 Open Access
 Total Downloads : 218
 Authors : Ankit Kumar, Atal Rai
 Paper ID : IJERTV3IS20178
 Volume & Issue : Volume 03, Issue 02 (February 2014)
 Published (First Online): 11022014
 ISSN (Online) : 22780181
 Publisher Name : IJERT
 License: This work is licensed under a Creative Commons Attribution 4.0 International License
Resonant Frequency of Microstrip Antenna using Artificial Neural Network
Ankit Kumar, Atal Rai
Student, Associate Prof.
Department of ECE, IIMT College of Engineering Meerut, India
Abstract : There are some key parameters in RFID reader antenna which are related closely with the antenna structure, such as resonant frequency, return loss and bandwidth in the antenna design process. Structure and properties of the antenna is a complex nonlinear system with complex state which is difficult to make mode by the mathematic method. In this case, neural network is used to express the nonlinear system in this article. Giving a large number of simulation data for the samples, adaptive artificial neural network is used to train network by simulation experiment which is used to verify the fitting degree of neural networks and simulation results. The experiment result shows that artificial neural network can improve the level of computeraided design of micro strip antenna and achieve the antenna design quickly.
Keywords: Microstrip Antenna, Artificial Neural Network(ANN), Resonant Frequency

INTRODUCTION
In 1990, Hansen and Salamon creatively put forward the NNE method. They proved that generalization ability of the neural computing system can be improved obviously by the way of training several NNs and then ensembling the results according to the rule of relative plurality voting or absolute plurality voting. Presently, the NNE studies mainly focus on two aspects: one is how to build/select every individual network, and the other is how to ensemble the outputs of every individual network. Taking the regression problem as an example, matrix inversion has to be carried out for getting combination weights of some conventional methods, and it is affected easily by multidimensional co linearity and noise in the data, which may decrease the generalization ability of the NNE. To solve the problem of multi dimensional co linearity, we can adopt some methods, such as avoiding matrix inversion and restricting combination weights, selective ensemble method, extracting principal components, etc. To decrease the influence of noise, we can also adopt some methods, such as restricting combination weights, adjusting objective function, etc.
Fig.1 Simple neural network

METHODOLOGY
The resonance frequency of microstrip antennas is a phenomenon driven by laws whose behavior can be determined from previously known input/output samples. Empirical models like ANN permit to estimate the resonance frequency for similar cases by interpolation, within the range where the samples were obtained. Feed forward Perceptions Multilayer (PML) and Radial Base Functions (RBF) models were developed, using experimental data presented in references, and compared with the deterministic results presented in through and the empirical results presented in (1) and (2).
In general, PML networks are valid alternatives, but their training is done using the backpropagation algorithm, which can present typical difficulties of the optimization algorithms based on gradient descendent methods. The major difficulties are velocity of convergence and susceptibility to local minima, increasing the computational effort and decreasing the interpretability/ transparency of the results. Good alternatives to the PML networks are the RBF ones. Many different algorithms can be used with the feed forward architecture. A classical algorithm used for the training of PML neural networks is the gradient conjugate, a heuristic method. The Newton analytical method is an alternative to the gradient conjugate method, for a rapid convergence. The basic equation for the Newton method is:
+1 = 1 (1)
Where H is a Hessian matrix (second partial derivatives) of the performance index at the current values of the weights and biases. Newtons method often converges faster than conjugate gradient methods. Unfortunately, it is complicated and expensive to compute the Hessian matrix for feed forward neural networks. There is a class of algorithms that is based on Newtons method, but it doesnt require calculation of second derivatives. These are called quasi Newton (or secant) methods. They update an approximate Hessian matrix at each iteration of the
algorithm. The update is computed as a function of the gradient. The quasiNewton method that has been most successful in published studies is the Broyden, Fletcher, Goldfarb, and Shanno (BFGS) update, whose basic equation is:
? = . () 1 . () (2)
Where J(x) is Jacobean matrix and e(x) represents the errors. BFGS algorithm was used for rectangular antennas. For triangular antennas, the algorithm that presented best results was the Rprop – Resilient Propagation, described in reference. In the Rprop algorithm the weights and learning rates are modified only once in each training epoch. Each weight wji has its own variation rate (ji), which depends of the time t, in the following way:
Here for the analysis design by has been presented by considering the ANN model with inputs as width (W), length (L), height (h) and permittivity constant (r) and output parameter as resonant frequency (fr).
This model was developed by considering scale without normalization. It gives accuracy up to 99.78% for training and 99.858% for testing. The NN Model has 4 inputs and 1 output as given in fig 5.1.
Fig.1.2: Four input & one output ANN Model
+ 1 ,
1 .
> 0
Here for the ANN model we are using input width between
0.93 to 1.49 ranges and length between the ranges 0.806 to 1.045, as well as for thickness 1.5 is minimum value and 9.5
=
1 ,
1 . > 0
is maximum value. Permittivity constant (r) value is constant between 2.082.50
1 ,
(3)
Where 0 < < 1 < +.
Changes on the weights occur only when the sign of the partial derivatives with respect to the weights change, and are independent of their magnitudes. For triangular antennas, PML networks with Rprop algorithm was more efficient than those based on conjugate gradient method or those based on second order gradient, like Levemberg Marquardt. However, in the applications of this paper, RBF models presented the best results for both, rectangular and triangular antennas. The basic equation for the RBF network is:

