Efficient concurrent error detection and correction of soft errors in NTT-based convolutions

A system for soft error detection and correction is proposed for digital Integrated Circuit (IC) implementation of convolution. The convolution is implemented in a Residue Number System using Fermat Number Theoretic Transforms. The flexibility afforded by the Modified Overlap Technique in allowing transforms of differing lengths in a convolution makes it possible to easily detect and correct soft errors by means of a Single Redundant Channel and pattern matching technique. The proposed system gives area reductions in the majority of cases examined, when compared with Triple Modular Redundancy. In the case of large (e.g. 28 and 32 bit) word lengths, the proposed system provides area reductions of up to 30%.