Simple, On-Device Positioning using Compressed Fingerprint Archives

This technical report describes (a) a method for efficient on-device positioning that is based on the radio frequency fingerprint model, (b) a method for compressing the fingerprints that are used with this positioning method and, (c) a method for estimating the physical distance between two or more devices. Our positioning method, called Weighted Gaussian Distance, achieves similar positioning accuracy to the state-of-the-art Bayesian method, but at lower compute costs. Our archival method for compressing large numbers of fingerprints allows large physical areas to be stored compactly on a mobile device, so that offline positioning is possible in most circumstances. The method reduces the size of a typical fingerprint archive by 1-2 orders-of-magnitude, while keeping accuracy roughly equivalent. Both methods can be used in survey-based and organic localization environments.