Generative AI for non-terrestrial networks: design, applications, and challenges

Abstract
Non-terrestrial networks (NTNs) have been identified as a fundamental component of future sixth generation (6 G) mobile networks, providing extended coverage to underserved and isolated regions and enhancing connectivity to fast mobility users (e.g., airplanes, vessels). Artificial intelligence (AI) techniques are set to revolutionize the way we orchestrate and optimize 6 G wireless networks and are especially appealing to anticipate and adapt to the complex NTN environments. In this context, generative AI (GAI) technologies have demonstrated remarkable potential in data generation and decision-making optimization, offering innovative solutions for dynamic and demand-driven network operations. This article investigates the role of GAI in 6 G-NTNs, emphasizing its transformative impact on resource allocation, beamforming, and security. We begin with the motivation and recent advances in GAI, exploring its applications in NTNs, including channel modeling, channel state information (CSI) estimation, intelligent network deployment, semantic communications, image processing, and enhanced network security and privacy. Subsequently, we discuss the fundamental challenges of GAI in NTNs and we propose a hybrid framework combining generative adversarial networks (GANs) and generative diffusion models (GDMs) to enable secure and efficient resource allocation and beamforming. Simulation results validate the effectiveness of the proposed framework.