Swarm and Evolutionary Computation 35 (2017) 26–40

Contents lists available at ScienceDirect

Swarm and Evolutionary Computation

journal homepage: www.elsevier.com/locate/swevo

An efficient side lobe reduction technique considering mutual coupling MARK

effect in linear array antenna using BAT algorithm
⁎
Avishek Dasa, D. Mandala, , S.P. Ghoshalb, R. Kara
a
Department of Electronics and Communication Engineering, NIT Durgapur, 713209, India
b
Department of Electrical Engineering, NIT Durgapur, 713209, India

A R T I C L E I N F O A BS T RAC T

Keywords: This paper presents an investigation of mutual coupling effect among the array elements in a symmetric linear
Linear array antenna array antenna with the aim of reducing the side lobe level and the null control for the radiation pattern synthesis
Side Lobe Level (SLL) using BAT Algorithm. PSO and DE optimization techniques are also adopted for the sake of comparison and to
First Null Beam Width prove the superiority of BAT algorithm based design. Reduced side lobe level and null control, with and without
Mutual Coupling
considering the mutual coupling effect in the cost function have been achieved by an optimum perturbation of
BAT algorithm
the array elements' current excitation amplitude weights and the inter-element spacing among the array
elements. The results are also compared with those of a uniform reference array having equal number of
elements with λ inter-element spacing. The approach proposed in this paper is a generic one and can be easily
2
applied to any type of symmetrical linear arrays having any number of elements. Five different design examples
are presented and their performances are studied to illustrate the capability of BAT algorithm based approach
over those of PSO and DE.

1. Introduction contrast to each other as an array with narrow beam width usually
does not produce lower side lobe levels and vice-versa [4,5], i.e., the
Modern communication system uses a variety of techniques to performance cannot be improved appreciably for one aspect without
improve the signal quality and to reduce the interference. Its perfor- compromising with the other. In many applications it becomes
mance significantly depends on the efficient design of antenna array [1] essential to sacrifice the gain and beam width in order to attain a
and it is extensively used in satellite, wireless, mobile and radar lower side lobe level. Also the rising pollution of the electromagnetic
communication systems. It also improves the system performance by environment has impelled the placing of nulls in undesired directions.
enhancing the directivity, improving the signal quality, extending the So it is required to design the antenna array with low side lobe levels
system coverage and ever-increasing the spectrum efficiency. It is while maintaining the fixed beam width and placing the nulls in the
composed of an assembly of radiating elements in an electrical or undesired directions [6–8].
geometrical arrangement. In most of the cases the elements of an array The performance of an antenna array is affected by the presence of
are alike. This is not necessarily true but is often expedient, simple and mutual coupling between the array elements [4]. But the effect of
more sensible. One can use either the array geometry or the array mutual coupling has not been considered in any of the literature
weights (amplitude and phase of the signal) to set the performance of reported so far. For the first time the authors are introducing this effect
the array. The total field of an array is calculated by the vector addition in this paper. The driving impedance of antenna depends on three
of the fields radiated by all individual elements [2]. parameters namely, current excitation weights, mutual impedance and
Antenna array has a supreme importance in detecting and dealing the self impedance. It has also been examined by induced emf method
with signals arriving from different directions. So antenna array with that the parameters are not directly related to each other [1]. Hence,
narrow first null beam width (FNBW) is preferred to obtain a high the decrement of any one of the parameter will decrease the value of
directivity. On the other hand, the system needs to uphold a low side the driving impedance of an antenna which is very crucial when
lobe level (SLL) to avoid the interference with the other systems attempting to match the antenna.
operating in the same frequency band [3]. This paper introduces a new method to compensate the effect of
The above mentioned prerequisites of SLL and FNBW are in mutual coupling among the array elements by reducing the driving

⁎
Corresponding author.
E-mail address: [email protected] (D. Mandal).

http://dx.doi.org/10.1016/j.swevo.2017.02.004
Received 5 March 2016; Received in revised form 20 December 2016; Accepted 14 February 2017
Available online 17 February 2017
2210-6502/ © 2017 Elsevier B.V. All rights reserved.
A. Das et al. Swarm and Evolutionary Computation 35 (2017) 26–40

admittance of each identical array elements. The summation of all

identical array elements' driving admittance is reduced effectively by
controlling the current excitation weights of all the individual array
elements by using different optimization techniques. The driving
admittance, which is the reciprocal of the driving impedance, is called
the mutual coupling among the array elements. It is an undesired effect
appears in the antenna array and degrades the performance of the
array signal processing [2].
In this paper, a symmetrical linear antenna array structure with the
optimum current excitation weights and inter-element spacing is
designed keeping the objective of suppressing the SLL and placing
the nulls in the undesired directions and also minimizing the mutual
coupling among the array elements.
This involves a non-monotonic deviation in antenna element
patterns which has become an enormously complex optimization
problem [31]. The classical optimization methods cannot solve such
complex optimization problem [6,9]. Classical optimization methods
have a number of disadvantages such as (i) highly reactive to
preliminary points when the number of solution variables and hence, Fig. 1. 2M-elements linear array geometry along the Z axis.

the size of the solution space increases; (ii) recurrent convergence to

local optimum solution or divergence or revisiting the same sub- Section 5 concludes the paper.
optimal solution; (iii) Necessity of continuous and differentiable cost
function (gradient search method [10]); (iv) necessity of the piecewise 2. Design equation
linear cost approximation (linear programming) and (v) obscurity of
convergence and algorithm complexity (non-linear programming). For An even number of isotropic array elements 2M (where M is an
the optimization of such complex, nonlinear, irregular and non- integer) are located symmetrically along the z- axis, as shown in Fig. 1.
differentiable array factor diverse heuristic search evolutionary tech- The partition between the elements is d and M elements are located on
niques such as GA [9,11], ABC [21] and some variants of PSO [12–16] each side of the origin. Assuming that the amplitude excitation is
were adopted. Various heuristic search evolutionary algorithm techni- balanced about the origin, the Array Factor (AF) for the given non-
ques are also used in the different fields of engineering and system uniform amplitude linear broadside array is given by (1).
modeling for the solution of complex, nonlinear, irregular and non- ⎛ 2n − 1⎞
M
differentiable optimization problems [34–54]. (AF )2M = 2 ∑ In cos [ ⎜ ⎟ . k. d . cos θ + ϕn]
⎝ 2 ⎠ (1)
In this paper, the authors have considered the constraints to design n =1

the linear array antenna in the presence and in the absence of mutual where ϕn =phase weight of nth element and is considered to be zero for
coupling effect. The constraints are: the design; In =excitation amplitude weight of nth element; 2M= total
number of elements in the array; k=propagation constant; d=spacing
• reduced side lobe levels, between the elements; θ =angle of radiation of electromagnetic plane
• deeper nulls at the interference region, wave.
• lower value of mutual coupling. All antenna elements are assumed to be isotropic only. The
excitation amplitude of each element and the inter-element spacing
It is achieved by optimizing the current excitation weights and the are used to vary the antenna radiation pattern. So this is an optimiza-
inter-element spacing between the array elements. The phase differ- tion problem and the Cost Function (CF) for suppressing the side lobe
ence between any two elements is kept at zero. The simulation results level with a null control in a desired direction in the presence of mutual
show the comparative analysis of PSO, DE and BAT algorithm based coupling is given by (2).
approaches for the linear antenna array synthesis and their perfor-
mance, for suppressing the side lobes and placing the nulls in an AF (θms1, Imi ) + AF (θms2, Imi ) 1
CF = W1 × + W2 × ( )
interference region in the presence and in the absence of the mutual AF (θ0, Imi ) ZD
coupling term in the cost function. + W3 × ∑ AF (ϕk ) 2
The main objective of this paper is to achieve the lowest SLL and k (2)
placing the nulls in the interference region keeping a comparatively where ;
lower value of FNBW of the array. For the first time the authors are
introducing the effect of mutual coupling in the cost function con- M ⎛ ∑M I ⎞
ZD = ∑ ⎜ i =1 mi ⎟
sidered in this paper. The performance of BAT algorithm based design ⎜ I ⎟
m =1 ⎝ mi ⎠
results is compared with PSO and DE algorithms to demonstrate the
effectiveness and superiority of BAT algorithm in terms of the accuracy Here, W1, W2 and W3 are the weighting factors and θ0 is the value of θ
and the convergence speed. where the crest of the main lobe is attained; θms1 is the angle of the
The complexity of an algorithm is determined by the maximum maximum side lobe in the lower band and θms2 is the angle of the
number of primitive operations that a program executes to obtain the maximum side lobe in the upper band; Imi is the current excitation
best solution. So, the complexity of an algorithm depends on the amplitude weight of mith element, and φk is the direction of the desired
execution time of the program which is known as time complexity. The angle for imposing nulls.
time complexity of BAT algorithm is also compared with PSO and DE The first term on the right hand side (RHS) of (2) is used for
algorithms to emphasize the superiority of BAT algorithm. suppressing the SLL; the second term of (2) is used to consider the
The rest part of the paper is organized as follows: Section 2 presents mutual coupling effect and the third term is used for the null control. In
the design equation of linear array antenna. In Section 3, the employed this paper, a symmetrical linear antenna array structure with optimum
algorithms for the design of linear array antenna are discussed. Section current excitation weights and inter-element spacing is designed
4 deals with the simulated and the statistical results and analysis. keeping the objective of suppressing the SLL and placing the nulls in

