GAN-Based Data Synthesis for LEO Satellite Internet Observations
A new framework using generative AI, specifically GANs and VAEs, has been proposed to synthesize high-fidelity data from incomplete Low-Earth orbit (LEO) satellite Internet observations. The research addresses the challenge of missing data in LEO network datasets, which limits data augmentation and representative dataset expansion. The framework evaluates performance on the WetLinks dataset using block-wise and point-wise missing scenarios to simulate real-world conditions. This approach aligns with the International Telecommunications Union's vision for 6G telecommunications networks, aiming to enhance ubiquitous connectivity. The study highlights the underexplored application of GenAI in this domain and demonstrates promising results for robust data synthesis.
Key facts
- Proposed a GenAI-based framework for synthesizing data from incomplete LEO satellite Internet observations.
- Evaluated GAN- and VAE-based models on the WetLinks dataset.
- Designed block-wise and point-wise missing scenarios.
- Aims to address missing data in LEO network observations.
- Aligns with ITU vision for 6G networks.
- Published on arXiv with ID 2607.24790.
- Focuses on data augmentation for satellite Internet.
- GenAI application in this domain has received little attention.
Entities
Institutions
- arXiv
- International Telecommunications Union