Dynamic range overflow is a serious problem in residue arithmetic systems. Contemporary overflow management schemes rely on inefficient scaling operations. In this correspondence a residue scaler is architectured which inhibits dynamic range overflow. The system uses the popular three moduli set {2n-1, 2n, 2n + 1}. Using a 4K memory model, practical 12-and 18-bit autoscalers are configured. An error model for the derived residue arithmetic unit is also derived and experimentally verified.

Four DFT's, namely the radix 2 and 4, Good-Winograd Fourier transform, and general-N Winograd transforms have been architected using the residue number system. The structure of these architectures, in light of the high degree of parallelism allowed by residue arithmetic, are critically compared. Scaling overheads are computed for residue based systems. Overhead factor of 3, 4.5, 5, and 10 were computed for radix 2, radix 4 FFT's, GWFTA, and NWFTA respectively. The effect of these overhead burdens are discussed and analyzed.