Learning-Based Fusion of Multipath Assisted Positioning and Fingerprinting

Kurzfassung

In multipath assisted positioning, multipath components (MPCs) are regarded as line-of-sight (LoS) signals from virtual transmitters. The locations of the physical and the virtual transmitters can be estimated jointly with the user position using simultaneous localization and mapping (SLAM). We have previously introduced such an approach called cooperative Channel-SLAM, where multiple users cooperatively estimate the locations of physical and virtual transmitters. Such schemes typically suffer from a high computational complexity due to expensive signal processing, though. Within this paper, we propose a novel approach that combines multipath assisted positioning with fingerprinting. In the first stage, multiple users estimate their own locations with cooperative Channel-SLAM. With the channel estimates and the estimated user positions from cooperative Channel-SLAM, a deep neural network (DNN) is trained. In the second stage, users can localize themselves making use of the DNN. In our novel approach, the positioning error is in the same order of magnitude as for cooperative Channel-SLAM, while the computational complexity is reduced drastically.