27
A. Das et al. Swarm and Evolutionary Computation 35 (2017) 26–40

the undesired directions and also minimizing the mutual coupling 3.2. Differential Evolution (DE) algorithm
effect among the array elements. BAT algorithm controls the ampli-
tudes of the current excitations and inter-element spacing among the DE algorithm is another stochastic search method of solving
array elements to minimize the cost function. multidimensional optimization problems. The basic concept of DE
algorithm is discussed in details in [20,22],. The main steps of DE
algorithm for the linear array antenna optimization problem are given
3. Evolutionary techniques employed
as a flowchart in Fig. 3.

Evolutionary algorithms are characterized as stochastic, adaptive

3.3. BAT algorithm
and learning in order to construct intelligent optimization schemes.
The optimization techniques employed individually for this study are
BAT algorithm is another stochastic search method used for solving
PSO, DE and BAT. The simulation results in this paper show the
multidimensional optimization problems. The basic concept of BAT
optimal set of current excitation weights and the inter-element spacing
algorithm is discussed in details in [23–26]. The main steps of BAT
between the array elements using PSO, DE and BAT algorithms. It is
algorithm [27–29] for the linear array antenna optimization problem
found that BAT algorithm is the finest optimizer among other relevant
are given as a flowchart in Fig. 4.
techniques, not only in the convergence speed but also in the optimal
performances of the synthesized linear array antenna.
4. Results and discussions

3.1. Particle Swarm Optimization (PSO) MATLAB simulation has been performed extensively for PSO, DE
and BAT algorithm to obtain the normalized current excitation weights
PSO algorithm is a well known evolutionary technique capable of and inter-element separations among the array elements for 12, 13, 14,
solving multidimensional optimization problems. The basic concepts of 15 and 16 elements, respectively, placed on the both sides with respect
PSO algorithm were discussed in details in [17–19]. The main steps of to the origin of the array.
the PSO algorithm employed for the linear array antenna optimization The best control parameters [33] for PSO, DE and BAT algorithms
problem are given as a flowchart in Fig. 2. are used for the different antenna array designs. The population size

Start

Specify the parameters of PSO; Maximum iteration Cycle (G) and desired accuracy; Set
iteration count K=1

Create a random initial population in D-dimensional search space within their allowable range

Evaluate the fitness of each particle

Yes
Current fitness of the Pbest = Current fitness
swarm < Pbest of the swarm

K = K+ 1 No

Yes
Current fitness of the swarm g best = Current fitness of
< gbest the swarm

No

Update the velocity and position of each particle

No
K > G and accuracy < desired
accuracy

Yes

Stop

Fig. 2. Flowchart for PSO algorithm.

28
A. Das et al. Swarm and Evolutionary Computation 35 (2017) 26–40

Start

Specify the parameters of DE, Maximum iteration Cycle (G) and desired accuracy. Set iteration
count K = 1.

Create a random initial population in D-dimensional search space within their allowable range

Set target vector

Evaluate the fitness of target vector

Generate mutant vector by mutation operation

K = K +1

Generate trial vector by cross over operation

Evaluate the fitness of trail vector

The best vector survives by selection operation

No
K > G and accuracy <
desired accuracy

Yes

Stop

Fig. 3. Flowchart for DE algorithm.

and the maximum iteration cycles are kept fixed for all these three compared and are presented in Tables 2–11.
different optimization techniques, the values are 50 and 500, respec- Table 1 shows the results of the uniformly excited linear arrays
tively. In order to obtain the best results near to the global best, each keeping the inter-element spacing between the array elements as λ
2
algorithm is run for 50 times in MATLAB 7.5 version on Intel core when 12,13,14,15 and 16 isotropic array elements are placed, respec-
(TM) 2 processor @ 3.00 GHZ with 3 GB RAM to get the best solution. tively, on both sides with respect to the origin of the symmetric linear
The best results are reported in this paper. array structure. The SLL values of the array structures are
−13.2108 dB, −13.2190 dB, −13.2378 dB, −13.2449 dB and
4.1. Computational results −13.2592 dB, respectively, whereas the FNBW values are 9.72°, 9.0°,
8.28°, 7.56°, 7.2° and the mutual coupling values are 0.0035, 0.0030,
The optimal results are computed by BAT algorithm and are 0.0026, 0.0022 and 0.0020, respectively.
compared with those of the uniformly excited and uniformly spaced Table 1 shows the inverse relationship between the number of array
λ elements of a uniformly excited linear array with the SLL, FNBW and
(d = 2 ) reference array pattern and also with the results obtained by
using PSO and DE for the same array structure. Table 1 shows the SLL the mutual coupling values for λ inter-element spacing among the array
2
and the FNBW values for uniformly excited (Imi =1) linear array with elements.
d=λ/2 inter-element spacing in the presence of mutual coupling. The results of Table 1 are graphically plotted as the reference
The best optimized values of current excitations for the design of patterns for the radiation pattern plots of 24, 26, 28, 30 and 32 array
linear antenna array with the reduced side lobe level and moderately elements. As all the elements of the array are uniformly excited with a
λ
narrow beam width with the null control (75–77° and 105–107°) on fixed inter-element spacing (d = 2 ) so the reference pattern plot will be
the radiation pattern in the presence and in the absence of mutual the same for both the cases irrespective of the mutual coupling effect in
coupling with three different algorithms (PSO, DE and BAT) are the cost function.

29
A. Das et al. Swarm and Evolutionary Computation 35 (2017) 26–40

Start

Specify the parameters of BAT, Maximum iteration cycle (G) and desired accuracy. Set iteration count K =1

Initialize BAT population

Evaluate the fitness function

Yes
K > G and accuracy <
Stop
desired accuracy
No

Generate new solution and update velocities and solutions

Yes No
Rand1(0,1) > pulse
rate

Select best solution and generate a Generate a new solution

local solution arbitrarily

Rand2(0,1)< Loudness & No

Present fitness < previous Yes
fitness

Accept the new solution

Rank the BATs and find the current best, K=K+1

Fig. 4. Flowchart for BAT algorithm.

4.1.1. In the presence of Mutual Coupling

Table 1
Table 2 depicts the comparative results of the non-uniformly
Initial value of SLL and FNBW for uniformly excited (Imi = 1) and uniformly spaced excited 24 elements symmetrical linear antenna array structure con-
λ
(d = ) linear array.
2
sidering the mutual coupling effect in the cost function, achieved by
using PSO, DE and BAT algorithms. The table contains the normalized
No. of SLL (dB) FNBW Mutual Inter-element
values of the current excitation weights, SLL, FNBW, mutual coupling,
element (deg) Coupling Spacing (λ)
inter-element spacing between the array elements and the execution
24 −13.2108 9.72 0.0035 0.5 time obtained by using PSO, DE and BAT algorithms. The SLL value for
26 −13.2190 9 0.0030 0.5 λ
the uniformly excited and uniformly spaced (d = 2 ) 24-elements
28 −13.2378 8.28 0.0026 0.5
reference array antenna is −13.2108 dB. The value of SLL improves
30 −13.2449 7.56 0.0022 0.5
32 −13.2592 7.2 0.0020 0.5 to −31.9715 dB and −33.1501 dB by using PSO and DE algorithm,
respectively. By using BAT algorithm far better results of SLL of
−34.7682 dB is achieved. The FNBW value of 10.80° is achieved for
all the cases obtained by using PSO, DE and BAT algorithm. The

Table 2
Current excitations, SLL, FNBW, inter-element spacing and Cost values achieved by using PSO, DE, BAT for 24-element array in the presence of mutual coupling.

