Methods for estimating the autocorrelation and power spectral density functions when there are many missing data values

Abstract

A new method for estimating the autocorrelation and the crosscorrelation has been developed. The resulting estimates are usually more accurate than the classical values. The method is particularly useful when there are many missing data values. For the case when there are many missing data values, it is suggested that a power spectral density (PSD) of the autocorrelation function can be developed. The resulting PSD can easily be mapped into the PSD of the original data. Towards this end, Burg's technique has been applied to the autocorrelation and the results of the application are presented.<>