PaperlayerView exact revision on arXiv

Source-linked AI summary

AI-Generated Incentive Mechanism and Full-Duplex Semantic Communications for Information Sharing

Hongyang Du, Jiacheng Wang, Dusit Niyato, Jiawen Kang, Zehui Xiong, Dong In Kim

arXiv:2303.01896v2eess.SP

TL;DR

MR headset limitations make repeated computation and high-definition information sharing difficult for Metaverse services. The paper proposes full-duplex D2D semantic communication with semantic matching, generalized fading analysis, and diffusion-model contract design. Its diffusion-model incentive mechanism achieves higher SIR utility than SAC and PPO in the reported comparison.

Abstract

from arXiv · show

The next generation of Internet services, such as Metaverse, rely on mixed reality (MR) technology to provide immersive user experiences. However, the limited computation power of MR headset-mounted devices (HMDs) hinders the deployment of such services. Therefore, we propose an efficient information sharing scheme based on full-duplex device-to-device (D2D) semantic communications to address this issue. Our approach enables users to avoid heavy and repetitive computational tasks, such as artificial intelligence-generated content (AIGC) in the view images of all MR users. Specifically, a user can transmit the generated content and semantic information extracted from their view image to nearby users, who can then use this information to obtain the spatial matching of computation results under their view images. We analyze the performance of full-duplex D2D communications, including the achievable rate and bit error probability, by using generalized small-scale fading models. To facilitate semantic information sharing among users, we design a contract theoretic AI-generated incentive mechanism. The proposed diffusion model generates the optimal contract design, outperforming two deep reinforcement learning algorithms, i.e., proximal policy optimization and soft actor-critic algorithms. Our numerical analysis experiment proves the effectiveness of our proposed methods. The code for this paper is available at https://github.com/HongyangDu/SemSharing

I. INTRODUCTION

The paper addresses redundant MR computation and costly image sharing with full-duplex D2D semantic communication, while modeling wireless impairments and incentivizing semantic-information exchange. It combines semantic encoding and matching, generalized fading analysis, and diffusion-model contract design for MR users.

I. INTRODUCTION

  • High-definition view sharing consumes substantial wireless resources, motivating smaller semantic representations for efficient MR information exchange.The framework transmits semantic features whose data size is much smaller than the original view image.Source: §I · p. 2, A. Semantic-Aware Free-Space… · pp. 3–4
  • Contract theory is used to incentivize semantic-information sharing, while multidimensional wireless conditions make traditional contract optimization difficult.The paper motivates generative AI because DRL methods can struggle with high-dimensional states or local optima.Source: §I · p. 2
  • Full-duplex D2D semantic communication shares free-space and semantic information among MR users, avoiding independent computation of virtual-object positions.Users transmit extracted interest points, descriptors, and free-space information rather than repeatedly computing and rendering similar scenes.Source: §I · p. 2, A. Semantic-Aware Free-Space… · p. 3

B. Full-Duplex Wireless Communications

  • The paper derives closed-form expressions for achievable rate and bit error probability under wireless fading and self-interference effects.These metrics characterize the performance of the full-duplex semantic communication system.Source: §I · p. 2

I. INTRODUCTION

  • A diffusion model generates optimal contract designs for semantic-information exchange and is evaluated against reinforcement-learning-based schemes.The paper identifies diffusion-model contract generation as its proposed incentive-mechanism approach.Source: §I · p. 2

B. Full-Duplex Communications

C. Mixed Reality-Aided Metaverse

  • The system targets Metaverse settings where repeated scene computation and high-definition sharing waste resources as user populations grow.The motivation spans virtual-object placement, safe walking-area computation, and communication among MR users.Source: C. Mixed Reality-Aided Metaverse · p. 3

A. Semantic-Aware Free-Space Information Sharing

  • The semantic encoder extracts free-space information, interest points, and descriptors from view images, and users perform lightweight matching on received information.SuperPoint-based encoding and SuperGlue-based matching support spatial correspondence between users’ views.Source: A. Semantic-Aware Free-Space… · p. 4

IV. SEMANTIC MATCHING AND CONTRACT MODELLING

The framework encodes view images into semantic features for efficient sharing, then uses those features to support downstream matching. A self-supervised SuperPoint-based encoder extracts interest points and descriptors efficiently for D2D transmission.

