥ 2, all CRSNs nodes
in ring 1 still act as independent CHs. This enables more nodes to share the inter-cluster data forwarding tasks
and reduce the energy consumption speed, which is beneficial for solving the energy hole problem. ETBCHs
metric should be compared among neighbors in the same ring (except ring 1) to compete for CHs, and the one
with the highest ETBCHs value in the neighborhood becomes final a CH. In round t, ETBCHs value of node j
(in ring i) is defined as follows:
�cj
ncommon_k
 
 
Ej (9)
 × log2 (1 + k=1
 
ETBCHs(j) =  ECH_j(i) +Esw ×intra_sw
)
C × neighj
neighj +1

where Ej is the residual energy of node j; ECH_j(i) is the expected energy consumption of node j for being a CH
once in ring i, and of course, it does not include the energy consumption of channel switching. Therefore, the
energy spent on intra-cluster channel switching is further added. Esw is the energy consumption per channel
switching, and intra_sw is the expected minimum times of channel switching performed by j. neighj is the num-
ber of neighbors of node j. ⌊x⌋ returns the nearest integer which is smaller than or equal to x. As CHs role will
rotate among node j and its neighbors as network operation goes on, j will become a CH every neighj + 1 rounds
on average. Therefore, the first item in Eq. (9) quantifies the maximum number of times to be a CH from energy
perspective for node j. If node j does not have enough energy for being a CH in this round, its ETBCHs value is
zero. This can avoid unnecessary energy waste during cluster formation and data transmission. cj is its number
of available channels. C is the total number of licensed channels. ncommon_k is the number of neighbors who share
channel k with j. Weall
cj
know that if node j has no available channel or it does not share common channels with
its neighbors, i.e., k=1 ncommon_k = 0, it should not become a CH. In order to guarantee this, we add 1 to the
log function. Actually, the log function item represents the expected probability of being a CH from channel
connectivity perspective. Therefore, ETBCHs value measures the potential times of node j for being a CH by
combining its energy potential and spectral potential.
In order to calculate ETBCHs value, we need to obtain cj, neighj, ECH_j(i), intra_sw and ncommon_k. Among them,
ECH_j(i) can be obtained through analysis in Theoretical analysis of cluster radius ri section, and we use Fig. 3 to
illustrate how to acquire the values of other parameters.
Assuming that 5 nodes A, B, C, D and E are located within the cluster radius range (ri) of node j, we have
neighj = 5. The bar with numbers next to each node shows the availability of all licensed channels. From Fig. 3,
we can see that 5 licensed channels are all available to j, therefore, cj = 5. Node j can reach nodes B, C and D on
channel 1, therefore, ncommon_1 = 3. Similarly, ncommon_2 = 2, ncommon_3 = 3, ncommon_4 = 4, ncommon_5 = 3. If node j can
become a final CH, it will broadcast time division multiple access (TDMA) schedule which assigns dedicated
time slot and channel to each CM within ri. The CH will choose the channel which is available to majority of
its CMs as cluster channel. If above channel is not available to a certain CM, the CH will select one common
available channel between them to enable their data delivery. We call this hybrid constraint, which means that

Scientific Reports | (2022) 12:14039 | https://doi.org/10.1038/s41598-022-18310-9 6

Vol:.(1234567890)
www.nature.com/scientificreports/

Figure 3.  Example of acquiring parameters for ETBCHs calculation.

