High-Speed Train Positioning Using Deep Kalman Filter With 5G NR Signals

Positioning is the most basic yet important process in a train control system. In practical train systems, the position of a train is determined with a track circuit, a radio frequency (RF) tag called a ``balise'', or a tachometer. A train control center gives each train an automatic train protection (ATP) speed profile computed from the positioning information of both the train and the next train in front it. To successfully operate in a conventional train control system such as ATP/ATO, each train derives its own automatic train operation (ATO) speed profile from the ATP speed profile provided by the control center. However, existing train positioning schemes face many difficulties in terms of installation, maintenance, and repair. To address these difficulties, we consider using 5G NR (New Radio) signals which have a high probability for guaranteeing line-of-sight (LOS) as well as a high sampling rate because they do not require the additional installation of any infrastructure for positioning. In this paper, we propose two positioning schemes for high speed trains (HSTs) based on Kalman filters that make use of 5G NR signals. The first is an HST positioning scheme using a modified Kalman filter and the second is an HST positioning scheme using a deep Kalman filter. Simulation results show that the two proposed schemes achieve better performance in nonlinear non-Gaussian systems as well as in nonlinear Gaussian systems in terms of mean and worst 5% position errors when compared to the existing positioning scheme for HSTs that utilize 5G NR signals.