Task scheduling for power optimisation of multi frequency synchronous data flow graphs

During recent years power optimisation has become one of the most challenging design goals in modern communication systems, particularly in the wireless domain. Many different approaches for task scheduling on single or multi-core systems exist, mostly addressing the minimisation of execution time or the number of processors used. The minimisation of the processor's clock frequency by adjusting the supply voltage or directly by frequency scaling according to the chosen task scheduling has shown good results in the reduction of power consumption. Most of the known approaches base their core algorithms on graph representations for multirate systems or synchronous data flow (SDF) graphs, in a single frequency domain. In many cases a signal processing system comprises several frequency domains, in which processes have to be fired according to their in- and output data rates as well as to their frequency domain. In this work the superposition of frequency domains and data dependencies is incorporated into the optimisation process and used as a another degree of freedom. Several algorithms have been implemented and evaluated to minimise the required processor's clock frequency, including a greedy, a simulated annealing, as well as a tabu search approach.