groupwise ­constraint29 is the main constraint and pairwise ­constraint30 is used as supplement. intra_sw is relevant
to above intra-cluster schedule. In Fig. 3, as j can reach 4 neighbors on channel 4, channel 4 is specified as cluster
channel. However, node A cannot use channel 4 in this round, therefore, CH j should select one common avail-
able channel to communicate with A, say channel 5. In order to minimize extra energy consumption brought by
channel switching, the best schedule strategy for node j is scheduling transmissions from nodes B, C, D and E in
4 consequent time slots on channel 4, and then it switches to channel 5 to communicate with node A. Therefore,
it performs once channel switching in this case, that is, intra_sw = 1.
After ETBCHs calculation, all living CRSNs nodes compete for CHs according to Algorithm1, as shown in
Fig. 4.
Lines 4–9 show the exchange of ETBCHs value. Each living node j (L(j) ≠ 1) broadcasts its ETBCHs value
within cluster radius rL(j), and its neighbors who are located in the same ring add j into their competing node set.
Each node, say j, compares its ETBCHs value with its neighbors in its competing node set Compenode(j).
If ETBCHs(j) is smaller than one of them, node j quits from CHs competition, becomes a non-CHs node and
waits for joining one cluster as a normal CM. It broadcasts quit message within rL(j) to notify its neighbors, and
those who receive this message remove j from their competing node set, as shown in lines 11–18. Otherwise, if
ETBCHs(j) is the highest in the neighborhood, j becomes a final CH and broadcasts CHs advertisement within
rL(j) to announce its role. The advertisement includes its node ID, residual energy and available channel list to
provide sufficient information for normal nodes. Nodes in Compenode(j) quit from CHs competition on receiv-
ing the CHs advertisement, as shown in lines 19–22.
During CHs selection, control packets such as node information, ETBCHs value, quit message and CHs
advertisement are broadcast within the cluster radius. The energy consumed by these packets largely depends
on the value of the cluster radius, and we will analyze how to obtain its value in the Theoretical analysis of cluster
radius ri section.
After determining CHs in each ring, cluster formation simultaneously begins in all rings except ring 1. For
each living non-CHs node, say j, it selects its CH according to the number of common available channels and
the Euclidean distance between them. Node j sends out join request to the final CH and requests to become its
member if such CH exists. The CH receives the join message and adds j into its CMs list. If such CH cannot be
found, node j becomes a separate CH, and it will broadcast CHs advertisement similar to that in Algorithm1.
Above process continues until all nodes are clustered.
For each CH, say k, if its ring index L(CH(k)) ≥ 2, it should select one CH (represented by CH(s)) from the
next inner ring towards the sink to relay its packets. In order to provide information for next-hop relay selec-
tion, each CH whose ring index is larger than 1 but smaller than Lmax broadcasts its own information such as
residual energy and Euclidean distance to the sink within Rt. The next-hop relay is chosen among candidate
next-hop relays according to Eq. (10). Here, the candidate next-hop relay should satisfy the following conditions:
L(CH(s)) = L(CH(k))-1 and dCH(k),CH(s) ≤ Rt.
  Econ (CH (k), CH (s))
WCH(k) (CH(s)) = CCH(k) ∩ CCH(s)  × ECH(s) × (10)
Econ (CH (s), sink)
where CCH(k) and CCH(s) are the available channel set of CH(k) and CH(s), respectively. The first item on the right
side of Eq. (10) represents the number of common channels shared by CH(k) and CH(s). ECH(s) is the residual
energy of CH(s). Econ(CH(k),CH(s)) is the energy consumption of data communication between CH(k) and
CH(s), while Econ(CH(s),sink) is the energy consumption of data transmission from CH(s) to the sink. Actually,
the ratio of these two items helps select appropriate next-hop relay which is located near the sink and a little
far away from CH(k). The inner CH with more common available channels, more residual energy and closer to
the sink has higher probability of being selected. A notification message is sent to the next-hop relay to inform

Scientific Reports | (2022) 12:14039 | https://doi.org/10.1038/s41598-022-18310-9 7

Vol.:(0123456789)
 www.nature.com/scientificreports/

Figure 4.  Pseudo code of CHs selection in ECMUC protocol.

about the final routing decision. If such CH cannot be found, one of its CMs who is closer to the sink is chosen
to help relay the data packet towards the next inner ring. It should be noted that all nodes in ring 1 transmit
their information which includes the residual energy, available channels and so on through direct link to the
sink. The purpose is to reduce the energy cost during route selection, as there are numerous CHs in ring 1. Then
the sink will broadcast above information to CHs in ring 2, and these CHs can choose them as next-hop relays
according to Eq. (10).

Theoretical analysis of cluster radius. In this section, we will theoretically analyze how to set cluster
radius ri according to specific network configurations. Taking nodes in ring i (i ≠ 1) as an example, energy is
consumed in the following 4 phases: CHs selection, cluster formation, route selection and data transmission.

• The energy consumption of all nodes in ring i in CHs selection phase.

  In order to compete for CHs, each CRSNs node broadcasts its own information and ETBCHs value within
cluster radius ri, and these information will be received by neighbors in the same ring. The corresponding
energy consumption are (Eelec + εfsri2) × l1 and Eelec × l1 × (ni − 1), respectively. ni is the number of nodes within
ri. By comparison, some powerful CRSNs nodes become CHs and broadcast CHs advertisement, while oth-
ers become non-CHs nodes and broadcast quit messages. These messages will also be received by neighbors
within ri. Therefore, no matter for a CH or a CM, the energy consumption in CHs selection phase can be
represented by:

eCHs = 3(Eelec + εfs ri2 ) × l1 + 3Eelec × l1 × (ni − 1) = 3(ni × Eelec + εfs ri2 ) × l1 (11)
  Accordingly, the energy consumption of ring i in CHs selection phase is:

ECHs(i) = 3(ni × Eelec + εfs ri2 ) × l1 × Ni (12)

where Ni is the number of CRSNs nodes in ring i.
• The energy consumption of ring i in cluster formation phase.
  Each non-CHs node, say node j, selects its own CH and sends out join request message. The CH receives
the message and records j as its member. The energy consumption of a CM and a CH in cluster formation
phase are shown in Eqs. (13) and (14), respectively.
2
eCluster_CM = (Eelec + εfs dtoCH ) × l1 (13)

eCluster_CH = Eelec × l1 × (ni − 1) (14)

  Consequently, the energy consumption of ring i in cluster formation phase is:
2
ECluster(i) = (Eelec + εfs dtoCH ) × l1 × (ni − 1) × Ki + Eelec × l1 × (ni − 1) × Ki
2 (15)
= Ki × (ni − 1) × (2Eelec + εfs dtoCH ) × l1

Scientific Reports | (2022) 12:14039 | https://doi.org/10.1038/s41598-022-18310-9 8

Vol:.(1234567890)
www.nature.com/scientificreports/

Figure 5.  Illustration of PinRt calculation.

where Ki is the number of CHs in ring i.

