Efficient implementation of the HMARM model identification and its application in spectral analysis

Abstract

The Hidden Markov Auto-Regressive model (HMARM) has recently been proposed to model non-Gaussian AutoRegressive signals with hidden Markov-type driving noise. This model has been shown to be suitable to many signals, including voiced speech and digitally modulated signals received through ISI channels. The HMARM facilitates a blind system identification algorithm that has a good computational efficiency and data efficiency. In this paper, we solve an implementation issue of the HMARM identification, which can otherwise degrade the efficiency of the model and hinder extensive evaluations of the algorithm. Then we study in more detail the properties associated with the autoregressive (AR) spectral analysis for signals of interest.