基于JDA-BP网络的MQAM信号调制识别  

Recognition method for multiple quadrature amplitude modulation signals based on the JDA-BP network

作　　者：张承畅[1] 李晓梦 李吉利 王艺培 黄彦豪 罗元[1] ZHANG Chengchang;LI Xiaomeng;LI Jii;WANG Yipei;HUANG Yanhao;LUO Yuan(School of Optoelectronic Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)

机构地区：[1]重庆邮电大学光电工程学院,重庆400065

出　　处：《实验技术与管理》2024年第5期31-37,共7页Experimental Technology and Management

基　　金：重庆市研究生教学改革项目(yjg222025,YKCSZ23112);重庆市教育教学改革研究项目(213153);重庆邮电大学教育教学改革重点项目(XJG20103);全国大学生创新创业训练项目(S202310617042)。

摘　　要：针对小样本条件下由信号调制识别准确率低和信道环境变化导致调制识别网络性能下降的问题,提出了一种基于联合分布适配-反向传播神经网络(JDA-BP)调制识别方法。通过改变信道环境生成概率分布不同的多进制正交振幅调制(MQAM)信号,提取MQAM信号的瞬时统计特征和高阶累积量组成样本,构建3个概率分布不同的数据集,使用联合分布适配(JDA)算法缩小数据集间的特征差异,并将适配后的数据集送入BP神经网络进行训练和测试。对比实验表明,在目标域为小样本的条件下,该文方法针对源域和目标域概率分布不同的情况,能有效地减小概率分布距离,信号调制识别平均准确率可达73.25%;相比于比未使用JDA-BP方法,调制识别准确率平均提高了6.80%。[Objective]The modulation recognition method based on feature extraction has achieved many achievements and applications in automatic modulation recognition,assuming that the training and test data have consistent probability distributions.However,when the communication scene changes,the trained classifier has low accuracy in modulation recognition of the new sample because of the difference in the probability distributions of the training classification network and test datasets.Moreover,data samples are difficult to obtain in real time in some specific scenarios,limiting the application of modulation recognition methods based on feature extraction to real-time scenarios with small sample conditions.To address these concerns,a JDA-BP modulation recognition network is proposed based on the back propagation(BP)neural network and joint distribution adaptation(JDA)method which uses the JDA algorithm to reduce the joint probability distribution distance between the two datasets after adaptation.This method makes the model robust on datasets with different communications and improves its generalization ability.[Methods]Multiple quadrature amplitude modulation(MQAM)with different probability distributions of data is generated by changing the channel environment.Three instantaneous statistical features and three high-order cumulants of MQAM signals are extracted to form data samples in six dimensions,and three datasets with probability distributions are constructed.These three datasets were used as the source and target domains for pairwise adaptation.The source domain dataset was used as the training set of the BP neural network,and the target domain dataset was tested on the BP neural network trained on the source domain dataset to obtain the signal modulation recognition accuracy of the JDA-BP neural network.[Results]Comparative experiments show that:1)when the probability distribution distance between the source and target domain datasets is large,the generalization of the classification model trained in the source

关 键 词：联合分布适配 多进制正交振幅调制 调制识别 反向传播神经网络 

分 类 号：TN971.1[电子电信—信号与信息处理]