Learn how 6G-Cloud validates service-based RAN composition, flexible control-fabric operation and AI/ML-assisted workflows across the BubbleRAN, VTT and BULL/AVANTIX testbeds.
From service-based RAN architecture to flexible control-fabric validation
This tutorial presents PoC-2, the second 6G-Cloud Proof of Concept, focused on service-oriented RAN composition with flexible control fabric.
PoC-2 demonstrates how Service-Based RAN functions, control applications, orchestration mechanisms, and AI/ML-assisted workflows can be deployed, interconnected, and validated in realistic testbed environments.
While PoC-1 focuses on the broader integration of the main 6G-Cloud frameworks, PoC-2 concentrates on the SBA-RAN side of the architecture. It explores how RAN Network Functions communicate through service-based interfaces, how RAN data is exchanged through message-bus-driven interactions, and how intelligent control applications can support network optimisation and runtime adaptation.
PoC-2 focuses on four main building blocks:
- the SBA-RAN flexible control fabric;
- MOF-aligned management and orchestration support;
- the AIMLF interaction layer;
- 6G-Cloud NwApps for both network exposure and intelligent RAN control.
Together, these building blocks support the validation of service-oriented, cloud-native and AI-assisted RAN control mechanisms for future 6G networks.
What you will learn
By the end of this tutorial, you will understand:
- how PoC-2 validates service-oriented RAN composition within the 6G-Cloud architecture;
- how the SBA-RAN flexible control fabric supports modular RAN control interactions;
- how RAN Network Functions communicate through a RAN Service Bus;
- how service discovery and message schema validation are supported through the RAN NRF;
- how RAN data can be collected, structured and consumed by other RAN functions and AI/ML workflows;
- how AIMLF-supported workflows contribute to training, inference and intelligent control applications;
- how interference detection and classification can support management-plane response.
Why PoC-2 matters
The 6G-Cloud project proposes a service-oriented, cloud-native and AI-driven architecture for future 6G networks. PoC-2 plays a key role in validating how these principles can be applied to the Radio Access Network domain.
Rather than relying on tightly coupled RAN control functions, PoC-2 explores a more modular and service-based approach. RAN functions, monitoring services, optimisation functions, and AI/ML components can interact through structured interfaces, publish/subscribe mechanisms, and service discovery procedures.
This makes it possible to compose RAN services more flexibly, process operational data more efficiently, and integrate AI-driven intelligence into control and management workflows.
The PoC-2 architecture in brief
PoC-2 is implemented across complementary testbed environments involving BubbleRAN, VTT, and BULL/AVANTIX.
At BubbleRAN, PoC-2 focuses on the SBA-RAN control fabric. The testbed supports cloud-native RAN and Core functions, service composition, lifecycle management, control-application deployment and runtime adaptation. The environment used for the PoC-2 demonstrations combines two NSs (each with different PLMN), one running an OAI RAN with the OAI core and one running an OAI RAN with Open5GS, a near-real-time RIC. For the Over-the-Air (OtA) environment, USRP B210 software-defined radios and Quectel RM520N-GL user equipment (controlled from containers within cloud environment) were used. This setup allows the same control fabric to be exercised over the air and under controlled interference conditions.
At VTT, the PoC-2 develops a service-based architecture for cloud-native RAN control, built around a RAN Message Bus implemented with Apache Kafka (deployed on Kubernetes cluster). The bus enables all inter-function interactions both publish/subscribe and request/response, so that RAN Network Functions (RAN NFs) interact through the bus, in a microservice style, instead of through point-to-point interfaces, this decoupling is what enables network-function composition. A RAN Network Repository Function is developed to let each RAN NF register and be discovered by others and serves the message schemas that NFs use to validate their data structure. The PoC demonstrates two operations: RAN data processing via data ingestion, schema-validated, and distribution to multiple independent consumers and RAN service composition workflows via a closed control loop spanning several RAN NFs.
The BULL/AVANTIX environment contributes to the AIMLF-supported and over-the-air validation aspects of PoC-2. By focusing on the interference detection problem, it supports use case characterisation and dataset generation, model training, and the integration of 6G-Cloud NwApps within the 6G-Cloud SBA-RAN domain.
Together, these environments show how service-based RAN composition, AI/ML-assisted workflows, and flexible control-fabric mechanisms can be instantiated and validated across different implementation contexts.
Main components
SBA-RAN Flexible Control Fabric
SBA-RANThe SBA-RAN flexible control fabric supports modular and service-oriented RAN control interactions. It enables RAN functions and applications to communicate through service-based mechanisms and supports service discovery, inter-function communication, and runtime control actions.
RAN Message Bus and RAN NRF
Message Bus + NRFThe RAN Service Bus is the communication backbone of the service-based RAN environment. It enables RAN Network Functions to publish, subscribe to and exchange operational data over named topics. Messages are schema-validated, and topics are partitioned so independent consumers can read the same stream at their own pace.
The RAN NRF is the environment's registry supporting NF registration, service discovery and holds the message schema for each topic. RAN NFs validate their data against these schemas before publishing, helping ensure the exchanged data follows the expected format.
MOF-aligned Orchestration and Management Support
MOF-alignedThe MOF-aligned orchestration and management layer supports deployment, configuration, update, and runtime orchestration procedures.
In the BubbleRAN implementation, this includes functions such as the Network Service Director, the Control, Coordination and Management Function (CCMF) and the NF Manager, which support service composition, control-application coordination and management-plane responses.
AIMLF and Intelligent NwApps
AIMLF + NwAppsThe AIMLF provides AI/ML lifecycle support, including model training, inference, performance monitoring, and model update workflows.
