You are on page 1of 6

Available online at www.sciencedirect.

com

ScienceDirect
ICT Express xxx (xxxx) xxx
www.elsevier.com/locate/icte

Robust spectrum sensing against malicious users using particle swarm


optimization
Noor Gula , Saeed Ahmedb , Su Min Kimc , Junsu Kimc ,∗
a Department of Electronics, University of Peshawar, Peshawar, Pakistan
bDepartment of Electrical Engineering, Mirpur University of Science and Technology, AJK, Pakistan
c Department of Electronics Engineering, Korea Polytechnic University, Gyeonggi-do, Republic of Korea

Received 10 September 2021; received in revised form 22 November 2021; accepted 21 December 2021
Available online xxxx

Abstract
The different issues in the wireless channel, such as multipath fading, shadowing, and hidden terminal problems, limit the sensing abilities
of the single user in a cognitive radio network (CRN). In this connection, it is more suitable to follow cooperative spectrum sensing (CSS).
However, the various categories of malicious users (MUs) pose threats to the performance of CSS. These MUs are eager to report false
sensing data to the fusion center (FC) to deceive its decision. This work considered the CSS performance in the presence of always yes MU
(AYMU), opposite MU (OMU), and always no MU (ANMU). Moreover, the considered particle swarm optimization (PSO) algorithm takes
out reports of the MUs using one to many sensing distances (OTMSD) and z-score outlier detection techniques before the final decision.
Simulation results validate the effectiveness of the proposed scheme in the presence of MUs.
© 2021 The Author(s). Published by Elsevier B.V. on behalf of The Korean Institute of Communications and Information Sciences. This is an open
access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Keywords: Cognitive radio network; Particle swarm optimization; Cooperative spectrum sensing; Malicious users

1. Introduction of the radio spectrum [7]. In this regard, spectrum sensing


has been legalized by the federal communication commission
The technological generations in wireless communications
(FCC) [8].
have played their roles to provide users a high data rates
The employment of cognitive radio (CR) in wireless com-
and minimum latency [1]. The evolution of 5G is expected
munication provides enough intelligence with efficient radio
to provide a vital contribution to public security, spectrum
spectrum utilization that learns and adjusts the communication
supervision, energy efficiency, and high data rates [2]. Internet
device’s parameters [9]. The primary users (PUs) are legal in
of Things (IoT) is the center of the 5G and 6G technologies.
cognitive radio networks (CRNs) to access the spectrum, while
Thus, the IoT-based devices are going to play an important
the secondary users (SUs) are opportunists in accessing the PU
role in the structure of the 5G and 6G networks [3].
The concept of IoT was introduced in [4] by Ashton as a spectrum. The SUs have to vacate the spectrum upon the PUs’
technological revolution. It can improve the landscape of nu- arrivals to avoid interference with the legitimate PUs.
merous industries and help in the improvised logistic learning, The uncertainties in the wireless channel like the fading,
intelligent transportation, automation, and health care units shadowing, and hidden terminal problems limit the individual
in [5]. The use of IoT without embedding cognition features user sensing performance [10]. Cooperative spectrum sensing
is a lake in intelligence [6]. Therefore, the rapid increase in (CSS) is considered a promising solution to overcome the
the number of devices and wireless technology demand new difficulties. In the CSS, the individual users sense and report
wireless services in both the utilized and underutilized portion their findings to the fusion center (FC) [11]. However, one
major problem in the CSS decision is the sensing reports of
∗ Corresponding author. malicious users (MUs), where false information is reported
E-mail addresses: noor@uop.edu.pk (N. Gul), to the FC to hijack the FC decision. Several studies are
saeed.ahmed@must.edu.pk (S. Ahmed), suminkim@kpu.ac.kr (S.M. Kim),
found in the literature to lower the abnormal sensing reports.
junsukim@kpu.ac.kr (J. Kim).
Peer review under responsibility of The Korean Institute of Communica- A recursive updating algorithm alleviates the effects of the
tions and Information Sciences (KICS). malicious attackers in [12]. The CR intelligently adapts the
https://doi.org/10.1016/j.icte.2021.12.008
2405-9595/© 2021 The Author(s). Published by Elsevier B.V. on behalf of The Korean Institute of Communications and Information Sciences. This is an
open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Please cite this article as: N. Gul, S. Ahmed, S.M. Kim et al., Robust spectrum sensing against malicious users using particle swarm optimization, ICT Express (2022),
https://doi.org/10.1016/j.icte.2021.12.008.
N. Gul, S. Ahmed, S.M. Kim et al. ICT Express xxx (xxxx) xxx

