Operators are looking for cost reduction in both CAPEX and OPEX – and RAN is one of the main cost drivers for operators. It is also one of the most difficult areas when introducing new features and services. Tackling RAN will have the greatest impact on optimizing costs and delivering innovation with more agility in the solution. With traditional legacy solutions, however, operators often have to wait several months to receive feedback on new services, hampering innovation.
Figure 1. Architecture of the O-RAN alliance. Source: O-RAN Alliance
RAN smart controller
The Intelligent RAN Controller (RIC) helps operators optimize and launch new services by enabling them to make the most of network resources. It also helps operators reduce network congestion. The Intelligent RAN Controller (RIC) is cloud-native and is a core component of an open, virtualized RAN network. See a summary of use cases in the table below.
A key feature of the RIC is its ability to support non-real-time applications (rApps) and near-real-time applications (xApps). Both types of applications help optimize network performance by controlling network responses with varying latencies with rApps handling latencies greater than one second and xApps control functions requiring latency less than one second.
Non real-time RIC takes a second or more to execute and therefore guides the RIC in near real time. The non-RT RIC exists within the Service Management and Orchestration (SMO) framework and hosts the policies that are enforced by the RT-near RIC. It also manages the ML models for the RIC close to the RT to be used for decision making based on the network status. The non-RT RIC provides the policies, data, and machine learning models needed for RAN optimization by the RT-near RIC.
Near real-time RIC performs functions between 10 milliseconds and one and communicates between 1. the application layer, 2. the non-RT RIC and 3. the infrastructure layer (O-CU & O-DU) where O-CU has control planes and d disaggregated users to add flexibility to the architecture. Near-RT RIC directly controls and optimizes lower levels of RAN and uses AI and ML to automate RAN and enforce policies that control routing and Quality of Service (QoS).
RICs host applications based on microservices, they are called xApps for Near-RT RIC and rApps for non-real-time RIC. With the help of rApps and xApps, Open RAN integrates AI/ML-based decision making into the solution.
The RIC provides advanced control functionality, which provides increased efficiency and better management of radio resources. These control features leverage analytical and data-driven approaches, including advanced machine learning and artificial intelligence (ML/AI) tools to improve resource management capabilities.
RIC enables a vendor-neutral platform to manage control and management planes. Thanks to the control and management planes, the RICs access the RAN as a whole: elements, connections and functions.
There are 3 control loops available for ORMs to leverage, depending on the needs of the application or service:
- The MNO can use non-RT RIC control loop (rApps) for services/applications with an execution time: 1 second or more.
- ORMs can use an RIC control loop close to RT (xApps) for applications with an execution time: between 10 ms and 1 second.
- And use the O-DU scheduler loop for applications requiring decision and execution time: less than 10 ms.
This enables the RIC to make intelligent decisions about the RAN to optimize performance, from resource and service optimization, to energy optimization and sustainability, and network slice assurance. Many use cases within the network such as optimized media, game streaming, AR/VR and metaverse can be enabled with efficient use of spectrum.
For efficiency and cost-effectiveness, the underlying hardware platform for RIC functions should be optimized for AI/ML-based learning and inference, as well as running all other workloads efficiently. normal working conditions at the node.
AI models fall into two categories: supervised and unsupervised learning. Being a real-time cellular network, it prefers models that are unsupervised learners to eliminate the challenge of the model and continuous training.
Near real-time RIC must include artificial intelligence (AI) as an xAPP responsible for predicting, preventing, and mitigating situations (i.e. handover) that affect the experience customer. The reason the AI needs to be in the near real-time RIC is that the AI will make time-sensitive decisions for network performance. All xAPPs must use unsupervised learning modes.
The AI software will use algorithms created by ML running as rAPP in the non-real-time RIC. All algorithms and formations can be built in non-real time. The reinforcement of these decisions must be done in real time by AI. A non-real-time RIC rAPP ML will help the AI xAPP in the real-time RIC to recognize traffic patterns and anomalies and adjust network health to deliver the appropriate RAN resources for an optimal subscriber experience.
AI/ML algorithms are responsible for:
• Forecast settings
• Detect anomalies
• Predict breakdowns
• Projection of heatmaps
• Classify components into groups
AI/Machine Learning: Enables intelligent operation by maximizing automation, eliminating the human element. This is a huge change for our industry. The architecture and applications are the platform on which you can implement AI and ML concepts.
As a result, it will enable proactive action and the ability to predict the future with some accuracy. Based on the prediction, preventive action can be taken to avoid a similar situation in the future.
Many mobile operators plan to use AI to automate network operations. AI coupled with ML will be the primary tools to ensure the quality of network performance and the resulting quality of end-user experience across all G.
Figure 1. RIC use cases and applications. Source: O-RAN Alliance
AI will be responsible for analyzing data and using ML algorithms to adjust network conditions, provide proper load balancing, and manage transfers seamlessly, all to ensure the subscriber has the best possible experience. .
All data sources, as in Big Data, will need to be considered to first classify the data, secondly recognize the anomaly pattern, then thirdly predict the behavior. Over time, ML algorithms will evolve and become better at predicting and helping AI make real-time network decisions. This will be critical for 5G when humans and objects are connected.
Any AI can only be as good as the data it contains. The data will need to cover different use cases and will include data from different vendors not only on all components of the RAN, but also on the entire network. This is where openness will play a critical role and where the ecosystem needs to be created.
Analytics is a tool to see and understand what is happening in the network and how these changes affect the subscriber experience. Analytics will provide a visual representation of patterns or anomalies and help a mobile operator understand what needs to be fixed to improve network performance for a better subscriber experience. This is an opportunity to review AI data and see reports on how ML is improving the network.
Analytics will be deployed as an rAPP under the non-RT RIC and will use Big Data to provide a holistic view of network conditions. There will be a need for more openness and better APIs between vendors that enable data mining.
In April 2022, the O-RAN Alliance released its second set of specifications for OpenRAN with a major focus on open intelligence.
This included the R1 interface between a rAPP and the non-RT RICs and SMOs, and the A1 interface which connects the non-RT RIC functions in the SMO layer with the near real-time RIC.
This release also included specifications for traffic steering, quality of service and experience, slicing, SMO, and the first release of physical layer acceleration abstraction and security specifications.
This brings O-RAN based components closer to larger deployments to open up and automate RAN.