Model-Based Machine Learning for Energy-Efficient UAV Placement

Serving as aerial base stations, unmanned aerial vehicles (UAVs) can provide wireless coverage for the ground users. To achieve the goal of energy-efficient coverage, this paper proposes a model-based UAV placement algorithm. Specifically, the UAV coordinate is determined via machine learning based UAV-to-ground channel estimation. First, We establish a probabilistic UAV-to-ground transmission pathloss model with unknown channel parameters, which are estimated based on supervised learning. Among them, we collect the training data from the diverse coordinates of the UAV and the ground users and the corresponding pathloss feedback from the ground users. Then, the optimal UAV placed altitude is solved by gradient descent via the learned channel model to perfectly cover the served area with the minimum transmit power. Simulation results are presented to validate that the proposed model-based learning scheme can precisely estimate the channel parameters. Moreover, the proposed model-based UAV placement algorithm cover the served area with lower transmit power compared with model-free UAV placement algorithm.