Q-Learning-based setting of cell individual offset for handover of flying base stations

Flying base stations (FlyBSs) are widely used to improve coverage and/or quality of service for users in mobile networks. To ensure a seamless mobility of the FlyBSs among the static base stations (SBSs), an efficient handover mechanism is required. We focus on the handover of FlyBSs among SBSs and we dynamically adjust the cell individual offset (CIO) of the SBSs based on their load to increase the sum capacity of the users served by the FlyBSs while considering also a handover cost. Due to complexity of the defined problem and limited knowledge of other parameters required for conventional optimization methods, we adopt Q-learning to solve the problem. For Q-learning, we define a reward function reflecting the tradeoff between the capacity of users and the cost of performed handovers. The proposed Q-learning based approach converges promptly and increases the sum capacity of the users served by the FlyBSs by up to 23% for eight deployed FlyBSs comparing to state-of-the-art algorithms. At the same time, the number of handovers performed by the FlyBSs is notably reduced (up to 25%) by the proposal.