Ben Cordes
Parallel Backprojection: A Case Study in High Performance Reconfigurable Computing
MS Thesis
Date: April 22, 2008
Abstract:
High-Performance Reconfigurable Computing (HPRC) is a novel approach to providing
the kind of large-scale computing power demanded by modern
scientific applications. One of these applications is Backprojection, an image
formation algorithm that can be used as part of a Synthetic Aperture Radar (SAR)
processing system. Backprojection is an embarassingly parallel application that
is ideal for implementation in hardware, especially on reconfigurable computing
devices such as FPGAs.
This thesis presents an implementation of backprojection for SAR on an HPRC
system. It is intended for use in the Swathbuckler project, a joint collaboration
of multiple national military research agencies. Using simulated data taken
at a variety of ranges, our HPRC implementation of backprojection runs over
200 times faster than a similar software-only program, with an overall application
speedup of better than 50x. In addition to excellent single-node performance,
the
application is easily parallelizable, achieving near-linear speedup when run
on multiple nodes of a clustered HPRC system.
Committee:
Miriam Leeser (advisor)
David Kaeli
Eric Miller, Tufts Univ.