基于FT-MT的RDSAR算法优化实现  

Optimization implementation of RD SAR algorithm based on FT-MT

作　　者：郑利华 杨辉 文楚 王耀华[2] 时洋 ZHENG Lihua;YANG Hui;WEN Chu;WANG Yaohua;SHI Yang(Hunan Greatwall Galaxy Science and Technology Co.,Ltd.,Changsha 410006,China;College of Computer Science and Technology,National Unibersity of Defense Technology,Changsha 410073,China)

机构地区：[1]湖南长城银河科技有限公司,长沙410006 [2]国防科技大学计算机学院,长沙410073

出　　处：《计算机应用文摘》2025年第9期88-93,96,共7页

基　　金：国家自然科学基金项目(6227477);湖南省自然科学基金项目(2023JJ40679)。

摘　　要：合成雷达孔径(Synthetic Aperture Radar,SAR)成像回波数据量大且算法复杂,这使得其在实时应用中的实现面临挑战。针对国防科技大学自主研制的高性能异构多核数字信号处理器(Digital Signal Processor,DSP)FT-MT的体系结构特征及距离-多普勒(Range-Doppler,RD)SAR成像算法的特点,设计了一种面向多核DSP架构的高性能并行RD SAR算法。该算法基于DSP的向量部件,实现了有限脉冲响应滤波(Finite Impulse Response,FIR)、行向和列向快速傅里叶变换(Fast Fourier Transform,FFT)以及快速傅里叶逆变换(Inverse Fast Fourier Transform,IFFT)的向量化计算。同时,结合算子融合、双缓冲和多核并行的优化策略,充分发挥FT-MT架构的优势,显著提升了计算密集型算法的运行效率。实验结果表明,采用并行RD SAR算法相比传统的串行算法具有明显的性能优势;在1.0 GHz频率下,FT-MT单个DSP核处理512 kB(1024×512)图像的时间为23.23 ms,而与德州仪器(Texas Instruments,TI)TMS320C6678在1.2 GHz频率下的单核运行相比,性能加速比可高达20.536;FT-MT四核并行处理512 kB图像的时间为6.089 ms,成功实现了实时性。Synthetic Aperture Radar(SAR)imaging has a large amount of echo data and complex algorithms,which poses challenges for its implementation in real-time applications.A high-performance parallel RD SAR algorithm for multi-core DSP architecture was designed based on the architecture characteristics of FT-MT,a high-performance heterogeneous multi-core digital signal processor(DSP)independently developed by the National University of Defense Technology,and the range Doppler(RD)SAR imaging algorithm.This algorithm is based on the vector components of DSP and implements vectorization calculations for Finite Impulse Response(FIR)filtering,row and column Fast Fourier Transform(FFT),and Inverse Fast Fourier Transform(IFFT).At the same time,by combining operator fusion,double buffering,and multi-core parallel optimization strategies,the advantages of FT-MT architecture are fully utilized,significantly improving the operational efficiency of computationally intensive algorithms.The experimental results show that the parallel RD SAR algorithm has significant performance advantages compared to traditional serial algorithms.At a frequency of 1.0 GHz,the FT-MT single DSP core takes 23.23 ms to process a 512 kB(1024×512)image,while compared to the single core operation of Texas Instruments(TI)TMS320C6678 at a frequency of 1.2 GHz,the performance acceleration ratio can be as high as 20.536.The FT-MT quad core parallel processing time for 512 kB images is 6.089 ms,successfully achieving real-time performance.

关 键 词：合成孔径雷达 多核数字信号处理器 距离-多普勒 向量化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]