This paper was accepted for presentation at the Sixth MIT Conference on Advanced Research in VLSI, April 2-4, 1990.
Very High Performance Signal Processing using the Residue Number SystemOddvar Søråsen, Yngvar Lundh and Jon Kristian Hagene(1)
Department of Informatics
University of Oslo
P O Box 1080 Blindern, N-0316 Oslo, Norway
AbstractHigh throughput Digital Signal Processing is performed by a special purpose VLSI circuit in CMOS. Parallel processing and complexity are exploited to overcome the need for high speed. Thereby low power consumtion is achieved The system is designed for an application in high performance spectrometry. The Direct Fourier Transformation is used. This eliminates the need for input buffering and facilitates programmable processing of selected spectral lines. Any number of input samples can be used.
High performance processing is achieved using the Residue Number System (RNS) and Residue arithmetics. A modular pipelined organization is employed. It comprises arithmetic units, accurate generation of trigonometric values and input and output conversion, all implemented on-chip. The arithmetic is performed by a number of 5 bit modulo adders and multipliers. Tables are used extensively for implementing the simple building blocks. Each VLSI circuit is able to continously perform Fourier transformations, either by calculating one frequency component for a given number of detectors or a specified number of frequency components for one detector. The typical speed of the input data is 80 ns per sample. The compactness and flexibility of the procedure are important for demanding instrumentation in remote sensing, environmental control and research applications.
(1) Jon Kristian Hagene is currently working as a Senior Research Scientist at Norsk Elektro Optikk AS, Norway