transmission parameters. Therefore, the use of artificial in-


telligence (AI) techniques is appropriate in CR technology.
A hybrid machine learning (ML) technique in [13] secure
the FC against malicious data. The collusion attackers are
tackled with the boosted tree ML scheme in [14]. In [15],
AI-based abnormality detection techniques are presented at
the physical layer in CR. In the equal gain combining (EGC)
and maximum gain combining (MGC), a global decision is
formed using the soft energy statistics of the SUs. Likewise,
in the hard combinations, users take their sensing decision
locally and report to the FC, where these decisions are united
in the logical OR, logical AND, and majority voting (MV)
schemes [16]. The genetic algorithm (GA) in [17] optimized
the detection and false alarm probabilities that precede a
reducing error probability. In [18], the GA handles the fitness
of the double-sided neighbor distance (DSND) to reduce the
error probabilities. Fig. 1. Centralized cooperative spectrum sensing environment.
This paper employs the particle swarm optimization (PSO)
algorithm to monitor the PU’s activity through the FC decision
enabling the FC to produce reliable sensing outcomes and and report their data to the FC. Based on the received sens-
mitigate the abnormal reports of the MUs. In the proposed ing notifications, the FC makes its global decision about the
scheme, the SUs at distinct terrestrial locations report their channel.
sensing data to the FC and employ the PSO algorithm to The binary hypothesis of the jth SU, which can be one of
get the most relevant outcome regarding channel availability. the SUs or the MUs, at the lth time slot is
{
The PSO fitness comprises a composite outlier score pre- H0 : n j (l)
pared using one to many sensing distances (OTMSD) and y j (l) = , (1)
H1 : h j s(l) + n j (l)
z-score (ZS) methods. The PSO selects the sensing report with
minimum out-of-the-way results in the PSO population on where H0 denotes the PU channel availability and H1 indicates
behalf of all cooperative SUs that adequately describes sensing the channel occupancy, n j (l) is the additive white Gaussian
reports’ reliability. The PSO selections are employed in the noise, h j is the channel gain, and s(l) is the PU’s signal. In
MGC, EGC, and MV schemes to produce the final decision. this work, we assume that the SUs perform energy detector-
The proposed schemes are validated at the participation of based sensing. The sensing period is divided into K received
the always-yes MUs (AYMUs), always-no MUs (ANMUs), samples. The energy measured by the jth SU in the ith sensing
and opposite MUs (OMUs). The AYMUs deliver an always period is as
⎧ l +K −1
high energy signal to the FC, while the ANMUs report lower i∑
⏐n j (l)⏐2 , H0

⎪ ⏐ ⏐
energy signals even if the PU channel is busy. Similarly, the ⎪


⎨ l=l
OMU senses the channel and reports opposite PU channel i
E i j = l +K −1 , (2)
information. ⎪ i∑
⏐2
⏐h j s(l) + n j (l)⏐ , H1

