Fast and Energy Efficient Image Formation of SAR Satellite Data using Embedded GPUs

This paper explores the feasibility of GPU accelerated near real-time image formation of spotlight SAR data acquired by small satellites using the backprojection algorithm. By introducing an approximation for the differential range, the task is computable in single precision. We demonstrate how this allows one to incorporate digital elevation models into the image formation process to reduce geometric distortions and enhance the geospatial accuracy of the resulting images. We provide figures for processing speed and energy consumption as tested on a Nvidia Jetson AGX Orin.