Design and implementation of discrete-time filters for efficient sampling rate conversion

Abstract

Rate-conversion systems are used in an array of applications, including the oversampled audio and video CODECs often found in entertainment and communications systems. It is common practice for many such systems to sample signals at rates which are much faster than the minimum required to represent some bandwidth of interest, and high-quality filters are often implemented at this fast rate. Therefore, their designs tend to be computationally expensive. A number of structures have been proposed to address this, including polyphase implementations and folded structures for linear-phase FIR filters. In this thesis, techniques which combine benefits from both classes of structures are discussed, and an efficient class of structures is proposed. The Generalized Transposition Theorem is also reviewed to demonstrate that an efficient downsampling structure also implies an equally efficient, closely-related upsampling structure. Techniques are investigated for designing minimum multiply filters for the class of structures presented, and methods are discussed for designing filters that, for a given set of frequency domain filter specifications, often require fewer multipliers and have smaller maximum error than Parks-McClellan designs.