Algorithm Current excitation SLL (dB) FNBW Mutual Coupling Inter-element Cost value Execution time
(deg) spacing (λ) (s)

PSO 0.6073 0.6060 0.5805 0.5347 0.4654 0.4320 −31.9715 10.80 0.00078342 0.7342 0.8442 1.0021
0.3347 0.2857 0.2301 0.1660 0.1464 0.0841
DE 0.5798 0.5499 0.5209 0.5145 0.4219 0.4044 −33.1501 10.80 0.00074318 0.7511 0.6632 0.9462
0.3146 0.2667 0.1959 0.1455 0.1222 0.0733
BAT 0.4172 0.4122 0.3749 0.3453 0.3020 0.2644 −34.7682 10.80 0.00031369 0.8289 0.4727 0.8892
0.2025 0.1651 0.1269 0.0953 0.0494 0.0209

30
A. Das et al. Swarm and Evolutionary Computation 35 (2017) 26–40

Table 3
Current excitations, SLL, FNBW, inter-element spacing and Cost values achieved by using PSO, DE, BAT for 26-element array in the presence of mutual coupling.

Algorithm Current excitation SLL (dB) FNBW (deg) Mutual Coupling Inter-element spacing (λ) Cost value Execution time (s)

PSO 0.4658 0.4521 −31.8074 10.80 0.00070556 0.6805 0.8614 1.2409

0.4228 0.3892
0.3812 0.3260
0.2703 0.2372
0.1900 0.1631
0.1251 0.0983
0.0658

DE 0.3994 0.4061 −33.9207 10.44 0.00019855 0.7387 0.6904 1.1275

0.3788 0.3551
0.3126 0.2771
0.2376 0.2034
0.1472 0.1132
0.0955 0.0611
0.0155

BAT 0.5834 0.5805 −35.6843 10.44 0.00043416 0.7546 0.5175 0.9923

0.5509 0.4830
0.4240 0.4274
0.3340 0.2819
0.2105 0.1705
0.1168 0.0764
0.0484

mutual coupling value of 24-element reference array is 0.0035. The using PSO, DE and BAT algorithms. The SLL value for the uniformly
λ
optimized value of mutual coupling for the PSO, DE and BAT excited and uniformly spaced (d = 2 ) 26-element reference array
algorithms are 0.00078342, 0.00074318 and 0.00031369, respectively antenna is −13.2190 dB. The value of SLL improves to −31.8074 dB
for the inter-element spacing of 0.7342 λ, 0.7511 λ and 0.8289 λ. The and −33.9207 dB by using PSO and DE algorithms, respectively. Now
execution times of PSO and DE algorithm are 1.0021 and 0.9462 s by using BAT algorithm far better results of SLL of −35.6843 dB is
whereas for the same iteration cycle and population size BAT algorithm achieved. The authors have obtained 10.80° of FNBW by using PSO
takes only 0.8892 s. The execution time shows that the time complexity algorithm. The FNBW value decreases to 10.44° by using DE and BAT
of BAT algorithm is less in compared to PSO and DE algorithm. The algorithms. The mutual coupling value of 26-element reference array is
results of Table 2 confirm that BAT algorithm yields the lowest value of 0.0030. The optimized values of the mutual coupling are reduced to
SLL as compared with those of the SLLs obtained by using PSO and 0.00070556, 0.00019855 and 0.00043416 by applying PSO, DE and
DE, keeping the same FNBW value. BAT algorithms, respectively for the inter-element spacing of 0.6805 λ,
The results obtained by using PSO, DE and BAT algorithms for non- 0.7387 λ and 0.7546 λ. Table 3 confirms that BAT algorithm yields the
uniformly excited 24 elements symmetrical linear antenna array lowest value of SLL as compared with those of the SLLs obtained by
structure in the presence of the mutual coupling effect in the cost using PSO and DE, keeping the FNBW value nearly the same. The
function are graphically plotted in Fig. 5. In Fig. 5 the plots are results of Table 3 also show BAT algorithm takes much lower execution
compared with a reference pattern obtained by adopting uniform time as compared to PSO and DE algorithms to complete the full
excitation to all the array elements having a half wavelength inter- iteration cycle for the same population size. Hence the time complexity
element spacing. The maximum peak value of the SLL using uniform of BAT algorithm is lesser as compared to PSO and DE.
current distribution is −13.2108 dB. The peak value of SLL reduces to The results of Table 3 are graphically represented in Fig. 6 which
−31.9715 dB, −33.1501 dB and −34.7682 dB, respectively, for PSO, confirms a substantial reduction in SLL with the optimal non-uniform
DE and BAT algorithms, respectively. Fig. 5 reveals that BAT algorithm current excitations weights, as compared with the case of uniform
yields the lowest SLL value as compared with PSO and DE, keeping the current excitations and spacing. The maximum peak of the SLL using
FNBW values nearly the same. uniform current distribution is −13.2190 dB. The peak values of SLL
Table 3 shows the comparative results of PSO, DE and BAT are −31.8074 dB, −33.9270 dB and −35.6843 dB, respectively, for
algorithms for non-uniformly excited 26 elements symmetrical linear PSO, DE and BAT algorithms. From Fig. 6 it is confirmed that BAT
antenna array structure considering the mutual coupling effect in the algorithm yields the lowest SLL as compared with those of PSO and
cost function. The table contains the normalized values of the current DE, keeping almost the same FNBW value.
excitation weights, SLL, FNBW, mutual coupling, inter-element spa- PSO, DE and BAT algorithm based results are compared in Table 4
cing between the array elements and the execution time obtained by which shows the optimized values of non-uniformly excited 28

Table 4
Current excitations, SLL, FNBW, inter-element spacing and Cost values achieved by using PSO, DE, BAT for 28-element array in the presence of mutual coupling.

Algorithm Current excitation SLL (dB) FNBW Mutual Inter-element Cost value Execution
(deg) Coupling spacing (λ) time (s)

PSO 0.6478 0.6886 0.5987 0.5672 0.5804 0.4869 0.4033 0.3662 −31.2520 10.80 0.0005133 0.6336 0.9469 1.2019
0.3242 0.3007 0.1913 0.1731 0.0960 0.0791
DE 0.5002 0.5168 0.4782 0.4399 0.4257 0.3717 0.3242 0.3029 −34.2589 10.44 0.00062165 0.6593 0.6648 1.1339
0.2432 0.1939 0.1429 0.1307 0.0788 0.0735
BAT 0.6285 0.6175 0.5876 0.5568 0.4863 0.4472 0.3884 0.3152 −35.7787 10.44 0.00010967 0.7253 0.4639 1.0359
0.2501 0.1892 0.1759 0.0918 0.0703 0.0148

31
A. Das et al. Swarm and Evolutionary Computation 35 (2017) 26–40

Table 5
Current excitations, SLL, FNBW, inter-element spacing and Cost values achieved by using PSO, DE, BAT for 30-element array in the presence of mutual coupling.

Algorithm Current excitation SLL (dB) FNBW Mutual Coupling Inter-element Cost value Execution time
(deg) spacing (λ) (s)

PSO 0.5721 0.5452 0.5374 0.5224 0.4762 0.4362 0.4236 0.3296 −32.7978 10.44 0.00042693 0.6083 0.8175 1.2263
0.3107 0.2312 0.2177 0.1679 0.1185 0.1059 0.0648
DE 0.5041 0.5435 0.4811 0.4631 0.4437 0.3866 0.3198 0.2922 −33.7336 10.80 0.000093164 0.6479 0.7087 1.1663
0.2590 0.2171 0.1561 0.1237 0.0675 0.0597 0.0121
BAT 0.6547 0.6855 0.6466 0.5556 0.5585 0.4771 0.4578 0.3782 −35.6877 10.68 0.00020612 0.6689 0.5595 1.0855
0.3072 0.2366 0.1901 0.1562 0.1157 0.0607 0.0341

