1
Dr. Guoqing Zhou
9. Data Processing (2)
CET 318
Book: p. 203-276
10. Data Processing (2)
From Download Data from Receiver to
Establishment of Data Processing Model and
Solution
3. Adjustment, Filtering, and
Smoothing-----Overview
3.1 Least Square Adjustment
1. Observation Equation
2. Adjustment
3. Accuracy Analysis
3.2 Kalman Filtering
1. Introduction
2. Prediction
3. Update
4. Example
3.3 Smoothing
See Explanation
2
4. Adjustment of Mathematical
GPS Model
4.1 Linearization of Observation Equation
Taylor series with respect to the approximate point
)Z-(t)(Z)Y-(t)(Y)X-(t)(X)(ρ
2
i
2
i
2
i
j
i
jjj
t ++=
)Z,Y,f(X
)Z-(t)(Z)Y-(t)(Y)X-(t)(X)(ρ
i0i0i0
2
i0
2
i0
2
i0
j
i0
=
++=
jjj
t
)Z,Y,(X
i0i0i0
Approximate point position
ij
i
i
j
ij
i
i
j
ij
i
i
j
Z
tp
ZtZ
Y
tp
YtY
X
tp
XtX
tt
?
?
?
?
?
??
?
?=
)(
)(
)(
)(
)(
)(
)(ρ)(ρ
0
0
0
0
0
0j
i0
j
i
iZiiYiiXi
ZaYaXatt ?????=
111j
i0
j
i
-)(ρ)(ρ
4.2 Linearization Model for Point
Positioning with Code Ranges
The elementary model for point positioning with code ranges
is given by
(t)δ-(t)δ)(p)(R
i
jj
i
j
i
cctt +=
In the model, only the clocks are modeled. The ionosphere,
troposphere, and other minor effects are neglected.
Lineaized Equation:
(t)δ-ZaYaXa(t)δ)(p)(R
ii
1
Zi
1
Yi
1
X
jj
i
j
i
iii
cctt ?+?+?=??
(t)δ-ZaYaXa
ii
1
Zi
1
Yi
1
X
1
iii
cl ?+?+?=
?a ?a ?a ?
1
Z
1
Y
1
X
1
iii
====l
The satellite clock bias is assumed to be known. This
assumption makes sense because satellite clock correctors
can be received from the navigation message.
Four Satellites:
(t)δ-ZaYaXa
ii
1
Zi
1
Yi
1
X
1
iii
cl ?+?+?=
(t)δ-ZaYaXa
ii
2
Zi
2
Yi
2
X
2
iii
cl ?+?+?=
!!!!!
3
=l
!!!!!
4
=l
Observation Equation:
Axl =
4.3 Linearization Model for Point
Positioning with Carrier Phases
(t)δ-λNZaYaXa(t)δ)()(λΦ
i
j
ii
1
Zi
1
Yi
1
X
j
0
j
iii
cctpt
j
ii
+?+?+?=??
The procedure is the same as code range
Compared to point positioning with code ranges, the number
of unknowns is now increased by the ambiguities.
Lineaized Equation:
Four Satellites again:
Axl =
The coefficients of the coordinate increments are supplemented
with the time parameter t. Obviously, the four equations do not
solve for the eight unknowns.
Observation Equation:
Above fact reflects
1. Point positioning with phases in this form cannot be
solved epoch by epoch.
2. Each additional epoch increases the number of
unknowns by a new clock term.
3. For two epochs there are eight equations and nine
unknowns (still an underdetermined problem). For three
epochs there are 12 equations and 10 unknowns, thus, a
slightly overdetermined problem.
Axl =
3
4.4 Linearization Model for Relative
Positioning
For the case of relative positioning, the investigation is
restricted to carrier phases, since it should be obvious how to
change from the more expanded model of phases to a code
model.
The model for the double-difference
jk
AB
jk
λN)()(λΦ += tpt
jk
ABAB
)()()()()( tptptptptp
j
A
k
A
j
B
k
B
jk
AB
+??=
Where:
Above equation reflects the fact of four measurement
quantities for a double-difference.
Each of the four terms must be linearized.
Lineaized Equation:
Axl =
jk
ABBZBYBX
AZAYAX
jk
AB
λNZaYaXa
ZaYaXa(t)δ
BBB
AAA
+?+?+?+
?+?+?=
jkjkjk
jkjkjk
)()()()()(λΦ)(
0000
jk
AB
tptptptpttl
j
A
k
A
j
B
k
BAB
jk
?++?=
?a ?a ?a
?a ?a ?a
BBB
AAA
ZYX
ZYX
===
===
jkjkjk
jkjkjk
Where:
Observation Equation:
5. Network Adjustment
5.1 Single Baseline Solution
1. The adjustment principle requires observations are
uncorrelated.
2. The single-differences are uncorrelated, whereas double-
differences and triple-differences are correlated.
3. The implementation of the double-difference correlation
can be easily accomplished. Alternatively, decorrelated
algorithm using a Gram-Schmidt orthogonalization.
4. The implementation of the correlation of the triple-
differences is more difficult. Furthermore, it is
questionable since the noise of the triple-differences will
always prevent to obtain a refined solution.
