Performance analysis of different special moduli sets for RNS-based inner product step processor

This paper studies the performance of RNS-based inner product step processor of different special moduli sets. The processor is generic with programmable number of modulo additions and multiplications. Such processor is the core of FIR filters and FIR filter-like functions such as correlators, convolvers, DFT/FFT/DCT, etc. In our study, five different RNSs with modulo arithmetic friendly moduli of the forms 2ⁿ and 2ⁿ±1 are selected for the implementation of inner product step processor. Their relative performances in terms of power dissipation, area and delay are compared. A good insight into these RNSs has been established to help with the choice of an appropriate special moduli set for a given application scenario.