Impulse response matching for z-Domain identification with application to model order reduction

This paper presents a deterministic time-domain approach to z-domain model identification. Coefficients for a discrete transfer function model of specified order are determined by matching the model's impulse response to that of an observed system. Interestingly, the specialized identification equations that result coincide with those from the conventional least-squares theory using regression models and thus provide a link between the deterministic and stochastic theories. The technique is applied to the model reduction problem for discrete linear systems.