Model Drift-Adaptive Resource Reservation in ISAC Networks: A Digital Twin-Based Approach

Reserving resources in integrated sensing and communication (ISAC) networks requires the predictive modelling of the demand and performance of a sensing service, which can be achieved by the synergy of mathematical and Artificial Intelligence (AI) models. However, under non-stationary network conditions, the accuracy of predictive modelling can degrade over time, which is referred to as model drift. In this paper, we propose a digital twin (DT)-based model drift-adaptive resource reservation scheme. The objective is to provide a statistical quality of service guarantee with minimal resource consumption for a sensing service. To achieve this objective, a DT is constructed at a network controller that collects the locations of sensing nodes and targets and captures their spatial distribution. To handle model drift, two candidate predictive spatial models are constructed in the DT. Then, a network emulation-based scheme is proposed to evaluate the resource reservation decisions made with the two candidate spatial models and select the optimal decision. Simulation results show that the proposed resource reservation scheme can reduce the resource under-/over-provision under non-stationary network conditions.