SPECIAL SECTION ON HUMAN-DRIVEN EDGE COMPUTING (HEC)

Received October 16, 2020, accepted November 3, 2020, date of publication November 16, 2020,
date of current version December 10, 2020.
Digital Object Identifier 10.1109/ACCESS.2020.3037707

Energy-Efficient Cooperative Spectrum Sensing

Strategy for Cognitive Wireless Sensor Networks
Based on Particle Swarm Optimization
YONGCUN CAO AND HAICHUAN PAN
School of Information Engineering, Minzu University of China, Beijing 100081, China
Corresponding author: Yongcun Cao ([email protected])
This work was supported by the Sub-Project of National Natural Science Foundation of China under Grant 60322507.

ABSTRACT Cognitive wireless sensor networks (CWSNs) can use the idle authorized frequency band
to solve the problem of spectrum resource shortage in traditional wireless sensor network. By employing
spectrum hole in the authorized frequency band, the spectrum sensing technology can degrade the coexistent
interference and enhance the performance of whole sensor network. Due to the characteristics of limited
battery energy and low processing capacity with sensor nodes, it is necessary to enhance the energy
efficiency while improving spectrum sensing performance. In this paper, a cooperative spectrum sensing
strategy for CWSNs based on particle swarm optimization is proposed. Firstly, the system throughput and
energy consumption are quantitatively analyzed, and the mathematical model related to energy efficiency
is established. Secondly, the particle swarm optimization (PSO) algorithm is used to obtain the optimal
selected nodes set under the limited conditions of false alarm probability and detection probability. To avoid
local optimization in the process of problem solving, Cauchy mutation method is introduced to optimize
the parameter selection of fitness function. The experimental results illustrate that our proposed method
can improve the throughput of the system while ensuring the sensing performance, and achieve the energy
efficiency effectively.

INDEX TERMS Energy-efficiency, cognitive wireless sensor networks, cooperative spectrum sensing,
particle swarm optimization.

I. INTRODUCTION multiple sensing nodes [4]. By taking advantage of dense sen-

As the most potential solution to overcome the spectrum sor nodes and obtaining more accurate signal measurements,
resource shortage, cognitive radio technology has attracted CWSNs also can organize multiple nodes to obtain better
a lot of attention. In cognitive radio networks, spectrum detection performance and overcome the problem of hidden
resources are allocated to the primary users (PUs), and terminal [5], [38]. However, CSS will generate more energy
the secondary users (SUs) have to access the spectrum consumption and sensing overhead. On the one hand, more
opportunistically by detecting the PU’s signal correctly and sensor nodes participating in CSS will result in additional
avoiding interference to the authorized user’s communica- energy consumption. In addition, the reporting time slot of
tion [1], [2]. The effectiveness and reliability of spectrum sensing results in the data frame will be greatly extended,
detection has become a key issue in cognitive radio networks. which will influences the throughput and energy efficiency
The performance of channel detection is affected by many of the system.
factors, such as the uncertainty of noise, multipath fading, From the perspective of system sensing performance,
shadow fading and the uncertainty of signal receiver [3]. plenty of cooperating SUs can provide the diversity with
In order to overcome those unfavorable factors, cooperative more signal measurements and thus obtain better detection
spectrum sensing (CSS) is introduced and regarded as an performance [6], [39]. However, the energy consumption
effective measure to improve the detection accuracy and also shows a linear increase with the number of cooperative
sensing performance by exploiting the spatial diversity of nodes. Especially for power constrained sensor nodes, plenty
of energy consumption will be includes mainly in terms
The associate editor coordinating the review of this manuscript and of the authorized user’s signal detection and sensing data
approving it for publication was Rongbo Zhu . reporting [7]. In addition, in correlated log normal shadow
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.
VOLUME 8, 2020 For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/ 214707
 Y. Cao, H. Pan: Energy-Efficient CSS Strategy for CWSNs Based on PSO