elements symmetrical linear antenna array structure, taking the mutual array antenna is −13.2449 dB. The value of SLL improves to
coupling effect into account the cost function. The table contains the −32.7978 dB and −33.7336 dB by using PSO and DE algorithm,
normalized values of current excitation weights, SLL, FNBW, mutual respectively. By using BAT algorithm far better results of SLL of
coupling, inter-element spacing between the array elements and the −35.6877 dB is achieved. The authors have also obtained FNBW of
execution times obtained by using PSO, DE and BAT algorithms. The 10.44° by using PSO algorithm. The FNBW values for DE and BAT
λ
SLL value for the uniformly excited and uniformly spaced (d = 2 ) 28- algorithms are 10.80° and 10.68°, respectively. The mutual coupling
element reference array antenna is −13.2378 dB. The value of SLL value of 30-element reference array is 0.0022. The optimized values of
improves to −31.2520 dB and −34.2589 dB by using PSO and DE the mutual coupling reduce to 0.00042693, 0.000093164 and
algorithm, respectively. By using BAT algorithm far better results of 0.00020612 by applying PSO, DE and BAT algorithms, respectively,
SLL is achieved as −35.7787 dB. The authors have also obtained 10.80° for the inter-element spacing of 0.6083 λ, 0.6479 λ and 0.6689 λ. The
FNBW by using PSO algorithm. The FNBW value decreases to 10.44° execution times for PSO and DE algorithms to complete the full
by using DE and BAT algorithms. The mutual coupling value of 28- iteration cycle are 1.2263 and 1.1663 s, respectively. The BAT algo-
element reference array is 0.0026. The optimized values of the mutual rithm takes only 1.0855 s to complete the same iteration cycle with the
coupling reduced to 0.0005133, 0.00062165 and 0.00010967 by same population size. Hence the results depict that the time complexity
applying PSO, DE and BAT algorithms, respectively for the inter- of BAT algorithm is lesser as compared to PSO and DE algorithms.
element spacing of 0.6336 λ, 0.6593 λ and 0.7253 λ. The results of Table 5 confirms that BAT algorithm based results yield the lowest
Table 4 confirm that BAT algorithm yields the lowest value of SLL as value of SLL as compared with those of the SLLs obtained by using PSO
compared with those of the SLLs obtained by using PSO and DE, and DE, keeping the same value of FNBW.
keeping the FNBW values the same. The results of Table 4 also depict Fig. 8 shows the graphical plots of the comparative results of the
that PSO and DE algorithms take 1.2019 and 1.1339 s, respectively, to optimized values of non-uniformly excited 30 elements symmetrical
complete the full iteration cycle whereas, BAT algorithm takes only linear antenna array structure. The graphical plots show a substantial
1.0359 s for the same population size and iteration cycle. So the results reduction of SLL value of the reference pattern with the optimal non-
show the time complexity of BAT algorithm is far better than PSO and uniform current excitation weights. The maximum peak value of the
DE algorithms. SLL using uniform current distribution is −13.2449 dB. The peak
The graphs of the radiation patterns for the tabulated results of values of SLL reduce to −32.7978 dB, −33.7336 dB and −35.6877 dB
Table 4 are shown in Fig. 7 which justifies a substantial reduction in for PSO, DE and BAT algorithms, respectively. From Fig. 8 it is evident
SLL with the optimal non-uniform current excitation weights, as that BAT algorithm leads to the lowest SLL value as compared with
compared with the case of uniform current excitations and spacing. those of the results obtained by using PSO and DE while keeping the
The maximum peak of the SLL using uniform current distribution is FNBW value almost the same.
−13.2378 dB. The peak values of SLL reduce to −31.2520 dB, Table 6 depicts the comparative results of the non-uniformly
−34.2589 dB and −35.7787 dB for PSO, DE and BAT algorithms, excited 32 elements symmetrical linear antenna array structure con-
respectively. Fig. 7 clearly shows that BAT algorithm yields the lowest sidering the mutual coupling effect in the cost function achieved by
value of SLL as compared with PSO and DE, keeping almost the same using PSO, DE and BAT algorithms. The table contains the normalized
value of FNBW. values of current excitation weights, SLL, FNBW, mutual coupling,
The comparative results of PSO, DE and BAT algorithms for non- inter-element spacing between the array elements and the execution
uniformly excited 30 elements symmetrical linear antenna array times obtained by using PSO, DE and BAT algorithms. The SLL value
λ
structure, considering the mutual coupling effect in the cost function, for the uniformly excited and uniformly spaced (d = 2 ) 32 elements
are shown in Table 5. The table contains the normalized values of reference array antenna is −13.2592 dB. The value of SLL improves to
current excitation weights, SLL, FNBW, mutual coupling, inter-ele- −32.6971 dB and −33.7741 dB by using PSO and DE algorithm,
ment spacing between the array elements and the execution time respectively. By using BAT algorithm far better results of SLL of
obtained by using PSO, DE and BAT algorithms. The SLL value for the −36.1407 dB is obtained. The FNBW value obtained by using PSO
λ
uniformly excited and uniformly spaced (d = 2 ) 30-element reference and DE algorithm techniques is same and is equal to as 10.80°. The

Table 6
Current excitations, SLL, FNBW, inter-element spacing and Cost values achieved by using PSO, DE, BAT for 32-element array in the presence of mutual coupling.

Algorithm Current excitation SLL (dB) FNBW Mutual Coupling Inter-element Cost value Execution time
(deg) spacing (λ) (s)

PSO 0.3658 0.3622 0.3625 0.3259 0.2949 0.2973 0.2390 0.2335 −32.6971 10.80 0.0000009647 0.6126 0.8495 1.2966
0.1798 0.1566 0.1542 0.1008 0.0718 0.0489 0.0460 0.0001
DE 0.7014 0.6791 0.6780 0.6418 0.6115 0.5645 0.5058 0.4447 −33.7741 10.80 0.00035665 0.5750 0.7079 1.2107
0.3795 0.3425 0.2927 0.2369 0.2076 0.1014 0.1611 0.0756
BAT 0.8120 0.7817 0.7401 0.7314 0.6579 0.6375 0.5288 0.5060 −36.1407 9.0 0.00031070 0.7284 0.5663 1.1219
0.4296 0.3580 0.2754 0.2406 0.2159 0.1315 0.0736 0.0715

32
A. Das et al. Swarm and Evolutionary Computation 35 (2017) 26–40

Table 7
Current excitations, SLL, FNBW, inter-element spacing and Cost values achieved by using PSO, DE, BAT for 24-element array in the absence of mutual coupling.

Algorithm Current excitation SLL (dB) FNBW Inter-element Cost Value Execution time
(deg) spacing (λ) (s)

PSO 0.5578 0.5688 0.5348 0.4769 0.4272 −34.4866 10.80 0.8013 0.7679 1.1293
0.3575 0.3177 0.2232 0.1850 0.1310 0.0948 0.0457
DE 0.6637 0.6280 0.6146 0.5454 0.4897 −35.1556 10.44 0.8153 0.5753 0.9966
0.4298 0.3471 0.2540 0.2014 0.1568 0.0908 0.0464
BAT 0.6627 0.6598 0.6080 0.5451 0.4649 −37.1382 10.80 0.9074 0.4003 0.8902
0.3651 0.3128 0.2128 0.1560 0.0970 0.0559 0

