Abstract
Semantic communication (SemCom) and edge computing are two disruptive solutions to address emerging requirements of huge data communication, bandwidth efficiency and low latency data processing in Metaverse. However, edge computing resources are often provided by computing service providers and thus it is essential to design appealingly incentive mechanisms for the provision of limited resources. Deep learning (DL)-based auction has recently proposed as an incentive mechanism that maximizes the revenue while holding important economic properties, i.e., individual rationality and incentive compatibility. Therefore, in this work, we introduce the design of the DL-based auction for the computing resource allocation in SemCom-enabled Metaverse. First, we briefly introduce the fundamentals and challenges of Metaverse. Second, we present the preliminaries of SemCom and edge computing. Third, we review various incentive mechanisms for edge computing resource trading. Fourth, we present the design of the DL-based auction for edge resource allocation in SemCom-enabled Metaverse. Simulation results demonstrate that the DL-based auction improves the revenue while nearly satisfying the individual rationality and incentive compatibility constraints.
Introduction
The Metaverse promises a seamless blend of physical and virtual worlds, enabling immersive experiences through high-fidelity digital twins. However, this vision faces a significant "resource bottleneck." To maintain real-time immersion, systems often target latencies as low as 7–15 ms, which traditional networks struggle to support given the massive volume of raw data generated by IoT devices and UAVs.
To overcome these hurdles, two key technologies are integrated: Semantic Communication (SemCom) and Edge Computing. While SemCom reduces bandwidth requirements by transmitting only meaningful information, Edge Computing Providers (ECPs) offer the necessary hardware for localized data processing. The challenge lies in the "Incentive Gap": ECPs are rational actors who require profitable reasons to share resources, while Virtual Service Providers (VSPs) need cost-effective access to maintain their utility.
Semantic Communication and Edge Offloading
Traditional communication transmits every raw pixel, which is inefficient for the Metaverse. SemCom uses deep learning to extract the "meaning" or semantic symbols from data. For example, in a vehicular Metaverse scenario using UAV imagery, instead of sending a multi-megabyte image, the system identifies and transmits only essential elements like "man riding bike" or "sign on pole."
In the proposed architecture, VSPs process what they can locally and offload the remaining heavy computation to ECPs. This synergy drastically reduces data size—in one case study, an original 3.59 MB image was reduced to 0.65 MB of bounding box data and a mere 56 bytes of text.
DL-Based Auction Design
To manage the trade of edge resources, the authors implement a neural network-based auction inspired by "RegretNet." This mechanism uses two Feed-Forward Neural Networks (FNNs):
- Allocation Network: Determines the probability of a VSP winning a computing unit.
- Payment Network: Determines the price the winner must pay.
The model is trained using an augmented Lagrangian method to minimize a loss function that penalizes violations of two critical economic properties:
- Individual Rationality (IR): Ensures VSPs do not lose money by participating.
- Incentive Compatibility (IC): Ensures VSPs have no incentive to lie about their true resource valuation.
Key Results and Insights
Simulation results using the UAV123 dataset and the RelTR (Relation Transformer) model demonstrate several key findings:
- Revenue Superiority: The DL-based auction consistently outperformed the standard Vickrey–Clarke–Groves (VCG) auction in revenue generation for the edge provider.
- Economic Stability: IR and IC penalties dropped to nearly zero during training, proving the auction is economically robust and fair.
- The Efficiency Paradox: Interestingly, while SemCom makes the system more efficient, it can reduce total ECP revenue because VSPs require fewer computing units to process the shrunken semantic data. However, as the density of the virtual world and the number of VSPs grow, competition drives the value back up, suggesting the model scales well.
Conclusion and Future Work
This research provides a robust framework for resource allocation in the SemCom-enabled Metaverse. Future work will address more complex market structures, such as "double auctions" involving multiple competing ECPs, and dynamic optimization where VSPs can adjust compression levels based on real-time market pricing.
Summary updated