About the quality of parametric power spectral estimates

Abstract

The quality of parametric power spectral estimates is analyzed and a lower bound based on the Cramer-Rao inequality for unbiased estimators has been computed. The quality of a least squares autoregressive (AR) ladder estimator is evaluated by simulation, in terms of bias and variance of the corresponding spectral density estimates. The AR ladder estimator and classical periodogram based moving average (MA) estimators are then evaluated against the theoretical Cramer-Rao bound for AR and ARMA process realizations.