channel, the correlation between sensing nodes will seriously the clusters [14]. In [15], a novel cluster-based CSS mech-
affect the detection results of the cooperative nodes set. The anism is proposed for CWSNs, which schedules the sen-
correlation between adjacent nodes will cause the redun- sor nodes into awake or sleep modes for energy saving.
dancy of the sensing results owing to the less useful infor- To improve the detection probability, a weighted CSS scheme
mation, and the increase of the number of cooperative SUs is proposed to assign different weights according to the
does not enhance the detection accuracy significantly [8]. signal-to-noise ratio of SUs [16]. However, due to lack of
Therefore, the issues about the sensor nodes selection to consideration of energy efficiency, it results in high energy
meet the performance constraints of the system on false consumption of the system when a large number of SUs coop-
alarm probability and missing detection probability should be erate in spectrum sensing. According to the local sensing data
investigated. and the intra-cluster fusion decisions, the intra-cluster and
The remainder of the paper is organized as follows. inter-cluster rules are combined for CSS to reduce the num-
In Section II, the related works are reviewed and the major ber of reports from the cluster member nodes [17]. In [18],
contributions of our paper are summarized. In Section III, an iterative algorithm is introduced to determine the optimal
the system model is described and the problem of energy effi- number of cooperative SUs, sensing and transmission time,
ciency is discussed. The proposed optimization algorithm is so as to obtain the maximization of the energy efficiency.
presented in Section IV. In Section V, the simulations results Most of the approaches presented choose all SUs to par-
and analysis are provided. The conclusions are presented in ticipate in cooperative sensing. However, in the actual envi-
Section VI. ronment, due to the influence of geographical location and
node’s distribution, there may be great differences in local
II. RELATED WORK sensing performance of each SU. Some studies have con-
In order to enhance the throughput or spectrum utilization, sidered various node selection schemes to reduce the over-
SUs often operate with high transmission power, which leads head and energy consumption during the process of spectrum
to the reduction of energy efficiency. Therefore, many studies sensing. In [19], the selection of cooperative sensor nodes is
not only examine spectrum sensing in terms of throughput formulated as binary knapsack problem, and a dynamic pro-
and spectrum utilization, but also study MAC protocol in gramming method is employed to resolve the minimization
CWSN from the perspective of energy efficiency. In [9], of the energy consumption. In [20], a joint sensing nodes and
the tradeoff between detection accuracy and energy effi- decision node selection method is introduced, which adopts
ciency is considered, and the optimal MAC frame structure is the convex optimization framework. Due to the NP-complete
designed to improve the energy efficiency. In [10], an energy- property, the problem of sensor selection is handled by map-
efficient spectrum access scheme and optimal sensing order ping the assignment index from an integer to a real number
are designed, in which SUs sequentially senses the channel field. To reduce energy consumption and sensing overhead,
until determining the licensed channel for data transmission. a selection probability scheme is proposed, which exploits
In [11], the transmission time and power for SUs for sensing the historical observations from adjacent sensor nodes and
multiple channels are determined to ensure the energy effi- excludes the nodes with low strength of the received sig-
ciency under the condition of the existence of the interference nals [21]. In [22], an energy-efficient CSS is proposed to
to PU. In [12], it is assumed that PU does not return and acquire the minimization of the energy consumption among
occupy the channel during the process of SU’s transmission, the distributed sensor nodes under the constraints of global
and the transmission time can be defined as a function of sens- detection probability and false alarm probability.
ing time. Then, the sensing time slot and transmission time Due to the hardware limitation and energy constraint of
slot are jointly optimized to enhance the energy efficiency of sensor nodes in CWSNs, it is important to make a trade-off
the system. In [13], the optimization problem of sensing time between detection accuracy and energy efficiency [23]. The
and transmission duration are discussed and a sub-optimal sensing performance will be enhanced with the number of
algorithm is proposed to achieve the maximization of the SUs, meanwhile the energy consumption in the transmission
network energy efficiency of the whole system and satisfy phase also increases. Therefore, it is very of significant to
constraint conditions of limited interference to the authorized study how to find the optimal number of cooperative sensor
signal. nodes to minimize energy consumption and provide reli-
Some researches focus on the specific topology or able detection results and transmission quality. Generally,
coordination among SUs, for instance, clustered-based col- the major contributions of our paper can be summarized as
laborative spectrum sensing. By grouping neighboring nodes follows:
logically, spectrum-aware clustering is regarded as one of i) The system throughput and energy consumption are
the most promising schemes to address the energy consump- quantitatively analyzed, and the mathematical model
tion problem. Focusing on increasing energy efficiency and related to energy efficiency is established.
prolonging the network lifespan, a cluster head selection ii) The PSO algorithm is used to solve the problem, which
algorithm base on fuzzy c-means (FCM) clustering is pro- makes the cooperative sensor nodes be selected opti-
posed according to several factors, which takes advantage of mally under the limited conditions of global false alarm
sensor node’s spatial diversity and residual energy to organize probability and detection probability.

214708 VOLUME 8, 2020

Y. Cao, H. Pan: Energy-Efficient CSS Strategy for CWSNs Based on PSO

iii) To avoid local optimization in the process of prob- probability can be given by
lem solving, Cauchy mutation method is introduced to N
optimize the parameter selection of fitness function.
Y

Pd = 1 − 1 − pd,i (4)
III. SYSTEM MODEL i=1
A. COOPERATIVE SPECTRUM SENSING YN


There are N sensor nodes deployed in a CWSN with a specific Pf = 1 − 1 − pf ,i (5)
fusion center (FC). The cognitive sensor node can sense the i=1

licensed channel periodically and conduct local decisions on Although CSS has obvious advantages in PU’s signal
the existence of the primary user according to its own obser- detection, the sensing performance is still affected by many
vations. To avoid any interference to primary users, secondary factors such as related shadows. Some researches show that
users should keep silent during detection process. The local with the increase of the number of sensing nodes in the fixed
decisions of all sensing nodes will be sent to FC continuously area, the performance of the sensor network will decline due
in the time slot based on TDMA scheme. Comparatively, to the stronger correlation between adjacent users, especially
the hard decision method needs to transmit one-bit decision in the correlated lognormal shadow [26]. In addition, coop-
to the FC, which can save more energy consumption than erative sensing has a great impact on resource consumption,
soft fusion. Thereupon, the FC employs hard decision fusion such as bandwidth consumption of control channel and trans-
instead of soft fusion to make the final decision. mission energy consumption of sensing reports. The energy
Assuming each node is independent in the sensing process consumption usually increases linearly with the number of
and makes local decisions by employing energy detection nodes participating in the cooperation [27]. Therefore, it is
method. For a given received signal, the detection of the more advantageous to select some uncorrelated cognitive
primary user can be formulated as a statistical problem [25], sensor nodes to incorporate during the sensing process under
and H1 and H0 denotes the hypothesis that the PU exists or certain constraints of sensing accuracy. On the one hand,
does not exist respectively. Therefore, the received signal by it can improve the robustness of decision results. On the other
i-th sensor node can be expressed by hand, it can help to reduce the cost of collaborative awareness.
To enhance the energy-efficiency during the process of
(
ui (k), H0
xi (k) = (1) CSS, sensing nodes should be selective to participate in spec-
hi (k)s(k) + ui (k), H1
trum sensing while maintaining the performance constraints.
where s(k) represents the primary user’s signal, and the noise Therefore, the global detection probability and false alarm
sample ui (k) can be assumed to be a cyclic symmetric Gaus- probability will be written as
sian random vector with zero mean and variance σn2 . Besides, YN
1 − ρi pd,i


