You are on page 1of 13

1

Abstract In this paper, an energy-efficient scheme is proposed


for cooperative spectrum sensing in cognitive sensor networks. In
our scheme, we introduce a technique to select the sensing nodes
and to set energy detection threshold so that energy saving can be
accomplished in the nodes. Our objective is to minimize the
energy consumed in distributed sensing subject to constraints on
global probability of detection and probability of false alarm by
determining the detection threshold and selection of the sensing
nodes. The energy detector is applied to detect the primary user
activity for the sake of simplicity. At first, it is assumed that the
instantaneous SNR for each node is known. Then, the optimal
conditions are obtained and a closed- form equation is expressed
to determine the priority of nodes for spectrum sensing and also
the optimum detection threshold. This problem is also solved
when the average SNRs of sensors are available according to real
situations. To achieve more energy saving, the problem of joint
sensing node selection, detection threshold and the decision node
selection (DN) is analyzed and an efficient solution is extracted
based on the convex optimization framework. Simulation results
show that the proposed algorithms lead to significant energy
saving in cognitive sensor networks.

I ndex Terms Cognitive wireless sensor networks, detection
probability and false alarm probability, cooperative spectrum
sensing, detection threshold.

I. INTRODUCTION
IVEN the limitations of the natural frequency spectrum,
it becomes obvious that the current static frequency
allocation schemes cannot accommodate the requirements of
an increasing number of higher data rate devices. As a result,
innovative techniques that can offer new ways of exploiting
the available spectrum are needed. Cognitive radio arises to be

Manuscript received
Copyright (c) 2013 IEEE. Personal use of this material is permitted.
However, permission to use this material for any other purposes must be
obtained from the IEEE by sending a request to pubs-permissions@ieee.org.
M. Najimi is with the Faculty of Electrical Engineering, Mazandaran
University of Science & Technology (e-mail:
Maryam_najimi1361@yahoo.com).
A. Ebrahimzadeh and S.M.Hosseini are with the Faculty of Electrical &
Computer Engineering, Babol University of Technology (e-mail:;
e_zadeh@nit.ac.ir; smh_andargoli@nit.ac.ir).
A. Fallahi is with RighTel, Tehran, Iran (e-mail: a.fallahi@rightel.ir).



