Navigation filter (RWTH/IRT)

GNSS Receiver for Railway Applications

In order to compensate varying influences and disturbances in the area of rail networks, for example, shadowing and multipath effects through urban canyons, it is necessary to develop aided navigation systems, which combine GNSS receiver, inertial sensors and possibly additional sensors to give resilient and integer localization. The simplest method of such a combination is represented by a loosely-coupled system, where a GNSS-only positioning solution is pre-calculated by the GNSS receiver. This method is dependent on the availability of at least four satellites.

Tightly-coupled filter: Since the availability of at least four satellites cannot be guaranteed under all environmental conditions, especially in forested or urban areas, more advanced methods of sensor aiding are necessary. Therefore, tightly-coupled filters allow aiding in the event of less than four satellites due to direct processing of raw GNSS measurements inside the filter. An overview of the structure of the tightly-coupled localization filter is given in figure 1.


Figure 1: Overview of filter structure

The main parts consist of GNSS measurement pre-processing, initialization algorithm, strapdown propagation and error-state extended Kalman filtering (EKF) for the measurement update. For prediction of the states at a constant rate of 100 Hz, the well-known strapdown approach is used, which is implemented by integrating the IMU measurements using a fourth-order Runge-Kutta integrator, [1]. In the GNSS measurement pre-processing, the required values such as satellite position and velocity, the pseudorange and Doppler observables and the signal-to-noise ratio for all available satellites can be calculated using the raw GNSS measurements. The EKF uses afterwards GNSS measurements to estimate the error of the current propagated state vector and updates the estimated state covariance matrix step to provide an increase in accuracy.

Experimental setup: In order to show the real-time capability of the proposed localization approach and its robustness, experiments are performed. Appropriate measuring equipment was installed into a car. Inertial measurements were performed using a LORD MicroStrain 3DM-GX3-25 with 100 Hz clock speed. GNSS raw data were supplied by a Septentrio AsteRx3 receiver. The Septentrio receiver was configured for the tests in such a way that raw data was displayed only by GPS L1 and L2C. The frequency was 10 Hz. Furthermore, smoothing functionalities, the detection of multipath propagation and clock correction were deactivated. The geodetic GNSS receiver Leica Viva GS10 was used as reference system and was supplied permanently with RTK correction data via an integrated GSM modem. Therefore, the standard deviation of the reference data, determined by the Leica machine, was among 2 and 8 centimeters during the overall test drive.

A detailed description of the used algorithms as well as detailed analyses can be found in [2].

Figure 2 shows the test results of the test drive and the developed navigation filter at different basic settings. The basic setting (red) uses GPS L1 and GPS L2C and so removes the current ionosphere error. The correction data to model the impact of the troposphere are also used. The average 2D error compared to the RTK reference (blue) is 0.75 meters. As soon as only one GNSS frequency is processed (green) and global ionospheric parameters need to be considered (sent together with the ephemerides, valid for one day and the entire earth), the average 2D error increases to 3.62 meters. The deactivation of the tropospheric model together with the use of L1 and L2C (orange) has a similarly serious impact: the 2D error is 2.34 meters compared to the reference.


Figure 2: Different configurations compared to reference (blue): L1/L2 and tropospheric modelling (red), only L1 and tropospheric modelling (green), L1/L2 without tropospheric modelling (orange). Map data: GeoBasis-DE/BKG, Google

Figure 3: In order to validate the robustness of the final filter, the scenario with sections of partial and full losses of GNSS signal reception is tested. Compared to Standard Code-Based Positioning (SPS), the proposed navigation filter (EKF) shows significantly better results. Sections with poor signal reception, caused by multipath and obstructions, are smoothly filtered. Moreover, complete outages in short tunnels are bridged by propagation of the IMU measurements.


Figure 3: 2nd scenario – Comparison between Standard Positioning Service (SPS, green), EKF (red) and reference (blue) in bad signal conditions: left detail: still stand, trees and near building, right detail: complete loss caused by short tunnel. Map data: GeoBasis-DE/BKG, Google



[1] J. Farrell: Aided navigation: GPS with high rate sensors. McGraw-Hill, Inc., 2008.
[2] M. Breuer, T. Konrad, D. Abel: High Precision Localisation in Customised GNSS Receiver for Railway Applications. In: Proceedings of ION GNSS+ 2016 Conference, Portland, Oregon, USA, 09/2016.