RA-CNN: An Effective Automatic Modulation Recognition method for Joint Communication and Radar System

Automatic modulation classification (AMC) is a crucial stage in the spectrum management, signal monitoring, and control of wireless communication systems. Recently, deep learning-based AMC (DL-AMC) methods have shown remarkable performance. However, existing DL-AMC methods usually focus on single systems in radar or communication and require a sufficient number of samples. To address these limitations, a novel residual attention convolutional neural network (RA-CNN) method is proposed. This method can effectively identify the modulation types in mixed communication and radar waveforms, achieving high classification accuracy even with small sample sizes. This method comprises two main steps: time-frequency image (TFI) processing and RA-CNN network. TFI processing involves converting commonly used IQ signals into time-frequency images, extracting the time-frequency features of the signals. RA-CNN network, on the other hand, can effectively classify modulation types based on these time-frequency images. The combination of the two steps can extract the signal features more efficiently. Simulation results demonstrate that our proposed TFI-based RA-CNN method outperforms the benchmark schemes.