采用图形处理器和最新的通用并行计算架构设计了射电天文多相滤波器组,并对其性能指标进行了测试和分析。利用图形处理器强大的浮点数计算和高效并行执行能力实现了多相滤波器、快速傅里叶变换算法加速,改善了多相滤波器组算法的执行效率。实验结果表明,设计的多相滤波器组具有一定的灵活性和可扩展性,能够实现射电信号的高速滤波及信道化,可有效提高射电望远镜数字终端算法的并行数据处理能力和计算效率。
A Polyphase filter bank is designed with the GPU and the latest NVIDIA CUDA parallel architecture for radio astronomy and its performance is tested and analyzed. Both GPU's powerful floating-point calculations and high performance parallel execution capabilities are adopted in this design and they accelerate PFB and FFT algorithms, therefore improve the efficiency of the filter bank. Experiment results show that the polyphase filter bank designed in this paper has certain flexibility and extensibility; it can implement high- speed filtering and high-speed channelization and improve computing efficiency and parallel data processing capability for astronomical digital backend algorithms.