电子学报 ›› 2020, Vol. 48 ›› Issue (9): 1695-1702.DOI: 10.3969/j.issn.0372-2112.2020.09.005

• 学术论文 • 上一篇    下一篇

基于核协同表示与鉴别投影的辐射源调制识别

李东瑾, 杨瑞娟, 李晓柏, 董睿杰   

  1. 中国人民解放军空军预警学院, 湖北武汉 430019
  • 收稿日期:2019-06-04 修回日期:2020-02-20 出版日期:2020-09-25 发布日期:2020-09-25
  • 作者简介:李东瑾 男,1992年出生,四川广元人,空军预警学院博士研究生,主要研究方向为一体化系统智能应用及表示学习.E-mail:li_dong_jin@163.com
    杨瑞娟 女,1964年出生,四川中江人,空军预警学院教授、博士生导师,主要研究方向为雷达通信一体化及雷达组网.E-mail:ruijuany@sohu.com
    李晓柏 男,1983年出生,甘肃陇西人,空军预警学院讲师、博士.主要研究方向为雷达通信一体化及波形设计.E-mail:lxb2cici@163.com
    董睿杰 男,1995年出生,新疆库尔勒人,硕士.主要研究方向为一体化系统智能应用.E-mail:drjkuw@126.com
  • 基金资助:
    国防科技创新特区基金(No.17H86304ZT00302201)

Emitter Signal Modulation Recognition Based on Kernel Collaborative Representation and Discriminative Projection

LI Dong-jin, YANG Rui-juan, LI Xiao-bai, DONG Rui-jie   

  1. PLA Air Force Early Warning Academy, Wuhan, Hubei 430019, China
  • Received:2019-06-04 Revised:2020-02-20 Online:2020-09-25 Published:2020-09-25

摘要: 针对辐射源识别中的特征稳定性不高和低信噪比环境适应性不足等问题,提出了一种基于二次时频分布、核协同表示与鉴别投影的识别方法.首先,通过时频变换、稀疏域降噪和二次特征提取的预处理算法降低噪声干扰和特征冗余,以获取高稳定性的二次时频分布特征;然后,采用核协同表示和鉴别投影思想进行降维学习和字典学习,以提升数据低维表征和类间鉴别能力;最后,通过离线训练完成系统优化并用于分类验证.仿真结果表明,二次时频分布特征具备较高稳定性,识别方法具备较强鲁棒性、时效性和适应性;当信噪比为-10dB时,该方法对8类辐射源信号的整体平均识别率达到96.88%.

关键词: 辐射源识别, 核协同表示, 鉴别投影, 二次时频分布, 批量随机梯度下降法

Abstract: Aiming at the problems of low feature stability in emitter signal recognition and poor adaptability to low signal-to-noise(SNR) environment,a recognition method based on secondary time-frequency distribution,kernel collaborative representation and discriminative projection(KCRDP) was proposed.First,the pre-processing algorithms of time-frequency transform,sparse domain noise reduction,and secondary feature extraction are used to reduce noise interference and feature redundancy,and secondary time-frequency distribution features with high stability were obtained.Then,the kernel collaborative representation and discriminative projection ideas are used to complete the dimensionality reduction learning and dictionary learning to improve the low-dimensional representation and inter-class discrimination capabilities of the data.Finally,the system is optimized through offline training and used for classification verification.Simulation results show that the secondary time-frequency distribution feature has high stability,and the recognition method has strong robustness,timeliness and adaptability.When the SNR is -10dB,the overall average recognition rate of the eight signals reaches 96.88%.

Key words: emitter signal recognition, kernel collaborative representation, discriminative projection, secondary time-frequency distribution(STFD), mini-batch stochastic gradient descent(MSGD)

中图分类号: