Summary

• The €1.4 million 6G-ROCKET project is led by the University of Oulu and supported by Business Finland.
• Its research combines GPU-accelerated radio systems, digital twins, sensing technologies and automated network management.
• Prototypes and an AI-RAN Perception Lab are planned over two years, rather than an immediate commercial deployment.

The University of Oulu has launched a €1.4 million research programme to develop radio networks that use artificial intelligence to anticipate changing conditions, bringing together telecommunications equipment suppliers, semiconductor companies and researchers working towards future 6G infrastructure.

The two-year 6G-ROCKET project is supported by Business Finland’s 6G Bridge programme and coordinated by the university’s 6G Flagship research initiative.

Its industry participants include Nokia, NVIDIA, Finnish mobile operator DNA, Keysight and MediaTek, alongside academic partners Northeastern University in the United States and the University of Tokyo.

The programme will build on an existing GPU-accelerated radio access network testbed at Oulu, investigating how wireless communications, computing and environmental sensing can operate within a more closely integrated architecture.

The university plans to develop technical designs, prototype demonstrations and a new AI-RAN Perception Lab over the next two years. The programme does not represent the commercial launch of a 6G service or a completed deployment of the technologies under investigation.

Networks begin anticipating disruption

Mobile networks continually adjust their operation in response to changing traffic, radio conditions and the movement of connected devices. Existing systems already use automated processes and analytical tools, but their behaviour can still depend heavily on responding after a problem has become apparent.

6G-ROCKET is investigating whether AI processing incorporated into radio infrastructure can anticipate changing conditions early enough to maintain connectivity before performance deteriorates.

The project is particularly interested in environments where relatively brief interruptions can disrupt physical operations, including factories, warehouses and healthcare facilities.

A mobile robot moving through an industrial building, for example, may encounter changes in signal quality when equipment or other objects obstruct its wireless connection. If the network recognises the developing condition early enough, it could adjust resources before the interruption affects the application.

Achieving that outcome requires reliable information about both the radio environment and the behaviour of connected devices. Predictive software must also determine whether a proposed adjustment is likely to improve performance without creating problems elsewhere.

An inaccurate prediction could move resources away from users that still need them, making validation and safe operating limits essential to any practical implementation.

AI moves closer to base stations

Radio access networks provide the connection between wireless devices and the wider mobile network. Their traditional functions include processing radio signals, managing communications resources and supporting connections as devices move between coverage areas.

AI-native radio architectures seek to integrate additional computing capabilities into that environment, allowing selected AI workloads to operate close to the radio equipment rather than exclusively within remote cloud facilities.

NVIDIA is contributing AI computing infrastructure and Aerial software tools, while Nokia provides expertise in radio systems and the evolution of mobile network architecture.

MediaTek’s involvement addresses the device side of the connection. Its experience with chipsets and user equipment is relevant because a change intended to improve network operation must also function reliably with the devices using that network.

Keysight will contribute measurement and validation technologies, including capabilities associated with simulation and digital twins.

A digital twin can provide a simulated representation of a network or physical environment, allowing researchers to examine how particular changes might affect performance without immediately applying those changes to live equipment.

Its value depends on whether the simulation reproduces relevant operating conditions accurately. A model based on incomplete or outdated information may produce recommendations that perform poorly when transferred to physical infrastructure.

DNA provides an operator perspective, helping researchers connect technical experiments with the requirements of organisations responsible for maintaining commercial networks.

Those requirements include service continuity, equipment compatibility, maintenance costs and the ability to manage changes across infrastructure that may remain in operation for many years.

Physical intelligence creates additional demands

The research also includes multi-sensor perception, allowing radio infrastructure to contribute to an understanding of the physical environment surrounding connected devices.

Such systems could support industrial automation and robotics, where communications and sensing increasingly interact with equipment operating in changing surroundings.

However, processing more information at the network edge introduces additional requirements for computing hardware, electricity and operational management.

Operators need to determine whether the benefits of moving AI functions closer to connected devices justify the costs of installing and maintaining additional processing capacity.

That calculation may vary considerably between different locations. A factory using mobile robots could place a greater value on predictable communications delay than an ordinary consumer network serving applications that tolerate temporary variation in performance.

The ability to support industrial applications also depends on the complete communications system. Improving a radio function cannot compensate for unreliable application software, inadequate coverage or computing infrastructure that cannot meet the workload’s requirements.

AI-native network management introduces further governance questions. Automated decisions must operate within defined technical limits, and engineers need to understand when interventions have occurred and why the system selected a particular response.

Diagnostic information is particularly important when a network supports equipment whose interruption may affect physical processes. Operators must be able to distinguish an AI control problem from failures elsewhere in the system.

Interoperability will determine adoption

The participation of equipment manufacturers, a mobile operator and specialist technology companies reflects the difficulty of introducing new network capabilities into an established telecommunications industry.

Commercial mobile infrastructure uses standards intended to support interoperability across suppliers and generations of equipment. New computing arrangements must coexist with those systems if they are to be adopted beyond limited research environments.

An architecture that performs well using carefully configured equipment may be more difficult to operate across networks containing different hardware, software versions and traffic patterns.

Researchers will consequently need to measure performance under varied conditions and identify which elements of the architecture can be implemented through existing interfaces and which would require future standards or equipment changes.

The project also provides a setting for investigating whether AI workloads can share computing resources with radio functions without compromising the reliability of either activity.

Shared infrastructure may improve hardware utilisation, but competing processing demands could create problems if one workload consumes resources needed for time-sensitive communications.

Those trade-offs affect both the technical feasibility and the financial attractiveness of the proposed architecture.

Research results will precede deployment

The University of Oulu intends to establish an AI-RAN Perception Lab to support continuing collaboration and experimental testing.

The planned facility will provide an environment for evaluating prototypes, comparing approaches and developing technical knowledge relevant to future wireless networks.

Further project information is expected at the 6G Summit in Helsinki on 28 October, where researchers and technology suppliers will discuss the convergence of wireless communications, artificial intelligence and security.

The programme will continue through 2028, allowing its participants to examine technical performance before the capabilities are considered for wider commercial use.

Finland has an established research base in mobile communications, and the project extends that capability into the integration of radio systems and accelerated computing.

Its success will be determined by measurable improvements in network reliability, responsiveness and resource efficiency under realistic conditions, rather than by the number of AI components incorporated into a prototype.

The immediate outcome is a funded research collaboration with access to existing test infrastructure. Demonstrating whether predictive AI can improve mobile networks without creating excessive cost or operational complexity remains the work ahead.