Large-scale linear computations with dedicated real-time architectures

Abstract

Understanding the connection between algorithms and solvers for large scale systems on the one hand, and appropriate architectures that execute them efficiently on the other is key to the effective design of modern signal processing applications. This trend has started with the emergence of digital media, large scale signal processing for image coding and analysis, digital mobile telephony and digital processing in medical imaging. In many cases dedicated, hardware or software dominated methods on a single processor have been used, only in recent times more generic methods based on general purpose array architectures, either dedicated to media processing or for general computing have become realizable. Massive use of parallelism becomes attractive and should, in the future, allow us to tackle large scale problems in a streamlined fashion.