基于深度学习的上行传输过程毫米波通信波束选择方法  

Deep learning based beam selection for uplink millimeter wave communications

作　　者：马文焱 戚晨皓[1] MA Wenyan;QI Chenhao(School of Information Science and Engineering,Southeast University,Nanjing 210096,China)

机构地区：[1]东南大学信息科学与工程学院

出　　处：《合肥工业大学学报（自然科学版）》2019年第12期1644-1648,共5页Journal of Hefei University of Technology：Natural Science

基　　金：国家自然科学基金资助项目(61871119);江苏省基础研究计划资助项目(BK20161428);上海航天科技创新基金资助项目(SAST2016072)

摘　　要：文章研究了多用户上行传输过程毫米波大规模多输入多输出(multi-input and multi-output,MIMO)系统的波束选择问题,提出了一种基于深度学习的波束选择方法。针对使用透镜的多用户毫米波大规模MIMO上行传输过程,提出一种面向波束选择的深度学习框架,通过信道数据预先对神经网络进行离线训练,然后将实测信号输入训练好的神经网络在线预测信道直达径对应的波束,从而实现波束选择;基于该深度学习框架制定了具体的训练细则,采用柔性最大值交叉熵函数作为损失函数,使用自适应矩估计优化器优化神经网络参数。仿真结果表明,该文提出的基于深度学习的波束选择方法优于现有的正交匹配追踪方法。In this paper, the beam selection for uplink multi-user millimeter wave massive multi-input and multi-output(MIMO) systems is studied, and a beam selection method based on deep learning is proposed. In view of the multi-user millimeter wave massive MIMO uplink transmission with lens antenna array, we first propose a deep learning framework for beam selection. The neural network is trained offline using the channel data, and then the received signal is input into the trained neural network to predict the beam corresponding to the line of sight(LOS) channel path. Then specific training rules are developed based on the proposed deep learning framework. We use the softmax cross entropy function as the loss function, and use the adaptive moment estimation(Adam) optimizer to optimize the parameters of the neural network. Simulation results show that the proposed deep learning based method outperforms the existing method based on orthogonal matching pursuit(OMP).

关 键 词：毫米波通信 波束选择 深度学习 透镜阵列 大规模多输入多输出(MIMO)系统 

分 类 号：TN928[电子电信—通信与信息系统]