On efficient digital filter structures with finite precision implementation consideration

Abstract

With the development of digital devices, digital signal processing has been considered as one of the key tools in modern technologies. Digital filters, as the essential elements in such processing systems, play a crucial role, and hence, digital filter design has become an active field. One of the important issues in digital filter design is the finite word length (FWL) implementation. In practice, the performance of a well-designed digital filter can be totally different from the desired one and hence a computationally efficient structure in infinite precision sense is no more efficient when FWL implementation is concerned. This leads to the optimal FWL structure problems. There are four types of FWL effects: parameter perturbation, roundoff noise, limit cycle, and overflow. The focus of this thesis is on the first two effects, which are measured by the transfer function sensitivity and roundoff noise gain, respectively.