Problem:
4
5. For an observed network, the use of the single baseline
method usually implies a baseline by baseline
computation for all possible combinations.
6. If ni denotes the number of observing sites, then ni(ni-1)
/2 baselines can be calculated. Note that only ni-1 of
them are theoretically independent.
7. The redundant baselines are either used for misclosure
checks or for an additional adjustment of the baseline
vectors.
1. The simple single baseline solution from the theoretical
point of view is that it is not correct because of the
correlation of the simultaneously observed baselines.
2. By solving baseline by baseline, this correlation is ignored.
Disadvantage:
5.2 Multi-Point Solution
1. In contrast to the baseline by baseline solution,
the multipoint solution considers at once all
points in the network.
2. The key difference compared to the single
baseline solution is that the correlations between
the baselines are taken into account.
3. The same theoretical aspects also apply to the
extended case of a network.
Single-difference Example for A Network
Taking A as reference site, two baselines A-B, A-C, the two
single-differences
j
A
B
C
Epoch t
B
a
s
e
li
n
e
)(Φ)(Φ)(Φ
j
A
j
ttt
j
BAB
?=
)(Φ)(Φ)(Φ
j
A
j
ttt
j
CAC
?=
?
?
?
?
?
?
=
)(Φ
)(Φ
j
j
t
t
SD
AC
AB
?
?
?
?
?
?
?
?
?
?
=Φ
)(Φ
)(Φ
)(Φ
j
j
j
t
t
t
C
B
A
?
?
?
?
?
?
?
?
=
101
011
C
Introducing
?
?
?
?
?
?
==
21
12
δδ)(
22 T
CCSDCov
Covariance matrix
A correlation of the single-differences of the two baselines
with a common point.
Double-difference Example for A Network
j
A
B
C
Epoch t
B
a
s
e
li
n
e
l
m
k
For ni points and nj satellites, (ni
-1)(nj-1) independent double-
difference
For ni =3; nj=4, 6 independent
double-difference
Matrix expression
Introducing
5
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
==
4
24
224
2114
12124
112224
δδ)(
22 T
CCDDCov
Covariance matrix
1. Solution without any correlation deviates from the
theoretically correct values by a greater amount.
2. It is estimated that the single baseline method deviates
from the multi-baseline (correlated) solution by a
maximum of 2δ.
symmetry
5.3 Single Baseline vs. Multi-point Solution
Some Arguments:
1. The correlation is not modeled correctly with the single
baseline solution because correlations between baselines
are neglected.
2. The computer program is, without doubt, much simpler
for the single baseline approach.
3. With modern software and hardware, the computational
time is not a real problem.
4. Cycle slips are more easily detected and repaired in the
multipoint mode.
5. It takes less effort in the single baseline mode to isolate
bad measurements and possibly to eliminate them.
6. The economic implementation of the full correlation for
a multipoint solution only works properly for networks
with the same observation pattern at each receiver site.
7. Even in the case of the multipoint approach, it becomes
questionable whether the correlations can be modeled
properly.
8. For the dual frequency receivers, the ionosphere-free
combination Lc is formed from L1 and L2 and
processed together with the L1 data of the single
frequency receivers. Thus, a correlation is introduced
because of the L1 data. A proper modeling of the
correlation biases the ionosphere-free Lc baseline by the
ionosphere of the L1 baseline, an effect which is
definitely undesirable.
5.4 Least Square Adjustment of Baselines
j
i
In networks, the number of measured baselines will
usually exceed the minimum amount.
Redundant information is available and the determination
of the coordinates of the network points may be carried
out by a least squares adjustment.
Xi
j
ijijX
AB
X-XXn ?=
ijijY
AB
Y-YYn ?=
ijijZ
AB
Z-ZZn ?=
6. Dilution of Precision
The geometry of the visible satellites is an important factor in
achieving high quality results especially for point positioning
and kinematic surveying.
The geometry changes with time due to the relative motion of
the satellites.
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
?
=
c
c
c
c
4
Z
4
Y
4
X
3
Z
3
Y
3
X
2
Z
2
Y
2
X
1
Z
1
Y
1
X
iii
iii
iii
iii
aaa
aaa
aaa
aaa
A
A measure of the geometry is the Dilution of Precision (DOP)
factor.
4 Satellites for point
positioning with code
ranges
6
1T
X
)A(AQ
?
=
DOP can be calculated from the co-variance matrix of the
solution.
?
?
?
?
?
?
?
?
?
?
?
?
?
?
=
tt
ZZZ
YYZYY
XXZXYXX
q
qq
qqq
qqqq
Q
t
t
t
X
Symmetry
The diagonal elements are used for the following DOP
definitions:
ttZZYYXX
qqqqGDOP +++=
ZZYYXX
qqqPDOP ++=
tt
qTDOP =
Geometric dilution
Position dilution
Time dilution
Why
7. Accuracy Measures
Why do we need accuracy measures ?
How many accuracy measures are in use!
Qxx, and RMS
Chi-square Distribution
One-dimensional accuracy measures
Two-dimensional accuracy measures
Three-dimensional accuracy measures
Summary
What have we learnt?
Which parts are important?