| Title |
Transformer-based Learning for Improved Localization Performance in UAV-RIS-supported Non-terrestrial Networks |
| Authors |
신승석(Seungseok Sin) ; 문상미(Sangmi Moon) ; 황인태(Intae Hwang) |
| DOI |
https://doi.org/10.5573/ieie.2025.62.9.3 |
| Keywords |
Deep learning; Line of Sight; Localization; Non-terrestrial network; Reconfigurable intelligent surface; Transformer; Unmanned aerial vehicle |
| Abstract |
Accurate and ubiquitous localization is a key requirement for the reliable operation of next-generation wireless communication systems. In particular, leveraging millimeter-wave (mmWave) signals in 5G and beyond communication systems enables precise positioning even under limited infrastructure. However, achieving centimeter-level localization accuracy requires a guaranteed line-of-sight (LOS) path, which forms the geometric basis between the channel parameters and the user position.To address this, we introduce solutions that improve LOS availability, such as utilizing non-terrestrial networks (NTNs), where high-altitude platforms act as base stations or user equipment, and employing reconfigurable intelligent surfaces (RISs) to control signal propagation. We further propose a deep learning-based refinement method that enables accurate localization even in scenarios with a single base station. The proposed method employs a Transformer-based learning network to refine initial location estimates derived from geometric approaches, thereby enhancing localization accuracy. Simulation results demonstrate that the proposed scheme achieves sub-meter accuracy for over 90% of users, confirming its effectiveness. |