Identification of non-minimum phase linear stochastic systems (Q1084071)

This paper considers the problem of estimating the parameters of non- minimum phase non-Gaussian ARMA processes from measurements corrupted by Gaussian white noise. The solution proposed consists of two steps. The first step estimates the parameters of a spectrally equivalent minimum phase ARMA model of the signal, by using a standard prediction error method. This step provides estimates of the system poles. The second step determines the location of the (possibly non-minimum phase) system zeros by minimizing a least-squares criterion involving a set of sample fourth order cumulants of the observed process and the corresponding theoretical cumulants. This two-step procedure is shown to produce consistent estimates under some weak conditions. Its performance is illustrated by means of some simulation examples.