Area efficient fault tolerant convolution using RRNS with NTTS and WSCA

An efficient fault tolerant digital integrated circuit implementation of a convolution is presented based on the redundant residue number system (RRNS) combined with the modified overlap save method. The moduli used are Fermat and Mersenne numbers. A combination of number theoretic transforms (NTT) and Winograd short convolution algorithms is used for the convolution. The resulting area is compared with an NTT RRNS system using the conventional overlap save (OS) method. Using the new system, significant area reductions are achieved. In particular, a convolution with wordlength 24 and transform length 15 can be implemented in the proposed system with 84% less area than that of an RRNS system using the conventional OS method.