te
A COMPRESSION FILTER FOR REAL-TIME SYNTHETIC
APERTURE RADAR (SAR) DATA
W. Dillen
G. Kluge
European Space Agency - Operations centre (ESOC)
Robert Bosch-Strasse 5
D-6100 DARMSTADT
Federal Republic of Germany
0. Abstract
A high-throughput hardware processor is presented which performs
real-time range compression of SAR-data. It is based on a "systolic
array" architecture which computes the convolution algorithm without
using Fourier transforms.
A prototype of this machine has been built in semi-custom CMOS gate
array IC's. Full advantage is taken of the regularity and repetitivi-
ty which is inherent to the architecture. Each chip implements a
(real) convolution of twelve kernel coefficients which results in an
arithmetic rate of 120 million operations per second (MOPS) when
operated at 5 MHz. At this speed, the SAR-range processor as a whole,
which contains 240 gate arrays to realise a 720-point complex convo-
lution, achieves 28.8 GOPS.
The price of one gate array chip being approximately $ 200, the cost
of arithmetic operations is under $ 2 per MOPS. This example shows
that special-purpose systolic arrays, based on commercially available
integrated circuits, can offer arithmetic processing power several
orders of magnitude cheaper than programmable array processors or so-
called supercomputers.
1. Introduction
Initially, the processing of SAR-data has been carried out by
analog computers, using optical methods. They had the advantage of
basically instantaneously accomplishing data compression in both
range and azimuth direction [1]. However, the drawbacks of analog
machines compared to digital processors - lack of flexibility, sensi-
tivity to noise, temperature changes, aging, etc. - became more
significant as the latter grew faster and more powerful.
401