Frequency errors in the spectral estimates of complex sinusoids using the "Tapered" burg technique

This paper discusses frequency errors in spectral estimates of complex sinusoids in white noise by the "tapered" Burg technique. The tapered Burg technique is a direct result of considering a weighted least-squares fit to the parameters of the all-pole model, subject to Levinson's recursion constraint, in place of the usual unweighted least-squares fit. The algorithmic consequence of this approach is the inclusion of an appropriate taper in the calculation of the partial-correlation coefficients in the usual Burg algorithm. Based on the expression for the frequency error in the spectral estimate of a sinusoid, an optimum taper is derived. The performance of this optimum taper is then compared with those of the rectangular (untapered Burg) and Hamming tapers. It appears that this taper makes MEM spectral estimates of sinusoids using Burg's technique more robust, without sacrificing its resolution.