基于MPI的Welch功率谱估计并行算法的实现

Implementation of parallel Welch power spectrum estimation based on MPI

Welch算法是一种应用很广的经典功率谱估计算法。但是面对现在日益膨胀的海量数据,单纯在串行Matlab环境下运行Welch算法势必耗费大量的运算时间。尽管Matlab也引入了并行计算工具箱,但是价格昂贵,不利于大范围推广使用。根据Welch算法的原理,在Linux集群环境及消息传递接口MPI的支持下,采用主从并行编程模式,实现了一个开源的Welch并行算法PMWelch。实验结果表明,PMWelch不仅具有Matlab下Welch算法一样的运算结果,还可以大幅减少运行时间。

It is well known that Welch algorithm is one of classical power spectrum estimation which is used widely. But facing a rapidly growing massive data, it will spend a plenty of time to operate Welch algorithm in the serial Matlab circum-stances. Though a parallel computing toolbox is involved in Matlab in 2004, it is too expensive to be used widely. A parallel Welch algorithm, which is implemented using the Master-Slave parallel programming model with the support of Linux clusters and MPI, is proposed here depending on the theory of Welch algorithm. The experiment results show that PMWelch can achieve the same accuracy as Matlab do and can highly improve the speed.