基于GPU的视频SAR加速处理架构

GPU-based processing architecture for video SAR acceleration

视频合成孔径雷达(Video Synthetic Aperture Radar,ViSAR)可对目标区域实现动态监测,具备低时延、高分辨率的特点。高性能计算是ViSAR系统的关键技术。为满足ViSAR系统的实时性要求,提出了一种基于图形处理器(Graphics Processing Unit,GPU)的ViSAR成像方案。该方案将极坐标格式算法(Polar Format Algorithm,PFA)部署到GPU上,使用并发流技术、异步并行技术等方法进行优化,并结合SAR成像的帧率与数据重叠率之间的关系,充分发挥GPU的计算性能。结果表明:本架构处理单张尺寸大小为2048×2048的数据耗时0.17 s,较中央处理器(Central Processing Unit,CPU)平均加速32.8倍,有效解决了ViSAR系统的实时成像问题。

Video Synthetic Aperture Radar(ViSAR)systems enable dynamic monitoring of the target area with low latency and high resolution.High performance computing is the key technology of ViSAR system.In order to meet the real-time requirements of ViSAR system,a ViSAR imaging scheme based on Graphics Processing Unit(GPU)is proposed.The scheme deploys the Polar Format Algorithm(PFA)to the GPU.Optimization using multi-stream processing techniques,asynchronous parallelism,etc.,which take full advantage of the GPU's computational performance.The results show that the architecture takes 0.17 seconds to process a single 2048×2048 size,which is 32.8 times faster than the average acceleration of the Central Processing Unit(CPU).It effectively solves the real-time imaging problem of ViSAR system.