Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Generative AI-Aided QoE Maximization for RIS-Assisted Digital Twin Interaction

Jiayuan Chen, Yuxiang Li, Changyan Yi, Shimin Gong

TL;DR

This work addresses QoE-aware resource allocation for RIS-assisted digital twin (DT) interactions in the presence of uncertain evolution, formulating a QoE objective across scene-specific problems. It introduces PG-ZFO, a framework that combines a prompt-guided decision transformer with a zero-forcing optimizer to jointly determine RIS phase shifts, beamforming, rendering resolutions, and computing resources across evolving DT scenes, while modeling each scene as an MDP and using returns-to-go in prompts. The approach leverages scene-specific prompts to generalize to unseen scenes without retraining and employs a ZF-based layer to efficiently compute transmit and receive beamformers, achieving superior QoE compared to baselines. Simulations show that PG-ZFO converges well, generalizes to new scenes, and yields notable QoE gains as DL latency decreases with higher transmit power, indicating strong practical potential for RIS-enabled DT systems in dynamic environments.

Abstract

In this paper, we investigate a quality of experience (QoE)-aware resource allocation problem for reconfigurable intelligent surface (RIS)-assisted digital twin (DT) interaction with uncertain evolution. In the considered system, mobile users are expected to interact with a DT model maintained on a DT server that is deployed on a base station, via effective uplink and downlink channels assisted by an RIS. Our goal is to maximize the sum of all mobile users' joint subjective and objective QoE in DT interactions across various DT scenes, by jointly optimizing phase shift matrix, receive/transmit beamforming matrix, rendering resolution configuration and computing resource allocation. While solving this problem is challenging mainly due to the uncertain evolution of the DT model, which leads to multiple scene-specific problems, and require us to constantly re-solve each of them whenever DT model evolves. To this end, leveraging the dynamic optimization capabilities of decision transformers and the generalization strengths of generative artificial intelligence (GAI), we propose a novel GAI-aided approach, called the prompt-guided decision transformer integrated with zero-forcing optimization (PG-ZFO). Simulations are conducted to evaluate the proposed PG-ZFO, demonstrating its effectiveness and superiority over counterparts.

Generative AI-Aided QoE Maximization for RIS-Assisted Digital Twin Interaction

TL;DR

This work addresses QoE-aware resource allocation for RIS-assisted digital twin (DT) interactions in the presence of uncertain evolution, formulating a QoE objective across scene-specific problems. It introduces PG-ZFO, a framework that combines a prompt-guided decision transformer with a zero-forcing optimizer to jointly determine RIS phase shifts, beamforming, rendering resolutions, and computing resources across evolving DT scenes, while modeling each scene as an MDP and using returns-to-go in prompts. The approach leverages scene-specific prompts to generalize to unseen scenes without retraining and employs a ZF-based layer to efficiently compute transmit and receive beamformers, achieving superior QoE compared to baselines. Simulations show that PG-ZFO converges well, generalizes to new scenes, and yields notable QoE gains as DL latency decreases with higher transmit power, indicating strong practical potential for RIS-enabled DT systems in dynamic environments.

Abstract

In this paper, we investigate a quality of experience (QoE)-aware resource allocation problem for reconfigurable intelligent surface (RIS)-assisted digital twin (DT) interaction with uncertain evolution. In the considered system, mobile users are expected to interact with a DT model maintained on a DT server that is deployed on a base station, via effective uplink and downlink channels assisted by an RIS. Our goal is to maximize the sum of all mobile users' joint subjective and objective QoE in DT interactions across various DT scenes, by jointly optimizing phase shift matrix, receive/transmit beamforming matrix, rendering resolution configuration and computing resource allocation. While solving this problem is challenging mainly due to the uncertain evolution of the DT model, which leads to multiple scene-specific problems, and require us to constantly re-solve each of them whenever DT model evolves. To this end, leveraging the dynamic optimization capabilities of decision transformers and the generalization strengths of generative artificial intelligence (GAI), we propose a novel GAI-aided approach, called the prompt-guided decision transformer integrated with zero-forcing optimization (PG-ZFO). Simulations are conducted to evaluate the proposed PG-ZFO, demonstrating its effectiveness and superiority over counterparts.

Paper Structure

This paper contains 20 sections, 4 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. System Model and Problem Formulation
  3. System Overview
  4. Uplink and Downlink Transmissions in DT Interaction
  5. QoE Model Analysis for DT Interaction
  6. Problem Formulation
  7. Prompt-Guided Decision Transformer Integrated with Zero-Forcing Optimization
  8. Problem Reformulation
  9. Prompt-Guided Decision Transformer in PG-ZFO
  10. Design of Prompt
  11. Structure of Prompt-Guided Decision Transformer
  12. ZF-based Optimization Algorithm Integrated in PG-ZFO
  13. Implementation of PG-ZFO
  14. Offline Training
  15. Online Execution
  16. ...and 5 more sections

Figures (4)

  • Figure 1: An illustration of the considered RIS-assisted DT interaction system.
  • Figure 2: Convergence of the proposed PG-ZFO approach.
  • Figure 3: Illustration of the sum of all mobile users' QoE of DT interaction across all time slots under various DT scenes within $\mathcal{I}$.
  • Figure 4: Comparison of the sum of all mobile users' QoE of DT interaction across all time slots w.r.t. maximum transmit power.