value of FNBW reduces to 9° by applying BAT algorithm. The mutual algorithm takes only 0.8902 s. The execution time shows that the time
coupling value of 32-element reference array is 0.0020. The optimized complexity of BAT algorithm is lesser as compared to PSO and DE
values of mutual coupling for PSO, DE and BAT algorithms are algorithm. The comparative results show that BAT algorithm based
0.0000009647, 0.00035665 and 0.00031070, respectively, for the results yield the lowest value of SLL as compared with those of the SLL
inter-element spacing of 0.6126 λ, 0.5750 λ and 0.7284 λ. The results obtained by using PSO and DE, keeping the FNBW value nearly the
given in Table 6 confirm that BAT algorithm based results yield the same.
lowest value of SLL as compared with those of the SLLs obtained by Fig. 10 shows the graphical representation of the radiation patterns
using PSO and DE, keeping the same value of FNBW. The results of obtained by using PSO, DE and BAT algorithms without considering
Table 6 also depict that PSO and DE algorithms take 1.2966 and the mutual coupling effect in the cost function. The figure also shows
1.2107 s, respectively, to complete the full iteration cycle, whereas, the comparison with the reference pattern obtained by using uniform
BAT algorithm takes only 1.1219 s for the same population size and current excitations to all the array elements keeping λ inter-element
2
iteration cycle. So the results show the time complexity of BAT spacing among the array elements. The maximum peak value of the
algorithm is better than PSO and DE algorithm. SLL using uniform current distribution is −13.2108 dB. The peak
The graphs of the radiation patterns for the results given in Table 6 values of SLL reduce to −34.4866 dB, −35.1556 dB and −37.1382 dB
are shown in Fig. 9 which justifies a substantial reduction in SLL with for PSO, DE and BAT algorithm, respectively. The graphical plots of
the optimal non-uniform current excitation weights as compared with Fig. 10 show clearly that BAT algorithm based results give the lowest
the case of uniform current excitations and spacing. The maximum SLL value as compared with those of the values obtained by using PSO
peak of the SLL using uniform current distribution is −13.2592 dB. The and DE keeping almost the same FNBW.
peak values of SLL reduce to −32.6971 dB, −33.7741 dB and Table 8 shows the comparative results of the non-uniformly excited
−36.1407 dB for PSO, DE and BAT algorithm, respectively. Fig. 9 26 elements symmetrical linear antenna array structure without
clearly depicts that BAT algorithm yields the lowest value of SLL as considering the mutual coupling effect in the cost function, achieved
compared with PSO and DE, keeping almost the same value of FNBW. by using PSO, DE and BAT algorithms. The table contains the
Therefore, the supremacy of BAT algorithm based approach is estab- normalized values of current excitation weights, SLL, FNBW, inter-
lished for all the cases studies, while considering the mutual coupling element spacing between the array elements and the execution times
effect. obtained by using PSO, DE and BAT algorithms. The SLL value for the
λ
uniformly excited and uniformly spaced (d = 2 ) 26 elements reference
array antenna is −13.2190 dB. The values of SLL improve to
4.1.2. In the absence of Mutual Coupling −34.7786 dB and −35.2633 dB by using PSO and DE algorithm,
The comparative results of PSO, DE and BAT algorithm for the non- respectively. Now by using BAT algorithm far better results of SLL of
uniformly excited 24 elements symmetrical linear antenna array −36.0351 dB is obtained. The FNBW value obtained by using PSO
structure without considering the mutual coupling effect in the cost algorithm is 10.80°. The authors have achieved FNBW value as 10.44°
function are shown in Table 7. The table shows the normalized values for DE and BAT algorithms. The results of Table 8 confirm that BAT
of current excitation weights, SLL, FNBW, inter-element spacing algorithm based results yield the lowest value of SLL as compared with
between the array elements and the execution time obtained by using those of the SLL obtained by using PSO and DE, keeping the FNBW
PSO, DE and BAT algorithm. The SLL value for the uniformly excited value nearly the same.
λ
and uniformly spaced (d = 2 ) 24 elements reference array antenna is The results of Table 8 also depict that PSO and DE algorithm
−13.2108 dB. The value of SLL improves to −34.4866 dB and individually takes 1.2654 and 1.1426 s to complete the full iteration
−35.1556 dB by using PSO and DE algorithm, respectively. By using cycle, whereas, BAT algorithm takes only 1.0206 s for the same
BAT algorithm far better results of SLL of −37.1327 dB is obtained. population size and iteration cycle. So the results show the time
The FNBW value obtained by using PSO is 10.80°. The value of FNBW complexity of BAT algorithm is better than that of PSO and DE
by applying DE and BAT algorithms are 10.44° and 10.80°, respec- algorithm.
tively. The execution times of PSO and DE algorithm are 1.1293 and The PSO, DE and BAT algorithm based results for non-uniformly
0.9966 s, whereas, for the same iteration cycle and population size BAT

Table 8
Current excitations, SLL, FNBW, inter-element spacing and Cost values achieved by using PSO, DE, BAT for 26-element array in the absence of mutual coupling.

Algorithm Current excitation SLL (dB) FNBW Inter-element Cost value Execution time
(deg) spacing (λ) (Sec)

PSO 0.6307 0.6477 0.6261 0.5409 0.4940 −34.7786 10.80 0.7776 0.7865 1.2654
0.4138 0.3604 0.2978 0.2169 0.1798 0.1023 0.0806 0.0107
DE 0.4455 0.4309 0.4121 0.3799 0.3454 −35.2633 10.44 0.7967 0.6170 1.1426
0.2820 0.2365 0.1849 0.1468 0.1115 0.0615 0.0465 0
BAT 0.6172 0.5928 0.5725 0.5342 0.4675 −36.0351 10.44 0.7846 0.4131 1.0206
0.3929 0.3664 0.2899 0.1931 0.1782 0.1172 0.0757 0.0297

33
A. Das et al. Swarm and Evolutionary Computation 35 (2017) 26–40

Table 9
Current excitations, SLL, FNBW, inter-element spacing and Cost values achieved by using PSO, DE, BAT for 28-element array in the absence of mutual coupling.

Algorithm Current excitation SLL (dB) FNBW Inter-element Cost value Execution time
(deg) spacing (λ) (s)

PSO 0.5451 0.5357 0.5226 0.4855 0.4477 −33.0134 10.44 0.6611 0.8647 1.2216
0.3829 0.3576 0.3164 0.2528 0.2098 0.1641 0.0974 0.1092 0.0368
DE 0.5196 0.4912 0.4924 0.4584 0.3933 −33.3692 10.44 0.7181 0.6391 1.1412
0.3670 0.3091 0.2440 0.2103 0.1664 0.1176 0.0798 0.0378 0.0001
BAT 0.5928 0.5738 0.5734 0.5132 0.4891 −36.0449 10.44 0.7184 0.4278 1.0610
0.4337 0.3609 0.3034 0.2500 0.1920 0.1371 0.1137 0.0631 0.0435

excited 26 elements symmetrical linear antenna array structure with- function are shown in Table 10. The table show the normalized values
out considering the mutual coupling effect in the cost function are of current excitation weights, SLL, FNBW, inter-element spacing
shown in Fig. 11. The figure shows a substantial reduction in SLL with between the array elements and the execution times obtained by using
optimal non-uniform current excitation weights, as compared with the PSO, DE and BAT algorithms. The SLL value for the uniformly excited
λ
case of uniform current excitations. The maximum peak of the SLL and uniformly spaced (d = 2 ) 30-element reference array antenna is
using uniform current distribution is −13.2190 dB. The peak values of −13.2449 dB. The value of SLL improves to −33.6705 dB and
SLL reduce to −34.7786 dB, −35.2633 dB and −36.0351 dB for PSO, −34.5724 dB by using PSO and DE algorithms, respectively. Now by
DE and BAT algorithm, respectively. Fig. 11 justifies the superiority of using BAT algorithm far better results of SLL of −37.2653 dB is
BAT algorithm over PSO and DE. achieved. The FNBW value is same for all the cases, as 10.44°, obtained
Table 9 depicts the comparative results of the non-uniformly by using PSO, DE and BAT algorithm. The results of Table 10 confirm
excited 28 elements symmetrical linear antenna array structure with- that BAT algorithm yields the lowest value of SLL as compared with
out considering the mutual coupling effect in the cost function, those of the SLLs obtained by using PSO and DE, keeping the FNBW
achieved by using PSO, DE and BAT algorithms. The table contains values the same. The results of Table 10 also depict that PSO and DE
the normalized values of current excitation weights, SLL, FNBW, inter- algorithms take 1.2641 and 1.1868 s, respectively, to complete the full
element spacing between the array elements and the execution times iteration cycle whereas, BAT algorithm takes only 1.0909 s for the same
obtained by using PSO, DE and BAT algorithms. The SLL value for the population size and iteration cycle. So the results show the time
λ
uniformly excited and uniformly spaced (d = 2 ) 28-element reference complexity of BAT algorithm is far better than PSO and DE algorithm.
array antenna is −13.2378 dB. The values of SLL improve to Fig. 13 shows the graphical plot of the results shown in Table 10.
−33.0134 dB and −33.3692 dB by using PSO and DE algorithm, The table shows a substantial reduction in SLL with the optimal non-
respectively. By using BAT algorithm far better results of SLL of uniform current excitation weights, as compared with the case of
−36.0449 dB is achieved. The FNBW value is same for all the cases, uniform current excitations. The maximum peak value of the SLL using
as 10.44°, obtained by using PSO, DE and BAT algorithm. The results uniform current distribution is −13.2449 dB. The peak values of SLL
of Table 9 confirm that BAT algorithm yield the lowest value of SLL as reduce to −33.6705 dB, −34.5724 dB and −37.2653 dB for PSO, DE
compared with those of the SLL obtained by using PSO and DE, and BAT algorithm, respectively. Fig. 13 shows clearly that BAT gives
keeping the FNBW value same. The results of Table 9 also show BAT the lowest SLL value as compared with those of the values obtained by
algorithm takes much lower execution time as compared to PSO and using PSO and DE keeping almost the same FNBW value.
DE algorithms to complete the full iteration cycle for the same PSO, DE and BAT algorithm based results are compared in Table 11
population size. Hence the time complexity of BAT algorithm is lesser which shows the optimized values of non-uniformly excited 32
as compared to PSO and DE. elements symmetrical linear antenna array structure, without consid-
The optimized value of Table 9 are graphically plotted in Fig. 12 ering the mutual coupling effect in the cost function. The table contains
which shows a substantial reduction in SLL with optimal non-uniform the normalized values of current excitation weights, SLL, FNBW, inter-
current excitation weights, as compared with the case of uniform element spacing between the array elements and the execution times
current excitations and spacing. The maximum peak of the SLL using obtained by using PSO, DE and BAT algorithms. The SLL value for the
λ
uniform current distribution is −13.2378 dB. The peak values of SLL uniformly excited and uniformly spaced (d = 2 ) 32-element reference
reduce to −33.0134 dB, −33.3692 dB and −36.0449 dB for PSO, DE array antenna is −13.2592 dB. The value of SLL improves to
and BAT algorithm, respectively. The figure shows visibly the super- −35.3438 dB and −35.7213 dB by using PSO and DE algorithm,
iority of BAT algorithm over PSO and DE. respectively. By using BAT algorithm far better results of SLL is
The results obtained by PSO, DE and BAT algorithm for non- achieved as −38.0849 dB. The FNBW value is same for all the cases,
uniformly excited 30 elements symmetrical linear antenna array as 10.44°, obtained by using PSO, DE and BAT algorithm. The results
structure without considering the mutual coupling effect in the cost of Table 11 confirm that BAT algorithm yields the lowest value of SLL