hi (k) is the channel gain between the i-th sensor node and the P̃d = 1 − (6)
primary user with the mean value 0 and the variance δi . i=1
For its simplicity, the energy detection method will be N
Y
1 − ρi pf ,i


employed for signal’s detection. For the i-th node, its energy P̃f = 1 − (7)
M i=1
statistics can be simply expressed as Ei = M1 |xi (m)|2 .
P
k=1 where ρi denotes the assignment index with the value 0
If the test statistics of the sensing node is greater than the or 1. Among them, 0 represents that the corresponding sensor
energy threshold, the presence of the primary user signal node is selected for spectrum sensing, and 1 indicates that the
will be made. Otherwise, the decision of the licensed chan- sensor is not involved in cooperation.
nel will be idle. According to the above decision rules and
B. ENERGY EFFICIENCY ANALYSIS
given threshold η, the false alarm probability and detection
probability can be given by Next, the energy efficiency of the system is analyzed mathe-
matically. In order to discover and utilize the idle authorized
η √
  
pf ,i = Q − 1 M (2) channel in time and reduce the interference to the primary
σn2 user, the SUs usually adopt the periodic frame structure
s !
η as shown in Fig. 1. The frame length of cognitive user

M
pd,i = Q − γi − 1 (3) can be fixed to T , including spectrum sensing time, spec-
σn2 1 + 2γi
trum and data transmission time. The spectrum sensing time
where γi represents the received signal-to-noise ratio of can be divided into two parts: local sensing and decision
the PU’s signal at i-th sensing node. Besides, Q(·) is result reporting. The time intervals assigned are Ts and KTr
a Gaussian tail function and is defined as: Q(t) = N
ρi denotes the number of SUs
 P
√1
R∞
exp − x2
dx. respectively, where K =
2π t 2 i=1
Since multipath fading and shadow fading will impact on participating in cooperative sensing. In order to ensure certain
the degradation of sensing performance of a single node, detection accuracy, the local sensing time is usually com-
the CSS can overcome this phenomenon. By using OR com- posed of M sensing time slots. In the data transmission stage,
bination rule, the global detection probability and false alarm cognitive sensor nodes decide whether to send data frames

VOLUME 8, 2020 214709

 Y. Cao, H. Pan: Energy-Efficient CSS Strategy for CWSNs Based on PSO

result to the FC through the control channel, ETs is the power

generated by the SU occupying the PU’s channel for data
transmission, and ETp is the power generated by the PU for
data transmission in the licensed channel. Then, the energy
consumption will be discussed in the following four cases:
Case 1: The PU does not transmit data in the channel, and
the SUs detect that the PU is idle. Within a frame length,
the energy consumption of the network can be calculated by
40,0 (T , K ) = K (Ts Es +Tr Er )+(T −Ts −KTr ) ETs (11)
FIGURE 1. The frame structure for CWSNs with spectrum sensing.  
with the probability p (H0 |H0 ) = p (H0 ) 1 − P̃f .
according to the spectrum sensing results, i. e., if the channel Case 2: The PU is idle but a false alarm occurs. The SUs
is decided to be busy, the sensor nodes will not send the data, detect that the PU is busy and the FC make a false spectrum
and vice versa. decision. In this case, the energy consumption of the network
When the FC decides that the PU exists, the SUs will can be calculated by
silently wait for the start of the next frame for the next
round. If the primary user channel is idle, the FC allocates the 41,0 (T , K ) = KTs Es + KTr Er (12)
channel to the sensor nodes for data transmission. In CWSNs, with the probability p (H1 |H0 ) = p (H0 ) P̃f
the throughput of the system can be composed of the PU’s Case 3: The PU transmits data in the channel, but the
throughput and the throughput from all sensor nodes. When sensing results from sensor nodes indicate that the PU is
the Bayesian risk is constant and the interference rate is below absent. In this case, both PU and the SUs will dissipate the
the threshold, the throughput per unit time of SUs in the energy consumption, which can be expressed by
channel can be expressed as follows: 40,1 (T , K ) = KTs Es +KTr Er +(T −Ts −KTr ) ETs +TETp
  
T − Ts − KTr  
 9s,0 (T , K ) = Cs 1 − P̃f P(H0 ) (13)


