05/07/03 12.540 Lec 22 1
12.540 Principles of the Global
Positioning System
Lecture 22
05/07/03 12.540 Lec 22 2
Kinematic GPS
processing
Prof. Thomas Herring
? The style of GPS data collection and
processing suggests that one or more GPS
stations is moving (e.g., car, aircraft)
? To obtain good results for positioning as a
function of time requires that the ambiguities
be fixed to integer values
? Track is the MIT implementation of this style of
1
05/07/03 12.540 Lec 22 3
Styles of kinematic GPS
? Kinematic GPS techniques go by a number of names
with features that are often receiver specific.
of lock while the receiver is moving. In survey mode, if loss of
lock occurs the antenna must be returned to a point of known
location.
resolve ambiguities. No need to maintain lock while receiver
moving. Surveying where position during static portion all that
is needed.
radio telemetry link. Analysis is done on-the-fly. Very popular
now with surveyors because results know instantly.
– Kinematic GPS: Early term which implies that there is no loss
– Rapid Static GPS: Technique that uses range and phase to
– RTK Real-time kinematic: Kinematic solution with real-time
05/07/03 12.540 Lec 22 4
General aspects
? The success of kinematic processing depends on
separation of sites
? The MIT software allows multiple stations to be used
in the positioning (may by kinematic or static)
? For separations < 10 km, usually easy (most RTK
systems work at these distances).
? 10>100 km more difficult but often successful.
Depends on quality of data and ionospheric activity.
? >100 km very mixed results
2
05/07/03 12.540 Lec 22 5
Issues with length
05/07/03 12.540 Lec 22 6
Track features
separately.
separately
? As site separation increases, the differential
ionospheric delays increases, atmospheric
delay differences also increase making
modeling of phase data more difficult
? For short baselines (<10 km), ionospheric
delay can be treated as ~zero and L1 and L2
resolved separately
? For longer baselines this is no longer true and
ambiguities must be resolved with LC (and
often the MW WL L1-L2 number of cycles)
? Track uses the Melbourne-Webena Wide
Lane to resolve L1-L2 and then a combination
of techniques to determine L1 and L2 cycles
? For short baselines uses a search technique
and floating point estimation with L1 and L2
? For long baselines uses floating point estimate
with LC and ionospheric delay constraint
3
05/07/03 12.540 Lec 22 7
Ambiguity resolution
differences.
– Searching methods: Two basic types
? Search over integer ambiguities checking RMS fit of phase
residuals
? Search over position, minimizing a fit function that does not
depend on integer part of ambiguity (e.g.. Cosine of phase
residuals)
– Estimation and then resolution using statistical
testing.
? Basic problem is determine the integer
number of cycles in the carrier phase double
? Two generic classes of approach:
05/07/03 12.540 Lec 22 8
? The most common method now is estimation with
statistical assessment of fitting to integer.
? Generic classes of cases:
NNNNN+1 also has >10% of being correct. Since 10-100
ambiguities need to be resolved 1-10 of them would be
incorrect in the above case.
very likely
far from integer,
Statistical Resolution
– NNNNN.01±0.01 Pretty clearly can be resolved
– NNNNN.35±0.40 Highest probability answer is NNNNN but
– NNNNN.01±0.55 clearly close to an integer but +1 value also
– NNNNN.35±0.01 should be resolvable to integer but value is
4
05/07/03 12.540 Lec 22 9
Statistical resolution
?
Formal estimates come for inversion but these
depend on data noise characteristics (most
importantly correlations in data).
?
rates (0.1-1Hz) so white noise assumptions generate
very small error estimates.
?
test (ie., ratio of c
2
with best and next best choice of
ambiguities and an impact on c
2
of setting the value to
an integer. Covers last case shown--no integer
seems correct implying modeling errors.)
Uncertainties of ambiguities are always uncertain.
Many kinematic surveys done with high sampling
Most testing methods use a “contrast” or “ratio” style
05/07/03 12.540 Lec 22 10
LAMBDA Method
? In addition to individual values: Each ambiguity that is
resolved, effects other estimates and thus there is a
cascading effect.
? The LAMBDA Method tries to account for these
correlations by projecting the ambiguities into an
orthogonal space. (Use of eigenvectors and
? Method is from a linear operator that preserves
estimates are nearly un-correlated. (Eigenvectors
would make estimates uncorrelated, by integers would
not be preserved).
eigenvalues discussed in earlier classes).
integer values and transforms ambiguities so that
5
?
05/07/03 12.540 Lec 22 11
Magnitudes of effects of ambiguities
?