a tempting solution to the spectral congestion problem by
introducing opportunistic usage of the frequency bands that
are not heavily occupied by licensed users [1]. Hence, one of
the main aspects of cognitive radio is related to autonomously
exploiting locally unused spectrum (spectrum holes) to
provide new paths to spectrum access. In cognitive radio
terminology, primary users can be defined as the users who
have higher priority or legacy rights to the usage of a specific
part of the spectrum. On the other hand, secondary users, who
have lower priority, exploit this spectrum in such a way that
they do not cause interference to primary users. Therefore,
secondary users need to have cognitive radio capabilities, such
as sensing the spectrum reliably to check whether it is being
used by a primary user and changing the radio parameters to
exploit the unused part of the spectrum.
Spectrum sensing has very important role in cognitive radio
networks. On the other hand, the status of the activity of the
primary user is determined by spectrum sensing. In fact,
spectrum sensing is employed to identify the spectrum holes.
However, a secondary user may declare a spectrum hole when
it has been already occupied by a primary user, that is called
missing detection. Alternatively, it may confirm the presence
of a primary transmission when the spectrum is not actually in
use that is referred to as false alarm. The attenuation caused
to the signal due to shadowing, fading or other impairments
can lead to erroneous detection of the primary user signal by a
single detector. It is believed that deploying multiple cognitive
sensors for sensing the spectrum can result in improving the
detection of primary signals [2], [3]. The method and the
algorithms employing this strategy are referred to as
cooperative spectrum sensing.
Cooperative spectrum sensing can be done by cognitive sensor
networks, such that the sensors reports their result to the
Fusion Center (FC) to make a final decision about the activity
of the primary user and therefore the available channels
(spectrum holes) are reported to other cognitive radio
networks. In this case, the sensor network acts as an advisor
network which provides primary user activities information
for cognitive radio networks. In cognitive sensor networks, it
is also possible that the provided information through
cooperative spectrum sensing is used for assigning idle
channels to available sensors in the network. However, this
will result in a trade-off between the sensors throughput and
sensing accuracy. In [4] the authors studied the cooperative
Sensor Selection and Optimal Energy Detection
Threshold for Efficient Cooperative Spectrum
Sensing
Ataollah Ebrahimzadeh, Maryam Najimi, Seyed Mehdi Hosseini Andargoli and Afshin Fallahi
G
This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TVT.2014.2331681
Copyright (c) 2014 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
2
incentives of the secondary users that seek to cooperate with
each other to improve the detection performance.
Various approaches have been proposed to detect the spectrum
holes in cognitive sensor networks. In [5]- [7] an optimal
technique is employed under stationary Gaussian noise,
entitled as matched filter detection. A matched filter is a linear
filter which maximizes the signal-to-noise ratio (SNR) by
correlating the received signal with the original signal
(without noise) and it is used in the optimal matched filter
formulas. But, in this method, the cognitive sensor needs to
have prior information of the primary user signal such as its
bandwidth, modulation type and order, pulse shaping, etc. So,
it is not feasible to develop such a system due to its expense
and complexity. A more general scheme for spectrum sensing
is energy detection [8]. In this technique, the energy received
on a licensed band is measured. Whenever the energy level
exceeds a predefined threshold, it means the primary user is
transmitting; otherwise, a spectrum hole will be confirmed.
This method is relatively simple to implement. Feature
detection is another technique for signal detection. Despite
stationary noise, the primary transmission signals have built in
periodicity and spectral correlation features which are unique
for different signals [9]. Feature detection technique utilizes
this property and is capable of distinguishing communication
signals at the cost of system complexity.
Another issue is that sensor nodes, employed for spectrum
sensing, operate under limited energy budgets. The physical
constraints of sensors and the prohibitive costs to replace the
failure sensors in the cognitive sensor network make energy as
a crucial consideration in designing a sensor network.
In [6], it is shown that increasing the number of cooperative
nodes can decrease the required detector sensitivity and
sensing time. However, the communication overhead will also
increase in terms of exchanged messages and processing
overhead. According to [10], it is not necessary for all
secondary users to cooperate in the network to achieve the
optimum performance. And secondary users with the highest
primary users signal-to-noise ratio (SNR) only participate in
spectrum sensing.
In this paper, energy detector is used for primary user signal
detection due to its simplicity and independency of having
prior knowledge about the primary user signal for channel
sensing. We propose an energy efficient cooperative spectrum
sensing scheme by determining the optimal sensing nodes and
energy detection threshold. The global probability of detection
and global probability of false alarm are the constraints of this
scheme. Hereby, an on/off scheme that is a major energy
efficient technique in [11] is considered. In an on/off
scheme, some sensors turn off their radios, while the other
nodes sense the spectrum.
In [12], minimizing energy consumption with constraints on
the detection performance is proposed. In fact, in [12], all
sensors have the same signal-to-noise ratio (SNR). It is a real
assumption especially when the sensors have approximately
the same distance to the primary user. However, this
assumption applies to simplify the problem. Therefore, the
number of sensing nodes is determined without giving which
nodes are sensing. Accordingly, we assume that the sensors
have different SNRs due to their distances from primary user
and small scale fading characteristics. This makes it more
difficult to solve the sensor selection problem. Therefore, for
minimizing energy, selection of the sensing nodes becomes
very important instead of only determining the number of
sensing node.
In [13], the authors optimize the detection threshold when
energy detection is employed to satisfy the detection
performance in an efficient and implementable way. In [14],
the detection threshold is optimized for minimizing the total
error rate. The analysis focuses on the derivation of a closed-
form expression for the average missed-detection probability
over Rayleigh fading and Nakagami- fading channels. In
[15], the authors get the Closed-form expression to the optimal
threshold of local detection to reach a trade-off between
missed detection probability and false alarm probability. In
[16], a method is proposed to change the threshold based on
the number of samples, to optimize the detection parameters,
namely probability of detection and probability of false alarm.
A control parameter is also introduced which is used to vary
the already-set threshold and, thereby, to obtain a response
more suited to the operational requirements than that obtained
by a fixed threshold based energy detector. In [17], the process
of threshold selection for energy detection is addressed by the
Constant False Alarm Rate (CFAR) method and selection is
carried out with regard to the present conditions of noise
levels. The results show that if they dynamically adjust
detection-threshold based to the noise level present during
detection process, the detection probability will be higher than
the one obtained when a fixed threshold value is considered.
Although, in these papers, SNR is different for the nodes,
energy consumption is not addressed and they just optimize
the detection parameters, namely probability of detection and
probability of false alarm by setting the detection threshold.
The contributions of this paper are as follows
Given that every node has a different SNR and
this instantaneous SNR is known for each node,
we minimize the energy by selecting the sensing
nodes and finding the optimal detection threshold
while the constraints on the global probability of
detection and global probability of false alarm are
satisfied.
Since it is difficult to obtain the
instantaneous SNR for every node, we take into
account the average SNR for each node for the
purpose of the sensor selection and threshold
setting. Then, the average probability of detection
is evaluated for each node. After obtaining the
optimal conditions based on KarushKuhnTucker
(KKT) conditions, an iterative algorithm is
proposed to search the sensing sensors for
spectrum sensing and the optimum detection
threshold.
Decision node (DN) selection is also proposed
so that more energy saving is obtained in
cooperative spectrum sensing. To move forward in
This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TVT.2014.2331681
Copyright (c) 2014 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
3
our problem solving, we address the decision node
selection criteria, its corresponding sensing nodes and
setting the detection threshold.
The remainder of the paper is organized as follows. In
Section II, we present the cognitive sensor network structure
and formulate the global probability of detection and the
global probability of false alarm. In Section III, the problem is
analyzed for the case where the instantaneous SNR is
available for each node. In Section IV, we propose an
algorithm to solve the problem and also find the optimum
sensing nodes and detection threshold. In Section V, the
analysis is provided for the case that the average SNR for each
node is available. In Section VI, decision node selection is
proposed for saving more energy. Simulation results are
discussed in Section VII. Conclusions are drawn in Section
VIII.
II. SYSTEM MODEL

Our system contains sensor nodes, one primary user and
fusion center (FC) as shown in Fig.1. It is assumed that is
the sensing time and

is the sampling frequency of the


received signal from primary user.and

are the same for all


sensors. is a multiple of

. Thus, the number of samples


is

which is assumed to be an integer value. Each node


decides which channel is busy or idle based on samples of its
receiving signal, that is,

, k=1,2,

and .
Therefore, there are two hypotheses,

and

as follows

(1)

(2)

where is the primary user signal which is assumed to
be deterministic;

is the channel gain between each node and


primary user and it has a model as follows

(3)

where

is a complex Gaussian random process with zero


mean and unit variance accounting for Raleigh fading and


has two components

(4)
where the first part is path loss component based on free-
space path loss (FPL) model which involves

as the
distance of each node from primary user,

as the carrier
frequency denoted by 2.4 and C is the speed of light. The
second part is a real Gaussian random variable with zero mean
and standard deviation of 3 according to large scale log-
normal shadowing [18].
The noise

is a Gaussian, independent and identically


distributed ( ) random process with zero mean and
variance

shows that the primary user is not active and


channel is idle while

states the primary user exists and


channel is busy.

is the primary user signal-to-noise


ratio(SNR) in the -th sensor under the hypothesis

The
energy detection scheme is used for signal detection for the -
th sensor as follows


(5)
where is the detection threshold. If the energy of the
receiving samples is greater than , the channel will be busy
(

; otherwise channel will be idle

. Sensors
use one bit to determine their results. Under

is a random
variable whose probability density function (pdf) is a central
chi-square distribution with 2

degrees of freedom. Under

has a non-central chi-square distribution with 2


degrees of freedom and a non-centrality parameter

. False
alarm probability states the probability in which the channel is
considered busy while it is really idle. In fact, it shows missing
the opportunity of using the spectrum holes. The probability of
false alarm is defined as [12],[19]

(6)

where is the incomplete gamma function given by

and is the gamma function.