• The energy consumption of ring i and the additional energy consumption introduced to the whole network
by ring i in route selection phase.
  Each CH in ring i − 1 broadcasts its own information within Rt to enable next-hop relay selection, and the
information will be received by CHs in ring i. Each CH in ring i determines its next hop towards the sink
and sends out routing notification. The corresponding inner CH receives this notification and becomes its
next-hop relay. Consequently, the energy consumption of ring i in route selection phase is:

ERoute(i) = Ki × Eelec × l1 × Ki−1 × PinRt + (Eelec + εfs Rt2 ) × l1 × Ki (16)

where Ki−1 is the number of CHs in ring i − 1, and PinRt is the probability that an inner-ring CH is in the
transmission range of ring-i CH. Therefore, Ki−1 × PinRt represents the excepted number of CHs in ring i − 1
from which ring-i CH can receive information. Fig. 5 is used to demonstrate how to calculate PinRt, and the
detailed processes are as follows:
  According to the Cosine Law, we have:
2
R2 + (i − 12 )Rt − [(i − 1)Rt ]2 i + 41

A = arccos t 1
= arccos (17)
2 × Rt × (i − 2 )Rt 2(i − 12 )

  Consequently, the arch area S1 is obtained by subtracting the area of the triangle from the area of corre-
sponding sector, as shown in Eq. (18).
A 1 1
S1 = πRt2 × − Rt2 × sin 2A = Rt2 × (A − sin 2A) (18)
π 2 2
  Similarly, we can obtain:
1
S2 = (i − 1)2 Rt2 × (B − sin 2B) (19)
2
  Therefore, the probability PinRt is:
1 1
S1 + S2 A− sin 2A + (i − 1)2 (B − sin 2B)
PinRt = = 2 2
(20)
STotal_i−1 (2i − 3)π
  The additional energy consumption brought to the whole network by ring i in route selection phase is:

�ERoute = Ki × Eelec × l1 + (Eelec + εfs Rt2 ) × l1 × Ki−1 (21)

• The energy consumption of ring i and the additional energy consumption introduced to the whole network
by ring i in data transmission phase.
  Each CH broadcasts TDMA schedule within its cluster, and it also receives and aggregates data from its
members and then transfers the aggregated information to the sink. In addition, each CH should relay data
packets from outer rings. Therefore, its energy consumption is calculated according to Eq. (22):

Scientific Reports | (2022) 12:14039 | https://doi.org/10.1038/s41598-022-18310-9 9

Vol.:(0123456789)
 www.nature.com/scientificreports/

Lmax
1 
eData_CH = (Eelec +εfs ri2 )×l1 +(ni −1)×Eelec ×l+ni ×EDA ×l+(Eelec +εfs Rt2 )×l+(2Eelec +εfs Rt2 )×l× Kj (22)
Ki
j=i+1

  Each CM receives the TDMA schedule from its CH and transmits data according to the schedule. Its energy
cost in this phase is:
2
eData_CM = Eelec × l1 + (Eelec + εfs dtoCH )×l (23)
  We can obtain the energy consumption of ring i in data transmission phase as follows:

EData(i) = Ki × {[(Eelec + εfs ri2 ) × l1 + (ni − 1) × Eelec × l + ni × EDA × l + (Eelec + εfs Rt2 ) × l
Lmax
1 
+ (2Eelec + εfs Rt2 ) × l × 2
Kj ] + (ni − 1) × [Eelec × l1 + (Eelec + εfs dtoCH ) × l]}
Ki
j=i+1
(24)
= Eelec × (l1 + l) × Ni + EDA × l × Ni + Ki εfs ri2 × l1 + Ki εfs Rt2 × l
L
max
+ (2Eelec + εfs Rt2 ) × l × 2
Kj + Ki × (ni − 1) × (Eelec + εfs dtoCH )×l
j=i+1

  Data packets from ring i should be relayed to the sink with the help of inner rings, therefore, the additional
energy consumption brought to the whole network by ring i can be obtained as follows:

�EData = (2Eelec + εfs Rt2 ) × l × Ki × (i − 1) (25)

• Analysis of the final results.
  The total energy consumption of ring i Etotal(i) can be obtained by merging Eqs. (12), (15), (16) and (24),
while the additional energy consumption brought to the whole network by ring i ΔEadd can be derived by
merging Eqs. (21) and (25).

Etotal(i) = 3(ni × Eelec + εfs ri2 ) × l1 × Ni + Ki × (ni − 1) × (2Eelec + εfs dtoCH

2
) × l1 + Ki × (Eelec
+ εfs Rt2 + Eelec × Ki−1 × PinRt ) × l1 + Eelec × (l1 + l) × Ni + EDA × l × Ni + Ki εfs ri2 × l1
L
max
(26)
+ Ki εfs Rt2 × l + (2Eelec + εfs Rt2 ) × l × 2
Kj + Ki × (ni − 1) × (Eelec + εfs dtoCH )×l
j=i+1

