Download as pdf or txt
Download as pdf or txt
You are on page 1of 25

Wireless Personal Communications (2020) 114:123–147

https://doi.org/10.1007/s11277-020-07354-7

Differential Evolution Based Reliable Cooperative Spectrum


Sensing in the Presence of Malicious Users

Noor Gul1,3 · Ijaz Mansoor Qureshi2 · Muhammad Sajjad Khan1 · Atif Elahi1 ·
Sadiq Akbar3

Published online: 13 April 2020


© Springer Science+Business Media, LLC, part of Springer Nature 2020

Abstract
In Cognitive Radio Network, access to the vacant spectrum holes in the licensed Primary
User (PU) channel is allowed to the Secondary Users. In order to use the PU spectral holes
accurately and on time, spectrum sensing is extremely important to avoid interference with
the PU transmission. In Cooperative Spectrum Sensing (CSS) individual users report the
Fusion Center (FC) about sensing data for a global decision which leads to better sensing
results as compared to non-cooperative sensing. The objective of this paper is to make the
CSS performance precise and accurate in the presence of Malicious Users (MUs) report-
ing false sensing information to the FC. The proposed scheme reduces falsifying effects of
the YES, NO and OPPOSITE categories of MUs in CSS using Differential Evolution (DE)
algorithm. A dynamic threshold value is determined using optimum coefficient vector by
the DE at the FC. This leads to minimum error probability in detecting PU channel when
the YES, NO and OPPOSITE categories of MUs reports FC in CSS. The weighting coef-
ficient vector is further employed at the FC to assign weights to the received sensing infor-
mation of cooperative users. Therefore, cooperative users with minutely erroneous reports
receive high weight as compared to MUs. The simulations performed for varying number
of MUs, total cooperative users, Signal to Noise Ratios and sensing samples show mini-
mum error in detecting PU activity for the proposed DE technique in comparison with the
Kullback–Leibler divergence, particle swarm optimization, Genetic Algorithm and Maxi-
mum Gain Combination schemes.

Keywords Cognitive radio network · Differential evolution · Genetic algorithm ·


Cooperative communication · Particle swarm optimization · Signal-to-noise-ratio ·
Multipath channels

* Noor Gul
noor.phdee51@iiu.edu.pk; noor@uop.edu.pk
Extended author information available on the last page of the article

13
Vol.:(0123456789)
124 N. Gul et al.

Fig. 1  Cognitive radio duty cycle

1 Introduction

The tremendous growth in wireless communications and search for high data rates in the
last decade has caused shortage of the radio spectrum. Therefore, Federal Communication
Commission (FCC) is considering a more flexible and comprehensive use of the available
spectrum resources using Cognitive Radio (CR) technology [1, 2]. Spectrum measure-
ments in different parts of the world show that more serious problem is the spectrum access
than its physicalinadequacy. These spectrum resources are most of the time either partially
occupied or completely unoccupied. Therefore, a more flexible and efficient management
policy with compatible technology, such as Dynamic Spectrum Access (DSA) is needed.
It allows allocation of the spectrum resources dynamically to different users operating in
similar allocated frequency range with unrestricted access. Therefore, regulatory bodies are
considering DSA to resolve this issue of spectrum underutilization.
The idea of CR was given by J. Mitola, which is derivation of a Latin word “cogno-
scere” meaning, to come to know or to become aware of something [3]. It provides a new
method for wireless communication design with the objective to provide wireless transmis-
sion services using DSA. CR resolves the spectrum underutilization problem by improving
wireless communication performance and frequency spectrum consumption. The typical
duty cycle of the CR consists of sensing analysis, reasoning and adaptation as illustrated
in Fig. 1, where, Primary User (PU) spectral holes are detected by the Secondary Users
(SUs). The analysis is to select the best available frequency band at the SUs and coordinate
the availability of the spectral resources with the CR neighbors. Finally, it has to vacate
the spectral resources whenever the PU reappears and starts accessing the radio spectrum.
Accurate sensing of radio spectrum is extremely requisite, in order to avoid disturbances
created by the SUs for the legitimate PUs. As sensing ability of an individual user might
be affected due to the multipath fading and hidden terminal issues, therefore, Cooperative
Spectrum Sensing (CSS) is helpful to reduce these difficulties. CSS can be categorized into
centralized or distributed, where in the centralized CSS sensing, users report the Fusion
Center (FC) their observations for a global decision. In distributed CSS, the channel sens-
ing and detection results are shared among cooperative users locally without any central
control and monitoring station [4, 5].

