Emitter signal recognition based on improved CLDNN

Traditional methods of emitter signal recognition often need to extract features manually, which not only requires high professional knowledge, but also can not guarantee that the features selected by humans are suitable for the recognition of most types of signals, meanwhile, the recognition accuracy and speed cannot be taken into account. To solve the above problems, convolutional long short-term deep neural network(CLDNN), a deep learning model commonly used in speech processing, is introduced into the recognition of emitter signal, and the long short-term memory (LSTM) layer in this model is changed into bidirectional gated recurrent unit (Bi-GRU) layer. The input of the model is the original time series data, and the processes of feature extraction and classification recognition are carried out in the network to avoid the incompleteness of artificial feature selection. Experimental results show that the proposed model can recognize the signal types effectively at low signal to noise ratio, and a balance between recognition accuracy and recognition speed is achieved compared with other models. © 2021, Editorial Office of Systems Engineering and Electronics. All right reserved.