�Eadd = Ki × Eelec × l1 + (Eelec + εfs Rt2 ) × l1 × Ki−1 + (2Eelec + εfs Rt2 ) × l × Ki × (i − 1) (27)
  As we know, the number of CRSNs nodes in ring i Ni = Ki × ni = Ki × πri2ρ, and here ρ is the node density.
According ­to31, the expected squared distance from a CM to its CH ξ[dtoCH2] can be estimated by:
 2π  ri
r2
2
ξ [dtoCH ]= r 2 ρ(r, θ)drdθ = i (28)
θ =0 r=0 2
  Our goal is to minimize the weighted sum of the energy consumption of ring i and the additional energy
consumption brought to the whole network by it. The objective function can be expressed as:
Minimize (α × Etotal(i) + �Eadd ) (29)
where α is the weighted coefficient assigned to Etotal(i). Etotal(i) is closely related to the number of clusters Ki.
Under given topology, by minimizing the energy consumption of each ring Etotal(i), we can achieve the goal of
minimizing the network energy consumption and extending network lifespan. ΔEadd is the extra energy bur-
den brought by ring i which is also related to the number of clusters Ki, and if we can minimize the additional
energy consumption brought to inner rings by each ring, the residual energy among nodes in different rings
can be well balanced and the energy hole problem can be solved effectively. When the area of ring i and the
total number of nodes in ring i Ni are given, as can be seen from Eqs. (26) and (27), increasing the number
of clusters Ki, Etotal(i) will decline while ΔEadd will rise. In this case, the energy hole problem may occur and
network surveillance capability will be negatively affected. On the contrary, decreasing Ki, Etotal(i) will increase
while ΔEadd will reduce. The increase of Etotal(i) will lead to faster node death in ring i, while the decrease of
ΔEadd will lead to slower node death in inner rings. In this case, network lifetime will be shortened. In other
words, there is a compromise between Etotal(i) and ΔEadd, i.e., there exists a reasonable Ki which can help avoid
energy holes while extending network lifetime as much as possible. Therefore, by taking the derivative of
Eq. (29) with respect to Ki and setting the result to 0, we can obtain the optimal number of clusters Ki, and
then we can obtain the optimal cluster radius ri through Eq. (30).

Ni
ri = (30)
πρKi

Scientific Reports | (2022) 12:14039 | https://doi.org/10.1038/s41598-022-18310-9 10

Vol:.(1234567890)
www.nature.com/scientificreports/

Parameters Values
Maximum node transmission range Rt 50 m
Node density ρ 1/25π/m2
Initial energy of each CRSNs node E0 0.5 J
Energy consumption of transceiver electronics per bit Eelec 50nJ/b
Energy consumption of amplifier in free-space loss model εfs 10pJ/b/m2
Energy consumption of amplifier in multi-path loss model εmp 0.0013pJ/b/m4
Energy consumption of data aggregation EDA 5nJ/b/packet
Data packet size l 1,000b
Control packet size l1 100b
Energy consumption per channel switching Esw 10 μJ
Number of PUs P 5
Total number of licensed channels C 5
Probability vector of ON states [0.2, 0.3, 0.4, 0.5, 0.6]

Table 2.  Simulation parameter settings.

  In addition, by combining Eqs. (11), (14), (16) and (22), the expected energy consumption of being a CH
in ring i ECH_i is shown in Eq. (31), and by substituting Rt with the actual Euclidean distance from node j to
the sink, the obtained ECH_j(i) can be used in Eq. (9).

ECH_(i) = 3(ni × Eelec + εfs ri2 ) × l1 + Eelec × l1 × (ni − 1) + (Eelec + εfs Rt2 ) × l1
+Eelec × l1 × Ki−1 × PinRt + (Eelec + εfs ri2 ) × l1 + Eelec × l × (ni − 1)
Lmax (31)
1 
+EDA × l × ni + (Eelec + εfs Rt2 ) × l + (2Eelec + εfs Rt2 ) × l × Kj
Ki
j=i+1

Results and discussion

In this section, we test the performance of ECMUC protocol through extensive simulations by using Matlab
which is one of the most commonly used simulation tools in evaluating the performance of clustering protocols
designed for C­ RSNs14,20,21,26. In order to capture the impact of network size, we keep the node density ρ unchanged
and analyze the performance of ECMUC protocol in three cases. In Case 1, N = 100 (Lmax = 1) is picked to assess
the performance of ECMUC when all CRSNs nodes can reach the sink through single-hop communication. In
Case 2 and Case 3, N = 400 (Lmax = 2) and N = 900 (Lmax = 3) are picked to evaluate the performance of ECMUC
with multi-hop inter-cluster relay. Other parameter settings are shown in Table 2.
ECMUC protocol is compared with current clustering protocols for CRSNs to verify its effectiveness, such
as ­IMOCRP14, ­CogLEACH15, ­NSAC16, ­EACRP17, ­WCM20 and ­LEAUCH21. These protocols have been explained
in the Related work section. Evaluation metrics, such as the number of living nodes at the end of round t, the
ratio of effective data gathering nodes, the total control overhead and the number of selected CHs per round,
are leveraged for performance comparison.

