Performance of normalized correlation estimators for complex processes

Direct and fast techniques for estimating normalized second-order moments of complex processes are discussed. First, the accuracy of the direct estimate is explicitly given in terms of bias and covariance for the Gaussian processes. Then, a class of estimators based on the complex invariance property is considered. Their theoretical accuracy is given for some cases of paramount interest from a computational point of view in the Gaussian case. In particular, applications in the areas of autoregressive spectrum analysis and spectral centroid estimation are presented. General analytic results and simple expression for immediate evaluation are provided