GAN Framework Synthesizes Robust Data for Satellite Internet

A new generative AI (GenAI) framework, based on Generative Adversarial Networks (GANs), has been developed to synthesize robust data for Low-Earth Orbit (LEO) satellite Internet observations. Current LEO datasets frequently suffer from missing information, complicating data augmentation and limiting the creation of representative datasets. The proposed framework addresses this by directly generating high-fidelity data from incomplete LEO network observations. Researchers designed specific block-wise and point-wise missing scenarios to accurately simulate real-world data loss in satellite networks. Evaluations using the WetLinks dataset showed the effectiveness of the GAN-based framework. The GT-GAN model consistently outperformed other GAN and VAE-based GenAI models, demonstrating the highest robustness. Even when 40% of the input data was missing, GT-GAN maintained its ability to capture the underlying data distribution, proving least affected by generalization issues. This work highlights promising directions for GenAI in satellite network data augmentation and research.