FLOW CHART
The flow chart for the analysis of the artificial neural network is given in fig. (1.3). Where in first decision box we decide that if the no. of epoch achieved, it stop otherwise it goes to the next stage. In next stage if desired accuracy have been achieved, it stop otherwise we will check the program. If it is learning over the result, we delete the neurons which are not requiring for the process. On the other hand if learning is under the result than we have to add some more neurons according to the requirement. We will perform the same process still training close to the target. After this whole process we will test the trained model. And after this whole process we achieve required result. It gives the accuracy up to 99%.
=0
(n+1) =
(4)
This is main reason so we are using artificial
Where is a scalar function radially symmetric with as its center. The operator . is the Euclidian norm and gives the modulus of the argument vector. Further details about the training of this kind of ntwork can be found in references.
The data sheet for input and output variables is given in table which has 26 input and output data
.
Fig1.3 Flow Chart
RESULTS We calculate the resonant frequency a Microstrip antenna, using its parameters like width (W), length (L), permittivity of the substrate (r) and height (h) of the substrate. We apply artificial neural network technique for optimization of L. The optimized L is used for calculating the resonant frequency of rectangular Microstrip patch antenna.
we are using input variables width (w), length (L), permittivity of the substrate (r) and thickness (h) and at output parameter is resonant frequency (fr).
Here, we are using input width (w) between 0.93 to 1.49 ranges, length (L) between the ranges 0.806 to 1.045, as well as for thickness (h) 1.5 is minimum value and 9.5 is maximum value. Permittivity constant (r) value is constant between 2.082.50.
TABLE1
Input
Output
(W)
Mm
(L)
mm
(T)
(h) mm
permittivity (r)
Resonant frequency (fr)
124.9
91.9
2.2
2.08
1050
92.2
80.6
1.9
2.08
1040
103.0
84.8
4.2
2.50
1060
137.0
94.4
9.2
2.10
890
143.3
91.3
5.1
2.10
930
111.3
103.2
1.7
2.20
860
114.0
81.3
5.6
2.50
960
93.0
81.0
8.8
2.20
1060
98.1
83.0
4.5
2.25
960
113.7
93.7
9.5
2.25
870
127.1
82.4
2.6
2.25
1040
149.2
82.4
2.3
2.25
890
110.1
104.5
1.5
2.25
840
122.2
110.0
8.5
2.50
770
120.4
100.4
6.7
2.50
955
130.4
102.0
8.0
2.50
992
134.0
102.0
6.7
2.50
812
126.0
100.0
7.8
2.50
995
100.0
100.6
1.8
2.50
980
135.6
100.0
2.3
2.50
1010
132.0
103.4
2.4
2.25
1039
124.5
104.5
2.5
2.25
974
137.4
100.7
2.4
2.25
1012
123.4
98.7
3.4
2.25
886
100.0
94.5
4.4
2.25
942
104.5
88.8
5.5
2.25
988
100.2
89.3
3.3
2.25
987
122.4
89.3
5.6
2.25
1096
Fig.2The performance plot for the epoch is
(W)
(L)
(h)
(r)
1.101
1.045
1.5
2.25
1.222
1.100
8.5
2.25
1.204
1.004
6.7
2.50
1.304
1.020
8.0
2.50
1.340
1.020
6.7
2.50
1.260
1.000
7.8
2.50
1.224
0.893
5.6
2.25
Fig.2.1 The regression plot for the epoch is
Testing output of resonant frequency for Microstrip antenna is:
y1 = 10.6227
y2 = 9.1522
y3 = 10.0329
y4 = 9.6663
y5 = 10.0867
y6 =9.6709
y7 =9.5835
Fig.2.2 The performance plot for training is given in
Fig2.4The training state plot for the training
Fig.2.5 The regression plot for the training

