A distributed signal processing architecture

Abstract : An architecture is described for a multi-processor implementation of real-time signal processing algorithms. A butterfly network is used to provide simultaneous, conflict-free interprocessor communication for multi-dimensional convolution and Fourier transformation. A hardware demonstration test-bed using four active processors was used to validate the concepts of shared algorithm execution, conflict-free data transfers, distributed network control, dynamic fault tolerance, and identical software in all processors. The Distributed Signal Processor (DISP) is designed to perform real-time signal processing on large data sets. Potential applications for the DISP include image processing (enhancement, restoration, segmentations, and coding), radar imaging, synthetic aperture radar, and infrared imaging (passive or active). These applications share three important characteristics. First, the data sets are large, generally multi-dimensional, with high computation rates. Second, the applications make extensive use of convolution and Fourier transformation algorithms. Finally, real-time signal processing applications generally have a regular, well defined time-line.