13
Differential Evolution Based Reliable Cooperative Spectrum… 125

2 Related Work and Contributions

CSS leads to high detection with minimum false alarm that results in reduce error prob-
abilities in decisions. Despite considerable advancement in CSS, the system security
still needs further improvement. The study of Malicious Users (MUs) to secure the CSS
against inaccurate sensing information is a challenging research area these days. These
attackers create confusion at the FC about the actual status of the PU spectrum. There-
fore, the CSS needs protection and immunity against these spectrum falsification attacks
[6, 7].
In CSS, the most commonly used sensing detectors in the literature are the Matched
Filter Detector (MFD), Cyclo-stationary Detector, Feature Detector and Energy Detec-
tors. Although MFD are able to perform better than other detection schemes but the
prior knowledge requirements of the PU channel makes their implementation difficult.
Due to minimum complexity and simple hardware requirements, Energy Detectors are
easy to implement in CSS. The use of Genetic Algorithm (GA) in CSS that leads to
optimal global decision against the cooperative users is investigated in [8–10].
The collaboration among SUs is vulnerable to serious security threats by malicious
nodes altering the sensing information for their personal benefits. The work in [11], has
quantified the number of affected nodes due to false reports of malicious nodes with
derivation of formulas validated in simulation scenarios for a Poisson point process.
The byzantine users, jammers and Primary User Emulation Attackers (PUEA) can seri-
ously affect performance of the CSS. Byzantine user is a type of spectrum sensing data
falsification attack in which MUs report false information to the FC. Traditional jam-
ming attacks target the frequency band of the operating radio with injecting malicious
signals that interfere with desired signal. PUEA attackers prevent access to the licensed
user spectrum by masquerading as the PU in order to prevent other users access to the
spectrum [12].
Byzantine users with a novel corruption strategy is implemented in [13]. This strat-
egy allow FC to make the attackers statistics distinguishable from the normal cooper-
ative users using message passing algorithm. As the selfish cooperative users are not
willing in sharing accurate sensing data with the FC, therefore, a contract theory based
approach is proposed in [14], as an incentive design mechanism. It rewards the coopera-
tive users in order to strengthen their cooperation in future. A novel algorithm proposed
in [15], deals with an arbitrary number of byzantine attackers. The parameter server is
tuned to select a small clean dataset for computing noisy gradients to filter out compro-
mised participants information. The research in [16], has formulated an abnormal power
emission due to device fault, selfish motivation and malicious attacks as a composite
binary hypothesis test. A recursive updating algorithm and order statistics proposed in
[17], helps in the selection of SUs with higher sensing reputation to reduce the impact
of MUs in CSS. The honest SUs in [18] are following the decision results of the CSS
as a final decision and use its local sensing information to guarantee the CSS reliability.
The use of privacy preserving protocol discussed in [19], use different cryptographic
schemes to preserve the location privacy of the SUs along with maintaining reliable
sensing performance. A privacy preserving scheme using additional architectural entity
along with cryptographic techniques to user location privacy, while performing spec-
trum sensing is in [20]. Similarly, the effect of Spectrum Sensing Data Falsification
(SSDF) users is reduced by measuring and updating the cooperative users credit in [21].
A data fusion scheme with effective fusion rule is investigated in [22], to counter SSDF

13
126 N. Gul et al.

attacks in CSS. Security issues of the cognitive radio wireless sensor network are dis-
cussed at the participation of PUEA and SSDF. In [23], a novel attack proof method
with M-ary quantized data for CSS, including MU identification and adaptive linear
combination technique is discussed.
In CSS, individual attackers sometime collude and share attacking information to
establish more stronger attack at the FC. These collusion attacks may not only increase
the individual attackers power at the FC, but can also easily avoid any defense mecha-
nism employed at the FC. A defense scheme from the XOR distance analysis perspective to
suppress collusion SSDF attackers is discussed in [24]. One problem with the trust based
CSS is that the feedback data initiator SUs remain unchecked at the FC. The PU spec-
trum status information feedback by the FC is often exploited by the attackers to disturb
trust mechanism in a collusive manner [25]. A Bayesian-inference using trust model slid-
ing window identify and reduce individual and collaborative probabilistic SSDF attackers
in [26]. In the penalty based collision attack prevention scheme in [27], system security is
enhanced against all possible security attacks of the individual and collaborative environ-
ment. A dynamic collusive SSDF attacker in [28], is maintaining high level trust at the FC
by reporting accurate sensing data dynamically and share fake sensing data to strengthen
the attack behavior. Similarly, performance of the randomly false attack model is analyzed
at the contribution of the individual and collaborative attackers at the FC in [29]. The tra-
ditional schemes assume pattern of the attacks at the FC. The work in [30] adopted an arbi-
trary behavior to restrict pre-assumed patterns and assumptions of the attacker by following
novel learning-empowered framework named Learn-Evaluate-Beat (LEB) to mislead FC.

2.1 Contributions

In our previous work in [31, 32], statistical Kullback–Leibler (KL) divergence schemes
are employed to determine dissimilarity in the Probability Distribution Function (PDF),
in respect of the reported soft energy information of the users to guarantee users trust,
while making global decision. Similarly, in [33, 34] we employed GA, to govern suitable
PU sensing information for a global decision based on the reported hard binary data of
the users. A static threshold mechanism is deviced for energy detectors to decide without
assigning weights to the individual users sensing data. In our outlier detection method in
[35], we used one-to-many sensing distance and Z-score statistical results of the user’s hard
binary decisions to identify and declare reported data from MUs as outlier amongst the
sensing data reported by all cooperative users with static threshold mechanism.
The current research work improves the CSS sensing performance in the presence
of YES, NO and OPPOSITE categories of fraud sensing MUs using Differential Evolu-
tion (DE) optimization. The proposed technique employs energy detection scheme at
each cooperative user, where the observed energy is compared with an adaptive thresh-
old in each sensing period determined by the DE instead of keeping static threshold as
in [31–35] against the user sensing information. In this paper the sensing users send soft
energy reports to the FC instead of local hard binary decisions as in [31, 32]. Cooperative
users are assumed at different geographical locations and experience independent Raleigh
fading effects, therefore, it is not suitable to treat their sensing notifications similar in the
global decision and equal weight assignment. The weighted coefficient vector that gives
optimum threshold point is selected in the proposed DE technique based on the soft energy
reports of all cooperative users before making global decision of the PU channel. This
optimum threshold is further selected as the global best threshold against the user reports.

13
Differential Evolution Based Reliable Cooperative Spectrum… 127

The individual users sensing reports are assigned with different weights at the FC in the
global decision. The cooperative users with trusted sensing information receive priority at
the FC with the assignment of high weights. MUs with false sensing reports that mislead
FC global decision are given low weights. This leads to a minimum false alarm with high
detection and low error probability at the FC as compared with [31–36]. Therefore, FC
decision in CSS is made secure and reliable in the presence of different categories of MUs.
An optimal weighting coefficient vector was investigated in [31, 36] using Particle
Swarm Optimization (PSO), GA, KL divergence and Maximum Gain Combination (MGC)
techniquesin a non-malicious environment. In this paper, the optimal weighted coefficient
vector results are determined for the proposed DE along with investigating the performance
of PSO, GA, KL and MGC schemes in the presence of OPPOSITE, YES and NO catego-
ries of MUs. The final weighted results of the DE are further utilized by the Soft Decision
Fusion (SDF) to reach to a final global decision at the FC. Simulations are performed for
different number of SUs, Signal-to-Noise Rations (SNRs) and sensing samples.
The results demonstrate a reliable and accurate detection performance of the PU chan-
nel by the proposed DE based CSS as compared to the PSO, GA, KL and MGC schemes
in [31, 36], with minimum error probability, high detection probability and low false alarm
for the proposed DE based scheme.
This paper is further categorized into the following sections: System model is discussed
in Sect. 2. The proposed model for determining optimal weighted results using GA is in
Sect. 3. Section 4 illustrates the simulation outcomes. Finally, concluded remarks and fur-
ther research directions are included in Sect. 5.

