Energy-Efficient and QoS Guaranteed BBU Aggregation in CRAN Based on Heuristic- Assisted Deep Reinforcement Learning

• DOI: 10.1109/jlt.2021.3120874
• Publication Date: 2021-10-20T18:23:07Z
• Authors: Min Zhu; Jiahua Gu; Tianyu Shen; Chenglin Shi; Xueqi Ren
• Keywords: Sleep mode

The surging mobile traffic poses serious challenges for operators, one of which is the unsustainable growth caused by high energy consumption massively deployed base stations (BSs). Cloud radio access network (CRAN), as a new architecture, proposed to confront this challenge. By isolating baseband unit (BBU) from its remote head (RRH) in BS, BBUs are consolidated into common place (i.e., BBU pool). Since "any-to-any" connection between and RRHs realized CRAN, low utilized can be switched sleep mode save during valley, effectively reduce CRAN. However, when enters mode, connected with must another BBU, would degrade quality service (QoS) RRHs. In paper, simultaneously guarantee RRH migration, both interrelated mutual restraint, we propose deep reinforcement learning (DRL) based aggregation scheme. Furthermore, train DRL fast well, introduce several heuristic algorithms assist training. Extensive numerical evaluations show that our heuristic-assisted (HA-DRL) power less migration. When compared baselines, HA-DRL achieve up 18.3% cost reduction 32.8% migrated at most 8.4% higher attains lowest all cases considered.