Parallel signal-processing for everyone

Abstract

We designed, implemented, and evaluated a signal-processing environment that runs on a general-purpose multiprocessor system, allowing easy prototyping of new algorithms and integration with applications. The environment allows the composition of modules implementing individual signal-processing algorithms into a functional application, automatically optimizing their performance. We decompose the problem into four independent components: signal processing, data management, scheduling, and control. This simplifies the programming interface and facilitates transparent parallel signal processing. For tested applications, our system both runs efficiently on single-processors systems and achieves near-linear speedups on symmetric-multiprocessor (SMP) systems.