基于卷积神经网络的5G无线信道参数学习方法  

A learning method for 5G wireless channel parameters based on convolutional neural network

作　　者：黄骏 唐慧 柴利[1] Huang Jun;Tang Hui;Chai Li(Engineering Research Center for Metallurgical Automation and Detecting Technology of Ministry of Education,Wuhan University of Science and Technology,Wuhan 430081,China)

机构地区：[1]武汉科技大学冶金自动化与检测技术教育部工程研究中心,湖北武汉430081

出　　处：《武汉科技大学学报》2022年第2期149-154,共6页Journal of Wuhan University of Science and Technology

基　　金：国家自然科学基金资助项目(61800339,61625305).

摘　　要：深度学习是解决5G无线信道建模中参数学习问题的有效手段,但学习网络的超参数选择对网络性能的影响较大,而常规的手动调参方法往往难以达到令人满意的学习效果,为此提出一种基于卷积神经网络(CNN)的5G无线信道参数学习方法,其中CNN网络超参数采用贝叶斯优化进行自动设置。利用仿真软件Wireless InSite建立了5G无线通信室外场景数据集,针对不同信道参数设计了相应的卷积神经网络,通过实验对比分析了贝叶斯自动寻优和手动调整超参数的学习效果,结果表明本文方法优势明显。Deep learning is an effective method for parameter learning in 5G wireless channel modeling,but hyperparameters of the learning network have great impact on network performances,and it is difficult to achieve satisfactory learning effect by using conventional manual parameter tuning method.Therefore,this paper proposed a learning method for 5G wireless channel parameters based on convolutional neural network(CNN),in which CNN network hyperparameters were automatically set by Bayesian optimization.The outdoor scenario dataset of 5G wireless communication was established by using Wireless InSite simulation software,and corresponding convolutional neural networks were designed for different channel parameters.The learning effects of Bayesian automatic optimization and those of manual adjustment of hyperparameters were compared by experiments.The results show that the proposed method has obvious advantages.

关 键 词：5G无线通信 信道建模 信道参数 卷积神经网络 贝叶斯优化 超参数 深度学习 

分 类 号：TN929.5[电子电信—通信与信息系统]