Abu dhabi: Khalifa University of Science and Technology's Digital Future Institute announced the launch of 'RF-GPT', a groundbreaking radio-frequency AI language model capable of interpreting wireless signals. This development addresses a significant challenge in telecommunications AI, where language models have primarily focused on text and structured network data.
According to Emirates News Agency, RF-GPT has demonstrated significant performance improvements in radio-frequency spectrogram tasks, surpassing existing baseline models by up to 75.4 percent. The model's ability to accurately count the number of signals in a spectrogram nearly 98 percent of the time is a rare achievement for general-purpose AI models.
RF-GPT functions by converting radio signals into visual patterns interpretable by AI systems. These systems then analyze the patterns and respond to queries about activity within the wireless spectrum using plain language. The model supports the UAE National Artificial Intelligence Strategy, contributing to the development of more autonomous and intelligent wireless networks.
The project was spearheaded by Khalifa University researchers under the leadership of Professor Merouane Debbah, with contributions from postdoctoral fellows Hang Zou and Yu Tian, research scientists Dr. Lina Bariah, Dr. Samson Lasaulce, and Dr. Chongwen Huang, along with PhD student Bohao Wang from Zhejiang University.
Professor Ahmed Al Durrah, Associate Provost for Research at Khalifa University, highlighted the model's alignment with national priorities, focusing on innovation in digital infrastructure and the advancement of AI integration. This initiative is crucial for the UAE's growing human capital and research capabilities, which are essential for supporting the country's evolving digital ecosystem.
Professor Merouane Debbah emphasized that RF-GPT represents a shift towards a unified RF-language interface, enhancing spectrum intelligence. By enabling AI-native radio systems, the model allows RF perception to directly support network optimization and policy decisions, paving the way for future AI-native 6G networks.
Trained with approximately 625,000 computer-generated radio signal examples, RF-GPT is designed for telecommunications operators, network engineering teams, and spectrum authorities. The model excels in identifying signal types, detecting overlapping transmissions, recognizing wireless standards, estimating device usage in Wi-Fi networks, and extracting data from 5G signals.