基于CUDA-GPU的宽带高速频谱分析系统的研究

A high-speed wideband spectrometer system based on CUDA-GPU

针对射电信号观测的需求，设计并实现了一种基于统一计算设备架构（CUDA）和图形显示卡（GPU）的宽带高速频谱分析系统。该系统通过运用库利．图基（Cooley-Tukey）快速傅立叶变换算法与谱分析算法实现实时宽带高速频谱分析。系统的关键部分在于通过CUDA来完成运用线程合并算法对数据在CPU-9GPU之间传递时的转换，并使用并行流水算法在总线中共享多核GPU来降低实时运算时间。该系统主要为500m口径射电望远镜工程的高分辨率微波巡视项目而设计，实测中满足目标需求，并可应用在射电信号观测或类似的高速密集数据运算中。

This paper presents the design and implementation of a high-speed wideband Spectrometer system based on the compute unified device architecture （CUDA） and graphics processing unit （GPU） for radio signal observation. This system executes the Cooley-Tukey FFT algorithm and the spectrum analysis algorithm to achieve high-speed analysis of a wideband spectrum. This was accomplished by the algorithm design in CUDA, with an optimized combination of Multi-Memory access, parallel and pipeline processing to reduce the redundant calculation. In particular, this system was designed for the HRMSFF project （high resolution microwave survey for the radio telescope FAST）, and it also can be utilized on other radio signal processing and related computationally intensive calculation.