Microwave Modeling Using Neural Networks and Iir Filters in Time and Frequency Domains

A method combining Infinite Impulse Response (IIR) filters and Neural Networks (NN) is proposed for precise electromagnetic (EM) of complex structures true the use of efficient optimization and parameterization techniques. This method is based on the characterization of the time response as a transfer function using a digital filter. The use of the neural network allows then the modeling of the geometric variation of the studied structure. The method is then also expanded in the frequency domain for studying the variation of the scattering parameters for the studied structures as a function of frequency. This is achieved by building a spectral transfer function. The main difference between time and spectral analyses is discussed and it is shown that it is more advantageous to interpolate the transfer function's poles and zeros rather than the polynomial coefficients The validity of our proposed technique is demonstrated on a microstrip step in width discontinuity and a microstrip filter, both analyzed in the time domain and also on a chamfered bend analyzed in the frequency domain.