Accelerating Time-Domain SAR Raw Data Simulation for Large Areas Using Multi-GPUs

Large areas synthetic aperture radar (SAR) raw data simulation, which contains various actual system errors, is one of the main tasks in SAR system design and development. The growth of swath and resolution results in a significant increase in data volume as well as the simulation time. This poses a challenge for SAR raw data simulation considering system errors. For recent years, the graphics processing unit (GPU)-based scalable parallel method has been applied to raw data simulation. In this paper, we investigate time-domain SAR raw data simulation for large areas on multi-GPUs architecture, which can not only simulate raw data of large areas by task partitioning and scheduling, but also improve the efficiency of current GPU-based algorithm by access conflict optimization and fine-grained parallel pipeline. Experimental results show that the proposed multi-GPUs-based raw data simulation method achieves a 5× speedup compared to the current GPU-based method on single GPU, and a significant over 500× speedup on 4 GPUs compared to traditional CPU-based simulation. These results verify that multi-GPUs-based time-domain method is very suitable for large data volume raw data simulation, especially for the case of wide swath and high resolution.