R H I N O S Railway High Integrity Navigation Overlay System Assessing & Mitigation of risks on railways operational scenarios Rome, June 22 nd 2017 Anja Grosch, Ilaria Martini, Omar Garcia Crespillo (DLR) and Simon Roberts (University of Nottingham)
Overview Detection and Identification of Local Hazardous Errors in the Railway Environment Introduction of several error sources and their impact on GNSS measurements Typical methods of mitigation Obscuration mapping using 360o photography Investigation of Three Local Fault Detection Methods based on Consistency Checks Two snapshot algorithms using pseudorange residuals One sequential algorithm using EKF innovations Train Navigation Integrity Performance Minimum Detectable measurement Bias (MDB) RAIM Slopes, Probability of Misdetection and FDE availability gain
GNSS Threats Satellite, system and propagation errors can be can be partially mitigated, i.e., dual frequency methods, or externally monitored such as SBAS, GBAS, ARAIM ISM, 2tiers Augmentation Satellite Clock Ionospheric Delay Orbit Signal Def. Antenna PCV Free Space Local faults cannot be detected by any external facility detection needs to be done onboard the train Tropospheric Delay RX noise Multipath
Local GNSS Threats Propagation effects: Multipath (LoS+reflection) Phasor diagram θ c θ LoS θ MP Each reflection changes polarization Typical measurement error/offsets: Pseudorange: several meters Doppler Frequency: up to hundreds of Hz Phase: several centimeter LoS Reflected signal
Local GNSS Threats Propagation effects: Multipath (LoS+reflection) Non LoS reception (only reflections) No LoS information anymore All Measurements depend only on reflected paths LoS Reflected signal
Local GNSS Threats Propagation effects: Multipath (LoS+reflection) Non LoS reception (only reflections) Radio frequency inference: Jamming intentional or unintentional Complete loss of signal tracking No measurements LoS Reflected signal
Local GNSS Threats Propagation effects: Multipath (LoS+reflection) Non LoS reception (only reflections) Radio frequency inference: Jamming intentional or unintentional Spoofing LoS Reflected signal
Spoofing Effect Spoofer tries to take over the code and phase tracking of the GNSS receiver to control the computed PVT onboard
Local Fault Detection/Mitigation (I) Antenna-based techniques Antenna design, e.g., reflected signals are attenuated by ground plane and changed signal polarization Antenna arrays enable differentiation between LoS and reflection and employ spatial filtering by pre and post correlation techniques Requires dedicated and more complex antenna and Rx front-end 2x2 Array by DLR 7-Element Array by SATIMO (France) 7-element Conformal Array by DLR M. Sgammini et al, Blind Adaptive Beamformer Based on Orthogonal Projections for GNSS, ION GNSS, Sept. 2012, Nashville, TN, USA
Local Fault Detection/Mitigation (II) Signal Processing Techniques Vector tracking loops and inertial aiding of the tracking loops, to coast signal outages Detection of signal distortion within the correlator banks, e.g., narrow correlator, double-delta and strobe correlator techniques Requires more complex digital receiver part Example with 11 correlators: Channel estimation possible Matched filtering Mitigation of distortion Enhance LoS reception w.r.t. improved SNR
Local Fault Detection/Mitigation (III) Navigation-processor-based techniques Analyzing code, Doppler frequency and phase measurements of signals from the same satellite (e.g. CMC monitors or smoothed pseudoranges) Signal to noise ratio (SNR) of the signal (e.g. SNR depending weighting of the PRN) Dual receiver architectures: investigation of measurement differences Rx2 Exploiting external sensors, which are independent to GNSS local faults such as 3D map information, odometers, camera pictures and inertial sensors Measurement consistency check algorithms: Residual and innovation based fault detection schemes b Rx1
Obscuration mapping using 360 o photography
Obscuration mapping using 360 o photography
Signal Blockage G22 R03 R20 Platform roofs, buildings, trees, and, more in general, any part of elevated terrain, may reduce the sky visibility. G18 R13 G24 R02 G12 R12 G15 G28 R21 G17 G30 MITIGATION: Use of Multiple Constellations Gxx Visible GPS Satellite G12 G13 R22 Gxx Masked GPS Satellite R01 Rxx Visible GLONASS Satellite G20 G05 Rxx Masked GLONASS Satellite
