GPU平台下针对SAR地面快视系统的RD算法优化与实现  

Optimization and Implementation of RD Algorithm in SAR Ground Quick Look System According to GPU Platform

作　　者：杨仁忠[1] 张萄[1] 林波涛[1] 韦宏卫[1] 

机构地区：[1]中国科学院对地观测与数字地球科学中心,北京100094

出　　处：《遥感技术与应用》2012年第2期237-242,共6页Remote Sensing Technology and Application

基　　金：遥感卫星应用国家工程实验室遥感卫星数据接收系统项目(O92601101C)

摘　　要：针对SAR地面系统实时关键成像技术进行研究,采用GPU(Graphic Processing Unit)硬件平台CUDA(Computer Uniformed Device Architecture)编程模型,对传统合成孔径雷达的RDA(Range Doppler Algorithm)算法核心部分进行了针对性的设计与实现,并在GPU专用科学计算平台Tesla C1060上进行了实验。结果表明其处理速度是一台主流4核心8线程CPU的20倍以上,并且相对RadarSat-1卫星可以达到10倍左右的实时率;基于GPU的处理方式较好地实现了SAR实时成像系统。Synthetic Aperture Radar（SAR） imaging requires huge computation load,thus it needs high-end computation platform to construct the system when comes to real time imaging process.Typical SAR real time imaging systems adopt DSP,FPGA and cluster.This paper presents a new approach that realizes SAR real time imaging processing based on Computer Unified Device Architecture（CUDA）,which is a new parallel programming model introduced recently on Graphic Processing Unit（GPU）.The study had analyzed the difference between data parallel mode of CUDA and task parallel mode of typical platform,including studied RD algorithm structure and optimized each step according to CUDA parallel programming model,then the corresponding RD algorithm structure is implemented.This poper also conducted the experiment results showed the speed on this platform is about 20~30 times of a common high-end PC（Intel Core-i7 950）,and 10 times of the speed required for real time imaging for RadarSat-1.It proves a promising way to realize SAR real time imaging system.

关 键 词：合成孔径雷达(Synthetic APERTURE Radar) 实时成像 并行计算 高性能计算 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]