Comparative Performance of QPSO and BSA for

Design of Linear Array of Parallel Dipole Antennas

With Mutual Coupling
Hemant Patidar, Ananya Mahanti, Rohit Sharma and G.K. Mahanti
Department of Electronics and Communication Engineering,
National Institute of Technology, Durgapur- 713209, India
E-mail: [email protected],[email protected], [email protected],[email protected]

Abstract—The authors presented two different optimization In this paper, the authors compared the performance of QPSO
methods- one is quantum particle swarm optimization (QPSO) with backtracking search optimization algorithm in terms of
and another is backtracking search algorithm (BSA). Combined
antenna parameters and statistical analysis.
with inverse fast fourier transform (IFFT), the paper deals with
the synthesis of array of linearly spaced and mutually coupled The objective of the paper is to search for optimal set of
half wavelength very thin vertical and parallel dipoles with fixed excitation voltage amplitude distribution using the above two
value of voltage standing wave ratio (VSWR) and side lobe level. optimization algorithms that will minimize voltage standing
The performances of the two algorithms are compared in terms wave ratio and side lobe level and to a value close to the
of different antenna parameters like side lobe level, VSWR as desired. Lower value of VSWR is required to provide better
well as statistical parameters such as best fitness value, mean, match between antenna and the feed network
standard deviation and computation time. There exists mutual Another contribution in the paper is reduction of computation
coupling among the parallel half–wavelength dipole antennas. time by calling an inverted mutual coupling matrix at every
Self and mutual impedances are derived from induced EMF
iteration stored separately in a mat file. This inverted mutual
method assuming distribution of current on each dipole is
sinusoidal. The pattern so generated is broadside in the coupling matrix is just multiplied by excitation voltage
horizontal plane. There exists a relationship between the IFFT amplitude distribution to get excitation current distribution at
and the array factor for an array of linearly and uniformly every iteration. This will eliminate mutual coupling matrix
spaced isotropic antennas. Array factor is found out by applying inversion at every iteration because excitation current
inverse fast fourier transform on the element excitations that in distribution is obtained by dividing excitation voltage
turn reduces the computational time. It is assumed that the amplitude distribution by mutual coupling matrix.
element pattern of vertical half–wavelength dipole antenna is Computation time is further reduced by calculating the array
omni-directional in the horizontal plane. One example is factor using IFFT instead of calculating it by conventional
presented to prove the effective use of the approach proposed.
method.
Keywords—Dipole Antenna; QPSO; BSA; VSWR; Mutual
Coupling II. THEORY

I. INTRODUCTION The far-field pattern in free space FF (u) in azimuth plane (x-y
plane) [1,5] for a uniformly spaced linear array of mutually
Antenna array [1] finds applications in various fields such as coupled parallel half-wavelength dipole antennas placed along
wireless, satellite and radar communications due to their high the x-axis with spacing d between the elements is given by
directivity. Several synthesis techniques related to antenna eqn. (1):
array were reported in the literature [2-5]. GA combined with Fig. 1 shows the array geometry.
FFT for synthesis of array antenna has been reported in [2]. M. ⎡N ⎤
Thevenot et.al. [3] proposed design of antenna array and
parasitic antenna array with mutual coupling. K. M. Lee [4]
∑
FF (u) = AF (u ) × EP(u ) = ⎢ I ne j ( n−1) kdu ⎥ × EP(u)
⎢⎣ n=1 ⎥⎦
(1)

