Parallel Processing of SAR Imaging Algorithms for Large Areas Using Multi-GPU

The procedure of Synthetic Aperture Radar (SAR) data processing is extraordinarily time-consuming. The traditional processing modes are hard to satisfy the demand for real-time which are based on CPU. There have been some implementations on singe GPU owing to its excellent ability of parallel processing. But there is no implementation on multi-GPU for larger areas. A multi-GPU parallel processing method is proposed including task partitioning and communication hiding in this paper. Furthermore, a detailed comparison of implementation effect among Range Doppler algorithm (RDA), Chirp Scaling algorithm (CSA) and \( \omega K \) algorithm (\( \omega KA \)) has been shown in this paper by implementing them on multi-GPU. Experimental results show \( \omega KA \) has the longest execution time and the highest speedup compared to RDA and CSA. All the algorithms satisfy real-time demand on multi-GPU. Researches can select the most suitable algorithm according to our conclusions. The parallel method can be extended to more GPU and GPU clusters.