In PoC-2, AIMLF-supported workflows interact with the RAN control environment through data processing, inference-based optimisation and an interference detection and classification of NwApp.
Video modules
Track A - Interference management and AI-assisted NwApp workflow
This module introduces an integrated PoC-2 demonstration showing how the 6G-Cloud architecture supports service-oriented RAN composition and continuous data exposure. It presents the workflow towards the AIMLF, where NwApps are trained through MLOps pipelines. A key contribution is an ML model for real-time interference detection and classification, deployed as a 6G-Cloud NwApp and evaluated in a live network scenario with inter-cell co-channel interference.
The demonstration highlights how continuous RAN data exposure and AI-assisted analysis enable automated network monitoring and pave the way for intelligent interference mitigation within the 6G-Cloud architecture.
This module presents the BubbleRAN/BULL/AVANTIX interference-management workflow.
The demonstration shows how RAN metrics are collected, selectively exposed towards the AIMLF and used by an interference detection and classification NwApp. When a management-level response is required, the NF Manager consolidates the relevant context and escalates it to the management plane. The CCMF then interprets the escalated context in the scope of the affected service and applies the resulting management-plane reconfiguration, which in this demonstration is an ARFCN update of the main service domain.
The workflow shows how AI-assisted analysis can support adaptive network management within the 6G-Cloud SBA-RAN architecture.
Track B - SBA-RAN, Service Bus and service composition
This module introduces the VTT implementation of the service-based RAN architecture, including RAN Network Functions, the RAN Service Bus, the RAN NRF, the AIMLF and the service-based interfaces used to support communication across the system.
This module explains how RAN data is processed within the VTT implementation.
The demonstration shows how data generated by gNBs is ingested, structured and published through the RAN Service Bus. It also illustrates how topic-based routing and schema-aware communication allow different RAN Network Functions to consume the data they need.
This module presents the RAN service composition procedure implemented in the VTT testbed.
The demonstration shows how independent RAN Network Functions collaborate through event-driven communication, from monitoring data publication to AI-assisted inference, optimisation decisions and delivery of network configuration changes.
Step-by-step walkthrough
The RAN SBA environment is deployed
The PoC-2 environment begins with the deployment of the main RAN SBA components in the selected testbed, including RAN Network Functions, control applications, orchestration components, monitoring services, the RAN Service Bus, the RAN NRF and AIMLF-related components.
RAN Network Functions register and discover services
RAN Network Functions register their available services and retrieve information about other functions, topics and schemas. This enables modular interaction between producers, consumers and control applications.
RAN data is generated and collected
gNBs generate operational data, including counters, metrics and monitoring information. This data supports monitoring, optimisation, AI/ML workflows and control-loop decisions.
Data is published through the RAN Service Bus
The collected RAN data is structured and published to the RAN Message Bus. Message schemas help ensure that the data can be consumed consistently by downstream functions.
RAN Network Functions and AIMLF consume relevant data
Different RAN NFs subscribe to the topics that are relevant to their role. AIMLF-related components can also consume selected data streams to support feature engineering, model training, inference and analytics workflows.
AI/ML or control applications generate outputsF
The AIMLF or intelligent NwApps process the available data and generate outputs such as predictions, classifications or recommendations.
The output is routed to the relevant control or management function
Depending on the workflow, the output is consumed by the relevant control or management component. This may involve an optimisation function, the NF Manager or the management plane.
The network action is applied
The required action is applied through the appropriate control or management path. This closes the loop from RAN data collection to AI-assisted decision-making and adaptive network response.
Validation focus
PoC-2 supports the validation of the 6G-Cloud architecture by demonstrating that service-oriented RAN composition, flexible control-fabric mechanisms and AI/ML-assisted workflows can be deployed, interconnected and operated across different testbed environments.
The tutorial focuses on functional validation, including:
- deployment and integration of service-oriented RAN functions and control applications;
- operation of the flexible control fabric in emulated and over-the-air conditions;
- RAN Service Bus communication between RAN Network Functions;
- service discovery and message schema validation through the RAN NRF;
- RAN data processing and topic-based routing;
- RAN service composition through event-driven workflows;
- interaction between SBA-RAN mechanisms and AIMLF-supported analytics workflows;
- integration of intelligent NwApps, including interference detection and classification;
- preparation for future KPI evaluation.
The final consolidated KPI measurement results will be reported in later project validation activities. PoC-2 establishes the technical and functional foundation for those evaluations by confirming that the required components, workflows and interactions are operational.
Key takeaways
- PoC-2 shows how service-based RAN composition can be implemented and validated within the 6G-Cloud architecture.
- It demonstrates how RAN Network Functions, the RAN Service Bus, the RAN NRF, orchestration components and AIMLF-supported workflows can interact in practical testbed environments.
- It provides a foundation for validating future 6G capabilities such as flexible RAN control, AI-assisted optimisation, efficient data processing, service composition and adaptive network management.
Glossary
| Term | Meaning |
| CCMF | Control, Coordination and Management Function |
| SBA-RAN | Service-Based Radio Access Network |
| RAN | Radio Access Network |
| NF | Network Function |
| NS | Network Service |
| NwApp | Network Application |
| xApp | Near-real-time RAN Intelligent Controller application |
| MOF | Management and Orchestration Framework |
| NF Manager | Network Function Manager |
| AIMLF | AI/ML Framework |
| RAN NRF | RAN Network Repository Function |
| gNB | 5G base station |
| PLMN | Public Land Mobile Network |
Related resources
- Deliverable D5.3: Service-oriented 6G composition PoC implementation (soon)
