Professor Shugong Xu and his team chart path to 6G Native AI communications

A research team at Xi’an Jiaotong-Liverpool University (XJTLU) has drawn a roadmap to realising intelligent 6G communications, publishing a detailed framework for its Channel Foundation Model (CFM).

Unlike 5G, which relies on predefined rules and engineering optimisation, 6G needs stronger environmental awareness, task adaptation, and autonomous optimisation. However, conventional supervised-learning AI models will struggle here due to their dependence on manually labelled data, leading to poor generalisation across scenarios and tasks.

Introduction to the GREAT-X simulation platform

To address this, Professor Shugong Xu and his team at XJTLU’s School of Advanced Technology first proposed the CFM concept in 2025. Now, Professor Xu and co-author Jun Jiang, a PhD student, have laid out a technical pathway in a paper published in the ZTE Technology Journal.

CFM treats the wireless channel as the core object of study, using a “pre-training and fine-tuning” paradigm: models learn general-purpose channel representations from large, diverse datasets, then adapt to downstream tasks.

Inputs can include traditional channel data, such as channel state information (CSI) and channel impulse response, and multimodal information like images and location data, enabling richer understanding of physical obstructions, multipath reflections, and other environmental conditions.

Plaque presentation ceremony for the Suzhou Low-Altitude Economy Technology Innovation Centre

At the model level, using pre-training to extract generalisable features from large datasets reduces the need for manually labelled data. And unlike conventional wireless AI models, typically designed for a single task, CFM aims to provide a unified channel representation framework that supports multiple tasks and enables more efficient transfer, generalisation, and adaptation across scenarios.

“This shift from AI-assisted communications to ‘AI native to the channel’ will help communication systems move beyond passively adapting to their environment to actively understanding and learning from it and optimising themselves accordingly,” Professor Xu explains.

Unveiling of the Low-Altitude Technology Joint Transformation Centre

From simulation to reality

The team is also moving CFM towards real-world use, drawing on its strengths in industrial and academic collaborations to advance technical validation and technology transfer.

Through the XJTLU Low-Altitude Innovation Research Institute, researchers at the University are working closely with companies including Suzhou Aviation Industry Group and Suzhou Low-Altitude Tech, and leading research institutions like the Gusu Lab and Suzhou Institute of Nano-tech and Nano-bionics, Chinese Academy of Sciences.

Professor Shugong Xu delivers a talk on Native AI and CFM via the Teaspark platform

Together, they have established the Low-Altitude Technology Joint Transformation Centre, which focuses on core 6G areas, and are providing a strong industrial foundation for deploying CFM in low-altitude scenarios.

The team has also autonomously developed the GREAT-X platform to simulate and test CFM across complex environments including dense urban areas, open suburban landscapes, and low-altitude flight scenarios.

Meanwhile, two related projects, CSI-CLIP and CSI-MAE, have been open-sourced to support global research in intelligent 6G communications, as the team works towards integrating CFM into international standards.

Text and images courtesy of XJTLU Low-Altitude Innovation Research Institute and Research Engagement and Innovation Office

 Translated by Xiangyin Han

Edited by staff editor and Xinmin Han