An Adaptive Grey Wolf Optimization Algorithm for

Secrecy Rate Optimization in Interference Limited

Wireless Networks
Md. Samiur Rahman Md. Monzurul Haque
Department of Electrical and Electronic Engineering Department of Electrical and Electronic Engineering
Islamic University of Technology Islamic University of Technology
Gazipur, Bangladesh Gazipur, Bangladesh
[email protected] [email protected]

Zubayer Kabir Eisham Mohammad Tawhid Kawser

Department of Electrical and Electronic Engineering Department of Electrical and Electronic Engineering
Islamic University of Technology Islamic University of Technology
Gazipur, Bangladesh Gazipur, Bangladesh
[email protected] 0000-0002-8165-3606

Mohammad Rubbyat Akram Samin Ziyan Rahman

Technology Division Department of Electrical and Computer Engineering
Robi Axiata Ltd. Virginia Polytechnic Institute and State University
Dhaka, Bangladesh Virginia, USA
[email protected] [email protected]

Abstract—Throughout the revolving generations of cellular modeled. In the field of network security, choosing minimum
technologies, data security has been one of the biggest concerns. power, based on the required secrecy rate, is one such objective
In an interference-limited wireless network, this security concern function. Solving this function may efficiently help at the
becomes quite vital due to the intervention of eavesdroppers
in the network. As a result, the max-min secrecy throughput transmitter’s end to transmit at low power and at the same
problem appears to be one of the most significant optimization time, ensure the required secrecy rate. This problem is known
problems in the fields of wireless communication and network as max-min secrecy throughput optimization [6].
security. Nature-inspired optimization algorithms appear to be Deterministic algorithm has high computational time com-
quite vital tools for this kind of optimization problem. In pared to stochastic algorithm. So, solving such computation-
this paper, a problem specific adaptive version of the Grey
Wolf Optimization Algorithm has been used to solve this max- ally heavy objective function with stochastic algorithm can
min throughput problem and the performance of the proposed provide the transmitter enough useful information within the
algorithm has been compared with the existing methods and with limited time to take the optimum decision.
a few existing meta-heuristic algorithms. The balance between Swarm-based algorithms are famous in the optimization
the exploration and exploitation phase was controlled to enhance genre. Among many swarm-based algorithms Grey Wolf
the convergence speed to yield the optimal solution in the lowest
possible time. Optimization(GWO) algorithm [7] is famous for its unique
Index Terms—Wireless Communication, Secrecy, Data Security, cognitive component, modeled after the hunting nature of grey
Meta-heuristic, Optimization Algorithm, GWO, AGWO wolf. Any algorithm’s performance can be enhanced for a
particular problem by changing different parameters, such as,
I. I NTRODUCTION exploration-to-exploitation coefficient, applying Chaos equa-
Stochastic algorithms are really efficient in solving tion [8]in place of different random number, and so on.
multi-dimensional function–uni-modal or multi-modal. These In this work, GWO algorithm’s exploration-to-exploitation
stochastic algorithms are useful in the cellular communication, coefficient was changed for this problem, and its performance
specially in the era of fifth-generation (5G) cellular technology. in finding the optimal power in max-min secrecy through-
Many such complex functions in 5G are modeled for dynamic put optimization has been compared in terms of computa-
cellular environment. For example, in spectral management tional time and convergence curve with other famous swarm-
[1], resource allocation [2], wireless caching [3], edge com- based algorithms, namely, Particle Swarm Optimization (PSO)
puting [4], and network security [5], many such complex [9], Whale Optimization Algorithm (WOA) [10], and Harris
functions, which are also known as objective functions, are Hawks Optimization Algorithm (HHO) [11].
 II. O BJECTIVE F UNCTION the Omega (ω). The flow of the solution set is dictated by the
The objective function is modeled after a interference- position of the Alpha, Beta and Delta. Initial solution set is
limited wireless network consisting of M number of users randomly allocated; however, solution set changes after each
where receiver a receives information from transmitter a, iteration based on the following equations modeled after grey
and both transmitter and receiver consist of single antenna. wolf [7].
The system-model is depicted in “Fig. 1” [6]. The user In “(2)”, X vector represents the position vector of a search
agent. The ω layer search agents – modeled after wolves–
update their positions according to the position of the α, β,
δ wolves to encircle the the solution that gives the desired
objective function value – modeled after prey of wolves.
The approximate position of the desired prey is calculated or
assumed by the positions of the α, β and δ wolves.
The encircling behavior is mathematically represented by
these equations.

