数据驱动的双通道CNN-LSTM调制分类算法

Data-driven Modulation Classification Algorithm Based on Dual-channel CNN-LSTM

为同时利用卷积神经网络(Convolutional Neural Network, CNN)的空间特征提取能力和长短时记忆(Long Short-Term Memory, LSTM)网络的时序特征提取能力,提出了一种由双通道一维CNN与LSTM相互串联的调制分类算法。算法采用数据驱动的方式,直接将信号送入至2路CNN提取其在不同维度的空间特征信息;把2个通道的特征融合信息输入至LSTM学习其时序上的特征;与全连接网络连接实现对5种目标信号的调制分类。实验结果表明,CNN与LSTM相互串联能够学习到更加丰富的特征信息,更有利于分类;与传统方法相比,提出的方法无需人工提取信号特征,减少了预处理步骤并有效提升了识别性能。

To utilize the spatial feature extraction ability of Convolutional Neural Network(CNN) and the temporal feature extraction ability of Long Short-Term Memory(LSTM) network simultaneously, a modulation classification algorithm based on two-channel one-dimensional concatenated CNN and LSTM is proposed. Firstly, the proposed algorithm adopts the data-driven method and directly feeds signals to the two-channel CNN module to extract the spatial feature information in different dimensions. Then, the LSTM network is followed to learn temporal feature information from the CNN module. Finally, LSTM is connected with the fully-connected network to realize modulation classification of five target signals. Experimental results show that the concatenated structure of CNN and LSTM can learn richer feature information, which is more conducive to modulation classification. Compared with the traditional methods, the proposed method does not need to manually extract signal features, reduces the preprocessing steps and effectively improves recognition performance.