摘要
RFT(Radon-Fourier Transform)是一种广义的MTD算法,可沿着目标径向运动轨迹进行相参积累。然而对距离-速度二维搜索产生的巨大计算量使得其难以快速实现和工程化。针对这个问题,根据雷达信号的回波数据结构和RFT算法思路,提出一种基于GPU的RFT并行化算法。通过实验,GPU平台实现的RFT算法与标准RFT和快速RFT相比,获得了巨大的加速比。另外,通过对比在CPU平台执行的MTD算法,得到在GPU平台上的RFT计算结果在不需要传回主机内存的条件下,计算速度快于在CPU平台上MTD算法。
RFT(Radon-Fourier Transform) is a kind of generalized MTD which can integrate along the track of target. However, it is not easy for RFT to be engineered due to the computing burden. Aiming at this problem, a kind of RFT parallelization strategy is put forward based on GPU and CUDA. Through ex- perimental verification, the execution time of RFT on GPU platform proved higher speedup compared with that of RFT and fast RFT on CPU. In addition, the results show that the execution time is faster than MTD if RFT results are saved in global memory.
出处
《雷达科学与技术》
北大核心
2016年第5期505-509,516,共6页
Radar Science and Technology
关键词
拉东傅里叶变换
并行化
图形处理器
通用并行计算架构
radon Fourier transform(RFT)
parallelization
graphic process unit(GPU)
compute uni- fied device architecture(CUDA)