A Game-Theoretic Online Optimization for Federated Digital Twin Construction via Wireless Sensing

In this paper, we propose a novel federated digital twin (DT) construction framework enabled by wireless sensing under edge-cloud collaboration. In this framework, the DT to be built at the cloud is viewed as a global one splitting into and aggregating from multiple functional components, i.e., partial-DTs, created at distributed edge servers (ESs) using feature data collected by associated wireless sensors. Considering time-varying DT evolutions, we formulate an online problem that jointly and dynamically optimizes partial-DT assignments, ES-sensor associations, and computation and communication resource allocations, for maximizing the constructed DT's model quality while minimizing induced costs in long runs. To this end, we transform the original problem into an equivalent two-layer hierarchical game. After game analyses in detail, we design a deep reinforcement learning-based solution integrated with stable matching and overlapping coalition formation, called DMO, to obtain the long-term equilibrium of the hierarchical game, thereby efficiently solving the original problem. Simulations show the effectiveness of the introduced framework, and demonstrate the superiority of the proposed solution over counterparts.