Optimum Segmentation and Windowing in Nonparametric Power Spectral Density Estimation

In averaging power spectral density (PSD) estimation methods, such as Bartlett and Welch approaches, segmented version of data is used. However, no systematic method for the choice of optimum segmentation is available. In this paper, we provide a novel approach to nonparametric PSD estimation that not only provides the optimum segmentation in these approaches, but also combines these approaches with a new optimum windowing within the segments. The desired criterion in this method is PSD mean square error that is estimated for windows and segments of different length. The new optimum windowing approach outperforms the existing nonparametric approaches.