Efficient Algorithms for Air-to-Ground Channel Reconstruction in UAV-Aided Communications

This paper develops an efficient algorithm to learn and reconstruct from a small measurement samples an air-to-ground radio map with fine- grained propagation details so as to predict the signal strength between a wireless equipped UAV and arbitrary ground users, and ultimately the optimal position of the UAV as a mobile relay. In this paper, a joint data clustering and parameter estimation algorithm is developed to learn an multi-segment propagation model from energy measurements that may contain large observation noise. To reduce the reconstruction complexity, we propose to learn a hidden multi-class virtual obstacle model to help efficiently predict the air-to-ground channel. Numerical results demonstrate that the channel prediction error is significantly reduced, and meanwhile, the radio map reconstruction time is reduced to 1/300.