What if a mobile network could anticipate changing demand and continuously reorganise how it uses its spectrum as users move around?
This is the idea behind a new UNITY-6G demonstration developed by Four Dot Infinity (FDI) as part of the project’s work on AI-driven resource allocation. The solution brings together two levels of intelligence within the Open Radio Access Network (O-RAN): longer-term traffic prediction through an rApp and near-real-time spectrum optimisation through an xApp.
By allowing these two applications to work towards a common objective, the approach aims to make better use of limited spectrum while reducing interference and ensuring that network resources are shared more fairly among users.
The challenge: making O-RAN intelligence work together
O-RAN is making the Radio Access Network more open, programmable and flexible. One important part of this architecture is the introduction of intelligent controllers that can host applications optimising how the network operates.
But these applications do not all work at the same speed.
The Non-Real-Time RAN Intelligent Controller (Non-RT RIC) hosts rApps that can analyse historical network information, predict longer-term trends and establish policies. The Near-Real-Time RIC, meanwhile, hosts xApps that respond much more quickly to current radio conditions.
This creates an opportunity that is not yet fully exploited: using the broader and more predictive view available to an rApp to guide the rapid decisions made by an xApp. Existing xApp-based decisions often depend primarily on instantaneous radio conditions without taking advantage of longer-term traffic information available at the management level.
That becomes particularly relevant when macro and micro cells operate over the same spectrum. As users move and demand changes, neighbouring transmissions can increasingly interfere with one another. Avoiding that interference by simply restricting spectrum reuse, however, can leave valuable radio resources underused.
The challenge is therefore not simply to eliminate interference. It is to reuse spectrum as efficiently as possible while maintaining a fair level of service across users.
One network, two control loops
The UNITY-6G approach addresses this through a nested control system operating at two different timescales.
The first is a slower, predictive loop running through an rApp in the Non-RT RIC. Historical network measurements are collected and used to train a traffic prediction model offline, without affecting live network operation. During operation, the model then estimates how traffic is likely to evolve for each radio unit over the coming minutes.
These predictions help the rApp establish policies governing how spectrum should be managed. Among other parameters, these policies can specify user priorities, acceptable service tolerances, the allocation and fairness mechanisms to use, and how the available spectrum should be divided into Physical Resource Blocks, or PRBs.
Those policies are passed through the standard O-RAN A1 interface to the xApp running in the Near-RT RIC.
Here, the second and much faster control loop takes over. The xApp combines the policy guidance from the rApp with live information about users and radio conditions to continuously reconsider which users should receive which spectrum resources.
The distinction is important: the rApp anticipates and guides; the xApp reacts and optimises.
And because the information exchange uses standard O-RAN interfaces rather than proprietary connections between the two applications, the framework is designed around the openness and interoperability that O-RAN is intended to provide.
A mobile network as a moving, colourable graph
The demo provides a particularly intuitive way of visualising what happens next: the network becomes a moving graph.
Every active user is represented as a point, or vertex. If two users would create unacceptable interference by sharing the same spectrum resource, an edge is drawn between them.
Spectrum allocation can then be treated as a graph-colouring problem. Each colour represents a PRB, and connected users should not receive the same colour.
As users move, their radio conditions and interference relationships change. The graph therefore changes as well and must continually be recoloured.
This allows the same spectrum resources to be reused in different parts of the network when doing so does not create problematic interference — increasing the opportunity for efficient spatial reuse of limited spectrum.
There is, however, another problem to solve. Under difficult conditions, graph colouring alone may repeatedly leave some users without sufficient resources.
To address this, the xApp includes a Modified Proportional Fair (MPF) mechanism. It considers both what a user could achieve with a particular resource and how much service that user has received over time. Users that have consistently been underserved therefore receive greater priority.
In practical terms, the system is trying to achieve three things at once: avoid harmful interference, make efficient use of available spectrum and prevent the same users from continually losing out.
What do the results show?
The approach was evaluated in simulation using a heterogeneous network with one macro and two overlapping micro radio units, mobile users and varying service demands. The evaluation ran 96 rApp episodes, each containing 900 xApp scheduling episodes.
Two aspects were particularly important: how successfully resources could be assigned to satisfy users, and how fairly service was distributed over time.
Across the evaluated configurations, fairness remained above 85%, including the most demanding scenario shown in the demonstration.
The accompanying research also compared different resource-allocation and fairness strategies. Graph-based approaches achieved allocation success rates above 90% in the comparative scenario, while the combination of Welsh-Powell graph colouring and Modified Proportional Fair scheduling produced particularly strong results. In that comparison, the MPF approach achieved a Jain’s fairness service-share of 97.7%, compared with 61.6% without an explicit fairness mechanism.
These results point to an important benefit: greater spectrum efficiency does not necessarily have to come at the expense of fair access to network resources.
Why could this matter beyond the test environment?
The underlying problem is relevant wherever operators need to extract more capacity from finite spectrum while dealing with changing demand, mobility and interference.
Consider a dense urban network or transport hub, for example. Demand can shift rapidly as large numbers of people move through different coverage areas. A network capable of anticipating traffic trends while continually adjusting resource allocation could make better use of the spectrum already available rather than relying exclusively on static configurations or reactive interventions.
The same principle could be particularly useful for private and industrial networks. A factory may simultaneously connect workers, sensors, cameras, autonomous mobile robots and production equipment. Devices move, workloads change and applications can have very different service requirements. Policy-driven spectrum allocation provides a way of taking those requirements into account while dynamically managing interference between overlapping cells.
There is also a broader operational argument.
One of O-RAN’s strengths is that disaggregation and open interfaces make the network more programmable and allow specialised applications from different sources to optimise it. But greater programmability can also mean greater complexity. The accompanying research notes that this complexity and the resulting need for continuous parameter configuration remain concerns around O-RAN adoption.
Interoperable rApp/xApp control can help turn that programmability into something more manageable: longer-term intelligence establishes policies, near-real-time intelligence adapts them to current conditions, and the underlying network executes the resulting decisions.
The potential value is therefore not just more spectrum, but better use of the spectrum and infrastructure that operators already have.
Towards more anticipatory 6G network management
The demo illustrates a broader direction being explored within UNITY-6G: moving from network management that is predominantly reactive towards management that can combine prediction, policy and real-time adaptation.
Rather than placing all intelligence in one controller, the framework distributes different responsibilities according to the timescale at which they are most useful. Traffic prediction informs decisions over minutes; the xApp responds to changing radio conditions much faster; and feedback continuously closes the loop.
The work is still progressing. The results presented so far come from simulation, and the next step announced in the demonstration is to integrate and test the framework in an O-RAN-compliant UNITY-6G testbed.
Future research is also expected to explore learning-enhanced graph-colouring techniques, energy-aware optimisation that considers both spectrum and power consumption, and eventually cross-domain orchestration spanning radio, transport and cloud resources.
Together, these developments point towards a 6G network in which intelligence at different levels does not operate in isolation, but cooperates to make network resources more adaptive, efficient and responsive to changing service needs.
