.png){kind=icon}
Return
Generative diffusion-driven AO framework for energy-efficient downlink STAR-RIS aided RSMA systems
.png){kind=icon}
PRE
AI
DOI：10.1016/j.comnet.2026.112650.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This paper investigates the energy efficiency (EE) optimization for downlink STAR-RIS-aided rate-splitting multiple access (RSMA) systems in IIoT networks. To decouple the non-convex interaction among beam-forming, rate allocation, and STAR-RIS configurations, we propose a hybrid framework, termed AO-ADAC (Alternating Optimization with Diffusion Actor-Critic). In the outer layer, a Dinkelbach-WMMSE solver ensures theoretical interpretability, while the inner layer introduces a generative diffusion-based agent to optimize high-dimensional RIS control variables. Unlike conventional policy-based methods, our diffusion Actor leverages a reverse denoising mechanism and self-attention modules to capture complex spatial dependencies among RIS elements and generate multimodal policy distributions, effectively escaping local optima. Specifically, the AO-ADAC algorithm alternately solves three subproblems, ensuring monotonic EE improvement and convergence to KKT-stationary points. Extensive simulations demonstrate that AO-ADAC significantly outperforms the traditional state-of-the-art (SOTA) iterative optimization (AO-SCA) and advanced DRL baselines (AO-SAC, AO-DDPG) in terms of EE, convergence stability, and scalability. Furthermore, the results verify that STAR-RIS achieves superior bidirectional coverage and power utilization compared to reflective-only RIS, validating the proposed diffusion-driven framework as an efficient paradigm for green and intelligent IIoT communications.
Keywords:
IIOT
RSMA
STAR-RIS
Generative diffusion model
Actor-Critic
Reinforcement learning
Journal
IF:
4.6
Papers:
1.7K
Citations:
1.6W
Organization
View more organizations
Cited Papers
No cited papers available