Non-Real-Time RAN Intelligent Controller (Non-RT RIC)

The Non-RT RIC communicates declarative policies and enrichment information to the Near-RT RIC via the standardized A1 interface. Unlike direct control commands, these policies define high-level optimization targets—such as maximizing energy efficiency or guaranteeing slice SLAs—while allowing the Near-RT RIC to autonomously determine the specific radio resource management actions. This hierarchical separation of concerns ensures that long-term business objectives are enforced without micromanaging millisecond-level control loops.

The Non-RT RIC serves as the exclusive hosting platform for rApps—modular, microservice-based applications that leverage AI/ML analytics to generate optimization recommendations. Each rApp addresses a specific use case:

• Traffic steering rApps analyze historical patterns to optimize load distribution policies
• Energy saving rApps predict cell utilization to recommend carrier shutdown schedules
• Anomaly detection rApps identify degradation trends before failures occur rApps consume data from the R-NIB and output policy guidance, never executing direct RAN control.

The Non-RT RIC orchestrates the complete end-to-end lifecycle of machine learning models used throughout the RAN:

• Training: Leverages offline data lakes with historical network telemetry to train models without real-time constraints
• Validation: Tests model accuracy against held-out datasets before deployment
• Deployment: Pushes trained model artifacts to the Near-RT RIC for inference execution by xApps
• Monitoring: Continuously tracks model drift by comparing live inference accuracy against baselines
• Rollback: Automatically reverts to previous model versions if performance degrades

A critical component that converts high-level business intents expressed in natural language or declarative syntax into machine-executable policies. For example, an operator intent stating 'Ensure premium subscribers receive at least 50 Mbps during peak hours' is decomposed into:

• Specific QoS Class Identifier (QCI) mappings
• Resource block allocation priorities
• Inter-frequency load balancing thresholds This translation layer bridges the gap between business operations and network engineering, enabling true intent-based networking within the O-RAN framework.

Multiple rApps operating simultaneously may generate contradictory policy recommendations—for instance, an energy-saving rApp requesting cell shutdown while a capacity rApp demands additional carriers. The Non-RT RIC implements conflict detection and resolution mechanisms that:

• Analyze overlapping policy scopes across rApps
• Apply operator-defined priority rules to resolve contradictions
• Coordinate with the Near-RT RIC to ensure only coherent, non-conflicting guidance is enforced This safeguard prevents the network instability that could arise from uncoordinated autonomous optimization loops.

The Non-RT RIC ingests and processes long-term performance data from multiple sources to provide contextual enrichment to the Near-RT RIC:

• Historical PM (Performance Measurement) counters from the O1 interface
• UE mobility patterns and geolocation trends
• External data sources such as weather forecasts or event schedules
• Cross-cell and cross-vendor topology information This enriched data enables xApps to make more informed near-real-time decisions by understanding broader temporal and spatial context beyond their immediate E2 data streams.