International Conference on Current Trends in Computer, Electrical, Electronics and Communication (ICCTCEEC-2017)

Optimized Design and Analysis of Microstrip

Antenna using Artificial Neural Network
Chetana Nayak Pramod Kumar
B.Tech Student Department of Electronics & Communication Engineering
Department of Electronics & Communication Engineering Manipal Institute of Technology, Manipal- 576104, India
Manipal Institute of Technology, Manipal-576104, India Email: - 79. [email protected]
Email: - [email protected]

Abstract— An Antenna is a device made of conducting

material which transmits and receives electromagnetic signals. II. ARTIFICIAL NEURAL NETWORKS (ANN)
Antennas are widely used as device for transmission of signals.
Microstrip antennas are gaining more importance due to their Artificial neural networks (ANNs) are a family of learning
small and compact size, and wide band which yields in the models in particular that are used to estimate functions that
reduction of material and applications are in cell phones, aircraft depend on a large number of inputs and are not known [2].
and satellite communications. In this paper a microstrip antenna Artificial neural networks presented as systems of
with rectangular patch configuration is modeled with metal
mounted over a large metal ground plane Its reduced size results
interconnected ‘neurons’ which exchange messages
in narrow bandwidth and the patch dimension parameter between each other. This functions appropriately as a
optimized. The optimization methods include cavity model, CST, human brain. It consists of three layers, input variables,
and TL. In this paper, to reduce the percentage of error and to hidden nodes, and outputs as shown in Fig. 1. Inputs can be
increase the efficiency of design, a model based on artificial of any number and provided in the initial learning phase.
neural network (ANN) is simulated and compared the results. Outputs are the ones obtained after the analysis. During the
process, many intermediate hidden nodes are created that
Keywords— Microstrip antennas, Artificial Neural Networks are essential for optimization. However, the user does not
(ANN), rectangular patch, Radial basis function (RBF).
have any control over them.

I. INTRODUCTION
With the evolution of time, microstrip antennas are gaining
importance, and their sizes are being extremely pushed to
reduce day by day to accomplish more space. However the
random reduction in size results in shifting of resonant
frequency and bandwidth. Hence keeping size and
performance parameters into account, the design of optimized
antenna has become crucial. Several optimization algorithms
have been proposed [1]. The traditional antenna design
methods involve two types;
i) Cavity Method (Analytical)
ii) Full wave method (Numerical)

The cavity method is better than full wave method

with respect to analysis. However both the methods have very
less control over operating frequency and bandwidths. Figure 1: The general structure of ANN
Because it reduces the size of the patch, the operating
frequency shifts to a higher side and bandwidth reduces The optimization process includes two important stages [1]:
gradually. Therefore, a better method of design which can
i) Training or learning phase
optimize both size and frequency is required. In this paper, to
solve the above problem, an optimization algorithm with ii) Analysis phase
artificial neural networking technique is used. Learning phase: This phase uses all the different input signals
to predict the possible outcomes of outputs using precautions
The paper organized as follows; the artificial neural networks learning of previous experiments. These can be accomplished
described in section II. microstrip antennas design procedure either by conducting a large number of experiments or by
explained in Section III. Section IV described the microstrip using values of previously conducted experiments. These
antennas analysis using ANN and section V concludes the values will be stored and used for the upcoming analysis
paper. phase.

978-1-5386-3243-7/17/$31.00 ©2017 IEEE.

