Data augmentation of linear frequency modulation radar signal and active jamming signal based on CGAN

Radar signals are difficult to obtain due to their secrecy and complexity, which to some extent limits the development of deep learning in the field of radar. Conditional Generative Adversarial Networks (CGAN) have made great achievements in the field of image generation, but there are very few researches on the application of CGAN to radar signal augmentation. In this paper, a one-dimensional convolution-based CGAN is proposed for radar signal augmentation. We use MATLAB to simulate linear frequency modulation (LFM) signals and their corresponding jamming signals. Then, the proposed CGAN is used to augment this simulated radar signal dataset. The experimental results show that our augmented radar signal is similar with the simulated radar signal in terms of visual appearance and objective similarity metrics. The similarity of augmented signals with simulated signals is also verified by means of time-domain comparison, frequency-domain comparison, and pulse compression.