Multi-Agent Q-Learning for Drone Base Stations
[ 1 ] Instytut Radiokomunikacji, Wydział Informatyki i Telekomunikacji, Politechnika Poznańska | [ P ] employee
2023
chapter in monograph / paper
english
- wireless communication
- energy consumption
- base stations
- Q-learning
- mobility models
- satellite broadcasting
- interference
EN We formulate a drone base stations (DBSs) localization problem that improves the users’ received signal-to-interference-plus-noise ratio (SINR) and the fairness between users in terms of receiving sufficient channel quality. In contrast to other works, our algorithm adapts to the actual users distribution on the ground without knowing their locations but rather their channel measurements history. We also leverage the fact that moving a DBS results in reduced aerodynamic energy consumption and illustrate that moving DBSs intelligently at a certain speed actually reduces energy consumption. We propose a multi-agent Q-learning formulation to solve this problem which requires less computation than its single agent counterpart and show by extensive simulations the improvements in terms of system fairness and link reliability relative to the benchmark solutions while leveraging a realistic users mobility model.
261 - 266
20
70