Linear Combinations of GNSS Phase Observables to Improve and Assess TEC Estimation Precision Brian Breitsch Advisor: Jade Morton Committee: Charles Rino, Anton Betten
1
Background and Motivation Linear Estimation of GNSS Parameters TEC Estimate Error Residuals Application to Real GPS Data
2
Earth's Ionosphere
3 J. Grobowsky / NASA GSFC
Ionosphere Effects on Electromagnetic Propagation ionosphere = cold, collisionless, magnetized plasma
X=
ωp2 ω2
ω = 2πf
±
1 O( f 3 )
shift / phase ion distort
n=1−
1 2X
here ionosp
for L-band frequencies (1-2 GHz) refractive index given by:
radio source
higher-order terms on the order of a few cm
ωp = √ Nϵ0eme
2
f = wave frequency
e = fundamental charge
Ne = plasma density
m = electron rest mass
ϵ0 = permittivity of free space
4
Global Navigation Satellite Systems (GNSS)
“
...a useful everyday radio source for geophysical remote-sensing!
GPS - Global Positioning System
GPS
GLONASS Beidou Galileio ...etc.
32-satellite constellation transmit dual-frequency BPSK-moduled signals new Block-IIF and next-gen Block-III satellites transmitting triple-frequency signals Signal Frequency (GHz) L1CA
1.57542
L2C
1.2276
L5
1.17645
5
GNSS Carrier Phase Observable accumulated phase (in meters) of demodulated GNSS signal at receiver for a particular satellite and signal carrier frequency fi FREQUENCY INDEPENDENT EFFECTS
IONOSPHERE RANGE ERROR
SYSTEMATIC ERRORS
Φi = r + cΔt + T + Ii + λi Ni + Hi + Si + ϵi CARRIER AMBIGUITY
HARDWARE BIAS
STOCHASTIC ERRORS
6
Ionosphere Range Error consider first-order term in ionosphere refractive index
n≈1−
1 2X
Ii = ∫
tx
rx
=1−
κ N fi2 e
κ=
e2 8π 2 ϵ0 me
≈ 40.308
tx κ (n − 1) ds ≈ − 2 ∫ Ne ds fi rx TOTAL ELECTRON CONTENT
rx
tx plasma /
units:
electrons m2
often measured in TEC units:
1TECu = 1016 electrons m2
7
vTEC
Ionosphere Plasma Density TE
C
v dis erti tri cal bu tio n
TEC and vertical TEC (vTEC) used to image plasma density structures
profile from CDAAC
tr ion avel dis osp ling tu rb here (TI anc Ds es ) ho dis rizo tri nta bu tio l n
map from IGS
image from Saito et al.
8
TEC Estimation Using DualFrequency GNSS neglecting systematic and stochastic error terms:
Φ1 − Φ2 = (I1 − I2 ) + (λ1 N1 − λ2 N2 ) + (H1 − H2 ) ≈ −κ ( f12 −
1 ) TEC f22
1
+ (λ1 N1 − λ2 N2 ) + ΔH1,2 carrier ambiguities
after resolving bias terms:
TEC =
Φ2 − Φ1 κ ( f12 − 1
1 ) f22
satellite and receiver interfrequency hardware biases
bias terms
9
Resolving Bias Terms carrier ambiguity resolution LAMBDA code-carrier-levelling [3] derives improved code-carrier leveling / ambiguity resolution using triple-frequency GNSS
hardware bias estimation must apply ionosphere model e.g. global ionosphere model using data assimilation and receiver networks e.g. single receiver and linear 2D-gradient in vTEC (such as work by [2])
Example of L1/L2 TEC before and after codecarrier-levelling / ambiguity estimation, for satellite G01 and receiver at Poker Flat, Alaska.
10
Examples of Dual-Frequency TEC Estimates Using methods similar to [2] and [3] to solve for bias terms, we compute dual-frequency TEC estimate TECL1,L2 and TECL1,L5
Poker Flat, Alaska, 2016-01-02
11
TECL1,L5 − TECL1,L2
Poker Flat, Alaska, 2016-01-02
Can we characterize / find the source of these discrepancies? Can we relate them to errors in dual-frequency TEC estimates?
