The present and future of signal processing.
Vito Cappellini, Pier Luigi Emiliani · Czech digital mathematics library · 1990
The paper reviews many important activites in digital signal processing in particular some topics interesting as a starting point in present and future depth analyses and discussions, with emphasis on implementation problems.The necessary changes and trends in design and implementation techniques of digital signal processing systems are described, which result from development of related technology with respect to costs and performances suitable for many application fields (image processing, speech recognition, pattern recognition, etc.).A significant new feature of signal processing follows from the fact that the number of arithmetical operations is not quite a correct figure of merit for the evaluation of relative efficiency of algorithms.Instead, other parameters, as segmentability of the algorithms and, particularly, the amount of communication between building blocks is becoming of increasing importance.From this point of view the following, three different classes of architectures are discussed in the paper:1. Unconstrained architectures.In this class, at least in principle, no constraints of control or arithmetic are imposed by the available components.This means that any architecture can be designed to be adapted to the chosen algorithm.2. Constrained architectures.This class consists of the implementations based on (i) general purpose computers, minicomputers and microcomputers, (ii) digital signal processors.3. Highly parallel architectures.In this class several approaches to the implementation of parallel processors with different complexity and performance levels can be considered.