T  
with the probability p (H0 |H1 ) = p (H1 ) 1 − P̃d
 
T − Ts − KTr  
 9s,1 (T , K ) = Cs 1 − P̃d P(H1 )


T Case 4: The SUs can successfully detect the PU’s signal,
(8) and FC will notify all sensor nodes not to occupy the licensed
channel. In this case, the energy consumption of the network
where Cs denotes the channel capacity that can be utilized
can be calculated by
under the transmitted signal power of the secondary user.
Besides, P(H0 ) and P(H1 ) represent the probability that the 41,1 (T , K ) = KTs Es + KTr Er + TETp (14)
primary user does not exist in the authorized channel and the
with the probability p (H1 |H1 ) = p (H1 ) P̃d
probability that the PU exists, respectively.
Therefore, we can obtain the average energy consumption
Hence, the PU’s throughput per unit time can be calculated
per frame as
by X
4 (T , K ) = 4i,j (T , K ) P Hi |Hj


9P,0 (T , K ) = CP P̃d P(H1 ) (15)
(
i∈{0,1},j∈{0,1}
(9)
9P,1 (T , K ) = CP 1 − P̃d P(H1 ) The energy efficiency can be expressed as follows:
where CP denotes the channel capacity that can be utilized 9 (T , K , φ)
EE = T × (16)
under the transmitted signal power of the PU. 4 (T , K )
When the FC makes the prediction errors in channel The FC needs to determine the sensing duration which can
estimation according to the reports from the SUs, the PU maximize energy efficiency and the number of sensor nodes
and the SUs may transmit data in the channel at the same participating in CSS. Hence, the optimization problem can be
time. It will cause interference between them and be almost formulated as
impossible to complete normal communication. Evidently,
the throughput generated by the system in the case of self maxτ ≥0,L<N {EE}
interference can be ignored. Since the detection probability s.t. P̃d ≥ α, P̃f ≤ β
of the system is known, based on the sensing time and the KTr + Ts ≤ T . (17)
number of sensing nodes, we can obtain the energy detection
Since P̃f is not dependent on the γi and using Eq. (7),
threshold of each SU. Therefore, the average throughput of
the upper limit for the number of sensing nodes is
network can be expressed as follows:
obtained as: 
9 (T , K , φ) = 9P,0 (T , K ) (1−φ)+φ9s,0 (T , K ) (10)

 
 ln (1 − β) 
K ≤  (18)
 
Suppose that Es is the power of the SU for sensing the 
η
q
M
 
channel, Er is the power when the SU reports the sensing ln 1 − Q σ 2 − 1 1+2γi
n

214710 VOLUME 8, 2020

Y. Cao, H. Pan: Energy-Efficient CSS Strategy for CWSNs Based on PSO

Therefore, the above optimization problem can be stated as The probability density function of one-dimensional
follows: Cauchy distribution is defined as
maxτ ≥0,L<N {EE}; θ
f (x) = (21)
π x +θ


N 2
Y
1 − ρi pd,i ≤ 1 − α,


s.t. where θ is the coefficient and θ > 0, and the distribution
i=1
  function of one-dimensional Cauchy distribution is
π
N
 
X  ln (1 − α) 
F (x) =
1
ρi =  arctan x + (22)
 
π 2
ln 1 − Q ση2 − 1
  q  
M
i=1 1+2γi
n when θ = 1, the above function obeys the standard Cauchy
KTr + Ts ≤ T . (19) distribution. In order to improve the convergence speed of
PSO and make the fitness value jump out of the local optimum
IV. PROPOSED METHOD quickly, the variable Cauchy mutation factor is employed.
PSO is an intelligent algorithm which imitates the behavior In the early and middle period of the PSO algorithm, the large
of birds, which was jointly proposed by Kennedy and Eber- value of θ will be generated to avoid falling into local opti-
hart [28] in 1995. Due to its simplicity and easy implemen- mum. While in the later stage of the optimization resolution,
tation, we apply it for above optimization problem for sensor the value of θ should be small to improve the convergence
node’s selection. By initializing a random group of particles, speed. Thus, the Cauchy variation factor can be defined as:
each particle represents a feasible solution to the problem. tmax − t
To seek the optimal solution, each particle moves in the λ= × V̂j (23)
t
direction to its historical best position and the global best
position [29]. where V̂j denotes the average velocity of swarm.
The PSO model includes a D-dimensional search space and After introducing Cauchy variation factor, the position and
m particle nodes. The whole particle swarm is represented by velocity of the i-th particle can be updated by
X = {x1 , x2 , · · · , xm }, and xi = (xi1 , xi2 , · · · , xid ) denotes (
the D-dimensional position vector of the i-th particle. Accord- vid (t + 1) = vid (t + 1) + λC (0, 1)
(24)
ing to the actual optimization problem, the fitness function is xid (t + 1) = xid (t) + vid (t + 1)
specified as the energy efficiency. The current fitness value
where C (0, 1) is the random number generated by the
will be calculated, and then compared with the current parti-
standard Cauchy distribution function.
cle position and determine whether the particle’s position is
Furthermore, during the search process, the smaller inertia
optimal or not. vi = (vi1 , vi2 , · · · , vid ) represents the velocity
weight can make the PSO gain faster convergence speed. But
of i-th particle. Besides, pi = (pi1 , pi2 , · · · , pid ) indicates the
out of the range of global position, the global optimal solution
current best position, and pg = pg1 , pg2 , · · · , pgd shows the


