On the relationship of maximum likelihood sampled-data power spectrum identification and optimum predicition filters

Methods for estimating the sampled power spectral density of a stochastic process in terms of a rational function of
z
have been presented in the literature. A method based on the maximum likelihood criterion for Gaussian processes leads to the minimum residual criterion.
^[1],[2]
This correspondence points out the relationship of the minimum residual criterion to optimum prediction filters and justifies the use of the criterion even for non-Gaussian processes.