Table 10
Current excitations, SLL, FNBW, inter-element spacing and Cost values achieved by using PSO, DE, BAT for 30-element array in the absence of mutual coupling.

Algorithm Current excitation SLL (dB) FNBW Inter-element Cost value Execution time
(deg) spacing (λ) (s)

PSO 0.5153 0.4863 0.4694 0.4388 0.4143 −33.6705 10.44 0.6439 0.7951 1.2641
0.4078 0.3322 0.2910 0.2407 0.2125
0.1615 0.1310 0.0922 0.0544 0.0386
DE 0.6454 0.6176 0.6238 0.6045 0.5321 −34.5724 10.44 0.6246 0.6088 1.1868
0.4985 0.4245 0.3815 0.3264 0.3073
0.2212 0.1635 0.1218 0.0923 0.0772
BAT 0.5968 0.5983 0.5534 0.5353 0.4669 −37.2653 10.44 0.7181 0.4633 1.0909
0.4421 0.3683 0.3241 0.2616 0.1889
0.1525 0.1115 0.0845 0.0432 0.0060

34
A. Das et al. Swarm and Evolutionary Computation 35 (2017) 26–40

Table 11
Current excitations, SLL, FNBW, inter-element spacing and Cost values achieved by using PSO, DE, BAT for 32-element array in the absence of mutual coupling.

Algorithm Current excitation SLL (dB) FNBW Inter-element Cost value Execution time
(deg) spacing (λ) (s)

PSO 0.4290 0.4227 0.4211 0.3948 0.3587 0.3221 0.3177 0.2611 0.2128 0.1901 −35.3438 10.44 0.6322 0.6710 1.3116
0.1597 0.1248 0.0871 0.0631 0.0354 0.0284
DE 0.4991 0.4805 0.4729 0.4239 0.4030 0.3702 0.3132 0.2707 0.2304 0.1958 −35.7213 10.44 0.6772 0.5995 1.2473
0.1474 0.1029 0.0806 0.0346 0.0392 0.0001
BAT 0.6825 0.6668 0.6246 0.6114 0.5437 0.5100 0.4519 0.3923 0.3249 0.2699 −38.0849 10.44 0.6628 0.3244 1.1332
0.2198 0.1543 0.1156 0.1093 0.0397 0.0085

Fig. 5. Array Patterns obtained by using different algorithms for the 24-element linear Fig. 7. Array Patterns obtained by using different algorithms for the 28-element linear
array in the presence of mutual coupling. array in the presence of mutual coupling.

Fig. 6. Array Patterns obtained by using different algorithms for the 26-element linear Fig. 8. Array Patterns obtained by using different algorithms for the 30-element linear
array in the presence of the mutual coupling. array in the presence of mutual coupling.

as compared with those of the SLLs obtained by using PSO and DE, uniformly excited 32 elements symmetrical linear antenna array
keeping the FNBW values the same. The results of Table 11 also depict structure in the absence of the mutual coupling effect in the cost
that PSO and DE algorithms take 1.3116 and 1.2473 s, respectively to function are graphically plotted in Fig. 14. In Fig. 14 the plots are
complete the full iteration cycle whereas, BAT algorithm takes only compared with a reference pattern obtained by adopting uniform
1.1332 s for the same population size and iteration cycle. So the results excitation to all the array elements having a half wavelength inter-
show the time complexity of BAT algorithm is far better than PSO and element spacing. The maximum peak value of the SLL using uniform
DE algorithms. current distribution is −13.2592 dB. The peak value of SLL reduces to
The results obtained by using PSO, DE and BAT algorithms for non- −35.3438 dB, −35.7213 dB and −38.0849 dB, respectively, for PSO,

35
A. Das et al. Swarm and Evolutionary Computation 35 (2017) 26–40

Fig. 9. Array Patterns obtained by using different algorithms for the 32-element linear Fig. 12. Array Patterns found by using different algorithms for the 28-element linear
array in the presence of mutual coupling. array in the absence of mutual coupling.

Fig. 10. Array Patterns found by using different algorithms for the 24-element linear Fig. 13. Array Patterns obtained by using different algorithms for the 30-element linear
array in the absence of mutual coupling. array in the absence of mutual coupling.

Fig. 11. Array Patterns found by using different algorithms for the 26-element linear Fig. 14. Array Patterns obtained by using different algorithms for the 32-element linear
array in the absence of mutual coupling. array with suppressed SLL and null control in absence of mutual coupling.

36
A. Das et al. Swarm and Evolutionary Computation 35 (2017) 26–40

Table 12 Table 15
Minimum, Maximum, Mean and Standard deviation of the Cost Function values for t-test values obtained for comparison of BAT algorithm with PSO and DE without
different algorithms considering the mutual coupling effect. considering the mutual coupling effect.

Number of Algorithm Minimum Maximum Mean value Standard Number of elements Algorithms t-test value
elements value value deviation
24 BAT/PSO 9.6863
24 PSO 0.8442 1.2352 1.0005 0.1166 BAT/DE 5.8333
DE 0.6632 1.0717 0.9128 0.1700
BAT 0.4727 0.8057 0.6352 0.1101 26 BAT/PSO 12.5099
BAT/DE 7.0442
26 PSO 0.8614 1.2532 1.0470 0.1236
DE 0.6904 1.0901 0.8890 0.0985 28 BAT/PSO 10.4622
BAT 0.5175 0.9027 0.7210 0.1038 BAT/DE 7.7073

28 PSO 0.9469 1.2460 1.0424 0.1028 30 BAT/PSO 8.0837

DE 0.6648 1.0506 0.8659 0.0988 BAT/DE 6.8651
BAT 0.4639 0.8233 0.6341 0.0929
32 BAT/PSO 10.0640
30 PSO 0.8175 1.2050 0.9375 0.0913 BAT/DE 8.5364
DE 0.7087 1.1506 0.9064 0.1168
BAT 0.5595 0.9223 0.8004 0.1037

32 PSO 0.8495 1.2390 0.9514 0.0908

DE 0.7079 1.1006 0.8954 0.0965
BAT 0.5663 0.9463 0.7935 0.1358

Table 13
t-test values obtained for comparison of BAT algorithm with PSO and DE considering the
mutual coupling effect.

Number of elements Algorithms t-test value

24 BAT/PSO 16.3056
BAT/DE 9.6916

26 BAT/PSO 14.2819
BAT/DE 8.3016

28 BAT/PSO 20.8369
BAT/DE 12.0861

30 BAT/PSO 7.0166
BAT/DE 4.7988

32 BAT/PSO 6.8348
BAT/DE 4.3251
Fig. 15. Convergence profiles for the non-uniformly excited 24-element linear array in
the presence of mutual coupling.

Table 14
Minimum, Maximum, Mean and Standard deviation of the Cost Function values for
different algorithms without considering the mutual coupling effect.

Number of Algorithm Minimum Maximum Mean value Standard

wlements value value deviation

24 PSO 0.7679 1.2390 0.9917 0.1479

DE 0.5753 1.1006 0.8824 0.1664
BAT 0.4003 0.9463 0.6903 0.1629

26 PSO 0.7865 1.1675 0.9868 0.1253

DE 0.6170 1.0043 0.8318 0.1073
BAT 0.4131 0.8162 0.6586 0.1368

28 PSO 0.8647 1.2152 0.9523 0.1655

DE 0.6391 1.0290 0.8303 0.1143
BAT 0.4278 0.8075 0.6480 0.1221

30 PSO 0.7951 1.1635 0.8753 0.2177

DE 0.6088 0.9888 0.7269 0.0907
BAT 0.4633 0.7635 0.6112 0.0773

32 PSO 0.6710 1.1225 0.8332 0.1631

DE 0.5995 0.9955 0.7226 0.0856
BAT 0.3244 0.7066 0.5471 0.1175 Fig. 16. Convergence profiles for the non-uniformly excited 26-element linear array in
the presence of mutual coupling.

