CAD tools for the automatic synthesis of parallel algorithms in control

Abstract

The paper is to promotes the concept that the design of parallel control algorithms can be pursued by employing algorithm engineering. It is demonstrated that a particular class of algorithms, control laws based on systems of recurrence equations with time as an independent variable, can be engineered at an algorithmic level to exploit parallelism. The problem of engineering an algorithm so that it models certain design objectives such as minimum latency, pipelined data flow etc. is considered. The author considers control system design as a 2-stage process. The levels are distinct and separated into conventional control system and parallel algorithm design respectively. The whole design process embraces a fusion between control theory and the theory of algorithms requiring skills in both mathematics and computer science. A number of seemingly different methods of synthesis are considered. They have considerable overlap and can in fact be combined to provide a unified and viable methodology for the systematic and reasoned mapping of control algorithms onto parallel architectures. It is well known that control elements can be implemented via physical devices or through software. Furthermore control systems are inherently parallel in nature.