Evolutionary Burg spectral estimation

We propose a new method to estimate the evolutionary spectrum of nonstationary signals. By defining the instantaneous stationary series, this technique applies the Burg spectral estimation method to this series at each time to obtain the time-varying autoregressive (AR) spectrum of the nonstationary signal. The time-varying model coefficients are efficiently found by a Levinson-like algorithm. This evolutionary Burg (EB) method can also be used to estimate the instantaneous frequencies of the sinusoids present in the nonstationary signal. Compared to the recently proposed estimators of the evolutionary spectrum, the new technique has better frequency resolution and lower sidelobe behavior.