VodafoneThree plans UK AI-RAN infrastructure trials

VodafoneThree plans UK AI-RAN infrastructure trials

VodafoneThree will test Nokia radio network software on Nvidia accelerated computing infrastructure, examining whether AI can improve mobile network performance and support shared computing resources for radio processing and other…

VodafoneThree plans UK AI-RAN infrastructure trials
Summary
  • VodafoneThree plans to test Nokia anyRAN software on Nvidia Aerial RAN computing infrastructure, initially in a laboratory and subsequently at a selected UK field site.
  • The trials will assess machine learning-based channel estimation, link adaptation, capacity, reliability and spectrum efficiency.
  • The proposed platform combines Nokia software, Nvidia accelerated computing and Red Hat OpenShift, but commercial deployment and measured performance gains remain unconfirmed.

VodafoneThree will test an AI-enabled radio access network using Nokia software and Nvidia accelerated computing, examining whether mobile network processing and other AI workloads can operate on shared infrastructure.

The programme, disclosed on 18 November, will begin in a controlled UK laboratory environment before progressing to a field trial at a selected site. VodafoneThree intends to compare the resulting performance with its existing network to assess potential changes in capacity, reliability and spectrum efficiency.

The planned system uses Nokia’s AI-native anyRAN software running on Nvidia’s Aerial RAN computing platform, with Red Hat OpenShift supporting the software environment. The architecture is intended to provide a common computing platform capable of accommodating radio network functions alongside selected AI applications.

The operator has not identified the field trial location, specified the number of computing systems involved or provided dates for the completion of either testing phase. No commercial rollout decision has been announced.

The initial work will concentrate on machine learning-based channel estimation and link adaptation, two functions associated with determining radio conditions and adjusting how information is transmitted between mobile devices and the network.

These are potential improvements rather than demonstrated results from the proposed installation. VodafoneThree will need measured evidence from the trials before establishing whether the architecture offers advantages over its existing equipment.

Radio functions move towards accelerated computing

A radio access network connects mobile devices to the wider telecommunications infrastructure through antennas, radio equipment, processing systems and transport connections. Its operation depends on the available spectrum, signal conditions, computing capacity and the movement of information between network components.

Radio processing includes digital signal operations and decisions about how transmission resources are allocated. Some functions must be completed within tightly defined time limits to maintain reliable communication.

That requirement distinguishes radio processing from many conventional data centre workloads, where small variations in processing time may have less immediate effect on service operation.

Nokia’s anyRAN software is intended to support radio functions on more flexible computing platforms. Nvidia’s Aerial architecture provides accelerated processing resources that can perform the relevant computations and potentially support additional applications.

VodafoneThree intends to assess whether sharing that hardware can provide sufficient performance for radio functions while making computing resources available for other work.

The first tests will examine how machine learning can improve the estimation of radio channel conditions. A wireless connection changes as devices move, obstacles affect signals and other transmissions create interference.

Channel estimation helps the network determine the characteristics of the connection. Link adaptation then adjusts transmission parameters according to those conditions, seeking to maintain reliable communication while using spectrum efficiently.

Machine learning could help these functions respond to more complex patterns, but it also introduces processing requirements and depends on the quality of the information used by the model.

The laboratory work will need to establish whether any improvements in radio performance outweigh the additional computing demands and whether the system can maintain predictable behaviour under different operating conditions.

Field testing will provide another level of evidence because actual networks encounter variations in user activity, propagation conditions, equipment loading and interference that are difficult to reproduce completely in a laboratory.

Shared computing introduces operational dependencies

Combining radio network processing with other AI applications creates potential advantages in hardware utilisation. Rather than maintaining entirely separate computing platforms for each function, an operator may be able to allocate resources according to workload requirements.

However, radio functions with strict timing requirements cannot necessarily share computing resources under the same conditions as applications that tolerate variable response times.

Resource allocation, workload isolation and scheduling therefore become important. A processing-intensive AI task should not be permitted to compromise the timely execution of functions needed to maintain mobile connections.

Red Hat OpenShift provides a software environment for deploying and managing containerised applications. Its inclusion supports the proposed programmable architecture, but telecom-specific timing, hardware access and resilience requirements still need to be addressed.

Equipment reliability also affects the design. A shared computing platform can consolidate hardware, but its failure may affect several functions that would otherwise operate on separate systems.

Operators must therefore assess how computing resources are protected, how workloads respond to faults and whether alternative capacity remains available when equipment is taken out of service.

Accelerated computing introduces additional electrical and thermal requirements. GPUs, network interfaces and supporting processors consume electricity and produce heat, while their suitability for distributed network locations depends on the available space, power supply and cooling infrastructure.

Using shared hardware does not automatically reduce energy consumption. The result depends on equipment utilisation, the electrical demand of the computing platform and whether replacing dedicated systems reduces the total supporting infrastructure required.

VodafoneThree has identified further potential applications involving energy efficiency, anomaly detection, mobility, coverage optimisation and quality-of-service assurance. These remain areas for exploration rather than capabilities validated through the newly announced tests.

The programme also examines whether AI applications could operate closer to users through computing infrastructure associated with the mobile network. Processing information nearer its source can reduce some network transport requirements, although overall application latency depends on routing, processing and the location of other services.

The operator’s parent company, Vodafone, has separately opened an AI-RAN testing hub in Málaga, Spain, to examine computing options for future radio network architectures. The UK tests form another part of that wider technical evaluation, rather than a confirmed decision to procure the technology for national deployment.

VodafoneThree plans to benchmark the Nokia and Nvidia platform against existing network performance before assessing wider adoption. Results from the laboratory and field trials will determine whether the proposed shared infrastructure can meet the performance, reliability and operating requirements of a commercial mobile network.


Stay updated with the latest insights and trends in the data centre industry by subscribing to our newsletter.

← Back

Thank you for your response. ✨