37
A. Das et al. Swarm and Evolutionary Computation 35 (2017) 26–40

Fig. 17. Convergence profiles for the of non-uniformly excited 28-element linear array Fig. 19. Convergence profiles for the non-uniformly excited 32-element linear array in
In the presence of mutual coupling. the presence of mutual coupling.

Fig. 18. Convergence profiles for the non-uniformly excited 30-element linear array in Fig. 20. Convergence profiles for the non-uniformly excited 24-element linear array in
the presence of mutual coupling. the absence of mutual coupling.

DE and BAT algorithms, respectively. Fig. 14 reveals that BAT 4.2. Comparisons of accuracy
algorithm yields the lowest SLL value as compared with PSO and DE,
keeping the FNBW values same. Therefore, the supremacy of BAT The results obtained by using PSO, DE and BAT algorithms are also
algorithm is established from all the cases studies, without considering compared for the sake of accuracy. Statistical tests [32] are conducted
the mutual coupling effect. for the independent samples to judge whether the results obtained by
The overall conclusion regarding the simulation results given above using BAT algorithm differ significantly from the results achieved by
are as follows: The simulation results show the proposed BAT based using PSO and DE algorithm. To determine the statistical significance
approach for linear antenna array synthesis and its performance, for of the difference between two algorithms, two sample t-test are
suppressing the side lobes and placing the nulls in a interference region conducted between BAT/PSO and BAT/DE algorithms. The positive
(75–77° and 105–107°) in the presence or in the absence of the mutual value of t-test signifies that first algorithm is better than the second
coupling term in the cost function. The BAT's optimizing capability and whereas negative value means the first algorithm performs poorer than
time complexity is better than PSO and DE. In the plot of radiation the second algorithm [30]. The t-value is defined in Eq. (3).
patterns three optimized patterns obtained by PSO, DE and BAT, α2 − α1
t=
respectively and a reference pattern (uniformly excited all array ⎛ σ22 ⎞ ⎛ σ12 ⎞
elements with half wavelength spacing between the array elements) ⎜ β + 1⎟ + ⎜ β + 1⎟
⎝ ⎠ ⎝ ⎠ (3)
are compared. All the above plots of radiation patterns reveal the
superior optimizing capability of BAT over PSO and DE. where α1 and α2 are the mean values of the first and the seconds

38
A. Das et al. Swarm and Evolutionary Computation 35 (2017) 26–40

Fig. 21. Convergence profiles for the non-uniformly excited 26-element linear array in Fig. 23. Convergence profiles for the non-uniformly excited 30-element linear array in
the absence of mutual coupling. the absence of mutual coupling.

Fig. 22. Convergence profiles for the non-uniformly excited 28-element linear array in Fig. 24. Convergence profiles for the non-uniformly excited 32-element linear array in
the absence of mutual coupling. the absence of mutual coupling.

methods, respectively;σ1 and σ2 are the standard deviations of the first 4.3. Effectiveness and convergence profiles of PSO, DE and BAT
and the second methods, respectively; and β is the value of degree of
freedom. Figs. 15–19 show the convergence plots of the adopted algorithms
If the t-test value is higher than 1.645 for the β value of 49, means utilized to design the non-uniformly excited linear antenna arrays with
there is a significant difference between the two algorithms with a 95% 24, 26, 28, 30, 32 elements with mutual coupling effect. For all these
confidence level. The t-test value between the BAT/PSO and BAT/DE cases BAT algorithm is found to yield the lowest CF values in compared
algorithm methods are presented in Tables 13, 15. The authors with PSO and DE.
achieved t-value is larger than 2.15, for the degree of freedom 49, Figs. 20–24 show the convergence plots of the adopted algorithms
meaning that there is a significant difference between BAT/PSO and utilized to design the non-uniformly excited linear antenna arrays with
BAT/DE algorithms methods with a 98% confidence level. Thus from 24, 26, 28, 30, 32 elements without mutual coupling effect. For all these
the statistical analysis, it is evident that BAT algorithm based technique cases BAT algorithm is found to yield the lowest CF values in compared
offers robust and promising results. with PSO and DE.
Tables 12 and 14 depict the minimum value, maximum value, mean All above plots of convergence, where the cost values are plotted
value and standard deviation of the cost function for 24, 26, 28, 30 and against the number of iterations, expose the excellent optimizing
32 element arrays. The results of Tables 12 and 14 show clearly that the capability of BAT over PSO and DE in terms of precision and
minimum value, maximum value, mean value of the cost function convergence speed.
achieved by using BAT algorithm are lower than those of PSO and DE
algorithm. So, from all the results discussed above it is established that
5. Conclusions
BAT algorithm based results are found to be the best in terms of
accuracy.
In this paper, BAT optimization technique is applied to find out the

