Practical Recommendations for the Spectral Analysis of Stationary Gaussian Processes

Estimates obtained for the spectral density of a stationary Gaussian discrete-time stochastic process on the basis of the differential properties of the estimated spectral density are discussed. It is shown that the rejection of nonnegative weighting functions makes it possible in many situations to diminish significantly the mean-square error of estimation. Recommendations are formulated for the selection of weighting functions in accordance with the number n of readings, which is assumed to be finite.