부산대학교 도서관

We design an incentive mechanism for edge computing trading between virtual service providers (VSPs) and an edge computing provider (ECP). The VSPs deploy unmanned aerial vehicles (UAVs) to collect sensing data from physical objects to update their digital twins (DTs) to serve their Metaverse users. To process the huge data, the VSP offloads a part of data computation to the ECP. Given the limited computing capacity, we propose a DL-based auction using the augmented Lagrangian method for determining the winning probabilities of the VSPs and their payments. The DL-based auction aims to maximize the ECP's revenue and holds incentive compatibility (IC) and individual rationality (IR) while satisfying budget (BG) constraints. To reduce the offloading cost, a semantic communication (SemCom) technique is deployed at the UAVs of the VSPs. The SemCom technique allows the UAVs to generate and transmit semantic symbols rather than the raw images to their corresponding VSP, which significantly reduces the offloading cost. To train the neural networks used in the DL-based auctions, we use a dataset including valuations of the computing resources to the VSPs, which is a function of the age of DT, the size of the semantic symbol, the sensing time and communication time of the UAVs, and the available computing capacity of the VSP. Simulation results clearly show that the proposed DL-based auction outperforms the classical auctions in terms of ECP's revenue, IR, IC, and BG. The results further show that the use of SemCom reduces the offloading cost for the VSPs.