Multi-Agent Deep Q-Learning for Antenna Tilt Optimization in Wireless Networks

Abstract

The configuration of an antenna installed at a base station involves the quality of communication in wireless networks. For example, at each transmitter, the antenna tilt must be optimized such that the desired and undesired receivers obtain the highest and lowest signal strength, respectively. In this work, we propose to use multi-agent deep Q-learning to optimize the antenna tilt. Our channel model includes the three-dimensional antenna gain, the Ericsson path loss model, and the digital elevation model (DEM). Our simulation indicates that multiagant deep Q-learning provides good signal quality.