Summary
- VodafoneThree will test Nokia's AI-native anyRAN software on Nvidia Aerial RAN computing infrastructure, beginning in a laboratory before a selected-site UK field trial.
- Initial technical work includes channel estimation and link adaptation, with performance to be benchmarked against the existing network.
- The trials will also explore shared computing resources for radio and AI workloads, but no commercial rollout or measured improvement has been announced.
VodafoneThree is preparing UK trials of artificial intelligence within its mobile radio network, working with Nokia and Nvidia to investigate whether machine learning can improve wireless capacity and reliability. The programme will test new radio processing methods and examine the possibility of running network functions alongside other AI applications on shared computing infrastructure.
The collaboration, confirmed on 18 November, will begin in a controlled laboratory environment before progressing to a selected UK field site. It will use Nokia’s AI-native anyRAN software running on Nvidia Aerial RAN computing technology, with Red Hat OpenShift supporting the proposed software architecture.
Initial testing will concentrate on channel estimation and link adaptation, two processes that influence how a mobile network responds to changing signal conditions. VodafoneThree intends to compare the new methods with its existing network, measuring performance in areas including capacity, reliability and spectral efficiency.
The announcement establishes a testing programme rather than a completed deployment. The companies have not released performance results from the planned UK trials or committed to a commercial rollout based on the proposed architecture.
AI could improve decisions inside the radio network
Mobile radio networks must manage constantly changing conditions as devices move, network demand fluctuates and signals encounter interference. Their software determines how available radio resources are used and adjusts transmission settings to maintain connections.
Conventional systems already employ sophisticated algorithms for these tasks, but machine learning introduces alternative ways to estimate network conditions and select suitable operating parameters.
Channel estimation involves assessing the characteristics of a wireless connection so that transmitted information can be recovered reliably. Its accuracy affects how effectively the network can use available radio capacity, particularly when signals are changing.
Link adaptation adjusts transmission settings according to the quality of the connection. Selecting appropriate parameters helps balance performance and reliability, although the network must make those decisions quickly enough to respond to actual conditions.
VodafoneThree will examine whether AI methods can improve these activities under realistic operating conditions. The trial results will need to account for the processing resources required by the new models as well as any changes in radio performance.
Using machine learning is not automatically more efficient than established signal processing. A model may produce more accurate estimates while requiring additional computing capacity, potentially changing the economics of the network even where technical performance improves.
The operator will therefore need to establish whether any capacity or reliability gains justify the hardware, electricity and operational costs associated with deployment.
Shared computing changes infrastructure requirements
The programme will also examine an architecture in which accelerated computing infrastructure supports both radio network processing and other AI workloads. Nvidia’s Aerial RAN technology is intended to provide computing resources capable of handling these different functions within a software-defined environment.
Traditionally, telecommunications operators have relied heavily on specialised infrastructure for radio processing. Sharing computing resources with other applications could create more flexible operating arrangements, but it also introduces technical requirements that differ from those of conventional cloud services.
Radio processing is time-sensitive. Delays can affect live communications, meaning computing resources must be available when network functions need them rather than simply allocated according to average demand.
Other AI applications may have more variable processing requirements, particularly where their workloads depend on user activity. Running both categories together requires effective resource management so that demand from one service does not undermine the performance of another.
The proposed combination of Nokia software, Nvidia hardware and Red Hat OpenShift will allow the partners to examine how such functions can be deployed and managed within a common platform.
Reliability is another consideration. Sharing infrastructure can reduce duplication, but a problem affecting common computing equipment may also have consequences for several services that would otherwise operate separately.
Before adopting such an architecture commercially, an operator would need evidence about resilience, isolation between workloads and the cost of maintaining sufficient spare capacity to handle failures or sudden demand.
Commercial services may benefit from greater control
VodafoneThree is considering the trials alongside its development of differentiated connectivity products, including Vodafone SuperMobile. The operator is examining whether more responsive management of radio resources could eventually support services with particular performance characteristics.
Those possibilities extend beyond increasing peak network speed. Some applications may depend on consistent connectivity, predictable latency or more effective access during periods when many customers are using the same radio infrastructure.
AI-based optimisation could potentially help allocate resources according to changing conditions, although the operator has not demonstrated that the proposed technology can deliver specific improvements to existing commercial services.
Radio performance also depends on factors outside the processing software. Coverage, device capabilities, spectrum holdings and the physical characteristics of network sites can influence customer experience even when the underlying algorithms perform effectively.
VodafoneThree will consequently need to evaluate the technology across different operating conditions, not solely during a controlled demonstration.
The proposed shared computing infrastructure could also support applications operating closer to users, commonly described as edge AI. Processing selected workloads within telecommunications infrastructure may reduce the distance information travels, although its value depends on the application and the resources available at the site.
Not every AI workload needs to run near a mobile network. Many services can operate effectively through centralised computing facilities, making the economics of locating additional processing capacity within the network an important consideration.
UK operators are testing competing approaches
VodafoneThree’s plans form part of broader experimentation with AI-enabled radio access networks. Other operators, including BT and Virgin Media O2, have announced or undertaken related testing with technology suppliers, while equipment manufacturers are developing different approaches to combining AI and telecommunications infrastructure.
That activity indicates substantial industry interest, but it does not establish a settled commercial architecture. Operators still need to compare performance, cost, energy consumption and the ability to integrate new technology with equipment already deployed.
Replacing radio processing infrastructure can be expensive, particularly where an operator has substantial investment in existing network hardware. Solutions that preserve useful equipment and introduce new functionality gradually may have different commercial advantages from approaches requiring extensive replacement.
VodafoneThree’s wider UK network development programme provides a practical setting for evaluating these alternatives. The operator has ongoing commitments to improve coverage and capacity, but the AI-RAN tests remain distinct from infrastructure upgrades already being delivered.
Following laboratory evaluation, the selected-site trial will provide evidence from a less controlled environment, where customer movement, changing radio conditions and operational constraints can affect results.
The programme’s significance will be determined by those measurements. Until the partners demonstrate consistent gains and establish a viable deployment model, AI-RAN remains a technology under evaluation rather than an established improvement to VodafoneThree’s live UK mobile services.












