Long-Range RTK Positioning Using

Virtual Reference Stations

Ulrich Vollath, Alois Deking, Herbert Landau,

Christian Pagels

Trimble Terrasat GmbH

BIOGRAPHY The paper presents results from practical tests with a RTK
system using standard reference stations and virtual
Alois Deking received a degree in Computer Science at reference stations. A special set-up was used to allow the
the FH Regensburg in 1994 and a Master of Science comparison of the two different RTK techniques using the
Degree in Computer Science and Engineering at the identical GPS raw data stream by two different rovers.
University of Connecticut in 1995. Since 1995 he is While the rover data was identical the RTCM data stream
working as development engineer at Trimble Precision is different for the two RTK rovers, one is using the
Terrasat, a company that actively works in the area of standard RTCM coming directly from the next reference
high-precision differential GPS since several years. station and the other is using the virtual reference station
Currently he is active in the area of Virtual Reference (VRS) data created from the GPS-Network.
Stations.
In that way we are able to compare the positioning results
Dr. Herbert Landau is Managing Director of Trimble nicely with respect to accuracy, reliability and
Precision Terrasat. He has many years of experience in initializations time performance.
GPS and has been involved in a large variety of GPS and
GLONASS developments for high precision positioning It is shown, that the use of VRS data results in a
systems and applications. significant performance gain of the RTK rover system.
While the rover with the standard RTCM data stream
Christian Pagels has a degree in Physics from the experiences a lot of problems with ambiguity
University of Erlangen (1993). He is Senior Development initialization, the VRS rover has a far higher success rate
Engineer at Trimble Precision Terrasat and working on and significantly less time for ambiguity initialization.
the design of reference station systems. The VRS rover reaches accuracies of 2 cm in horizontal
coordinates with a high reliability on distances of up to 32
Dr. Ulrich Vollath received a Ph.D. in Computer Science km.
from the Munich University of Technology (TUM) in
1993. He is Senior Manager of the department of INTRODUCTION
kinematic positioning and real-time systems of Trimble
Precision Terrasat. Differential GPS is a well-established technique to
improve the positioning precision since many years. This
ABSTRACT allows even centimeter-level accurate positioning using
the so-called integer ambiguity resolution technique.
Long-Range RTK positioning is limited due to the growth
of the influence of systematic errors with the baseline The basic concept is to mitigate the main error sources,
length. Especially during periods of high ionospheric ionospheric and tropospheric delay, orbit errors and
influences the maximum distance from the reference satellite clock errors (where we had S/A effects, too until
station allowing high productivity work for RTK recently) by receiving satellite data at a well-known
positioning might become as low as 10 km or less. location. All common errors between this reference
receiver and the user receiver cancel out. Though this
The concept of virtual reference stations allows already works quite well for many applications, the
performing RTK positioning in reference station networks special de-correlation (i.e. the change of the errors when
with distances of up to 40 km or more from the next moving away from the reference station) of the errors
reference station while providing the performance of short leaves large error contributions in the corrected signals.
baseline positioning.

470
A new concept addresses this problem. The idea is to • The navigation solution is transferred to the
generate Virtual Reference Stations (VRS) that simulate a computing center.
local reference station near by the user receiver. Thus, the
errors cancel out better than by using a more distant • The computing center immediately starts to send
reference station.

Using networks of reference stations with distances of

B
about 70 km between the reference receivers,
performance tests were done. Two brands of real-time
kinematic receivers were operated on 16 km and 32 km
baselines during a period of medium solar activity. For
every network, one receiver received normal single base
station reference data. The other was operating with Virtual Ref. C
Virtual Reference Station data.

Rover D
The results are very promising, showing improvements in
time to fix as well as positioning accuracy by a factor of
2. A
VIRTUAL REFERENCE STATION OVERVIEW

The basic principles of Virtual Reference Stations Virtual Reference Station data to the field user.
operation are given in the following overview:
Figure 1: Virtual Reference Station Network
• Data from the reference station network is
Figure 2 shows a typical setup procedure.
transferred to a computing center.

• The network data is used to compute models of GPS Receiver

ionospheric, tropospheric and orbit errors.
GPS Receiver

• The carrier phase ambiguities are fixed for the

network baselines.
LAN RS232 LAN

• The actual errors on the baselines are derived in Access Server Router
Mobile Phone GPSNet Server
centimeter accuracy using the fixed carrier phase RTCM VRS

observations.

• Linear of more sophisticated error models are Displacement &

Interpolation
used to predict the errors at the user location.