IV. SEMANTIC MATCHING AND CONTRACT MODELLING

  • The proposed semantic matching and contract-modelling framework uses SuperPoint for encoding and SuperGlue for matching received semantic information.Source: §IV · p. 5

A. Self-Supervised Semantic Encoding

  • SuperPoint-based semantic encoding extracts interest points and fixed-length descriptors from each user’s view image in one forward pass.A shared CNN reduces dimensionality before separate encoders produce interest points and descriptors.Source: A. Self-Supervised Semantic Encoding · p. 5
  • The semantic encoder combines interest-point and descriptor encoders and trains them with synthetically warped image pairs.Its loss is the sum of the two encoder losses.Source: A. Self-Supervised Semantic Encoding · p. 5
  • After training, the network quickly extracts the semantic features needed for transmission, with SuperPoint-based architectures supporting real-time extraction.The extracted features are received by other users through D2D full-duplex wireless communication.Source: A. Self-Supervised Semantic Encoding · p. 5

B. Semantic Information Matching

Semantic matching compares received features with locally extracted features, while contract modelling links semantic-information quality to payments and participant utilities. The payment plan accounts for communication quality, encoding costs, and transmission costs.

B. Semantic Information Matching

  • SuperGlue matches received interest points and descriptors with those extracted from the receiving user’s own view image.The system focuses on pairwise matching because each recipient independently performs matching.Source: B. Semantic Information Matching · pp. 5–6
  • The matching network predicts a partial soft assignment matrix from two sets of local features, with correspondence confidence values for possible matches.Source: B. Semantic Information Matching · p. 6
  • SuperGlue combines an attentional graph neural network with an optimal matching layer to reason over feature relationships and produce partial assignments.The optimal matching layer constructs a score matrix, augments it with dustbins, and applies Sinkhorn iterations.Source: B. Semantic Information Matching · p. 6

C. Contract Modelling

1) Payment Plan:

3) Utility of The Semantic Information Receiver:

4) Contract Modelling:

The paper formulates semantic-information incentives as an optimal contract problem and uses a conditional diffusion model to generate contract designs from wireless-system environments.

4) Contract Modelling:

  • The SIR chooses contract items {c_q, c_f} to maximize utility while satisfying incentive compatibility and individual rationality for the SIP.The contract model incorporates the SIP’s utility threshold and derives optimal power for a given contract.Source: 4) Contract Modelling · p. 7
  • The contract-design problem is difficult because optimal contracts depend on high-dimensional wireless conditions, motivating PPO and SAC baselines alongside diffusion generation.The state includes factors such as interference power, interference-path count, and small-scale-fading parameters.Source: 4) Contract Modelling · p. 7

1) Diffusion Model:

2) AI-Generated Contract:

  • The contract-design policy maps environment states to contracts that maximize expected cumulative SIR utility, with the final sample obtained from the reverse diffusion chain.The method trains a contract-quality network and contract-generation network, then generates the contract by denoising Gaussian noise.Source: 2) AI-Generated Contract · pp. 8–9
  • The proposed diffusion-based contract algorithm is presented as more robust and effective than direct neural-network output, and its performance is compared with conventional DRL algorithms.The paper identifies the diffusion model as its first use for incentive design and evaluates it against PPO and SAC.Source: 4) Contract Modelling · p. 7, 2) AI-Generated Contract · p. 9

B. Performance Analysis

The performance-analysis section derives closed-form SINR distribution expressions for full-duplex D2D semantic transmission under fading, interference, self-interference, and noise.