Performance analysis in Case 1. Performance comparison results with respect to above evaluation met-
rics in Case 1 are provided in Fig. 6. Among them, Fig. 6a shows the variations of number of living nodes with
time (represented by round number) when these clustering protocols are applied. From it, we can observe that
the number of living nodes of all competing protocols decreases as round number increases. CRSNs nodes are
powered by limited-capacity battery, and tasks such as information transmission and reception, data aggrega-
tion and so on all consume energy and result in rapid reduction in residual energy. When the energy left in the
battery is exhausted, the node is neglected from network operation and the number of living nodes decreases.
In order to compare these protocols explicitly, we record the rounds in which the first death node appears.
Among these competitors, ECMUC protocol achieves the best performance. The first death node of ECMUC
appears in round 6,668 which is much later than other protocols. The reasons can be analyzed from the sources
of energy dissipation, i.e., the total control overhead and the number of selected CHs per round. Detailed analysis
is listed below:

• As shown in Fig. 6b, the total control overhead of ECMUC in each round is always 0. In other words, there is
no control information exchange among neighboring CRSNs nodes or between CRSNs nodes and the sink,
which helps conserve energy. In this case, CRSNs nodes know that the sink is located within their communi-
cation range. They become separate CHs and directly deliver their monitored data to the sink. This explains
why the number of selected CHs per round of ECMUC in Fig. 6c is obviously higher than its competitors,
and this result is consistent with our theoretical analysis in Design details of ECMUC protocol section. Each
living CRSNs node becomes an independent CH, which can result in the minimum network energy con-
sumption. The only source of energy dissipation is data transmission from CRSNs nodes themselves to the

Scientific Reports | (2022) 12:14039 | https://doi.org/10.1038/s41598-022-18310-9 11

Vol.:(0123456789)
 www.nature.com/scientificreports/

Figure 6.  Performance comparison results in Case 1.

sink, and their short Euclidean distance consumes a small amount of energy. Therefore, nodes can survive
longer, which ensures strong network surveillance capability.
• As for IMOCRP, it is a centralized clustering protocol in which CHs selection is carried out by the sink
instead of normal nodes. Therefore, each CRSNs node should transmit necessary information to the sink
once per round, such as location or residual energy, that is, control overhead of CHs selection is incurred
by information exchange between CRSNs nodes and the sink, not among neighboring CRSNs nodes. In
IMOCRP, with the purpose of minimizing the weighted sum of the average node energy consumption and
the standard deviation of node residual energy, the sink leverages intelligent algorithm to determine CHs and
cluster membership. Therefore, its number of selected CHs is always higher than others except ECMUC. In
addition, no information is exchanged among neighboring nodes for cluster formation and cluster channel
selection, and the total control overhead of IMOCRP is equal to the number of living nodes at the beginning
of that round. Thanks to its low control overhead, the first death node of IMOCRP appears in round 5,411.
However, in Case 2 and Case 3, many CRSNs nodes cannot satisfy the single-hop communication require-
ment to the sink. Therefore, they cannot run IMOCRP effectively, and it is omitted from these two cases.
• Other protocols are distributed in nature, and their CHs selection and cluster formation rely on information
exchange in the neighborhood. In CogLEACH, each CRSNs node can decide whether itself can become a CH
or not by comparing its CHs weight with a randomly generated number (∈ [0,1]). Therefore, its control over-
head for CHs selection is 0. However, it requires twice information exchange between CHs and normal nodes
for temporary and final request and permission. The total control overhead of CogLEACH is about twice
the number of living nodes at the beginning of each round. The random selection of CHs cannot guarantee
the exact number and positions of optimal CHs, which can be observed from Fig. 6c. The number of CHs of
CogLEACH in each round oscillates heavily. However, the optimal number of clusters used by CogLEACH is
obtained by minimizing the total energy consumption of data transmission which is not quite comprehensive.
This is the reason why their performance is worse than ECMUC. EACRP continuously merges neighboring
clusters according to some predetermined rules until the termination condition is satisfied. A large number
of control packets are exchanged during this process. Therefore, its total control overhead is very high. In
addition, its control overhead is highly related to PUs activity, so dramatic oscillations can be observed from
the total control overhead curve of EACRP in Fig. 6b. LEAUCH is an uneven clustering protocol. Its CHs
selection is done through competition among candidate CHs, and its cluster formation process is the same
as that of CogLEACH. Therefore, its total control overhead is usually smaller than four times the number of
living nodes at the beginning of each round. It should be noted that candidate CHs in LEAUCH are selected

Scientific Reports | (2022) 12:14039 | https://doi.org/10.1038/s41598-022-18310-9 12

Vol:.(1234567890)
www.nature.com/scientificreports/

Figure 7.  Performance comparison results in Case 2.

by inheriting the CHs weight function from CogLEACH, and nodes whose weight is higher than the threshold
can become candidate CHs. However, in Case 2 and Case 3, due to the high threshold defined by LEAUCH,
none available candidate CHs can be picked out, not to mention the final CHs. Therefore, the performance
of LEAUCH is omitted from Case 2 and Case 3. WCM and NSAC also require massive information exchange
for CHs selection and cluster formation, so their number of living nodes reduces rapidly.