Probability of detection is the probability that channel is
considered busy when it is really busy. This probability
protects transmission of the primary user from the interference
produced by the secondary user. Therefore, it is stated as
follows [12],[19]

(7)

where

is the generalized Marcum -function,

, in which

is the modified Bessel function of the first kind and


order .


Fig.1. Cooperative spectrum sensing configuration
In [20], cooperative spectrum sensing is considered as
the solution to overcome the problems such as fading and
shadowing effects and also to increase the sensing
performance. In cooperative spectrum sensing, each sensor
determines the channel status and sends its result to the fusion
This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TVT.2014.2331681
Copyright (c) 2014 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
4
center (FC). FC makes the final decision about the channel
based on the receiving results and a fusion rule. We consider
OR rule used in FC in which if at least one sensor reports that
the primary user is active, final decision means the channel is
busy. Therefore, the global probability of false alarm and
probability of detection can be written as

(8)

(9)

We propose an energy-efficient cooperative spectrum
sensing by selecting the sensors that participate in spectrum
sensing and the optimal detection threshold setting while
maintaining constraints on global probability of detection and
probability of false alarm. In the proposed algorithm, some
sensors sense the spectrum; therefore, (8) and (9) should be
modified as fallows

(10)

(11)

where

is the assignment index and its value is 0 or 1.


0 indicates that the sensor node is not sensing the spectrum
while 1 shows that the sensor is a sensing node.
The model in [12] is used to obtain the total energy
consumption for cooperative spectrum sensing. Therefore,
total energy consumed can be calculated through the following
formula

(12)

It has two parts: first,

is the energy consumption for


sensing the channel, decision about the channel status and
signal processing like modulation type, signal shaping, etc.

is assumed the same for all sensors and denoted by

. The
second part,

, is the consumed energy to transmit one


reliable bit from each sensor node to FC.

is used to derive
the radio electronics and the power amplification. Then,

is
expressed as follows

(13)

where

is the transmitter electronics energy,

is
the required amplification and

is the distance between -th


node and FC. We consider a case study with IEEE
802.15.4/ZigBee radios as a communication technology
between sensor nodes which cooperate with each other.
Therefore,

and

is
considered

to satisfy a receiver sensitivity of -


90 [21],[22].
Two constraints are considered on

and

, so that

and

. It is desired that to be zero and to be


one. According to (6) and (7),

and

are dependent on
the detection threshold (, so that the value of this parameter
affects the selection of the nodes for spectrum sensing.
Therefore, with proper selection of , it is possible to save
energy while maintaining the constraints on

and

.
Therefore, the problem is stated as follows

(14)



Since,

is an increasing function of

and it is
independent of

, the maximum number of sensing nodes can


be obtained using (10) and

.Therefore, we have

(15)

where is the maximum number of sensing nodes and

is the number of nodes participating in spectrum


sensing. Then, our problem can be modified as follows

(16-1)

(16-2)

(16-3)

According to (16), our problem is minimizing energy
consumption by selecting the sensing nodes and setting the
detection threshold while satisfying the detection performance.
It should be noted that

is a different discrete parameter for


each sensor. Therefore, our problem has unknown
parameters while there are only two parameters in [12]: the
on/off and censoring parameters which are the same for all
sensors. In addition, with the same SNR, the local probability
of detection is also the same for all nodes. Hence, the global
probability of detection in (16-2) is only dependent on the
number of sensing nodes while in our paper, it is important to
select the nodes which sense the spectrum due to different
SNRs. In fact, this problem is the integer programming
problem with discrete parameters and one continuous
parameter. It makes it more difficult than other pervious
works. In [13]-[17], the problem becomes simpler because
minimizing the energy consumption is not considered and the
problem is only optimizing the detection performance by
setting the detection threshold. This problem is also solved
and analyzed for two scenarios: knowing instantaneous SNR
or knowing the average SNR. In the next section, the solution
for the problem is analyzed.
This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TVT.2014.2331681
Copyright (c) 2014 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
5
III. NODE SELECTION ANALYSIS FOR INSTANTANEOUS SNR
In the analysis, it should be noted that

and

are
increasing functions of

and decreasing functions of . [23]


shows that

is a discrete parameter; and solving the


problems similar to (16) is NP-complete. In order to simplify
this problem, we assume

a continuous parameter with


values in [0,1]. Such an assumption converts the problem to
more tractable form. After solving our problem,

is mapped
to discrete space again. Now, the problem can be rewritten as
follows

(17-1)

(17-2)

(17-3)

In the above problem, the objective function, (17-1) and(17-
3) are convex with respect to

while (17-2) is not. Although


(17) is not a standard convex optimization problem, we can
still exploit the convex optimization methods to find a local
instead of a global. Using Lagrangian function and Karush
KuhnTucker (KKT) conditions, the priority of nodes for
sensing the spectrum is determined. For this purpose, in the
first step, the detection threshold is considered as a fixed
parameter and

s are obtained with the fixed . Therefore, the


Lagrangian function is determined as follows [24],[25]

(18)

where and are the Lagrangian multipliers for (17-1) and
(17-2) constraints, respectively. From KKT conditions, we
have[24],[25]


(19)

In order to determine the priority of each node for spectrum
sensing, we use the following cost function [23]

