The aim of this research is to solve the stochastic realization problem for linear time-invariant non-minimum phase systems which possess rational transfer functions. This problem consists of estimating the parameters, or equivalently the poles and zeros, of an autoregressive moving average model. The novelty of the research lies in the consideration of non-minimum phase models, ie. those in which the zeros of the underlying discrete system lie outside the unit circle. Two approaches to the stochastic realization problem for non-minimum phase systems are proposed, and it is expected that they will allow the reconstruction of both amplitude and phase information from output data only. Solution of the stochastic realization problem will provide a method for parametric spectral and polyspectral estimation which is important in such applications as estimating the source signature in sonar and seismic data processing, speech coding, adaptive control and constructing equalizers for communication systems. +
