Improved Waveform Classification for Integrated Radar-Communication 6 G Systems Via Convolutional Neural Networks

To overcome the spectrum congestion problem in next-generation wireless networks, an integrated radar-communication system with spectrum sharing becomes a promising solution, wherein radar and communication signals can be discriminated by means of modulated waveforms. This letter presents an efficient radar-communication waveform classification method by taking advantage of the combination of smooth pseudo Wigner-Ville distribution-based time-frequency analysis and deep learning to achieve a good trade-off between complexity and accuracy. To this end, a high-performance convolutional network, namely the radar-communication waveform recognition network (RadComNet), is designed with multiple cutting-edge techniques and advanced structures, including depthwise convolution for complexity reduction and residual connection and multi-level attention mechanisms for learning efficiency enhancement. Relying on the simulation results acquired on a synthetic signal dataset of 12 radar and communication waveform types with the presence of channel impairments, our proposed method shows superiority over other classification approaches and deep models in terms of accuracy and complexity.