摘要： FFT作为雷达信号处理的关键计算步骤之一，本质上是一个基于数据流的处理过程。以往的FFT计算大多集中在通用计算平台上进行并行计算实现，计算系统存在扩展性和鲁棒性问题。随着科学计算应用在Flink上的逐渐兴起，将FFT在Flink上进行并行和优化，不仅可以很好地利用框架自身良好的系统扩展性和鲁棒性，同时也能使其具备高吞吐的实时性能。基于Flink对FFT流处理算法流程进行了设计和优化，同时针对Flink对适用于FFT计算的缓存窗口机制进行了设计，实验结果表明，改进后FFT并行算法在多个大规模点数下计算速度均有所提高。

中图分类号： TP311.1
文献标识码： A
DOI： 10.19358/j.issn.2096-5133.2021.08.009
引用格式： 钟旭阳，徐云. 基于Flink流处理框架的FFT并行及优化[J].信息技术与网络安全，2021，40(8)：53-59.

FFT parallel algorithm and optimization based on Flink stream processing framework

Abstract： As one of the key calculation steps of radar signal processing, FFT is essentially a processing process based on data stream. In the past, most of the previous FFT calculations concentrated on the implementation of parallel calculations on a general-purpose computing platform, and the computing system has problems with scalability and robustness. With the increasing popularity of scientific computing applications on Flink, parallelizing and optimizing FFT on Flink can not only make good use of the framework′s own strong system scalability and robustness, but also enable it to have high-throughput real-time performance. Based on Flink, this paper designs and optimizes the FFT stream processing algorithm flow. At the same time, it designs a buffer window mechanism suitable for FFT calculation in Flink. The experimental results show that the improved FFT parallel algorithm has a better calculation speed at multiple large-scale points.

Key words : FFT parallel algorithm；radar signal processing；distributed stream processing；Apache Flink