On the optimum tapered burg algorithm

Kaveh and Lippert's recently published optimum window technique is reexamined from a frequency-domain viewpoint. It is shown that the overall variance of the error in the frequency estimator is dominated by contributions from those frequency regions near 0 and π where the error model itself is unreliable. Excluding these regions from the variance calculations reveals that the optimum window may not be the best choice under all circumstances.