An analysis of noise influence to time-varying spectra estimators is presented. Hankel and Toeplitz kernel forms of distributions are considered. It is shown that these two forms of kernels have the same variance, if the windows have the same form. Since the Hankel kernel results in a highly concentrated distribution for a noisy, frequency-modulated signal, this form can be more reliable for the estimation of signal parameters in noise. The results are generalized for multiplicative noise. For frequency-modulated signals, the same conclusions hold in this case.
PDF version of the full paper