User Tracking with Multipath Assisted Positioning-based Fingerprinting and Deep Learning

Abstract

Multipath assisted positioning schemes allow localizing a user with only a single physical transmitter by treating multipath components (MPCs) as line-of-sight signals from virtual transmitters. The user position and the locations of the physical and virtual transmitters can be estimated jointly with simultaneous localization and mapping (SLAM). While such approaches often show very good positioning performance, they come at the cost of a high computational complexity. To reduce this complexity, multipath assisted positioning schemes based on SLAM may be combined with fingerprinting, where the fingerprints are features of the wireless radio channel. Within this paper, we present such an approach, where a deep neural network (DNN) is trained on data from a multipath assisted positioning scheme to predict the user position and the corresponding uncertainty from channel information. Based on the DNN, a Kalman filter can accurately and efficiently track the user position. We show by simulations that the positioning performance is improved by a factor of 1.5 while the computational complexity is crucially lower than that of multipath assisted positioning-based SLAM.