Intelligent Open RAN Control: From Vision to Reality (Reader Forum)

Among the most famous principles of the Open RAN paradigm are softwareization and disaggregation, which are central to discussions about the commercialization and deployment of O-RAN systems. A less commonly discussed principle that these two elements enable is programmability – the ability to dynamically and algorithmically adjust the configuration and behavior of the RAN to optimize performance for specific conditions and use cases. With programmability and open interfaces comes closed-loop control, allowing the use of telemetry and data from the RAN itself (and possibly other sources) to automatically assess the health of the RAN, align it with the optimal action or configuration, and Apply it within the programmability RAN system.

Programmability and closed-loop control enable, among other things, energy optimization within the RAN, support of different traffic profiles with conflicting requirements through dynamic slicing, and dynamic load balancing of users across the network. These capabilities are achieved through a combination of artificial intelligence (AI) and machine learning (ML) techniques that use data streams and telemetry provided by the RAN for control, classification and prediction. This architecture also facilitates the integration of different views of the RAN, ranging from highly granular, device-specific views of individual base stations to more generalized, centralized views that aggregate data from dozens of base stations and hundreds of users.

Programmability and control therefore have the potential to redefine the way the RAN is managed and optimized. These principles are anchored in the O-RAN RAN Intelligent Controllers (RICs), which enable control and analysis at non- and near real-time scales. The non-real-time RIC (non-RT) handles more extensive orchestration and policy definition tasks and operates in loops of one second or more. The near real-time (Near-RT) RIC manages control loops at intervals between 10 milliseconds and 1 second, directly influencing network performance through radio resource management in the RAN. The Non-RT-RIC manages scalable control policies for thousands of devices, while the Near-RT-RIC ensures fast, local responses to network conditions.

However, there is a disconnect between the importance and value that the RICs, including the Near-RT RIC, bring to the O-RAN architecture and their commercial adoption. ATIS recently released a Minimum Viable Profile (MVP) document for the North America region that includes the Near-RT RIC as an optional component. The availability of E2 and O1 interface implementations on commercial RAN stacks is rare, with open source frameworks representing the most advanced solutions in this area. Overall, this highlights the challenges that the RIC ecosystem and the wider Open RAN community still need to address to make intelligent and programmable wireless networks a reality:

Full specifications including test and interoperability profiles.
These are required at different levels and across different interfaces, including those between the RICs and the RAN and between the application logic and the RAN (e.g. testing E2 service models across xApps and the RAN). As functional specifications evolve rapidly and innovations are introduced every six months, there is a growing need for automated and continuous testing, integration and validation across the end-to-end ecosystem, taking into account both the RAN and the applications running on the RICs become. To address this issue, O-RAN is developing test specifications for RIC interfaces, and O-RAN PlugFests is intensifying its efforts to test RIC-related interfaces. In addition, the test must also include the energy efficiency of these systems. From a continuous integration perspective, we recently released a framework that optimizes support for open source stacks such as OpenAirInterface and srsRAN for xApps on the O-RAN Software Community Near-RT RIC.

Manage network complexity. The RICs and associated AI-driven Open RAN software components introduce additional complexity in terms of configuration, versioning, and resource and infrastructure management. Addressing this complexity through intelligent automation was a key focus of our research and led to the development of an intelligently orchestrated operating system for the deployment of end-to-end, full-stack and fully managed O-RAN systems. This solution is now being commercialized by a Northeastern spin-off, zTouch Networks.

Design efficient and effective AI/ML for the RICs. Further research on AI/ML for the RIC is needed to ensure the development of algorithms that are both efficient and effective under different network conditions. AI/ML-based control solutions must also undergo rigorous validation and testing in controlled environments after training to minimize the risk of disrupting production networks. However, these test environments must also be realistic enough to deliver meaningful results. This requires consideration of user load, traffic patterns, and RF characteristics that reflect real-world deployments in which these models will ultimately be implemented. By improving both the design and testing of AI/ML models, we can better ensure their reliability and performance in dynamic network environments.

Data and role of digital twins. Developing robust and scalable AI/ML solutions that can effectively translate to various real-world deployment scenarios requires leveraging comprehensive datasets of RAN telemetry, operational data, and performance metrics. Although network operators are able to collect such datasets, they are often not suitable for use in research and development due to privacy and security concerns. Wireless digital twins like Colosseum with embedded O-RAN environments provide a viable alternative to address these challenges and enable the creation of accurate and secure test environments for the development and validation of AI/ML models.

Extensions to the real-time domain. Today’s RIC control range concerns time scales above 10 ms or a 5G NR frame. Furthermore, the optimization only interacts with the control plane, without access or influence to user plane data entities. However, real-time control and interaction with the data plane are important requirements for dynamic control in a wide range of use cases, from spectrum capture and sharing to fronthaul optimization to integrated capture and communications. In this context, the O-RAN ALLIANCE nGRG has taken on a research contract on dApps, a real-time extension of the RIC architecture that will provide plug-and-play programmable loops within the RAN itself. This research has received contributions from Northeastern University, NVIDIA, Mavenir, Qualcomm, MITER, and reviews from Ericsson, Samsung, Verizon, Jio, and Keysight.

Addressing these issues is key to developing and deploying programmable, intelligent control loops in mobile systems and unlocking important performance gains through agile and tailored network configurations.

Watch the full Open RAN Forum 2024 on-demand now.