基于截断迁移与并行残差网络的调制识别算法  

Modulation Recognition Algorithm Based on Truncated Migration and Parallel ResNet

作　　者：郭业才[1,2] 王庆伟 Yecai Guo;Qingwei Wang(School of Electronic and Information Engineering,Nanjing University of Information Science&Technology,Nanjing 210044,China;Jiangsu Collaborative Innovation Center on Atmospheric Environment and Equipment,Nanjing University of Information Science&Technology,Nanjing 210044,China)

机构地区：[1]南京信息工程大学电子与信息工程学院,江苏南京210044 [2]南京信息工程大学江苏省大气环境与装备技术协同创新中心,江苏南京210044

出　　处：《系统仿真学报》2022年第9期2009-2018,共10页Journal of System Simulation

基　　金：国家自然科学基金(61673222,61371131);江苏省高校自然科学研究重大项目(13KJA510001);江苏高校品牌专业建设项目(PPZY2015B134)

摘　　要：针对卷积神经网络提取的信号时序特征受限问题,提出一种截断迁移的数据预处理算法,将采样矩阵一端的距离单位截断,迁移到另一端,依次合并成新的矩阵,使卷积神经网络提取到更多的采样点,比较更多的符号信息。同时提出一种改进的并行残差神经网络,通过两路并行的支路同时关注水平和垂直2个方向的特征。结果表明,该算法比普通卷积网络提高约10%的准确率,改进的网络在信噪比为14 dB时,准确率为93.78%,信噪比大于0 dB时,准确率均在91%以上。A truncated migration data preprocessing algorithm is proposed for the problem of limited time series characteristics of the signal extracted by convolutional neural network.The distance unit at one end of the sampling matrix is truncated,migrated to the other end to form a new matrix,allowing the convolutional neural network to extract more sampling points and compare more symbolic information.An improved parallel ResNet is proposed,which focuses on features in both horizontal and vertical directions simultaneously by two parallel branches.The results show that the algorithm has an accuracy rate of about 10%higher than that of ordinary convolutional networks.When the signal-to-noise ratio is 14 dB,the improved network has an accuracy rate of 93.78%and when the signal-to-noise ratio is greater than 0 dB,the accuracy rate is above 91%.

关 键 词：卷积神经网络 截断迁移 数据预处理 并行残差神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP391[自动化与计算机技术—控制科学与工程]