Rate optimal scheduling of recursive DSP algorithms by unfolding

摘要

We propose a novel criterion called modified generalized perfect rate graph (MGPRG) to facilitate efficient synthesis of recurrent digital signal processing (RDSP) algorithms on application specific multiprocessor systems. We prove that a rate optimal schedule, which is a multiprocessor realization with maximum throughput rate, exists if and only if the dependence graph of the corresponding RDSP algorithm is unfolded sufficient number of times to satisfy the MGPRG condition. We show that MGPRG is a more general and unified criterion compared to earlier proposed criteria. An advantage of MGPRG is that the original algorithm needs not to be unfolded too many times before it can be guaranteed to have a rate optimal schedule. Consequently, the resulting MGPRG will consist of fewer nodes and edges, making it easier to find the rate optimal schedule