PaperlayerView exact revision on arXiv

Source-linked AI summary

Attention-aware Resource Allocation and QoE Analysis for Metaverse xURLLC Services

Hongyang Du, Jiazhen Liu, Dusit Niyato, Jiawen Kang, Zehui Xiong, Junshan Zhang, Dong In Kim

arXiv:2208.05438v6cs.AI

TL;DR

Conventional resource allocation does not account for users’ differing attention to virtual objects, motivating personalized Metaverse xURLLC. The paper combines an MSP–InP contract framework, a QoE metric, and attention-aware rendering allocation; on UOAL data, it reports a 20.1% average QoE improvement over uniform allocation.

  • Problem

    Rendering capacity should reflect users’ differing attention to virtual objects to support personalized Metaverse service quality.

    Source: A. Background · p. 2, 3) Utility of MSP · p. 5

  • Method

    The paper proposes Meta-Immersion using the Weber–Fechner Law and a two-step attention-aware rendering allocation algorithm within a contract-based MSP–InP service framework.

    Source: B. Contributions · p. 3, §IV · p. 6

  • Results

    20.1% average QoE improvement is reported for xURLLC attention-aware allocation over conventional URLLC with uniform resource allocation.

    Source: B. Contributions · p. 3

  • Takeaways & Limitations

    The reported average improvement reaches 40% when total resources for Metaverse xURLLC services are limited.

    Source: B. Contributions · p. 3

Abstract

from arXiv · show

Metaverse encapsulates our expectations of the next-generation Internet, while bringing new key performance indicators (KPIs). Although conventional ultra-reliable and low-latency communications (URLLC) can satisfy objective KPIs, it is difficult to provide a personalized immersive experience that is a distinctive feature of the Metaverse. Since the quality of experience (QoE) can be regarded as a comprehensive KPI, the URLLC is evolved towards the next generation URLLC (xURLLC) with a personalized resource allocation scheme to achieve higher QoE. To deploy Metaverse xURLLC services, we study the interaction between the Metaverse service provider (MSP) and the network infrastructure provider (InP), and provide an optimal contract design framework. Specifically, the utility of the MSP, defined as a function of Metaverse users' QoE, is to be maximized, while ensuring the incentives of the InP. To model the QoE mathematically, we propose a novel metric named Meta-Immersion that incorporates both the objective KPIs and subjective feelings of Metaverse users. Furthermore, we develop an attention-aware rendering capacity allocation scheme to improve QoE in xURLLC. Using a user-object-attention level dataset, we validate that the xURLLC can achieve an average of 20.1% QoE improvement compared to the conventional URLLC with a uniform resource allocation scheme. The code for this paper is available at https://github.com/HongyangDu/AttentionQoE

I. INTRODUCTION

The paper motivates personalized xURLLC for Metaverse QoE and combines attention-aware resource allocation with an MSP–InP contract framework. It reviews attention, VR, and contract-theory research before introducing the wireless network model.

A. Background

  • Conventional URLLC focuses on objective KPIs, while Metaverse services require personalized stimulation and subjective user feelings for an immersive experience.The paper frames QoE as a comprehensive service-design goal encompassing technical indicators and user feelings.Source: A. Background · p. 1

B. Contributions

  • The paper proposes an MSP–InP contract framework that maximizes MSP utility while giving the InP incentives to participate.The contract is designed around incentive compatibility and individual rationality.Source: A. Background · p. 2, B. Contributions · p. 3
  • Meta-Immersion models QoE using both objective service quality and users’ subjective feelings, drawing on the Weber–Fechner Law.The paper also derives closed-form downlink data-rate and uplink bit-error-probability expressions for its xURLLC system.Source: B. Contributions · p. 3
  • 20.1% average QoE improvement is reported for attention-aware xURLLC allocation over conventional URLLC with uniform resource allocation; the improvement reaches 40% when total resources are limited.The evaluation uses UOAL attention values from 30 users and 96 objects.Source: B. Contributions · p. 3

