Noisy input/output system identification using cumulants and the Steiglitz-McBride algorithm

By transforming the input/output system identification problem into the high signal to noise ratio (SNR) cumulant domain, the Steiglitz-McBride algorithm is extended, yielding an autocumulant and cross-cumulant based approach for autoregressive moving average (ARMA) modeling. The autocumulant approach requires that the ARMA parameters be estimated by first estimating the cumulants of the ARMA parameters. The cross-cumulant formulation permits the ARMA parameters to be estimated directly. Possible convergence points and convergence issues are investigated. Simulations are presented to illustrate the performance of these algorithms