(20)
Here, nodes with the lowest cost function are selected for
spectrum sensing until the constrain on

is satisfied.
Complimentary slackness conditions imply that [24]



It should be noticed that node selection can be affected by
(21-3) and (21-4). If the number of selected nodes for
spectrum sensing becomes fewer than , then (21-3) will be
satisfied and if number of selected nodes equals to , then,
(21-4) is true. It can be proved that (21-2) is the optimal
condition, because

and

are increasing functions of

s. Therefore,

s can be decreased so that

is
satisfied. Under this reduction, there are smaller

and


which leads to more desirable answer. Therefore, is the
true condition in (21-2).
In [23] a novel algorithm, modified energy efficient sensor
selection (MEESS), is introduced with linear order of
complexity to select the sensing nodes. It consumes less
energy and satisfies

. The details of MEESS can be


expressed as follows:
In this algorithm, an iterative method is used in order to find
the optimum value of in which the cost function in (20) is
computed for all sensors and then the nodes with lower cost
function are selected (i.e.,

s for the selected sensing nodes


are denoted by 1 and for other nodes are denoted by 0
[26].) until

is satisfied. Anyhow, the maximum


number of selected nodes for spectrum sensing is as noted
earlier. After the selection of the nodes, total energy
consumption is evaluated and the value of is updated
according to

conditions. The algorithm ends when the


accuracy of becomes smaller than.In the case that
reduces while

increases, it indicates that more nodes are


participating in spectrum sensing. So, we select the sensing
nodes in the last iteration, which consumes less energy while
maintaining

. MEESS algorithm is a good algorithm


for reducing energy consumption with a pre-specified
detection threshold. To save more energy, we propose an
algorithm in the next section to obtain optimum values for
and the detection threshold, jointly.
IV. PROPOSED ALGORITHM FOR NODE SELECTION AND
DETECTION THRESHOLD SETTING FOR INSTANTANEOUS SNR
In this section, an algorithm for finding the optimum values
of and is proposed. Get noticed that

is an increasing
function of the detection threshold as shown in Fig.2. In fact,
the feasible points to minimize the energy consumption and
satisfy the detection performance are searched. The set of all
feasible points is called the feasible set or the constraint set
[24]. In this problem, the feasible set is dependent on and

.
It should be noted that, there is no feasible point for some
values of . In Fig.3, it has been shown that increasing leads
to decrement in the global probability of detection. In fact, as
decreases,

and

increase and fewer number of sensing


nodes is required to satisfy the detection performance;
therefore the energy consumption reduces. Hence, the
minimum should be found in feasible set so that the global
pobability of detection and the global probability of false
alarm constraints can be satisfied.
This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TVT.2014.2331681
Copyright (c) 2014 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
6
1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 2.1 2.2
x 10
-10
3.25
3.3
3.35
3.4
3.45
3.5
3.55
x 10
-7
Threshold
E
n
e
r
g
y

Fig.2. Energy is an increasing function of the threshold

0 0.5 1 1.5 2 2.5 3 3.5
x 10
-9
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Threshold
P
r
o
b
a
b
i
l
i
t
y

O
f

D
e
t
e
c
t
i
o
n
Total Nodes=35
Total Nodes=20
Total Nodes=5
feasible set

Fig.3.The global probability of detection varies for different number of total
nodes when threshold changes
In order to achieve the minimum, first, we should find
the range for in which the feasible set is not empty. In
order to determine the feasible set, maximum probability of
detection algorithm (MPDA) is proposed. Using this
algorithm, for each, the probabilities of detection for all
nodes are computed and sorted in descending order. Then,
according to the maximum number of sensing nodes, the
sensors with maximum probability of detection are
determined. These nodes are selected until the


constraint can be maintained. That is, if the feasible set is
not empty, the answer for the problem will be found. The
value of is updated with the following rule: If the global
probability of detection becomes more than , will
increase; otherwise, will decrease. By this way, the
maximum is obtained. Now, for finding the minimum,
we use MPDA similar to the one used to obtain the
maximum threshold. However, the difference lies in
updating, where decreases if the global probability of
detection is greater than , and vice versa. Pseudo code for
this method has been shown below.

Bisection algorithm for determining the valuable range for
the detection threshold

=0

=(a large enough number)

= multiplier of noise power

=0
is a small number
% number of iteration
While (


n=0 % number of sensing nodes
compute

for every sensor node


compute


compute

for each node


sort (

, 'descend')

While (select sensor with higher probability of
detection<)


If

then
break
end if
end while
If

then


else


end if

end while

There is similar to this algorithm for obtaining the

.
But only the difference in updating

as
If


else


end if



After obtaining the desirable range of

, the minimum value of will be selected as the


optimum detection threshold because the energy
consumption decreases while the detection performance is
satisfied. Then, MEESS algorithm is used with this. The
This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TVT.2014.2331681
Copyright (c) 2014 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
7
new proposed algorithm is called joint optimization of
threshold and sensing nodes (JOTSN). In this algorithm,
and

s are obtained simultaneously. The complexity of


JOTSN is in the order of . To solve (17), it should be
noted that the objective function and

are convex with


respect to . However, super level of a convex function is a
non- convex set. In fact, although (17) is not a standard
convex optimization problem, it is possible to still exploit
the convex optimization methods for findinga local solution
instead of a global one. On the other hand, although the
problem is non-convex, the solution obtained by the convex
optimization methods is often the global optimal answer. It
is especially true when the objective function is convex
while the feasible set is non- convex. In JOTSN, in addition
to (18) and (19) we also have

(22)

The cost function and complimentary slackness conditions
are similar to (20) and (21), respectively. The flowchart of the
problem solution is shown in Fig.4.

