Wordlength optimization of an MPEG-2 audio decoder

Abstract

A set of internal wordlengths for implementing an MPEG-2 audio decoder has been determined to minimize the hardware cost while limiting the fixed-point noise to be below the perceptual masking curve. The hardware architecture is based on a data path containing a multiplier and an adder. The finite wordlength effects of three DCT algorithms for the efficient implementation of the subband filtering are also compared. For the bit-accurate fixed-point simulation, the fixed-point optimization utility, which has been developed for the rapid evaluation of finite wordlength effects, is used.