基于深度神经网络的高频谱效率频分复用系统的信道估计方法  

A DNN-Based Channel Estimation Method for Spectral Efficient Frequency Division Multiplexing Systems

作　　者：陈嘉润 余宝贤 王剑莹 张涵[1] CHEN Jiarun;YU Baoxian;WANG Jianying;ZHANG Han(Guangdong Provincial R&D Center of Cardiovascular Individual Medicine & Big Data/School of Physics and Telecommunication Engineering, South China Normal University, Guangzhou 510006, China)

机构地区：[1]广东省心脑血管个体化医疗大数据工程技术研究中心/华南师范大学物理与电信工程学院,广州510006

出　　处：《华南师范大学学报（自然科学版）》2020年第3期17-21,共5页Journal of South China Normal University(Natural Science Edition)

基　　金：国家自然科学基金项目(61471176);教育部蓝火计划(惠州)产学研项目(CXZJHZ201705);广东省科技计划项目(2017A010101015,2017B030308009,2017KZ010101);广东省特支计划项目(2016TQ03X100);广东省自然科学基金项目(2018A030313990,2019A1515011940);广州市科技计划项目(2020-02-03-06-3008-0007);华南师范大学青年教师科研培育基金项目(19KJ16)。

摘　　要：针对高频谱效率频分复用(SEFDM)系统,提出了一种基于深度神经网络(DNN)的信道估计方法.该方法使用等间隔相互正交的导频符号,将其接收信号作为DNN的输入信号,通过4层的全连接DNN结构提取信道特征,最终实现了时域上的信道估计.仿真结果表明:提出的信道估计方法在同等条件下的均方误差(MSE)明显比最小二乘法(LS)的低,对应的解调性能也更优,且对导频数量具有更强的鲁棒性,由此反映出该方法的优越性.A channel estimation method based on deep neural network(DNN)for spectral efficient frequency division multiplexing(SEFDM)systems is proposed.The method employs uniform spaced orthogonal pilot symbols to achieve the channel estimation.To be specific,the received pilot signals are used as the input of the four-layer DNN in order to extract the channel features.Simulation results show that the proposed scheme can yield a smaller mean square error(MSE)and,in turn,perform better demodulation in comparison with the conventional least square(LS)method.In particular,the DNN-based method is more robust to the number of pilots,which indicates its superiority.

关 键 词：深度神经网络 高频谱效率频分复用 信道估计 

分 类 号：TN929.531[电子电信—通信与信息系统]