Pole-zero modeling of transient waveforms: a comparison of methods with application to acoustic signals

Abstract

The modeling of damped signals as the impulse response of a pole-zero system is considered for a broad range of pole-zero modeling algorithms. The goal is to obtain the best possible fit between the model impulse response and the modeled signal. Prony's method, the least squares modified Yule-Walker equations (LSMYWE), iterative prefiltering, and the Akakie maximum likelihood estimator are compared on known test sequences for a variety of model degrading situations (e.g., additive noise) to develop an understanding of which methods are most suitable for modeling real world signals. A correlation domain version of iterative filtering (including the correlation domain version). Modeling several laboratory generated short duration acoustic signals confirmed the robustness of LSMYWE and iterative prefiltering. It is shown that correlation domain iterative prefiltering outperforms standard iterative prefiltering when large model orders are required for accurate modeling. Shank's method was determined to be the most effective method of determining the zeros of a pole-zero model when a time domain match is required.