系统工程与电子技术 ›› 2025, Vol. 47 ›› Issue (10): 3512-3519.doi: 10.12305/j.issn.1001-506X.2025.10.34

• 通信与网络 • 上一篇    

基于深度学习的复合干扰信号识别

刘佳楠(), 宋志群, 李勇, 刘丽哲, 夏金涛, 王斌, 李行健, 汪畅   

  1. 中国电子科技集团公司第五十四研究所先进通信网全国重点实验室,河北 石家庄 050081
  • 收稿日期:2024-04-22 出版日期:2025-10-25 发布日期:2025-10-23
  • 通讯作者: 李勇 E-mail:jiananliu@stu.xidian.edu.cn
  • 作者简介:刘佳楠(2000—),男,硕士研究生,主要研究方向为抗干扰通信
    宋志群(1963—),男,研究员,博士,主要研究方向为无线通信系统设计、抗干扰通信、认知无线电
    刘丽哲(1978—),女,研究员,硕士,主要研究方向为无线通信系统
    夏金涛(1994—),男,助理工程师,硕士,主要研究方向为无线通信系统
    王 斌(1968—),男,研究员,硕士,主要研究方向为无线通信系统
    李行健(1992—),男,工程师,博士,主要研究方向为无线通信系统
    汪 畅(1995—),男,助理工程师,硕士,主要研究方向为无线通信系统
  • 基金资助:
    先进通信网全国重点实验室基金(FFX23641X002,SCX23641X011)资助课题

Composite interference signal recognition based on deep learning

Jianan LIU(), Zhiqun SONG, Yong LI, Lizhe LIU, Jintao XIA, Bin WANG, Xingjian LI, Chang WANG   

  1. National Key Laboratory of Advanced Communication Networks,China Electronics Technology Group Corporation No.54 Research Institute,Shijiazhuang 050081,China
  • Received:2024-04-22 Online:2025-10-25 Published:2025-10-23
  • Contact: Yong LI E-mail:jiananliu@stu.xidian.edu.cn

摘要:

在无线通信环境中压制式复合干扰信号对通信系统的正常工作有着严重的影响,针对其特征提取和识别较为困难的问题,提出一种基于短时傅里叶变换(short time Fourier transform, STFT)和残差卷积网络的复合干扰识别算法。该算法将STFT得到的时频域信息作为输入,同时对复合干扰信号的种类和干噪比进行识别,为了使模型更加适合部署在移动端上,采用幻影卷积代替普通卷积。仿真结果表明,在干噪比为?15~10 dB的范围内,该算法在5种单一干扰及其复合而成的10种复合干扰信号种类识别任务上准确率可以达到99.97%,在干噪比识别任务上准确率可以达到99.04%。相比于残差卷积网络,该算法在几乎不降低准确率的前提下可以使模型参数量减小38.4%,计算复杂度降低46.6%,更加符合移动端的要求。

关键词: 深度学习, 复合干扰识别, 短时傅里叶变换, 残差卷积网络, 幻影卷积

Abstract:

In the wireless communication environment suppressed composite interference signal has a serious impact on the normal operation of the communication system, for its feature extraction and identification is more difficult, a composite interference identification algorithm based on short time Fourier transform (STFT) and residual convolution network is proposed. The algorithm takes the time-frequency domain information obtained from the STFT as the input, and identifies the type of composite interference signal and the jamming-to-noise ratio at the same time. In order to make the model more suitable to be deployed on the mobile terminal, the ghost convolution is used instead of the ordinary convolution. Simulation results show that in the range of jamming-to-noise ratios from ?15 to 10 dB the algorithm can achieve 99.97% accuracy on the task of identifying five single interferences as well as ten composite interference signal types compounded by them, and 99.04% accuracy on the task of identifying jamming-to-noise ratio. Compared with the residual convolutional network, the proposed algorithm can reduce the number of model parameters by 38.4% and the computational complexity by 46.6%, which is more in line with the requirements of the mobile terminal, with almost no reduction in the accuracy.

Key words: deep learning, composite interference recognition, short time Fourier transform (STFT), residual convolutional network, ghost convolution

中图分类号: