You are on page 1of 4

Design Optimization Of Loop Antenna Using

Competitive Learning ANN


Kumaesh Sarmah Kandarpa Kumar Sarma
Department of Electronics and Communication Engineering Department of Electronics and Communication Technology
Tezpur University Gauhati University
Assam-784 028, INDIA Guwahti-781014, INDIA
Email: kumaresh4you8@gmail.com Email: kandarpaks@gmail.com

Abstract—Out of several antenna design techniques the Artificial  Finite Element Analysis and Modeling (FEM)
Neural Network (ANN) based method is suitable for prediction of  Use of Genetic Algorithm (GA)
characteristic parameters of loop antenna by considering
 Use of Artificial Neural Network (ANN)
transmit – receive conditions of practical communication set-ups.
The predicted set of parameters can be used to fix dimensions of
a loop antenna which involves theoretical calculations. This work All these methods are efficient and widely used for
proposes an approach to determine the best suitable combination various types of antenna design problems. Antenna design
of conductor thickness and loop radius using Competitive problems are formulated in terms of integral equations and
Learning ANN from predicted values of antenna parameters. The solving these integral equations we can get the exact solution
proposed method uses the ANN predicted parameters to find the of the model. The Method of Moment (MOM) technique is
optimized set of conductor thickness and loop radius using Self mostly used for solving these integral equations. In the MOM
Organizing Map (SOM) to fix the layout of a loop antenna for technique, complex mathematical terms are converted to
applications with electrically driven finite element grids.
simple matrix equation known as discretization [7], [8]. Two
Keywords- Loop antenna, ANN, SOM, Optimization. other numerical approaches for antenna analysis are finite
element frequency domain method and finite difference time
domain method. The Finite Difference Time Domain Method
I. INTRODUCTION
(FDTD) is a computational procedure for solving Maxwell’s
Antenna design using theoretical approaches and custom – equation based on clever algorithm. In FDTD, an antenna
built softwares provide satisfactory results despite respective problem is formulated differently for transmitting and
limitations. But there always exists possibilities to explore receiving antennas [9], [10]. The Finite Element Analysis and
innovative approaches for such design involving soft Modeling (FEM) is a numerical procedure to obtain solution to
computational tools like Artificial Neural Network (ANN) boundary value problems of antenna in time and frequency
which can fix design requirements considering environmental domain with the aid of a computer [11], [12]. For solving these
variations as observed in practical communication set-ups.
numerical techniques different custom based software are
ANNs like Multi Layer Perceptron (MLP) s have been used for
several antenna designs [1], [2], [3] due to their capacity to available. Antennas such as micro strip patch antennas,
learn given patterns and use such knowledge for predicting dielectric lens antennas, broadband antennas, reflector
design parameters. The works cited in [5], [6] have shown how antenna, horn antenna, phased array antennas, smart antennas
MLPs can be used to predict different characteristic features of are widely modeled by using these techniques and these
a loop antenna for high frequency applications. This work is a models are coded and computed in different customized
follow up to those approaches for electrically small designs software. Genetic algorithm (GA) is another technique used
where conductor thickness and loop radius of a loop antenna is for design optimization of different antenna structures by
optimized using Self Organizing Map (SOM) so that physical minimizing or maximizing the objective parameter or cost
dimensions can be ascertained for assumed frequency range of function. Simulated annealing and particle swarm optimization
200 to 300 MHz’s. (PSO) are some techniques of GA used in antenna optimized
design [13], [14]. ANNs are suitable for prediction and
II. ANTENNA DESIGN TECHNIQUES- A BRIEF optimization of design layouts with their capacity to learn
OVERVIEW complex patterns fed to them. ANNs are used for design of
micro - strip patch antenna where apriori knowledge of fields
There are five well known methods for antenna design and and radiation patterns help in optimizing physical dimensions
modeling [7]. These techniques are, [1], [2], [3],[4].
 Integral Equation (IE) and Method of Moment (MoM)
 Finite Difference Time Domain Method (FDTD)

978-1-4244-9581-8/11/$26.00 © 2011 IEEE

Authorized licensed use limited to: National Institute of Technology- Delhi. Downloaded on February 05,2021 at 11:21:07 UTC from IEEE Xplore. Restrictions apply.
TABLE I. LIMITS OF LOOP RADIUS AND WIRE RADIUS FOR TWO
DIFFERENT RANGES OF RADIATION EFFICIENCY.

