车联网V2I场景下基于GNN的SC-FDMA智能信道估计  

Intelligent Channel Estimation of SC-FDMA Based on GNN for V2I Scenarios in Internet of Vehicles

作　　者：廖勇[1] 尹子松 田肖懿 LIAO Yong;YIN Zi-song;TIAN Xiao-yi(School of Microelectronics and Communication Engineering,Chongqing University,Chongqing 400044,China)

机构地区：[1]重庆大学微电子与通信工程学院,重庆400044

出　　处：《电子学报》2024年第3期772-782,共11页Acta Electronica Sinica

基　　金：国家自然科学基金(No.61501066);重庆市自然科学基金(No.cstc2019jcyj-msxmX0017)。

摘　　要：随着车联网的迅猛发展,车对路基础设施(Vehicle to Infrastructure,V2I)通信对车联网的可靠性和时延提出了更高的要求,而信道估计是接收机高可靠低时延通信的重要保障.为解决传统信道插值算法不能有效拟合V2I信道快时变特性、自适应多普勒频移能力弱和传统神经网络可解释性不强的问题,本文提出基于图神经网络(Graph Neural Network,GNN)的单载波频分多址(Single Carrier-Frequency Division Multiple Access,SC-FDMA)智能信道估计算法.该算法将信道频率响应中的数据点作为图的节点、符号间时域相关性作为边,将图化后的数据送入GraphSAGE信道插值器(GraphSAGE Channel Interpolator,GCI)中,通过边更新、聚合操作、节点更新三大模块进行网络训练,同时采用多普勒频移矢量作为节点特征控制网络拟合不同多普勒条件的信道,使得网络具备可解释性.最后,系统仿真验证了在不同速度环境下算法的有效性和鲁棒性,较线性插值、样条插值以及全连接网络,本文所提GCI在低、中和高速移动环境下具有最优的误码率(Bit Error Rate,BER)和归一化均方误差(Normalized Mean Square Error,NMSE)性能,特别地,在200 km/h高速移动条件下GCI的优势更为明显.With the rapid development of the Internet of vehicles,vehicle to infrastructure(V2I)communication puts forward higher requirements for the reliability and delay of vehicle networking.Channel estimation is an important guaran⁃tee for high reliable and low-latency communication of receiver.To solve the problems that the traditional channel interpola⁃tion algorithm cannot effectively fit the fast time-varying characteristics of V2I channel,the ability of adaptive Doppler fre⁃quency shift is weak,and the interpretability of traditional neural network is not strong,this paper presents a single carrierfrequency division multiple access(SC-FDMA)intelligent channel estimation algorithm based on graph neural network(GNN).The proposed algorithm takes the data points in the channel frequency response as the nodes of the graph and the in⁃ter-symbol time domain correlation as the edges.The graphical data is fed into the GraphSAGE channel interpolator(GCI).The network training is carried out through the three modules of edge update,aggregation operation and node update.At the same time,the Doppler shift vector is used as the node feature control network to fit the channels with different Doppler conditions,making the network interpretable.Finally,the system simulation verifies the effectiveness and robustness of the algorithm in different speed environments.Compared with linear interpolation,spline interpolation and fully connected net⁃work,the proposed GCI has the best performance of bit error rate(BER)and normalized mean square error(NMSE)in low,medium and high-speed mobile environments,especially,the advantage of GCI is more obvious under the condition of 200 km/h high-speed movement.

关 键 词：车联网 V2I 双选衰落信道 高速移动 多普勒频移 GNN 信道估计 信道频率响应 

分 类 号：TN929.5[电子电信—通信与信息系统]