Smart Sampling for Ultra-Wideband Nonparametric Belief Propagation Indoor Localization

Abstract:
We propose a novel sampling distribution for message multiplication in nonparametric belief propagation which draws samples smartly, i.e. samples reside in regions where the product of messages has significant probability mass. This inherent property of the sampling distribution allows for a significant reduction in the number of samples used for message multiplication without impairing localization accuracy notably. Reducing the number of samples, in turn, enables a considerable reduction in terms of computational complexity. The sampling distribution arises under realistic assumptions on the indoor ultra-wideband radio channel. Through simulations, we show that the proposed sampling distribution enables reduced complexity and results in faster convergence when compared to typical sampling distributions from literature.