Upload
others
View
5
Download
0
Embed Size (px)
Citation preview
Design and Evaluation of Integrity Algorithms for PPP in
Kinematic ApplicationsKazuma Gunning, Juan Blanch, Todd Walter, Stanford University;
Lance de Groot, Laura Norman, Hexagon Positioning Intelligence, Canada
Bottom line up front:
We use solution separation techniques developed for aviation combined with a PPP engine to produce meter-level protection levels for static, driving, and flight scenarios.
2
Solution Separation
Position Solution
Covariance
Nominal model-no faults are present
Protection Level?
PRN 1
PRN 2
PRN 3
PRN 4
PRN 5
3
Solution Separation
Faulted MeasurementClock and/or ephemeris
Large multipath
PRN 1
PRN 2
PRN 3
PRN 4
PRN 5
Producing fault-tolerant subsets
4
Solution Separation
PRN 1
PRN 2
PRN 3
PRN 4
PRN 5
Protection Level 5
Precise Point Positioning (PPP)• External precise orbit and clock corrections
• Kalman filter to estimate float carrier phase ambiguities, tropospheric delay
• Able to achieve sub-decimeter accuracy after convergence
• Globally available
6User Receiver
𝚽𝒊𝒇(𝒊)
= 𝒙𝒔(𝒊)− ෝ𝒙𝒓𝒙 + 𝒄 𝒃𝒓𝒙,𝒄 − 𝒃𝒔
(𝒊)+ 𝑻(𝒊) + 𝒄𝜹𝒕𝒓𝒆𝒍 + 𝒃𝒑𝒘𝒖
𝒊 − 𝑨𝒊 + 𝚫𝐫𝐫𝐜𝐯𝐫𝒊
+ ො𝝐 𝒊 + 𝝐 +⋯
IGS sat position
Rx position
IGS sat clock
Rx clock tropodelay
relativistic effects
phase wind-up
carrier phase
ambiguity
site displacement
effects
PPP Algorithm
Estimated errors
Measurement noise
• General strategy: • Use precise external inputs and model as many effects as possible,
and estimate the rest
Estimated error could include multipath, orbit and clock error, etc.
PPP Algorithm
Kalman filter state vector (real-time sequential estimator)
𝑠𝑡𝑎𝑡𝑒 = 𝑥𝑟𝑥 , 𝑦𝑟𝑥 , 𝑧𝑟𝑥 , 𝑏𝑢,1,… , 𝑏𝑢,𝑛, Δ𝑇, 𝐴1, … , 𝐴𝑚, 𝜖1, … , 𝜖𝑚′
Measurements:
𝚽𝒊𝒇(𝒊)
= 𝒙𝒔(𝒊)− ෝ𝒙𝒓𝒙 + 𝒄𝒃𝒓𝒙,𝒄 +𝒎(𝒊)𝚫𝑻 − 𝑨𝒊 + 𝒐𝒕𝒉𝒆𝒓 𝒎𝒐𝒅𝒆𝒍𝒔
…
𝝆𝒊𝒇(𝒊)
= 𝒙𝒔(𝒊)− ෝ𝒙𝒓𝒙 + 𝒄𝒃𝒓𝒙,𝒄 +𝒎(𝒊)𝚫𝑻 + 𝒐𝒕𝒉𝒆𝒓 𝒎𝒐𝒅𝒆𝒍𝒔
…
Given measurement noise characteristics, process noise, etc., estimate!
Subsets and Parallel Filters
12345
PRN
12345
PRN
12345
PRN
…
All-in-View
Subset 1
Subset 5
Solution separation requires multiple Kalman filters running in parallel
Steps have been taken to reduce computational complexity
Simplified error modeling to reduce statesSharing computations across subsets
t = 1 2 3 4 5 6 7 9
Ingest available measurements
Subset management
New subset based on new measurements
State management
Time update
Range Modeling(non-estimated)
f(x(0),t)
Measurement update and
residual checks (measurement
exclusion)
Subset with faulted PRN
Protection level computation and potential signal
exclusion
1 computation per subset
1 computation total-significant time savings
SS PPP Implementation
10
Protection Level Algorithms
PL Algorithm Speed Fault Detector
Separation Based- Basic Fast No
Cov. Based Exact Search Too Slow Yes
Cov. Based Approx. Fast Yes
Cov. Based Approx. Coarse Fast Yes
11
Dataset overview
Stanford, California, USA
1. Static rooftopNominal conditions
Injected step error
Injected ramp error
Calgary, Alberta, CA
2. Open sky drivingNominal ConditionsInjected ramp error
3. Suburban drivingAtlantic City, New Jersey, USA
4. Flight
All runs post-processed
MGEX precise clock and ephemeris
PHMI = 10-7
12
• Stanford Aero/Astro department rooftop receiver STFU (IGS MGEX Network)
• Trimble NetR9
• 1 hour of static data on November 7, 2017
• 1 Hz GPS (L1C-L2P semi-codeless), GLONASS (L1C-L2P)
• Truth position from IGS station solution
• Processed in dynamic mode
Dataset 1- Stanford rooftop
13
Step fault injection• Step error of 20 meters added to precise clock of GPS PRN 8 five
minutes into the run
• Bad measurements are caught in the residual check
14
Ramp fault injection• Ramp error injected into PRN 8 precise clock- 9 meters per hour
• Slow enough that much of the ramp is pulled into the error states associated with PRN 8.
• Once the PL threshold is tripped, all PRN 8 measurements are excluded henceforth, and the filter is reset
Subset with PRN 8 excluded becomes new all-in-view solution
15
• Just outside Calgary International Airport
• Benign, open sky environment
• 1 Hour Driving on March 1, 2018• Seven laps
• Receiver: NovAtel OEM 7500
• 1 Hz GPS (L1C-L2P semi-codeless), GLONASS (L1C-L2P)
• Truth: Novatel OEM729 with tactical-grade IMU
Dataset 2 - Open Sky Driving
16
Open sky driving bounding and solution error
17
18
Dataset 3- suburban driving• Suburban Calgary
• Largely benign environment with occasional full measurement outages
• 1 Hour Driving on March 1, 2018
• 1 Hz GPS (L1C-L2P semi-codeless), GLONASS (L1C-L2P)
• Truth: Novatel OEM729 with tactical-grade IMU
19
Suburban driving bounding and solution error
Overpass
Heavy foliage
Tall buildings
20
Dataset 4- flight• FAA Global 5000 Aircraft
• 1 Hz GPS (L1C-L2P semi-codeless), GLONASS (L1C-L2P)
• Trimble BD935
• 2 Hours of Flight Data on 6-30-2017
• Truth: NRCan PPP
21
22
Flight bounding and solution error
Normalized Position Error Statistics
23
Thank you to Hexagon Positioning Intelligence for partneringThank you to FAA for flight dataThank you to IGS for Precise Products
Solution separation techniques have been used to produce meter-level protection levels for automobile and aviation scenarios
Min. Hor. PL Median PL
Static 1.70 m 2.13 m
Car 1 2.05 m 2.76 m
Car 2 2.86 m 3.84 m
Flight 1.08 m 1.54 m
24