Robust signal ARMA model estimation using pre-filtering and data sectioning

The authors consider the robust estimation of signal autoregressive and moving average (ARMA) model parameters in noise. The original noisy measurement with high sampling rate is prefiltered by using a moving average process. The fast modes of the signal are obtained by using the leading section of the data and the closely spaced slow modes are calculated from the decimated trailing section of the data. The MA coefficients are extracted from the estimated AR coefficients by solving a normal equation with singular value decomposition (SVD) and spectral factorization in the frequency domain.<>