TLS-based prefiltering technique for time-domain ARMA modeling

摘要

Modeling time-series with linear pole-zero AutoRegressive-Moving Average (ARMA) models has numerous applications in signal processing. This problem is in general non-linear and most ARMA modeling techniques are iterative in nature. The Iterative Prefiltering (IP) method has the advantage of computing potential non-minimum phase representations which may be useful in time-domain modeling. The original IP minimization procedure is an ill-conditioned problem which has classically been solved using a leastsquares approach. This work presents a modification of the classical IP technique in which the least-squares iteration step is replaced by a Total Least Squares (TLS) step to take advantage of the statistical properties of the TLS method. Results show that improvements in the modeling performances may be obtained by using the TLS-based IP method when modeling signals distorted by white Gaussian noise. ARMA modeling, Total least squares, Transient modeling