Linear estimation of ARMA processes

The authors propose an identification algorithm for autoregressive moving average (ARMA) processes. Given a finite length sample drawn from an ARMA (p/sub 0/, q/sub 0/) model, the technique provides the estimated values of the orders p/sub 0/ and p/sub 0/, as well as the AR and MA coefficients. They are obtained from the reflection coefficient sequence estimated directly from the data. The order selection scheme is based upon the minimization of a functional that measures the mismatch of the data to any ARMA (p,q) assumed as its model.<
>