摘要
针对多天线组阵接收中未同步多路信号的联合调制识别问题,提出了一种基于深度神经网络的多天线组阵信号联合调制识别方法。该方法将多天线组阵信号联合调制识别问题转化为多元分类判决问题,使用深度神经网络实现信号调制方式的识别。首先,构建了多个卷积层、循环层和全连接层级联结构的深度神经网络;然后,基于广泛采用的RadioML框架构建训练、测试和验证数据集,数据集中单个信号样本为存在参数差异的多天线信号时域波形;最后,使用基于CPU+GPU架构的服务器对网络进行训练,并对其性能进行测试。结果表明,所提方法能够有效抑制多路信号间参数差异的影响,实现多路信号联合调制识别,接收单元数目越多,分类判决性能越好,且在高信噪比下优势体现更加明显。
In order to solve the problem of joint modulation recognition of unsynchronized multi-antenna signals in reception of multi-antenna array,a DNN-based joint modulation recognition method of multi-antenna array signals is proposed in this paper.In this method,the joint modulation recognition of multi-antenna array signals is transformed into multielement classification decision,and the DNN(deep neural network)is utilized to realize signal recognition in modulation mode.The DNN with a cascaded structure of multiple convolution layer,cycle layer and full connection layer is constructed.The training,testing and verification dataset are constructed on the basis of the widely-used RadioML framework.The single signal sample in the dataset is the multi-antenna signal time domain waveform with different parameters.The network was trained by the server based on CPU+GPU architecture,and its performance was tested.The results show that the proposed method can effectively suppress the influence of parameter differences between multi-antenna signals and realize joint modulation recognition of multiantenna signals.The more the number of receiving units is,the better the classification performance becomes,and the advantages are more obvious in high SNR.
作者
张凯
田瑶
刘义
ZHANG Kai;TIAN Yao;LIU Yi(Key Laboratory of Optoelectronic Countermeasure Test and Evaluation Technology,Luoyang 471000,China;Unit 96862 of PLA,Luoyang 471000,China)
出处
《现代电子技术》
2022年第17期24-28,共5页
Modern Electronics Technique
基金
国家自然科学基金项目(62001476)。
关键词
多天线组阵
联合调制识别
特征提取
分类判决
深度神经网络
训练数据集
网络训练
multi-antenna array
joint modulation recognition
feature extraction
classification discrimination
DNN
training dataset
network training
