FREQUENCY ESTIMATION PERFORMANCE OF SEVERAL WEIGHTED BURG ALGORITHMS

The frequency estimation performance of a set of weighted Burg algorithms is investigated for single complex and real sinusoids in noise. It is demonstrated that in the complex case an extension of Ibrahim's modification of Kaveh and Lippert's optimum tapered Burg algorithm approaches the Cramer-Rao lower bound for a signal-to-noise ratio sufficiently high and a model order sufficiently large. This is in contrast with the original Burg method, for instance. In the case of a real sinusoid, the same algorithm most closely approaches this bound, thus making it the technique of choice for both applications. The issue of spectral resolution is also addressed briefly.