Evaluation of Methods for AR Coefficients Estimation Using Monte Carlo Analysis

Abstract

Aim of this paper is to give recommendation for work with methods used for estimation of coefficients of autoregressive process. We applied Monte Carlo simulations to investigate performance of Burg, Yule-Walker and covariance methods. Evaluation of precision of spectral estimation is done with focus on signal length and lag order. The results are presented in graphical form and briefly discussed. Taking these results into account, Yule-Walker method shows better performance in case of long length signals and in case of overvalued lag order. Burg and covariance methods provide better results in case of short length signal and undervalued lag order.