Abstract
A new technique is proposed to estimate frequencies and amplitudes of sinusoids embedded in background signals with unknown colored spectrum. In such cases Pisarenko's harmomic retrieval (PHR) method can give no unbiased estimates. The present approach can eliminate the biases by whitening the colored noise in observed data which is modelled as an autoregressive (AR) process, and adopting the PHR method to the filtered signal. The parameters of the whitening FIR filter are determined so as to minimize a specified criterion whose minimum value is attained when the parameters agree with those of the AR process. The criterion also plays an important role of deciding the number of noise-corrupted sinusoids. The paper also gives a method of obtaining the gradient vector of the criterion with respect to the filter parameters, which is needed to perform the conjugate gradient method. The validity of the present method is examined in numerical examples.