analyzed mutual coupling between a finite phased array of Where n=element number, j=imaginary quantity, d=inter
dipoles and its feed network. Reconfigurable array antennas element spacing =0.5λ, k=2π/λ= wave number, λ=wavelength,
synthesis with minimum variation of active impedances was u=cosφ, φ being the azimuth angle of far-field measured from
reported in [5]. x-axis (0o to 180o), In =excitation current amplitude of the n-th
Formulation of QPSO has been presented in the article [6, 7]. element, N is total number of elements. In our case, Vn is
Backtracking Search Algorithm [8], a population based excitation voltage at the n-th dipole terminal. All the elements
heuristic evolutionary algorithm, has been used for numerical are fed with constant voltage excitation phase of zero degree.
problems related to optimization. AF(u) is the array factor. EP (u) is element pattern of each z-
978-1-4673-6621-2/16/$31.00 © 2016 IEEE
directed vertical half-wave length dipole anntenna, which is If the transmission line has a chaaracteristic impedance not
assumed omnidirectional in x-y plane, i.e., EP
P (u) =1. being equal to the terminating impeedance, some of the power
is reflected back from the termiination. The ratio of the
maximum to minimum voltage along a transmission line is
defined as Voltage Standing Wave Ratio.
R
Once the reflection coefficient has been calculated, it can be
used to determine maximum VSWR R by the following formula:

1 + RC max
m
VSWRmax = (5)
1 − RC max
m
A low value of VSWRmax is desired d to ensure tolerance of the
impedance matching condition fo or all the elements of the
array. This will deliver more po ower to the antenna. The
objective is now to search for th he optimal set of voltage
amplitude distribution of the elemeents with the help of QPSO
and BSA that will minimize the folllowing cost function:

Cost or Fitness = F12 + F2[ ] (6)

Fig. 1. An array of z-directed parallel half-wave llength dipoles Where

The terminal current distribution matrix of ssize (N×1) at the ⎧SLL0 − SLLd , if → SLLo > SLLd
F1 = ⎨ (7)
antenna is given by [1, 5]:
⎩0, if → SLLo ≤ SLLLd

I=Z¯1V

Where V is the driving voltage matrix of size (N×1) applied to ⎧⎪VSWRmax,o −VSWRmax,d , if →VSWR
V max,o > VSWRmax,d
F2 = ⎨ (8)
dipole antennas and Z is the mutual couppling impedance ⎪⎩0, if →VSWRmax,o ≤ VSWRmax,,d
matrix of size (N×N).
Where SLLo and SLLd are obtained d and desired value of side
Self-impedances Znn and mutual Impedancees Znm of Z are
derived from induced EMF method [1, 5]], which assume lobe level, VSWR max, o and VSWR max, d are obtained and
dipoles to have sinusoidal current distribution.. desired value of maximum volttage standing wave ratio
The integration is solved using 16-point Gauss-Legendre respectively.
quadrature integration formula.
The voltage across the n-th dipole [1, 5] is givven by: III. RRAY FACTOR
IFFT AND AR

Vn = Z nn I n + ∑Z
m≠n
nm I m (2)
Array factor (AF) in x-y plane is giv
ven by

N
Where Znn is the self-impedance of dipole n and Znm is the
mutual impedance between dipoles n andd m. The active
AF (u ) = ∑I e
n =1
n
j ( n −1) kdu

impedance [1, 5] across dipole n, ZnA is givenn by:

N
Let t = 1+ k ,
kdu
Z nA = V n / I n = Z nn + ∑Z
m≠ n
nm ( I m / In) (3)
Then
2π

Considering the feed network with characterisstic impedance of N ⎛ 2π ⎞

j⎜ ⎟ ( n−1)(t −1)
50 ohm, the refection coefficient (RC) at thee input of of n-th
dipole antenna [1] is given by:
AF (t ) = ∑I
n=1
n e ⎝N⎠ (9)

Equation (9) is similar to standard

d form of one dimensional
⎡ ZA −Z ⎤ IFFT. This shows that the arraay factor can be directly
n o
RC n = ⎢ ⎥ (4) calculated by operating IFFT on n the current In and the
⎢ZA +Z ⎥
⎣⎢ n o⎥
⎦ complexity involved in the compu utation can be reduced to a
Zo=characteristic impedance. Now the maxximum reflection great extent. The program uses 4096
6-point IFFT.
coefficient among all elements RCmax is calcullated.
 produced by QPSO and BSA. Fig.5 depicts the normalized
power pattern in dB using QPSO and BSA.