B. Performance Analysis

  • The analysis targets closed-form achievable-rate and bit-error-probability expressions for D2D full-duplex links affected by wireless impairments.The considered impairments include small-scale fading, self-interference, co-channel interference, and noise.Source: B. Performance Analysis · p. 9, 1) PDF and CDF expressions · p. 9
  • Algorithm 1 trains contract-generation and contract-quality networks through exploratory contract execution, replay-buffer updates, and target-network soft updates.During inference, Gaussian noise is denoised using the trained generator to return the optimal contract design.Source: 1) PDF and CDF expressions · p. 9

1) PDF and CDF expressions:

  • The SINR probability density and cumulative distribution functions are derived in closed form using a generalized special-function representation.Lemma 1 states the closed-form PDF and CDF expressions, with the multivariate Fox’s H-function used in the representation.Source: 1) PDF and CDF expressions · pp. 9–10

2) Approximation Analysis:

The approximation analysis simplifies the full-duplex performance model in an interference-limited setting and validates the resulting outage-probability approximation against the exact expression.

2) Approximation Analysis:

  • In the interference-limited scenario, the analysis assumes negligible noise and perfect self-interference cancellation before deriving simplified PDF and CDF expressions.The assumptions are σ_N^2 = 0 and υ = 0.Source: 2) Approximation Analysis · p. 10
  • The outage probability is expressed through the received SNR falling below an outage threshold, with a high-transmit-power simplification obtained from the nearest Mellin-Barnes pole.The approximation is developed for the interference-limited case and uses the derived distribution expressions.Source: 2) Approximation Analysis · p. 10
  • As transmit distance increases, the kth user’s outage probability increases; at high P_j, the approximate expression closely matches the exact expression.Figure 7 compares outage probability against the jth user’s transmit power under different transmit distances.Source: 2) Approximation Analysis · p. 10, Figure 7 · p. 10

3) Achievable Rate in Full-duplex Communications:

The paper derives closed-form achievable-rate and BEP expressions for generalized full-duplex D2D channels, then evaluates how the framework supports semantic information sharing under wireless impairments.

3) Achievable Rate in Full-duplex Communications:

  • The paper derives the kth user’s achievable rate in closed form from the SINR distribution for full-duplex communications.The achievable rate is defined as the maximum channel-supported data rate per Hz.Source: 3) Achievable Rate in Full-duplex… · p. 10

4) Bit Error Probability in Full-duplex Communications:

  • The paper derives a closed-form BEP expression covering multiple modulation and detection combinations.The parameterization includes coherent BPSK, coherent BFSK, non-coherent BFSK, and differentially coherent BPSK.Source: 4) Bit Error Probability in… · pp. 10–11
  • The derived performance indicators incorporate large-scale fading, generalized small-scale fading, self-interference, and arbitrary-multipath co-channel interference.The α-µ fading model extends the metrics to multiple fading conditions.Source: 4) Bit Error Probability in… · p. 11

VI. NUMERICAL RESULTS

  • The framework combines semantic encoding, contract incentives, diffusion-based contract optimization, and full-duplex transmission so users can match shared information instead of recomputing tasks.The end-to-end process shares free-space information, interest points, and descriptors for semantic matching in users’ view images.Source: 4) Bit Error Probability in… · p. 11
  • The numerical experiments examine information-sharing effectiveness, incentive effects on SIR utility, and the influence of wireless channels and self-interference.The simulations use Monte Carlo channel realizations on a CPU/GPU experimental platform.Source: §VI · p. 11

A. Effectiveness of semantic communication framework (for Q1)

The experiments evaluate semantic information sharing, AI-generated contract incentives, and full-duplex channel effects. Results show reduced redundant computation, higher contract utility, and strong sensitivity to transmission power, distance, and self-interference.

A. Effectiveness of semantic communication framework (for Q1)

  • The proposed framework lets users recover free-space information through spatial matching while consuming fewer resources than independent computation.The receiver uses transmitted interest points, descriptors, and free-space information, then performs spatial correspondence and rotation instead of repeating the full computation.Source: A. Effectiveness of semantic… · p. 11
  • As BEP decreases, the recovered free-space information and matching results become more accurate, highlighting the need for reliable semantic transmission.Different user viewpoints can provide missing details through feedback and real-time information updates.Source: A. Effectiveness of semantic… · p. 11, Figure 8 · p. 12

