基于多特征信息的深度学习网络调制识别算法  

Modulation recognition algorithm for multiple feature information based on deep learning

作　　者：张航 吴泓霖 余勤[1] 黄承钊 欧阳厚德 雒瑞森 ZHANG Hang;WU Hong-lin;YU Qin;HUANG Cheng-zhao;OUYANG Hou-de;LUO Rui-sen(College of Electrical Engineering,Sichuan University,Chengdu 610065,China)

机构地区：[1]四川大学电气工程学院,四川成都610065

出　　处：《计算机工程与设计》2022年第10期2762-2768,共7页Computer Engineering and Design

基　　金：四川省科技厅重点研发基金项目(2020YFG0051);四川大学校企合作基金项目(19H0355、19H1121);四川大学研究生培养教育创新改革基金项目(GSALK2021019)。

摘　　要：无线电调制方式盲识别存在受电磁环境影响较大和识别种类较多的问题,为此提出一种基于多特征信息结合的调制识别算法。将调制信号的瞬时幅值和瞬时相位值与原始I/Q信号相结合,送入结合深度可分离卷积块与双向长短期记忆网络,引入注意力机制的神经网络模型中。其丰富了调制信号的有效特征信息,可实现调制信号潜在的时序特征和空间特征的互补。实验结果表明,所提算法在高信噪比下,识别准确率可达93%,优于目前流行的基于深度学习的调制识别算法,缓解了对于多进制正交幅度调制方式的识别困难,说明了所提算法的有效性。Aiming at the difficulties of the influence of electromagnetic environment and many modulation types in the blind radio modulation recognition,a modulation recognition algorithm based on the combination of multiple feature information was proposed.The instantaneous amplitude and instantaneous phase value of the modulated signal were combined with the original I/Q signal,and sent to the neural network that combined the deep separable convolution blocks with the bidirectional long short-term memory,and the attention mechanism was introduced.It not only enriches the effective feature information of the modulated signal,but realizes the complementation of the potential temporal features and spatial features of modulated signals.Experimental results show that the recognition accuracy of the proposed algorithm can reach 93%under high signal-to-noise ratio,which is better than that of the current popular modulation recognition algorithms based on deep learning,especially alleviating the recognition difficulty of multiple quadrature amplitude modulation,which illustrates the effectiveness of the proposed algorithm.

关 键 词：调制识别 深度学习 瞬时特征 神经网络 注意力机制 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]