An ARMA system identification scheme in the presence of noise

摘要

This paper presents a scheme for the identification of minimum-phase autoregressive moving average (ARMA) systems in the presence of noise. For the identification of the AR part, we propose to repeat the correlation operation on an enhanced autocorrelation function of the observed signal based on a decision criterion and then employ the resultant to an extended form of Yule-Walker equations. A frequency domain noise-compensation scheme is introduced which operates on the noise-contaminated residual signal. The MA parameters are then estimated via the spectral factorization performed on the power spectrum of the noise-compensated residual signal. Computer simulations exhibit a superior estimation performance even at low levels of signal-to-noise ratio (SNR).