#Qatar #UAE #telecom - Telecom operators can now test an AI model that reads radio signals and answers questions about them in plain language. Abu Dhabi research university Khalifa University of Science and Technology, mobile industry body GSM Association (GSMA) and US chipmaker Advanced Micro Devices (AMD) have launched an RFGPT operator trial at MWC Doha 2026. Operators get two weeks of API access to the radio frequency foundation model, powered by AMD compute. The goals are to enable test uses in network planning, spectrum management and interference analysis.
SO WHAT? - Radio frequency (RF) data normally needs specialist engineers to interpret it. First announced earlier this year, RFGPT is first radio-frequency language model that integrates RF spectrograms** into a multimodal model. The model allows operators to query RF signals in natural language, which could make network planning and troubleshooting both faster and cheaper. The trial tests this hypothesis, allowing the people who run networks to trial the breakthrough RFGPT model.
KEY POINTS:
Khalifa University’s Digital Future Institute has developed RFGPT, a radio frequency foundation model that combines wireless signal understanding with natural-language interaction. Users can ask about observed signals, identify their characteristics and interpret scenes containing several signals at once.
In collaboration with GSM Association (GSMA) and Advanced Micro Devices (AMD), the university is now making the RFGPT model available to telecom operators. Operators can request two weeks of API access by submitting a use case to the research team, who will then agree the evaluation scope, access arrangements and trial dates with each operator.
Running the RFGPT model on AMD compute, operators will be able to use it to explore use cases across network operations, planning and performance management, including radio access network (RAN) optimisation.
RFGPT identifies wireless technologies including Wi-Fi, Bluetooth, LTE, 5G New Radio (NR), UMTS and DVB-S2. The model also distinguishes uplink from downlink transmissions where that applies to the technology.
The model recognises modulation schemes and analyses wideband environments with several simultaneous signals. That makes it relevant to crowded spectrum, where operators need to know what is transmitting and how.
For coexistence analysis, RFGPT finds time-frequency overlap between transmissions and estimates the strength of interference or signal interaction. Operators could use this to support interference mitigation and spectrum management decisions.
RFGPT estimates wireless activity by analysing WLAN and NR user densities within a given observation window. This gives operators a view of network utilisation patterns that could feed into capacity planning.
The model interprets key 5G NR parameters, including subcarrier spacing, synchronisation signal block (SSB) patterns, channel state information reference signal (CSI-RS) configurations and sounding reference signal (SRS) structures. Engineers can ask about these in plain language.
Participating telecom operators will be asked to feed back on performance, usability and relevance to their own applications. GSMA says the results will show where the model adds value and what it needs to go further.
GSMA Foundry, the innovation arm of the GSMA, and Khalifa University began a strategic partnership in November 2025, aimed at producing open telecom AI assets and benchmarking tools.
The RFGPT trial also allows telecom operators to trial AMD’s AI compute infrastructure, which has become a core focus for the vendor. Last week AMD announced that data centre segment revenue rose 97% in its third quarter, driven by record sales of Instinct AI GPUs and EPYC CPUs. Overall, the company’s third-quarter 2026 revenue hit a record $13.4 billion, up 45% year on year.
ZOOM OUT - Khalifa University’s Digital Future Institute first announced RFGPT in February. According to researchers, it is the first radio-frequency language model to feed RF spectrograms into a multimodal model, allowing one system to process and reason over wireless signals in text. Prior to the research, telecoms AI relied on separate deep-learning models for each signal-processing task. The team trained RFGPT on about 625,000 synthetic instruction examples drawn from roughly 12,000 simulated scenes covering six wireless technologies. No manual labelling was needed. In benchmark tests, RFGPT beat OpenAI’s GPT-4 and Alibaba’s Qwen 2.5 on tasks including modulation classification and 5G NR information extraction.
[Written and edited with the assistance of AI]
Source: Khalifa University
LINK
Apply to participate in the RFGPT operator trial (Google Form)
Read more about telecom GPT and related research:
Khalifa University built AI model tops telco leaderboard (Middle East AI News)
Khalifa University unveils breakthrough RF AI model (Middle East AI News)
GSMA, KU to develop new TelecomGPT version (Middle East AI News)
GSMA releases telecom AI benchmarks ahead of MCW 2025 (Middle East AI News)
New leaderboard to support development of telecom LLMs (Middle East AI News)