The accuracy of center frequency estimators using linear predictive methods

Linear prediction has been established as a reliable and accurate technique for spectrum estimation. One of its most interesting properties is its high resolution. A computer simulation of a signal processor based on this method has been implemented in order to study the errors associated with the estimation of the spectral center frequency of narrow-band processes. Both the Yule-Walker, or correlation method, and Burg´s method were considered. The sensitivity of the results was determined for variations in the signal-to-noise ratio, number of samples, process bandwidth and relative center frequency. The results are then compared to lower bounds on spectral center frequency estimation based on periodogram analysis and radar principles.