Parameter estimation of AR systems at a very low SNR using prefiltering in the autocorrelation domain

A simple prefiltering technique in the autocorrelation of a noisy signal in the Fourier domain has been presented here to identify the system parameters using the high-order Yule-Walker (HOYW) equations. Presence of noise in the autocorrelation domain is the main reason for failure of the conventional system identification techniques at low signal to noise ratio (SNR). As system modes are observed in the Fourier spectrum of the autocorrelation sequence, proper threshold reduces the randomly distributed noise spectra leaving the significant portion of the peaks of the system spectra undistorted. Reconstruction of the autocorrelation sequence from the filtered spectrum reduces the unwanted fluctuations making the system identification more consistent even at low SNR. Results for estimated system parameters at a low SNR of -5 dB has been presented to show the efficacy of the proposed technique.