A. User-Object-Attention Mechanism

  • Eye-movement records can provide user-object attention values that help allocate rendering resources for xURLLC.The related work describes eye tracking as a source of situational attention information.Source: A. User-Object-Attention Mechanism · p. 3

B. Virtual Reality

  • Prior gaze-based VR approaches predict gaze regions or adapt content in real time, whereas this paper motivates allocating rendering capacity according to attention to virtual objects.The cited discussion notes that gaze-based allocation requires real-time computing and may ignore users’ interest in objects.Source: B. Virtual Reality · p. 3

C. Contract Theory

  • Prior work applies contract theory in wireless networks, but MSP–InP contracts for wireless Metaverse services have not been fully investigated.The paper applies contract theory to this service-market setting and targets QoE through attention-aware resource allocation.Source: C. Contract Theory · p. 3

III. SYSTEM MODEL

  • The system model places a multi-antenna cloudlet-enabled base station in communication with edge rendering servers that receive and render virtual-object data for users.It considers wireless-environment factors including antennas, fading, transmit power, and co-channel interference.Source: A. Wireless Network Model · p. 4

B. Service Market Model

The service-market model links MSP payments to the InP’s resources and users’ QoE, then models the InP’s costs, service KPIs, and risk preferences.

1) Payment Plan:

  • The MSP–InP payment plan makes the InP’s revenue depend on the computing and communication resources supplied and the QoE of MSP users.The fixed payment F_s covers hardware infrastructure use, while u_M is the fee per unit QoE and M_i is user i’s quantified QoE.Source: 1) Payment Plan · pp. 4–5

2) Utility of the InP:

  • The InP invests in downlink power, bandwidth, uplink power, and rendering capacity, with a quadratic-form cost based on unit-resource costs.The resources affect objective KPIs and users’ subjective service experience.Source: 2) Utility of the InP · p. 5
  • The modeled service indicators include downlink data rate, uplink bit-error probability, and users’ subjective service experience.The text links high downlink rate to low transmission latency and low uplink error probability to accurate user-object interactions.Source: 2) Utility of the InP · p. 5
  • The InP’s utility accounts for risk aversion using constant-relative-risk-aversion preferences, with risk neutrality at τ = 0.The model considers 0 ≤ τ < 1 and uses a power utility function.Source: 2) Utility of the InP · p. 5

3) Utility of MSP:

The section frames MSP utility around users’ QoE and designs a contract that maximizes MSP utility while preserving incentives for the InP. It models Meta-Immersion from objective network KPIs and subjective experience, then motivates attention-aware rendering allocation.

3) Utility of MSP:

  • Under the same resource investment, higher user QoE raises both InP and MSP utility; users’ differing attention motivates assigning more rendering capacity to more interesting objects.The paper proposes predicting attention from sparse historical records to guide this allocation.Source: 3) Utility of MSP · p. 5, A. QoE Analysis for Metaverse xURLLC… · p. 6

4) Optimal Contract Design:

  • The MSP’s utility is maximized under a contract that must also satisfy the InP’s incentive-compatibility and participation constraints.The contract includes a fixed fee and a user-service fee; the InP’s utility is tied to maximizing user Meta-Immersion.Source: 4) Optimal Contract Design · pp. 5–6

A. QoE Analysis for Metaverse xURLLC Services

  • Meta-Immersion combines conventional network KPIs with users’ subjective service experience, modeled here through rendering quality and the number of objects seen.The model uses the Weber–Fechner Law, with rendering capacity as stimulus intensity and a connection coefficient determined by URLLC KPIs.Source: A. QoE Analysis for Metaverse xURLLC… · p. 6
  • The QoE model represents the user-object-attention value as a constant and uses a connection coefficient based on downlink data rate and uplink BEP.The transformation from stimulus to subjective feeling is described as influenced by interaction between the virtual and real worlds; the minimum threshold and maximum provider value bound the normalization.Source: A. QoE Analysis for Metaverse xURLLC… · p. 6
  • The QoE metric can accommodate service-specific KPI priorities by adjusting its connection coefficient, including a latency term for services with a delay limit.The example uses T(L_max − L_k), where L_max is the maximum delay the service can tolerate.Source: A. QoE Analysis for Metaverse xURLLC… · p. 6

