基于DRSN的通信信号调制方式识别方法  

Identification of Communication Signal Modulation Methods Based on DRSN

作　　者：竹杭杰 郭建新 张雨帅 朱锐 黄磊 丁自立 ZHU Hangjie;GUO Jianxin;ZHANG Yushuai;ZHU Rui;HUANG Lei;DING Zili(School of Electronic Information,Xijing University,Xi’an 710123,China;National Defense Engineering Research Institute,Academy of Military Science,Beijing 100036,China)

机构地区：[1]西京学院电子信息学院,陕西西安710123 [2]军事科学院国防工程研究院,北京100036

出　　处：《无线电工程》2024年第7期1643-1651,共9页Radio Engineering

基　　金：陕西省重点研发计划(2021GY-341)。

摘　　要：针对现有的通信信号调制方式识别方法在低信噪比(Signal to Noise Ratio, SNR)条件下存在的识别率较低、调制类型较少和信道类型不够丰富等问题,提出了一种基于深度残差收缩网络(Deep Residual Shrinkage Network, DRSN)的通信信号调制方式识别方法。根据调制识别领域的特点,构建改进的深度残差收缩网络模型,充分利用该网络的注意力机制和软阈值化进行降噪处理,提高模型在低SNR条件下的调制识别能力。实验结果表明,相比残差网络(Residual Network, ResNet)、卷积长短时深度神经网络(Convolutional Long-short-term Deep Neural Network, CLDNN)和卷积门控循环深度神经网络(Convolutional Gated recurrent Deep Neural Network, CGDNN)模型,所提方法在低SNR和5种信道类型条件下对26种调制信号的识别中有效地降低了噪声的影响,在4 dB以上时识别率达到了91.70%,10 dB时识别率在98%以上,取得了较好的识别表现。Existing communication signal modulation recognition methods face challenges such as lower recognition rates under conditions of low Signal to Noise Ratio(SNR),limited modulation types,and a lack of diversity in channel types.A method for communication signal modulation recognition based on the Deep Residual Shrinkage Network(DRSN)is proposed.With the specific features of the modulation recognition domain in mind,an improved deep residual shrinkage network model is constructed.This network fully utilizes its attention mechanism and soft thresholding for noise reduction,enhancing the modulation recognition capability in low SNR conditions.Experimental results demonstrate that,compared to Residual Network(ResNet),Convolutional Long-short-term Deep Neural Network(CLDNN),and Convolutional Gated recurrent Deep Neural Network(CGDNN),the proposed method effectively minimizes noise interference in recognizing 26 types of modulated signals under low SNR and 5 types of channel conditions.The recognition rate achieves 91.70%when the SNR is above 4 dB,and surpasses 98%at 10 dB,showcasing commendable recognition performance.

关 键 词：通信信号 调制识别 深度残差收缩网络 注意力机制 软阈值化 

分 类 号：TN911.3[电子电信—通信与信息系统]