IV. OVERVIEW OF QPSO AND BSA

The QPSO algorithm that was described in [6, 7], is an

evolutionary algorithm based on the principles of particle TABLE I
swarm and properties of quantum mechanics. COMPARATIVE RESULTS OF QPSO AND BSA
Velocity vector is absent in QPSO and that way it is different Design Parameters Case1(Obtained by Case2(Obtained by
from standard PSO. It has only position vector. QPSO has one QPSO) BSA)
controlling parameter called α=0.75, being the coefficient Global Best Fitness 0.0246 0.0867
dealing with the convergence. Value
Standard Deviation 0.0114 0.0089
Mean 0.0355 0.1031
Backtracking search algorithm (BSA), an evolutionary Maximum Active 54.86–3.76i 57.41–5.40i
algorithm that was proposed by P. Civicioglu [8]. BSA has Impedance (Ohm)
two control parameters, one is called dim_rate, which in our Side Lobe Level –26.12 –26.66
problem is set as unity, another one is scale factor, which is set (dB)
Voltage Standing 1.1244 1.1866
as 3*randn; randn being random normal distribution. Wave Ratio
Description of the above two algorithms are available in the Computation Time 905.74 813.96
paper [6-8]. Readers are requested to go through these articles (Seconds)
for further reading.
TABLE II
NORMALIZED EXCITATION VOLTAGE AMPLITUDE DISTRIBUTION OBTAINED BY
V. SIMULATED RESULTS QPSO AND BSA

We present an example of linearly and uniformly spaced array Element Using Using Element Using Using
Number QPSO BSA Number QPSO BSA
of 35 center-fed z-directed dipole antennas of length 0.5λ and 1 0.0973 0.1738 19 0.8212 0.7449
spaced 0.5λ apart along x-axis in order to produce a broadside 2 0.2123 0.3206 20 0.8331 0.6941
pattern in azimuth plane (x-y plane) with specified SLL of - 3 0.2741 0.3022 21 0.8504 0.7497
25dB or less and desired voltage standing wave ratio of 1.1 or 4 0.2772 0.2768 22 0.7703 0.6956
less. The excitation voltage amplitude distribution is 5 0.2399 0.3190 23 0.7484 0.7334
6 0.2991 0.2792 24 0.8050 0.5782
asymmetric. Each dipole has a diameter of 0.003λ. IFFT is 7 0.3460 0.3282 25 0.7783 0.5667
used for calculation of array factor. Number of field point in 8 0.3973 0.5129 26 0.7523 0.7145
IFFT is 4096 points. Two cases are discussed here. 9 0.4772 0.4487 27 0.6326 0.5071
10 0.5495 0.3964 28 0.6058 0.4417
Case1: For generation of power pattern in presence of mutual 11 0.6284 0.5989 29 0.5185 0.4115
12 0.6369 0.5710 30 0.4878 0.3831
coupling QPSO is run 5 times each with 500 generations with 13 0.6033 0.6745 31 0.4429 0.2754
a population size of 60. A set of 35 voltage amplitude values is 14 0.7574 0.6585 32 0.4146 0.1630
generated. Each run generates one best fitness value and 15 0.8343 0.6627 33 0.4009 0.2713
global best is regarded as best among such five best cost 16 0.8557 0.8320 34 0.3123 0.3141
values. Mean and standard deviation of five best cost values 17 0.8492 0.8214 35 0.1428 0.1715
are then calculated. 18 0.8963 0.6722 -------- -------- --------
Case2: For generation of power pattern in presence of mutual
coupling and to compare the obtained results of BSA with that 50
of QPSO, BSA is also run 5 times each with 500 iterations
with a population size of 60. A set of 35 voltage amplitude 40
values is generated.
B est F itn ess V alu e

Matlab was used for writing the program. Measurement of

30
computational time is done with a PC with Intel processor of
clock frequency 2.93GHz and 2 GB of RAM. Table I shows
comparative results produced by quantum particle swarm 20

optimization and backtracking search optimization algorithm.