12
Systematic Errors in GNSS Observations hardware bias drifts
Hi terms not constant
multipath reflected signals interfere with primary signal at receiver → causes fluctuations in phase / signal amplitude
ray-path bending r ≠ line-of-sight range
antenna phase effects relative displacement of satellite antenna phase centers changes as satellite moves / rotates
higher-order ionosphere terms need to consider orientation / strength of geomagnetic field
13
Objectives Derive optimal triple-frequency estimation of TEC Investigate the discrepancy in TECL1,L5 − TECL1,L2 Provide a (partial) characterization of TEC estimate residual errors 14
Motivation Improve / understand TEC estimate precision
Push the boundaries of TID signature detection from earthquakes, explosions, etc. Understand / address the errors in TEC estimates from low-elevation satellites Improve user range error for precise positioning applications 15
Approach Develop framework for linear estimation of GNSS parameters Apply framework to derive triple-frequency estimates of TEC and systematic errors Relate to impact on TEC estimate error residuals 16
Background and Motivation Linear Estimation of GNSS Parameters TEC Estimate Error Residuals Application to Real GPS Data
17
Simplified Carrier Phase Model neglect bias terms By neglecting bias terms, we address estimation precision, rather than accuracy
Φi = r + cΔt + T + Ii + λi Ni + Hi + Si + ϵi "geometry" term
Φi = G + Ii + Si + ϵi zeromean
zero-mean normallydistributed
18
Linear Inverse Problem Φ = Am + ϵ Φ = [Φ1 , ⋯ , Φm ]T
m = [G, TEC, S1 , ⋯ , Sm ]T
observations
⎡1 ⎢1 A=⎢ ⎢⋮ ⎣1
− fκ2 1 κ − f2 2
− fκ2
m
model parameters
1 0 0
0 1 ⋯
forward model
⋯ ⋯ ⋱
0⎤ 0⎥ ⎥ ⎥ 1⎦
ϵ = [ϵ1 , ⋯ , ϵm ]T stochastic error
19
Linear Estimation ∗ ^ ≈ A Φ m
^ m
A∗ = ?
model estimate
model estimator
AT
−1 T (AA )
Poor results; treats each parameter with equal weight
We must apply a priori information about model parameters 20
A Priori Information Under normal conditions, we know that:
∣G∣ ≫ ∣Ii ∣ ≫ ∣Si ∣ G∼
20,000 km
I∼
1 - 150 m
S∼
several cm 21
Using A Priori Information We could apply ∣G∣ ≫ ∣Ii ∣ ≫ ∣Si ∣ using Gaussian priors Instead we derive each row separately:
C = [c1 , ⋯ , cm ]T ∈ Rm estimator
(written as row vectors here)
⎡ CG ⎤ CTEC ⎥ ⎢ ⎥ A∗ = ⎢ C S 1 ⎢ ⎥ ⎢ ⋮ ⎥ ⎣ CS ⎦ m
geometry estimator TECu estimator
systematic-error estimators
22
How to Choose Optimal C Linear combination E given by inner-product:
E = ⟨C∣Φ⟩ Goals: 1. produce desired parameter with unity coefficient 2. remove / reduce all other terms Approach: First, constrain C to satisfy Goal 1 Then, constrain / optimize C to achieve Goal 2 23
Linear Coefficient Constraints Use one or two of the following constraints to reduce search space for optimal estimator coefficients:
