Performance analysis of manifold learning methods for radio map construction

摘要

Wifi fingerprinting is an appealing method for indoor positioning as it can utilizes available wireless infras-tructures while retaining an acceptable level of accuracy. For the method to work, radio map of target environment has to be constructed in advance. The process is laborious and time consuming as signal strength has to be collected from mobile devices at a large number of reference locations. An attractive idea is to harness unlabeled wireless signal data from crowd to help construct the radio map. Various manifold learning methods have been employed for this purpose. However, results are not comparable as they are highly dependent on experiment environment. In this works, we implement four different manifold learning methods for the radio map construction task and report the localization performance of the resulting map on a simulated data and a public real dataset.