Convolutional Neural Network for Joint Communication and Radar Signals Classification

Automatic modulation classification (AMC) plays an important role in the development of cognitive radio and cognitive radar systems. Due to the distinct aims of radar and communication systems, their commonly used modulation techniques may differ significantly. However, AMC in the radar and communication domains can be closely aligned. Considering the widespread applications of deep learning architectures, particularly, convolutional neural networks (CNN) in classification problems, we propose a CNN-based framework for the joint classification of communication and radar signals. In the proposed framework, a CNN is first trained by signals in the in-phase and quadrature domain and then another CNN is trained by using constellation diagrams to differentiate between close modulations. A publicly available benchmark dataset, which consists of different modulation schemes in the communication domain, has been augmented with our simulated linear frequency modulated radar signals under the same noise condition to validate the accuracy of the proposed framework.