Optimization of recursive digital filters for practical implementation

Abstract

This thesis concentrates on two basic issues. The first one considers the development of effective recursive digital filter structures. The main purpose is to generate structures for which the output noise is very low and the multiplier coefficients can be represented as a sequence of shifts and adds. Filters of this kind are very attractive in VLSI implementations where a general multiplier is very costly. A very useful structure is a cascade connection of filters that are parallel connection of two allpass filter sections.Another issue of work is to develop an efficient algorithm for coefficient quantization. For this purpose, the multicriterion optimization based on algorithm of Dutta and Vidyasagar is modified and used. A very crucial matter is the design of a starting point filter for further optimization. The starting point filter has to be in relative neighborhood of the optimal solution. A very efficient design scheme for generating such a filter is proposed for the filter structures under consideration.Several examples prove the suitability of the proposed overall design method to be a universal tool in digital filter synthesis. It may also be applicable in many other areas of signal processing.