Nonlinear system identification using multilayer perceptrons with locally recurrent synaptic structure

Abstract

It is proved that a multilayer perceptron (MLP) with infinite impulse response (IIR) synapses can represent a class of nonlinear block-oriented systems. This includes the well-known Wiener, Hammerstein, and cascade or sandwich systems. Previous methods used to model these systems such as the Volterra series representation are known to be extremely inefficient, and so the IIR MLP represents an effective method of modeling block-oriented nonlinear systems. This was demonstrated by simulations on two models within the class. The significance of the IIR MLP is that it demonstrates that a useful range of systems can be modeled by a network architecture based on the MLP and adaptive linear filters.<>