基于Storm的分布式实时信号处理系统

Storm-based distributed real-time signal processing system

针对传统基于数字信号处理器(DSP)的实时信号处理系统存在的编程复杂、硬软件耦合紧、可靠性和扩展性差等问题,提出了一种基于大数据流式实时处理系统Storm的新型信号处理机架构。该架构利用通用处理器和Storm系统替代DSP处理器和其相应的数据流程序框架,对阵列信号处理算法进行模块化分解,并揉入Storm拓扑中,由Storm通用处理器集群完成数据的传输和处理。利用常规波束形成算法对新型系统进行测试,其对一拍数据进行傅里叶变换与波束形成计算的时间分别为59.13 ms和5.27 ms,远小于两个数据节拍之间的时间间隔。利用人工去除节点来模拟节点失效,系统能够进行自我修复并继续运行。结果与分析表明新型架构完全能够满足水声阵列信号处理系统对实时性的要求,且可靠性高、可扩展性与程序可复用性好。

Concerning the disadvantages of conventional signal processing system based on Digital Signal Processor（ DSP） like low stability, low expansibility and tight coupling of software and hardware, a new kind of signal processing system based on steaming real-time signal processing system Storm was proposed. Without using DSP, the new system used generalpurpose processor and Storm for data transmission and processing. Array signal processing algorithms were modularized and distributed on the topology of Storm. A test on the system using conventional beamforming algorithm was conducted. In the test, it took 59. 13 ms and 5. 27 ms respectively to do Fourier transformation and beamforming when the situation that one computing node failed was simulated, the system still successfully finished the tasks. The results and the analyses show that the new system meets the requirements of underwater acoustic signal processing system and it has a good performance of stability, expansibility and reusability.