Adaptive windowing in nonparametric power spectral density estimation

Abstract

In this paper we consider nonparametric power spectral density (PSD) estimation. We use finite length observed data not only to estimate the PSD by one of the existing periodogram estimators, but also to estimate the mean square error (MSE) in PSD estimation. Minimizing this criterion provides the optimum window length for the Blackman-Tukey approach. It can also provide extra adaptive window for PSD estimates of periodogram averaging methods such as Bartlett and Welch approaches. We demonstrate that the new additional optimum windowing improves the performance of the existing averaging periodogram approaches.