摘要： 在研究信号稀疏分解理论及其最常用的匹配追踪算法的基础上，针对MP算法存在的计算量过大的问题，提出一种基于并行计算系统实现信号稀疏分解的方法。该方法利用8台微机，采用MPI消息传递机制，以100 M高速以太网作为互联网络，构建了一套Beowulf 并行计算系统，在此系统上通过编制并行程序来实现MP算法。实际测试表明这种方法具有很高的并行计算效率，分解时间从单机75 min左右下降到8机并行11 min左右，大大提高了信号稀疏分解的速度。
关键词:
稀疏分解,
匹配追踪,
并行计算,
MPI消息传递
Abstract: After studying Matching Pursuit(MP) algorithm of signal sparse decomposition, this paper proposes a new approach to improve the speed of MP algorithm, and it describes how to build a Beowulf parallel computing system with 8 PCs. Its parallel computation is implemented by Message-Passing-Interface(MPI), and a 100Mb/s high speed Ethernet network interconnects all PCs. Test is made using parallel computing program to measure the parallel efficiency of the system, results show that this parallel can reduce the MP algorithm computing time-cost from 75 minutes with a PC to 11 minutes with 8 PCs.
Key words:
sparse decomposition,
Matching Pursuit(MP),
parallel computing,
MPI message passing
中图分类号:
刘 浩;杨 辉;尹忠科;王建英. 基于MPI并行计算的信号稀疏分解[J]. 计算机工程, 2008, 34(12): 19-21.
LIU Hao; YANG Hui; YIN Zhong-ke; WANG Jian-ying. Signal Sparse Decomposition Based on MPI Parallel Computing[J]. Computer Engineering, 2008, 34(12): 19-21.