Ionospheric Integrity Lessons from the WIPP
Todd Walter
Stanford University
http://waas.stanford.edu
Todd Walter
Stanford University
http://waas.stanford.edu
History
Ionospheric Storms and Disturbances Originally Tested Via ScenariosSimulated disturbances added to
simulated ionosphereGenerally, large geographic features were
placed near center of network
Ionospheric Algorithm Originally Based on JPL GIM CodeTuned to work on scenariosLive data from WRSs not yet available
Ionospheric Models Provides Truth
Good for initial algorithm validation
Very SmoothAverage TECLoses small-scale
variationsSpatial gradients
smoothed as well
Useful Tool Before Data Was Available However Does Not Faithfully Represent
Real-World Instantaneous Ionosphere
Example Scenario
From “Ionospheric Specification for the Wide Area Augmentation System (WAAS) Simulation Studies” by Steve Chavin, ION GPS-96
dTEC/dt = 0.74 TECU/min
Gradient = 0.085 TECU/km
Shell Height = 360 km
Problem Algorithm Tuned to Work on Scenarios
All passed easily
Did not Work as Well on Real Data Required extensive retuning
Simulated Ionosphere Did Not Faithfully Reproduce Real Ionosphere Real disturbances worse than predicted Real slant-to-vertical errors better than predicted
Failing Scenario Could Prove Loss of Integrity However, Passing All Scenarios Would Not Demonstrate
Positive Integrity Worst-case scenario is algorithm dependent Does not demonstrate probability of missed detection
requirement is met
National Satellite Test-Bed
Early Prototyping Dual-Frequency
Survey Receivers Single Threaded Initiated in 1993 Full Deployment
Started in 1996
Prototyping Occurred During Solar Minimum No significant ionospheric disturbances observed
Caused Us to Become Overconfident Performance Dominated by Receiver Artifacts
Reasonability checks instituted to mitigate these errors Too aggressive, would remove much of solar max observed behavior
11 Year Solar Cycle Solar activity changes dramatically over an
11 year solar cycle Ionosphere at the peak is much worse than
at minimum Most disturbances at peak and declining
phase
NSTB
WIPP
WIPP
At the End of 1999 FAA Certification Required a Change in the Safety AnalysisLevel D code not considered reliableThreat models required for all monitorsRigorous accounting for monitor observability
Certification of Ionospheric Algorithms Left to Ionospheric Experts
Experts Created Threat Models From DataReliable threat not hypotheticalMust protect against worst observed conditionsMust overbound historical observationsMust have a demonstratable probability of missed detection
Supertruth Data 25 WRS - 3 Threads Each - Carrier Leveled - Biases
Removed - Voting to Remove Artifacts Clean Reliable Data Collected at the Peak of the Solar
Cycle Contained Worst Observed Gradients (Temporal and
Spatial over CONUS) Most Severely Disturbed Days Formed the Basis for Threat
Model Ionospheric Disturbances Are Deterministic, but Sampled
Randomly Worst cases are sampled over time
Will appear in the data as they move w.r.t IPPs
Apply Data Deprivation to Model Effects of Poor Observability
Ionospheric Measurements
Storm Example
Differences in Vertical Delay
Difference in Vertical Delay vs. IPP Separation Distance for Two Days:
Quiet Day July 2nd 2000
Disturbed Day July 15th 2000
CONUS Ionosphere Threat
Not Well-Modeled by Local Planar FitIonosphere well-sampled1
Ionosphere poorly sampled2
Ionosphere Changes Over the Lifetime of the Correction
User Interpolation Introduces Error
1. “Robust Detection of Ionospheric Irregularities,” Walter et al. ION GPS 2000
2. “The WAAS Ionospheric Threat Model,” Sparks et al. Beacon Symposium 2001
Well-Sampled Ionosphere
Chi-Square Metric Acts as “Storm-Detector”Test using small decorrelation value
Nominally ~ 35 cm one-sigma
Passing test accepts larger valueTypically ~ 85 cm one-sigma
Analytic ApproachDoes not require data except as validationFully specified before the first storm data of
April 2000
Under-Sampled IonospherePurely an Empirical Threat Model Worst Storm Data UsedIPPs Removed From Estimation to
Simulate Poor SamplingThree quadrant removal schemesStorm detector used on remaining data
Threat Based on Worst Deviation for Given Set of Metrics
Threat Region Is 5x5 Degree Cell Centered on IGP
CORS Data
Goal of Sigma Undersampled
To protect against an unfortunate sampling of the ionosphere such that we fail to detect an existing disturbance
Presumes the ionosphere is non-uniform near the IGP, i.e. it is divided into at least two states: a quiet one that is sampled, and a disturbed one that is not
Goal of Data Deprivation To divide the IPPs into two groups: one that
samples a relatively quiet ionosphere and does not trip the chi-square, the other that samples ionosphere not well-modeled from the quiet pointsData deprivation is used to simulate conditions that
were not actually experienced but may reasonably be experienced in the future
It allows us to investigate threats that occurred in well observed regions as though they had occurred in poorly observed regions
Want to have the quiet IPP distribution match those that may occur on operational system
Storm DaysOver the last 5 years, ~100 active days
have been identified~50 affected or would have affected
WAAS performance~45 supertruth files generated
Working on files for the lesser days
16 days affect our empirical threat model (serious effect)
All supertruth files available to international SBAS community
Temporal Threat Model
Also an Empirical Threat Model Worst Storm Data Used
Storm detector used
Look to See Largest Change With Respect to Planar Estimate
Overbound of Worst Rate of Change Ever Observed
Correlation with Spatial GradientErrors are currently double counted
No Storm Detector> 3 m / min> 6 m /2 min> 7 m over 5 min
With Storm Detector< 0.5 m / min< 1.25 m /2 min< 2.5 m over 10 min
Temporal Threat Model
Post IOC Storms
October 29-31 and November 20, 2003 Were Some of the Worst Storms Ever Observed
Conservatism in GIVE Calculation Protected UsersNo HMI observed at any location
None even close
However Worse Than PredictedStill much uncertainty in ionosphere
Conclusions
FAA Certification Required All Users Bounded Under All ConditionsIonospheric deviations are deterministicIonospheric deviations are observable
Threat Models Essential for Limiting Ionospheric Behavior
Each Monitor Must Account for the Limits of Its Observability
Approach is Very ConservativeWe are still learning!