State Mirroring

State mirroring is not a periodic batch upload; it is a continuous, bidirectional data flow. Configuration changes pushed to the physical network are instantly reflected in the twin, while simulated optimizations from the twin can be safely committed to the live network. This creates a closed control loop where the digital state and physical state are kept in lockstep, often using streaming telemetry protocols like gNMI or NETCONF.

The mirror must capture the full depth of the physical entity's state, not just surface-level KPIs. This includes:

• Hardware State: CPU load, memory utilization, FPGA register values, and temperature of each RU.
• Protocol State: Active RRC connections, bearer contexts, and HARQ process status.
• RF Environment State: Current channel state information (CSI), interference levels, and beamforming weights.
• Software State: Running firmware versions, patch levels, and active feature licenses.

The value of a mirror is defined by its synchronization freshness. For a RAN digital twin to be useful for near-real-time control loops, the mirroring latency must be deterministic and bounded. This requires a hard real-time communication fabric between the physical gNB and the twin, ensuring that the state delta between the physical and virtual entity is always within a known, acceptable window, typically measured in milliseconds for PHY-layer use cases.

In a distributed system, maintaining a single source of truth is critical. State mirroring must implement consistency models to resolve conflicts. If a local autonomous function on the gNB changes a parameter at the same time the twin pushes an update, a conflict resolution policy—such as last-writer-wins or a custom merge function—must execute. This ensures the twin does not diverge into a corrupted or impossible state.

Physical network elements from different vendors (e.g., Ericsson, Nokia, Samsung) represent identical operational states using proprietary data models. State mirroring requires a semantic normalization layer that translates vendor-specific YANG models and MIBs into a unified, vendor-agnostic information model. This abstraction is what allows a single digital twin to simulate and manage a multi-vendor RAN deployment seamlessly.

Beyond the current state, the mirror maintains an append-only, time-series log of all state transitions. This immutable history is essential for:

• Root Cause Analysis: Rewinding the twin to the exact state preceding a network fault.
• Auditability: Providing a verifiable record of every configuration change for compliance.
• Training Data: Generating high-quality, sequential datasets for training predictive AI models on real operational patterns.