Ensemble Learning Method-Based Slice Admission Control for Adaptive RAN

The future 5G system is expected to support an additional diverse range of services with performance targets of low latency, high reliability, and high capacity. We have proposed an adaptive radio access network (RAN) system to support communication services called the advanced 5G system. The RAN slice is defined as a logical network able to provide a variety of services separately in a RAN. In adaptive RAN, an efficient slice admission control (SAC) scheme for RAN slices is required. Recently, research on applying a machine learning (ML) to facilitate efficient slice resource management and SAC in 5G RAN slicing environments has been discussed. However, the training time for the existing SAC scheme based on the Deep Reinforcement Learning Framework (DRLF) takes too long to achieve an acceptable level of performance. Since it takes a long time to learn, it is difficult to guarantee sufficient performance in the early phase of the operating period. In this paper, we propose a SAC based on an ensemble learning method (ELM) to reduce learning time and improve performance for the adaptive RAN. Our evaluation with ELM for the improved utilization of resource blocks (RBs) shows that the proposed approach outperforms the other approaches.