Φi = G + Ii + Si + ϵi
∑i ci = 0
∑i ci = 1
geometry-free
geometry-estimator
ci ∑i f 2 i
=0
ionosphere-free
κ ∑i − f 2 ci i
=1
TEC-estimator
24
Reduction of Error Linear combination stochastic error variance:
σϵ2 = CT Σϵ C where Σϵ is the covariance matrix between ϵi Optimal C for minimizing stochastic error variance:
C∗ = arg min CT Σϵ C C
ϵi equal-amplitude and uncorrelated
C∗ = arg min ∑ c2i C
i
25
TEC Estimator 1. apply TECu-estimator constraint 2. apply geometry-free constraint (since ∣G∣ ≫ ∣Ii ∣ ) Dual-Frequency Example TEC-estimator geometry-free
−
κ c f12 1
−
κ c f22 2
=1
c1 + c2 = 0 ⇒ c1 = −c2 ⇒ − κc1 ( f12 − 1
⇒ c1 = −
1 ) f22
=1 TEC =
1 κ ( f12 1
−
recall:
1 ) f22
Φ2 − Φ1 κ ( f12 − 1
1 f22 )
26
Triple-Frequency TEC Estimator Applying constraints yields following system of coefficients (with free parameter denoted x:
c1 = c2 =
( f12 − f12 )
1 κ +x
3
1 1 − f22 f12
− κ1 −x(
c3 = x
2
1 1 − f32 f12
1 − 1 f22 f12
)
To satisfy C∗ = arg min ∑ c2i , choose C
x∗ =
1 κ
(
1 − 1 f12 f22
2
i
( f22 − f12 − f12 )
) +(
3
2
1 − 1 f22 f32
1 2
) +(
1 − 1 f32 f12
2
)
denote corresponding coefficient vector CTEC1,2,3 and its corresponding estimate TEC1,2,3 27
TEC Estimator Using TripleFrequency GPS
CTECL1,L2,L5 CTECL1,L5 CTECL1,L2 CTECL2,L5
Estimate
c1
c2
c3
∑i c2i
TECL1,L2,L5 TECL1,L5 TECL1,L2 TECL2,L5
8.294 7.762 9.518 0
−2.883 0 −9.518 42.080
−5.411 −7.762 0 −42.080
10.314 10.977 13.460 59.510
28
Geometry Estimator 1. apply geometry-estimator constraint 2. apply ionosphere-free constraint since Ii are the next-largest terms For triple-frequency GNSS:
c1 = c2 = c3 =
−
(
)
(
)
1 +x 1 − 1 f22 f22 f32 1 − 1 f12 f22
1 −x 12 − 12 2 f1 f1 f3 1 − 1 f12 f22
x
To satisfy C∗ = arg min ∑ c2i , C
x∗ =
1 κ 2
i
( f22 − f12 − f12 ) 3
2
1 2
2
( f12 − f12 ) +( f12 − f12 ) +( f12 − f12 ) 1
2
2
3
3
1
We call this coefficient vector CG1,2,3 and its corresponding estimate G1,2,3 the optimal "ionosphere-free combination"
29
Geometry Estimator Using Triple-Frequency GPS
CGL1,L2,L5 CGL1,L5 CGL1,L2 CGL2,L5
Estimate
c1
c2
c3
∑i c2i
GL1,L2,L5 GL1,L5 GL1,L2 GL2,L5
2.327 2.261 2.546 0
−0.360 0 −1.546 12.255
−0.967 −1.261 0 −11.255
2.546 2.588 2.978 16.639
30
Systematic Error Estimator Since ∣G∣ ≫ ∣Ii ∣ ≫ ∣Si ∣, must apply both geometryfree and ionosphere-free constraints note this requires m ≥ 3
For triple-frequency GNSS:
c1 = x
1 − 1 f32 f22 1 1 − f22 f12
c2 = −x c3 = x
1 − 1 f32 f12 1 − 1 f22 f12
system is linear subspace 31
Geometry-Ionosphere-Free Combination We call the linear combination that applies both geometry-free and ionospherefree constraints the geometry-ionosphere-free combination (GIFC)
FACT: The difference between any two TEC estimates produces some scaling of the GIFC FACT: CGIFC and CTEC1,2,3 are perpendicular, i.e.
