PARALLEL PROCESSING APPLICATIONS FOR AVIONICS

A SINGLE FORM OF A "MASSIVELY PARALLEL" COMPUTER ARCHITECTURE IS BEING SOUGHT TO HOST THE COMPLETE SUITE OF A VISION MODEL-BASED OBJECT RECOGNITION ALGORITHM. TWO APPROACHES WILL BE INVESTIGATED DURING PHASE I: (1) A SIMD ARCHITECTURE REPRESENTED BY THE AMT DISTRIBUTED ARRAY PROCESSOR DAP, AND (2) A MIMD ARCHITECTURE BASED UPON THE SHILOACH/VISHKIN ALGORITHM. THE DAP 510 WILL BE EMULATED ON I-MATH'S VAX COMPUTERS USING THE COMMERCIALLY AVAILABLE DAP SIMULATOR; THE MIMD ARCHITECTURE WILL BE STUDIED USING THE NYU ULTRACOMPUTER. COMPLETE EXPLORATION OF BOTH TECHNOLOGIES IS ASSURED BY KEY I-MATH CONSULTANTS; DR. ROBERT HUMMEL (NYU COURANT INSTITUTE) AND DR. WINTRHROP SMITH (AMT CHIEF APPLICATION SCIENTIST). FOR PHASE I, THE SCERPO (SPATIAL CORRESPONDENCE, EVIDENTIAL REASONING, PERCEPTUAL ORGANIZATION) MODEL-BASED ALGORITHM WILL BE USED AS THE CANDIDATE FOR SIMD/MIMD IMPLEMENTATION. IT'S PERFORMANCE IN EACH ARCHITECTURE WILL BE BENCHMARKED AGAINST ITS CURRENT PROCESSOR IMPLEMENTATION.