⎪ ⏐
The paper is divided into these sections. Section 2 explains ⎪


the system model. Section 3 specifies the interpretation of the l=li

PSO to attain accurate sensing data before the EGC, MGC, and where li denotes the starting sample of the ith sensing period.
MV techniques. Section 4 justifies the proposed scheme using The energy measured converges to the Gaussian random vari-
the simulation results, while Section 5 concludes this paper. able when the value of K is large enough according to the
central limit theorem and expressed as
{ (
2. System model N µ0 = K , σ02 = 2K , H0
)
Ei j ∼ , (3)
N µ1 = K (η j + 1), σ12 = 2K (η j + 1) , H1
( )
As the individual user sensing experiences disturbances
due to the wireless channel effects. Therefore, the cooperative where η j is the average signal-to-noise ratio (SNR) between
environment in Fig. 1 helps in reducing sensing problems jth SU and the PU. Similarly, (µ0 , σ02 ) and (µ1 , σ12 ) are the
experienced by the single user. The objective of this paper is mean and variance values of the energies reported under the
to reduce the false alarm P f and misdetection probabilities Pm H0 and H1 hypotheses.
that leads to a reduction in the error probability Pe = P f + Pm . The FC store the reports of all SUs in the N0 intervals and
Please note that Pm = 1 − Pd , where Pd denotes the detection try to find optimal sensing with the best fitness as minimum
probability. outlier score. The best selection is then employed in the
In Fig. 1, the reporting users consist of the normal SUs and final decision through the EGC, MGC, and majority voting
MUs (AYMUs, ANMUs, OMUs) that sense the PU channel schemes.
2
N. Gul, S. Ahmed, S.M. Kim et al. ICT Express xxx (xxxx) xxx
⏐ ⏐
⏐ 1 ∑M
where E ′ i j = ⏐ M−1 k=1,k̸= j E ik ⏐ indicates the average re-

ports of the other SUs’ while taking out the results of the jth
user.

Step 2.1: Outliers identification using OTMSD

The results in (4) and (5) determine the first outlying factor
based on the OTMSD d j (i) for the jth SU of the ith particle
as
d j (i) = ⏐ E i j − E i′ j ⏐ ,
⏐ ⏐
(6)
The total outlying score of the ith particle is determined as
M

d(i) = d j (i). (7)
j=1

The measurement in (7) is repeated for the N0 particles and


the results are collected as
d = [d(1) d(2) · · · d(N0 )]T . (8)
d in (8) is the outlier score results for all the N0 sensing
periods.
Fig. 2. Proposed CSS scheme based on PSO.
Step 2.2: Outliers identification using ZS

3. Proposed CSS scheme based on PSO In this step, ZS is employed as an outlier score measure-
ment as
The PSO algorithm in the paper has a fixed number of ⏐ E i j − µ(i) ⏐
⏐ ⏐
particles. Each particle is a vector with several elements. o j (i) = ⏐⏐ ⏐, (9)
These elements are randomly initialized that enter an iterative σ (i) ⏐
In (9), µ(i) = M1 j=1 E i j , and σ (i) denote the average
∑M
process. At the end of each iteration, the performance of each
particle is estimated in terms of its closeness to the objective and the standard deviation of the ith particle. o j (i) is the ZS
function by substituting in the objective function. Moreover, at outlying factor for the jth user in the ith period. The ZS
the end of every iteration, one of the particles comes with the measurements for all particles are accumulated as
best results. A flowchart of the proposed PSO-based scheme M

is shown in Fig. 2. Detailed steps for the proposed scheme are o(i) = o j (i). (10)
as follows. j=1

Hence, the total z-score result for the N0 particles are


3.1. Step 1: Sensing data collection
o = [o(1) o(2) · · · o(N0 )]T . (11)
The history matrix is formed at the FC that consists of the The suitability of particles selection using the results in (7) and
reported soft energies of the SUs in the N0 sensing periods as (10) is as
f (i) = d(i) + o(i). (12)
...
⎡ ⎤
E 11 E 12 E 1M
[ ] ⎢ E
⎢ 21 E 22 ... E 2M ⎥ Step 3: Changes in the PSO population
E = Ei j = ⎢ . .. .. .. ⎥, (4)

⎣ .. . . . ⎦
The particle with the minimum outlying score in E based
E N0 1 E N0 2 ··· E N0 M
on (12) is selected as the global best. Let P denote the local
where i = 1, . . . , N0 , j = 1, . . . , M. Here M is the total best particles initialized as P = E. Furthermore, let V denote
number of SUs, N0 is total number of sensing periods to form the velocity matrix with the same size of E, i.e., N0 × M. Then,
the population matrix. The FC further modifies the particles the position and velocities of all these particles are initialized
position in (4) as zero and modified as
. . . E ′ 1M
⎡ ′
E ′ 12

