Covariance decompositions via elementary transformations. Applications to filtering

Abstract

This paper depicts the connection between algorithms for the factorization of covariance matrices, a basic operation in linear estimation, and whitening or modeling filters. General algorithms for arbitrary covariances are presented and related to the four fundamental types of cascade filters: feed-forward and feed-back tapped delay line or ladder filter. Systolic implementations of the algorithms and realizations of the associated filters illustrate the correspondence between algorithms and filters. Finally, fast algorithms for special covariances are introduced through two important examples: constant-parameter tapped delay line and ladder filter.