首页|基于JDA-BP网络的MQAM信号调制识别

基于JDA-BP网络的MQAM信号调制识别

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
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%。
Recognition method for multiple quadrature amplitude modulation signals based on the JDA-BP network
[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 domain is poor,and it cannot meet the accuracy requirements of the modulation recognition model.2)After the joint probability distribution adaptation of the target and source domains,the probability distribution distance between these domains is reduced,and the modulation recognition network model is retrained and tested.The average modulation recognition rates of the support vector machines classification model,K-nearest neighbor classification model,and BP neural networks increased by 13.92%,8.85%,and 6.80%,respectively.[Conclusions]To improve the accuracy of MQAM signal modulation recognition under small samples and to solve the problem of degradation of MQAM signal modulation recognition performance caused by a change in channel environment,this paper proposes a JDA-BP modulation recognition method based on transfer learning.Results show that using the JDA algorithm increases the recognition accuracy of the modulation recognition network by 9.87%on average,and the modulation recognition accuracy of the proposed methods is higher than that of other methods,and the average modulation accuracy is as high as 73.25%.The overall recognition accuracy has obvious advantages.

joint distribution adaptationmultiple quadrature amplitude modulationmodulation recognitionback propagation neural network

张承畅、李晓梦、李吉利、王艺培、黄彦豪、罗元

展开 >

重庆邮电大学 光电工程学院,重庆 400065

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

重庆市研究生教学改革项目重庆市研究生教学改革项目重庆市教育教学改革研究项目重庆邮电大学教育教学改革重点项目国家级大学生创新创业训练计划

yjg222025YKCSZ23112213153XJG20103S202310617042

2024

10.16791/j.cnki.sjg.2024.05.005

实验技术与管理
清华大学

实验技术与管理

CSTPCD北大核心
影响因子：1.651
ISSN：1002-4956
年,卷(期)：2024.41(5)