Gigapixel spotlight synthetic aperture radar backprojection using clusters of GPUs and CUDA

Synthetic aperture radar (SAR) image formation via backprojection offers a robust mechanism by which to form images on general, non-planar surfaces, without often restrictive assumptions regarding the planarity of the wavefront at the locations being imaged. However, backprojection presents a substantially increased computational load relative to other image formation algorithms that typically depend upon fast Fourier transforms. In this paper, we present an image formation framework for accelerated SAR backprojection that utilizes a cluster of computing nodes, each with one or more graphics processing units (GPUs). We address the parallelization of the backprojection process among multiple nodes and the scalability thereby obtained, several optimization approaches, and performance as a function of both allocated resources and desired precision. Finally, we demonstrate the achieved performance on a simulated gigapixel-scale data set.