Identification of resonant systems using Kautz filters

Abstract

It is pointed out that by approximating the impulse response of a linear time-invariant stable system by a finite sum of given exponentials, the problem of estimating the transfer function is considerably simplified. The author shows how the complexity can be reduced further by using orthogonalized exponentials. The analysis is based on the result that the corresponding normal equations will then have a Toeplitz structure. The z-transform of orthogonalized exponentials corresponds to discrete Kautz functions, which generalize discrete Laguerre functions to the several, possibly complex, poles case. Hence, by appropriate choice of time constants Kautz models give low-order useful approximations of many systems of interest. In particular, resonant systems can be well approximated using Kautz models with complex poles. Several basic results on transfer function estimation are extended to discrete Kautz models.<>