Author: Dr. Sameh Yamany, Chief Technology Officer, VIAVI Solutions
By 2030, close to 40% of wireless traffic is expected to be AI traffic. Every operator will face the question of who tests the AI, and with what data.
6G will be designed against AI traffic that does not yet exist, for applications not yet built, by operators who cannot afford to find out in production whether it works. AI will make decisions in the network, and those must be tested. Testing AI requires AI trained on real data and deployed in a digital twin that behaves like the live network.
AI Traffic as a Design Constraint
Understanding how AI traffic behaves alongside other traffic is now a 6G design constraint. Most discussion centers on AI inside the network, such as optimization agents and failure-prediction models. AI traffic through the network matters equally: inference traffic between data centers, model updates pushed to edge devices, and sensing data from drones, vehicles, and ships. All must be tested and designed against.
The Data Is the Test
A model trained on the wrong distribution will drift, and one trained on synthetic data can fail on meeting a real subscriber, radio environment, or city reshaped by construction. The data is the test, and the data is the design. The requirement is data for AI, not just emulation: the right data, fed to the right twin, to train and test the models that will run the network.
What a Realistic Twin Requires
Most digital twins use synthetic data, such as a rendered city block, a ray-traced beam, or a single simulated user. Such twins can fail against a real network; research environments often behave differently from live metropolitan coverage.
A more realistic approach repurposes field data. Some 25 years of field measurements across 2G to 5G-Advanced networks, including KPI verification and root-cause and troubleshooting data, were gathered to fix problems but can train an AI engine that emulates how a network responds to change. Three perspectives combine:
- High-granularity time-series data from the live network.
- 3D spatial information, including building geometry, beam shapes, and handover patterns.
- Field-collected sensor and instrument data correlated with AIOps and location-intelligence telemetry.
Scenario Generation
Consider a city fire, with emergency services activated, drones flying, and coverage shifting in real time. The operator needs to know how the network will behave with its actual gNB configuration, transport, and subscriber load. The twin generates the scenario and the AI predicts the response before the live network is changed. Each scenario improves the twin, which also tests applications and interfaces.
The Generative Reality Digital TwinTM
This is a high-fidelity replica of a live network, continuously synchronized with real data, in which AI tests AI in real time. It mirrors the network’s traffic, conditions, and scale as it evolves. Every change, model, configuration, and upgrade is validated in the twin first, speeding deployment while protecting production.
Use Cases
Use cases include:
- Construction sites: Coverage changes daily as cranes go up. The twin predicts heat maps and runs what-if scenarios, such as moving a 5G antenna to a vehicle or drone.
- Ports: Physical and network interactions are emulated together, with congestion modeled against ship counts and container heights. Dynamic beamforming is tested, and an antenna moving on a barge is tracked within milliseconds. Ports are among the most complex commercial wireless environments.
- Developer sandboxes: A US operator gave application developers a sandbox trained on its own network, so they could build against realistic behavior without live access.
- Agriculture and IoT: IoT interactions are emulated at scale for low-coverage rural environments.
- Railways: As European operators migrate railway communications from 2G to 5G and 5G-Advanced, the twin runs what-if scenarios for signaling and communications that no live track can host.
- Non-terrestrial networks: Satellite emulation plus real terrestrial data models Doppler effects, delay, handover, and NB-IoT scenarios.
AI-RAN, AIOps, and the Data Layer
An O-RAN Alliance-certified lab, once focused on emulating radio access networks, now also creates open, shareable datasets for any party testing or training AI-RAN models. This sits upstream of AIOps. AIOps closes the day-two loop: predict, detect, resolve. The twin closes the day-zero and day-one loop: simulate, validate, optimize before deployment. Both rely on real network telemetry rather than synthetic generation.
Quantum-Safe from Day One
6G is being specified with a non-negotiable requirement to be quantum-safe from day one, so testing must begin now, in the twin, before the standard is finalized. There is no time to test it in the live network afterward.
Integrated Sensing and Communication
ISAC merges sensing into the communication layer. The twin must model environments through sensing data, then test communication scenarios on the same model.
Implications for 6G
Real data is the most valuable element, and twins trained on years of real telemetry will matter most. The twin is where AI learns.
Trust starts with VIAVI. Learn more about the VIAVI Generative Reality Digital Twin – and how it federates domain-specific twins into a single validation layer for autonomous network operations – by visiting Generative Reality Digital Twin (GRDT) | VIAVI Solutions Inc.