• A Virtual Reference Station (VRS) is created at GPS Receiver

GPS Receiver
the user location.
Figure 2: Virtual Reference Station Set-Up
• The VRS data is transmitted to the user in
THE TEST SET-UP
standard formats (RTCM).
The following set-ups were used to assess the
This concept is visualized in Figure 1. Base station data
performance of VRS operation in comparison with
from A,B,C and D are used to predict the errors at the
traditional reference station methods.
Rover location.
1. The SAPOS networking system of the Bavarian
The user set-up in the field follows this procedure: land surveying authorities (BLVA) was used to
generate VRS and traditional uncorrected data on
• The field receiver determines the user location
a baseline of 16 km (Figure 3). Rover receiver
with a navigation solution (no reference) or by
brand A was used.
DGPS (uncorrected data)
2. The Trimble Terrasat test network generated
• The receiver dials into the computing center via VRS and uncorrected reference data on a 32 km
mobile phone and is authenticated. baseline. (Figure 4). Rover receiver brand B was
used.

471
 In a post-mission step, these data were analyzed. The
deviations of the computed positions from the true
position, established with static post-processing a week of
data, were analyzed statistically in their North, South and
vertical components to produce the error distributions and
statistical moments.

Also, the times to fix (TTF) were assessed by inspecting

the gaps in the precise positions due to new initialization
after ambiguity resets.

The data presented here was recorded in week 36, May

2001, during a period of medium solar activity. The rover
B setup on a roof of a building produced high multipath,
as many reflecting material was ubiquitous on the roof.
Rover A was using an antenna of the company’s test
network, setup in low multipath conditions.

Figure 3: The SAPOS test setup FIXING PERFORMANCE (TTF)

For every reference station network, the VRS data was To get a practical measure of the fixing performance of
used on one geodetic quality RTK rover and the
traditional uncorrected reference data was fed into another TTF VRS/Raw 16km
identical RTK receiver. Both receivers were configured to 100
Raw Data
operate in kinematic mode. 90
Acc. Rel. Frequency [%] 80
VRS

70
60
50
40
30
20
10
0
0 20 40 60 80 100 120 140 160 180
TTF [sec]
the two systems, the distributions of the accumulated
times to fix are suitable. For every time span, the
percentage of fixes that have a time to fix below or equal
to this time is computed. Figure 5 shows the TTF
distributions of the raw data system and the VRS-operated
system for the 16 km baseline data collected.

Figure 5: Time to fix VRS/Raw, 16km

To get an idea of the performance experienced by a field

user, two numbers are interesting: the maximum time to
Figure 4: The Trimble Terrasat test network setup. fix for 50 % of the samples giving the typical TTF and the
90 % value for the typical maximum time.
A file server recorded the computed positions from the
two real-time engines in NMEA format. The RTK For VRS, the figure returns a 50 % value of 15 seconds, a
systems were programmed to reset all ambiguity fixing value perfect for productive work, and a 90 % time of 65
information after the computation of a few positions in seconds. Using raw data, these numbers increase to 55
centimeter level carrier phase precision. and more than 180 seconds, an improvement by a factor
of more than 4.

472
For the 32 km baseline, the results are shown in Figure 6. independent of the time of day. This proves that the VRS
The Time-To-Fix values obtained in this set-up are 75 sec method has reduced ionospheric influences at day time
(50%) and 115 sec (90%) for VRS mode. For raw data the down to the night time level.
times increase to 115 sec (50%) and more than 180 sec
(90%). Again, the improvement is from long initialization Figure 9 and Figure 10 show the same analyses for the 32
times typically experienced during ionospheric activities km baseline. Surprisingly, daytime performance is
to a good performance allowing productive work with the
TTF VRS 16km Day/Night
VRS method. 100 Day

Acc. Rel. Frequency [%]

90 Night
TTF VRS/Raw 32km 80
100
Acc. Rel. Frequency [%]

70
VRS
90 60
80 Raw Data 50
70 40
60 30
50 20
40 10
30 0
0 20 40 60 80 100 120 140 160 180
20 TTF [sec]
10 somewhat better than nighttime performance. This effect
0 could not be explained yet. Still, the improvements by
0 20 40 60 80 100 120 140 160 180 using VRS are large at both times of the day.
TTF [sec]
Figure 6: Time to fix, VRS/Raw, 32km Figure 8: Time to fix, VRS, 16km, Day/Night

DAYTIME/NIGHTTIME COMPARISON
TTF Raw 32km Day/Night
90
Acc. Rel. Frequency [%]

