北京航空航天大学学报 ›› 2019, Vol. 45 ›› Issue (9): 1855-1863.doi: 10.13700/j.bh.1001-5965.2018.0724

• 论文 • 上一篇    下一篇

基于深度学习的无人机数据链信噪比估计算法

孙宇航1, 曾国奇2, 刘春辉2, 张多纳1   

  1. 1. 北京航空航天大学 电子信息工程学院, 北京 100083;
    2. 北京航空航天大学 无人系统研究院, 北京 100083
  • 收稿日期:2018-12-17 出版日期:2019-09-20 发布日期:2019-09-29
  • 通讯作者: 曾国奇 E-mail:zengguoqi@buaa.edu.cn
  • 作者简介:孙宇航,男,硕士研究生。主要研究方向:电磁信号智能感知方法;曾国奇,男,博士,高级实验师,硕士生导师。主要研究方向:无人机遥测系统、相控阵天线;刘春辉,男,博士,工程师。主要研究方向:电磁信号智能感知方法;张多纳,男,博士研究生。主要研究方向:电磁信号智能感知方法。
  • 基金资助:
    国防基础科研计划(JCKY2017601C006)

SNR estimation algorithm for UAV data link based on deep learning

SUN Yuhang1, ZENG Guoqi2, LIU Chunhui2, ZHANG Duona1   

  1. 1. School of Electronic and Information Engineering, Beihang University, Beijing 100083, China;
    2. Institute of Unmanned System, Beihang University, Beijing 100083, China
  • Received:2018-12-17 Online:2019-09-20 Published:2019-09-29
  • Supported by:
    National Defense Basic Research Program (JCKY2017601C006)

摘要: 无人机数据链通信受到各种自然与人为的干扰,信噪比(SNR)是信道状态和通信质量的有效评估指标。为解决传统估计算法信噪比估计精度不足的问题,提出了一种卷积神经网络(CNN)与长短时记忆(LSTM)网络结合的估计模型。利用仿真与实测相结合的方式,构建了一个包含不同信噪比、调制方式、衰落信道等信息的无人机通信信号数据集;在网络训练阶段,将样本序列进行分割,对分割后的每一部分序列使用CNN-LSTM网络提取深度特征,多次训练并保存模型参数;在测试阶段,利用构建好的测试集完成对算法的验证与测试,得到信噪比估计值。实验表明,相比于传统信噪比估计算法与单一网络结构的深度学习算法,所提算法的均方误差最低,实现了对信噪比的高精度估计。

关键词: 无人机, 信噪比(SNR)估计, 深度特征, 卷积神经网络(CNN), 长短时记忆(LSTM)网络

Abstract: UAV data link communication is subject to natural and artificial interferences. The signal-to-noise ratio (SNR) is an effective evaluation indicator of channel state and communication quality. In order to address insufficient SNR estimation accuracy involved in traditional estimation algorithm, an estimation model which combines convolutional neural networks (CNN) and long short term memory (LSTM) network is proposed. By means of both simulation and actual measurement, a data set of UAV communication signals is constructed with multiple SNRs, modulation modes, fading channels and other information included. In the network training phase, the sample sequence is segmented, CNN-LSTM is used to extract the deep feature of each part, and the model parameters are saved through multiple trainings. In the test phase, the constructed test set is used to verify and test the algorithm, and the SNR estimation value is obtained. Experiments show that compared with traditional SNR estimation algorithm and single-network deep learning method, the proposed algorithm can help achieve the lowest mean square error for different levels of SNR, thus achieving the high-precision estimation of SNR.

Key words: UAV, signal-to-noise ratio (SNR) estimation, deep feature, convolutional neural networks (CNN), long short term memory (LSTM) network

中图分类号: 


版权所有 © 《北京航空航天大学学报》编辑部
通讯地址:北京市海淀区学院路37号 北京航空航天大学学报编辑部 邮编:100191 E-mail:jbuaa@buaa.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发