基于生成对抗与卷积神经网络的调制识别方法  

Modulation recognition method based on generative adversarial and convolutional neural network

作　　者：邵凯[1,2,3] 朱苗苗 王光宇 SHAO Kai;ZHU Miaomiao;WANG Guangyu(School of Communication and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China;Engineering Research Center of Mobile Communications of the Ministry of Education,Chongqing 400065,China;Chongqing Key Laboratory of Mobile Communications Technology,Chongqing 400065,China)

机构地区：[1]重庆邮电大学通信与信息工程学院,重庆400065 [2]移动通信教育部工程研究中心,重庆400065 [3]移动通信技术重庆市重点实验室,重庆400065

出　　处：《系统工程与电子技术》2022年第3期1036-1043,共8页Systems Engineering and Electronics

基　　金：中国电子科技集团公司第二十九研究所资助课题。

摘　　要：自动调制识别在频谱监测和认知无线电中占有重要地位。针对现有调制识别算法在低信噪比条件下识别率低的问题,提出一种基于生成对抗网络(generative adversarial network, GAN)和卷积神经网络(convolutional neural network, CNN)的数字信号调制识别方法。在利用平滑伪Wigner-Ville分布将调制信号转换为时频图像(time-frequency images, TFIs)后,在经典GAN中嵌入了剩余密集块(residual dense block, RDB)结构,保证了对TFIs的去噪和修复。通过对经典的剩余网络(residual network, ResNet)模型微调,满足了TFIs的识别与分类。仿真结果表明,所提方法在低信噪比情况下有效地降低了噪声对TFIs的干扰,提高了识别性能。Automatic modulation recognition occupies an important position in spectrum monitoring and cognitive radio. Aiming at the low recognition rate problem of existing modulation recognition algorithms under the condition of low signal to noise ratio, a digital signal modulation recognition method combined generative adversarial network(GAN) and convolutional neural network(CNN). After using the smooth pseudo Wigner-Ville distribution to convert the modulated signal into time-frequency images(TFIs), the residual dense block(RDB) structure is embeded in the classic GAN network to guarantees the denosing and repairmen of TFIs. By fine-tuning the classic residual networkl(ResNet) model of CNN network, the recognition and classification of TFIs is satisfied. The simulation results show that the proposed method effectively reduces the interference of noise on TFIs and improves the recognition performance under the condition of low signal to noise ratio.

关 键 词：自动调制识别 时频分布 卷积神经网络 生成对抗网络 剩余密集块 

分 类 号：TN911.3[电子电信—通信与信息系统]