Linear approximations of nonlinear FIR systems for separable input processes (Q1776437)

The paper deals with the approximation of the nonlinear finite-memory SISO systems by LTI models, optimum in the mean-square output error sense. It is shown that a necessary and sufficient condition for the linear approximation to be an adequate finite impulse response (FIR) model is the separability of proper order of the input signal. The condition is discussed in detail for Gaussian inputs as a special case of separable processes, and some applications of the result are presented.