Find the minimum detection
threshold
Compute The instantaneous probability of
detection for each sensor & Maximum
number of sensing nodes
MEESS algorithm to find
the sensing nodes
START
END

Fig.4. The JOSTN algorithm for finding the sensing nodes and optimum
detection threshold
V. ANALYSIS FOR SELECTION OF SENSING NODES AND
SETTING THE DETECTION THRESHOLD FOR AVERAGE SNR
In the previous section, the problem was solved for the case
that instantaneous SNR for each sensor has been given;
however, it is very difficult to obtain the instantaneous SNR
for each node. In realistic situations, the average SNR is
available for every sensor. In this case, the average probability
of detection is derived by taking the average of (7) over fading
statics. Therefore, we have

(23)

where

is the probability density function (pdf) of


SNR under fading channels. Under Rayleigh fading,


has an exponential distribution

(24)

where

is the average SNR for each sensor node and is as


follows

(25)

where indicates the expected value,

and

are the
fading and shadowing effects according to (3) and (4),
respectively.

is the transmitting power of the primary user


and

is the distance between each node and primary user.


Therefore, a closed-form expression for

is as follows [19]

(26)

Here, (26) shows that

is dependent on the average SNR


for the -th sensor. Note that

is independent of SNR
because there is no primary user signal under

. Therefore,

is the same for all sensors similar to (6) for instantaneous


SNR. Then, the average global probability of detection and the
average global probability of false alarm for the problem are
defined as

(27)

(28)
Hence, the new problem is written as

(29-1)
This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TVT.2014.2331681
Copyright (c) 2014 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
8

(29-2)

(29-3)


To solve it, we use convex optimization methods.
Therefore, the Lagrangian function is determined as follows

(30)

Similar to the previous section, JOTSN algorithm is used
for solving (29), obtaining the sensing nodes and setting the
detection threshold. The only difference is that, instead of
instantaneous SNR, the average SNR is considered for each
node. That is due to the fact that the estimation of the
instantaneous SNR for each sensor is very difficult in practice.
VI. DECISION NODE SELECTION TO IMPROVE ENERGY
EFFICIENCY
In the previous sections, the problem of minimizing energy
consumption was solved using the selection of the sensing
nodes and setting the detection threshold. Also, it is possible
for sensing nodes to send their results to a decision node (DN)
instead of FC. DN makes a decision on such results and
reports it to FC. Accordingly, the nodes which are farther from
FC consume less energy. DN decides about the channel status
based on the results sent from the sensing sensors via a fusion
rule. Fig.5 shows how one DN transmits the final results to
FC. OR rule is the fusion rule which is used to make the final
decision about the activity of the primary user.


Fig.5. Sending the final decision from DN to FC
Therefore, the new problem selects the decision node in
addition to the sensing nodes selection and setting the
detection threshold that satisfies the detection performance. In
this problem, the average SNR for each sensor is assumed to
be known. Then, the total energy consumption is stated as
follows

(31)

where

is DN selection index such that

. 1
indicates that the corresponding node is selected as a DN
while 0 means not being selected as a DN.

and

indicate
the location of the -th sensor node and

is the distance
between node and FC. Therefore, the new problem is

(32)


The last constraint shows that one node cannot be both a
sensing node and a decision node, simultaneously.
One solution for (32) is the exhaustive search algorithm in
which for possible candidates as a DN and each feasible,
all combinations of remaining sensors are tested for
selection of the sensing nodes. The DN candidate besides the
corresponding sensing nodes which save the most energy and
also satisfy

are selected as a decision and sensing


nodes. But this algorithm has the complexity with the order of

(33)

which is high. This leads us to look for an algorithm with
lower complexity. In doing so, discrete parameters of

and

are considered as continuous parameters to solve the


problem. Thereafter, they are mapped to the discrete space.
This way, the convex optimization methods can be used to
solve (32). Therefore, the Lagrangian function is as follows
[23]

(34)

To solve (32), the last constraint is removed. Nevertheless
one node cannot be considered as both DN and sensing node,
simultaneously. KKT conditions reveal that

(35)
This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TVT.2014.2331681
Copyright (c) 2014 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
9

(36)

Note that

is the same as (22). Due to (35) and similar to


the results of section III, the priority of sensors for being
selected as the sensing nodes is determined according to the
following cost function

(37)

which indicates the cost function for node when node is
selected as DN. The following cost function determines the
priority of sensors to be selected as a DN

(38)

Again, the complimentary slackness conditions for this
problem imply that




It should be noted that the first and the second conditions
are similar to (21) while the third condition indicates that only
one sensor can be selected as a decision node. Therefore,

is a true condition. The forth condition implies


that one node cannot be both a decision node and sensing
node, simultaneously; therefore,

is true.
To solve (32), an algorithm similar to JOTSN algorithm
introduced in the previous section is used. In this algorithm,
the minimum is used at each iteration. This way, is
updated and (37) is computed for every node as a decision
node candidate. Therefore, the minimum number of sensing
nodes which satisfies

is obtained. Then, every sensor


with minimum cost in (38) is selected as a DN and goes to the
next iteration. The flowchart in Fig.6 indicates the JOSTN
algorithm for finding the DN.
Find the minimum detection
threshold
Compute The average probability of
detection for each sensor & Maximum
number of sensing nodes
Updating Lambda
For each node as DN Obtain the sensing nodes and
the energy consumption based on (20 ) & (12),
respectively
Compute the cost function
for selecting the DN using
(38)
START
END