Radiation Efficiency( η) 70% 90%


Operating Loop Radius ‘a’ 0.048 m 0.048 m
Frequency at
Wire radius ‘ b’ 1.28 × 10-3 m 4.49 × 10-3 m
200 MHz
Operating Loop Radius ‘a’ 0.032 m 0.032 m
Frequency at
Wire radius ‘ b’ 1.044 × 10-3 m 4.035× 10-3 m
300 MHz

Radiation resistance of loop antenna with a being the loop


radius and b being the wire radius is given as,
2
 a2 
Rr  31,171  2


Where S = Loop area = πa2 , λ = wavelength.
Figure 1: Block Diagram of the proposed Model. Loss resistance is given as,

III. PROPOSED MODEL AND EXPERIMENTAL DETAILS a f 0


Rl 
b 
Previously a work was reported in [5], which deals with The constant parameters are σ =5.8 × 107 mho/m and μ0 =
the prediction of four characteristic parameters for electrically 4π × 10-7. The radiation resistance and loss resistance of an
small and large loop antennas for several radius – thickness antenna determines the radiation efficiency. It is the ratio
pairs. The optimization was with respect to dimensions. between the radiation resistances to the total resistance of the
Antoher work is reported in [6], where a single dimension pair loop.
of thickness and radius is considered and four different
Rr
parameters are predicted. That work has been extended further eloop ( ) 
to include two more parameters. These are-. for electrically Rr  Rl
driven finite element based loop antenna designs. From these The loop antenna with constant current approximation is
six parameters, the dimesion of the loop antenna is to be given as 2πa ≤ 0.2λ.
optimized in terms of frequency of operation at a given point
of time for an electrically small loop antenna. Considering a loop antenna working between 200 MHz
and 300 MHz the loop radius ‘a’ ranges between 0.048 m to
A. Generation of Radiation Efficiency: 0.032 m. The wire radius ‘b’ can be approximated from its
In this present model (Figure 1), the optimization of the radiation efficiency η. In the proposed model we consider the
dimension of a single turn loop antenna is carried out by cases of efficiency greater than or equal to 70 %. The ratio of
considering its radiation efficiency (η). The radiation radiation resistance and loss resistance for this efficiency range
efficiency η depends on radiation resistance (Rr) and loss is given as,
resistance (Rl) of an antenna. Again the radiation resistance Rr
and loss resistance is directly related to the physical dimension For 70 % efficiency Rl 
of loop antenna i.e. loop radius a and wire radius b and the
2.33
frequency of operation ‘f’ [15],[16],[17]. Radiation resistance R
of a loop antenna is defined as the resistance in series with the
And for 90 % Rl  r
9
antenna that will consume the same amount of power as it Table 1 summarizes the loop radius and wire radius for
actually radiated. It is the ratio between power radiated by the different operating frequencies.
antenna and the square of the root mean square value of the
maximum current flowing in the antenna at the best matched Now, considering the frequency range 200 – 300 MHz, a
condition. The radiation and loss resistances of an antenna total of 121 values of efficiency are generated for 11 pairs of a
determine the radiation efficiency. It is the ratio between the and b. The optimization problem now is to determine the best
radiation resistances to the total resistance of the loop. Loss combination of a and b such that efficiency falls within the
resistance of an antenna depends on several factors. A few of stipulated range and the problem of determining the physical
them are resistance due to dielectric losses, brushing losses, layout of the antenna becomes feasible. This can be carried out
loss by leakage over insulation, resistance due to conductor by doing an inspection of the generated values and by plotting
losses in antenna and earth and eddy current losses. graphs of efficiency values against varying frequencies for
given pairs of a and b.

Authorized licensed use limited to: National Institute of Technology- Delhi. Downloaded on February 05,2021 at 11:21:07 UTC from IEEE Xplore. Restrictions apply.
TABLE II. PARAMETERS USED FOR CONFIGURING SOM
OPTIMIZATION BLOCK

Item Description of Values


Initial input values Between minimum and maximum values
of input
Size of SOM neighborhood 11×2
Competitive layer topology Hexagonal
Distance function Linkdist can be Dist or MANDIST
Ordering phase learning rates 0.5 to 0.9
Tuning phase learning rates 0.02 to 0.5
Tuning phase neighborhood 1
distance

