An Improved Autoregressive Spectral Estimation Method Using the Kalman Filter System Identification Technique

This paper presents an improved multichannel autoregressive spectral estimation method by smoothing the autoregressive (AR) model obtained by the least-squares technique. The smoothing is based on the relationship between a state-space model and an AR model of a stochastic sigal. The method starts with the classical least-squares estimation of an AR model of the signal, and then uses the Kalman filter system identification method to obtain a state-space model and the corresponding Kalman filter gain. The model and filter gain are in turn used to reconstruct a smoothed AR model, which is used to produce an improved spectral estimation. A numerical example is included to illustrate the feasibility of the method.