D = |Xp (t).C − X(t)|

(2)
X(t + 1) = Xp (t) − A.D

In “(2)”, Xp denotes the position of the target or the prey, X

denotes the position of the wolf, D is the distance between
Fig. 1. Model of the system. that wolf and the prey, t denotes the number of the current

Pa Gaa
 iteration, and the values of vectors A, C are determined by
data rate is given by χa (P ) = log2 1 + , where the following equations.
N0 + Ia
Pa indicates transmitted power of a − th transmitter, Gab
gives the channel gain from transmitter a to receiver b, and A = 2ar1 ; C = 2r2 (3)
XM
Ia = Pa Gab and P is the transmit power vector and
b=1 In “(3)”, r1 ,r2 are random vectors whose values are in [0, 1],
N0 is the power of the noise. If there is an unwanted user and the value of a starts from 2 and goes to 0 by decreasing
in the system with single-antenna, similar data rate equation linearly according to the “(4)”.
can be developed. The transmitter a and the single-unwanted
user or eavesdropper has a certain data
 rate between  them

t

Pa Gua a=2 1− (4)
and it is given by Xa (P ) = log2 1 + where T
N0 + Iau
XM
Iau = Pa Gub and Gub gives the channel gain from the In “(4)”, T is the maximum number of iteration.
b=1 To emulate the hunting mechanism, assumption that the
receiver b to the eavesdropper, so the definition of the secrecy current alpha wolf, beta wolf, delta wolf know better about
rate of transmitter a is ζa (P ) = max {χa (P ) − Xa (P ), 0}, the location of the potential prey was made, so their solutions
and the whole equation is given as follows. are saved and the positions of the other wolves according
max ζa (p) = min [χa (p) − Xa (P )] to their positions (the best search agents) are updated. The
P a=1,...,M mathematical model is in the “(5)”.
(1)
s.t. 0 ≤ Pa ≤ Pamax , ∀a = 1, ...M
X1 + X2 + X3
X(t + 1) =
3
III. G REY W OLF O PTIMIZATION
X1 = Xα − A1 Dα
A. Normal GWO
(5)
This meta-heuristic algorithm was modeled after the social
X2 = Xβ − A2 Dβ
hierarchy and the hunting behavior of grey wolf. The fittest
solution is called the Alpha(α), and followed on from that,
the second and third best solutions are called the Beta(β) and X3 = Xδ − A3 Dδ
the Delta(δ) respectively. The rest of the solutions are called
The value of D in “(5)” are given by “(6)”. the algorithm gets more number of iteration to improve the
result, and this AGWO also gives the best solution in less
Dα = |Xα C1 − X|
time which gives this method an advantage.
This can be achieved by many number of equations that
Dβ = |Xβ C2 − X| increase the exploration phase. In this work, “(7)” was used.
(6) 
 k1 !k2

t
Dδ = |Xδ C3 − X| a = 2 1 −  k1 = 12, k2 = 2 (7)
T

In “Fig. 3”, for first 250 iterations, “(7)” gives the highest value
of a, so it gives the algorithm the opportunity to explore more,
Grey wolves hunt with the pack. Their positions are updated
and the stopping criterion condition isn’t met during the first
according to the positions of α, β and δ wolves. The wolves
few iteration. This improves the result and at the same time
diverge when they have to go to search for a better prey or a
AGWO does this in short amount of time.
more possible prey, and they converge – modeled after their
pursuing for the kill. A depicts this behaviour, and The range
of A depends on the value of a. When |A| < 1, the wolves
move to attack the prey. When |A| > 1, the wolves diverge to
go for a better prey.i.e. better solution.
Floswchart of the GWO is in “Fig. 2”.

Fig. 3. Exploration to exploitation graph of the eqautions.

IV. R ESULTS AND A NALYSIS

An interference limited wireless network has been assumed
and modeled for the purpose of simulation and analysis.
The channel parameters and other network and simulation
properties have been kept as it was done in [12] which can be
found from Table I. Here is to mention that “(8)” expresses
Fig. 2. Flowchart of GWO. the Users Gain Matrix, i.e, channel gains among the users
themselves.
 