114
Authorized licensed use limited to: Universitas Indonesia. Downloaded on December 15,2022 at 00:45:09 UTC from IEEE Xplore. Restrictions apply.
 International Conference on Current Trends in Computer, Electrical, Electronics and Communication (ICCTCEEC-2017)

Analysis phase: After learning all the calculation procedure mapping (activation functions of the individual hidden units).
from the previous phase and is now ready for successfully The kernel used for the proposed control technique is an
providing outputs for any inputs provided. Specially designed exponential Radial Basis Function (ERBF), and it is defined
algorithms can not only predict the accurate values but also by the equation 2.
provide more optimized and best solution for the problem. For
better results, it is advisable to provide a predicted output so xi − x j
that it will be easier for the networks to produce values in that K ( xi , x j ) = e 2σ 2 (2)
range.
Where,  denotes the Gaussian bandwidth. Given an arbitrary
A. RBF algorithm (Radial basis function) point, x, K(x,xi) is a column vector whose ith entry is k(x,xi).
The kernel matrix has the property to be positive definite and
The design of a control system is to establish a model based symmetric which saves the computation time. It follows that
on the equation which maps inputs, x, to the outputs, y. The the objective is to solve for the coefficients, wi, which will best
goal is to make the system produce an optimal response to a fit the function fitness subject to smoothness constraints. The
given excitation. In a supervised neural network using radial coefficients depend on data and satisfy the following linear
basis function (RBF), it is achieved by mapping the inputs into system of equation 3 [3, 5].
the high-dimensional feature space through some nonlinear
mapping [3]. A brief scheme and the technique of RBF is ( K + λI )ω = y d (3)
shown in Fig. 2 and explained below.
Where I is the identity matrix and λ is called the
regularization parameter and yd is the desired response vector.
Above procedure to solve the problem using RBF network
may be summarized as:

Kernel function K(xi,xj) for solving the problem using RBF is

constructed using the input data point, weight vector, w, for a
specified desired response vector, yd, and appropriate
regularization parameter, λ is obtained from the equation (3).
The optimized solution is then defined using equation (1).

III. MICROSTRIP ANTENNAS DESIGN PROCEDURE

A Microstrip or a patch antenna is a metal patch suspended
over a ground plane made of any dielectric [6]. It’s very
simple to fabricate, easy to modify and customize, and
constructed on a dielectric substrate, usually employing the
same sort of lithographic patterning as used to fabricate
Figure 2: Schematic of RBF printed circuit boards. In its simplest form, it consists of a
radiating patch on one side of a dielectric substrate which has
The Fig.2 depicts the RBF network, consisting of three layers.
a ground plane on the other side. They usually use half or
The first layer is the input layer. It consists of m number of
quarter wavelength patch for better performance but a larger
input vectors x. The second layer is a hidden layer, composed
size. One of the key drawbacks of such devices is their narrow
of nonlinear units that are connected directly to all the nodes
bandwidth. To achieve wider bandwidth a relatively thick
in the input layer [4]. There is one hidden unit for each data
substrate used. However, the antenna substrate supports tightly
point xi, i = 1, 2 …N, Where N is the number of the training
bound surface-wave modes which represent a loss in the
sample. Output of ith hidden unit is K(x, xi), and it connected
antenna. The loss due to surface wave modes increases with
to the output layer. The output of the network is a linearly
substrate thickness.
weighted sum of the outputs of the hidden units.
The Microstrip antennas may have a square, rectangular,
The optimized solution for the given data is of the form, and it
circular, triangular or elliptical shape. Use of shapes of a well-
is defined by the equation 1. defined geometry not only simplifies analysis but also helps in
performance prediction [7]. They are also mechanically robust

N
y= w i K(x, x i ) (1) when mounted on rigid surfaces, compatible with MMIC
i=1
designs, and when particular patch shape and mode are
Where, wi is the ith element of the weight vector w and K(x,xi) selected, they are very versatile regarding resonant frequency,
is the kernel function which defines the generalized inner polarization, pattern, and impedance [8]. The size of a
Microstrip antenna is highly proportional to its frequency.
product and is a commonly used tool to perform nonlinear

115
Authorized licensed use limited to: Universitas Indonesia. Downloaded on December 15,2022 at 00:45:09 UTC from IEEE Xplore. Restrictions apply.
 International Conference on Current Trends in Computer, Electrical, Electronics and Communication (ICCTCEEC-2017)

