Abstract:
Several fields of application such as: astronomy, acoustics, image and signal processing, etc., have used of higher order statistics. They played a crucial role in the identification of a non-minimal phase linear system. Thus, the object of this thesis is the identification of the parameters of a non-minimal phase 2D MA models using higher order cumulants. First, the almost-sure convergence properties of sample estimates of higher order spatial statistics are derived. As a practical framework, we address the problem of identification of 2D moving average (MA) models with non-Gaussian errors based on cumulants alone under a nonminimum phase assumption first and on a generalized method of moments approach after. A simulation study verifies the performance of the proposed methods.
