Estimação de Freqüências por Predição Linear com Desempenho de Máxima Verossimilhança<br />DOI: 10.14209/jcis.1998.4

Abstract

This work establishes a unifying framework to the use of linear prediction for estimating the unknown frequencies of signals corrupted by additive noise. A method is proposed that optimizes linear prediction in total least squares sense. Also presented is a signal subspace approach to the method. This method is shown to be competitive in relation to that of principal components. In order to make linear prediction perform like maximum likelihood, recent methods take into account the structure imposed by the linear prediction filter on the data matrix. These methods are shown to minimize the same objective function and yield equivalent solutions.