International Journal of Electronics and Computer Science Engineering 639
Available Online at www.ijecse.org ISSN- 2277-1956
ISSN 2277-1956/V1N2-639-641
Analysis of EM scattering in Waveguide Filter using
Neural Network
Manidipa Nath 1, Bhaskar Gupta 2,
1
Jadavpur University, ETCE department, Kolkata700032, West Bengal, India
2 Jadavpur University, ETCE department,
Kolkata700032, West Bengal, India
Email: 1 manidipa.deoghar@gmail.com, 2gupta_bh@yahoo.com,
Abstract- This paper discusses the application of Neural Network (NN) technique in the modeling of a typical electromagnetic (EM) field scattering problem in a waveguide filter structure [1]. The structure under consideration is a rectangular waveguide with four dielectric circular rod inserted symmetrically. The Neural Network takes into consideration the geometrical and material parameters of the EM model of the structure and the reflection coefficient values for different frequencies in the frequency band of interest. The Neural Network is trained off line using training data sets generated by the image theory. The work is to find out an acceptable Neural Network model of the waveguide structure having four dielectric post, verification using Method of moment by simulation and the corresponding filter structure practically implemented in hardware for verification.
Keywords: Neural Network, EM scattering, Waveguide filter.
1. Introduction
A filter transmits energy in on one or more passbands and attenuates energy in one or more stopbands. The response function of the filter is defined in terms of location of the poles of the insertion loss function and zeros within the passband. Microwave and Millimeter wave filters are essential component in modern broadband communication systems. Their electrical performance is crucial for overall system design. Low insertion loss, high return loss and high slope selectivity are simultaneous requirements along with a low cost, mass producible, tuning free, compact design. So very few filter structures among a large variety becomes suitable and realizable to satisfy the required criterion. Thus modern communication systems require computer aided design (CAD) techniques for accurate modeling, analysis and design of filters.
2. Neural Network Approach
IJECSE,Volume1,Number 2
Manidipa Nath and Bhaskar GuptaISSN 2277-1956/V1N2-639-641
3. Formulation of the problem
Several techniques have already been applied for analysing the scattering behavior of rectangular waveguides having circular dielectric posts as discontinuities. The structures are investigated with the moment method and as well as mode-matching method. Here moment method was used for analyzing the scattering properties of vertical full height dielectric post and theoretical computation based on image theory [2] was used to generate training data for the neural network model The reflection and transmission matrices for four dielectric post can be derived in closed form [3] using the lattice sums and the T-matrix of a circular cylinder in free space. TE10 mode is
assumed to be propagating inside the waveguide at X band for which the frequency response of the power reflectance and transmittance has to be calculated from the ANN model. The scattering properties of the vertical posts are analyzed by method of moments and the scattered fields are calculated from the lattice sum and the transition matrices. Analysis results have been used to generate S parameter data for the configuration in order to train the proposed neural network.
Figure (1): 3D view of the Four Circular Post in Rectangular waveguide
To develop a neural network model we need to identify input and output parameters of the structure under consideration to generate and preprocess data for carrying out ANN training. Generally Multi Layer Perceptron (MLP) model is being chosen where weights are initialized by assigning with small random values. The purpose of the neural network training is to adjust the weights such that the error function is minimized. As the error function is a nonlinear function of the adjustable weight parameters, iterative algorithms are used to update it with an appropriate learning rate. The back propagation training algorithm updates weights along the negative direction of the gradient of the training vector. Finally the quality of the neural network model is evaluated with an independent set of data and a quality measure is performed based on average test error and standard deviation.
