物联网学报 ›› 2019, Vol. 3 ›› Issue (1): 8-13.doi: 10.11959/j.issn.2096-3750.2019.00085

• 理论与技术 • 上一篇    下一篇

一种基于深度学习的物联网信道状态信息获取算法

廖勇,姚海梅,花远肖   

  1. 重庆大学通信与测控中心,重庆 400044
  • 修回日期:2019-02-14 出版日期:2019-03-01 发布日期:2019-04-04
  • 作者简介:廖勇(1982- ),男,四川自贡人,博士,重庆大学副研究员、博士生导师,主要研究方向为下一代无线通信技术、AI 及其在行业中的应用。|姚海梅(1992- ),女,江西吉安人,重庆大学硕士生,主要研究方向为AI及其在无线通信中的应用。|花远肖(1994- ),男,四川阆中人,重庆大学硕士生,主要研究方向为AI及其在无线通信中的应用。
  • 基金资助:
    国家自然科学基金资助项目(61501066);重庆市基础与前沿研究计划项目(cstc2015jcyjA40003);中央高校基本科研业务费基金资助项目(106112017CDJXY500001)

Channel state information acquisition algorithm based on deep learning for IoT

Yong LIAO,Haimei YAO,Yuanxiao HUA   

  1. Center of Communication and TT&C,Chongqing University,Chongqing 400044,China
  • Revised:2019-02-14 Online:2019-03-01 Published:2019-04-04
  • Supported by:
    The National Natural Science Foundation of China(61501066);The Fundamental Research Funds for the Central Universities(106112017CDJXY500001)

摘要:

针对基于大规模多输入多输出(MIMO)的物联网系统中用户侧将信道状态信息(CSI)发送到基站时反馈开销大的问题,提出一种基于深度学习的CSI反馈网络用来反馈CSI。该网络首先使用卷积神经网络(CNN)提取信道特征矢量和最大池化层通过降维来达到压缩CSI的目的,然后使用全连接和CNN将压缩的CSI解压,恢复原始信道。仿真结果表明,与现有的CSI反馈方法相比,所提出的CSI反馈网络恢复的CSI更接近原始信道,重构质量明显提高。

关键词: 大规模MIMO, 物联网, CSI反馈, 深度学习

Abstract:

To solve the problem of high feedback overhead when the user sends channel state information (CSI) to the base station in massive multiple input multiple output (MIMO) based on Internet of things system,a CSI feedback network based on deep learning was proposed to feedback CSI.Firstly,the proposed network used convolutional neural network (CNN) to extract channel feature vectors and maxpooling to compress the data.Then the compressed CSI was decompressed by using full connection and CNN to restore the original channel.The simulation results show that compared with the existing CSI feedback methods,the CSI recovered by the proposed CSI feedback network is closer to the original channel,and the reconstruction quality is improved significantly.

Key words: massive MIMO, Internet of things, CSI feedback, deep learning

中图分类号: 

No Suggested Reading articles found!