Multiparadigm Computing for Space-Based Synthetic Aperture Radar.
Adam Jacobs, Grzegorz Cieslewski, Casey Reardon, Alan D. George · 2008
Abstract—Projected computational requirements for future space missions are outpacing technologies and trends in conventional embedded microprocessors. In order to meet the necessary levels of performance, new computing technologies are of increasing interest for space systems, such as reconfigurable devices and vector processing extensions. These new technologies can also be used in tandem with conventional general-purpose processors in the form of multiparadigm computing. By using FPGA resources and AltiVec extensions, as well as MPI extensions for multiprocessor support, we explore possible hardware/software designs for a synthetic aperture radar application. Design of key components of the SAR application including range compression and azimuth compression will be discussed, and hardware/software performance tradeoffs analyzed. The performance of these key components will be measured individually, as well as in the context of the entire application. Fault-tolerant versions of range and azimuth compression algorithms are proposed and their performance overhead is evaluated. Our analysis compares several possible multiparadigm systems, achieving up to 18 × speedup while also adding fault tolerance to a pre-existing SAR application. I.