TWIST: New token sync framework slashes wireless digital twin costs

Wireless digital twins rely on repeatedly synchronizing a physical scene with its digital counterpart over bandwidth-limited links. Traditional pixel-domain or uniformly protected bitstreams often waste resources on semantically irrelevant details. In a new arXiv preprint, Sige Liu and Kezhi Wang introduce TWIST, a closed-loop token synchronization framework that shifts the focus from visual reconstruction to semantic state delivery.

TWIST represents each physical observation as a token and transmits it over a wireless link. Tokens are grouped by task relevance and protected via mode-conditioned unequal error protection across low, medium, and high synchronization modes. At the receiver, decoding confidence flags unreliable hard decisions as erasures, which a completion model restores before updating the twin state. This recovered state enables traffic inference and generates compact feedback on channel quality, receiver uncertainty, semantic drift, and application priority to adapt the next transmission mode. Experiments on a dynamic road-scene digital-twin scenario demonstrate that TWIST improves traffic-state inference and reduces average synchronization cost compared to fixed-mode and channel-only adaptive approaches.