GAN-Based Data Augmentation for AI-Enabled ATP in Free Space Optical Communication

Free space optical communication (FSO) has been becoming one of the main directions of future communication network due to its advantages of high transmission rate, small divergence angle, and strong confidentiality. The acquisition, tracking, pointing (ATP) sub-system is an important guarantee for stable communication of FSO system. In the AI-enabled ATP system, data augmentation algorithms are effective to reduce the size of the dataset for spot prediction, and thus to lower the system structural complexity. In this letter, we propose a novel data augmentation algorithm that introduces wavelet transform into the fusing-and-filling generative adversarial network (F2GAN). The wavelet transform is utilized to convert the image information from time domain to frequency domain. The diversity of the generated images obtained from the GAN model is thus enhanced with the richer details. The experimental results show that, compared to the F2GAN algorithm, the proposed algorithm can improve peak signal to noise ratio by 14.5%, structure similarity index measure by 11.8%, and learned perceptual image patch similarity by 21%, respectively.