Abstract
Immersive communication in sixth-generation mobile networks (6G) demands ultra-high data rates, seamless connectivity, and consistent service quality, posing stringent requirements on mobility management in dynamic environments. Fully-decoupled radio access network (FD-RAN), as a promising 6G architecture, supports decoupled uplink/downlink access and multi-base-station cooperation, enabling uplink-intensive multi-connectivity mobility management. Realizing this capability in FD-RAN entails joint optimization of uplink base station (UBS) selection and user equipment (UE) power control, which is inherently challenging due to their mixed-variable nature. In this paper, we investigate the uplink mobility management problem in FD-RAN by formulating it as a mixed-variable optimization that minimizes the mismatch between network serving capability and dynamic QoS demands. To solve this efficiently, we propose a hybrid-action-space deep reinforcement learning (HDRL) framework and further design a dimension-reduced discrete action space to improve training efficiency. Based on the HDRL framework, we develop a hybrid proximal policy optimization (HPPO) algorithm with two independent sub-actors for discrete UBS association and continuous power control, jointly trained via a shared global critic. Simulation results validate the effectiveness of the HPPO algorithm in enabling seamless and demand-aware uplink services for mobile UEs.
Funding Information
  • National Natural Science Foundation of China (62271244)
  • Jiangsu Province Innovation and Entrepreneurship Team Project (JSSCTD202202)

This publication has 37 references indexed in Scilit: