Frequency Weighted Identification and Model Reduction of 2-D Separable Denominator Systems

Abstract

A model reduction scheme for causal recursive separable denominator (CRSD) 2-D models is outlined. Frequency weightings are introduced to improve the model accuracy over the operating bandwidth. The weighting matrices are defined using the impulse responses of input and output weighting filters, and then the model is reduced using balanced truncation. The models are demonstrated to correspond to a frequency weighted balanced realization. An example is provided to illustrate that the proposed method provides improved accuracy over the desired frequency range.