As fixing performance and precision are very much Day

influenced by the level of solar activity present, it is 80
Night
interesting to view daytime and nighttime performance 70
separately. The data collection was running over several 60
days and nights, so the previous figures build the average
of day and night. By separating the times to fix of day and 50
night data, interesting insights are derived (Figure 7). 40
30
20
TTF Raw 16km Day/Night 10
100
Acc. Rel. Frequency [%]

Day 0
90 0 20 40
60 80 100 120 140 160 180
Night TTF [sec]
80
70 Figure 9: Time to fix, Raw, 32km, Day/Night
60
50
40 TTF VRS 32km Day/Night
100
Acc. Rel. Frequency [%]

30
Day
20 90
Night
10 80
0 70
20 40 60 80 100 120 140 160 180
TTF [sec] 60
Figure 7: Time to fix, Raw, 16km, Day/Night 50
40
The figure shows that long initialization times can be 30
attributed to the ionospheric activities at daytime mainly.
20
For VRS mode on 16 km, Figure 8 gives the day/night 10
comparison. The performance is now similar and almost 0
0 20 40 60 80 100 120 140 160 180
TTF [sec]

473
Figure 10: Time to fix, VRS, 32km, Day/Night International Geodetic Symposium on Satellite
Positioning, Las Cruces, New Mexico, March 13-17,
PERFORMANCE OVERVIEW 1989.

Table 1 shows the performance numbers derived in the [2] Sjöberg, L: The Best Linear Combination of L1 and
tests presented. In all disciplines, VRS shows significant L2 Frequency Observables in the Application of
improvements. Also, the time to fix for VRS is almost Transit/Doppler and GPS, manuscripta Geodetica 15
independent of the time-of-day. (1990), p. 17-11.

In addition, the horizontal and vertical position standard [3] Beutler, G., Schaer, S. and Rothacher, M.: Wide Area
deviations are given. The precision is improved Differential GPS, Study in the Context of AGNES.
marginally for the 16 km data. The improvement is large Printing Office, University of Bern, September 17,
for the 32 km data. As the night performance is worse 1999
than the day performance here, additional investigations
are needed to identify the reason for the strong [4] Collins, J.P. and Langley, R.B.: Possible Weighting
positioning improvement and the not anticipated Schemes for GPS Carrier Phase Observations in the
properties of the night data. A possible explanation is the Presence of Multipath, Geodetic Research
presence of strong tropospheric effects as the height Laboratory, Department of Geodesy and Geomatics
difference was 120 m between reference and rover station Engineering, University of New Brunswick, Canada.
here. For United States Army Corps of Engineers
Topographic Engineering Center, March 1999.
Last but not least, the errors column displays the number
of wrong fixes for each data set. Wrong fixes have only [5] Jonge de, P.J., and C.C.J.M. Tiberius (1996)
occurred on the long baseline in raw mode, proving that Integer ambiguity estimation with the LAMBDA
current RTK systems are already very reliable. method. Proceedings IAG Symposium No. 115, GPS
Nevertheless, VRS mode was able to eliminate the few trends in precise terrestrial, airborne and spaceborne
wrong fixes on the 32 km baseline, another important applications, XXI General Assembly of IUGG, July
result. 2-14, Boulder, CO, July 2-14, 1995, G. Beutler et.al.,
(Eds.), Springer Verlag, pp. 280-284.
CONCLUSIONS
[6] Jonge de, P.J., and C.C.J.M. Tiberius (1996)
The Virtual Reference Station technique has proven its The LAMBDA method for integer ambiguity
performance in several existing networks. The tests estimation: implementation aspects. Delft Geodetic
performed showed improvements of the performance Computing Centre LGR series, No. 12.
numbers important for the field user, i.e. time to fix and
position accuracy. [7] Vollath, U., Buecherl, A., Landau, H. Pagels, C.,
Wagner, B.: Multi-Base RTK using Virtual
REFERENCES Reference Stations, Proceedings of the ION-GPS
2000, Salt Lake City, Utah.
[1] Wild, U., Beutler, G., Gurtner, W. and Rothacher, M:
Estimating the Ionosphere using One or More Dual
Frequency GPS Receivers.Proceedings of the 5th
Parameter Errors TTF 50% [sec] TTF 90% [sec] Std.dev. Horiz. [mm] Std.dev.Vert. [mm]

Set-up [%] All Day Night All Day Night All Day Night All Day Night

16km Raw 0.0 55 80 50 >180 >180 145 19 20 17 25 29 22

16km VRS 0.0 15 15 15 65 75 60 17 19 16 24 26 22

32km Raw 0.3 115 110 120 >180 >180 250 57 56 57 47 47 51

32km VRS 0.0 75 70 80 115 105 120 22 21 23 28 28 29

Table 1 Performance evaluation results

474