Model reduction of discrete-time systems via frequency response matching

Abstract

An algorithmic approach to reduced order modelling of discrete-time systems is developed. It preserves the dominant eigenvalues of the original system while minimizing the frequency response deviations between the original system and the reduced model. The coefficients of numerator polynomials of the reduced order pulse transfer function matrix are obtained by utilizing a Fletcher-Powell unconstrained optimization algorithm. The proposed algorithm is illustrated by two examples. © 1997 Elsevier Science Ltd.