Khalifa University Unveils RF-GPT for Smarter Networks

Khalifa University of Science and Technology has unveiled a groundbreaking artificial intelligence model, RF-GPT, marking a major leap forward in telecommunications technology. The newly launched model is designed to interpret radio-frequency (RF) signals—an area where traditional AI systems have long faced limitations.

Unlike conventional language models that rely on text and structured data, RF-GPT introduces a novel approach by converting wireless signals into visual spectrogram patterns. These patterns are then analyzed by the AI, enabling it to understand and respond to queries about wireless spectrum activity using natural language.

Performance Breakthrough in RF Intelligence

RF-GPT has demonstrated significant improvements in performance, outperforming existing baseline models by up to 75.4% in radio-frequency spectrogram tasks. Notably, the model achieved nearly 98% accuracy in counting signals within a spectrogram—an advanced capability rarely seen in general-purpose AI systems.

Advancing Future Wireless Networks

The model directly aligns with the UAE’s national AI ambitions and supports the development of more autonomous and intelligent wireless systems. By making the electromagnetic spectrum accessible through natural language queries, RF-GPT lays the foundation for AI-driven network optimization and policy-making—critical for the evolution of future 6G networks.

Research Leadership and Collaboration

The project was led by Merouane Debbah, Senior Director of the Digital Future Institute, along with a global team of researchers and scientists. Key contributors include Hang Zou, Yu Tian, Dr. Lina Bariah, Dr. Samson Lasaulce, Dr. Chongwen Huang, and PhD researcher Bohao Wang.

According to Ahmed Al Durrah, Associate Provost for Research, the launch reflects Khalifa University’s long-term commitment to advancing digital infrastructure and strengthening AI integration across strategic sectors.

A Step Toward AI-Native Connectivity

RF-GPT was trained on approximately 625,000 simulated radio signal datasets, enabling it to handle complex wireless environments. The model excels in tasks such as identifying signal types, detecting overlapping transmissions, recognizing wireless standards, estimating Wi-Fi device usage, and extracting insights from 5G signals.

Experts believe RF-GPT represents a turning point in spectrum intelligence, shifting from isolated RF analysis tools to a unified AI-powered interface. This innovation is expected to play a key role in shaping next-generation, AI-native communication systems.