39
A. Das et al. Swarm and Evolutionary Computation 35 (2017) 26–40

solution of highly nonlinear values of the current excitations for the global optimization over continuous spaces, Technical Report, International
Computer science Institute, Berkley, 1995.
symmetric linear antenna array element. The problem of designing [21] N. Karaboga, A new design method based on artificial bee colony algorithm for
linear array antenna with optimal current excitations and inter- digital IIR filters, J. Frankl. Inst. 346 (2009) 328–348.
element separations using PSO, DE and BAT optimization technique [22] N. Karaboga, Digital IIR filter design using differential evolution algorithm,
EURASIP J. Appl. Signal Process. 2005 (2005) 1269–1276 (Hindawi Publishing
has been investigated with and without considering the mutual Corp., Article ID. 8).
coupling term in the cost function. The main objective of this work is [23] J. Altringham, Bats: Biology and Behaviour, Oxford University Press, 1998.
to achieve the lowest side lobe level and placing the nulls in the [24] C. Tudge, The Variety of Life: A Survey and a Celebration of all the Creatures That
Have Ever Lived, Oxford University Press, 2002.
interference region along with minimizing the mutual coupling effect [25] P. Richardson, Bats. Natural History Museum, London, 2008.
among the array elements. For the first time the authors are introdu- [26] P. Richardson, The secrete life of bats 〈http://www.nhm.ac.uk〉.
cing the effect of mutual coupling in the cost function considered in this [27] X.-S. Yang, A new metaheuristic bat-inspired algorithm, in: J.R. Gonzalez et al.
(Eds.), Nature Inspired Cooperative Strategies for Optimization, NISCO 2010,
paper. The results obtained by using BAT algorithm show that it leads
Studies in Computational Intelligence, vol. 284, Springer, Berlin, 2010, pp. 65–74.
to an improved design in terms of suppressing the side lobe level while [28] X.S. Yang, A.H. Gandomi, Bat algorithm: a novel approach for global engineering
maintaining nulls in the interference region. The performance of BAT optimization, Eng. Comput. 29 (5) (2012) 464–483.
algorithm is found to be the best in terms of the accuracy, the time [29] M. Kumar, A. Aggarwal, T.K. Rawat, Bat algorithm: application to adaptive infinite
impulse response system identification, Arab. J. Sci. Eng. 41 (9) (2016)
complexity and the convergence speed as compared with PSO and DE 3587–3604.
algorithms. Thus the obtained results confirm that the BAT algorithm [30] R.E. Walpole, R.H. Myer, Probability and Statistics for Engineers and Scientists,
has a good prospective to be used as an efficient optimizer for the Macmillan, New York, 1978.
[31] A. Zhou, B.-Y. Qu, H. Li, S.-Z. Zhao, P.N. Suganthan, Q. Zhang, Multiobjective
solution of obtaining the optimal values of current excitation weights evolutionary algorithms: a survey of the state of the art, Swarm Evolut. Comput. 1
and the inter-element spacing between the array elements for the (1) (2011) 32–49.
synthesis and pattern modeling of a symmetric linear array antenna. [32] J. Derrac, S. García, D. Molina, F. Herrera, A practical tutorial on the use of
nonparametric statistical tests as a methodology for comparing evolutionary and
swarm intelligence algorithms, Swarm Evolut. Comput. 1 (1) (2011) 3–18.
Acknowledgement [33] A.E. Eiben, S.K. Smit, Parameter tuning for configuring and analyzing evolutionary
algorithms, Swarm Evolut. Comput. 1 (1) (2011) 19–31.
[34] A. Aggarwal, M. Kumar, T.K. Rawat, D.K. Upadhyay, Optimal design of 2-D FIR
This work has been supported by SERB, Department of Science and
filters with quadrantally symmetric properties using fractional derivative con-
Technology—Government of India under the grant SB/EMEQ-319/ straints, Circuits, Syst. Signal Process. 35 (6) (2016) 2213–2257.
2013. The authors are grateful to Science and Engineering Research [35] A. Aggarwal, M. Kumar, T.K. Rawat, D.K. Upadhyay, Optimal design of 2-D FIR
digital differentiator using L1-norm based Cuckoo-Search algorithm, Multidimens.
Board for funding the project.
Syst. Signal Process. (2016) 1–19.
[36] M. Kumar, T.K. Rawat, Optimal fractional delay-IIR filter design using cuckoo
References search algorithm, ISA Trans. 59 (2015) 39–54.
[37] M. Kumar, T.K. Rawat, Optimal design of FIR fractional order differentiator using
cuckoo search algorithm, Expert Syst. Appl. 42 (7) (2015) 3433–3449.
[1] C.A. Ballanis, Antenna Theory Analysis and Design, 2nd ed., John Willey and Son's [38] M. Kumar, T.K. Rawat, Design of a variable fractional delay filter using compre-
Inc, New York, 1997. hensive least square method encompassing all delay values, J. Circuits Syst.
[2] R.S. Elliott, Antenna Theory and Design, Revised ed., John Willey, NJ, 2003. Comput. 24 (8) (2015) 1550116.
[3] Hoi-Shun Lui, Hon Tat Hui, Mook Seng Leong, A note on the mutual-coupling [39] R. Barsainya, T.K. Rawat, M. Kumar, Design of minimum multiplier fractional
problems in transmitting and receiving antenna arrays, IEEE Antennas Propag. order differentiator based on lattice wave digital filter, ISA Trans. (2016). http://
Mag. 51 (5) (2009). dx.doi.org/10.1016/j.isatra.2016.09.024.
[4] X. Li, M. Yin, Optimal synthesis of linear antenna array with composite differential [40] M. Kumar, T.K. Rawat, A. Aggarwal, Adaptive infinite impulse response system
evolution algorithm, Sci. Iran. D 19 (6) (2012) 1780–1787. identification using modified-interiorusing modified-interior search algorithm with
[5] L. Pappula, D. Ghosh, Linear antenna array synthesis using cat swarm optimiza- levy flight, ISA Trans. (2016). http://dx.doi.org/10.1016/j.isatra.2016.10.018.
tion, Int. J. Electron. Commun. (AEU) 68 (6) (2014) 540–549. [41] Ali. R. Yildiz, A novel particle swarm optimization approach for product design and
[6] D. Mandal, S.P. Ghoshal, A.K. Bhattacharjee, Wide null control of symmetric linear manufacturing, Int. J. Adv. Manuf. Technol. 40 (5–6) (2009) 617–628.
antenna array using novel particle swarm optimization, Int. J. RF Microw. Comput. [42] Ali. R. Yildiz, A new hybrid particle swarm optimization approach for structural
Eng. 21 (4) (2011) 376–382. design optimization in automotive industry, J. Automob. Eng. 226 (D10) (2012)
[7] M.A. Mangond, H.M. Elragal, Antenna array pattern synthesis and wide null 1340–1351.
control using enhanced particle swarm optimization, Prog. Electromagn. Res. B 17 [43] Hakan Gökdağ, Ali R. Yildiz, Structural damage detection using modal parameters
(2009) 1–14. and particle swarm optimization, Mater. Test. 54 (6) (2012) 416–420.
[8] V. Robinovich, N. Alexandrov, Antenna Arrays and Automotive Application [44] Ali Rıza Yıldız, Enes Kurtuluş, Emre Demirci, Betul Sultan Yıldız, Selçuk Karagöz,
(Chapter 2), in: Typical Array Geometries and Basic Beam Steering Methods, Optimization of thin-wall structures using hybrid gravitational search and Nelder-
Springer Science, New York, 2013. Mead algorithm, Mater. Test. 58 (1) (2016) 75–78.
[9] R.L. Haupt, D.H. Werner, Genetic Algorithms in Electromagnetics, IEEE Press and [45] Betul Sultan Yildiz, Huseyin Lekesiz, Ali Riza Yildiz, Structural design of vehicle
Willey-Interscience, 2007. components using gravitational search and charged system search algorithms,
[10] T. Ciloglu, An efficient local search method guided by gradient information for Mater. Test. 58 (1) (2016) 79–81.
discrete coefficient FIR filter design, Signal Process. 82 (10) (2002) 1337–1350. [46] Ali R. Yildiz, Kiran N. Solanki, Multi-objective optimization of vehicle crash-
[11] R.L. Haupt, An introduction to genetic algorithm for electromagnetics, IEEE worthiness using a new particle swarm based approach, Int. J. Adv. Manuf.
Antennas Propag. Mag. 37 (2) (1995) 7–15. Technol. 59 (1–4) (2012) 367–376.
[12] D. Gie, Y. Rahmat-Samii, Particle swarm optimization for reconfigurable phased [47] Morteza Kiani, Ali R. Yildiz, A comparative study of non-traditional methods for
differentiated array design,”differentiated array design, Microw. Opt. Technol. Lett. vehicle crashworthiness and NVH optimization, Arch. Comput. Methods Eng. 23
38 (3) (2003) 168–175. (4) (2016) 723–734. http://dx.doi.org/10.1007/s11831-015-9155-y.
[13] D.W. Boeringer, D.H. Werner, Particle swarm Optimization versus genetic algor- [48] Ali R. Yildiz, Comparison of evolutionary based optimization algorithms for
ithmalgorithm for phased arrayarray synthesis,”synthesis, IEEE Trans. Antenna structural design optimization, Eng. Appl. Artif. Intell. 26 (1) (2013) 327–333.
Propag. 52 (3) (2004) 771–779. [49] Ali R. Yildiz, A comparative study of population-based optimization algorithms for
[14] J. Robinson, Y. Rahmat-Samii, Particle swarm optimization in electromagne- turning operations, Inf. Sci. 210 (2012) 81–88.
tics,"electromagnetics, IEEE Trans. Antenna Propag. 52 (2) (2004) 397–407. [50] Ali R. Yildiz, Hybrid Taguchi-differential evolution algorithm for optimization of
[15] G. Ciuprina, D. Ioan, I. Munteanu, Use of intelligent-particle swarm optimization in multi-pass turning operations, Appl. Soft Comput. 13 (3) (2013) 1433–1439.
electromagnetics, IEEE Trans. Magn. 38 (2) (2004) 1037–1040. [51] Ali R. Yildiz, Hybrid immune-simulated annealing algorithm for optimal design and
[16] M.M. Khodier, C.G. Christodoulou, Linear array geometry synthesis with minimum manufacturing, Int. J. Mater. Prod. Technol. 34 (3) (2009) 217–226.
sidelobe level and null control using particle swarm optimization, IEEE Trans. [52] Ali R. Yildiz, A new hybrid differential evolution algorithm for the selection of
Antenna Propag. 53 (8) (2005) 2674–2679. optimal machining parameters in milling operations, Appl. Soft Comput. 13 (3)
[17] J. Kennedy, R. Eberhart, Particle swarm optimization, Proc. IEEE Int. Conf. Neural (2013) 1561–1566.
Netw. 4 (1995) 1942–1948. [53] Nursel Ozturk, Ali R. Yildiz, Necmettin Kaya, Ferruh Ozturk, Neuro-genetic design
[18] R. Eberhart, Y. Shi, Comparison between genetic algorithm and particle swarm optimization framework to support the integrated robust design optimization
optimization, in: Evolutionary Programming VII, Springer, 1998, pp. 611–616. process in CE, Concurr. Eng. Res. Appl. 14 (1) (2006).
[19] S.K. Saha, R. Kar, D. Mandal, S.P. Ghoshal, IIR filter design with craziness based [54] Ali R. Yildiz, Optimization of multi-pass turning operations using hybrid teaching
particle swarm optimization technique, World Acad. Sci. Eng. Technol. 60 (2011) learning-based approach, Int. J. Adv. Manuf. Technol. 66 (9–12) (2013)
1628–1635. 1319–1326 (6/1).
[20] R. Storn, K. Price, Differential evolution- a simple and efficient adaptive scheme for

40