LLM4PG: Adapting Large Language Model for Pathloss Map Generation via Synesthesia of Machines
IEEE Transactions on Machine Learning in Communications and Networking · 2026 · 4 · 864-881
DOI: 10.1109/tmlcn.2026.3693982 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Machine Learning in Communications and Networking, 2026. Verfasst von Mingran Sun und 3 weiteren Autor:innen. Die Arbeit nennt 33 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, New Cornerstone Science Foundation through the XPLORER PRIZE, Taishan Scholars Program.
Abstract (Originalsprache)
In this paper, a novel large language model (LLM)-based pathloss map generation model (LLM4PG) is proposed for sixth-generation (6G) artificial intelligence (AI)-native communication systems via Synesthesia of Machines (SoM). To explore the mapping mechanism between sensing images and pathloss maps, a new synthetic intelligent multi-modal sensing-communication dataset for SoM in uncrewed aerial vehicle (UAV)-to-ground (U2G) scenarios, named SynthSoM-U2G, is constructed, including multiple U2G scenarios with multiple frequency bands and multiple flight altitudes. By adapting the LLM to the cross-modal pathloss map generation for the first time, a novel framework that enables effective cross-domain alignment between the multi-modal sensing-communication domain and the natural language domain is introduced. Furthermore, a task-specific adaptation of the LLM is achieved through fine-tuning, with a properly designed layer selection and activation scheme tailored to the unique demands of massive-scale and high-quality pathloss map generation. Compared with the conventional artificial intelligence generated content (AIGC) models, the proposed LLM4PG enables accurate pathloss map generation and demonstrates strong generalization across various scenarios, frequency bands, and flight altitudes under three-dimensional (3D) high-mobility U2G scenarios. The accuracy and generality of the proposed LLM4PG are validated by comparing simulation results and ray-tracing (RT)-based results. Simulation results demonstrate that the proposed LLM4PG can achieve accurate pathloss map generation with a normalized mean squared error (NMSE) of 0.0428, outperforming the conventional AIGC model by more than 3.16 dB. The generality of the proposed LLM4PG across different conditions achieves an NMSE of 0.0492, outperforming the conventional AIGC model by more than 4.52 dB.
Förderung
- National Natural Science Foundation of China
- New Cornerstone Science Foundation through the XPLORER PRIZE
- Taishan Scholars Program
- Young Elite Scientists Sponsorship Program by China Association for Science and Technology
- Shandong Natural Science Foundation
- Open Research Fund of National Mobile Communications Research Laboratory, Southeast University
Zitieren
APA
Sun, M., Bai, L., Cheng, X., & Wu, J. (2026). LLM4PG: Adapting Large Language Model for Pathloss Map Generation via Synesthesia of Machines. IEEE Transactions on Machine Learning in Communications and Networking, 4, 864-881. https://doi.org/10.1109/tmlcn.2026.3693982
Vancouver
Sun M, Bai L, Cheng X, Wu J. LLM4PG: Adapting Large Language Model for Pathloss Map Generation via Synesthesia of Machines. IEEE Transactions on Machine Learning in Communications and Networking. 2026;4:864-881. doi:10.1109/tmlcn.2026.3693982
BibTeX
@article{Sun2026,
author = {Sun, Mingran and Bai, Lu and Cheng, Xiang and Wu, Jianjun},
title = {LLM4PG: Adapting Large Language Model for Pathloss Map Generation via Synesthesia of Machines},
journal = {IEEE Transactions on Machine Learning in Communications and Networking},
year = {2026},
volume = {4},
pages = {864--881},
doi = {10.1109/tmlcn.2026.3693982}
}
EUVIMED