3 System Model

In the conventional CSS model illustrated in Fig. 2, users sense the PU channel and report
their statistics for a global decision to the FC. The normal users in this scheme report the
actual information of the channel to the FC, while the MUs report the FC with false sens-
ing information to mislead the system global decision. Energy distribution of the normal,
OPPOSITE, YES, NO and RANDOM categories of MUs are shown in Fig. 3. The OPPO-
SITE user reports an always opposite energy statistics to the FC without considering actual
PU status. It reports low energy statistics if the channel is busy and likewise forward high
statistics in the absence of PU. The presence of OPPOSITE user in CSS leads to a low data
rate and increased interference with the PU. The YES category of MU always report high
energy statistics, that leads to a high data rate for the SUs. The NO category of MU sense
the PU activity and report FC with low energy statistics to indicate an always idle condi-
tions of the channel that result in an interference for the PU. Similarly, the RANDOM user
probabilistically operate equally as OPPOSITE with probability (P) and operate normally
with the probability (1 − P). The involvement of any user from these categories leads FC to
decide inaccurately about the PU activity, therefore, they can reduce the data rate and cre-
ate unacceptable interference for the legitimate users.
Cooperative users report FC with their local sensing observations, where appropriate
weighting coefficient vector is determined against the sensing users to take a global deci-
sion using SDF scheme.
The weighted SDF scheme is illustrated in Fig. 4. The FC receives sensing informa-
tion from the cooperative users including normal and MUs. The proposed system takes
global decision based on the received sensing notifications from the cooperative users with

13
128 N. Gul et al.

Fig. 2  Conventional centralized


CSS

optimum weighting coefficient vector and threshold determined using DE. Performance of
the proposed scheme is investigated in the presence of cooperative users under different
channel conditions and YES, NO and OPPOSITE categories of MUs.
The i th user observes the PU channel with binary hypothesis H0 and H1 as;
{ }
H0 ∶ Xi [n] = Wi [n]
H1 ∶ Xi [n] = gi S[n] + Wi [n]
, i ∈ 1, 2, … , M, n ∈ 1, 2, … , K (1)

where H0, is the hypothesis that shows no PU activity in the channel and, H1, is the pres-
ence hypothesis of the PU channel. In Eq. (1), Xi [n], is the PU received signal at the ith
user in the nth sensing time slot. The total number of sensing samples K = 2BTs, are con-
sidered large enough so that to make the energy distribution Gaussian under both H0 and
H1 hypothesis. gi , is the channel gain between PU and ith sensing users. S[n], is sensing
sample of the PU channel that is independent and identically distributed Gaussian random
process in the nth sensing slot with zero mean and variance 𝜎S2. Similarly, Wi [n], is the
Additive White Gaussian Noise (AWGN) at the PU and ith user channel with zero mean
and variance, 𝜎W2
.
i
The sensing energy reported by the ith user to the FC in the total K sensing samples is
represented as;


K
Zi = |Ui [n]|2 (2)
| |
n=1

In Eq. (2),Ui [n] = PR,i hi Xi [n] + Ni [n], is the received signal at the FC which is reported
by the ith user in the nth sensing slot. Similarly,PR,i, is the ith cooperative user transmitting

13
Differential Evolution Based Reliable Cooperative Spectrum… 129

Fig. 3  The PDF of the normal and malicious sensing reports: a Normal, b OPPOSITE, c YES, d NO, e
RANDOM

power to the FC.hi, is the channel gain between the ith user and FC. The SU to FC reporting
channel noise is also assumed to be AWGN with variance,𝛿i2, and zero mean.
The FC decides globally about the PU activity based on the received signal notifications of
the M cooperative users and weighting coefficient vector as:


M
( )
Z= wi Zi (3)
i=1

here wi , is the weighting coefficient determined at the FC against the ith user sensing
employing DE. In Eq. (3), as the ith user reporting energy, Zi , is normally distributed,
therefore, the resultant, Z , is also assumed to be of the normally distributed nature as in
[37];

( ) ∑M
E Z ||H0 = 2
wi K𝜎0,i (4)
i=1

( ) ∑M
E Z ||H1 = 2
wi K𝜎1,i (5)
i=1

13
130 N. Gul et al.

Fig. 4  Proposed CSS model

( ) ∑M
var Z ||H0 = 2w2i K(𝜎0,i
2
⃗ T 𝛷H0 w
+ 𝛿i2 )2 = w ⃗ (6)
i=1

( ) ∑M
var Z ||H1 = 2w2i K(𝜎1,i
2 2 2
+ 𝜎0,i ⃗ T 𝛷H1 w
) =w ⃗ (7)
i=1

here 𝜎0,i
2
, and, 𝜎1,i
2
, are the variances of, Ui [n], under the H0 and H1 hypothesis made by the
ith user that are equivalent to, 𝜎0,i = PR,i ||hi || 𝜎W + 𝛿i2, and, 𝜎1,i = PR,i ||gi || ||hi || 𝜎s2 + 𝜎0,i ,
2 2 2 2 2 2 2
i
respectively. [ ]T
The optimal coefficient vector, w ⃗ = w1 w2 … wM , in Eqs. (4)–(7) is to be optimized
with the goal to determine appropriate threshold value, 𝛽 , that further leads to a minimum
sensing error.