Apart from above performance metrics, we also care about the information collection capability each proto-
col can provide. This capability can be reflected by the ratio of effective data gathering nodes, and higher ratio
means stronger capability. The comparison results among these protocols are shown in Fig. 6d. From it we can
see that ECMUC can achieve the highest ratio of effective data gathering nodes. Each CRSNs node becomes
a separate CH in ECMUC and it transmits data directly to the sink. The communication channel is randomly
chosen from its available channel list, and the random channel selection decreases collision with PUs. Of course,
channel reclaim from PUs is inevitable in ECMUC, as it only considers about the channel occupancy state at
sensing time for simplicity purpose. If prediction of channel availability is included, the ratio of effective data
gathering nodes can be further improved. As for other protocols, their curves of ratio of effective data gather-
ing nodes oscillate heavily which depend on PUs activity. We can see from Fig. 6c that the number of clusters
of these protocols is much fewer than that of ECMUC, and this means more CMs in each cluster. As common
channels are required by each cluster, more CMs mean higher probability of collision with PUs, which will result
in data transmission failures.

Performance analysis in Case 2 and Case 3. Case 2 and Case 3 are multi-hop communication scenarios
which require inter-cluster relay. Performance comparison results in these two cases are shown in Figs. 7 and 8,
respectively.
From Figs. 7 and 8, we can obtain the following observations:

• For ECMUC protocol, the time when the first death node appears is much later than EACRP, NSAC and
WCM, but it is earlier than CogLEACH. The reasons are explained as follows: ECMUC is a multi-hop clus-
tering routing protocol, and CRSNs nodes in it should consume extra energy to search for available routing
paths and relay inter-cluster data packets. Therefore, compared with CogLEACH, the energy consumption
of data packet transmission in ECMUC is higher, but thanks to its low energy consumption of control infor-

Scientific Reports | (2022) 12:14039 | https://doi.org/10.1038/s41598-022-18310-9 13

Vol.:(0123456789)
 www.nature.com/scientificreports/

Figure 8.  Performance comparison results in Case 3.

mation exchange, its node death rate is slower after the first death node appears. From Figs. 7b and 8b, we
can see that the total control overhead of ECMUC is heavier than CogLEACH, which means the number of
control packets exchanged in ECMUC is more than that in CogLEACH. However, in ECMUC, the control
information for CHs selection and cluster formation are both exchanged within the cluster radius which is
much smaller than the maximum node transmission range Rt, which helps reduce its energy consumed in
control information exchange. Random CHs selection of CogLEACH results in the fluctuating number of
CHs and their random positions, this may aggravate depletion of node energy. However, only CHs which
can reach the sink through single-hop communication can successfully deliver their data, and this can help
conserve energy to a certain extent. As for other protocols, especially EACRP, higher total control overhead
of CHs selection and cluster formation together with routing overhead leads to faster node death. Its number
of living nodes deteriorates rapidly.
• Similar to Case 1, the number of selected CHs in ECMUC is still much higher than its competitors, and it
decreases as network operation goes on. In ECMUC, CRSNs nodes with the sink in their single-hop com-
munication range automatically become CHs, which means each living CRSNs node in ring 1 is an inde-
pendent CH. CHs competition in other rings is performed within cluster radius, which restricts the cluster
size. Therefore, many small clusters are formed in ECMUC, which is beneficial for conserving energy. In
EACRP, neighboring clusters can be merged only if any two nodes in these clusters are within each other’s
transmission range. This illustrates why more clusters are formed in EACRP than in CogLEACH and NSAC
under the same network conditions.
• As shown in Figs. 7d and 8d, the ratio of effective data gathering nodes of ECMUC is much higher than other
competing protocols. As a multi-hop clustering routing protocol, ECMUC relies on effective intra-cluster
data collection and inter-cluster relay to forward data packets towards the sink. It determines reasonable
routing paths for packet delivery. Therefore, most data packets can reach the sink successfully. Similarly, living
CRSNs nodes in EACRP can transmit their data successfully to the sink through the multi-hop forwarding
paths which are found during route selection process, and this explains why EACRP can achieve high ratio
of effective data gathering nodes at the beginning of network operation. However, fast energy exhaustion
leads to poor network surveillance capability which cannot satisfy requirements of practical applications.
CogLEACH, WCM and NSAC can only achieve low ratio of effective data gathering nodes. The reason is
that they are all designed for scenarios which require single-hop communication between CHs and the sink,
so they consider nothing about inter-cluster routing. Due to the limited node transmission range and long

Scientific Reports | (2022) 12:14039 | https://doi.org/10.1038/s41598-022-18310-9 14

Vol:.(1234567890)
www.nature.com/scientificreports/

distance between CRSNs nodes and the sink in Case 2 and Case 3, data packets from many nodes cannot
reach the sink at all.
• By comparing Figs. 6 with 7 and 8, we can observe the impact of number of nodes on network performance.
In general, when the node density ρ is kept unchanged, more nodes will lead to shorter network lifespan,
higher total control overhead, higher number of selected CHs in each round and a little lower ratio of effective
data gathering nodes. In order to compete for CHs in the neighborhood and form clusters, each CRSNs node
j (L(j) ≠ 1) needs to broadcast or unicast related information, that is, the total control overhead is positively
proportional to the number of living nodes. In addition, more nodes in outer rings will increase the relay
burden of nodes in inner rings. Even though the energy consumption of control information exchange can
be effectively controlled by reasonably setting the cluster radius, heavier inter-cluster data relay will accelerate
the energy exhaustion of nodes in inner rings. In this case, network lifetime will be shortened.

