Abstract
Pisarenko's method is an approach to estimate the frequencies of sinusoidal signals corrupted by a white noise. Because this method provides theoretically an unbiased estimation independent on the signal-to-noise ratio, many researchers have improved Pisarenko's method and developed efficient fast algorithms for it.
In this paper, we first present a procedure for the analytic signal transform, which has zero phase property using the fast Fourier transform. It can generate N analytic signals from N measurement data, so it is suitable for spectral analysis with finite data. Secondly, we show that we could get a good frequency estimation by using Pisarenko's method with the analytic signals. Furthermore, we drive three procedures which use effectively and averagely the decimated analytic signals by two and the remainder signals in the decimation. We compare the three procedures by computer simulations and, as a result, insist that the procedure with autocorrelation function averaging is the best of the three.
This method gives much more accurate frequency estimation than the conventional Pisarenko method. Also, we could perform a real-time processing with the proposed method if we use several efficient algorithms developed recently.
