I. Introduction
Statistical data are characterized by their autocorrelation function or by their power spectral density. Together with the mean value, these quantities represent all statistical information of stationary stochastic data with a normal distribution. It is also valuable information for other distributions. The discovery of the fast-Fourier-transform algorithm in 1965 [1] had a great impact on the early practice of spectral and autocorrelation estimation. The reduced computer effort enabled for the first time the routine analysis of extensive sets of data with windowed and tapered periodograms and the estimation of the lagged product (LP) autocorrelation functions as their inverse Fourier transform [2]. Until about the year 2000, the computational demands for time-series (TS) analysis were too heavy for the automatic use in data with unknown model type and order. For that reason, periodograms have been the main practical tool for spectral analysis for such a long time. However, the vast literature on windowing and tapering did not solve the problem that the best choice for type and width of the spectral window requires the knowledge of the exact spectral density [2]. Optimal windows cannot be deduced with sufficient reliability from the estimated spectrum of random data.