B. Attention-Aware Rendering Capacity Allocation

The allocation scheme first predicts user-object attention from sparse historical records, then uses those estimates to guide rendering capacity. Its predictor uses a semantic factorization approach and coordinate-level updates to support online prediction.

1) User-Object-Attention Values Prediction:

  • The predictor maps users and objects into a joint semantic space and estimates their attention values as inner products of user and object feature vectors.Object labels can be grouped into semantic topics; user attention to those features and object-feature similarities yield the user-object-attention matrix.Source: 1) User-Object-Attention Values… · p. 7
  • The method fits predicted attention values with weighted regression and uses alternating least squares to update user and object semantic vectors.The regression uses a regularization strength to prevent over-fitting.Source: 1) User-Object-Attention Values… · p. 7
  • Element-level updates avoid matrix inversion and reduce the predictor’s iteration complexity to O(N_U N_O S^2).With precomputed terms, the stated complexity is further reduced to O(N_U N_O S), supporting online predictions.Source: 1) User-Object-Attention Values… · pp. 7–8
  • The predicted-attention allocation achieves QoE only 2% below the case where all user-object-attention values are perfectly known.The paper describes the predictor as simple and fast enough to support online user-object-attention prediction.Source: 1) User-Object-Attention Values… · p. 8

2) Optimal Rendering Capacity Allocation:

Given predicted user-object attention values, the scheme allocates rendering capacity to maximize Meta-Immersion. The paper presents an optimal allocation rule and an algorithm that converges efficiently.

2) Optimal Rendering Capacity Allocation:

  • The optimal rendering capacity allocation is designed to maximize Meta-Immersion using the predicted attention values for each object.The user’s total rendering capacity bounds the sum of the resolutions assigned to virtual objects.Source: 2) Optimal Rendering Capacity Allocation · p. 8
  • Algorithm 2 converges efficiently, with iterations tied to the number of objects assigned the minimum rendering capacity.The optimal rule’s parameter is obtained by solving the function specified in the proposition.Source: 2) Optimal Rendering Capacity Allocation · p. 8
  • Algorithms 1 and 2 together form the attention-aware allocation algorithm, enabling the InP to make better use of wireless MIMO resources for Metaverse xURLLC services.Source: 2) Optimal Rendering Capacity Allocation · p. 8

V. QOE ANALYSIS AND CONTRACT SOLUTION

The section derives URLLC KPI expressions to obtain a closed-form Meta-Immersion expression and support contract design, alongside an attention-based rendering-capacity allocation procedure. It also reports an insight about how antenna counts affect downlink interference resistance.

V. QOE ANALYSIS AND CONTRACT SOLUTION

  • The optimal rendering-capacity scheme takes predicted user-object-attention values and total capacity as input and returns an allocation, updating allocations using attention values and a recalculated μ*.Its procedure initializes temporary lists and reallocates capacity as objects are processed.Source: 1) Downlink Data Rate · p. 9

1) Downlink Data Rate:

  • Proposition 2 gives a closed-form expression for the downlink data rate R(D)_k, including the bandwidth B_k allocated to user k.The expression uses the Meijer’s G-function and Gamma function.Source: 1) Downlink Data Rate · p. 9
  • When co-channel interference is high, Λ_k approaches zero, and the downlink rate expression can be approximated using the nearest pole of the Meijer’s G-function.The passage describes this as an insight for URLLC system design.Source: 1) Downlink Data Rate · p. 9
  • Increasing the number of RS antennas improves URLLC anti-interference capability more than increasing the number of CBS antennas.Source: 1) Downlink Data Rate · p. 9

2) Uplink Bit Error Probability:

The section derives an uplink bit-error-probability expression and examines its behavior under high interference and increasing transmit power. It notes that short-packet transmission remains future work for approximating achievable xURLLC packet rates.

