DSX Blueprint: Building the Omniverse-Connected Telecom Digital Twin

NVIDIA Omniverse was designed for industrial digital twins — manufacturing plants, autonomous vehicle simulation, architectural visualization. Telecom was not the primary use case. So why did OranSense build its digital twin platform on Omniverse?

The answer is physics fidelity.

A telecom digital twin that is useful for network optimization needs to accurately simulate radio propagation in complex urban environments. This requires ray tracing — the same computational technique used for photorealistic rendering. Omniverse's USD-based scene graph and GPU-accelerated ray tracing engine provide exactly this capability, at a scale and performance level that no telecom-specific tool can match.

When we simulate a 5G mmWave deployment in a dense urban canyon, we need to model reflections off glass facades, diffraction around building edges, and absorption through foliage. These are rendering problems as much as they are physics problems. Omniverse solves them.

The DSX Blueprint defines a layered architecture that connects the physical network to its digital representation and back. Each layer has a specific responsibility, and the interfaces between layers are standardized.

The foundation is the physical network — base stations, antennas, fiber, edge compute nodes. This layer is represented in the digital twin as a geometrically accurate 3D model, sourced from operator GIS data, site surveys, and photogrammetric reconstruction.

The physical layer is not static. As infrastructure changes — new sites deployed, antennas retilted, equipment replaced — the digital twin is updated to reflect the current state. This synchronization is automated through integration with the operator's network inventory system.

Above the physical infrastructure sits the radio environment model. This layer represents the propagation characteristics of the deployment area — building geometry, terrain, vegetation, and the electromagnetic properties of surfaces.

The radio environment model is built from a combination of sources: LiDAR point clouds for building geometry, satellite imagery for land cover classification, and empirical measurements from the live network for calibration. The calibration step is critical — it ensures that the simulated propagation matches the measured propagation, not just in aggregate but at the individual cell level.

The network state layer maintains a real-time representation of the live network's operational state — which cells are active, what configurations they are running, what KPIs they are reporting. This layer is continuously synchronized with the live network through streaming telemetry interfaces.

The network state layer is the bridge between the static physical model and the dynamic operational reality. It is what makes the digital twin a live representation rather than a historical snapshot.

The sensing layer integrates data from the InfraSense platform — the environmental intelligence derived from radio propagation measurements. This includes mobility patterns, environmental conditions, and physical event detections.

In the digital twin, sensing data is visualized as overlays on the 3D scene — heat maps of pedestrian density, vectors of vehicle flow, indicators of environmental conditions. This visualization capability is one of the most powerful aspects of the Omniverse integration: complex multi-dimensional data becomes spatially intuitive.

The AI inference layer runs the optimization and prediction models that make the digital twin actionable. This includes the AODT optimization engine, the anomaly detection models, and the capacity planning algorithms.

The inference layer operates on the combined state of layers 1–4, using the full context of the physical environment, radio propagation, network state, and sensing data to produce recommendations and predictions.

The simulation engine is the computational core of the digital twin. It takes proposed changes — configuration modifications, new site deployments, spectrum assignments — and simulates their effects across the full network model.

The simulation engine runs on GPU clusters, using Omniverse's physics simulation capabilities for propagation modeling and OranSense's custom RAN simulation for protocol-level behavior. A full simulation of a metropolitan network can be completed in minutes, enabling rapid iteration on optimization hypotheses.

The top layer is the human interface — the dashboards, 3D visualizations, and control surfaces through which operators interact with the digital twin. This layer is built on Omniverse's visualization capabilities, providing photorealistic 3D rendering of the network environment with real-time data overlays.

The visualization layer is not just for display. It is also a control surface — operators can select cells, modify configurations, and trigger simulations directly from the 3D view. The spatial interface makes network operations more intuitive, particularly for complex multi-site scenarios where the geographic relationships between cells are important.

The hardest engineering problem in building the DSX Blueprint was not the simulation or the visualization. It was synchronization — keeping the digital twin consistent with the live network in real time.

The challenge is that the live network is continuously changing. Cells go in and out of service. Configurations are modified. Traffic patterns shift. Environmental conditions evolve. The digital twin must track all of these changes without falling behind.

Our synchronization architecture uses a change-data-capture approach: rather than periodically polling the network for its full state, we subscribe to change events and apply them incrementally to the twin. This reduces the synchronization latency from minutes (polling) to seconds (event-driven), and dramatically reduces the data volume that needs to be processed.

For the radio environment model, which changes more slowly, we use a hybrid approach: continuous updates for network state and sensing data, periodic recalibration (daily or weekly) for the propagation model.

The DSX Blueprint is designed to integrate with the existing tools and systems in a network operator's environment. We maintain certified integrations with eight major platforms:

• NVIDIA Aerial — GPU-accelerated RAN processing and AI inference
• NVIDIA Omniverse — 3D simulation and visualization
• Nokia Network Services Platform — OSS/BSS integration for network state synchronization
• Ericsson AMOS — Configuration management and alarm integration
• Red Hat OpenShift — Container orchestration for edge deployment
• Grafana — Metrics visualization and alerting
• Kafka — High-throughput telemetry streaming
• InfluxDB — Time-series storage for network measurements

Each integration is tested against the current production version of the target platform and updated with each DSX Blueprint release.

The DSX Blueprint is not an end in itself. It is the foundation for a new class of network operations capabilities that are not possible without a high-fidelity, real-time digital twin.

Predictive maintenance — By correlating physical sensor data with network performance trends, the twin can identify hardware degradation before it causes failures.

Autonomous optimization — With a validated simulation environment, AI-driven optimization can be deployed with confidence, knowing that recommendations have been tested before execution.

Capacity planning — Traffic growth scenarios can be simulated against the current network model, identifying capacity constraints months before they become problems.

Incident response — When a network event occurs, the twin provides immediate context — what changed, what the propagation impact is, what the recommended response is.

The DSX Blueprint is the infrastructure for the intelligent network. Everything else is built on top of it.