A. Theoretical analysis method: The ANN model is built using four parameters as follows:

The formulas that have been using for the design till now a) Size parameters (W, L, h): The size has been kept as
include the following, Effective dielectric constant is low as possible. In extreme conditions thickness h
calculated by equation 4 [9]. can also be kept varying. However, their variation
will not matter much hence they are almost kept
fixed.
(4)
b) Material parameters (r): It is fixed for a particular
material hence cannot be controlled, and are used
Where  is the dielectric constant of the material, h is the only in the training phase.
height and w are the width of the antenna. c) Performance parameters: (fr, , D): These are the
most critical parameters and its control is important.
Width of the antenna can be calculated using equation (5) It must be optimized properly for better
and (6), performance.

w 8e A
= 2A w
≤2
h (e − 2)
, for h (5)
V. USING THE TEMPLATE
w 2 εr −1
= [B −1− ln(2B −1) + [ln(B −1) +
h ∏ 2εr
0.61
0.39 − ]
εr (6)

For

Where,
Figure 3: The structure of Microstrip antenna [9]

(7) The ANN based approach takes many values of all the
experimental results and then learns to optimize it accurately.
And (8) All the four parameters mentioned above are used as variables
in the neural network [10-12]. In the initial training phase, the
frequency given as an output parameter and the other four as
(9) input parameters and in the next phase, size as the output
parameter and others as the input parameter which shown in
Where, Le=L+L, L is the fringing parameter. Fig.4.

By observing all these, we can see that all the parameters

cannot be controlled or optimized easily.

IV. MICROSTRIP ANTENNA ANALYSIS USING ARTIFICIAL

NEURAL NETWORKS(ANN)
The traditional methods used till now had lots of
disadvantages since size and frequencies are proportional.
Optimizing both the values at the same time has become the
tedious task. Hence an alternative approach for their design
has been proposed here. The ANN design implemented for all
type of patch antennas, but the focus is here mainly on
(a)
microstrip rectangular patch antenna as shown in Fig.3.

116
Authorized licensed use limited to: Universitas Indonesia. Downloaded on December 15,2022 at 00:45:09 UTC from IEEE Xplore. Restrictions apply.
 International Conference on Current Trends in Computer, Electrical, Electronics and Communication (ICCTCEEC-2017)

1.27 1.35 0.163 2.55 6560

1.5 1.621 0.163 2.55 5600

(b)

Figure 4: The schematic ANN model for the (a) Analysis of

parameters and (b) Design of the antenna [1, 10].

