基于GPU实现射电干涉阵列信号实时处理的加速研究

项目名称： 基于GPU实现射电干涉阵列信号实时处理的加速研究

项目编号： No.U1231123

项目类型： 联合基金项目

立项/批准年度： 2013

项目学科： 天文学、地球科学

项目作者： 田海俊

作者单位： 三峡大学

项目金额： 66万元

中文摘要： 射电干涉仪的相关器通常包括射电干涉仪信号的傅里叶变换和信号间交叉互关联两部分，它是射电干涉仪数据实时处理过程中最关键也是计算量最大的部分，随着干涉仪阵元数目N的增加，信号间交叉相关的计算量以O(N^2)急剧增加，如何以低廉而有效的办法来解决目前万亿次每秒甚至亿亿次每秒的实时计算需求，一直是国际关注的一个问题。近年GPU的迅速发展为这一难题的解决带来了新的机遇，我们基于先导实验，找到了一种比传统方法更佳的解决方案。传统的解决方案往往需要为具体项目专门研制FPGA或专用集成电路ASIC等硬件，而该方案基于GPU以软件的形式来实现，具有开发周期较短、扩展性和通用性强，同时费用低廉等优势。该课题的研发成果可为我国重大基础科学工程，尤其射电设备（如“天籁计划”、VLBI等）的实时数据处理带来一种通用、低廉且具有高性能的解决方案。

中文关键词： 相关器；GPU；ROACH2；射电干涉阵；

英文摘要： The cross-correlator of an radio interferometer, which is the key and most complex computational segment during the realtime processing of the signal, usually includes two main parts: one is the Fourier transform of the time serial signals from each antenna channel, and the other one is the cross correlation between all the dual channels. With the increasing of the number N of the antenna in the interferometer, the computational complexity sharply increases as the O(N^2). Facing the current exascale computational requirements, the traditional ways is to developing the special FPGAs or ASICs for the concrete projects. However, its expensive cost and the inflexibility limits its widely usage in the future large interferometer array. Fortunately, the rapid developing of GPU brings a revolutionary way to handle this intractable tasks. Basing on our experiments, we find another better way to meet the requirements using the advanced GPU cards. In this way, a novel and creative implementation of cross correlator could be realized in this project with much lower cost and much higher performance for the large radio interferometer array, especially useful for "TianLai" or "VLBI" etc.

英文关键词： Correlator；GPU；ROACH2；Radio interferometer array；