An optimum tapered Burg algorithm for linear prediction and spectral analysis

Abstract

Several problems associated with the Burg algorithm are for the most part alleviated by the use of an appropriate energy taper used in the derivation of the algorithm. The "tapered" Burg technique uses a weighted forward-backward least-squares fit to the parameters of the all-pole model, subject to Levinson's recursion constraint. Based on a minimum average variance of the estimated frequency of a real sinusoid, an optimum taper is derived. Finally the performance of the optimally tapered Burg technique is compared with the least-squares, Yule-Walker, and the usual Burg techniques.