?
a change of 0.56 cycles in LC and only 0.22 cycles in LG
(variations in LG can be several cycles)
? Combinations such as N1=3, N2=4 and N1=4 and N2=5 can
cause small effects in LC (ie., geodetic fit looks good but
ionospheric delay in error: if small can be detected but when
large can be difficult).
DLC =
1
1 - ( f
2
/ f
1
)
2
N
1
-
f
2
/ f
1
1- ( f
2
/ f
1
)
2
N
2
= 2.54N
1
-1.98N
2
DLG ( f
2
/ f
1
)N
1
+ N
2
0.78N
1
+ N
2
Basic changes in phase with ambiguities
Notice that N1=N2=1 (not detectable in the MW Widelane) cause
= - = -
05/07/03 12.540 Lec 22 12
processing
? Detecting cycles slips can be difficult in kinematic
processing because of vehicle movement. Normally
in static GPS, coordinates are well enough known that
of data can be used. If a change is larger an a
tolerance level (usually 0.2 to 0.5 cycles) then a cycle
slip is detected.
? Cycle slips are resolved to integers to fitting with
?
known.
Cycle slip detection in kinematic
changes on phase (LC combination) between epochs
simply polynomials across the epoch with the jump.
In kinematic processing this is much more difficult
because the position of the moving receiver is not
6
05/07/03 12.540 Lec 22 13
Cycle slips
? Large jumps can be detected by using the
pseudorange position estimate. In some
? Small slips (just a cycle or so can be difficult). More
common than expected because the receivers try to
fix cycle slips and they often get it wrong be a small
amount (slips based in SNR).
? MW WL and ionospheric delay jumps are common
methods but can still leave slips un-detected.
05/07/03 12.540 Lec 22 14
situations
cases,Doppler shift is also available and can be used.
Basics of MIT track program
? Track runs using a command file
? The base inputs needed are:
? Obs_file specifies names of rinex data files.
Sites can be K kinematic or F fixed
? Nav_file orbit file either broadcast ephemeris
file or sp3 file
? Mode air/short/long -- Mode command is not
strictly needed but it sets defaults for variety of
7
05/07/03 12.540 Lec 22 15
Basic track use
– Kinematic site rovr appears dynamic Coordinate
RMS XYZ 283.44 662.53 859.17 m.
– For 2067 Double differences: Average RMS
17.85 mm
? Recommended to start with above commands
and see how the solution looks
? Usage: track -f track.cmd >&! track.out
? Basic quality checks:
? grep RMS of output file
05/07/03 12.540 Lec 22 16
Basic track use and evaluation
– grep FINAL output file
– A 3 in column Fixd means fixed, 1 means still
floating
? Check on number of ambiguities fixed
? If still non-fixed biases or atmospheric delays
are estimated then smoothing solution should
be made (back_type smooth)
? output in NEU and/or geodetic coordinates
8
05/07/03 12.540 Lec 22 17
More advanced features
– Works by running track and extracting FINAL lines
into an ambiguity file. Setting 7 for the Fixd column
will force fix the ambiguity. ambiguity file is then
read into track (-a option or ambin_file)
05/07/03 12.540 Lec 22 18
Advanced features
? Track has a large help file which explains
strategies for using the program, commands
available and an explanation of the output and
how to interpret it.
? It is possible to read a set of ambiguities in.
? Commands allow control of how the biases
are fixed and editing criteria for data
? Editing is tricky because on moving platform,
jumps in phase could simply be movement
? Ion delay and MW WL used for editing.
? Explicit edit_svs command allows removal of
problematic data
9
05/07/03 12.540 Lec 22 19
Some results
rovr.LC and
d016f.NEU.rovr.L1+L2
? Examine results from phase processing of
homework #3.
? Solutions with LC and with L1+L2 (less noise
but larger ionospheric delay.
? Output of processing in track.016f.out
? Solutions in North, East Up differential position
from etab: d016f.NEU.
05/07/03 12.540 Lec 22 20
North Evolution
-6000.00
-4000.00
-2000.00
0.00
2000.00
4000.00
16.80 16.85 16.90 16.95 17.00 17.05
Fractional Day
dN LC (m) dN L1+L2 (m)
dN
LC
(m)
10
05/07/03 12.540 Lec 22 21
-766.15
-766.10
-766.05
-766.00
-765.95
-765.90
16.85 16.85 16.85 16.85 16.85 16.85
Fractional Day
Zoom of features
dN LC (m) dN L1+L2 (m)
dN
LC
(m)
05/07/03 12.540 Lec 22 22
Summary
? Track is still developmental and performance
depends on quality of GPS data
? For short baselines it usually works well, for
longer baselines it can be difficult
? see $HELP_DIR/track.hlp for more details.
? There are frequent updates to the program
11