J4 ›› 2013, Vol. 40 ›› Issue (3): 145-151.doi: 10.3969/j.issn.1001-2400.2013.03.022

• 研究论文 • 上一篇    下一篇

软件雷达信号处理的多GPU并行技术

秦华1;周沫1;察豪1;左炜2   

  1. (1. 海军工程大学 海洋电磁环境研究所,湖北 武汉  430033;
    2. 海军工程大学 电子工程学院,湖北 武汉  430033)
  • 收稿日期:2012-08-26 出版日期:2013-06-20 发布日期:2013-07-29
  • 通讯作者: 秦华
  • 作者简介:秦华(1982-),男,海军工程大学博士研究生,E-mail: blueqinhua@yahoo.com.cn.
  • 基金资助:

    国家部委专项基金资助项目

Research on multi-GPU parallel technology in  software radar signal processing

QIN Hua1;ZHOU Mo1;CHA Hao1;ZUO Wei2   

  1. (1. Electromagnetism Institute of Ocean, Naval Engineering Univ., Wuhan  430033, China;
    2. School of Electronic Engineering, Naval Engineering Univ., Wuhan  430033, China)
  • Received:2012-08-26 Online:2013-06-20 Published:2013-07-29
  • Contact: QIN Hua

摘要:

针对中央处理器(CPU)平台难以满足雷达信号处理实时性不足的问题,利用图形处理器(GPU)并行运算能力强的特点,在CPU-GPU异构系统中采用任务级、数据级和线程级并行策略,设计了基于多图形处理器的雷达信号处理并行算法.新算法根据图形处理器的访存机制进行优化设计,充分利用了图形处理器的并行计算资源.实验结果表明:基于4块图形处理器的多任务并行化计算平台与中央处理器平台相比较,加速比最大可达42.78,并且能够满足雷达信号处理的实时性要求.

关键词: 软件雷达, 信号处理, 并行计算, 图形处理器

Abstract:

Since the conventional central processing unit (CPU) platform could hardly fulfill the real-time requirement of radar signal processing, a software radar signal parallel processing algorithm based on multi-GPU is proposed. To make the best of GPU's strong computational capability, we design a pipeline workflow with 3 levels, task level, data level and thread level in the CPU-GPU heterogeneous system. We optimize the algorithm according to the access mechanism of GPU and take full advantage of the GPU's parallel computational resources. Experimental results show that the proposed GPU platform with 4 GPUs is 42.78 faster than the CPU platform, and manages to meet the real-time requirement of radar signal processing.

Key words: software radar, signal processing, parallel computing, graphic processing unit

Baidu
map