LIMITATIONS ON THE USE OF DISCRETE LINEAR-MODELS OF CONTINUOUS RANDOM-PROCESSES

A continuous stationary signal possessing non-Gaussian higher order st atistics cannot be correctly modelled by any discrete process based on passing independently and identically distributed noise through a lin ear filter. In particular, it is shown that at third order there exist s no discrete skewed linear model with a discrete bispectrum that is t he same as that obtained from the Nyquist samples of any continuous st ationary process. The nature of the problem is elucidated and an alter native method for modelling the third order statistics of continuous s tationary processes is proposed.