On the Application of the E-M Algorithm for Autoregressive Modeling of Noisy Sources

In this paper we examine the problem of autoregressive (AR) modeling of information sources which have been degraded by noise. An elegant iterative procedure for estimating the parameters of Gaussian AR sources, corrupted by Gaussian noise whose power spectral density (PSD) is known, was first proposed by Lim [1]. This procedure was extended in [2] to non-Gaussian noise processes whose power spectral density is unknown.