Quality measures of the modified biperiodogram: selection of estimator parameters
R.L. Turcotte, Mark A. Wickert · 1991
The mean, variance and normalized mean-squared error of the modified biperiodogram are calculated for a finite sample of the observed signal. It is assumed that an overall record of length N is available, which can be divided into K subrecords, each of length L, and data windowing, can be used. The normalized mean-squared error is used to obtain the best selection of L and K given the overall record length N, and a specified window type. A numerical example is given for a process created by exciting a finite impulse response filter with non-zero skewness zero mean white noise.>