A Reinforcement Learning Approach to Energy Efficiency and QoS in 5G Wireless Networks

Abstract

Satisfying the huge demand for high bandwidth in 5G networks is in part achieved by vertical densification of the network infrastructure with so-called small-cell base stations. As a direct consequence, the performance of 5G networks intrinsically suffers from: 1) a huge inter-cell interference due to the network density and 2) a large energy waste that results from the high redundancy of lightly loaded, always-on, and small-cell base stations. In this paper, we use a game theoretic approach to design a distributed energy efficient bandwidth sharing mechanism for small-cell networks. To circumvent the combinatorial nature of searching in the huge strategy space, we invoke a reinforcement learning approach to intelligently guide the search toward good solutions. More precisely, our solution relies on dynamically learning good strategies for the user-equipment association and orthogonal frequency division multiple access (OFDMA) scheduling to strike a balance between energy efficiency and high bandwidth utilization. Our model takes into account inter-cell interference and the diversity of user quality of service requirements.