Block-floating-point implementation of recursive computations on a multiple datapath DSP
Shingo Kobayashi, Gerhard Paul Fettweis · 2003
The realization of recursive computation on multiple datapath digital signal processors (DSPs) is studied. A block-floating-point implementation is also proposed. This implementation requires only one guard bit in the accumulator, thus leading to a low complex hardware. This implementation allows a superior signal processing accuracy compared to that of short-word floating-point or fixed-point.