B. Efficiency of the proposed incentive mechanism (for Q2)

  • The AI-generated contract algorithm converges more effectively and achieves higher final reward than SAC and PPO.Diffusion-based sampling uses repeated fine-tuning and supports longer-term dependencies in contract generation.Source: B. Efficiency of the proposed… · p. 11
  • For a fixed environment state, the AI-generated contract reaches an SIR utility of 189.1, versus 185.9 for SAC and 184.3 for PPO.This comparison is reported for the trained models’ optimal contract design capability.Source: B. Efficiency of the proposed… · p. 12, Figure 9 · p. 12

C. Influence of channel and self-interference (for Q3)

  • Increasing the transmitter’s power from 0 to 20 dBW reduces BEP from about 10^-1 to 10^-5, whereas greater interfering power raises BEP.The interference penalty becomes more pronounced at higher transmit power.Source: C. Influence of channel and… · p. 12, Figure 10 · p. 13
  • Increasing a user’s own transmit power can raise BEP through self-interference, while stronger channel parameters reduce BEP.For μ=2, BEP rises from below 10^-4 at 5 dBW to nearly 10^-2 at 30 dBW.Source: C. Influence of channel and… · p. 12, Figure 11 · p. 13
  • Achievable rate increases with transmit power but decreases with distance, reaching approximately 50 Mbit/s at 25 dBW over a 5-meter link.Full-duplex can underperform half-duplex under strong self-interference at low interfering-user power, but benefits more from bandwidth sharing as transmitter power increases.Source: C. Influence of channel and… · p. 13

VII. CONCLUSION

APPENDIX A PROOF FOR LEMMA 1

The appendix derives the PDF and CDF-related expressions for the received signal model using fading distributions, integral transformations, and multivariate special functions.

APPENDIX A PROOF FOR LEMMA 1

  • Rayleigh channel gains and sums of independent Rayleigh-fading signals are converted into distributions used to derive the PDF of the SINR-related variable.The derivation uses a Nakagami-m signal amplitude with m=N and constructs intermediate variables through convolution and substitution.Source: A. Proof of the PDF expression · p. 14
  • Mellin-Barnes integral representations and beta-function identities transform the intermediate integrals into a multivariate Fox’s H-function expression.The appendix completes the PDF derivation after variable substitutions and standard integral identities.Source: A. Proof of the PDF expression · p. 14

B. Proof of the CDF expression

The appendix derives the CDF expression from the previously obtained PDF by applying the CDF definition and evaluating the resulting integral.

B. Proof of the CDF expression

  • The CDF of γ_k is obtained by integrating the derived PDF and applying the stated special-function identities.The proof substitutes the PDF expression into the CDF definition before rewriting the result in closed form.Source: B. Proof of the CDF expression · p. 15

APPENDIX B PROOF FOR LEMMA 2

The appendices derive PDF, CDF, and achievable-rate expressions for the modeled wireless channel using substitutions and special-function identities.

A. Proof of the PDF expression

  • The PDF of γ′_k is derived from earlier expressions through substitution and the cited Mellin-Barnes representation.This completes the first distributional step of the appendix derivation.Source: A. Proof of the PDF expression · p. 15

B. Proof of the CDF expression

  • The CDF is derived from the PDF using its definition and then expressed in closed form with the cited identity.The result is labeled as equation (28).Source: B. Proof of the CDF expression · p. 15

APPENDIX C PROOF FOR THEOREM 1

  • The achievable-rate expression is obtained by substituting the derived distribution into the rate formula and evaluating the remaining integral.The proof uses the cited special-function identities and concludes with equation (31).Source: Appendix C · p. 15

APPENDIX D PROOF FOR THEOREM 2

The appendix completes the proof of Theorem 2 through substitutions and identities involving the Gamma function and cited equations.

  • The derivation rewrites E_k using the definition of the Gamma function [35, eq. (8.350)].Source: Appendix D · p. 15
  • Substitution of (25) into (D-1) produces the next proof expression.Source: Appendix D · p. 15
  • I_D is solved using [35, eqs. (3.351.3) and (8.339.1)].Source: Appendix D · p. 15
  • Substituting I_5 into (D-2) and applying [51, eq. (A-1)] derives (33), completing the proof.Source: Appendix D · p. 15

Related papers

Loading 2303.01896v2…