摘要
自动调制识别(AMC)是目前各种通信场景中信息获取的前提,由于无线通信过程中的干扰等因素,致使自动调制识别几十年以来一直是一个通信领域的研究难题。针对该难题提出了一种新型深度学习特征融合方案,该方案包含两个分支模型,第一个分支模型为基于注意力机制的双向长短时记忆模型(Attention mechanism based BiLSTM,AMb BiLSTM),该模型从IQ数据中提取信号的幅度相位信息,并使用BiLSTM双向提取信号的语义信息。第二个分支模型为基于多尺度特征提取技术的卷积神经网络(Multi-scale feature extraction CNN,MFE CNN),该模型通过提取IQ数据的浅层特征和深层特征,很好地学习到数据之间的重复特征(Local repeat features)。该方案结合了上述两种模型的优势。通过开源数据集RML2016.10a的验证,证明了所提方案的先进性。
AMC(Automatic modulation classification)is the premise of information acquisition in various communication scenarios.Due to interference and other factors in wireless communication process,AMC has always been a difficult problem in communication field for decades.Aiming at this problem,a new feature fusion-based AMC scheme is proposed,and this scheme consists of two branch models.The first branch model is attention mechanism based BiLSTM(AMb BiLSTM),this model extracts the amplitude and phase information of the signal from IQ data,and uses BiLSTM to extract the semantic information of the signal,while the second branch model is a multi-scale feature extraction-based convolutional neural network(MFECNN),and by extracting low-level and high-level features of IQ data,this model can well learn the local repetitive features between data.This scheme combines the advantages of the above two models,and the verification of open source dataset RML2016.10a proves the advanced nature of the proposed scheme.
作者
陈昱帆
邵尉
李淑丰
彭云飞
CHEN Yu-fan;SHAO WEI;LI Shu-feng;PENG Yun-fei(Army Engineering University of PLA,Nanjing Jiangsu 210007,China)
出处
《通信技术》
2020年第10期2404-2410,共7页
Communications Technology
关键词
自动调制识别
特征融合
CNN
LSTM
注意力机制
automatic modulation classification
feature fusion
CNN
LSTM
attention mechanism
