CLPNet:基于深度学习大规模MIMO的CSI反馈网络  

CLPNet:A Deep-learning-based CSI Feedback Network for Massive MIMO Systems

作　　者：刘为波 颜彪[1] 沈麟 丁宇舟 LIU Weibo;YAN Biao;SHEN Lin;DING Yuzhou(College of Information Engineering,Yangzhou University,Yangzhou 225009,China)

机构地区：[1]扬州大学信息工程学院,江苏扬州225009

出　　处：《无线电工程》2022年第9期1660-1665,共6页Radio Engineering

基　　金：国家自然科学基金(61601403)。

摘　　要：在频分双工(Frequency Division Duplex,FDD)模式下,大规模多输入多输出(Multiple-Input Multiple-Output,MIMO)系统的容量增益依赖于精确的信道状态信息(Channel State Information,CSI)。因此,终端(User Equipment,UE)需要准确地将CSI反馈给基站(Base Station,BS)。同时,天线数量的增多导致CSI反馈开销巨大,已经成为大规模MIMO系统的一个瓶颈。利用深度学习(Deep Learning,DL)在处理数据上的优势,提出了一种CSI反馈网络——CLPNet,该网络在CLNet的基础上进行了改进,利用注意力机制、多感受野与大卷积通道的相互结合,可以恢复更丰富的特征。仿真结果表明,CLPNet的收敛速度和恢复质量得到进一步提升。In Frequency Division Duplex(FDD)mode,the capacity gain of massive Multiple-Input Multiple-Output(MIMO)systems depends on accurate Channel State Information(CSI).Therefore,User Equipment(UE)needs to accurately feed CSI back to Base Station(BS).Meanwhile,the increase in the number of antennas leads to huge overhead of CSI feedback,which has become a bottleneck problem in massive MIMO systems.Using the advantages of Deep Learning(DL)in processing data,a CSI feedback network,CLPNet,is proposed.Improved on the basis of CLNet,this network utilizes the combination of attention mechanism,multiple receptive fields and large convolution channels to recover richer features.Experimental simulations prove that the convergence speed and recovery quality of CLPNet are further improved.

关 键 词：大规模多输入多输出 深度学习 信道状态信息反馈 感受野 注意力机制 

分 类 号：TN974[电子电信—信号与信息处理]