A framework for rendering high resolution synthetic aperture radar images on heterogeneous architectures

This paper presents a modular, extensible, framework for rendering images from synthetic aperture radar (SAR) data on an array of heterogeneous processor architectures. Our design supports real-time reconstruction of a two-dimensional image from a matrix of echo pulses and their corresponding response values using Backprojection. Key to our design is the division of the Backprojection problem into atoms, which decomposes Backprojection along both it´s input and output data dimensions, and allows scheduling algorithms to explicitly minimize communication overhead through the deliberate assignment of atoms to processors. Performance analysis on a cluster of 10 Tesla C2050 GPUs, and 5 Intel Xeon Processor X5650 CPUs has shown speedup that closely follows the number of processors in the cluster.