Learning radio maps for UAV-aided wireless networks: A segmented regression approach

Abstract

This paper targets the promising area of unmanned aerial vehicle (UAV)-assisted wireless networking, by which communication-enabled robots operate as flying wireless relays to help fill coverage or capacity gaps in the networks. In order to feed the UAV's autonomous path planning and positioning algorithm, a radio map is exploited, which must be, in practice, reconstructed from UAV-based measurements from a limited subset of locations. Unlike existing methods that ignore the segmented propagation structure of the radio map, this paper proposes a machine learning approach to reconstruct a finely structured map by exploiting both segmentation and signal strength models. A data clustering and parameter estimation problem is formulated using a maximum likelihood approach, and solved by an iterative clustering and regression algorithm. Numerical results demonstrate significant performance advantage in radio map reconstruction as compared to the baseline.