13
Differential Evolution Based Reliable Cooperative Spectrum… 131

The covariance matrices based on the resultant hypothesis H0 and H1 are;


( )
(8)
4
𝛷H0 = diag 2K𝜎0,i

( )
𝛷H1 = diag 2K(PR,i |gi |2 |hi |2 𝜎S2 + 𝜎0,i
2 2
) (9)

where 𝛷H0 , and, 𝛷H1, are the square diagonal matrices in Eqs. (8) and (9). The detection
and false alarm probability results at the FC can be represented as;

⎛ ⎞ ⎛ ⎞


� ⎜ 𝛽 − E(Z ��H0 ) ⎟ ⎜ 𝛽−w ⃗ T 𝜇⃗0 ⎟
Pf = P Z > 𝛽 �H0 = Q⎜ � ⎟ = Q⎜ � ⎟ (10)
⎜ var(Z ��H0 ) ⎟ ⎜ w ⃗ T 𝛷H0 w ⃗⎟
⎝ ⎠ ⎝ ⎠

⎛ ⎞ ⎛ ⎞


� ⎜ 𝛽 − E(Z ��H1 ) ⎟ ⎜ 𝛽−w ⃗ T 𝜇⃗1 ⎟
Pd = P Z > 𝛽 �H1 = Q⎜ � ⎟ = Q⎜ � ⎟ (11)
⎜ var(Z ��H1 ) ⎟ ⎜ w ⃗ T 𝛷H1 w ⃗⎟
⎝ ⎠ ⎝ ⎠

� �
⎛ wT 𝛷H1 w𝜇0T w ⃗ H1 w𝜇1T w ⎞⎟
⃗ + w𝛷

𝛽=⎜ � � ⎟ (12)
⎜ wT 𝛷H0 w + wT 𝛷H1 w ⎟
⎝ ⎠

Let us assume that, Pf = Pm, where, Pm, is the misdetection probability and,
Pf = 1 − Pd , therefore, the total error probability, Pe, is determined as;

⎛ ⎞ ⎛ ⎞
⎜ 𝛽−w ⃗ T 𝜇⃗0 ⎟ ⃗ T 𝜇⃗1 − 𝛽 ⎟
⎜ w
Pe = Pf + Pm = Q⎜ � ⎟ + Q⎜ � ⎟ (13)
⎜ w ⃗ T 𝛷H0 w ⃗⎟ ⎜ w ⃗ T 𝛷H1 w⃗⎟
⎝ ⎠ ⎝ ⎠

In Eq. (13), it is noticeable that the error probability is highly depending on the selec-
tion of w⃗ . Therefore, optimal threshold, 𝛽 , determined using Eq. (12) substituted into Eq.
(13) will leads to a high detection, minimum false alarm and low� error probability results.
∑M 2
In the proposed technique, selection of the, w ⃗ , is 0 < wi < 1 and, w = 1, to reduce
i=1 i
selection of the search space.
In the PSO based scheme in [36], SUs sense PU spectrum and forward their energy sta-
tistics to the FC for a number of observations to form a PSO population. FC then applies
PSO technique in identifying sensing report that better resemble with the actual status of
the PU transmission. The FC measure the fitness scores under all sensing iterations and
declare the minimum outlying particle as the actual channel information of the PU for a
final decision. The problem in hand is to minimize the objective function, Pe (w), in Eq.
(13) by increasing the, Pd (w), and lowering, Pf (w).
The GA introduced by john Holland in 1974 is a computational approach, motivated
by evolution. GA is biological inspired method used in searching an optimized solution
to science and engineering problems [8–10, 37, 38]. The chromosomes are the strings of

13
132 N. Gul et al.

binary symbols that represent one candidate solution to the problem. The GA tries to find
the optimal set of weighted coefficient vector against the sensing reports received from all
cooperative users as in [10]. In the random normalized set of coefficient vector population
the vector with low error probabilities are elected as the optimal set of vector.

4 Proposed Differential Evolution Algorithm Based Weighting Method

DE is a population based algorithm with crossover, mutation and selection procedures [39,
40]. The difference between GA and DE in searching the best fitness is that GA relies on
crossover while DE is dependent on mutation. The DE algorithm also uses a non-uniform
crossover that can take child vector parameters of one parent often higher than others.
A non-uniform crossover enables DE to consider child vector parameters of one parent
often higher than others. The benefit with the use of DE is the global minimum identifica-
tion irrespective of the initial parameters, quick convergence to solution and less control
parameters.
One major advantage associated with DE is the self-adaptability introduced through
mutation and selection processes. Storm and Price suggested DE as population based sto-
chastic search algorithm for continuous functions optimization in [41].
In the proposed DE based CSS scheme, optimal coefficient vector is determined that
assign high weights to the sensing information of the normal users with priority assign-
ment in comparison with MUs. The channel availability is finalized with the DE using
weighted SDF scheme that results in low error probability. A flow chart of the proposed
model is demonstrated in Fig. 5.
The proposed DE algorithm finds optimal threshold and sensing weights as follows;
Step 1 Initial Population
A total number of, N , chromosomes are considered for the DE. The algorithm is initial-
ized with initial population of randomly generated, N , chromosomes, which consists, M ,
T
⃗ x = [ w1 w2 … wM ] , x ∈ 1, … , N .
total dimensions, i.e., w
wx = L + (H − L) × rand(N, M) (14)
In Eq. (14), L , is the lower and, H , upper limit of the coefficient matrix. The fitness of
each weight vector, Pe ( w ⃗ x ( 1 ) ), Pe ( w ⃗ x (N) ), is determined and the vector
⃗ x ( 2 ) ), … , Pe ( w
with minimum error probability and corresponding threshold value is determined as the
global best.
Step 2 Selection of the Random Numbers
In this step three dissimilar random numbers are generated.
Step 3 Mutation
Mutation operation is performed based on the initialized population, wx , and random
numbers, n1, n2 and n3 as;
wy = wx (n1 ) + F × ( wx (n2 ) − wx (n3 ) ) (15)
where wy, is a mutant or mutation vector. Due to this difference in Eq. (15), this algorithm
is called DE. The constant, F , selection is problem dependent and should be smartly placed
keeping the value of genes between L and H.
Step 4 Crossover
In this step, a crossover operation is performed using initial weights, wx , and mutant
vectors, wy, which show selection of the genes between wy and wx as;

13
Differential Evolution Based Reliable Cooperative Spectrum… 133

Fig. 5  The proposed model flowchart

13
134 N. Gul et al.