CONCLUSION
From the above experimental results, artificial neural network has the faster convergence rate and the smallest error. This proves the effectiveness of neural network to solve the problem of the design and calculation of micro strip antennas different parameters. The neural network method can significantly improve the level of computer aided design of micro strip antenna to achieve the calculation of antennas parameters quickly.
In future we can use PSO to solve the problem of the design and to calculation the different parameters of micro strip antenna. Because the optimization of the Microstrip Patch is partially realized this concludes that the PSO code is functioning correctly
REFERENCES

Guo zhitao, Yuan jinli and GU Junhua, Design and Research of RFID Microstrip Antenna Based on Improved PSO Neural Networks IEEE Trans. 2010 pp. 251255.

E. R. Brinhole, J. F. Z. Destro, A. A. C. de Freitas, and N. P. de Alcantara Jr, Aetermination of Resonant Frequencies of Triangular and Rectangular Microstrip Antennas, Using Artificial Neural Networks Progress In Electromagnetic Research Symposium 2005, Hangzhou, China, August 2226, pp. 579582.

M.NaserMoghaddasi, Pouya Derakhshan Barjoei, Alireza Naghsh, A Heuristic Artificial Neural Network for Analyzing and synthesizing Rectangular Microstrip Antenna IJCSNS International Journal of Computer Science and Network Security, VOL.7 No.12, December 2007.pp.278281.

Kapil Goswami, Ashutosh Dubey, Girish Chandra Tripathi And Birbal Singh, Design and Analysis of Rectangular Microstrip Antenna with PBG Structure for Enhancement of Bandwidth Global Journal of Research Engineering Volume 11 Issue 2 Version 1.0 March 2011.

Maja Sarevska, and AbdelBadeeh and M. Salem, Antenna Array Beam forming using Neural Network World Academy of Science, Engineering and Technology 24 2006 pp. 115121.

Abhilasha Mishra, Vaishali Bhagile, Smita Kasar, P.M.Patil, Effect of Normalized Scale on Design of Rectangular Microstrip Antenna by using FFBP Abhilasha Mishra et al. / (IJCSE) International Journal on Computer Science and Engineering Vol. 02, No. 03, 2010, 626 629.

J.Lakshmi Narayana,Dr.K.SriRamaKrishna and Dr.L.Pratap Reddy, Design of Microstrip Antennas Using Artificial Neural Networks International Conference on Computational Intelligence and Multimedia Applications 2007, IEEE DOI 10.1109/ICCIMA.2007.53.

Alka Verma and Neelam shrivastava, Analysis and Design of Rectangular Microstrip Antenna in X Band MIT Interational
Journal of Electronics and Communication Engineering Vol. 1, No. 1, Jan 2011, pp (3135).

R. Garg, P. Bhartia, I. Bahl, and A. Ittipiboon, Microstrip Antenna Design Handbook, Artech House, 2000.

K. F. Lee, Ed., Advances in Microstrip and Printed Antennas, John Wiley, 1997.

C. A. Balanis, Antenna Theory, Analysis and Design second edition, John Wiley & Sons, New York, 2009.