641 Analysis of EM scattering in Waveguide Filter using Neural Network
ISSN 2277-1956/V1N2-639-641
Figure (2a) and (2b): Comparison of response of the Neural Network model with the computed values of Magnitude of reflectance and transmittance of the dielectric posts embedded in rectangular waveguide
5.Observations
The trained neural network model is found to be working within the error boundary limited by the practical error function. After proper training the average relative error is found to be 0.1 for Mag (S11) and 0.05 for Mag (S21) over a set of test data spanning the entire X band of frequencies and different values for the position and dielectric constant of the dielectric posts, which have not been used in the training process. The trained neural network model is found to be working within the error boundary limited by the practical measurement error of mag (S11) (fig 2a.) & mag(S21) (fig 2b.) for a four dielectric post in rectangular waveguide.
Figure 3. The Simulated Filter response of the structure(with the four dielectric rod inserted within the waveguide)
5.00 7.50 10.00 12.50 15.00
Freq [GHz] -37.50
-25.00 -12.50 0.00
Y
1
XY Plot 13
ANSOFTCurve Info
dB(S(WavePort1,WavePort1)) Setup3 : Sw eep1
dB(S(WavePort1,WavePort2)) Setup3 : Sw eep1
0.4 0.42 0.44 0.46 0.48 0.5 0.52 0.54 0.56 0.58 0.6
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Normalized Fequency
M
a
g
S
1
1
Neural Network Model of composite dielectric post
NN-output Target-output
0.4 0.42 0.44 0.46 0.48 0.5 0.52 0.54 0.56 0.58 0.6
0.7 0.8 0.9 1 1.1 1.2
Normalized Frequency
M
a
g
S
2
1
Neural Network Model of composite dielectric post
IJECSE,Volume1,Number 2
Manidipa Nath and Bhaskar GuptaISSN 2277-1956/V1N2-639-641 Figure 4. The fabricated waveguide filter
Conclusion
Microwave filters on EBG structure is being designed using multi-post neural network models and their performance can be optimized using Genetic algorithm. The trained network with a suitable optimization programme using GA can yield a faster and accurate filter solution. In this work neural network technique is applied in the design of microwave filters consist of multiple vertical dielectric posts in rectangular waveguide and performance can be optimized using GA. Here NN based model is used as starting point for optimizer to reduce computational efforts significantly. NN has been used for non-linear modeling of the frequency response of the scattering properties of four circular dielectric cylinder in a three dimensional rectangular waveguide in X band. Analysis using lattice sum and T– matrix method has been done to generate S11 and S21 data for the structure and 80% of them are used for proper training of the neural network. Rest of the data generated is used for testing the model for the required accuracy. This type of model is very useful in optimization problems [4] where a fast and accurate response is required which cannot be obtained from rigorous electromagnetic simulation. This novel technique of modeling electromagnetic field problem implementing NN approach has great potential in the analysis of various problems in the field of electromagnetic.
Microwave filters on EBG structure is being designed using multi-post neural network models and their performance can be optimized using Genetic algorithm. The trained network with a suitable optimization programme using GA can yield a faster and accurate filter solution. The NN is applied in the design of microwave filters consist of multiple vertical dielectric posts in rectangular waveguide and performance can be optimized using GA. Here NN based model is used as starting point for optimizer in order to reduce computational efforts significantly.
References
[1] Q. J. Zhang, K. C. Gupta, and V. K. Devabhakuni, “Artificial Neural Networks for RF and Microwave Design-From Theory to Practice,” IEEE Trans. on MTT -51, No.4, 1339-1350,April 2003.
[2] K.Yasumoto and H. Jia, “Modeling of photonic crystals by multi layered periodic arrays of circular cylinders,” in Electromagnetic Theory
and Applications for Photogenic Crystals (Optical Science and Engineering Series, Vol. 103), ed. K. Yasumoto, Chap 3, pp 123-190,CRC
Press 2005.
[3] K.Yasumoto, N.Koike, H.Jia, and B. Gupta, “Analysis of Electromagnetic Band gap Based Filters in a Rectangular Waveguide,” IETCE Trans. Electron., Vol. E89-C, No. 9, pp 1324-1329, September 2006.