基于LSTM与多特征融合的高铁无线信道场景识别  

Multi-feature fusion for high-speed railway propagation scene recognition based on LSTM networks

作　　者：王英捷 周涛[1] 陶成[1] WANG Yingjie;ZHOU Tao;TAO Cheng(School of Electronics and Information Engineering,Beijing Jiaotong University,Beijing 100044,China)

机构地区：[1]北京交通大学电子信息工程学院,北京100044

出　　处：《电波科学学报》2021年第3期453-459,476,共8页Chinese Journal of Radio Science

基　　金：国家自然科学基金(61701017)。

摘　　要：为满足5G移动通信系统中用户通信业务质量的需求,提出了一种基于长短时记忆(long short term memory,LSTM)与多特征融合的识别方法准确识别高铁无线信道场景,该方法能够与智能决策系统相结合,提高通信系统的整体性能.首先,对不同信道场景的特点及信道特征参数进行阐述,并对整体数据集进行训练集与测试集的划分.然后,提出一种基于LSTM网络的加权平均后融合的方法识别无线信道场景,并与三种常用的特征融合方式的结果相比较.结果表明,本文所提方法在验证集上的识别准确率达到92.2%,受试者工作特征(receiver operating characteristic,ROC)曲线下面积(area under curve,AUC)最大,优于其他特征融合方式.因此,该方法能够为高铁通信系统提供一种精准识别传播场景的方法.In order to fulfill the requirement of service quality of the 5G system,we consider multi-feature fusion methods in this paper to recognize the propagation scene of high-speed railway precisely.After that,the performance of system is improved by using some adaptive technologies.Firstly,we make an explanation on propagation scenes and channel feature parameters,and the dataset is split into two sets:training set and testing set.Then we propose a multilayer long short term memory(LSTM)architecture with a novel weighted score fusion scheme to learn classification from different propagation scenes and compare the results with usage of three regular fusion schemes.The result shows that the recognition accuracy of proposed model on testing dataset reaches 92.2%,the area under curve(AUC)of this model is better than the other three fusion schemes.Therefore,the proposed method provides an accurate recognition application for high-speed railway communication systems.

关 键 词：高铁无线信道 信道场景识别 多特征融合 长短时记忆(LSTM)神经网络 混淆矩阵 受试者工作特征(ROC)曲线 

分 类 号：TN929.5[电子电信—通信与信息系统]