Professional Documents
Culture Documents
To cite this article: C. Satirapod & M. Luansang (2008) Comparing stochastic models
used in GPS precise point positioning technique, Survey Review, 40:308, 188-194, DOI:
10.1179/003962608X290988
ABSTRACT
GPS observables are normally processed using the least-squares method. In order to ensure high
accuracy positioning results, both the functional model and the stochastic model must be correctly
defined. In GPS Precise Point Positioning (PPP) mode, an ionosphere-free linear combination is
generally used for constructing the functional mode, while post-mission information and error
mitigation methods are employed to eliminate many biases. Nevertheless, some unmodelled biases still
remain in the GPS observables and eventually cause errors in the coordinate results. It is, however,
possible to further improve the accuracy and reliability of GPS results through an enhancement of the
stochastic model. This research aims to investigate stochastic models used in the GPS PPP technique.
Three different stochastic models; equal weight for all GPS observables, different weight for each GPS
observable based on satellite elevation angle and different weight for each GPS observable based on the
MINQUE (Minimum Norm Quadratic Unbiased Estimation) procedure, have been compared in this
paper. Test results indicate that the stochastic model estimated from the MINQUE method produces the
most accurate coordinate results both in horizontal and vertical components.
INTRODUCTION
GPS PPP mode only. The performance of three different stochastic models are
compared, namely; equal weight for all observables, different weight for each GPS
observable based on satellite elevation angle and different weight for each GPS
observable based on the MINQUE (Minimum Norm Quadratic Unbiased Estimation)
procedure [14]. The functional and stochastic models are described in the following
section. The test data and processing strategy are explained, followed by the analysis
of results for assessing the performance of each stochastic model. Some concluding
remarks are made in the final section.
The undifferenced observation equations for code and carrier phase measurements
are ([1], [2], [15])
f 12 .P( L1 ) - f 22 .P( L 2 )
P( L3 ) = = ρ + c( dt - dT ) + d orb + d trop + ε p (3)
f 12 - f 22
f 12 .φ( L1 ) - f 22 .φ( L 2 )
φ( L3 ) =
f 12 - f 22
(4)
cf 1 N 1 - cf 2 N 2
= ρ + c( dt - dT ) + d orb + d trop + + εφ
f 12 - f 22
where f 1 and f 2 are the GPS frequencies on L1 and L2 (Hz) respectively and
cf 1 N 1 - cf 2 N 2
is an ionosphere-free ambiguity term.
f 12 - f 22
Equations (3) and (4) are typically used as a traditional functional model for the PPP
technique as they can eliminate the ionospheric bias ([1], [16], [17], [18]). It should be
189
COMPARING STOCHASTIC MODELS USED IN GPS PRECISE POINT POSITIONING
noted that the ionosphere-free ambiguity term in equation (4) is no longer an integer
number, and hence treated as a float number. Error mitigation methods are introduced
to reduce systematic errors. Other unknown parameters (such as ambiguity terms,
receiver clock errors, receiver coordinates) are to be estimated in the least-squares
process.
[19] pointed out that the estimated unknown parameters are dependent on the
stochastic model adopted for the measurements. Any mis-specifications in the
stochastic model may lead to inaccurate results (e.g. [20], [21], [22]). In this study, the
three different stochastic models, called Models I, II and III, are investigated. Model I
assumes that all GPS observables are statistically independent and have the same
variance. Model II assumes that all GPS observables are statistically independent and
have different variances based on the satellite elevation angle. Model III assumes that
all GPS observables are statistically dependent and have different variances based on
the MINQUE procedure. The details of each stochastic model are outlined below.
