From RF Channel to Federated Defence: An Extended Review of Digital Twin Methods in 6G-Enabled Internet of Vehicles

Sixth-generation cellular networks and digital twin networks are approaching the Internet of Vehicles from different directions, and the convergence is reshaping how the field thinks about channel modelling, twin lifecycle, learning-based control, and defence at the same time. From the wireless side, sixth-generation networks contribute terahertz spectrum, integrated sensing and communication, reconfigurable intelligent surfaces, and a space-air-ground fabric whose roadside units, unmanned aerial vehicles, and low-earth-orbit satellites operate as co-equal edge tiers. Digital twin networks add a continuously synchronised virtual counterpart for every vehicle, road segment, and radio environment, on which learning algorithms can operate as if it were the physical network itself. Each technology has a substantial literature on its own; the joint deployment they are now becoming raises questions that no single paper resolves, clustered around how a radio-frequency twin is grounded in the physics of a millimetre-wave channel, how vehicle twins migrate across heterogeneous edge servers, how learning is split between vehicles and their twins without leaking the sensor data that gave the twin its fidelity, and how the resulting pipeline is defended against active falsification. This review reads fourteen recent peer-reviewed frameworks against those questions and treats them as components of a single deployed pipeline rather than as isolated proposals. The anchor work is the 3D ray-tracing RF digital twin of Liu et al., given a dedicated discussion in Section 2.2 because most of the learning-oriented frameworks reviewed here rest on a channel abstraction that, in deployment, would have to be served by some form of RF twin. Three comparison tables consolidate the readings: a physical-layer and architecture grid, a learning and resource-management grid, and a security and trust grid. Six research gaps are then identified: unverifiable twin fidelity, fragmented benchmark practice, opaque synchronisation cost, weak active-adversary threat models, absent end-to-end energy accounting, and absent post-quantum readiness. Each is paired with an incremental, testable next step.