B. Optimization of Dimension:
Figure 2: Initial SOM topology
For an automated system, a better solution shall be certain
computational aid which in this case is executed using a SOM with a grid topology of 11 × 11 which is designed to map to a
based optimization. The SOM is trained with 11 sets of 11 × 2 neighbourhood implying the fact that there are 11 sets
efficiency values generated for 11 different frequencies for 11
of (a, b)s. At the end of the stipulated number of training
pairs of a and b. The optimization Problem we can define as
sessions, the SOM converges to the required limit. At this
best {ai, bi} such that η is maximum. Let η be the expected
efficiency and  be the actual efficiency. The error component point for a single set of value of, say a, the neighbourhood
takes a form as depicted in Figure 3.
is given as,

1 n
e[ f ]  [ [ f ]  i [ f ]]
2 i 1 i

The error can be minimized using 


e[ f ] for a given
f
set of a and b so that  is maximum which is greater from
ANN predicted values.
The SOM block is used to carry out the approximation such
that yj = y0j T Wj , j = 1,2,………….N.
Where y0j represent the 11×11 matrix of efficiency values
for 11 combinations of a and b and frequencies between 200 to
300 MHz. Also, Wj is the, random set of weights assigned
initially to the competitive layer of the SOM. The “winners
take all’’ based updation creates a value assigned to the winner Figure 3: SOM topology after training for a single parameter set
neuron [19] such that,
The results provided by this convergence are used to
y j  max j [ yoj ,W j ] calculate the radiation efficiency which in this case reaches the
optimum level which is verified using theoretical
For this, a SOM structure is created. The parameters considerations. The model is thus ready to handle dimension
considered are shown in Table II. optimization problems despite variations in the transmit –
receive condition. The system is tolerant to ±50% variation in
IV. RESULT AND DICUSSION input parameters. Some of the (a, b) values generated by the
The 11 sets of (a and b) used for generating the radiation SOM optimizer is provided in Table IV. These are the values
efficiency for 11 different frequencies provide values some of generated during training phases of the system which provides
which are shown in Table III. To start with the SOM the best case combinations for the frequency range considered.
neighborhood is as in Figure 2. These sets also provide the most suitable radiation efficiency
values despite input parameter variations. The most likely set
After training, the SOM generates the required set of (a, b) of (a, b) values is provided by the SOM with a neighbourhood
such that the efficiency is not only within the stipulated range as depicted in Figure 4. Thus the system provides a means to
but also maximizes the values. The SOM initially is configure

Authorized licensed use limited to: National Institute of Technology- Delhi. Downloaded on February 05,2021 at 11:21:07 UTC from IEEE Xplore. Restrictions apply.
generate an optimized dimension which can be translated for
activating finite elements of electrically driven grid arrays.
TABLE IV. OPTIMIZED SET OF (A, B) GENERATED BY SOM.

