Rate-optimal static scheduling for recursive DSP algorithms by retiming and unfolding

Abstract

A set of novel techniques for rate-optimal scheduling of recursive digital signal processing (DSP) algorithms is presented. The underlying multiprocessor system is based on the RISC architecture. The fully-extended precedence graph and the optimization method of retiming and unfolding is introduced. The retiming process can evenly redistribute the delays and optimize the schedule time of each iteration. The unfolding of recursive data flow graph exploits intra-iteration and inter-iteration parallelisms, and makes the schedule more compact. The optimal unfolding factor can be derived by using the proposed method, which is better than previous methods. It is shown that after evenly retiming and unfolding with the optimal unfolding factor, the rate-optimal design can be easily obtained by a simple contiguous scheduling algorithm.