Multi-Task Wireless-Native Foundation Model for 6G

RIS CUTTING EDGE FOURM

面向6G的多任务无线原生基座模型

Multi-Task Wireless-Native Foundation Model for 6G

The wireless foundation model is pre-trained on large-scale datasets to learn a universal representation of wireless propagation environments, and is adapted to various communication and sensing downstream tasks through diverse task-specific heads, enabling a “one model, multiple uses” paradigm and forming an intelligent backbone of AI-native 6G networks. Although large language models (LLMs) have achieved strong performance in natural language processing, computer vision, and code generation, their architectures are not fully suitable for communication problems; therefore, it is necessary to integrate domain knowledge in communications into the Transformer architecture and develop a wireless-specific foundation model. This talk proposes a Channel Transformer design, which introduces an adaptive interference and noise suppression module to improve channel representation in open environments, redesigns the self-attention mechanism by combining physical correlations with feature similarity to better capture channel properties, develops a channel-consistency-driven lightweight computation scheme to reduce inference latency without sacrificing performance, and introduces a gating-based multi-task generalization mechanism to improve adaptability across tasks. Finally, the talk presents prototype validation results showing that the proposed model achieves superior performance in downstream tasks such as channel estimation and channel prediction.

Li Sun received the B.S. and Ph.D. degrees in information and communications engineering from Xi'an Jiaotong University, China, in 2006 and 2011, respectively. He is currently a Full Professor of Pengcheng National Laboratory (PCNL), China. From Oct. 2021 to Jun. 2025, he was with the Wireless Technology Lab, 2012 Labs, Huawei Technologies, as a Senior Expert and the Chief Scientist of Wireless PHY/MAC Layer Security. From Jan. 2012 to Sep. 2021, he was with Xi'an Jiaotong University as a professor and the Deputy Director of the Wireless Communications Institute (2019-2021). From 2019 to 2020, he was with the Department of Electrical Engineering and Computer Science, University of California at Irvine, as a Visiting Professor. His research interests include Physical Layer Security, Privacy-preserving Sensing, Wireless AI, and AI-native Edge. He has published over 170 papers and has more than 80 granted patents with 20+ adopted into standards or commercialized. He is the Solo Principle Investigator of the National Science and Technology Major Project on Wireless Foundation Models. He was listed among the World's Top 2% Scientists by Stanford University and Elsevier in 2020 and 2021, received the Best Paper Award of China Communications in 2017 and 2020, the Best Paper Award of the IEEE ICCCS in 2023, the Outstanding Scientific Paper Award of Shaanxi Province of China in 2016, the Outstanding Master Thesis Supervisor Award of Chinese Institute of Electronics in 2020, the First Price of the Teaching Achievement Award of Shaanxi Province of China in 2018, the Innovation Pioneer Award at Huawei in 2022 and 2023, and the IEEE COMMUNICATIONS LETTERS Exemplary Reviewers Certificate from IEEE ComSoC in 2014 and 2016. He is the Founding member and Online Content Chair of the IEEE ComSoC Large Generative AI Models in Telecom Emerging Technology Initiative (GenAINet ETI), and is serving or has served as the Associate Editor of IEEE Transactions on Dependable and Secure Computing, and the Guest Editors of IEEE WCM, IEEE Network, etc. He is a Senior Member of IEEE.

RIS CUTTING EDGE FOURM