An Enhanced Dual-Currency VCG Auction Mechanism for Resource Allocation in IoV: A Value of Information Perspective

Wei Wang, Nan Cheng, Conghao Zhou, Haixia Peng, Haibo Zhou, Zhou Su, Xuemin Shen

TL;DR

This paper introduces an enhanced Dual-Currency VCG Auction mechanism designed to solve the complex problem of resource slicing in the Internet of Vehicles (IoV). By using "Value of Information" (VoI) as a novel currency and a Mean-Field Multi-Agent Reinforcement Learning (MFMARL) framework, the authors create a system that maximizes social welfare even when vehicle service providers' utility functions are unknown, while effectively preventing bidder collusion.

What Problem Does This Paper Address?

In the era of 6G and the Internet of Vehicles (IoV), Network Service Providers (NSPs) use Network Function Virtualization (NFV) to "slice" resources (like bandwidth, compute, and cache) for different Vehicle Service Providers (VSPs). The goal is to maximize social welfare—the sum of the benefits all VSPs receive.

However, two major hurdles exist:

  1. The "Black Box" Utility Problem: NSPs usually don't know the VSPs' internal utility functions (how much a VSP actually values a specific resource). These functions are often private or too complex to express mathematically.
  2. Selfishness and Collusion: VSPs are profit-driven. In a standard auction, they might lie about their needs or collude with other VSPs to manipulate prices, leading to unfair or inefficient resource distribution.

Traditional auction mechanisms like the Vickrey-Clarke-Groves (VCG) auction are great for truth-telling but typically require explicit knowledge of these "black box" utility functions to calculate fair payments.

Methods

The authors propose a three-pronged solution:

  • Value of Information (VoI) as Currency: Instead of bidding with traditional money, VSPs bid using "VoI Currency" (VC). VoI acts as a proxy for the expected benefit a VSP gets from a resource.
  • The "Bank" (A Third-Party Entity): The mechanism introduces a neutral "Bank" that issues VC, provides loans to VSPs for bidding, and calculates taxes. This separates the financial incentive management from the actual resource allocation performed by the NSP, protecting VSP privacy.
  • MFMARL (Mean-Field Multi-Agent Reinforcement Learning): Standard multi-agent learning struggles to scale as the number of vehicles grows (the "curse of dimensionality"). MFMARL simplifies this by having each agent react to the average (mean field) behavior of its neighbors rather than every individual agent, allowing the system to scale to 50+ VSPs efficiently.

The auction happens in three phases:

  1. Issuance: The Bank estimates total VoI and creates a currency pool.
  2. Bidding: VSPs submit "intention" bids in VC to the NSP.
  3. Disclosure & Taxation: The NSP allocates resources. The Bank then compares the VSPs' bids against their actual performance (gross profits) and applies a "VC interest" tax to ensure that bidding truthfully was the most profitable strategy.

Datasets

The study relies on simulated IoV environments rather than a static external dataset.

  • Environment: A multi-VSP network (10 to 50 VSPs) with three types of resources (bandwidth, computation, cache).
  • Utility Modeling: VSPs are assigned utility functions based on exponential/logarithmic models (e.g., $u = \beta(1 - e^{-(\xi s)})$) to represent realistic diminishing returns on resources.
  • Benchmarks: The proposed DVCG-MFMARL is compared against DVCG-MADDPG (standard multi-agent DDPG), DVCG-MARL (independent learning), and QVCG (a traditional VCG auction using quantization).

Key Insights and Ideas

  • Truthfulness is the Best Policy: The authors mathematically prove the mechanism is Dominant-Strategy Incentive Compatible (DSIC). Because the "Bank" taxes the difference between a VSP's bid and its actual realized utility, a VSP that overbids or underbids ends up losing more in taxes than it gains in resources.
  • Breaking Collusion: In many auctions, two players can agree to bid low to keep prices down. Here, the MFMARL framework introduces enough exploration-based randomness that it becomes nearly impossible for conspirators to maintain a stable, collusive strategy.
  • Privacy Preservation: Because the NSP learns the relationship between bids and resources through reinforcement learning (rather than direct data access), VSPs never have to reveal their sensitive internal business logic or raw utility data.

Key Results

  • Faster Convergence: DVCG-MFMARL reached a stable, high social welfare state significantly faster than MADDPG-based versions, especially as the number of VSPs increased.
  • Efficiency: The mechanism requires drastically less information exchange volume. For 50 VSPs, the proposed method exchanged only $0.025 \times 10^6$ floating-point values, compared to $50.0 \times 10^6$ for high-precision quantized VCG (a 2000x reduction).
  • Social Welfare: The system achieved a total social welfare that closely approached the theoretical "Optimal Allocation" (the pink line in their charts), outperforming traditional quantized VCG methods.
  • Robustness: Simulations showed that even when VSPs intentionally tried to collude or shift their bidding behavior, their net profits were always highest when they returned to truthful bidding.

Limitations and Future Work

  • Computational Overhead: While MFMARL is more efficient than standard MARL, the "Bank" and NSP still need to run complex neural network updates, which might be challenging in ultra-low-latency 6G environments.
  • Multi-Dimensional Complexity: The current model focuses on three resource types. The authors note that future work should expand this to MIMO systems involving more complex constraints across spatial, temporal, and power domains.
  • Real-World Dynamics: The study assumes a relatively stable set of VSPs during the auction rounds; testing the mechanism against highly mobile vehicles entering and leaving the network rapidly remains a future challenge.

Summary updated

Abstract

The Internet of Vehicles (IoV) is undergoing a transformative evolution, enabled by advancements in future 6G network technologies, to support intelligent, highly reliable, and low-latency vehicular services. However, the enhanced capabilities of loV have heightened the demands for efficient network resource allocation while simultaneously giving rise to diverse vehicular service requirements. For network service providers (NSPs), meeting the customized resource-slicing requirements of vehicle service providers (VSPs) while maximizing social welfare has become a significant challenge. This paper proposes an innovative solution by integrating a mean-field multi-agent reinforcement learning (MFMARL) framework with an enhanced Vickrey-Clarke-Groves (VCG) auction mechanism to address the problem of social welfare maximization under the condition of unknown VSP utility functions. The core of this solution is introducing the ``value of information" as a novel monetary metric to estimate the expected benefits of VSPs, thereby ensuring the effective execution of the VCG auction mechanism. MFMARL is employed to optimize resource allocation for social welfare maximization while adapting to the intelligent and dynamic requirements of IoV. The proposed enhanced VCG auction mechanism not only protects the privacy of VSPs but also reduces the likelihood of collusion among VSPs, and it is theoretically proven to be dominant-strategy incentive compatible (DSIC). The simulation results demonstrate that, compared to the VCG mechanism implemented using quantization methods, the proposed mechanism exhibits significant advantages in convergence speed, social welfare maximization, and resistance to collusion, providing new insights into resource allocation in intelligent 6G networks.
BibTeX
@article{wang2026enhanced,
  author = {Wei Wang and Nan Cheng and Conghao Zhou and Haixia Peng and Haibo Zhou and Zhou Su and Xuemin Shen},
  title = {{An Enhanced Dual-Currency VCG Auction Mechanism for Resource Allocation in IoV: A Value of Information Perspective}},
  journal = {IEEE Trans. Mobile Computing},
  year = {2026},
  doi = {10.1109/tmc.2025.3594358},
}

Loading related content...

Figure