一种基于CGAN+CNN的水声通信信号调制识别方法  

Method for Modulation Recognition of Underwater Acoustic Communication Signal Based on CGAN and CNN

作　　者：王彬[1] 王海旺 李勇斌 WANG Bin;WANG Haiwang;LI Yongbin(Information Engineering University, Zhengzhou 450001, China)

机构地区：[1]信息工程大学,河南郑州450001

出　　处：《信息工程大学学报》2021年第1期1-7,共7页Journal of Information Engineering University

基　　金：国家自然科学基金资助项目(61602511)。

摘　　要：为提高复杂海洋环境下水声通信信号调制识别的性能和实用性,提出一种基于条件生成对抗网络和卷积神经网络的调制识别方法。首先,构造一种基于条件生成对抗网络的降噪模块,用于降低海洋环境噪声对通信信号调制特征的影响;然后,采用卷积神经网络完成降噪数据的特征提取和分类识别;同时,利用数据迁移思想构造迁移学习训练数据集,并通过两步迁移学习策略解决目标水域信道下训练数据不足的问题。仿真实验和实际信号测试结果验证了算法的有效性,相比现有方法,低信噪比下的识别率明显提升,在目标水域信道小样本条件下也具有较好的识别性能。To improve the performance and applicability of modulation classification in complex marine environments,an approach based on conditional generative adversarial network(CGAN)and convolutional neural network(CNN)is proposed.First,a CGAN-based noise reduction module is built to reduce the effects of the marine ambient noise on modulation characteristics;second,a CNN-based classification module is utilized to perform feature extraction and recognition on denoised data.Meanwhile,the idea of data transfer is introduced to generate a transfer learning training set,based on which a two-step transfer learning strategy is adopted to address the problem of data scarcity in channels of target water regions.Simulation experiments and practical signal tests both demonstrate the effectiveness of the proposed method.Compared with existing algorithms,the classification accuracy of our method is significantly improved under low SNR condition,and the performance is still robust even with a small sample set available in channels of target water regions.

关 键 词：调制识别 条件生成对抗网 卷积神经网络 降噪 数据迁移 

分 类 号：TN911.7[电子电信—通信与信息系统]