Cm = σ2 Inm (5)
For a session solution with t epochs of data, the covariance matrix of all
undifferenced GPS observations is
C = σ2 D (6)
Model II: Different Weight for each observable based on satellite elevation angle
In this stochastic model, GPS observations are assumed to be statistically
independent and have different variances. These variances are defined as a function of
satellite elevation angle. The basic assumption of using the satellite elevation angle
information is that each satellite has a different precision and that a low elevation angle
satellite tends to be noisier than a high elevation angle satellite. The simple
relationship between satellite elevation and variance (σe2) can be expressed as ([5],
[23])
The covariance matrix of all undifferenced GPS observations for a session solution
with t epochs of data is
C = σe2 D (9)
190
C SATIRAPOD AND M LUANSANG
Model III: Different Weight for each observable based on the MINQUE procedure
Model III assumes that all GPS observations are statistically dependent and have
different variances. The basic assumption is that residuals obtained from the least-
squares process represent the same characteristic as true errors if the observation
period is long enough to remove any systematic error. Thus, the precision of each
satellite relies on the residuals obtained from the least-squares process. Based on the
use of least-squares residuals, the rigorous statistical MINQUE procedure is employed
to estimate a full covariance matrix. The full covariance matrix at epoch m can be
expressed as
k
C m = ∑ θ i Tim (10)
i =1
where θ1,θ2,… ,θk are the variance-covariance components of the GPS observations to
be estimated; k is the number of variance-covariance components; and T1,T2,… ,Tk are
the accompanying matrices. Comprehensive details of the MINQUE procedure for the
estimation of variance-covariance components of the GPS observations can be found
in [22] and [24]. For a session solution with t epochs of data, the covariance matrix of
all undifferenced GPS observations is
k
C = ∑ θ i Ti (11)
i =1
Test Data
The data was collected on the rooftop of Vidhayanives building, at Chulalongkorn
University, Bangkok, Thailand, using a dual-frequency GPS receiver (Leica system
500). All data was collected in static mode between 25th and 30th October 2002, at 15-
second rate. The cut-off elevation angle was set to 15 degrees.
191
COMPARING STOCHASTIC MODELS USED IN GPS PRECISE POINT POSITIONING
ANALYSIS OF RESULTS
After all data segments were processed, the final coordinates were compared with the
reference coordinates. The performance of the PPP technique under each model is
characterized by the Root Mean Square Error (RMSE) value. Both the horizontal and
vertical RMSE values are computed to estimate the accuracy of results obtained from
all session lengths. These are shown in Table 1. Further visual comparisons are shown
in Figures 1 and 2, for the different session lengths.
With reference to Figure 1, all the three models tend to produce comparable values in
horizontal RMSE. However, Model III consistently gives the smallest horizontal
RMSE in all session lengths. There is also clear improvement in vertical accuracy for
all the models (Figure 2). Model III shows an overall improvement in vertical accuracy
of approximately 7% compared with Model I. Similarly, Model III gives the smallest
vertical RMSE value in all session lengths.
CONCLUDING REMARKS
In this paper, three different stochastic models have been tested with the GPS PPP
technique. Based on the results, the use of Model III (different weight for each GPS
observable based on the MINQUE procedure) improves the accuracy of solutions best,
192
C SATIRAPOD AND M LUANSANG
for both the horizontal and vertical components. Since the functional model and error
mitigation methods were kept identical in all data processing (only varied the
stochastic models, the improvements in positioning results are entirely from the use of
more realistic stochastic model. Thus, model III is identified to be the best model
among the three stochastic models tested. However, it should be noted that this model
is designed for static data processing, since it requires an adequate number of
observations to produce reliable estimates of all variance-covariance components.
Fig. 1. Horizontal RMSE values obtained from the 5-min, 10-min, 15-min and 30-min session lengths
Fig. 2. Vertical RMSE values obtained from the 5-min, 10-min, 15-min and 30-min session lengths
ACKNOWLEDGEMENTS
References
1. Leick, A. 2004. GPS Satellite Surveying. (3rd edition), John Wiley & Sons, Inc., New York,
435pp.
2. Rizos, C. 1997. Principles and Practice of GPS Surveying, Monograph 17, School of Geomatic
Engineering, The University of New South Wales, 555pp.
193
COMPARING STOCHASTIC MODELS USED IN GPS PRECISE POINT POSITIONING
3. Zumberge, J.F., Heflin, M.B., Jefferson, D.C., Watkins, M.M. and Webb, F.H. 1997. Precise
point positioning for the efficient and robust analysis of GPS data from large networks, Journal
of Geophysical Research, 102(B3), 5005-5017.
4. Satirapod, C. 2006. Stochastic Models Used in Static GPS Relative Positioning, Survey Review,
38(299), 279-286.
5. Barnes, J.B. and Cross, P.A. 1998. Processing model for very high accuracy GPS positioning,
Journal of Navigation, 51, 180-193.
6. Cross, P.A., Hawksbee, D.J. and Nicolai, R. 1994. Quality measures for differential GPS
positioning, Hydrographic Journal, 72, 17-22.
7. Gao, Y. and Shen, X. 2002. A New Method for Carrier-Phase-Based Precise Point Positioning,
Navigation, 49(2), 109-116.
