Mixture Density Networks for Multipath Assisted Positioning-based Fingerprinting

Kurzfassung

In multipath assisted positioning schemes, the spatial information contained in multipath propagation of wireless radio systems is exploited for localization of a receiver. However, such schemes suffer from a high computational complexity. We have proposed before a fingerprinting localization system based on multipath assisted positioning, where the fingerprinting database is encoded in a deep neural network (DNN). Within this paper, we propose and evaluate a mixture density network approach in our DNN to analyze ambiguities among fingerprints at different locations. We show that our scheme shows a very good positioning performance with an error of around 2m for the most part, while having a low computational complexity in the online stage and a very low effort compared to traditional fingerprinting schemes.