Channel Prediction with Location Uncertainty for Ad-Hoc Networks

Artikel i vetenskaplig tidskrift, 2018

Multi-agent systems (MAS) rely on positioning technologies to determine their physical location, and on wireless communication technologies to exchange information. Both positioning and communication are affected by uncertainties, which should be accounted for. This paper considers an agent placement problem to optimize end-to-end communication quality in a MAS, in the presence of uncertainties. Using Gaussian processes (GPs), operating on the input space of location distributions, we are able to model, learn, and predict the wireless channel. Predictions, in the form of distributions, are fed into the communication optimization problems. This approach inherently avoids regions of the workspace with high position uncertainty and leads to better average communication performance. We illustrate the benefits of our approach via extensive simulations, based on real wireless channel measurements. Finally, we demonstrate the improved channel learning and prediction performance, as well as the increased robustness in agent placement.