Fig. 6  Probability of error versus SNRs with YES category of MUs

{ }
wy if a ≤ CR or j = jrand
wz =
wx Otherwise (16)

here a, is a random number and, jrand , is a random integer from 1 to M and, 0 ≤ CR ≤ 1.


Step 5 Fitness of the Particles
The suitability of each coefficient vector of the,wz , is determined as
⃗ z ( 1 ) ), Pe (w
Pe ( w ⃗ z (N) ).
⃗ z ( 2 ) ), … , Pe ( w
Similarly, the vector with minimum error probability results is selected as the new
global best coefficient vector and the corresponding threshold as the new global best
threshold.
Step 6 New Population and Updating the Global Best Coefficient Vector
In this step, the fitness of each coefficient vector of the crossover over popula-
tion, wz , is compared with the initial population weighting matrix, wx , to update the
population.
{ }
wz Fitness(wz ) > Fitness(wx )
wx =
wx Otherwise (17)

Similarly, the global best coefficient vector, wz , is compared with, wx , to update the
global best coefficient vector and threshold.
Step 7 Stopping Criteria
The DE starts repeating step 2 if the fitness functions, i.e., minimum Pe is not
achieved or the given number of iterations is not met by the system. At the end of
iterations the final global best vector and optimal threshold results are returned by the
algorithm.
A pseudo code of the proposed DE based CSS technique is as below;

13
Differential Evolution Based Reliable Cooperative Spectrum… 135

13
136 N. Gul et al.

Fig. 7  Probability of error versus SNRs with NO category of MUs

5 Simulations and Results

In the simulations, parameters of the CRN are adjusted for 10 and 14 cooperative users.
The results are simulated in the varying SNRs environment, i.e., − 30 dB to 0 dB. The
sensing interval is selected 1 ms having 270 to 335 samples. Cooperative users placed
in different SNRs sense the PU channel independently. Population size of the GA, PSO
and DE consists of M genes in each N chromosome. The maximum size of the sensing

Fig. 8  Probability of error versus SNRs with OPPOSITE category of MUs

13
Differential Evolution Based Reliable Cooperative Spectrum… 137

iterations is kept 50. A crossover point is randomly selected in the range of 1 to M . The
GA, PSO and proposed DE scheme returns the optimum coefficient vector that is further
used in the global decision of the SDF at the FC.
Performance of the proposed DE based SDF scheme is simulated and compared with
the GA SDF, PSO SDF, KL SDF and MGC SDF schemes. The simulation environment is
categorized into four different cases by collecting the error probabilities. Error probabili-
ties are obtained against the varying SNRs with fixed total cooperative users and sensing
samples in case 1. Case 2 discusses the error probabilities against the varying total coop-
erative users for all combination schemes with fixed SNRs and sensing samples. Similarly,
case 3 shows the error probabilities against the varying number of sensing samples with
fixed SNRs and cooperative users. Case 4 discusses the probability of detection and the
false alarm probability in the presence of the OPPOSITE category of MUs under varying
SNRs, cooperative users and sensing samples.

Case 1

Case 1 shows the error probability results against the increasing SNRs values for 14 coop-
erative users and 270 sensing samples for the reporting users. Figure 6 shows the results for
the DE, PSO, GA, KL and MGC weighting schemes in the presence of YES category of
MUs that always reports busy status of the PU channel. The result shows minimum error
probability among all for the proposed DE scheme in the presence of YES MUs. It is also
clear from the results in Fig. 6 that the MGC scheme has the worst performance at the
participation of YES MUs in cooperation. Figure 7 illustrates the error probability results
under the same parameter settings as in Fig. 6 in the presence of NO category of MU. The
NO user reports availability of the PU channel irrespective of actual status. The NO behav-
ior of the cooperating user severely affects the sensing ability of the MGC, KL, PSO and
GA, while the DE shows less vulnerability to sensing errors caused by the NO. The results
collected in Fig. 8 are for OPPOSITE category of MUs in the cooperative environment.

Fig. 9  Probability of error versus cooperative users with YES category of MUs

13
138 N. Gul et al.

Fig. 10  Probability of error versus cooperative users with NO category of MUs

Fig. 11  Probability of error versus cooperative users with OPPOSITE category of MUs

The OPPOSITE user report FC with low energy statistics when the PU transmission is
active, while sending high energy statistics about the PU channel when there is opportunity
for SU transmission and misguide the FC about the channel status. DE shows minimum
error probability in sensing, while the PSO and GA schemes sensing performance is con-
siderably equal up to − 12 dB SNR. Likewise, any further increase in the SNRs lead GA to
dominate the performance of PSO. The MGC and KL schemes show worst performance in
the presence of OPPOSITE category of MU as compared with the soft computing methods.

13
Differential Evolution Based Reliable Cooperative Spectrum… 139

Fig. 12  Probability of error versus sensing samples with YES category of MUs

Fig. 13  Probability of error versus sensing samples with NO category of MUs

Case 2

Case 2 discuss error probabilities for the proposed and traditional combination schemes with
increased total cooperative users while keeping the average SNRs fixed at − 21.5 dB and 270
sensing samples. Error probabilities are collected in Figs. 9, 10 and 11 for the YES, NO and
OPPOSITE categories of MUs in cooperation. The results in Fig. 9 is collected to measure
sensing performance of the DE, PSO, GA, MGC and KL divergence schemes when YES user

13
140 N. Gul et al.

Fig. 14  Probability of error versus sensing samples with OPPOSITE category of MUs

participate in reporting their sensing data to the FC. These results demonstrate that the DE and
PSO error probabilities reduces significantly as compared with GA, MGC and KL schemes by
increasing total number of cooperative users. The results of the DE and PSO is followed by the
GA while the traditional MGC scheme shows poor performance as compared with other com-
bination schemes employed at the FC. A similar comparison is shown in Fig. 10 for these com-
bination schemes at the participation of NO users in CSS. The results in Fig. 10 demonstrate
that with the participation of NO user, DE error probability significantly reduces in comparison
with all other schemes. Similarly, the PSO and GA results are almost similar in this scenario,
while the KL scheme is showing better sensing performance as compared to traditional MGC
scheme. In continuation of Case 2 results, sensing reliabilities of these schemes were tested at
the participation of OPPOSITE users in Fig. 11. Figure 11 shows that with the contribution of
OPPOSITE users, sensing results achieved through DE is far more superior as compared to
all other schemes and produces little sensing error with increasing total number of cooperative
users. Similarly, in this case as was the case in Fig. 10, GA and PSO have similar result char-
acteristics that surpass MGC and KL divergence schemes. The MGC scheme in Fig. 11 shows
poor performance in sensing PU channel as compared to DE, PSO, GA and KL schemes.