may not be obtained [30], [31]. The larger inertia weight
best position by the whole swarm. At iteration t, the particle
can make the PSO not fall into the local optimum, but the
swarm will update the speed and position according to the
convergence speed is slow. In the early stage, it should be paid
following rule:
 more attention not to fall into the local optimum. But in the
 vid (t + 1) = wvid (t) + c1 r1 (pid (t)
− xid (t))
 later stage, the particles are already near the global optimal
+ c2 r2 pgd (t) − xid (t) (20) position, so the particle convergence speed should be more
important. According to the above analysis, the inertia weight
xid (t + 1) = xid (t) + vid (t + 1)


should be reduced linearly with time and can be updated by
where t is the number of iterations. r1 and r2 is a random wmax − wmin
number to maintain the diversity of the population. As accel- w = wmax − ×t (25)
tmax
eration factors, c1 and c2 represent the ability of particles,
who learn from themselves or groups and approach to the where wmax is the initial weight value and wmin is the final
optimal position. The inertia weight w is used to measure the weight value.
influence of local or global optimal ability in particle swarm The algorithm is described as follows:
optimization. V. PERFORMANCE EVALUATION
The standard particle swarm optimization algorithm is In this section, we conduct the experiments to evaluate the
easy to fall into local optimization and slow convergence performance of the proposed strategy for CWSNs, and com-
speed [32]. By applying the appropriate weighting factor pare with Counting-based Selection [36], CogLEACH [37]
for updating the position, it can reduce the blindness of and Joint selection [38] schemes in terms of average through-
the search process, and the computational efficiency can be put and energy efficiency. In our simulation environment,
effectively improved [33]. Consequently, Cauchy mutation the cognitive sensor nodes change from 10 to 100 in quantity,
will be introduced to accelerate the convergence of standard and the values of the parameters used are listed as: Es =
PSO algorithm. 0.01W, Er = 0.05W,ETp = 5W, ETs = 2.5W, M = 50,

VOLUME 8, 2020 214711

 Y. Cao, H. Pan: Energy-Efficient CSS Strategy for CWSNs Based on PSO

Input: PU, Number of sensor nodes N , tmax .

Output: Selected sensing nodes.
1. Initialize particles xi ∈ X as described in Section IV
2. While t < tmax
3. For i = 1 to N do
4. Calculate the fitness value of xi ;
5. If EE(xi ) > pbesti then
6. xi will be replaced by pbesti ;
7. End if
8. If EE(xi ) > gbest then
9. gbestwill be replaced by xi ;
10. End if
11. The velocity of particle xi is updated
according to Eq. (24);
FIGURE 2. False alarm probability with different sensor nodes.
12. The position of particle xi is updated
according to Eq. (24);
13. End for
14. For j = 1 to N do
15. Obtain the value of V̂j ;
16. Update the Cauchy variation factor
according to Eq. (23);
17. End for
18. End while
19. Working out the optimal selected sensing nodes.
20. End

σ 2 = 1, γ = −20dB, α = 0.9, β = 0.1, tmax = 100,

wmax = 0.9 and wmin = 0.3.
FIGURE 3. Sensing performance gain with different sensor nodes.
Figure 2 shows the curve of the global false alarm proba-
bility with the increase of the number of cooperative sensing
nodes, and Fig. 3 shows the curve of the sensing perfor-
mance gain. The sensing performance gain is defined as
1Pfj = Pf (j − 1) − Pf (j), which indicates the sensing
performance gain caused by adding the j-th cooperative node
to the cooperative sensing node set, and Pf (j) is the false
alarm probability of j cooperative nodes. Since the cognitive
sensor nodes are randomly distributed in the coverage of the
network, the simulation results are obtained by executing
1000 times. It can be seen that with the increase of cooperative
sensor nodes, the false alarm probability shows an obvious
downward trend. When the number of cooperative nodes
increases to a certain extent, the sensing performance gain is
gradually contracted and tends to zero. It also demonstrates
that there is an asymptotic performance lower bound for CSS. FIGURE 4. Missed detection probability with different sensor nodes.
From the experimental results in Fig. 3, it can be found
that when the number of cooperative nodes is greater than Figure 4 shows the missed detection probability with dif-
30, the sensing performance gain is close to zero. From ferent sensor nodes. It demonstrates that with the increase
the perspective of system performance, large number of of the number of nodes, the number of selected cooperative
cooperative sensing nodes can obtain better sensing perfor- sensing nodes increases accordingly as well as improving the
mance. However, with the increase of the number of cooper- sensing performance. Compared with the false alarm proba-
ative nodes, the energy consumption and the traffic increase bility, the missed detection probability can be maintained at
approximately linearly. Therefore, it is necessary to make a relatively low level even when the node density is small.
a trade-off between energy efficiency and detection perfor- The reason is that the optimization algorithm can meet the
mance in accordance with the selection of the number of requirements of sensing performance and select the appro-
cooperative sensing nodes. priate set of sensing nodes. However, it should be noted that

214712 VOLUME 8, 2020

Y. Cao, H. Pan: Energy-Efficient CSS Strategy for CWSNs Based on PSO

FIGURE 5. Number of selected sensing nodes with different sensor nodes. FIGURE 6. Total error probability with different sensor nodes.