2) Uplink Bit Error Probability:

B. Convexity Analysis

The paper analyzes Meta-Immersion’s resource convexity, solves the provider contract numerically, and evaluates attention prediction and rendering allocation using UOAL. The attention-aware scheme improves MI over uniform allocation, while prediction errors are mostly 0 or 1.

B. Convexity Analysis

  • Meta-Immersion is linear in downlink bandwidth and concave in downlink transmit power, uplink transmit power, and rendering capacity.Source: B. Convexity Analysis · p. 10

C. Contract Solution

  • Using the KPI expressions and convexity analysis, the authors obtain a closed-form Meta-Immersion expression and solve the optimal contract numerically.For a given contract, Algorithms 1 and 2 and convex optimization tools yield resource allocations and corresponding Meta-Immersion values.Source: B. Convexity Analysis · p. 10, C. Contract Solution · p. 10

A. Metaverse Dataset Preparation

  • UOAL attention records are made sparse by grouping images into scenario options and having users randomly select options and view random numbers of images.The resulting sparse user-object-attention records are input to Algorithm 1.Source: A. Metaverse Dataset Preparation · p. 10

B. Performance Analysis

  • The performance analysis evaluates attention-value prediction, optimal rendering-capacity allocation, and optimal contract design.Source: B. Performance Analysis · p. 10

1) The accuracy of the attention-aware user-object-attention values prediction Algorithm 1 (for Q3):

  • With 41.8% of input attention-matrix entries empty, 62.8% of unknown-value prediction errors are 0, 34.1% are 1, and 3.16% are 2.Across all entries, the error values are 0 for 67.2%, 1 for 30.7%, and 2 for 2.12%.Source: 1) The accuracy of the attention-aware… · p. 10

2) The effectiveness of the optimal rendering capacity allocation Algorithm 2 (for Q3):

  • The allocation comparison uses random, uniform, attention-aware, and ground-truth upper-bound schemes, with resolution as rendering capacity and a 15 K minimum threshold per object.The total capacity per user is set to 20 K times the number of virtual objects.Source: 2) The effectiveness of the optimal… · pp. 10–11
  • The attention-aware rendering scheme achieves an average 20.1% MI improvement over uniform allocation, with a 25.5% maximum and 6.26% minimum; the theoretical upper bound has only 2% QoE improvement over the scheme.Source: 2) The effectiveness of the optimal… · p. 11

3) The superiority of the xURLLC over conventional URLLC (for Q2 and Q3):

The section compares attention-aware xURLLC with conventional URLLC and examines the MSP–InP contract. It reports QoE gains and explains how participation constraints shape feasible contracts, while identifying cold-start, cross-layer, and security directions.

3) The superiority of the xURLLC over conventional URLLC (for Q2 and Q3):

  • At high transmit power, conventional URLLC can achieve satisfactory KPIs, yet attention-blind rendering gives lower QoE than xURLLC.The reported KPIs are downlink data rate and uplink BEP; users and hardware also differ in wireless performance.Source: 3) The superiority of the xURLLC over… · p. 12
  • The attention-aware scheme improves MI more than uniform allocation, especially when total rendering capacity is small.The comparison uses the third user and 56 objects; the ground-truth attention values provide an upper bound.Source: 3) The superiority of the xURLLC over… · pp. 12–13, Figure 10 · p. 12

4) The effectiveness of the optimal contract design (for Q1): :

