Energy-Efficient RIS-aided Uplink Cell-Free Massive MIMO-NOMA
Résumé
Cell-free massive MIMO (CF-mMIMO) promises uniform coverage for beyond-5G networks but faces critical energy efficiency (EE) challenges in dense and uncoordinated Internet-of-Things deployments. This paper addresses uncoordinated uplink transmission where devices transmit autonomously without coordination, potentially causing severe multi-user interference and leading to inefficient resource utilization. We propose a reconfigurable intelligent surface (RIS)-aided CF-mMIMO system with non-orthogonal multiple access that jointly optimizes subband allocation, RIS phase configuration, access point (AP) selection, and decoding order of successive interference cancellation to maximize EE while satisfying heterogeneous quality-of-service requirements. The mixed-integer nonconvex optimization problem is decomposed into three interconnected stages: lightweight device-driven subband selection via upper confidence bound bandits requiring minimal feedback, RIS phase optimization using sequential programming, and centralized deep reinforcement learning for dynamic AP selection that intelligently deactivates redundant links and enables AP sleep mode. A parallel training framework ensures generalization across diverse channel conditions and traffic types. Numerical results demonstrate substantial improvements over baseline schemes, achieving significantly higher EE and successful transmission rates, while reducing the number of active AP-device associations and enabling intelligent sleep-mode operation.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| licence |

