Non-iterative Subspace-based Method for Estimating AR Model Parameters in the Presence of White Noise with Unknown Variance

We consider the problem of estimating the parameters of autoregressive (AR) processes in the presence of white observation noise with unknown variance, which appears in many signal processing applications such as spectral estimation, and speech processing. A new non-iterative subspace-based method named extended subspace (ESS) method is developed. The basic idea of the ESS is to estimate the variance of the observation noise via solving a generalized eigenvalue problem, and then estimate the AR parameters using the estimated variance. The major advantages of the ESS method include excellent reliability and robustness against high-level noise, and also estimating the AR parameters in a non-iterative manner. Simulation results help to evaluate the performance of the ESS method, and demonstrate its robustness.