Three GPU-based Parallel Schemes for SAR Back Projection Imaging Algorithm

Back projection (BP) algorithm has the ability of accurate imaging for synthetic aperture radar (SAR). However, the drawback of low computational efficiency limits the wide application of BP. In this paper, three different parallel schemes based on GPU for BP are presented. According to the features of serial structure of BP, the first two schemes are parallelized through mapping the pulse repetition intervals (PRIs) and pixels in the scene onto the parallel threads in GPU, respectively; another scheme combines the first two schemes together. In order to select the best scheme for application, we also provide a coarse criterion after detailed analysis of the three schemes from aspects of parallelization degree, memory occupancy and memory access efficiency. The simulation results are carried out and compared, which validates the efficacy of the three schemes.