Case 3

This case investigates sensing performance of cooperative users with increased sensing sam-
ples from 270 to 335. The system is tested for the three different categories of MUs, YES
user that report high energy statistics irrespective of the actual PU status, NO user always
report low energy sensing statistics irrespective of the regular PU activity and OPPOSITE
user that always negate actual PU channel activity. The average SNRs is kept constant at
− 21.5 dB and total cooperative users at 14. Error probabilities are plotted against varying
sensing samples in Figs. 12, 13 and 14 for the three different scenarios of the YES, NO and
OPPOSITE MUs. Figure 12 shows that with increasing sensing samples, the DE error prob-
ability results get further lower and achieves better sensing results as compared to PSO, GA,

13
Differential Evolution Based Reliable Cooperative Spectrum… 141

Fig. 15  Probability of detection versus probability of false alarm with OPPOSITE MUs at varying SNRs

Fig. 16  Probability of detection versus probability of false alarm with OPPOSITE MUs at varying coopera-
tive users

KL and MGC schemes. Similarly, in this part, the PSO scheme performance is also predom-
inantly improved as compared to GA, MGC and KL, while the GA, MGC and KL schemes
have shown less decrease in error probability with increasing the sensing samples size.
A similar comparison is illustrated in Fig. 13 with the contribution of NO users in
cooperation. The DE scheme shows better sensing results with minimum sensing errors

13
142 N. Gul et al.

Fig. 17  Probability of detection versus probability of false alarm with OPPOSITE MUs at varying sensing
samples

in Fig. 14 as compared to all other schemes. Figure 14 shows the error probabilities versus
increasing sensing samples that shows superiority of the DE in producing minimum sens-
ing error. DE is able to produce comparatively less sensing error even at low sensing sam-
ples. GA shows improved detection performance with minimum sensing error in Fig. 14 as
compared to PSO, MGC and KL schemes.

Case 4

In this case, results for the probability of detection against the false alarm probability is
collected in the presence of OPPOSITE category of MUs in different simulation environ-
ments in Figs. 15, 16 and 17.
Figure 15 shows the probability of detection against false alarm probability at the
participation of the OPPOSITE category of MUs under varying SNRs, fixed sensing
samples and total cooperative users. The results in Fig. 15 shows that the proposed DE
scheme has improved Receiver Operating Characteristic (ROC) results as compared to
all other schemes. DE results are next followed by the GA scheme, while the MGC
scheme has its worst performance in the varying SNRs environment. Similarly, Fig. 16
shows probability of detection against the probability of false alarm in the presence of
OPPOSITE category of MUs with varying the number of cooperative users from 10
to 22. In this case, the SNRs and sensing samples are kept constant. The proposed DE
based selection method has improved ROC in this case of varying total number of coop-
erative users as compared to Fig. 15. In comparison with Fig. 15 here the PSO scheme
shows high detection with minimum false alarm in comparison with GA, KL and MGC
schemes. Similarly, the detection and false alarm results are collected for the proposed
DE, GA, PSO, KL divergence and MGC schemes under varying sensing samples in

13
Differential Evolution Based Reliable Cooperative Spectrum… 143

Fig. 17. The total number of cooperative users and SNRs are kept constantfor the simu-
lation results in Fig. 17. The results shows that the proposed DE based CSS has better
performance with high detection probability and low false alarm results in comparison
to all other schemes. The results obtained for the traditional GA, PSO, KL divergence
and MGC schemes did not improved significantly in the detection probability against
false alarm probability in this case of varying sensing samples.

6 Conclusion and Future Work

The fading and shadowing effects of the Raleigh fading channel reduces sensing perfor-
mance of an individual user. The proposed DE based CSS is able to determine suitable
optimal coefficient vectors and adaptive threshold against the reporting users before
SDF scheme is allowed to take decision at the FC. The optimal coefficient vector is
able to demonstrate improved detection performance with minimum sensing error for
the proposed DE-SDF scheme compared to the PSO-SDF, GA-SDF, MGC-SDF and
KL-SDF schemes at different levels of SNRs, sensing samples and cooperative users
participation.
Reliability of all cooperative schemes, such as DE, PSO, GA, MGC and KL is
investigated in the presence of YES, NO and OPPOSITE categories of MUs. The DE-
SDF scheme surpassed the other competing schemes in authenticity and accuracy of
PU spectrum sensing. Further research can be carried out to investigate these schemes
for PUEA and LAZY categories of MUs in CSS. Similarly, reliability of the proposed
scheme can be analyzed in the presence of collusion attackers that share their attacking
information locally to device a more severe attack on the FC.

Compliance with Ethical Standards


Conflict of interest The authors declare no conflict of interest.

References
1. Haykin, S. (2005). Cognitive radio: Brain-empowered wireless communications. IEEE Journal on
Selected Areas in Communications, 23(2), 201–220.
2. Ghasemi, A., & Sousa, E. S. (2008). Spectrum sensing in cognitive radio networks: Requirements,
challenges and design trade-offs. Communications Magazine, IEEE, 46(4), 32–39.
3. Giupponi, L., Galindo-Serrano, A., Blasco, P., & Dohler, M. (2010). Docitive networks: An emerg-
ing paradigm for dynamic spectrum management [dynamic spectrum management]. IEEE Wireless
Commun., 17(4), 47–54.
4. Cabric, D., Mishra, S. M., & Brodersen, R. W. (2004). Implementation issues in spectrum sensing
for cognitive radios. In Conference record of the thirty-eighth asilomar conference on signals, sys-
tems and computers, 2004 (Vol. 1, pp. 772–776).
5. Tandra, R., & Sahai, A. (2005). Fundamental limits on detection in low SNR under noise uncer-
tainty. In 2005 international conference on wireless networks, communications and mobile comput-
ing (Vol. 1, pp. 464–469).
6. Hiep, K. V.-V. (2012). A robust cooperative spectrum sensing based on Kullback-Leibler diver-
gence. IEICE Transactions on Communications, E95–B(4), 1286–1290.
7. Sharifi, A. A., & Niya, J. M. (2016). Securing collaborative spectrum sensing against malicious
attackers in cognitive radio networks. Wireless Personal Communications, 90(1), 75–91.

