Guided Generative Adversarial Network for Super Resolution of Imaging Radar

We propose a super-resolution algorithm for radar data that combines the ability to generate realistic results based on adversarial loss with the accurate estimation performance of subspace-type direction-of-arrival technique. The proposed algorithm forces the generator to completely incorporate information in multiple channels through a guide for the discriminator in the adversarial learning architecture, enabling reliable super resolution of radar data. In addition, the proposed algorithm can be generalized and applied to all dimensions of radar data, thereby contributing toward overcoming the physical limitations in obtaining 4D high-resolution radar data.