THE GEORGIA TECH DIGITAL SIGNAL MULTIPROCESSOR

Techniques for prototyping of digital signal processing (DSP) algorith ms on synchronous multiprocessors have been studied at the Georgia Ins titute of Technology for the past ten years. This research can be char acterized as a theoretical investigation of optimality criteria for th e implementation of DSP algorithms, and the development of a unique ne w approach that can achieve these optimality criteria. A recent result of this ongoing research program has been the development of a set of formal techniques for the automatic realization of optimal synchronou s supercomputer implementations for a large class of iterative or recu rsive DSP algorithms. Three experimental laboratory digital signal mic roprocessor (DSMP) systems have been designed at Georgia Tech to provi de experimental verification of the theoretical results. This has prov ided important feedback to drive the theoretical developments. In addi tion, compilers that take flow graph specifications of algorithms and generate optimal multiprocessor code have been developed. This paper s urveys the theory and architectural explorations in the design and imp lementation of the Georgia Tech digital signal multiprocessor.