北京航空航天大学学报 ›› 2022, Vol. 48 ›› Issue (3): 533-543.doi: 10.13700/j.bh.1001-5965.2020.0591

• 论文 • 上一篇    下一篇

基于调制卷积神经网络的空地数据链信道估计

刘春辉1, 王美琳2, 董赞亮2, 王沛2   

  1. 1. 北京航空航天大学 无人系统研究院, 北京 100083;
    2. 北京航空航天大学 电子信息工程学院, 北京 100083
  • 收稿日期:2020-10-19 发布日期:2022-03-29
  • 通讯作者: 刘春辉 E-mail:liuchunhui2134@buaa.edu.cn
  • 基金资助:
    科技创新2030-“新一代人工智能”重大项目(2020AAA0108200);北京市自然科学基金(4204102)

Channel estimation of air-ground data link based on modulated convolutional neural network

LIU Chunhui1, WANG Meilin2, DONG Zanliang2, WANG Pei2   

  1. 1. Institute of Unmanned System, Beihang University, Beijing 100083, China;
    2. School of Electronics and Information Engineering, Beihang University, Beijing 100083, China
  • Received:2020-10-19 Published:2022-03-29
  • Supported by:
    Science and Technology Innovation 2030-Key Project of “New Generation Artificial Intelligence” (2020AAA0108200); Beijing Municipal Natural Science Foundation (4204102)

摘要: 针对复杂环境下空地数据链正交频分复用(OFDM)系统信道估计精度不足的问题,提出了一种基于调制卷积神经网络(MCNN)和双向长短时记忆网络(BiLSTM)结合的信道估计算法。利用最小二乘算法(LS)提取初始信道状态信息(CSI);利用MCNN网络提取初始CSI的深度特征,并对网络模型进行压缩;利用BiLSTM网络对最终CSI进行预测,实现信道估计。利用构建的空地信道模型生成信道系数数据集,实现神经网络模型的训练与测试。仿真结果表明:与传统算法和现有深度学习方法相比,所提出的信道估计算法具有更小的估计误差,高信噪比条件下的系统误码率(BER)性能提升接近一个数量级;由于引入了调制滤波器技术,随着神经网络层数增加,网络模型参数量大幅减少。

关键词: 正交频分复用(OFDM), 深度学习, 信道估计, 空地信道模型, 多径效应

Abstract: Aimed at the inaccuracy of channel estimation of orthogonal frequency division multiplexing (OFDM) system in the complex air-ground data link environment, this paper proposes a channel estimation algorithm based on the modulated convolutional neural network (MCNN) and bidirectional long short-term memory (BiLSTM) network. First, least square (LS) algorithm is used to extract the initial channel state information (CSI), then MCNN network is used to extract the depth characteristics of the initial CSI while compressing the network model, and finally BiLSTM network is used to predict the final CSI and realize channel estimation. In the aspect of experimental verification, the air-ground channel model constructed is used to generate the channel coefficient dataset, so as to realize the training and testing of neural network model. The simulation results show that compared with the traditional methods and the existing deep learning method, the proposed channel estimation method has a lower estimation error, and the performance of the bit error ratio (BER) of the system under the condition of high SNR is improved by nearly an order of magnitude. Due to the introduction of the modulation filter technology, the number of network model parameters decreases remarkably with the increase of the number of neural network layers.

Key words: orthogonal frequency division multiplexing (OFDM), deep learning, channel estimation, air-ground channel model, multipath effect

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