Actualité
Vodafone tests an AI twin to predict 5G network performance before upgrades
Vodafone and Cirrus360 say their joint trials let engineers model a 5G radio site, test configurations and spot potential bottlenecks before hardware or software changes reach the network.
Vodafone says it has successfully tested an AI-driven predictive digital twin that allows engineers to assess how a mobile mast could perform before new hardware is installed or a network upgrade is deployed at scale. The work, carried out with US technology company Cirrus360, targets one of the most difficult parts of modern telecoms: predicting how complex radio equipment, software and customer traffic will interact in the field.
The announcement concerns a trial rather than a consumer-facing product launch. Its significance lies in moving more network testing into a virtual environment, where engineers can explore possible outcomes before committing to physical changes. That could help operators reduce deployment risk, shorten laboratory work and make better use of existing sites as demand for mobile data continues to grow.
A virtual model of a live 5G site
Vodafone describes the system as a dynamic virtual model of a physical 5G radio access network site. Cirrus360’s Declarative Digital Twin, powered by its Gabriel platform, is designed to represent the behaviour of the compute resources inside a mobile mast and to simulate how those resources may respond to future workloads.
The model can be used to forecast potential hardware or system problems, including memory pressure, cache limits and central processing unit usage. It can also estimate customer traffic and likely latency before an operator introduces an upgrade or deploys a new radio configuration. Instead of discovering a bottleneck only after equipment has been installed, engineers can investigate the likely impact earlier in the planning process.
This is particularly relevant to Open RAN deployments, where radio networks can combine hardware and software from multiple suppliers. That flexibility can encourage innovation, but it also increases the number of interactions that operators need to validate. A virtual model gives teams a way to examine those interactions without relying entirely on repeated physical installations.
Testing network changes before they reach customers
Vodafone says engineers typically need to test between six and eight different RAN configurations before installing or upgrading a mobile network across a country. Each configuration can involve different combinations of radio equipment, processing resources, software and traffic assumptions. Testing those options in a lab is valuable, but it can also consume time and leave gaps between controlled conditions and real network behaviour.
The companies say their joint tests in the UK and the United States are intended to make that process more predictive. Rather than relying on a rigid digital model that must be rebuilt for every change, the declarative approach uses high-level rules describing what the system should do. The model can then adapt as the physical asset moves from design to testing and eventually to operation.
That distinction matters because a mobile site does not remain static. Traffic patterns change, software is updated and radio hardware can be reconfigured as coverage requirements evolve. A model that can follow those stages may help engineers maintain a more consistent view of performance throughout the asset’s lifecycle.
What the trial does — and does not — change
Vodafone is not announcing a nationwide customer rollout or claiming that the digital twin has already improved everyday speeds. The company says it is considering how to extend the collaboration to additional vendors and make greater use of real-world test environments in Málaga, Spain, and Newbury in the UK.
For customers, any future benefit would therefore be indirect. If the approach moves from trials into operational deployments, operators could identify capacity constraints earlier, select more suitable configurations and reduce the chance that a network change creates unexpected performance problems. That could support more stable connectivity for activities such as video calls, live streaming, cloud gaming and other latency-sensitive applications.
The announcement also highlights a practical role for AI in telecoms that is different from the assistants found on smartphones. Here, AI is being used to reason about infrastructure, simulate possible states and help engineers compare choices. It does not replace the physical network, and it does not remove the need for testing, monitoring or human approval before a change is introduced.
That last point is important. A prediction is only useful when it is checked against real measurements and updated as conditions change. Vodafone and Cirrus360 have described the technology and the scope of their trials, but they have not published independent performance figures showing how much faster or more accurate deployment becomes in a live commercial network.
The broader direction is clear: as 5G networks become more software-driven and support more demanding mobile applications, operators need better tools for understanding the consequences of change. Vodafone’s work with Cirrus360 suggests that predictive digital twins could become part of that toolkit, provided future deployments validate the models against real-world network data and demonstrate measurable benefits for reliability, capacity or energy use.