Video-SAR Implementation on Various Computing Architectures

Forming synthetic aperture radar images in real-time using the backprojection algorithm (BPA) requires optimal use of available hardware. We discuss the challenges and feasibility of implementing the BPA to achieve good performance on graphics processor units (GPU) platforms. We will compare the achieved performances between high-end GPUs, CPUs (x86-64 and ARM) as well as edge devices with low-performance GPUs. Benchmark results prove the feasibility of real-time processing with embedded GPUs using less than 40W of power.