Machine Learning Based Node Selection for UWB Network Localization

In location-aware networks, only a subset of nodes provides representative measurements for position inference. Therefore, efficient high-accuracy localization calls for strategies to select an appropriate subset of active nodes. While node selection strategies benefit efficient localization, determining an optimal subset of active nodes relies on knowledge of channel state information whose acquisition overhead can be prohibitive. This paper presents a probabilistic node selection strategy for ultra-wideband network localization based on machine learning. We formulate the node selection problem as a classification task given a position estimate and determine near-optimal access probabilities from training data obtained via model-based optimization. A case study in a 3rd Generation Partnership Project scenario validates the proposed strategy and compares it against uniformly distributed random node selection.