RaComNet: High-Performance Deep Network for Waveform Recognition in Coexistence Radar-Communication Systems

In this paper, an efficient deep learning-based waveform recognition method is introduced for coexistence radar-communication systems in the presence of channel impairments. The method first leverages smooth pseudo Wigner-Ville distribution (SPWVD) to analyze signals in the time-frequency domain, which in turn provides a high-resolution visual representation of a signal with alleviation of cross-term interference. To effectively learn waveform patterns from a noisy-confusing dataset of transformed time-frequency images, we design a convolutional neural network (CNN), namely radar-communication waveform recognition network (RaComNet), which has several processing modules in cascade to extract representative features automatically. Each module is competently designed by incorporating residual connection and attention connection in a sophisticated structure to attain the following multifold advantages: feature diversity, gradient preservation, and feature refinement (i.e., strengthen relevant features and weaken irrelevant features), thus enhancing learning efficiency. Simulation results show that RaComNet is robust under impaired channel conditions and outperforms other existing CNN-based approaches in terms of accuracy.