AR, ARMA, and AR-in-noise modeling by fitting windowed correlation data

A method for autoregressive (AR) modeling of stationary stochastic signals is extended to AR moving-average (ARMA) models, including the special case of AR signals in white noise. Both AR and ARMA examples are presented. The method differs from the well-known method of overdetermined normal equations in that fitting error, not equation error, is minimized, and significantly improved performance is obtained. Iterative algorithms patterned after the Steiglitz-McBride (1965) deterministic method are derived to solve the resulting nonlinear equations.<
>