Coloured noise compensation by using autoregressive estimators and side information

Abstract

The use of a priori knowledge about the process under analysis is very effective to obtain high quality spectral estimators. On the other hand, one of the simplest and most effective schemes that has been successfully employed in spectral analysis is the well-known autoregressive model. The air of this paper is to introduce side information into autoregressive spectral estimators, in order to get a powerful procedure to estimate spectral densities. Specifically, we shall center our interest in methods to compensate the coloured noise that corrupts a desired signal. We show the exact equations, a linearized version of it, and the corresponding algorithms. Some examples are considered to illustrate the proposed methods. Although focused on the noise compensation problem, the method should have general application in cases where a spectrum shape pre-estimator can be defined using a priori knowledge.