基于AlexNet卷积神经网络的5G信号调制方式识别  

Modulation Recognition of 5G Signals Based on AlexNet Convolutional Neural Network

作　　者：张清 胡国兵 赵嫔姣 Zhang Qing;Hu Guobing;Zhao Pinjiao(School of Electronics and Optical Engineering,School of Microelectronics,Nanjing University of Posts and Telecommunications,Nanjing 210023,China;School of Electronic Information Engineering,Jinling Institute of Technology,Nanjing 211169,China)

机构地区：[1]南京邮电大学电子与光学工程学院、微电子学院,南京210023 [2]金陵科技学院电子信息工程学院,南京211169

出　　处：《信息化研究》2020年第2期36-43,共8页INFORMATIZATION RESEARCH

基　　金：江苏省基础研究计划(自然科学基金)(NO.BK20161104);金陵科技学院高层次人才引进项目(jit-b-201630,jit-b-201724);江苏省第十二批次六大人才高峰项目(DEXX022)。

摘　　要：针对非协作条件下信号调制识别对信号的先验信息要求较高,且人工选取特征复杂等问题,文章提出一种基于AlexNet卷积神经网络的5G信号调制方式识别算法。针对π/2-BPSK、QPSK、16QAM、64QAM、256QAM这5种常用5G信号(3GPP R15协议建议),选择其星座图作为AlexNet网络的输入特征,构建识别分类算法。仿真结果表明,该算法在15dB信噪比下对5种常用5G信号的平均识别正确率达90%,相较于已有基于信号散布图特征的识别算法,其性能更优。Aiming at the problem that the signal modulation recognition under the non-cooperative condition requires high priori information of signals and the complex artificial selection for the features are complex,this paper proposes a modulation recognition method based on AlexNet convolutional neural network for 5 G(5 th-Generation)signal modulation.For the five commonly used 5 G signals(3 GPP R15 protocol recommendations)ofπ/2-BPSK,QPSK,16 QAM,64 QAM and 256 QAM,the constellation is selected as the input feature of AlexNet network to construct the recognition classification algorithm.The simulation results show that the average recognition accuracy of the five commonly used 5 G signals is up to 90%under the 15 dB SNR.Compared with the existing recognition algorithms based on signal scatter plots,the performance is better.

关 键 词：调制识别 AlexNet卷积神经网络 星座图 5G信号 

分 类 号：TN91[电子电信—通信与信息系统]