A Novel Hankel Approximation Method for Arma Pole-Zero Estimation from Noisy Covariance Data*

Abstract

Model based methods have been gaining popularity in high resolution spectral estimation, and have recently demonstrated a great deal of success. Such methods allow us to parameterize the spectrum in terms of a relatively small number of unknown parameters, and thus reduce the spectral estimation problem to that of first, selecting the appropriate model, and second, estimating its parameters. The most popular models used today, are 1) Autoregressive model (AR), 2) Sinusoids plus noise model (S+N) and 3) Autoregressive moving average model (ARMA)