B. Adaptive GWO 0.4310 0.0002 0.0129 0.0011
In the main GWO algorithm, the convergence factor de- 0.0002 0.3018 0.0005 0.0031

0.2605 0.0008 0.4266 0.0099
 (8)
creases linearly from 2 to 0. In this particular problem, max-
min secrecy rate, the stopping criteria is 10−6 . So, when the 0.0039 0.0054 0.1007 0.0634
difference of the two consecutive α solution, one from current “(9)” expresses the Eavesdropper Gain Matrix, a term
iteration and other from previous iteration, goes below the related with the channel gains among the users and Eaves-
stopping criteria, the number of desired α solution or Flag dropper.
value is increased, and based on the user’s total number of
α solution or flag, the search for optimum solution will be 4.1454 3.6180 6.4587 4.9546


(9)
terminated.
So, when exploration phase is higher, meaning a is of high The performance of proposed Adaptive Grey Wolf Al-
value, the search agents changes their value with large enough gorithm was evaluated on the max-min secrecy throughput
difference, so they don’t fall under the stopping criteria. Hence, problem, and the effectiveness of the proposed algorithm was
compared to the existing path following algorithm and three TABLE II
other single objective meta-heuristic optimization algorithms, M INIMUM S ECRECY T HROGHPUT ( IN BPS /H Z )
namely Particle Swarm Optimization (PSO), Whale Opti-
mization Algorithm (WOA), and Harris Hawks Optimization Path Following Algorithms 1.3481
(HHO). Because of the stochastic nature of nature-inspired
meta-heuristic algorithms, all simulations were run 100 times PSO 1.2639
and the mean results were obtained to ensure data integrity.
The maximum number of iterations and the number of search
WOA 1.3522
agents were fixed at 500 and 30, respectively. The simulations
were performed in MATLAB 2018b running in Windows
10, 64-bit operating system on intel i3 8th generation with HHO 1.3176
RAM size 12 GB. The AGWO algorithm’s superiority can be
solemnly demonstrated by the attained outcome. AGWO 1.3101