the better sensing performance will increase the requirement

of algorithm computation and increase the number of itera-
tions. In addition, too many cooperative nodes may increase
the energy consumption for the sake of the improvement of
sensing performance, which is not conducive to ensuring the
energy efficiency of the whole system.
Fig. 5 shows the number of selected cooperative nodes
with different sensor nodes. From the experimental results,
we can observe that the number of sensing nodes selected by
optimization algorithm under the constraints increases with
the increase of node’s density, but the obvious growth rate
slowly slows down. The reason is that the relative distance
between the cooperative sensing nodes decreases with the
increase of node’s density, and the sensing performance gain
is also relatively small even if too many sensor nodes par- FIGURE 7. Average network throughput with different sensor nodes.

ticipate in cooperation. It means that the relative distance

between cooperative sensing nodes can impact on the correla-
tion between cooperative sensing nodes, which is conducive
to the improvement of sensing performance of cooperative
sensing node.
Fig. 6 shows the total error probability with different sen-
sor nodes. The total error probability is defined as the sum
of missed detection probability and false alarm probability
[39, 40]. By utilizing particle swarm optimization, our
proposed method can select the optimal sensor nodes to
meet target detection probability and false alarm probability.
It effectively guarantees the sensing data of all cooperative
nodes being combined at the FC. Due to the SUs whose
sensing data is subjected to error, the total error probability in
CogLEACH is degraded with large number of sensor nodes. FIGURE 8. Energy efficiency with different sensor nodes.
In joint selection scheme, the SUs will be selected for coop-
eration depending on the sensing measurement is error-free results reporting. Therefore, by selecting the appropriate
or not, in which the error probability is reduced effectively. cooperative sensing nodes, the optimal ratio of the network
The comparison of the average throughput of our proposed throughput to the energy consumption can be achieved.
method with traditional schemes is shown in Fig. 7. From Fig 8 shows the efficiency of our proposed methods com-
the results, it can be observed that when the density of sen- pared with traditional scheme. It can be observed that with
sor nodes is too high, the more cooperative sensing nodes increasing number of sensor node, the energy efficiency
participate. Although more accurate sensing results can be increases rapidly especially when the node density is small.
obtained, it will lead to the decline of throughput contrarily. It also clearly shows that efficiency of our proposed method
Moreover, more nodes participating in cooperative sensing is higher than other schemes when the number of sensor
will cause more energy consumption in sensing and sensing nodes exceeds 50. Due to increasing number of cooperative