As demonstrated above, in Case 1, ECMUC can achieve the best performance among all competing cluster-
ing protocols. All living CRSNs nodes automatically become independent CHs and transmit data directly to the
sink. No control information is exchanged among CRSNs nodes, which results in the minimum network energy
consumption. In Case 2 and Case 3, ECMUC is still superior to other competing protocols in terms of the ratio
of effective data gathering nodes, which means ECMUC can guarantee strong network surveillance capability.
However, its number of living nodes metric is inferior to CogLEACH, and this is the cost paid for route selection
and multi-hop inter-cluster data relay. However, there are much more effective data gathering nodes in ECMUC
during long-term network operation. In other words, ECMUC can provide powerful surveillance capability while
compromising the lifespan of a portion of nodes.

Conclusions
Previous clustering protocols for CRSNs only consider about minimizing the total energy consumption of data
communication and omit the impact of control overhead. Aiming at conquering this limitation and solving the
multi-hop inter-cluster routing problem in large-scale CRSNs, an uneven clustering routing protocol ECMUC
is proposed in this paper. ECMUC combines energy consumption of data communication and control informa-
tion exchange together for the first time to form the first part of the clustering objective. Then minimizing the
network energy consumption and balancing the residual energy among nodes in different rings are innovatively
transformed into minimizing the weighted sum of the energy consumption of each ring and the additional
energy consumption introduced to the whole network by it. Cluster radiuses of different rings are theoretically
derived, and they control the cluster size as well as the corresponding energy consumption of control information
exchange. ECMUC applies varied CHs selection and cluster formation strategies to different rings, and it selects
nodes with the greatest energy and spectral potential as CHs. Through massive well-designed simulations, we
demonstrate that control overhead has a large impact on energy consumption, and taking it into consideration
is beneficial for improving network performance. In addition, ECMUC is superior to most current clustering
protocols in terms of network surveillance capability and network lifetime. In the future, we will leverage energy
harvesting and simultaneous wireless information and power transfer technology to further prolong network
lifespan. In addition, we will research on how to determine the cluster radius automatically according to specific
network configurations.

Data availability
The datasets generated during and/or analysed during the current study are available from the corresponding
author on reasonable request.

Received: 26 April 2022; Accepted: 9 August 2022

References
1. Wang, J. H. & Li, S. ECE: A novel performance evaluation metric for clustering protocols in cognitive radio sensor networks. IEEE
Internet Things J. 8(3), 2078–2079 (2021).
2. Liu, Z. X., Zhao, M. Y., Yuan, Y. Z. & Guan, X. P. Subchannel and resource allocation in cognitive radio sensor network with wire-
less energy harvesting. Comput. Netw. 167, 1–10 (2020).
3. Prajapat, R., Yadav, R. N. & Misra, R. Energy efficient k-hop clustering in cognitive radio sensor network for Internet of Things.
IEEE Internet Things J. 8(17), 13593–13607 (2021).
4. Ren, Q. & Yao, G. S. An energy-efficient cluster head selection scheme for energy-harvesting wireless sensor networks. Sensors.
20(1), 1–17 (2020).
5. Afsar, M. M. & Younis, M. A load-balanced cross-layer design for energy-harvesting sensor networks. J. Netw. Comput. Appl. 145,
1–19 (2019).
6. Shahraki, A., Taherkordi, A., Haugen, O. & Eliassen, F. A survey and future directions on clustering: from WSNs to IoT and modern
networking paradigms. IEEE Trans. Netw. Service Manag. 18(2), 2242–2274 (2021).
7. Wang, J. H. & Shi, W. X. Survey on cluster-based routing protocols for cognitive radio sensor networks. J. Commun. 39(11), 156–169
(2018).
8. Ozger, M. & Akan, O. B. Event-driven spectrum-aware clustering in cognitive radio sensor networks. In 2013 Proceedings IEEE
INFOCOM. 1483–1491 (2013).
9. Ozger, M., Fadel, E. & Akan, O. B. Event-to-sink spectrum-aware clustering in mobile cognitive radio sensor networks. IEEE Trans.
Mob. Comput. 15(9), 2221–2233 (2016).
10. Tabassum, M. et al. An energy aware event-driven routing protocol for cognitive radio sensor networks. Wirel. Netw. 22(5),
1523–1536 (2016).
11. Huang, L. Y. et al. Stochastic event-based sensor schedules for remote state estimation in cognitive radio sensor networks. IEEE
Trans. Automat. Contr. 66(5), 2407–2414 (2021).