CTEC1,2,3 CGIFC
⟨CGIFC ∣CTEC1,2,3 ⟩ = 0 FACT: ⟨CTEC ∣CTEC1,2,3 ⟩ ∣∣CTEC1,2,3 ∣∣
= ∣∣CTEC1,2,3 ∣∣
i.e. CTEC projected onto direction CTEC1,2,3 lands at CTEC1,2,3
32
GIFC Triple-Frequency GPS We (arbitrarily) choose:
CGIFCL1,L2,L5 = CTECL1,L5 − CTECL1,L2 = [−1.756, 9.520, −7.764]T
Note: the triple-frequency GIFC does not have a welldefined unit. GIFC in our results section have the scaling shown here. 33
Background and Motivation Linear Estimation of GNSS Parameters TEC Estimate Error Residuals Application to Real GPS Data
34
Estimate Residual Error Define the error residual vector R with components:
Ri = Si + ϵi The residual error impacting the TEC estimate is:
RTEC = ⟨CTEC ∣R⟩ Note that:
GIFC = ⟨CGIFC ∣R⟩ 35
A Convenient Basis Note that U3 ⊥ CTEC since
We transform R using the orthonormal basis:
U1 =
CTEC1,2,3 ∣∣CTEC1,2,3 ∣∣
U2 =
CGIFC ∣∣CGIFC ∣∣
U1 and U2 span the geometry-free plane
U3 = U1 × U2 ⎡U1 ⎤ U = U2 ⎣U3 ⎦
Note that:
R = UR
R1′
Ri′ = ⟨Ui ∣R⟩
R2′ =
′
=
RTEC1,2,3 ∣∣CTEC1,2,3 ∣∣ GIFC ∣∣CGIFC ∣∣
36
TEC Estimate Residual Error Express RTEC as residual error components in transformed coordinate system:
U1 =
CTEC1,2,3 ∣∣CTEC1,2,3 ∣∣
U2 =
CGIFC ∣∣CGIFC ∣∣
⟨U3 ∣CTEC ⟩ = 0
RTEC = ⟨UCTEC ∣UR⟩ = ⟨U1 ∣CTEC ⟩R1′ + ⟨U2 ∣CTEC ⟩R2′ = RTEC1,2,3 +
⟨CGIFC ∣CTEC ⟩ ∣∣CGIFC ∣∣2 RGIFC
common TEC estimate residual error component
GIFC residual error component
37
TEC Estimate Residual Error Discussion Term
⟨CGIFC ∣CTEC ⟩ ∣∣CGIFC ∣∣2
= amplitude of GIFC residual error
component in TEC estimate TEC1,2,3 is optimal in the sense that it completely removes the GIFC component of residual error Term RTEC1,2,3 = unobservable "TEC-like" residual error component But can we say anything about the overall TEC estimate residual error?
38
Argument for Using GIFC to Assess Overall Residual Error Assume R has an overall distribution that is joint symmetric about the origin with distribution function fR (x)
Ri equal amplitude and uncorrelated
By definition, UR ∼ symmetric with fR (x) for any orthonormal transformation U The distribution of a scaled version aR for some scalar a is fR ( xa ) x fGIFC (x) = fR ( ∣∣CGIFC ∣∣ ) x fRTEC (x) = fR ( ∣∣CTEC ∣∣ )
GIFC ∣∣ fRTEC (x) = fGIFC ( ∣∣C ∣∣CTEC ∣∣ x)
39
Overall TEC Residual Error Discussion The assumption that R has joint symmetric distribution is wrong We can do better by carefully assessing a priori knowledge about the error components in each Φi investigating GIFC is first-step in this process
GIFC ∣∣ fRTEC (x) = fGIFC ( ∣∣C ∣∣CTEC ∣∣ x) is a coarse approximation
relates deviations as: devRTEC ≈
∣∣CTEC ∣∣ ∣∣CGIFC ∣∣ dev
could be very wrong if RTEC1,2,3 ≫ GIF C
GIFC 40
Relation Between GIFC and TEC Estimate Residual Errors amplitude of GIFC error signal in TEC residual
Estimate TECL1,L2,L5 TECL1,L5 TECL1,L2 TECL2,L5
⟨CGIFC ∣CTEC ⟩ ∣∣CGIFC ∣∣2
0 0.303 −0.697 4.723
relates deviation in GIFC and TEC residual
∣∣CTEC ∣∣ ∣∣CGIFC ∣∣
0.831 0.885 1.085 4.796
41
Background and Motivation Linear Estimation of GNSS Parameters TEC Estimate Error Residuals Application to Real GPS Data
42
Experiment Data Alaska, Hong Kong, Peru 2013, 2014, 2015, 2016 Septentrio PolarXs 1 Hz GPS L1/L2/L5 measurements
GPS Lab high-rate GNSS data collection network
