Distributed Channel Prediction for Multi-Agent Systems

Multi-agent systems (MAS) communicate over a wireless network to coordinate their actions and to report their mission status. Connectivity and system-level performance can be improved by channel gain prediction. We present a distributed Gaussian process regression (GPR) framework for channel prediction in terms of the received power in MAS. The framework combines a Bayesian committee machine with an average consensus scheme, thus distributing not only the memory, but also computational and communication loads. Through Monte Carlo simulations, we demonstrate the performance of the proposed GPR.