Deep Learning for Coexistence Radar-Communication Waveform Recognition

This paper presents a deep learning-based method to automatically recognize the waveform types of wireless signals in coexistence radar-communication systems. In this method, we leverage smooth pseudo Wigner-Ville distribution (SPWVD) to analyze signals in time and frequency domains concurrently, and then design an efficient deep convolutional neural network (CNN) with several processing blocks which integrate residual connection and attention connection to improve feature learning efficiency. Based on the simulations on a synthetic dataset of eight common radar and communication waveforms, the proposed method achieves the average recognition accuracy of 97.59% at 30 dB SNR and outperforms a Fuzzy support vector machine approach by approximately 17.36%.