Table II shows voltage amplitude distribution of the elements 10
obtained by QPSO and BSA. Fig.2 shows the best -run fitness
value versus iterations produced by QPSO. Fig.3 shows the
0
best-run fitness value versus iterations obtained by BSA. Fig.4 0 50 100 150 200 250 300 350 400 450 500
depicts the normalized voltage amplitude distribution Iterations
Fig. 2. Best-run fitness value versus iterations produced by QPSO.
 though both the algorithms suited well
w for synthesis of power
pattern in presence of mutual couplling, the processing time of
QPSO is slightly more than BSA A. The desired values of
antenna parameters such as SLL L and VSWR match very
60 closely with each other with QPSO slightly better than BSA in
producing VSWR value. The same is the case with the
50 statistical parameters where both h algorithms match very
closely with each other, with QPS SO showing slightly better
Best Fitness Value

40 results than BSA in global best fitn

ness value and mean. This
comparative analysis can be extend ded to other antenna array
30 configurations also.

20 REFERENC
CES

10 [1] Balanis C.A., Antenna Theory: Analysis and Design, 2nd ed.
Singapore: John Wiley and Sons (Asiaa), 2003.
[2] L.L. Wang, D.G.Fang and W.X.S Sheng, “Combination of genetic
0
0 100 200 300 400
0 500 algorithm and fast Fourier transfform for synthesis of arrays”,
Iterations Microwave and Optical technology letters,
l vol. 37, no. 1, pp. 56-59,
April 2003.
Fig. 3. Best-run fitness value versus iterations generated by BSA. [3] M. Thevenot, C. Menudier, A. El Saayed Ahmad, G. Z. El Nashef, F.
Fezai, Y. Abdallah, E. Arnaud, F. Torres, T. Monediere, “Synthesis
of antenna arrays and parasitic antennna arrays with mutual couplings”,
International Journal of Antennas andd Propagation, vol. 2012, pp.1-22,
2012.
[4] K. M. Lee, R. S. Chu, “Analysis of mutual coupling between a finite
phased array of dipoles and its feed network”, IEEE Transactions on
Antennas and Propagation, vol. 36, no. 12, pp.1681-1699, December
1988.
[5] G.K. Mahanti, S. Das, A. Chakrabaarty, J.C. Brégains and F. Ares,
“Design of reconfigurable array anteennas with minimum variation of
active impedances”, IEEE Antennas anda Wireless Propagation Letters,
vol.5, no.1, pp.541-544, 2006.
[6] J. Sun, B. Feng, W. Xu, “Particle sw warm optimization with particles
having quantum behavior”, in Proc. of Congress on evolutionary
computation, Portland, Oregon, USA, pp. 325-331, 2004.
[7] J. Sun, W. Xu, B. Feng, “A global search strategy of quantum behaved
Particle swarm optimization”, in Proc. IEEE Conference on
Cybernetics and Intelligent systems, Singapore,
S pp. 111-116, 2004.
[8] P.Civicioglu, “Backtracking search optimization for numerical
optimization problems”, Applied Ma athematics and Computation, vol.
219, no. 15, pp. 8121-8144, 2013.
Fig. 4. Voltage amplitude distribution generated by Q
QPSO and BSA.

0
N o rm alized P o w er P attern in d B

-10 Pattern for

f QPSO
Pattern for
f BSA
-20

-30

-40

-50

-60

-70

-80
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0..6 0.8 1
u
O and BSA.
Fig. 5. Normalized power pattern using QPSO

VI. CONCLUSIONS

This paper presents a comparative analysis between QPSO

and BSA algorithms in terms of the design paarameters as well
as statistical ones. Tables providing the resultts show that even