A modular approach to the computation of convolution sum using distributed arithmetic principles

Inner product evaluation can be considered to be basic to all digital signal processing algorithms. Distributed arithmetic (DA) is a bit serial memory-oriented method that implements inner products. In DA, memory increases exponentially as the number of coefficients increases. This deters the use of DA in the implementation of DSP algorithms. This paper explores the possibility of combining DA principles with the residue number system (RNS) to exploit the advantages of both. A new technique for the computation of the inner product in the RNS domain using modular multipliers and DA principles is proposed, The resulting architecture is elegant, uses minimum memory and adders, and hence is VLSI efficient. The proposed architecture is extremely efficient in terms of speed and does not involve conversion of data into residues although computation is performed in residue domain.