Non-stationary spectral estimation based on robust time varying AR model excited by a t-distribution process

A new robust time variant spectral estimation method is proposed. We use the parametric autoregressive (AR) model to obtain the desired spectra. For robust estimation, we assumed that the residual signal is identically and independently distributed. The probability density function (PDF) of the residual signal is a t-distribution with small α degrees of freedom. We put a certain base function to the parameter of the AR model, so that the obtained spectra is time variant within the considered window. Simulation results show that by using a small α, the obtained running spectra is closer to the ideal spectra than that by using a large α. The mean square error (MSE) between the estimation result and the ideal spectra derived by using a small α is smaller than that by utilizing a large α.