一种面向透镜阵列毫米波多用户MIMO的CNN天线选择算法  

A CNN Antenna Selection Algorithm for Lens Array Millimeter-wave Based Multi-user MIMO System

作　　者：程文铭 钱蓉蓉 任文平[1] CHENG Wenming;QIAN Rongrong;REN Wenping(School of Information Science&Engineering,Yunnan University,Kunming 650500,China)

机构地区：[1]云南大学信息学院,云南昆明650500

出　　处：《无线电工程》2023年第5期1145-1152,共8页Radio Engineering

基　　金：国家自然科学基金青年科学基金项目(61701433);云南省科技厅面上项目(2018FB099)。

摘　　要：毫米波通信中大规模多输入多输出(MIMO)系统的实现需要使用大量射频链,导致了系统硬件成本和能耗过高的问题。为了解决这一问题,考虑利用透镜天线阵列基于方向的能量聚焦特性并结合天线选择技术能够在性能损失不明显的同时有效地减少射频链的使用数量。针对具有透镜天线阵列的多用户MIMO系统,提出了一种基于卷积神经网络(CNN)的天线选择算法,将信道状态信息和以信道容量为指标而制作的标签作为网络的输入,对神经网络进行训练,用训练好的网络模型为新的信道状态信息选择出最优的天线组合。仿真结果表明,所提方案获得的和速率性能接近全数字方案,并且所提天线选择算法的分类准确率可以达到98%左右,优于其他基于机器学习的天线选择算法。The implementation of a massive MIMO system in millimeter-Wave(mmWave)communication requires a large number of radio frequency chains,which leads to the problems of high system hardware cost and energy consumption.To solve this problem,it is considered that utilizing the direction-based energy focusing properties of lens antenna arrays combined with antenna selection techniques can effectively reduce the number of radio frequency chains without significant performance loss.For the multi-user MIMO system with lens antenna array,an antenna selection algorithm based on Convolutional Neural Network(CNN)is proposed.The channel state information and the label made with the channel capacity as the index are used as the input of the network to train the neural network.The trained network model selects the optimal antenna combination for the new channel state information.The simulation results show that,the sum-rate performance of the proposed scheme is close to the full-digital scheme,and the classification accuracy of the proposed antenna selection algorithm can reach about 98%,which is better than other antenna selection algorithms based on machine learning.

关 键 词：毫米波通信 大规模多输入多输出 透镜天线阵列 天线选择 卷积神经网络 

分 类 号：TN928[电子电信—通信与信息系统]