The process involves feeding all the available set of values

into the network and allowing it to analyze. Back propagation
algorithm used here uses the sigmoid function for the
optimization. After this phase, the network will be able to
predict the output of any new set of values provided in Table 1
and it is implemented in MATLAB. Fig.5 summarizes the
output resonant frequency obtained from the analysis phase
for varying input widths by keeping other parameters constant Figure 5: Graphical representation of resonant frequency
as, L=0.45 cm, h=0.163 cm, = 2.5. Fig. 5 describes the versus optimized width
optimized value of the width of microstrip patch antennas
differs at the different values of resonant frequency of antenna
at the constant value of all other parameters. V.CONCLUSIONS
The optimization technique using ANN gives best possible
Table I: DATA USED FOR LEARNING PHASE [1] optimization values for all the parameters. It also requires less
time to compute and design. It’s very simple to analyze the
W (cm) L (cm) h (cm)  fmax
(MHz) data. The above results show that the computation errors with
ANN method are less than cavity and TL technique.
5.7 3.8 0.3175 2.33 2310
4.55 3.05 0.3175 2.33 2890
References
2.95 1.95 0.3175 2.33 4240
1.95 1.3 0.3175 2.33 5840 [1] Francismari Noronha dos Santos, Sueli Souza Nascimento, Vitaly F.
1.7 1.1 0.3175 2.33 6800 Rodriguez- Esquerre, and Fabricio G. Simões Filho, “Analysis and
Design of Microstrip Antennas by Artificial Neural Networks,”
1.4 0.9 0.3175 2.33 7700 IEEE MTT-S International Conference on Microwave &
Optoelectronics Conference (IMOC), pp.226-230,2011.
1.2 0.8 0.3175 2.33 8270
[2] Prakash, Geetha, Muralidhar Kulkarni, and U. Sripati, ”Using RBF
1.05 0.7 0.3175 2.33 4730 Neural Networks and Kullback-Leibler distance to classify channel
1.7 1.1 0.1524 2.5 2228 models in Free Space Optics,” IEEE International Conference on
Optical Engineering (ICOE), pp.1-6,2013.
10.8 4.14 0.1524 2.5 2181 [3] THEODOROS EVGENIOU, MASSIMILIANO PONTIL AND
0.85 1.29 0.017 2.22 7740 TOMASO POGGIO, “Statistical Learning Theory: A Primer”,
International Journal of Computer Vision 38(1), 9–13, 2000.
0.79 1.185 0.017 2.22 8450 [4] Baoming Hong and C.H. Chen. "Radial basis function neural Network
2 2.5 0.017 2.22 3970 based non parametric estimation approach for missing data
reconstruction of non-stationary series," International Conference on
1.063 1.183 0.079 2.55 7730 Neural Networks and Signal Processing, pp.75-78,2003.
0.91 1 0.127 10.2 4600 [5] Simon Haykin, “Neural Networks”, A comprehensive foundation,”
2004
1.72 1.86 0.157 2.33 5060
[6] Bablu Kumar Singh, “Design of Rectangular Microstrip Patch Antenna
1.81 1.96 0.157 2.33 4805 based on Artificial Neural Network Algorithm,” IEEE 2nd International

117
Authorized licensed use limited to: Universitas Indonesia. Downloaded on December 15,2022 at 00:45:09 UTC from IEEE Xplore. Restrictions apply.
 International Conference on Current Trends in Computer, Electrical, Electronics and Communication (ICCTCEEC-2017)

Conference on Signal Processing and Integrated Networks (SPIN), pp.1- International Conference on General Assembly and Scientific
4,2015. Symposium (URSI GASS),pp.1-4, 2014.
[7] John D Kraus, Ronald J Marhefka, Ahmad S khan,“Antennas and Wave [11] Zhongbao Wang, Shaojun Fang, Qiang Wang, and Hongmei Liu, “An
propagation,” Fourth edition, Tata McGraw Hill education Pvt.Ltd. ANN-Based Synthesis Model for the Single-Feed circularly
[8] L. Merad, L.Merad, F.T.Bendimerad, Sidi Mohammed Meriah,"Design Polarized Square Microstrip Antenna with Truncated Corners,” IEEE
and resonant frequency calculation of rectangular microstrip antennas," Transactions on antennas and propagation, vol. 60, No. 12, 2012.
International Journal of Numerical Modelling Electronic Networks [12] Abhilasha Mishra, Ganesh Janvale, B. V. Pawar, A.J.Patil, “The Design
Devices and Fields, 2010. of Circular Microstrip Patch Antenna by using Conjugate Gradient
[9] C.A. Balanis, Antenna Theory, Analysis and Design, 3rd ed., John algorithm of ANN,” IEEE International Conference on Applied
Wiley & Sons, New York, 2005 Electromagnetics Conference (AEMC),pp.1-4,2011.
[10] Linh Ho Manh, F. Grimaccia, M. Mussetta, P. Pirinoli, R.E. Zich,
“Multi-Net ANN approach for improving antenna optimization” , IEEE

118
Authorized licensed use limited to: Universitas Indonesia. Downloaded on December 15,2022 at 00:45:09 UTC from IEEE Xplore. Restrictions apply.