TABLE I
C HANNEL PARAMETERS

complexity among all algorithms since it needs to invoke

Number Of Users 4
the solution of a convex optimization problem. On the other
hand, the proposed AGWO needed 16 iterations to converge
Noise Power 10−6 W to the optimal solution, but with comparatively a lot higher
convergence speed than the path following algorithm. Despite
Minimum Transmit Power 10−10 W the fact that other meta-heuristic algorithms converge at a
faster rate with less computational complexity than the path
following technique, AGWO’s convergence speed or comput-
Maximum Transmit Power 10−3 W
ing time outperforms all others. The AGWO needed only
0.0054 second to converge to the optimal solution in 16
iterations while even the closest computational time which
The minimization function of “(1)” was optimized using was provided by WOA (0.0075s) is significantly higher for
the proposed Adaptive Grey Wolf Algorithm to yield the a scenerio of only 4 users. For more number of users in the
minimum secrecy throughput where the search agents updated network, the optimization problem becomes more complex and
their positions (which is the transmit power for our problem) thus the AGWO appears to be the most effective algorithm
based on the position of the leader alpha wolf in every iteration in this problem in terms of dealing with the computational
to reach to the optimal value until the stopping criteria was complexity. For the computational time assessed for AGWO
satisfied. The stopping condition was defined in such a way versus other algorithms, Wilcoxon’s Ranksum [13] test was
that the execution would come to a halt if the difference in used, and the findings matched the significance and superiority
values in four consecutive iterations fell below a particular of AGWO over all other methods.
threshold (10−6 ). Table II represents the results of minimum
secrecy throughput found by different algorithms in bps/Hz. TABLE III
The results lie within a close proximity but the suggested C OMPUTATIONAL C OMPLEXITY
Adaptive Grey Wolf Method clearly gives the best optimal
outcome for minimum secrecy throughput, beating the path Algorithm Name Number of Iterations Computational Time
following algorithm and other single objective optimization
algorithms. Although the Particle Swarm Algorithm provides
PSO 16 0.0271s
lower value than AGWO in this respect, the standard deviation
of PSO for this case is quite significant as evident from
Table IV, which is not the case for AGWO. WOA 31 0.0075s
The most overwhelming performance shown by the pro-
posed algorithm is its excessively lower computational time. HHO 28 0.0149s
Table III denotes the average number of iterations and the
computational time needed for each algorithm to yield the
AGWO 16 0.0054s
solution by meeting proper stopping criteria. The results reveal
that, among all existing algorithms, the suggested AGWO
has a significant advantage in terms of exceptionally low
computational complexity for the following problem. With the The standard deviations of the findings for different runs
given channel parameters, the path following algorithm needed were assessed in order to determine the degree of stability and
12 iterations but with comparatively the highest computational integrity of the achieved results for lowest secrecy throughput
discovered by AGWO and other algorithms. which is repre- algorithms. As a result, the AGWO can be considered as an
sented in Table IV. The results also demonstrate AGWO’s extremely important tool for wireless networks having numer-
superior performance in terms of stable outputs or results, ous users in a multipath enviornment for efficient, optimal and
since AGWO produces the second-most stable results when fastest computaion for the mentioned problem.
compared to other algorithms, behind only WOA by a slight
R EFERENCES
margin..
[1] G. P. Koudouridis and P. Soldati, ”Spectrum and Network Den-
sity Management in 5G Ultra-Dense Networks,” in IEEE Wireless
TABLE IV Communications, vol. 24, no. 5, pp. 30-37, October 2017, doi:
S TANDARD D EVIATION (STD) 10.1109/MWC.2017.1700087.
[2] H. Halabian, ”Distributed Resource Allocation Optimization in 5G
Virtualized Networks,” in IEEE Journal on Selected Areas in
PSO WOA HHO AGWO Communications, vol. 37, no. 3, pp. 627-642, March 2019, doi:
10.1109/JSAC.2019.2894305.
[3] G. Paschos, E. Bastug, I. Land, G. Caire and M. Debbah, ”Wireless
0.0537 0.0105 0.0812 0.0236 caching: technical misconceptions and business barriers,” in IEEE
Communications Magazine, vol. 54, no. 8, pp. 16-22, August 2016,
doi:10.1109/MCOM.2016.7537172.
[4] Y. Mao, C. You, J. Zhang, K. Huang and K. B. Letaief, ”A Survey
“Fig. 4” represents the convergence curves of each algo- on Mobile Edge Computing: The Communication Perspective,” in IEEE
rithms in the comparison list, for an individual run. The con- Communications Surveys & Tutorials, vol. 19, no. 4, pp. 2322-2358,
Fourthquarter 2017, doi: 10.1109/COMST.2017.2745201.
vergence curve shows the high convergence speed of AGWO [5] I. Ahmad, S. Shahabuddin, T. Kumar, J. Okwuibe, A. Gurtov and M.
comparing with other algorithms in less number of iterations Ylianttila, ”Security for 5G and Beyond,” in IEEE Communications
with a stable result. Surveys & Tutorials, vol. 21, no. 4, pp. 3682-3722, Fourthquarter 2019,
doi: 10.1109/COMST.2019.2916180.
[6] Z. Sheng, H. D. Tuan, A. A. Nasir, T. Q. Duong and H. V. Poor, ”Power
Allocation for Energy Efficiency and Secrecy of Wireless Interference
Networks,” in IEEE Transactions on Wireless Communications, vol. 17,
no. 6, pp. 3737-3751, June 2018, doi: 10.1109/TWC.2018.2815626.
[7] S. Mirjalili, S. M. Mirjalili, and A. Lewis, “Grey wolf optimizer,”
Advances in engineering software, vol. 69, pp. 46–61, Mar. 2014
[8] Saremi, S., Mirjalili, S. & Lewis, A. Biogeography-based optimisa-
tion with chaos. Neural Comput & Applic 25, 1077–1097 (2014).
https://doi.org/10.1007/s00521-014-1597-x
[9] J. Kennedy and R. Eberhart, ”Particle swarm optimization,” Proceedings
of ICNN’95 - International Conference on Neural Networks, 1995, pp.
1942-1948 vol.4, doi: 10.1109/ICNN.1995.488968.
[10] Seyedali Mirjalili, Andrew Lewis, The Whale Optimization Algorithm,
Advances in Engineering Software, Volume 95, 2016,Pages 51-67, ISSN
0965-9978, https://doi.org/10.1016/j.advengsoft.2016.01.008.
[11] Ali Asghar Heidari, Seyedali Mirjalili, Hossam Faris, Ibrahim
Aljarah, Majdi Mafarja, Huiling Chen, Harris hawks optimization:
Algorithm and applications, Future Generation Computer
Systems, Volume 97,2019, Pages 849-872, ISSN 0167-
739X,https://doi.org/10.1016/j.future.2019.02.028.
[12] Q. Pham, S. Mirjalili, N. Kumar, M. Alazab and W. Hwang, ”Whale
Optimization Algorithm With Applications to Resource Allocation in
Wireless Networks,” in IEEE Transactions on Vehicular Technology, vol.
69, no. 4, pp. 4285-4297, April 2020, doi: 10.1109/TVT.2020.2973294.
[13] J. Hajek and Z. Sidak, “Theory of Rank Tests,” Academic Press, New
York, 1967.

Fig. 4. Convergence Curve.

V. C ONCLUSION
This paper provides a performance evaluation of an Adap-
tive Grey Wolf Algorithm (AGWO) in max-min secrecy
throughput problem in an interference limited wireless net-
work. The results signify the superiority of AGWO over the
existing path following algorithm and other 3 meta-heuristic
algorithms, specially in terms of computational complexity
and convergence speed. The adaptive algorithm has a faster
convergence due to its longer stay in exploration phase, which
resulted in an optimal result of max-min secrecy throughput
problem with relatively much less computational time. The
result also yields greater stability comparing to most of the