ULTRAFAST SIGNAL PROCESSING ALGORITHMS AND ARCHITECTURES

A CRITICAL PROBLEM IN THE USE OF SIGNAL PROCESSING ALGORITHMS IN REAL-WORLD APPLICATIONS IS THAT OF FINDING EFFICIENT REALIZATIONS ON AVAILABLE HARDWARE. RECENT WORK IN SYMBOLIC SIGNAL ANALYSIS AND MANIPULATION HAS DEMONSTRATED THE POTENTIAL FOR THE AUTOMATIC TRANSFORMATION OF ALGORITHMS INTO EFFICIENT FORMULATIONS. THESE METHODS HAVE SHOWN PROMISE IN REDUCING THE HUMAN EFFORT REQUIRED IN DESIGNING EFFICIENT IMPLEMENTATIONS OF ALGORITHMS AND HAVE DEMONSTRATED THE ABILITY TO FIND NEW METHODS FOR IMPLEMENTING SOME SIGNAL PROCESSING OPERATIONS. CURRENTLY, HOWEVER, THESE METHODS ARE GENERAL-PURPOSE AND DO NOT MAINTAIN REALISTIC HARDWARE MODELS. AS A RESULT, IT IS DIFFICULT TO AUTOMATICALLY EVALUATE THE APPLICABILITY OF PARTICULAR ARCHITECTURES AND TO QUANTIFY PERFORMANCE. IT IS PROPOSED TO STUDY THE TECHNIQUES OF SYMBOLIC SIGNAL ANALYSIS AND MANIPULATION USING HARDWARE-SPECIFIC MODELS. THIS WORK WILL DEVELOP SPECIFICATIONS OF DIFFERENT ARCHITECTURES, INCLUDING BOTH REGULAR AND SPECIAL-PURPOSE ARCHITECTURES, AND WILL EXAMINE VARIOUS MEASURES OF MERIT, INCLUDING OPERATIONS COUNT, MEMORY, AND THROUGHPUT. SYMBOLIC METHODS FOR SIGNAL ANALYSIS AND TRANSFORMATION TAILORED TO THESE HARDWARE SPECIFICATIONS WILL BE DEMONSTRATED. PHASE I WILL ANALYZE CURRENT ALGORITHMS AND ARCHITECTURES FOR THE APPLICABILITY OF SYMBOLIC MANIPULATION, WILL DEVELOP AN EXPLICIT REPRESENTATION FOR A PARTICULAR HARDWARE ARCHITECTURE, AND WILL DEMONSTRATE SYMBOLIC ANALYSIS AND TRANSFORMATION OF SOME SIGNAL PROCESSING ALGORITHMS FOR THE CHOSEN ARCHITECTURE.