Fig.6. The JOSTN algorithm for finding the optimal DN

VII. NUMERICAL AND SIMULATION RESULTS
For simulation, a sensor network in which the number of
sensors varies from five to fifty nodes is considered. Sensor
nodes are distributed uniformly in the square field with the
length of 200. FC is located in the center of the square. The
primary user is located randomly in this square. Simulation
results are shown for and . The channel model
between every node and FC is based on free-space path loss
model. The number of samples equals to 6 corresponding to a
sensing time of . Simulation results are averaged over
1000 independent simulation runs.
The proposed algorithm is compared with the following
algorithms in simulation which have constant detection
threshold (as a multiple of the noise power:
MEESS: It has been already described in the
previous sections.
Minimum Energy Algorithm (MEA): In this
algorithm, sensors are sorted in ascending order
according to their distances from FC; and nodes
with minimum distances are selected for sensing
This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TVT.2014.2331681
Copyright (c) 2014 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
10
the channel so that

is satisfied. The number


of sensing nodes is kept fewer than . If MEA can
find the nodes which satisfy

constraint,
then, its answer is nearly optimal due to the
selection of nodes with minimum distances from
FC.
Maximum Probability of Detection Algorithm
(MPDA): In this algorithm, sensors are sorted in
descending order according to their

s. Therefore,
the nodes with the maximum

are selected so
that the number of selected nodes is kept fewer than
and

constraint is satisfied. MPDA will


find the solution for the problem if a solution exists.
Random Sensor Selection Algorithm (RSSA):
In this algorithm, sensors are randomly selected for
spectrum sensing so that the number of selecting
nodes is fewer than. This algorithm has the
minimum complexity for finding a solution for our
problem.
A. Sensing Nodes and Detection Threshold Setting
Simulation for Instantaneous SNR
In this section, we assume that the instantaneous SNR is
available for each node. In Fig.7 the successful percent of
finding a solution for the problem is given for different
number of nodes. It is important for all algorithms to find the
answer if the problem has the solution (i.e., the feasible set is
not empty). Successful percent of finding a solution, for each
algorithm is a desired indication for the algorithm. In some
scenarios, it is possible that the feasible set of problem
becomes empty. In such a case, there is no solution for
problem. In fact, this metric shows the ability of algorithms in
finding a solution in the feasible set of the problem.
It is clear that JOSTN has the most possible percentage in
comparison with the other algorithms due to the adaptive
setting for the detection threshold. Note that the algorithms are
compared when the problem has an answer, i.e., the constraint
on the global probability of detection is satisfied. It is shown
that MEA and RSSA have the minimum successful percentage
of finding a solution. They become distanced from the other
algorithms when the number of total nodes increases. The
reason is that, in large number of nodes, more sensors are
located close to FC. Therefore, according to MEA, the nodes
with minimum distances from FC are selected for spectrum
sensing. Of course, these nodes might have small probabilities
of detection. MEESS overlaps with MPDA, and this shows
that when the problem has a solution, MEESS can find it.
When the problem has no solution, the channel is assumed to
be idle.
Fig.8 shows the average energy consumption versus
different number of nodes. Clearly, the detection threshold
setting helps with more energy saving while maintaining

constraint. MEESS consumes more energy than


JOSTN due to the constant value of the detection threshold.
MEA has the maximum energy consumption because this
algorithm selects the sensors according to the distances from
FC. Therefore, when the probability of detection for each node
is low due to the value of the detection threshold, MEA needs
more nodes for spectrum sensing until the detection
constraints are satisfied.
5 10 15 20 25 30 35 40 45 50
20
30
40
50
60
70
80
90
Total nodes
s
u
c
c
e
s
s
f
l

p
e
r
c
e
n
t

o
f

f
i
n
d
i
n
g

s
o
l
u
t
i
o
n


MEESS
MEA
RSSA
JOTSN
MPDA

Fig.7. Successful percent of finding an answer for different nodes
5 10 15 20 25 30 35 40 45 50
200
400
600
800
1000
1200
1400
1600
1800
Total nodes
E
n
e
r
g
y

i
n

n
J


MEESS
JOTSN
MEA
MPDA
RSSA
Fig.8. Energy consumption with instantaneous SNR for each node
Fig.9 shows the average number of selected sensing nodes
versus different total number of nodes. JOSTN algorithm uses
fewer number of sensing nodes for spectrum sensing while
satisfying the probability of detection constraint. It is clear that
MEA and RSSA have the maximum number of sensing nodes
which affects the energy consumption. Again, the algorithms
are compared when all of them have an answer.

This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TVT.2014.2331681
Copyright (c) 2014 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
11
5 10 15 20 25 30 35 40 45 50
0
5
10
15
20
25
30
Total nodes
A
v
e
r
a
g
e

N
u
m
b
e
r

o
f

S
e
l
e
c
t
e
d

S
e
n
s
i
n
g

N
o
d
e
s


MEA
RSSA
JOTSN
MEESS
MPDA

Fig.9. The average number of selected sensing nodes for spectrum sensing
B. Sensing Nodes Selection, Detection Threshold Setting and
Decision Node Selection Simulation for Average SNR
In this section, the average probability of detection is
considered for sensor selection in the case of information
insufficiency which is a function of the average SNR. The
main question raised here is that how our algorithm works to
satisfy the instantaneous detection performance (

and energy consumption when the average probability of


detection in sensor selection is used due to information
insufficiency. It should be mentioned that the instantaneous
SNR is applied in order to evaluate the global probability of
detection and energy consumption by the selected nodes. Our
decision about sensor selection is made based on average
SNR; however, the effects of the decision on the energy
consumption and detection performance are evaluated based
on the instantaneous SNRs.
Fig.10 shows the successful percent of finding an answer
for different algorithms. JOTSN and JOTSN with DN have the
maximum probability rate in finding the solution while RSSA
has the minimum rate. It is shown that MEESS, MEESS with
DN, MPDA, JOSTN and JOST with DN get close to each
other when the total number of nodes increases. Again, with
an increase in the total number of nodes, the successful
percentage of finding the solution for MEA and RSSA
algorithms considerably decreases. In comparison with Fig.7,
the successful percentage of finding a solution increases
because more sensors are selected to satisfy the average global
probability of detection in case of less information for each
node on the instantaneous SNR. The larger number of sensing
nodes leads to more energy consumption compared to the case
of using instantaneous SNR.
In Fig.11, the average energy consumption versus different
number of nodes has been obtained. It shows that DN
selection helps save more energy. JOSTN with DN selection
consumes less energy than JOSTN. MEA consumes a lot of
energy to satisfy the constraint on probability of detection.
MEESS and MEESS with DN consume more energy than
JOSTN and JOSTN with DN. Therefore, it shows that the
proposed algorithms are very effective in saving energy. In
comparison with Fig.8, when only average SNR is available to
satisfy the detection performance, more sensors are needed.
Therefore, energy consumption increases.


