基于数字孪生信道的大规模MIMO信道低开销预测研究  

Low-Overhead Channel Prediction for Massive MIMO Based on Channel Twin

作　　者：沈子冰 于力[1] 张建华[1] 张宇翔 张振 王启星 姜涛 SHEN Zibing;YU Li;ZHANG Jianhua;ZHANG Yuxiang;ZHANG Zhen;WANG Qixing;JIANG Tao(Beijing University of Posts and Telecommunications,Beijing 100876,China;Inner Mongolia University,Hohhot 010021,China;China Mobile Research Institution,Beijing 100032,China)

机构地区：[1]北京邮电大学,北京100876 [2]内蒙古大学,内蒙古呼和浩特010021 [3]中国移动研究院,北京100032

出　　处：《移动通信》2024年第7期40-45,94,共7页Mobile Communications

基　　金：国家重点研发计划“面向6G复杂应用场景的高动态无线环境预测与重建”(2023YFB2904803);国家自然科学基金“智慧车间复杂传播环境感知、信道重构与资源配置理论研究”(92167202);国家杰出青年科学基金“无线信道的建模理论与实验研究”(61925102);北京邮电大学-中国移动研究院联合创新中心。

摘　　要：准确的CSI对于大规模多输入多输出系统的性能优化至关重要,而其庞大的天线数量会给CSI获取带来巨大的开销与时延。提出了一种基于Crossformer网络的信道预测方法,利用该方法可以在数字孪生信道中获得准确有效的CSI,从而实现对物理信道的精准指导。具体地,将使用间隔多个时隙的历史CSI数据与孪生环境中终端的位置坐标作为神经网络的输入,首先通过维度分段模块将输入的多元时间序列嵌入到二维矢量数组中,以保留时间、空间以及跨维度信息,然后通过二阶注意层来捕获多维度间的依赖关系,最后通过分层编码器-解码器结构利用挖掘到的信道空时依赖关系和环境与信道跨维度依赖关系联合预测未来多个时隙的CSI。最后将所提出的信道预测框架与传统的LSTM与Transformer模型的预测性能进行比较,验证该方法的优越性。Accurate Channel State Information(CSI)is critical for optimizing the performance of massive Multiple Input Multiple Output(MIMO)systems.However,the large number of antennas in these systems results in significant overhead and latency for CSI acquisition.This paper proposes a novel channel prediction method based on a Crossformer network,leveraging digital twin channels to achieve precise and efficient CSI prediction,thereby guiding the physical channels accurately.The method utilizes historical CSI data from multiple previous time slots and terminal position coordinates in the twin environment as inputs to the neural network.Initially,a dimensional segmentation module embeds the multivariate time series inputs into a two-dimensional vector array to preserve temporal,spatial,and cross-dimensional information.Subsequently,a second-order attention layer captures dependencies across multiple dimensions.Finally,a hierarchical encoder-decoder structure exploits the spatiotemporal dependencies of the channel and cross-dimensional dependencies between the environment and the channel to jointly predict the CSI for multiple future time slots.The proposed channel prediction framework is evaluated against traditional LSTM and Transformer models,demonstrating its superior performance.

关 键 词：大规模MIMO Crossformer网络 数字孪生信道 依赖关系 

分 类 号：TN929.5[电子电信—通信与信息系统]