  • The optimal contract maximizes MSP utility under IC and IR, while the marked contract point gives the highest utility for both parties among feasible contracts.The InP’s resource choice and MI depend on uM, not fixed payment Fs; Fs affects InP utility and IR feasibility, so it cannot be zero.Source: 4) The effectiveness of the optimal… · p. 13

3) The superiority of the xURLLC over conventional URLLC (for Q2 and Q3):

  • 20.1% average MI improvement is reported for attention-aware xURLLC over conventional URLLC with uniform allocation.The conclusion attributes the gain to the UOAL-based attention-aware rendering-capacity scheme.Source: §VII · p. 13

VII. CONCLUSION AND FUTURE DIRECTIONS

  • The proposed MI metric considers objective network KPIs and users’ subjective feelings; its attention-aware xURLLC allocation predicts attention values first, then allocates resources optimally, increasing MI by 20.1% on average versus conventional URLLC uniform allocation.Source: §VII · p. 13
  • Cold-start remains unresolved for new users because the scheme predicts attention from sparse interaction records.Graph neural networks are proposed as a possible approach to prediction without collecting extra user information beforehand.Source: §VII · p. 13
  • The study focuses on attention-guided upper-layer allocation; cross-layer design with emerging access technologies remains a future direction.The paper gives dynamically adjusting beamforming accuracy to service content as an example.Source: §VII · p. 13
  • The modeled wireless service does not address eavesdropping, which the authors identify as a user privacy risk.They suggest physical-layer security, covert communication, and encryption as possible techniques.Source: §VII · pp. 13–14

APPENDIX A DATASET PREPARATION

The appendix describes how sparse user-object-attention records are generated and presents derivation steps for downlink data rate and uplink BEP expressions.

APPENDIX A DATASET PREPARATION

  • Sparse records are generated by randomly selecting scenario groups and reserving part of the data, then scoring object frequency and mapping scores into five levels.The algorithm uses randomly selected groups and a reserved data fraction, then locates objects in segmentation labels.Source: Appendix A · p. 14

APPENDIX B PROOF OF PROPOSITION 2

  • For the downlink data rate per Hertz, substituting (1) into (B-1), rewriting its logarithmic function using [57, eq. (01.04.07.0003.01)] and an integration path, and applying [36, eq. (9.301)] yields closed-form R(D).Source: Appendix B · p. 14

APPENDIX C PROOF OF PROPOSITION 3

  • The BEP proof applies distribution-function and integral identities to derive the closed-form expression and complete Proposition 3.The derivation invokes the Gauss hypergeometric function and evaluates intermediate integrals.Source: Appendix C · pp. 14–15

APPENDIX D PROOF OF PROPOSITION 1

The proof establishes convexity of the rendering-capacity optimization problem, then uses KKT conditions to obtain its optimality conditions.

APPENDIX D PROOF OF PROPOSITION 1

  • Convexity of the rendering-capacity objective and Slater’s condition establish strong duality for problem (26).The associated Lagrangian is then formed.Source: Appendix D · p. 15
  • The proof states the KKT optimality conditions for the optimal rendering-capacity solution.The multipliers are λn for each object and μ.Source: Appendix D · p. 15
  • The solution is derived through two cases, with μ* obtained by solving the resulting condition.Source: Appendix D · p. 15

APPENDIX E PROOF OF LEMMA 1

The appendix proves a convexity lemma and uses it to establish concavity properties needed for Proposition 4, including linear dependence on downlink bandwidth.

APPENDIX E PROOF OF LEMMA 1

  • The lemma uses a second-derivative test and Jensen’s inequality to establish convexity for the stated functions.Source: Appendix E · p. 15

APPENDIX F PROOF OF PROPOSITION 4

  • The objective is linear in downlink bandwidth.Source: Appendix F · p. 15
  • Using the lemma, the proof establishes concavity of the objective in uplink transmit power and downlink power.The supplied proof fragments identify the respective concavity results.Source: Appendix F · pp. 15–16
Loading 2208.05438v6…