Fig.10 Successful percent of finding an answer for different nodes
5 10 15 20 25 30 35 40 45 50
0
500
1000
1500
2000
2500
3000
Total nodes
E
n
e
r
g
y

i
n

n
J


MEESS
JOTSN
JOTSN & DN
MEA
MPDA
RSSA
MEESS & DN
Fig.11. Energy consumption for different algorithms in average SNR

In Fig.12, the average number of selected sensing nodes
versus the total number of nodes is depicted. The proposed
algorithm utilizes fewer number of sensing nodes while it
saves more energy and finds the best solution for the problem.
In comparison with Fig.9, the number of sensing nodes
increases to satisfy the average probability of detection
constraint due to information insufficiency.



5 10 15 20 25 30 35 40 45 50
10
20
30
40
50
60
70
80
90
100
Total nodes
s
u
c
c
e
s
s
f
l

p
e
r
c
e
n
t

o
f

f
i
n
d
i
n
g

s
o
l
u
t
i
o
n


MEESS
MEA
RSSA
MEESS &DN
MPDA
JOTSN & DN
JOTSN
This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TVT.2014.2331681
Copyright (c) 2014 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
12
5 10 15 20 25 30 35 40 45 50
0
5
10
15
20
25
30
Total nodes
A
v
e
r
a
g
e

N
u
m
b
e
r

o
f

S
e
l
e
c
t
e
d


S
e
n
s
i
n
g

N
o
d
e
s


JOTSN
MEA
RSSA
MEESS & DN
MEESS
MPDA
JOTSN & DN
Fig.12. Average number of selected sensing nodes versus different number of
total nodes
VIII. CONCLUSION
Cooperative spectrum sensing is essential in cognitive
wireless sensor networks for mitigating the effects of
shadowing and fading and also for improving the detection
performance. However, it is not necessary for all sensors to
sense the channel in order to detect primary user activity. If
they do so, each sensor will spend a lot of energy while the
detection performance does not get considerably improved in
the network.
We proposed an energy efficient cooperative spectrum
sensing in cognitive sensor networks. Such a scheme selects
the sensing nodes and sets the value of detection threshold
subject to the constraints on the detection performance. The
problem was formulated in fading channels and convex
optimization methods were used to solve it. In order to save
more energy, the problem of decision node selection,
corresponding sensing nodes, and setting the detection
threshold was expressed and solved using mathematical
optimization methods. The problem was addressed in two
cases. First, the instantaneous SNR and as a result the local
probability of detection for each sensor is available. In the
second case, it was assumed that the average probability of
detection can be computed using the probability density
function (pdf) of SNR since knowing the instantaneous SNR
in real situations is difficult. The simulation results showed
that the proposed algorithm considerably helps reduce energy
consumption. Successful percentage in finding the solution
was a metric to show that our proposed algorithm can find the
answer whenever the problem has the solution. The solution
is independent of the probability of detection distribution and
type of the fusion rule. So, it can be extended to any other
distribution and any other type of the fusion rule in a
straightforward manner.