E 11 V(i+1) j = Vi j + C1 R1 (Pi j − E i j ) + C2 R2 (g j − E i j ), (13)
[ ] ⎢ ′
⎢ E 21 E ′ 22 . . . E ′ 2M ⎥
E′ = E i′ j = ⎢ . ⎥, where C1 and C2 are the learning coefficients that represent
.. .. .. (5)

⎣ .. . . . ⎦ the particles individually and social contribution, R1 and R2
E ′ N0 1 E ′ N0 2 · · · E ′ N0 M are uniformly distributed random numbers from 0 to 1. Please
3
N. Gul, S. Ahmed, S.M. Kim et al. ICT Express xxx (xxxx) xxx

note that g = [g j ] denotes the global best particle, and j =


1, . . . , M. The particle velocities are rounded as
{
max(V), Vi j > max(V)
V(i+1) j = . (14)
min(V), Vi j ≤ max(V)
Then, the particle position is updated as
E (i+1) j = E i j + V(i+1) j , (15)
where E (i+1) j denotes the updated population.

Step 4: Changes in the local best and the global best


Fig. 3. Pd vs. P f at SNRs1 and SNRs2 with fixed SUs for ANMU.

The local best particles are looked for any modification as


{
Ei , f (Ei ) < f (Pi )
Pi = , (16)
Pi , otherwise
where f (·) is the fitness function that denotes the selection
criteria in (12). The fitness of Pi in (16) is compared with the
g to look for any improvement as
{
Pi , f (Pi ) < f (g)
g= , i = 1, . . . , N0 . (17)
g, otherwise
The final global best particle is employed in the EGC, MGC,
and MV in the next section.
Fig. 4. Pd vs. P f at SUs1 and SUs2 with fixed SNRs for ANMU.

Step 5: Decisions
4. Numerical evaluation
The detection and false alarm probabilities of the EGC are The simulation is classified into three different cases: Case
1 exhibits the representation when 4 ANMUs are engaged
⎧ ⏐ ⎫
⎨1 ∑ M ⏐ ⎬
g j ≥ γ ⏐⏐ H1 , in sensing the PU channel; Case 2 shows the performance

Pd E GC = Pr
⎩M ⎭ of the schemes in the presence of 4 AYMUs; while Case 3
j=1 ⏐
⎧ ⏐ ⎫ (18) illustrates the results in the presence of the 4 OMUs. Some of
⎨1 ∑ M ⏐
the simulation parameters are listed in Table 1.

g j ≥ γ ⏐⏐ H0 .

P f E GC = Pr
⎩M ⎭
j=1 ⏐
4.1. Case 1
The detection and false alarm probabilities of the MGC are
⎧ ⏐ ⎫ Case 1 shows the detection and false probabilities for the
M ⏐
simple MV, simple EGC, simple MGC, proposed PSO-MV,
⎨∑ ⎬
w j g j ≥ γ ⏐⏐ H1 ,

Pd M GC = Pr
⎩ ⎭ PSO-EGC, and PSO-MGC in the attendance of ANMUs. The
j=1 ⏐
⎧ ⏐ ⎫ (19) results in Fig. 3 are obtained by soliciting first for the average
M
SNRs1 = −16 dB and then expanded to SNRs2 = −10 dB.
⎨∑ ⏐ ⎬
w j g j ≥ γ ⏐⏐ H0 .

P f M GC = Pr In Fig. 3, the total number of SUs is 12 with 4 ANMUs,
⎩ ⎭
j=1 ⏐
100 sensing iterations, 270 sensing samples, PSO population
where w j denotes the weight with the definition of w j = 20, and 1 ms sensing time. Fig. 4 exhibits the detection and
false alarm probabilities at two different numbers of SUs. The
η( j)/ M j=1 η( j). The detection and false alarm probabilities

number of sensing users was SUs1 = 11 which was raised to
of the MV schemes are as
⎧ ⏐ ⎫ SUs2 = 17, while the average SNRs is −14.5 dB.
M ⏐ The decisions in Figs. 3 and 4 depict that the PSO-based
⎨∑ M ⏐⏐ ⎬
Pd M V = Pr gj ≥ H1 , MV, EGC, and MGC schemes have more useful sensing results
⎩ 2 ⏐⏐ ⎭
j=1 with high detection and minimum false alarm probabilities in
⎧ ⏐ ⎫ (20)
M ⏐ contrast with the simple MV, EGC, and MGC schemes.
⎨∑ M ⏐⏐ ⎬
P f M V = Pr gj ≥ H0 . The results also reveal that sensing performance strongly

j=1
2 ⏐⏐ ⎭ relies on the SNRs in Fig. 3, where a moderate increase in the
4
N. Gul, S. Ahmed, S.M. Kim et al. ICT Express xxx (xxxx) xxx