VOLUME 8, 2020 214713

 Y. Cao, H. Pan: Energy-Efficient CSS Strategy for CWSNs Based on PSO

nodes, large amount of energy will be consumed and contrar- [12] Y. Wu and D. H. K. Tsang, ‘‘Energy-efficient spectrum sensing and trans-
ily the throughput will be decreased due to sensing errors. mission for cognitive radio system,’’ IEEE Commun. Lett., vol. 15, no. 5,
pp. 545–547, May 2011.
Our proposed method only chooses limited number of sensor [13] Z. Shi, K. C. Teh, and K. H. Li, ‘‘Energy-efficient joint design of sens-
nodes can degrade the coexistent interference and enhance ing and transmission durations for protection of primary user in cogni-
the performance of whole sensor network. tive radio systems,’’ IEEE Commun. Lett., vol. 17, no. 3, pp. 565–568,
Mar. 2013.
VI. CONCLUSION [14] B. Dost, S. Nasir, and N. Haewoon, ‘‘Fuzzy C-means clustering and energy
efficient cluster head selection for cooperative sensor network,’’ Sensors,
In this paper, a cooperative spectrum sensing strategy for vol. 16, no. 9, pp. 1459–1476, 2016.
cognitive wireless sensor networks based on particle swarm [15] A. Rauniyar and S. Y. Shin, ‘‘A novel energy-efficient clustering based
optimization is proposed to achieve energy-efficient trans- cooperative spectrum sensing for cognitive radio sensor networks,’’ Int.
J. Distrib. Sensor Netw., vol. 2015, Jun. 2015, Art. no. 198456.
mission. Firstly, the system throughput and energy consump- [16] X. Xu, J. Bao, and H. Cao, ‘‘Energy-efficiency-based optimal relay
tion are quantitatively analyzed, and the mathematical model selection scheme with a BER constraint in cooperative cognitive radio
related to energy efficiency is established. Secondly, the par- networks,’’ IEEE Trans. Veh. Technol., vol. 65, no. 1, pp. 191–203,
Jan. 2016.
ticle swarm optimization algorithm is used to solve the prob-
[17] T. Li, ‘‘Clustering weighted cognitive radio cooperative detection algo-
lem, which makes the cooperative sensor nodes be selected rithm,’’ Radio Eng., vol. 9, pp. 41–44, 2015.
optimally under the limited conditions of global detection [18] F. A. Awin, E. Abdelraheem, and M. Ahmadi, ‘‘Designing an opti-
probability and false alarm probability. To avoid local opti- mal energy efficient cluster-based spectrum sensing for cognitive
radio networks,’’ IEEE Commun. Lett., vol. 20, no. 9, pp. 1884–1887,
mization in the process of problem solving, Cauchy mutation Sep. 2016.
method is introduced to optimize the parameter selection [19] N. U. Hasan, H. S. Kim, W. Ejaz, and S. Lee, ‘‘Knapsack-based energy-
of fitness function. The experimental results show that the efficient node selection scheme for cooperative spectrum sensing in cogni-
tive radio sensor networks,’’ IET Commun., vol. 6, no. 17, pp. 2998–3005,
algorithm can improve the throughput of the system while Nov. 2012.
ensuring the sensing performance, and achieve the energy [20] M. Najimi, A. Ebrahimzadeh, S. M. H. Andargoli, and A. Fallahi, ‘‘A novel
efficiency effectively. sensing nodes and decision node selection method for energy efficiency of
cooperative spectrum sensing in cognitive sensor networks,’’ IEEE Sensors
In the future work, we will further improve the network J., vol. 13, no. 5, pp. 1610–1621, May 2013.
energy efficiency by optimizing the frame length, and also [21] T. Cui and K. S. Kwak, ‘‘Cooperative spectrum sensing with adaptive node
consider the multicast throughput for multi-channel wireless selection for cognitive radio networks,’’ Wireless Pers. Commun., vol. 78,
no. 4, pp. 1879–1890, Oct. 2014.
sensor networks.
[22] A. Ebrahimzadeh, M. Najimi, S. M. H. Andargoli, and A. Fallahi, ‘‘Sen-
sor selection and optimal energy detection threshold for efficient coop-
REFERENCES
erative spectrum sensing,’’ IEEE Trans. Veh. Technol., vol. 64, no. 4,
[1] K. Wang, Y. Wang, Y. Sun, S. Guo, and J. Wu, ‘‘Green industrial Internet pp. 1565–1577, Apr. 2015.
of Things architecture: An energy-efficient perspective,’’ IEEE Commun. [23] S. Zhang, S. Wang, H. Zhao, and A. S. Hafid, ‘‘Cross-layer
Mag., vol. 54, no. 12, pp. 48–54, Dec. 2016. aware joint design of sensing and frame durations in cognitive
[2] R. Deng, ‘‘Energy-efficient cooperative spectrum sensing by optimal radio networks,’’ IET Commun., vol. 10, no. 9, pp. 1111–1120,
scheduling in sensor-aided cognitive radio networks,’’ IEEE Trans. Veh. Jun. 2016.
Technol., vol. 61, no. 2, pp. 716–725, Feb. 2012.
[24] D. B. Rawat, T. Amin, and M. Song, ‘‘The impact of secondary user mobil-
[3] M. Zareei, A. K. M. Muzahidul Islam, N. Mansoor, S. Baharun,
ity and primary user activity on spectrum sensing in cognitive vehicular
E. M. Mohamed, and S. Sampei, ‘‘CMCS: A cross-layer mobility-aware
networks,’’ in Proc. IEEE Conf. Comput. Commun. Workshops (INFO-
MAC protocol for cognitive radio sensor networks,’’ EURASIP J. Wireless
COM WKSHPS), Hong Kong, Apr. 2015, pp. 588–593.
Commun. Netw., vol. 2016, no. 1, pp. 48–55, Dec. 2016.
[25] S. Mishra, A. Sahai, and R. Brodersen, ‘‘Cooperative sensing among
[4] C. Singhal and A. Rajesh, ‘‘Review on cross-layer design for cognitive
cognitive radios,’’ in Proc. IEEE Int. Conf. Commun., Jun. 2006,
ad-hoc and sensor network,’’ IET Commun., vol. 14, no. 6, pp. 897–909,
pp. 1658–1663.
Apr. 2020.
[26] J. A. Bazerque and G. B. Giannakis, ‘‘Distributed spectrum sensing for
[5] J. Abolarinwa, N. M. A. Latiff, S. K. S. Yusof, and N. Fisal, ‘‘Energy-
cognitive radio networks by exploiting sparsity,’’ IEEE Trans. Signal Pro-
efficient, learning-inspired channel decision and access technique for cog-
cess., vol. 58, no. 3, pp. 1847–1862, Mar. 2010.
nitive radio-based wireless sensor networks,’’ Int. J. Multimedia Ubiqui-
tous Eng., vol. 10, no. 2, pp. 11–24, Feb. 2015. [27] S. Anjana and S. Nandan, ‘‘Energy-efficient cooperative spectrum sensing:
[6] A. Bradai, K. Singh, A. Rachedi, and T. Ahmed, ‘‘EMCOS: Energy- A review,’’ in Proc. 2nd Int. Conf. Inventive Commun. Comput. Technol.
efficient mechanism for multimedia streaming over cognitive radio sensor (ICICCT), Apr. 2018, pp. 992–996.
networks,’’ Pervasive Mobile Comput., vol. 22, pp. 15–21, Sep. 2015. [28] J. Kennedy and R. Eberhart, ‘‘Particle swarm optimization,’’ in Proc.
[7] O. Ergul and O. B. Akan, ‘‘Energy-efficient cooperative spectrum sensing IEEE Int. Conf. Neural Netw., vol. 4. Perth, WA, Aust, Nov. 1995,
for cognitive radio sensor networks,’’ in Proc. IEEE Symp. Comput. Com- pp. 1942–1948.
mun. (ISCC), Jul. 2013, pp. 210–213. [29] J. Yang, H. Zhang, Y. Ling, C. Pan, and W. Sun, ‘‘Task allocation for wire-
[8] E. Fadel, M. Faheem, V. C. Gungor, L. Nassef, N. Akkari, M. G. A. Malik, less sensor network using modified binary particle swarm optimization,’’
S. Almasri, and I. F. Akyildiz, ‘‘Spectrum-aware bio-inspired routing in IEEE Sensors J., vol. 14, no. 3, pp. 882–892, Mar. 2014.
cognitive radio sensor networks for smart grid applications,’’ Comput. [30] P. Chauhan, K. Deep, and M. Pant, ‘‘Novel inertia weight strategies for par-
Commun., vol. 101, pp. 106–120, Mar. 2017. ticle swarm optimization,’’ Memetic Comput., vol. 5, no. 3, pp. 229–251,
[9] J. Zhang, F.-C. Zheng, X.-Q. Gao, and H.-B. Zhu, ‘‘Sensing-energy effi- Sep. 2013.
ciency tradeoff for cognitive radio networks,’’ IET Commun., vol. 8, no. 18, [31] M. Kanemasa and E. Aiyoshi, ‘‘Algorithm tuners for PSO methods and
pp. 3414–3423, Dec. 2014. genetic programming techniques for learning tuning rules,’’ IEEJ Trans.
[10] Y. Pei, Y.-C. Liang, K. C. Teh, and K. H. Li, ‘‘Energy-efficient design of Electr. Electron. Eng., vol. 9, no. 4, pp. 407–414, Jul. 2014.
sequential channel sensing in cognitive radio networks: Optimal sensing [32] W. Kan and S. Jihong, ‘‘The convergence basis of particle swarm opti-
strategy, power allocation, and sensing order,’’ IEEE J. Sel. Areas Com- mization,’’ in Proc. Int. Conf. Ind. Control Electron. Eng., Aug. 2012,
mun., vol. 29, no. 8, pp. 1648–1659, Sep. 2011. pp. 63–66.
[11] S. Wang, Y. Wang, J. P. Coon, and A. Doufexi, ‘‘Energy-efficient spec- [33] T. Kurihara and K. Jin’no, ‘‘Analysis of convergence property of PSO and
trum sensing and access for cognitive radio networks,’’ IEEE Trans. Veh. its application to nonlinear blind source separation,’’ in Proc. IEEE Congr.
Technol., vol. 61, no. 2, pp. 906–912, Feb. 2012. Evol. Comput., Cancun, Mexico, Jun. 2013, pp. 976–981.