REFERENCES
[1] S. Haykin, Cognitive radio: Brain-empowered wireless
communications, IEEE J. Sel. Areas Commun., vol. 23, no. 2, pp. 201
220, Feb.2005.
[2] S. M. Mishra, A. Sahai, and R. W. Brodersen, Cooperative sensing
among cognitive radios, in Proc. IEEE Int. Conf. Commun., Jun. 2006,
pp. 16581663.
[3] S. Maharjan, C. Yuen, Y. H. Chew, Y. Zhang, and S. Gjessing,
Distributed spectrum sensing for cognitive radio networks with
heterogeneous users: A game theoretical framework, in Proc. IEEE
CogART, Rome, Italy, Nov. 2010, pp. 15.
[4] W. Saad, Z. Han, T. Basar, M. Debbah, and A. Hjorungnes, Coalition
formation games for collaborative spectrum sensing, IEEE Trans. On
Veh.Technology, vol. 60, no. 1, pp. 276297, Jan. 2011.
[5] I.F. Akyildiz, W. Lee, M.C. Vuran, S.Mohanty, A Survey on spectrum
management in cognitive radio networks, IEEE Commun. Mag., vol.46,
no.4, pp. 40-48, Apr. 2008.
[6] A. Ghasemi, S. Sousa, Spectrum sensing in cognitive radio networks:
requirements, challenges and design trade-offs, IEEE Commun.
Mag.,vol.46, no.4, pp. 32-39, Apr. 2008.
[7] T. Ycek and H. Arslan, A survey of spectrum sensing algorithms for
cognitive radio applications, IEEE Comm. Surveys and Tutorials, vol.
11, no. 1,pp.116-130, 2009.
[8] A. Sahai, N. Hoven, and R. Tandra, Some fundamental limits on
cognitive radio, in Proc. Allerton Conf. Commun. Control Comp.,
Monticello, IL, Oct. 2004, pp. 16621671.
[9] D. Cabric, S. M. Mishra, and R. W. Brodersen, Implementation issues
in spectrum sensing for cognitive radios, in Proc. of Asilomar Conf.
on Signals, Systems and Computers, Pacific Grove, CA, USA, Nov.
2004, pp. 772776.
[10] E. Peh and Y. C. Liang, Optimization for cooperative sensing in
cognitive radio networks, in Proc. IEEE WCNC, Kowloon, Hong
Kong, Mar. 2007, pp. 2732.
[11] K. Yamasaki and T. Ohtsuki, Design of energy-efficient wireless
sensor networks with censoring, on-off, and censoring and on-Off
sensors based on information, IEEE Vehicular Technology Conf.
(VTC), May 2005,pp.1312-1316.
[12] S. Maleki, A. Pandharipande, and G. Leus, Energy-efficient distributed
spectrum sensing for cognitive sensor networks, IEEE Sensors Journal,
vol. 11, no. 3, pp. 565573, Mar. 2011.
[13] W. Zhang, R. K. Mallik, Kh. B. Letaief, Cooperative spectrum sensing
optimization in cognitive radio networks, IEEE Trans.on Wireless
Commun.,vol.8, no.12, pp.5761-5766, Dec.2009.
[14] S. Atapattu, Ch. Tellambura and H. Jiang, Spectrum sensing via
energy detector in low SNR , IEEE International Conf. on
Communications (ICC), Jun.2011, pp.1-5.
[15] Sh.Xie, L.shen and J.Liu, Optimal threshold of energy detection for
spectrum sensing in cognitive radio, IEEE International Conf. on
Wireless Communications & Signal Processing (WCSP),
Nov.2009,pp.1-5.
[16] P.R Nair, A.P.Vinod and A.K.Krishna, An adaptive threshold based
energy detector for spectrum sensing in cognitive radios at low SNR,
IEEE International Conf. on Communication Systems(ICCS),
Nov.2010,pp.574-578.
[17] D.Mercedes, M.Plata and A.G.A.Reatiga, Evaluation of energy
detection for spectrum sensing based on the dynamic selection of
detection-threshold, Elsevier on Procedia Engineering journal, vo.35,
pp.135-143,2012.
[18] B. Sklar, Rayleigh fading channels in mobile digital communication
systems Part1: characterization,IEEE Commun. Mag.,vol.35,no.7,
pp.90-100, Jul. 1997.
[19] F. F. Digham, M.-S. Alouini and M. K. Simon, On the energy detection
of unknown signals over fading channels, IEEE Trans. On Commun.
vol.55, no.1,pp.21-24, Jan.2007.
[20] Y.-C. Liang, Y. Zeng, E. C. Peh, and A. T. Hoang, Sensing-throughput
tradeoff for cognitive radio networks, IEEE Trans. on Wireless
Commun., vol. 7, no. 4, pp. 13261337, Apr. 2008.
[21] S. Maleki, A. Pandharipande, and G. Leus, Energy-efficient distributed
spectrum sensing for cognitive sensor networks, in Proc. 35th Annu.
Conf. IEEE Ind. Electron. Soc., Nov. 2009, pp. 26422646.
This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TVT.2014.2331681
Copyright (c) 2014 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
13
[22] S. Maleki, A. Pandharipande, and G. Leus, Energy efficient distributed
spectrum sensing with convex optimization, in Proc. 3rd Int.
Workshop Comput. Advances in Multi-Sensor Adaptive Processing,
Nov.2009, pp. 396399.
[23] M.Najimi, A. Ebrahimzadeh, S.M.Hosseni Andargoli, A. Fallahi, A
novel sensing nodes and decision node selection method for energy
efficiency of cooperative spectrum sensing in cognitive sensor
networks, IEEE Sensors Journal, vol.13,no.5, pp.1610-1621,May 2013.
[24] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, U.K.:
Cambridge Univ. Press, 2004.
[25] W. Yu and R. Lui , Dual methods for nonconvex spectrum optimization
of multicarrier systems, IEEE Trans. On Commun, vol.54, no.7,
pp.1310-1322, Jul.2006.
[26] B. Yang, Kh. B. Letaief and R. S. Cheng, Timing recovery for ofdm
transmission, IEEE J. Sel. Areas Commun, vol. 18, no. 11, pp.2278-
2291, Nov.2000.



























































Dr.Ataollah Ebrahimzadeh received
his Ph.D. degree in electrical
engineering from Ferdsi university,
Mashhad, Iran, in 2007. Now he is an
associate professor in the Faculty of
Electrical and Computer Engineering at
Babol University of Technology. His
current scientific interests are in general
area of signal processing and artificial
intelligence. He is the reviewer of
international conferences and journals.
He has published more than 50 papers in
international journals.



Maryam Najimi received her B.Sc in
electronics from Sistan & Baloochestan
University, Zahedan, Iran in 2004 and her
M.Sc in telecommunication systems
engineering from K.N.Toosi University of
Technology, Tehran, Iran and Ph.D.
degree in communication from Babol
University of Technology, Mazandaran,
Iran, in 2008 and 2014, respectively. Her
interests include Spectrum sensing in
wireless cognitive sensor networks.




Dr. Seyed Mehdi Hosseini Andargoli
received B.Sc degree in electronics
engineering from Shahed university,
Tehran, Iran, in 2004, the M.Sc. degree
and Ph.D. degree in telecommunication
systems engineering from K. N. Toosi
University of Technology, Tehran, Iran, in
2009 and 2011, respectively. He is
currently an assistant professor with
department of electrical and computer
engineering, Babol Noshirvani University
of technology, Babol, Iran. His research
interests include resource allocation of
cellular networks, cognitive radio networks, relay networks, sensor
networks, optimization and MIMO-OFDM systems.




Dr. Afshin Fallahi received the B.Sc.
degree from University of Tehran, Iran, in
1996, the M.Sc. degree from Tarbiat Modarres
University (TMU), Iran in 1999, and PhD
degree from University of Manitoba, Canada,
in 2008, all in Electrical Engineering. His
main research interests are in the area of
modeling, analysis and optimization for
wireless networks.



This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication.
The final version of record is available at http://dx.doi.org/10.1109/TVT.2014.2331681
Copyright (c) 2014 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

You might also like