43
Data Alignment and Correction align data by sidereal day = 23h 55m 54.2 s
GPS orbital period ≈ 1/2 sidereal day
must remove jumps in GIFC data due to ionosphere activity / multipath / interference
Outlier segments (∣GIFC∣ > 2) are removed from analysis
44
GIFC Examples Alaska G01
G25
G24
G27
45
GIFC Examples Hong Kong G01
G25
G24
G27
46
GIFC Examples Peru G01
G25
G24
G27
47
GIFC Calendar Alaska
G01
G25
G24
G27
48
GIFC Calendar Hong Kong G01
G25
G24
G27
49
GIFC Calendar Peru
G01
G25
G24
G27
50
Satellite Antenna Phase Effects? angle cosine between Earth center, satellite, and Sun
antenna phase effects relative displacement of satellite antenna phase centers changes as satellite moves / rotates
51
GIFC Heatmap Alaska
G01
G25
G24
G27
52
GIFC Heatmap Hong Kong
G01
G25
G24
G27
53
GIFC Heatmap Peru
G01
G25
G24
G27
54
GIFC Deviations and TEC Residual Error Estimates GIFC percentile deviations computed over aggregate of all data
Percentile 50 75 90
Overall 0.11 0.19 0.21
GIFC deviation multiplied by scaling factor
∣∣CTEC ∣∣ ∣∣CGIFC ∣∣
Percentile 50 75 90
TECL1,L2,L5 0.091 0.158 0.175
TECL1,L5 0.033 0.058 0.064
TECL1,L2 0.077 0.132 0.146
TECL2,L5 0.520 0.897 0.992
TECL1,L5 0.097 0.168 0.186
TECL1,L2 0.119 0.206 0.228
TECL2,L5 0.528 0.911 1.007
[TECu]
Percentile 50 ⟨CGIFC ∣CTEC ⟩ 75 ∣∣CGIFC ∣∣2 90
55
Recap simple phase observation model (ignore biases) methodology for choosing optimal linear estimators triple-frequency optima
l?
TEC1,2,3
GIFC characterize / relate to RTEC 56
Discussion TECL1,L2 residual error on order of 0.2 TECu includes large-scale trend → for TID detection, trend is removed and precision improves [4] cites 0.05 TECu fluctuations to be above noise for TID detection Improvement of TECL1,L2,L5 over TECL1,L5 seems minor:
∣∣CTECL1,L2,L5 ∣∣ = 10.314 ∣∣CTECL1,L5 ∣∣ = 10.977 ∣∣CTECL1,L2 ∣∣ = 13.460
...but it does eliminate GIFC component in TEC residual error
57
Next Steps Use characterization of GIFC to address residual errors Is the GIFC trend variation due to satellite antenna phase effects? Can we obtain and apply better information on residual error components Ri ? Can we use GIFC to validate mitigation techniques for multipath, higher-order ionosphere terms, ray-path bending, antenna phase effects? → enable TEC estimation from low-elevation satellites
58
Acknowledgements This research was supported by the Air Force Research Laboratory and NASA.
Thank you to my advisor, committee members, and all who provided me with feedback and criticism! 59
References [1] Saito A., S. Fukao, and S. Miyazaki, High resolution mapping of TEC perturbations with the GSI GPS network over Japan, Geophys. Res. Lett., 25, 3079-3082, 1998. [2] Bourne, Harrison W. An algorithm for accurate ionospheric total electron content and receiver bias estimation using GPS measurements. Diss. Colorado State University. Libraries, 2016. [3] Spits, Justine. Total Electron Content reconstruction using triple frequency GNSS signals. Diss. Université de Liège, Belgique, 2012. [4] M. Nishioka, A. Saito, and T. Tsugawa, “Occurrence characteristics of plasma bubble derived from global ground-based GPS receiver networks,” Journal of Geophysical
60