基于CNN-LSTM-NOMA的车车通信频谱效率研究  

Research on Spectral Efficiency of T2T Communication Based on CNN-LSTM-NOMA

作　　者：孙博 杨宗烨 张博文 马腾飞 陈维民 SUN Bo;YANG Zongye;ZHANG Bowen;MA Tengfei;CHEN Weimin(Engineering Design and Research Institute,CRSC Research&Design Institute Group Co.,Ltd.,Beijing 100070,China;School of Automation and Electrical Engineering,Lanzhou Jiaotong University,Lanzhou Gansu 730070,China;Guangzhou Signaling&Communication Depot,China Railway Guangzhou Group Co.,Ltd.,Guangzhou Guangdong 510610,China;Lanzhou Communication Depot,China Railway Lanzhou Group Co.,Ltd.,Lanzhou Gansu 730099,China)

机构地区：[1]北京全路通信信号研究设计院集团有限公司工程设计研究院,北京100070 [2]兰州交通大学自动化与电气工程学院,甘肃兰州730070 [3]中国铁路广州局集团有限公司广州电务段,广东广州510610 [4]中国铁路兰州局集团有限公司兰州通信段,甘肃兰州730099

出　　处：《中国铁道科学》2025年第1期192-199,共8页China Railway Science

基　　金：中国国家铁路集团有限公司科技研究开发计划课题(N2023G081)。

摘　　要：为解决城市轨道交通5G毫米波车车通信系统频谱资源不足的问题,对T2T通信频谱效率进行研究。首先,构建下行多用户簇的非正交多址接入(NOMA)系统模型,在保证簇内各天线能够通过串行干扰消除(SIC)成功分配功率的同时,完成NOMA系统中总频谱效率的设计;其次,提出簇内天线功率分配方案,在不设定速率限制的情况下实现最大功率分配;最后,采用卷积神经网络和长短期记忆神经网络辅助的NOMA通信方法 (CNN-LSTM-NOMA),对输入数据进行训练,得到最佳的频谱效率。结果表明:同一试验数据集下,所提方法的决定系数可达0.995 71,高于采用单一神经网络辅助的CNN-NOMA和LSTM-NOMA方法,两者决定系数分别为0.966 54和0.979 96,且CNN-LSTM-NOMA方法的频谱效率更接近最优无约束数字预编码。该研究可为未来城市轨道交通中提高T2T通信的频谱效率提供理论依据。In order to address the issue of spectrum resource scarcity in 5G millimeter-wave Train-to-Train(T2T)communication systems for urban rail transit,the spectral efficiency of T2T communications is investigated.Firstly,a system model of Non-Orthogonal Multiple Access(NOMA)with downlink multi-user clusters is constructed.This model ensures successful power allocation among antennas within each cluster through Successive Interference Cancellation(SIC)while also achieving the design of total spectral efficiency in the NOMA system.Secondly,a power allocation scheme for antennas within clusters is proposed to achieve the maximum power allocation without setting rate constraints.Lastly,a Convolutional Neural Network and Long Short-Term Memory Neural Network-Assisted NOMA communication method(CNN-LSTM-NOMA)is employed to train the input data and obtain the optimal spectral efficiency.The results demonstrate that,under the same experimental dataset,the coefficient of determination for the proposed method reaches 0.99571,which is higher than that of the CNN-NOMA and LSTM-NOMA methods assisted by a single neural network,and the determination coefficients of these two methods are 0.96654 and 0.97996,respectively.Moreover,the spectral efficiency of the CNN-LSTM-NOMA method is closer to the optimal unconstrained digital precoding.The research provides a theoretical basis for improving the spectral efficiency of T2T communications in future urban rail transit systems.

关 键 词：城市轨道交通 非正交多址接入 频谱效率 毫米波 T2T通信 

分 类 号：U285[交通运输工程—交通信息工程及控制]