Loop radius ‘a’ Wire radius ‘b’ in


in meter meter
0.0320 0.0010444
0.0335 0.001399
0.0350 0.0017544
0.0365 0.002109
0.0380 0.0024644
0.0395 0.0028194
Figure 4: SOM topology to generate the most likely set of (a, b)
0.0410 0.003174
values
0.0425 0.003529
0.0440 0.003884 [5] K. Sarmah and K. K. Sarma : “ Artificial Neural Network Based
Parameter Estimation and Design Optimization of Loop Antenna”,
0.0455 0.004239 International Journal of Electrical and Information Engineering, Vol. 4,
0.0470 0.00459 No. 3, pp.172 -178, 2010.
[6] K. Sarmah and K. K. Sarma : “ Prameter Prediction and Optimization of
V. CONCLUSION Loop Antenna Design using ANN”,4th Indian International Conference
on Artificial Intelligence(IICAI 09) pp.820-827, 2009.
This work proposes an approach to determine the best [7] C. A. Balanis, Modern antenna Handbook , John Willy & Sons Inc
suitable combination of conductor thickness and loop radius Publication, 2008.
[8] J. J. H. Wang: “Generalized Moment Methods in Electromagnetics
using Competitive Learning ANN from predicted values of Formulation and Solution of Integral Equations”, John Wiley & Sons,
antenna parameters. The proposed method uses the ANN Hoboken, NJ, 1991.
predicted parameters to find the optimized set of conductor [9] A. F. Peterson, S. L. Ray and R. Mittra: “Computational Methods for
thickness and loop radius using Self Organizing Map (SOM) Electromagnetics, IEEE Press”, New York, 1998.
[10] P. A. Tirkus and C. A. Balanis: “Finite-difference time-domain method
to fix the layout of a loop antenna for applications with for antenna radiation”, IEEE Trans. Antennas Propag., Vol. 40, pp. 334-
electrically driven finite element grids. The work shows how 340, 1992.
a SOM can be used to select the most likely set of values to [11] T. W. Hertel and G. S. Smith: On the convergence of common FDTD
determine the dimension of a loop antenna which can be used feed models for antennas, IEEE Trans. Antennas Propag., Vol. 51, pp.
1771 - 1779, 2003.
for constituting a physical layout so that the design can handle [12] J. M. Jin: “The Finite Element Method in Electromagnetics, 2nd ed.”,
variations observed in transmit-receive conditions. John Wiley & Sons, Hoboken, NJ, 2002.
[13] Z. Lou and J. M. Jin: A novel dual - field time - domain finite – element
domain - decomposition method for computational electromagnetics,
REFERENCES IEEE Trans. Antennas Propag., Vol. 54, pp. 1850 - 1862, 2006.
[1] V. R. Gupta and N. Gupta: “An Artificial Neural Network Model for [14] R. L. Haupt and S. E. Haupt: “Practical Genetic Algorithms,” 2nd ed,
Feed Position of the Microstrip Antenna”, Electronika Ir John Wiley & Sons, Hoboken, NJ, 2004.
Elektrotechnika, pp. 87-89, 2005 Nr. 4(60). [15] E. Altshuler and D. Linden: “Wire-antenna designs using genetic
[2] R. K. Mishra and A. Patnaik: “Neural Network Based CAD Model for algorithms,” IEEE Antennas Propag. Mag., Vol. 39, No. 2, pp. 33-43,
the Design of Square-Patch Antenna”. IEEE Transactions on Antenna 1997.
And Propagation, vol. 46, NO. 12, December 1998. [16] J. D. Kraus, Electromagnetics 4th Edition., McGrow-Hill Book. Co.,
[3] K. Siakavara: “Artificial Neural Network Employment in the Design of New York, 1992.
Multilayred Microstrip Antenna with Specified Frequency Operation”, [17] C. A. Balanis, Antenna Theory, Analysis and Design, John Willey and
Proceedings of PIERS , pp. 27 - 30, Prague, 2007. Sons INC, New York, 2005.
[4] A. A. Eldek: “ Design Of Double Dipole Antenna With Enhanced [18] G. S. Smith: “Loop antennas (chap 5)”, Antenna Engineering Handbook,
Usable Bandwidth for Wideband Phased Array Application”, Progress R.C. Johnson (Ed.), McGraw- Hill, New York, pp. 5.1-5.29, 1993.
In Electromagnetics Research, PIER 59, pp. 1-15, 2006. [19] S. Haykin, Neural Networks: A Comprehensive Foundation . 2nd Ed ,
Pearson Education, New Delhi, 2003.

TABLE III. NORMALIZED VALUES OF Η GENERATED BY 11 SETS OF (A , B) FOR THE FREQUENCY RANGE.
0.3604 0.478 0.580 0.665 0.733 0.787 0.830 0.863 0.889 0.910 0.926
0.386 0.506 0.608 0.690 0.755 0.806 0.845 0.876 0.900 0.918 0.933
0.412 0.533 0.633 0.712 0.774 0.822 0.859 0.88 0.909 0.926 0.939
0.438 0.559 0.657 0.733 0.792 0.837 0.871 0.897 0.917 0.933 0.945
0.462 0.583 0.679 0.752 0.808 0.850 0.882 0.906 0.924 0.939 0.950
0.486 0.606 0.699 0.770 0.822 0.862 0.891 0.914 0.931 0.944 0.954
0.509 0.628 0.718 0.786 0.835 0.872 0.900 0.921 0.936 0.949 0.958
0.531 0.649 0.736 0.800 0.847 0.882 0.907 0.927 0.941 0.953 0.961
0.553 0.668 0.752 0.813 0.858 0.890 0.914 0.932 0.946 0.956 0.965
0.573 0.686 0.767 0.826 0.868 0.898 0.921 0.937 0.950 0.960 0.967
0.593 0.703 0.781 0.837 0.877 0.905 0.926 0.942 0.954 0.963 0.970

Authorized licensed use limited to: National Institute of Technology- Delhi. Downloaded on February 05,2021 at 11:21:07 UTC from IEEE Xplore. Restrictions apply.

You might also like