MADRL Based Scheduling for 5G and Beyond

Abstract

Scheduling in cellular networks plays a critical role and is a key differentiating factor of network performance. The design of scheduling algorithms is challenging since it has to be both computationally efficient to meet the real-time Transmission Time Interval (TTI) requirements and robust to inaccurate/coarse channel-feedback information. Addressing the aforementioned challenges, this paper presents a novel multi-agent deep reinforcement learning (MADRL) based scheduling strategy. The simulation results, under the setting of Channel Quality Indicator (CQI) feedback, show that the proposed method outperforms conventional scheduling in different variants of Proportional Fair (PF) scheduling policies with low computational time.