Genetic algorithm based DSP multiprocessor scheduling

This paper presents recent work on the application of genetic algorithms to the NP-complete problem of multiprocessor scheduling for audio DSP algorithms. The genetic algorithm is used to schedule algorithms written in the form of data flow graphs onto specified multiprocessor arrays. A unique chromosome representation technique is described and a number of application-specific genetic operators are introduced. Comparisons of the performance of the genetic algorithm technique with heuristic scheduling techniques show that the choice of the most suitable technique varies with the structure and complexity of the scheduling problem. Finally, techniques for combining heuristic and genetic algorithm scheduling techniques are discussed.