Title :
On incorporating signal processing and neural network techniques with the FDTD method for solving electromagnetic problems
Author :
Litva, J. ; Wu, C. ; Navarro, E.
Author_Institution :
Commun. Res. Lab., McMaster Univ., Hamilton, Ont., Canada
Abstract :
The finite difference time domain method (FDTD) is a very powerful numerical method for solving electromagnetic (EM) problems. Due to its flexibility, it can be used to solve problems which have very complex boundaries. It is well known that the FDTD method requires long computation times for simulating the resonant or high-q structures. The reason for this is because the algorithm is based on the leap-frog formula. The quest to find a good predictor for enhancing the method is very interesting topic in EM modelling. In this paper, the autoregressive model (AR) and backpropagation neural network are designed as predictors. The total least squares (TLS) method is applied to obtain AR coefficients. A waveguide filter is used as an example and modeled using the FDTD method. We demonstrate that a short segment of an FDTD data set can be used to train the predictors and the predictors can predict later information very well
Keywords :
autoregressive processes; backpropagation; electromagnetic field theory; finite difference time-domain analysis; least squares approximations; multilayer perceptrons; rectangular waveguides; signal processing; waveguide filters; waveguide theory; AR coefficients; FDTD method; autoregressive model; backpropagation neural network; electromagnetic problems; finite difference time domain method; neural network techniques; predictors; signal processing; total least squares; training; waveguide filter; Backpropagation algorithms; Computational modeling; Finite difference methods; Least squares methods; Neural networks; Predictive models; Resonance; Signal processing; Signal processing algorithms; Time domain analysis;
Conference_Titel :
Signals, Systems, and Electronics, 1995. ISSSE '95, Proceedings., 1995 URSI International Symposium on
Conference_Location :
San Francisco
Print_ISBN :
0-7803-2516-8
DOI :
10.1109/ISSSE.1995.498000
