A Scalable Real-time Signal Processor for Object-oriented Data Flow Applications 

Daniel Scherrer, Hans Eberle
Swiss Federal Institute of Technology
CH-8092 Zurich, Switzerland

Abstract

We present a scalable multiprocessor for real-time signal processing. The basic 
building block is a node with eight processors. A switch-based network is used 
to interconnect multiple nodes. Arbitrary topologies are possible as demanded 
by applications. The network provides quality of service guarantees allowing 
continuous data to be exchanged at given rates and in bounded time. Rather than 
using DSPs we use low-cost RISC CPUs with an architecture extended by instructions 
to accelerate signal processing. On the resulting platform we implemented an 
object-oriented programming environment familiar from general-purpose systems. 
In correspondence to the data flow architecture of the underlying hardware, 
objects exchange data by method calls.

Keywords: signal processing, real-time processing, data flow architecture.