基于三阶运动模型的GRFT算法的并行化实现
Parallel Implementation for GRFT Algorithm Based on Three-order Motion Model
-
摘要: 广义随机傅里叶变换(GRFT:Generalized Radon-Fourier Transform)是一种广义的MTD(Moving Target Detection)算法,通过搜索目标的速度、加速度、加加速度等高阶运动信息,补偿多个脉冲间的相位来完成相参积累。这种采用搜索的方法完成众多脉冲的相参积累,必然会带来巨大的计算量,不利于雷达的实时检测。针对这个问题,根据目标各运动参数之间搜索的独立性和雷达回波信号的存储特点及GRFT算法思路,提出一种基于图形处理器(GPU:Graphic Processing Unit)的GRFT算法,实现了高维搜索并行化问题, 并采用通用并行计算架构(CUDA:Computer Unified Device Architecture)完成了GRFT算法的具体实现。仿真结果表明: GRFT算法的计算速度在GPU平台上得到显著提高。
Abstract: Generalized Radon-Fourier Transform (GRFT) is a generalized algorithm of Moving Target Detection (MTD), which achieves coherent integration by compensating the phase of multiple pulses through searching the target’s velocity, acceleration and jerk. This kind of algorithm inevitably brings a huge amount of computation because of multi-dimension search, which is not conducive to real-time detection of radar. To address this issue, a GRFT algorithm based on graphic processing unit (GPU) is proposed in this paper, which combines search independence between motion parameters, the storage of radar echo signal and idea of GRFT algorithm. Computer unified deceive architecture (CUDA) is adopted to realize this algorithm. Numerical simulation results show that the computing speed of GRFT algorithm is significantly improved on the GPU platform.
下载: