Evaluation of the Steiglitz algorithm for estimating the parameters of an ARMA process

Steiglitz (2) has recently described an algorithm for estimating the parameters of an autoregressive-moving-average (ARMA) process. This algorithm has application, for example, to the problem of determining the poles and zeroes of the vocal-tract transfer function. In this paper, we report and discuss the results of a number of simulations conducted using the Steiglitz algorithm. The bulk of the experiments involved driving the ten-pole, two-zero filter described in (2) with a single pulse, with a short pulse train, and with samples of white Gaussian noise. In each of these cases, we evaluated the effects of such processing options as windowing, preemphasis, and cepstral-domain filtering. We also discuss and compare results obtained by applying the Steiglitz algorithm and a Newton-Raphson conditional maximum-likelihood algorithm to a first-order process.