Table 1
Simulation parameters.
Parameters Case-I Case-2 Case-3
MUs 4 ANMU 4 AYMU 4 OMU
Total user SUs1 = 11 SUs1 = 11 SUs1 = 11
SUs2 = 17 SUs2 = 17 SUs2 = 17
Average SNRs SNRs2 = −10 dB SNRs2 = −10 dB SNRs2 = −10 dB
SNRs1 = −16 dB SNRs1 = −16 dB SNRs1 = −16 dB
Iterations 100 100 100
Sensing samples 270 270 270
Population 20 20 20
Sensing time 1 ms 1 ms 1 ms

Fig. 5. Pd vs. P f at SNRs1 and SNRs2 with fixed SUs for AYMU. Fig. 7. Pd vs. P f at SNRs1 and SNRs2 with fixed SUs for OMU.

Fig. 6. Pd vs. P f at SUs1 and SUs2 with fixed SNRs for AYMU. Fig. 8. Pd vs. P f at SUs1 and SUs2 with fixed SNRs for OMU.

4.3. Case 3
SNRs commences to improvement in detection performance.
The increasing number of SUs in Fig. 4 from SUs1 = 11 to In case 3, the character of MUs is taken over by OMU. The
SUs2 = 17 has low impact on the sensing performance as detection and false alarm probability are collected in Figs. 7
compared with the SNRs. and 8 for the recommended and simple-MV, simple-EGC,
and simple-MGC schemes. The results show that the OMU
4.2. Case 2 severely degrades the sensing performance of the traditional
schemes compared with the proposed PSO-based schemes.
Case 2 shows the detection and false alarm probabilities at The results validated that the PSO-EGC and PSO-MGC
the participation of the AYMU class of MUs in Figs. 5 and 6. schemes have enhanced detection with limited false alarm in
The proposed scheme can produce better sensing performance comparison with the simple-EGC, simple-MGC, and simple-
in the presence of AYMU. The resemblance in the perfor- MV schemes.
mance of the schemes in the presence of AYMU and ANMU is
due to the similarity in their reporting nature,i.e., The AYMUs 5. Conclusions
always report high energy, and ANMUs report low energy In this paper, a PSO-based scheme is proposed to enable
signals with similar outlier behavior. FC’s decision considering reliable sensing information. The
5
N. Gul, S. Ahmed, S.M. Kim et al. ICT Express xxx (xxxx) xxx