13
144 N. Gul et al.

8. Mehboob, U., Qadir, J., Ali, S., & Vasilakos, A. (2016). Genetic algorithms in wireless networking:
Techniques, applications, and issues. Berlin: Springer.
9. Bhattacharjee, S. (2015). Optimization of probability of false alarm and probability of detec-
tion in cognitive radio networks using GA. In Proceedings of ReTIS’15—2nd IEEE international
conference on recent trends in information systems, Kolkata, India. https​://doi.org/10.1109/retis​
.2015.72328​52.
10. Akbari, M., & Ghanbarisabagh, M. (2014). A novel evolutionary-based cooperative spectrum sens-
ing mechanism for cognitive radio networks. Wireless Personal Communications, 79(2), 1017–
1030. https​://doi.org/10.1007/s1127​7-014-1915-8.
11. Fathy, M., Tammam, A., & Saafan, A. (2019). Influence of relaying malicious node within coopera-
tive sensing in cognitive radio network. Wireless Network, 5, 2449–2458.
12. Sharifi, A. A., Sharifi, M., & MuseviNiya, M. J. (2016). Collaborative spectrum sensing under pri-
mary user emulation attack in cognitive radio networks. IETE Journal of Research, 62(2), 205–211.
13. Abrardo, A., Barni, M. Kallas, K., & Tondi, B. (2018). Decision fusion with unbalanced priors
under synchronized byzantine attacks : A message-passing approach. In 2018 Asia-Pacific sig-
nal and information processing association annual summit and conference (APSIPA ASC) (pp.
1160–1167).
14. Gupta, N., Dhurandher, S. K., Member, S., & Sehgal, A. (2019). A contract theory approach-based
scheme to encourage secondary users for cooperative sensing in cognitive radio networks. IEEE Sys-
tems Journal 1–11.
15. Cao, X., Lai, L., & Member, S. (2019). Distributed gradient descent algorithm robust to an arbitrary
number of byzantine attackers. IEEE Transactions on Signal Processing, 67(22), 5850–5864.
16. Zhang, L., Ding, G., & Member, S. (2019). Detecting abnormal power emission for orderly spectrum
usage B. generalized likelihood ratio detector with unknown. IEEE Transactions on Vehicular Technol-
ogy, 68(2), 1989–1992.
17. Lee, S., Zhang, Y., Yoon, S., & Song, I. (2019). Order statistics and recursive updating with aging fac-
tor for cooperative cognitive radio networks under SSDF attacks. ICT Express, 2019, 1–4.
18. Sun, Z., Xu, Z., Hammad, M. Z., Ning, X., Wang, Q., & Guo, L. (2019). Defending against mas-
sive SSDF attacks from a novel perspective of honest secondary users. IEEE Communications Letters,
23(10), 1696–1699.
19. Grissa, M., Member, S., & Yavuz, A. A. (2017). Preserving the location privacy of secondary users
in cooperative spectrum sensing. IEEE Transactions on Information Forensics and Security, 12(2),
418–431.
20. Grissa, M., Yavuz, A., & Hamdaoui, B. (2016). An efficient technique for protecting location privacy
of cooperative spectrum sensing users. In 2016 IEEE conference on computer communications work-
shops (INFOCOM WKSHPS) (pp. 915–920).
21. Du, H., Fu, S., & Chu, H. (2015). A credibility-based defense SSDF attacks scheme for the expulsion
of malicious users in cognitive radio. International Journal of hybrid information Technology, 8(9),
269–280.
22. Sharifi, A. A. (2019). An effective and optimal fusion rule in the presence of probabilistic spectrum
sensing data falsification attack. Journal of Communication Engineering, 8(1), 78–92.
23. Chen, H., Zhou, M., Xie, L., & Li, J. (2017). Cooperative spectrum sensing with M-ary quantized
data in cognitive radio networks under SSDF attacks. IEEE Transactions on Wireless Communica-
tions, 16(8), 5244–5257.
24. Feng, J., Zhang, M., Xiao, Y., & Yue, H. (2018). Securing cooperative spectrum sensing against
collusive SSDF attack using XOR distance analysis in cognitive radio networks. Sensors (Switzer-
land), 18(2), 1–14.
25. Feng, J., Li, S., Lv, S., Wang, H., & Fu, A. (2018). Securing cooperative spectrum sensing against
collusive false feedback attack in cognitive radio networks. IEEE Transactions on Vehicular Tech-
nology, 67(9), 8276–8287.
26. Fu, Y., & He, Z. (2019). Bayesian-inference-based sliding window trust model against probabilistic
SSDF attack. IEEE Systems Journal, 12, 1–12.
27. Shrivastava, S., John, S., Rajesh, A., & Bora, P. K. (2018). Preventing collusion attacks in coop-
erative spectrum sensing. In International conference on signal processing and communications
(SPCOM) (pp. 90–94).
28. Zhao, F., Li, S., & Feng, J. (2019). Securing cooperative spectrum sensing against DC-SSDF attack
using trust fluctuation clustering analysis in cognitive radio networks. Wireless Communications
and Mobile Computing, 2019, 1–11.