214714 VOLUME 8, 2020

Y. Cao, H. Pan: Energy-Efficient CSS Strategy for CWSNs Based on PSO

[34] G. Li, D. Jiang, Y. Zhou, G. Jiang, J. Kong, and G. Manogaran, ‘‘Human YONGCUN CAO received the B.S. degree in 1986.
lesion detection method based on image information and brain signal,’’ He is currently a Professor with the School of
IEEE Access, vol. 7, pp. 11533–11542, 2019. Information Engineering, Minzu University of
[35] W. Wei, S. Liu, W. Li, and D. Du, ‘‘Fractal intelligent privacy China, Beijing, China. His current research inter-
protection in online social network using attribute-based encryption ests include big data, parallel algorithm, and intel-
schemes,’’ IEEE Trans. Comput. Social Syst., vol. 5, no. 3, pp. 736–747, ligent information processing and systems.
Sep. 2018.
[36] Z. Khan, J. Lehtomaki, K. Umebayashi, and J. Vartiainen, ‘‘On the
selection of the best detection performance sensors for cognitive radio
networks,’’ IEEE Signal Process. Lett., vol. 17, no. 4, pp. 359–362,
Apr. 2010.
[37] R. Eletreby, H. Elsayed, and M. Khairy, ‘‘CogLEACH: A spectrum aware
clustering protocol for cognitive radio sensor networks,’’ in Proc. 9th Int.
Conf. Cognit. Radio Oriented Wireless Netw., Jun. 2014, pp. 179–184. HAICHUAN PAN received the B.S. degree from
[38] Y. Peng, F. Al-Hazemi, H. Kim, and C.-H. Youn, ‘‘Joint selection for the Luoyang Institute of Science and Technol-
cooperative spectrum sensing in wireless sensor networks,’’ IEEE Sensors ogy, China, in 2019. He is currently pursuing
J., vol. 16, no. 22, pp. 7837–7838, Nov. 2016. the master’s degree with the School of Infor-
[39] I. F. Akyildiz, B. F. Lo, and R. Balakrishnan, ‘‘Cooperative spectrum mation Engineering, Minzu University of China,
sensing in cognitive radio networks: A survey,’’ Phys. Commun., vol. 4, Beijing, China. His research interests include
no. 1, pp. 40–62, Mar. 2011. machine learning, image data analysis, and swarm
[40] Z. Huang, X. Xu, J. Ni, H. Zhu, and C. Wang, ‘‘Multimodal representation intelligence.
learning for recommendation in Internet of Things,’’ IEEE Internet Things
J., vol. 6, no. 6, pp. 10675–10685, Dec. 2019.

VOLUME 8, 2020 214715