Scientific Reports | (2022) 12:14039 | https://doi.org/10.1038/s41598-022-18310-9 15

Vol.:(0123456789)
 www.nature.com/scientificreports/

12. Akrem, L. A. & Qiu, D. Y. Energy efficient spectrum aware clustering for cognitive sensor networks: CogLEACH-C. In 2015 10th
International Conference on Communications and Networking in China. 515–520 (2015).
13. Kim, S. S., Mcloone, S., Byeon, J. H., Lee, S. & Liu, H. B. Cognitively inspired artificial bee colony clustering for cognitive wireless
sensor networks. Cognit. Comput. 9(2), 207–224 (2017).
14. Wang, J. H., Li, S. & Ge, Y. Y. Ions motion optimization-based clustering routing protocol for cognitive radio sensor network. IEEE
Access. 8, 187766–187782 (2020).
15. Eletreby, R. M., Elsayed, H. M. & Khairy, M. M. CogLEACH: A spectrum aware clustering protocol for cognitive radio sensor
networks. In 2014 9th International Conference on Cognitive Radio Oriented Wireless Networks and Communications. 179–184
(2014).
16. Zheng, M., Chen, S., Liang, W. & Song, M. NSAC: A novel clustering protocol in cognitive radio sensor networks for Internet of
Things. IEEE Internet Things J. 6(3), 5864–5865 (2019).
17. Yadav, R. N., Misra, R. & Saini, D. Energy aware cluster based routing protocol over distributed cognitive radio sensor network.
Comput. Commun. 129, 54–66 (2018).
18. Stephan, T., Al-Turjman, F., Suresh Joseph, K., Balusamy, B. & Srivastava, S. Artificial intelligence inspired energy and spectrum
aware cluster based routing protocol for cognitive radio sensor networks. J. Parallel Distrib. Comput. 142, 90–105 (2020).
19. Heinzelman, W. B., Chandrakasan, A. P. & Balakrishnan, H. An application-specific protocol architecture for wireless microsensor
networks. IEEE Trans. Wirel. Commun. 1(4), 660–670 (2002).
20. Wang, T. J., Guan, X. J., Wan, X. L., Shen, H. & Zhu, X. M. A spectrum-aware clustering algorithm based on weighted clustering
metric in cognitive radio sensor networks. IEEE Access. 7, 109555–109565 (2019).
21. Pei, E. R., Han, H. Z., Sun, Z. H., Shen, B. & Zhang, T. Q. LEAUCH: low-energy adaptive uneven clustering hierarchy for cognitive
radio sensor network. Eurasip J. Wirel. Comm. Netw. 2015(1), 1–8 (2015).
22. Zhang, M. C., Zheng, R. J., Li, Y., Wu, Q. T. & Song, L. R-bUCRP: a novel reputation-based uneven clustering routing protocol for
cognitive wireless sensor networks. J. Sensors. 2016, 1–9 (2016).
23. Wang, C. Q. & Wang, S. B. Research on uneven clustering APTEEN in CWSN based on ant colony algorithm. IEEE Access. 7,
163654–163664 (2019).
24. Stephan, T., Al-Turjman, F. & Balusamy, B. Energy and spectrum aware unequal clustering with deep learning based primary user
classification in cognitive radio sensor networks. Int. J. Mach. Learn. Cybern. 12(11), 3261–3294 (2021).
25. Yang, L., Lu, Y. Z., Zhong, Y. C., Wu, X. G. & Yang, S. X. A multi-hop energy neutral clustering algorithm for maximizing network
information gathering in energy harvesting wireless sensor networks. Sensors. 16(1), 1–22 (2015).
26. Wang, J. H. & Ge, Y. Y. A radio frequency energy harvesting-based multihop clustering routing protocol for cognitive radio sensor
networks. IEEE Sens. J. 22(7), 7142–7156 (2022).
27. Soro, S. & Heinzelman, W. B. Prolonging the lifetime of wireless sensor networks via unequal clustering. In 19th IEEE International
Parallel and Distributed Processing Symposium. 1–8 (2015).
28. Zheng, M., Wang, C. Q., Song, M., Liang, W. & Yu, H. B. SACR: a stability-aware cluster-based routing protocol for cognitive radio
sensor networks. IEEE Sensors J. 21(15), 17350–17359 (2021).
29. Zhang, H. Z., Zhang, Z. Y., Dai, H. Y., Yin, R. & Chen, X. M. Distributed spectrum-aware clustering in cognitive radio sensor
networks. In 2011 IEEE Global Telecommunications Conference. 1–6 (2011).
30. Shah, G. A. & Akan, O. B. Spectrum-aware cluster-based routing for cognitive radio sensor networks. In 2013 IEEE International
Conference on Communications. 2885–2889 (2013).
31. Peng, S., Wang, T. & Low, C. P. Energy neutral clustering for energy harvesting wireless sensors networks. Ad Hoc Netw. 28, 1–16
(2015).

Acknowledgements
This work was supported in part by the National Natural Science Foundation of China under Grant 61901102.

Author contributions
Conceptualization, J.W.; Methodology, J.W. and C.L.; Software, C.L.; Writing-original draft presentation, J.W.;
Writing-review & editing, J.W. and C.L.

Competing interests
The authors declare no competing interests.

Additional information
Correspondence and requests for materials should be addressed to J.W.
Reprints and permissions information is available at www.nature.com/reprints.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International
License, which permits use, sharing, adaptation, distribution and reproduction in any medium or
format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the
Creative Commons licence, and indicate if changes were made. The images or other third party material in this
article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the
material. If material is not included in the article’s Creative Commons licence and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder. To view a copy of this licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/.

© The Author(s) 2022

Scientific Reports | (2022) 12:14039 | https://doi.org/10.1038/s41598-022-18310-9 16

Vol:.(1234567890)