8. Han, S. 1997. Quality control issues relating to instantaneous ambiguity resolution for real-time
GPS kinematic positioning, Journal of Geodesy, 71(7), 351-361.
9. Lau, L. and Mok, E. 1999. Improvement of GPS relative positioning accuracy by using SNR,
Journal of Surveying Engineering, 125(4), 185-202.
10. Satirapod, C. 2002. Improving the GPS Data Processing Algorithm for Precise Static Relative
Positioning, Unisurv Report S-69, School of Surveying & Spatial Information Systems, The
University of New South Wales, Sydney, Australia, 131pp.
11. Teunissen, P.J.G. 1997. On the sensitivity of the location, size and shape of the GPS ambiguity
search space to certain changes in the stochastic model, Journal of Geodesy, 71, 541-551.
12. Tiberius, C.C.J.M. and Kenselaar, F. 2000. Estimation of the stochastic model for GPS code
and phase observables, Survey Review, 35, 441-454.
13. Wang, J. 1999. Modelling and Quality Control for Precise GPS and GLONASS Satellite
Positioning, Ph.D. thesis, School of Spatial Sciences, Curtin University of Technology, Perth,
Australia, 171pp.
14. Rao, C.R. 1971. Estimation of variance and covariance components- MINQUE, Journal of
Multivariate Analysis, 1, pp. 257-275.
15. Teunissen, P.J.G. and Kleusberg, A. 1998. GPS for Geodesy, Springer-Verlag, Berlin
Heidelberg New York, 650pp.
16. Witchayangkoon, B. 2000. Elements of GPS Precise Point Positioning, Doctoral
dissertation, Department of Spatial Information Science and Engineering, Graduate
School, University of Maine, 265pp.
17. Muellerschoen, R.J., Bar-Sever, Y.E., Bertiger, W.I. and Stowers, D.A. 2001. NASA’s Global
DGPS for High Precision Users, GPS World, 12(1), 14-20.
18. Kouba, J. and Heroux, P. 2001. Precise Point Positioning Using IGS Orbits Products, GPS
Solutions, 5(2), 12-28.
19. Wang, J., Satirapod, C., and Rizos, C. 2002. Stochastic assessment of GPS carrier phase
measurements for precise static relative positioning, Journal of Geodesy, 76(2), pp. 95-104.
20. Cannon, M. E. and Lachapelle, G. 1995. Kinematic GPS trends: equipment, methodologies and
applications, In: Beutler, Hein, Melbourne & Seeber (Ed.): GPS Trends in Precise Terrestrial,
Airborne, and Space-borne Applications, IAG Symposium No. 115, Springer-Verlag, Berlin
Heidelberg New York, 161-169.
21. Satirapod, C., Wang, J. and Rizos, C. 2001. A New Stochastic Modelling Procedure for Precise
Static GPS Positioning, Zeitschrift für Vermessungswessen, 126(6), 365-373.
22. Wang, J, Stewart, M.P. and Tsakiri, M. 1998. Stochastic modelling for static GPS baseline data
processing, Journal of Surveying Engineering, 121(4), 171-181.
23. Gerdan, G.P. 1995. A comparison of four methods of weighting double-difference pseudo-
range measurements. Trans Tasman Surveyor, Canberra, Australia, 1, pp. 60-66.
24. Satirapod, C., Wang, J. and Rizos, C. 2002. A simplified MINQUE procedure for the
estimation of variance-covariance components of GPS observables, Survey Review, 35(286),
582-590.
25. Dawson, J., Govind, R. and Manning, J. 2001. The AUSLIG Online GPS Processing System
(AUSPOS), Available from: http://www.ga.gov.au/geodesy/sgc/wwwgps/ [March, 15th 2007].
26. Satirapod, C. and Homniam, P. 2006. GPS Precise Point Positioning Software for Ground
Control Point Establishment in Remote Sensing Applications, Journal of Surveying
Engineering (ASCE), 132(1), 11-14.
27. Saastamoinen, J. 1971. Atmospheric Correction for the Troposphere and Stratosphere in Radio
Ranging of Satellite, Int. Symp. on the Use of Artificial Satellite, Henriksen (ed.), 3rd
Washington, 247-251.
28. Niell, A. E. 1996. Global Mapping Functions for the Atmosphere Delay at Radio Wavelengths,
Journal of Geophysical Research, 101(B2), 3227-3246.
194