Adaptive Digital Twin-Assisted 3C Management for QoE-Driven MSVS: A GAI-Based DRL Approach

Abstract

Communication, computing, and buffer control (3C) management is essential to enhance quality-of-experience (QoE) in multicast short video streaming (MSVS). The existing 3C management schemes mainly rely on static data processing methods and a general QoE model, which may not efficiently improve QoE when users’ swipe behaviors exhibit distinct spatiotemporal differences. In this paper, we propose an adaptive digital twin (DT)-assisted 3C management scheme to enhance QoE in MSVS. Particularly, DTs consist of user status data and data-based models, which can update multicast groups and abstract users’ swipe features. An adaptive DT management mechanism is developed to adapt to users’ swipe behavior dynamics. Then, a fine-grained QoE model is established by considering the impact of resource constraints and DT model accuracy, leading to accurate buffer control. Finally, a joint optimization problem of 3C management is formulated to maximize long-term QoE. To efficiently solve this problem, a diffusion-based deep reinforcement learning (DRL) algorithm is proposed, which utilizes the denoising technique to improve the action exploration capabilities of DRL. Simulation results based on a real-world dataset demonstrate that the proposed DT-assisted 3C management scheme outperforms benchmark schemes in terms of QoE, with improvements of 18.4% and 20.5% under low and high user dynamics, respectively.