An Accurate Frequency Estimator of Single Sinusoid in Gaussian Colored Noise

It is difficult to estimate the frequency of a single sinusoid in additive Gaussian colored noise. This paper firstly heterodynes the signal to band nearby zero by a coarsely estimated frequency got from nonlinear filtered psd (power spectrum density) of signal, and then gets more accurate frequency estimation by applying least square algorithm to differenced phase of signal after depressing noise by low-pass filtering and data decimating. The Monte Carlo simulation indicates that the new algorithm can work very well on condition that the psd fluctuation of colored noise nearby sinusoid is slight and the data decimation factor is large. Because the mean-squared error of new algorithm already reaches the CRB for white noise with only 0.5 dB deterioration above threshold, it is unnecessary to Karhunen-Loeve expand or whitening filter the signal when the frequency estimation of sinusoid in Gaussian colored noise is concerned