Data Collection and Distribution Framework

The framework ingests heterogeneous telemetry from diverse sources across the RAN, including O-CU, O-DU, and O-RU network functions. It normalizes disparate data formats into a unified schema for consumption by AI/ML applications.

• Collects Performance Measurements (PM) and Fault Management (FM) data via the O1 interface
• Ingests near-real-time UE-level metrics and cell-level KPIs via the E2 interface
• Aggregates application-layer metrics for QoE-aware optimization

Raw telemetry streams are processed through a high-throughput pipeline that filters, enriches, and routes data subsets to specific xApps and rApps based on their declared subscriptions. This prevents consumer overload and ensures each application receives only the data it requires.

• Applies topic-based subscriptions where xApps register for specific KPI streams
• Performs on-the-fly aggregation to reduce data volume before delivery
• Supports windowed queries for time-bounded analytics

A centralized or distributed database that serves as the persistent storage layer for near-real-time RAN state, UE context, and topology information. The R-NIB provides a consistent, queryable view of the network for all hosted xApps and rApps.

• Stores cell topology and neighbor relation tables
• Maintains UE context records including mobility history and bearer information
• Provides a publish-subscribe interface for real-time state change notifications

The framework enforces granular access controls that govern which xApps and rApps can access specific data streams. This ensures data privacy, security, and compliance with operator policies while preventing unauthorized consumption of sensitive network telemetry.

• Implements role-based access control (RBAC) for data subscriptions
• Supports data anonymization for UE-specific information before exposure
• Enforces rate limiting to protect RIC platform stability

Before delivery to training pipelines or inference hosts, the framework applies transformation operations to prepare raw telemetry for machine learning consumption. This includes feature engineering, normalization, and handling of missing or corrupted data samples.

• Performs feature extraction from raw PM counters
• Applies imputation techniques for gaps in telemetry streams
• Generates labeled datasets by correlating events with outcomes for supervised learning

The framework connects to long-term data storage systems to enable offline AI model training and retrospective analysis. It manages the tiering of data between hot, warm, and cold storage based on retention policies and access frequency.

• Archives historical PM data for model training in the Non-RT RIC
• Supports time-series databases optimized for high-cardinality telemetry
• Enables batch export to external analytics platforms and data lakes

A centralized or distributed database within the RIC platform that stores the aggregated output of the collection framework. The R-NIB maintains a near-real-time graph of RAN state, including:

• UE context and mobility history
• Cell topology and neighbor relations
• Current resource block allocations
• Active slice configurations xApps and rApps query the R-NIB as their primary source of truth, decoupling data producers from consumers and enabling stateless application design.

The automated pipeline within the SMO and Non-RT RIC that manages the end-to-end lifecycle of AI models. The data collection framework feeds this pipeline by providing curated, labeled datasets for training. The orchestrator handles:

• Data ingestion and preprocessing from O1 and E2 streams
• Feature engineering and dataset versioning
• Model training, validation, and deployment to inference hosts
• Continuous monitoring for model drift detection

The management framework that integrates the Non-RT RIC and provides unified orchestration of O-RAN network functions. The SMO hosts the data collection framework's aggregation layer, ingesting telemetry from the O1 interface across multiple network functions. It provides the data lake and analytics engine that transforms raw performance measurements into enrichment information distributed to rApps via the A1 interface for policy generation.

The control paradigm that the data collection framework enables. It follows a continuous cycle:

• Observe: The framework collects real-time telemetry from E2 and O1 interfaces
• Orient: AI models within xApps and rApps analyze the data against optimization targets
• Decide: The RIC determines corrective actions (e.g., handover parameter adjustment)
• Act: Commands are issued back through E2 to network functions This loop operates without human intervention, with the framework ensuring data freshness and integrity at each stage.