Methods to design low-sensitivity canonical digital filters using impulse response data

Abstract

We minimize a two-objective function to iteratively design canonical-structure digital filters. The frequency response is matched by approximating a desired impulse response, and the coefficient sensitivity is directly minimized using both forward and backward prediction error methods. Low sensitivity digital filters have low output quantization noise power. The described methods yield filters with variable MA and AR orders: examples show that filters with more transfer function zeros than poles possess low sensitivity and highly matched impulse responses.