13
Differential Evolution Based Reliable Cooperative Spectrum… 145

29. Wan, R., Ding, L., Xiong, N., & Zhou, X. (2019). Mitigation strategy against spectrum- sensing
data falsification attack in cognitive radio sensor networks. International Journal of Distributed
Sensor Networks, 15(9), 1–12.
30. Luo, Z., Zhao, S., Lu, Z., Xu, J., & Sagduyu, Y. E. (2019). When attackers meet AI : Learning-
empowered attacks in cooperative spectrum sensing. 1–15.
31. Gul, N., Qureshi, I. M., Omar, A., Elahi, A., & Khan, M. S. (2017). History based forward and
feedback mechanism in cooperative spectrum sensing including malicious users in cognitive radio
network. PLoS ONE, 12(8), e0183387.
32. Gul, N., Qureshi, I. M., Akbar, S., Kamran, M., & Rasool, I. (2018). One-to-many relationship
based kullbackLeibler divergence against malicious users in cooperative spectrum sensing. Wire-
less Communications and Mobile Computing, 2018, 1–14.
33. Gul, N., Qureshi, I. M., Elahi, A., & Rasool, I. (2018). Defense against malicious users in coopera-
tive spectrum sensing using genetic algorithm. International Journal of Antennas and Propagation,
2018, 1–11.
34. Gul, N., Naveed, A., Elahi, A., Khattak, T., & Qureshi, I. (2017). A combination of double sided
neighbor distance and genetic algorithm in cooperative spectrum sensing against malicious users.
In Proceedings of 2017 14th international bhurban conference on applied sciences & technology
(IBCAST) (pp. 746–753).
35. Gul, N., Mansoor, I., Aqdas, Q., Atif, N., & Imtiaz, E. (2019). Secured soft combination schemes
against malicious—users in cooperative spectrum sensing. Wireless Personal Communications,
0123456789, 1–20.
36. Akbari, M., Manesh, M. R., Saleh, A. A., & Ismail, M. (2012). Improved soft fusion based coop-
erative spectrum sensing using particle swarm optimization. IEICE Electronics Express, 9(6),
436–442.
37. Quan, Z., Cui, S., & Sayed, A. H. (2008). Optimal linear cooperation for spectrum sensing in cog-
nitive radio network. IEEE Journal of Selected Topics in Signal Processing, 2, 28–40m.
38. Zainab, A., & Sinha, P. (2016). A survey of cognitive radio reconfigurable antenna design and pro-
posed design using genetic algorithm. In IEEE students’ conference on electrical, electronics and
computer science, Bhopal, India (p. 1–6). https​://doi.org/10.1109/sceec​s.2016.75092​63.
39. Karaboga, D., & Okdem, S. (2004). A simple and global optimization algorithm for engineering
problems: differential evolution algorithm. Turkish Journal of Electrical Engineering & Computer
Sciences, 12(1), 53–60.
40. Veterstrom, J., & Thomsen, R. (2004). A comparative study of differential evolution, particle
swarm optimization, and evolutionary algorithms on problems. In IEEE congress on evolutionary
computation (pp. 980–987).
41. Storn, R., & Price, K. V. (1995). Differential evolution: A simple and efficient adaptive scheme
for global optimization over continuous spaces. ICSI, USA, Tech. Rep. TR-95-012, 1995 [Online].
Available: http://icsi.berke​ley.edu/∼storn​/liter​a.html.

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

Noor Gul was born in Peshawar, Pakistan in 1983. He received his


Ph.D degree in Electronics Engineering from the department of Elec-
trical Engineering, International Islamic University, Islamabad, Paki-
stan. He is currently working as Lecturer at the department of Elec-
tronics, University of Peshawar. His research interest include are
digital communication, cognitive radio network, machine learning,
statistical data analysis, artificial intelligence, optimization techniques
and signal processing.

13
146 N. Gul et al.

Ijaz Mansoor Qureshi was born in Rawalpindi, Pakistan in 1955. He


received his Ph.D. degree in HEP from University of Toronto, Canada.
He is currently Professor at Department of Electrical Engineering, Air
University, Islamabad. His research activities are in digital signal pro-
cessing, digital image processing, digital communication, soft comput-
ing, compressed sensing and heuristic optimization techniques.

Muhammad Sajjad Khan received his B.Sc. in Computer Information


Systems Engineering from University of Engineering and Technology
Peshawar, Pakistan in 2004, and his M.Engg. from Mehran University
of Engineering and Technology Jamshoro, Pakistan in 2007. He
received his Ph.D. degree from University of Ulsan, S. Korea. He is
currently pursuing his Post Doctorate from University of Ulsan, S.
Korea. His research interests include wireless communication, cogni-
tive radio networks and next generation wireless communication
systems.

Atif Elahi was born in Peshawar, Pakistan in 1982. He received his


Ph.D. degree in Electronics Engineering from the department of Elec-
trical Engineering, International Islamic University, Islamabad, Paki-
stan. He is currently Lecturer at Government Degree College Matra.
His research include are digital communication, signal processing,
heuristic techniques and compressed sensing.

13
Differential Evolution Based Reliable Cooperative Spectrum… 147

Sadiq Akbar was born in 1975 at District Charsadda, village Turang-


zai, KPK, Pakistan. He did his M.Sc. in Electronics from University of
Peshawar in 1998, and MS(CS) in Telecommunication & Networking
from Gandhara University Peshawar in 2010. He Completed his Ph.D.
in Electronics in 2018 from the department of Electronics, University
of Peshawar. Presently he is working as Assistant Professor in the
department of Electronics, University of Peshawar, KPK, Pakistan as
Assistant Professor. His area of research is parameter estimations, sys-
tem identification, and optimization methods.

Affiliations

Noor Gul1,3 · Ijaz Mansoor Qureshi2 · Muhammad Sajjad Khan1 · Atif Elahi1 ·
Sadiq Akbar3
Ijaz Mansoor Qureshi
imqureshi@mail.au.edu.pk
Muhammad Sajjad Khan
Khan.mcsl@gmail.com
Atif Elahi
atif.phdee40@iiu.edu.pk
Sadiq Akbar
sadiqakbar@uop.edu.pk
1
Department of Electrical Engineering, Faculty of Engineering and Technology, International
Islamic University, Islamabad 44000, Pakistan
2
Department of Electrical Engineering, Air University, Islamabad 44000, Pakistan
3
DepartmentofElectronics, University of Peshawar, Peshawar 25120, Pakistan

13

You might also like