Skyview obscuration & GNSS quality for selected sites on Nottingham testbed
In Progress Web-based GNSS ephemeris data with satellite elevation & azimuth data transmitted with NRTK corrections Knowledge of GNSS reliability and accuracy allows calibration of on-board sensors when these are high Obscuration maps used to predict when, where and for how long GNSS signals will be blocked and detect NLOS signal reception Calibrated sensor data used to enhance position during GNSS outages
Train Position Methods Snapshot Approaches - GNSS only Algorithm 1. Unconstrained GNSS with map matching (ucg) 2. Constrained GNSS (CG) Sequential Approach - Multi-sensor Algorithms 3. Total-State Extended Kalman Filter With 1D along track acceleration as control input (EKF)
Unconstrained GNSS with Map Matching State of the art GNSS WLS positioning: 3+l unknowns Applying map matching Along track position uncertainty
Constrained GNSS GNSS based solution directly within the track map: Reduction of unknowns (1+l): along track position and user clock offsets Along track position uncertainty Compared to unconstrained GNSS solution Less SV need thus higher availability and continuity Thanks to higher redundancy better accuracy
Residual based Fault Detection Test statistic: Weighted Sum Square Error (WSSE) of the pseudorange residuals r = ρ c ρ x q g = r T Wr q g ~χ 2 N m Define detection threshold T g for a given probability of false alert P FA = P(q g > T g H 0 ) and degrees of freedom N m P FA Fault monitor triggers q g > T g
Minimum Detectable Bias Under nominal conditions H0 Under non-nominal conditions H1 The Minimum Detectable Bias P MD
Total-state EKF x = s v b a b b T
Innovation based Fault Detection Test statistic: Normalized Innovation square (NIS) q NIS = γ T k S 1 k γ k q NIS ~χ 2 N γ k = z k h x k S k = H k P k H k T + R k Detection threshold T NIS for a given probability of false alert P FA = P(q NIS > T NIS H 0 ) and degrees of freedom N Under nominal conditions H0 q NIS H0 ~χ 2 N Under non-nominal conditions H1 q NIS H1 ~χ 2 N, λ 2 The Minimum Detectable Bias Fault monitor triggers q NIS > T NIS
Scenario Description A Railway Track in Brunswick Germany Scenario is based on measurement campaign B
MDB Comparison System Parameters: P FA = 5.6 x 10 8 P MD = 5.6 x 10 8 σ PR = 1m
Fault Detection Capability Detection Sensitivity RAIM Slopes = Ratio of fault impact on position to fault impact on test statistic Simulation parameters: Duration = 40 s Single fault on PRN 5 (worse case in terms of RAIM slopes) Fault Bias 30 m Fault Occurrence: @ Epoch 18, 28, 38
Monte Carlo Simulation Results
Summary Defined local threats, their affects and mitigation techniques in general Investigated the usage of 360 camera to extract obscuration maps and to identify NLOS signals Analyzed three suitable positioning techniques for simultaneous GNSS bias fault detection and exclusion Residual vs. Innovation based monitors Investigation of track discrimination in terms of accumulating track probabilities (sequential method to identifying current track reliably)
Future Steps Evaluation and analysis of the performance due to large number of measurements Extend DLR s Multi-sensor Simulator for railway scenarios Further investigation of additional fault detector such as sequential normalized innovation square (SNIS) to detect further fault profiles such as ramp like faults (ICNS 2017 paper) Further investigation of methods to detect and to cope with multi fault scenarios (ENC 2017 paper) Develop ARAIM based Protection Levels adopted to the railway environment Use the DLR s ARAIM demonstrator to investigate ARAIM protection levels
Thank you for your attention Any questions? Anja Grosch Anja.Grosch@DLR.de
Track Discrimination (I) Sequential algorithm for track identification based on maximum likelihood of measurement residuals Probability of track accumulates over time, identification is possible if track probability exceeds required reliability threshold Likelihood of measurements A priori probability Point of interest model evidence
Track Discrimination (III) For the given scenario and a track probability threshold of 1 P track 10 4 Track not identified Track identified