PSO selects optimal sensing information based on the results [5] R. Khan, S.U. Khan, R. Zaheer, S. Khan, Future internet: The internet
of the OTMSD and ZS as a fitness. The PSO sensing iden- of things architecture, possible applications and key challenges, in:
tifications are further employed in the EGC, MGC, and MV Proc. - 10th Int. Conf. Front. Inf. Technol. FIT 2012, 2012, pp.
257–260.
schemes for making a global decision. This allows the FC [6] Q. Wu, G. Ding, Y. Xu, S. Feng, Z. Du, J. Wang, K. Long, Cognitive
decisions to be less prone to the participation of the AYMU, internet of things: A new paradigm beyond connection, IEEE Internet
ANMU, and OMU categories of MUs in all soft and hard Things J. 1 (2014) 129–143.
decisions. The simulation results validate the effectiveness of [7] A. Ghasemi, E.S. Sousa, Spectrum sensing in cognitive radio networks:
the proposed PSO-based scheme that has high detection and requirements, challenges and design trade-offs, Commun. Mag. IEEE
46 (2008) 32–39.
low false alarm probabilities at distinct levels of the SNRs and [8] A.O. Abdul Salam, R.E. Sheriff, S.R. Al-Araji, K. Mezher, Q.
the total number of SUs. Nasir, Adaptive threshold and optimal frame duration for multi-taper
spectrum sensing in cognitive radio, ICT Express 5 (2019) 31–36.
[9] S. Haykin, Cognitive radio: Brain-empowered wireless communica-
Declaration of competing interest tions, IEEE J. Sel. Areas Commun. 23 (2005) 201–220.
The authors declare that they have no known competing [10] E. Axell, G. Leus, E.G. Larsson, H.V. Poor, Spectrum sensing for
cognitive radio: State-of-the-art and recent advances, IEEE Signal
financial interests or personal relationships that could have Process. Mag. 29 (2012) 101–116.
appeared to influence the work reported in this paper. [11] Y. He, J. Xue, T. Ratnarajah, M. Sellathurai, F. Khan, On the
performance of cooperative spectrum sensing in random cognitive
Acknowledgments radio networks, IEEE Syst. J. 12 (2018) 881–892.
[12] S. Lee, Y. Zhang, S. Yoon, I. Song, Order statistics and recursive
This work was supported in part by the MSIT (Ministry updating with aging factor for cooperative cognitive radio networks
of Science and ICT), Korea, under the ITRC (Information under SSDF attacks, ICT Express 6 (2020) 3–6.
Technology Research Center) support program (IITP-2021- [13] N. Gul, S.M. Kim, S. Ahmed, M.S. Khan, J. Kim, Differential
evolution based machine learning scheme for secure cooperative
2018-0-01426) supervised by the IITP (Institute for Informa- spectrum sensing system, Electron. 10 (2021) 1–29.
tion & Communications Technology Planning & Evaluation) [14] N. Gul, M.S. Khan, S.M. Kim, J. Kim, A. Elahi, Z. Khalil, Boosted
and in part by the National Research Foundation of Ko- trees algorithm as reliable spectrum sensing scheme in the presence
rea (NRF) funded by the Korea government (MSIT) (No. of malicious users, Electron. 9 (2020) 1–23.
2021R1A2C1013150). [15] A. Toma, A. Krayani, M. Farrukh, H. Qi, L. Marcenaro, Y. Gao,
C.S. Regazzoni, AI-based abnormality detection at the PHY-layer of
cognitive radio by learning generative models, IEEE Trans. Cogn.
References Commun. Netw. 6 (2020) 21–34.
[16] N. Marchang, R. Rajkumari, S.B. Brahmachary, A. Taggu, Dynamic
[1] A. Agarwal, G. Misra, K. Agarwal, The 5th generation mobile wireless decision rule for cooperative spectrum, in: 2015 IEEE Int. Conf. Electr.
networks- key concepts, network architecture and challenges, Am. J. Comput. Commun. Technol., 2015, pp. 1–5.
Electr. Electron. Eng. 3 (2015) 22–28. [17] S. Bhattacharjee, Optimization of probability of false alarm and
[2] T.Q. Duong, N.S. Vo, Editorial: Wireless communications and probability of detection in cognitive radio networks using GA, in:
networks for 5G and beyond, Mob. Netw. Appl. 24 (2019) Proc. ReTIS’15-2nd IEEE Int. Conf. Recent Trends Inf. Syst., Kolkata,
443–446. 2015, pp. 53–57.
[3] W. Ejaz, A. Anpalagan, M.A. Imran, M. Jo, M. Naeem, S. Bin Qaisar, [18] N. Gul, A. Naveed, A. Elahi, T. Khattak, I. Qureshi, A combination of
W. Wang, Internet of Things (IoT) in 5G wireless communications, double sided neighbor distance and genetic algorithm in cooperative
IEEE Access 4 (2016) 10310–10314. spectrum sensing against malicious users, in: Proc. 2017 14th Int.
[4] K. Ashton, Internet of Things: In the Real World, Things Matter More Bhurban Conf. Appl. Sci. Technol., Islamabad, Pakistan, 2017, pp.
than Ideas, Springer, Berlin, Germany, 2009. 746–753.

You might also like