Open radio access network (O-RAN) provides a clear path to evolve radio access networks toward open, interoperable interfaces, RAN virtualization, Big data and AI-enabled RAN intelligence. The RAN Intelligent Controller (RIC), which controls and optimizes RAN functions, is a critical component of the O-RAN architecture that enables and ultimately allows service providers to build an open, intelligent and smarter RAN. The closed RAN data, which was previously accessible only to RAN vendors, is now available for service providers and third-party vendors to build innovative applications using AI/ML technologies. These specialized, AI-powered RIC applications (also known as rApps/xApps) allow operators to enable new business models, personalize the service experience and optimize operational efficiency. Some examples include:
- Network Slicing: Network slicing is a key advance in 5G networks, with end-to-end connectivity and data processing tailored to specific customer requirements or workloads. The RIC and associated applications can continuously monitor slice performance metrics and initiate corrective action in case of SLA violation.
- Traffic Steering: The RIC and the associated apps can monitor the dynamically changing network load, using AI/ML-based steering algorithms to distribute the load to different frequencies within the same base station, to neighboring base stations or even to different radio access technologies, resulting in efficient utilization of operator resources.
- Energy Efficiency: AI-driven predictions and controls can look at insights from traffic, coverage, interference and other factors to optimize the energy efficiency of the RAN by switching off antennas as needed.
- Massive Multiple Input and Multiple Output (M-MIMO) Optimization: Massive multiple input and multiple output (M-MIMO) provides greater capacity and minimizes interference in 5G networks. By applying AI/ML and decision-making in real time, the RIC can proactively and continuously improve the subscriber’s experience even in dense areas or at times when demand is surging, such as in crowded cities or entertainment venues.
- Quality of Experience (QoE): Intelligent, real-time controls allow a better user experience for latency-sensitive or bandwidth-intensive applications like cloud virtual reality. The RIC and associated applications can use analytics to take policy-based actions, ensuring that priority users maintain a satisfactory QoS even during peak loads.
Vodafone AI-Assisted RAN Slice SLA Assurance Proof of Concept
Juniper Networks completed several RIC trials with Tier-1 service providers and ecosystem partners to demonstrate innovative AI/ML-powered use cases on top of the Juniper RIC platform. Among the more important ones is a proof of concept (PoC) with Vodafone to implement AI-assisted RAN Slice SLA Assurance for network slicing. Network slicing provides customized end-to-end connectivity and data processing to meet specific business needs, including high data rates, traffic densities, service availability, and low latency. Each slice specifies service requirements such as data rate, traffic capacity, user density, latency, reliability and availability, which are provided through a service level agreement (SLA) between mobile operators and business customers. Service providers are interested in mechanisms to ensure SLAs are met without any violations. O-RAN’s open interfaces and AI-/ML-based architecture can enable these mechanisms, making it easier for service providers to realize the opportunities of network slicing.
The traditional, deterministic RAN Slice Assurance approach would wait for an SLA violation and then initiate corrective action by using closed-loop automation. For example, it can adjust radio resource (PRBs – Physical Resources Blocks) allocation across different cells to meet the slice SLA requirements to bring the SLA back in conformance. In contrast, the AI-assisted RAN Slice SLA Assurance approach uses AI/ML techniques to forecast potential SLA violations and immediately initiates preventive action to proactively allocate PRBs while demand is still increasing so that the slice can support the increased demand beforehand without SLA violation. This capability would be very useful when an SLA violation might happen due to sudden changes in traffic patterns, such as public emergencies, special conditions, holidays etc.
Here is the high-level architecture for the AI-assisted RAN slice SLA assurance approach.
The architecture shows the Service Management and Orchestration (SMO) microservices used to implement AI-assisted RAN Slice SLA Assurance use case. The RAN Network Slice Sub-Net Management Function (NSSMF) manages the portion of the network slice within the RAN domain, including slice creation, activation, deactivation and deletion. The SMO Network Function Management Function (NFMF) collects the data from the network using the O1 interface from the O-RAN Central Units (O-CUs) and O-RAN Distributed Units (O-DUs) and accumulates them in the Performance Management Key Performance Indicator (PM KPI) database. This data is used to train an AI/ML model that can be deployed as a rApp on top of the Juniper RIC platform.
The AI-Assisted RAN Slice SLA Assurance use case consists of two phases: the AI/ML training phase and the AI/ML inference phase. In the AI/ML training phase, the data collection and ingestion pipeline microservice passes the PM KPI data in real time for model training. Once the model is trained and available, the AI/ML model packaging and deployment service takes care of deploying the model on AI/ML RAN Slice SLA Predictor rApp.
On the Non-RT RIC, we have two rApps deployed, AI/ML RAN Slice SLA Predictor rApp and RAN Slice SLA Assurance rApp. The AI/ML RAN Slice SLA Predictor rApp is running the trained AI/ML model and performing predictive SLA assurance together with RAN Slice SLA Assurance rApp/xApp. The AI/ML RAN Slice SLA Predictor rApp predicts potential future SLA violations based on the current traffic patterns in the network. It works in tandem with the RAN Slice SLA Assurance rApp/xApp to take corrective action based on the current network data as well as the forecasted SLA data. The RAN Slice SLA Assurance rApp issues A1 policies to RAN Slice SLA Assurance xApp on Juniper Near-RT RIC to execute assurance actions on the network using E2 interface.
The AI/ML models were trained using anonymized cell-level data from Vodafone and simulated datasets capturing real-world network conditions and anomalies. The results were promising: The PoC demonstrated that AI-assisted approach outperformed the traditional RAN Slice Assurance by detecting and preventing 15% to 30% of SLA violations. The AI/ML model running on the AI/ML RAN Slice SLA Predictor rApp can be updated on a need basis depending on changes in traffic patterns.
Accelerating the Development of AI-powered RIC Applications
The Juniper PoC with Vodafone successfully proved that the RIC with AI-/ML-enabled applications can analyze real-time network data to predict and identify optimal network resource allocation. This ensures that corrective actions can be taken by the application, ensuring smooth operation even before a user notices any service degradation.
Juniper believes that AI-powered applications on the RIC can reduce operational expenditures (OpEx) and enhance network efficiency and reliability for network operators. To accelerate progress, the industry should embrace a more structured approach to training, deploying, monitoring and refreshing machine learning models. Additionally, radio data should be made available so that new, innovative companies can better use the data to realize sustainable and scalable AI-powered RIC applications.
Click here to learn more about the Juniper RAN Intelligent Controller (RIC) solution and its benefits.
