Input recoding for reducing power in distributed arithmetic

Abstract

Digital signal processing algorithms rely heavily on the efficient computation of inner products. Distributed arithmetic provides a multiplication-free method for calculating inner products of fixed-point data, based on table lookups of precalculated partial products. A method is proposed for reducing switching activity, and hence power dissipation, in distributed arithmetic systems used for processing signed data. By performing a recoding of the two's complement inputs into a nonredundant signed-digit representation, the table lookups and accompanying accumulations corresponding to the sign extensions in the higher-order bit positions can be reduced.