One approach for training of recurrent neural network model of IIR digital filter

Abstract

One approach for training of recurrent neural network model of 1-D IIR digital filter is proposed. The sensitivity coefficients method has been applied in the training process of the neural network. The set of time domain data is generated and used as a target function in the training procedure. The modeling results have been obtained for two different cases - for 4-th order bandpass IIR digital filter and for partial response IIR digital filter. The frequency domain behavior of the neural network model and the target IIR filter has been investigated. The analysis of the frequency responses shows good approximation results. © Springer Science+Business Media B.V. 2010.