基于混合信号多域特征和Transformer的干扰识别  

Interference identification based on mixed signal multidomain feature and Transformer framework

作　　者：阳鹏飞 何羚 王茜 王睿笛 张明志 YANG Pengfei;HE Ling;WANG Qian;WANG Ruidi;ZHANG Mingzhi(School of Aeronautics and Astronautics,University of Electronic Science and Technology of China,Chengdu 611731,China;Aircraft Swarm Intelligent Sensing and Cooperative Control Key Laboratory of Sichuan Province,Chengdu 611731,China)

机构地区：[1]电子科技大学航空航天学院,四川成都611731 [2]飞行器集群智能感知与协同控制四川省重点实验室,四川成都611731

出　　处：《系统工程与电子技术》2024年第6期2138-2145,共8页Systems Engineering and Electronics

基　　金：四川省自然科学基金(2022NSFSC0545,2023NSFSC0494)资助课题。

摘　　要：针对无线通信信道易受到蓄意射频信号干扰问题,提出了一种从混合信号中识别干扰类型的方法。通过改进经典Transformer结构,形成新型网络模型Multidomain-former,以提取多域特征和识别信号干扰类型。首先,通过特定的序列划分机制对输入频谱进行预处理,并通过线性嵌入和位置编码保留原始顺序特征;其次,设计了逆傅氏变换和傅氏变换结合的编码模块,使Multidomain-former能同时提取频域和时域特征。使用通用仪器和收发天线搭建了无线收发信道,在不同干信比条件下对混合信号频谱进行采集,得到训练集和测试集。干扰对比实验通过所提Multidomain-former网络模型完成,并将经典的Transformer结构和其他常见的深度学习模型与所提网络模型进行了对比。对比实验结果表明,在干信比小于10 dB时,所提模型性能相较于经典Transformer在识别正确率方面有2%~3%的提升;在干信比等于-5 dB时,所提模型以最少参数量和次低计算复杂度获得了比另外5种基准网络高3.0%~9.3%的识别率。Aiming at the vulnerability of wireless communication channels to interference from intentional radio frequercy signals,a method of interference identification from mixed signals is presented.In this work,a novel Transformer model named Multidomain-former is devised to serve as the multidomain features extractor and jamming types recognizer.The proposed model is built with the following characteristics:the original spectral data is firstly preprocessed by a specific sequence partitioning mechanism.Meanwhile,the initial sequence features are reserved by linear embedding and position coding.Secondly,an encoding module which jointly adopting inverse Fourier transform and Fourier transform is designed,by this means Multidomain-former can obtain both frequency domain and time domain features.A real wireless transceiver channel is established utilizing universal instruments and transceiver antennas,and the mixed signal spectrum is collected subject to different jamming-to-signal ratios(JSR)to form training and test data sets.The interference classification experiments are carried out sequentially by the proposed Multidomain-former,in comparison with the classic Transformer and other popular deep learning networks as well.It is shown that Multidomain-former achieves the best performance with the least number of parameters and lower complexity.With the condition of the JSR is less than 10 dB,the probability for correct classification of Multidomain-former is 2%~3%higher than that of classic Transformer.When the JSR is equal to-5 dB,the performance of Multidomain-former is proofed to increase by 3.0%~9.3%on correct classification rate compared with other benchmarks.

关 键 词：混合信号 多域特征提取 干扰识别 TRANSFORMER 干信比 

分 类 号：TN973.1[电子电信—信号与信息处理]