Strong consistency of recursive least square estimates on amplified harmonics for line spectrum

For a sampled signal with length n, it is shown that the method of maximum least-squares regressions on amplified harmonics yields estimates for discrete frequencies with the variance of order n/sup -3/ and with the bias of an order almost surely smaller than n/sup -3/2/ (log n)/sup 1/2+ delta /. Hence, these estimates are strongly consistent and very accurate. It is shown that these estimates can also be obtained in an iterative way based on recursive least-squares estimates on amplified harmonics.<
>