Multiple window correlation estimation with applications in adaptive filtering

A novel autocorrelation estimator is developed using Slepian sequences as multiple windows. The estimator has more degrees of freedom than any single-window estimate, including the sample average, with the same frequency domain resolution. Because the Slepian sequences are orthogonal, confidence intervals can be estimated by jackknifing as well as by standard χ² methods. The proposed multiple window estimator is applied to batch autoregressive (AR) parameter estimation and recursive least squares equalization. Both applications show significant improvement, especially for small data lengths, while only linearly (in the number of windows) increasing computational complexity. Generalizations to higher-order correlation estimators are delineated