Inspired by an exchange on the RPLS

Introduction

Inspired by an exchange on the RPLS.com board a group of six posters joined to process the same data using five different baseline processors and compare the results. The five participants are:

J. Anthony Cavell: Trimble Geomatics Office (TGO), WAVE processor float solution

Christian Clausen: Topcon Tools

Doug Copeland: Ashtech Solutions

Peter Lazio: PAGE-NT

Loyal Olson: Novatel SoftSurv

Mike Strutt: Trimble Geomatics Office (TGO), WAVE Processor fixed solution

Using data from four CORS sites, ANP1 (Annapolis, Md.), GODE (Goddard Space Flight Center, Md.), RED1 (Reedy Point, De.) and PSU1 (Penn State, State College, Pa.), three baselines were processed holding ANP1 fixed:

ANP1 - GODE: 18.9 km

ANP1 - RED1: 108.6 km

ANP1 - PSU1: 225.9 km

Each baseline was processed four times using session lengths of one hour, two hours, four hours and eight hours, for twelve baselines. The derived coordinates for GODE, RED1 and PSU1, from each session, were compared to the published coordinates and misclosures were computed. CORS were deliberately chosen with different antenna types to test the antenna phase center modeling of the processing software.

Discussion of Methods

Twenty-four hours of data collected on day 230 (August 18, 2003) at each CORS were downloaded from NGS' National CORS site. The original observations at ANP1 and RED1 were collected at five-second intervals; observations at PSU1 were collected at one-second intervals. Using TEQC, from UNAVCO (http://www.unavco.ucar.edu/software/teqc/teqc.html) the observations from ANP1, RED1 and PSU1 were decimated to 30-second epochs. Again using TEQC, the twenty-four hour observation files were windowed into four files:

1 hour: 01:00 - 02:00

2 hours: 01:00 - 03:00

4 hours: 01:00 - 05:00

8 hours: 01:00 - 09:00

Each participant received identical RINEX files. Included with the RINEX observation files were the RINEX navigation file (xxx2300.o3n), the IGS precise ephemeris (igs12321.sp3), the NGS data sheet, the coordinate sheet and the CORS log file.

NGS publishes National CORS coordinates referenced to both NAD83 and ITRF00. All processing was done using ITRF00 coordinates. ITRF00 was chosen as the reference frame to match the reference frame of the IGS precise ephemeris. The IGS precise ephemeris is a realization of the ITRF00 coordinate system at the epoch of the observations, in this case 2003.628. By international convention ITRF00 coordinates for the CORS are published at an epoch of 1997.0. Before processing any baselines the coordinates of the CORS were transformed to 2003.628 using the published velocities.

CORS		1997.00	2003.628
ANP1	Lat	39— 00' 37.03252" N	39— 00' 37.03331" N
	Lon	076— 36' 33.28941" W	76— 36' 33.29393 W
	h	20.250 m	20.253 m
GODE	Lat	39— 01' 18.21820" N	39— 01' 18.21883" N
	Lon	076— 49' 36.58375" W	076— 49' 36.58772" W
	h	14.524 m	14.511 m
RED1	Lat	39— 33' 41.22537" N	39— 33' 41.22594" N
	Lon	075— 34' 11.93140" W	75— 34' 11.93587" W
	h	-27.061 m	-27.061 m
PSU1	Lat	40— 48' 24.84041" N	40— 48' 24.84084" N
	Lon	077— 50' 59.27863" W	77— 50' 59.28284" W
	h	311.233 m	311.231 m

There was some confusion related to the coordinates of CORS GODE. The RINEX files for GODE indicate an antenna height of 0.0614 meters. There is a monument associated with the CORS GODE with the CORS ARP 0.0614 meters above this monument. After discussion it was decided that, because we were not directly comparing vectors but rather differences in coordinates, equivalent results would be obtained if one used the ARP coordinate with an antenna height of 0.000 or the monument coordinate with an antenna height of 0.0614.

Participants processed the twelve baselines using the default settings for the software. Each participant could optionally tinker with additional solutions to try to improve the results. All results were compiled into a table included below.

Results were evaluated by comparing the vertical and horizontal misclosures.

Summary of Processing

Ashtech Solutions

Doug Copeland

Ashtech Solutions returned a mixture of fixed, partially fixed and float solutions. Ashtech defines a fixed solution as one in which all ambiguities are fixed to integer values. If a subset of more than 50% of the possible ambiguities is fixed, Ashtech defines this solution as a partial solution. If fewer than 50% of the possible ambiguities are fixed to integers then the solution is defined as a float solution. Solution definitions will be revisited when the TGO solutions are discussed.

Baselines were processed with both precise and broadcast ephemeredes. Interestingly, the broadcast ephemeris often produced results closer to the published coordinates of the CORS than the precise ephemeris. Ashtech Solutions required the precise ephemeris for the previous day to process solutions using the precise ephemeris. Solutions using both the precise and broadcast ephemeris are included in the tabulated results below.

Novatel (SoftSurv)

Loyal Olson

With the exception of the one-hour session, Novatel returned fixed solutions for all the ANP1-GODE sessions. For the baselines ANP1-RED1 and ANP1-PSU1 Novatel returned float solutions for every session. Baseline were processed using the processor defaults. Additional processing was fine-tuned by increasing the mask angle and removing some SVs. These tuned solutions are included in the tabulated results.

PAGE-NT

Peter Lazio

PAGE-NT is a session baseline processor. Session processing maintains the correlations between baselines observed at the same time. PAGE-NT allows one to attempt to fix ambiguities to integer values or specify a float solution. Baselines were processed three times:

As sessions while attempting for fix ambiguities
As individual baselines while trying to fix ambiguities
As sessions with no attempt to fix ambiguities

Results from all three baseline solutions are included in the tabulated results below.

Only one solution fixed 100% of the ambiguities. The percentage of fixed integers in the partially fixed solutions ranged from 50% to 94%.

TGO Fixed Solutions

Mike Strutt

Processing the baselines using TGO with fixed solutions proved challenging. Only two of the twelve baselines returned results without warnings. The other ten baselines returned results with high RMS, low ratio or a combination of low ratio, high RMS and high reference variance. The one-hour, two-hour and four-hour baselines from ANP1 to RED1 failed to return a fixed solution.

Trimble defines a fixed solution as one in which at least three ambiguities have been fixed to integers. A minimum of four satellites are needed to fix a position in three dimensions. When processing baseline solutions, TGO combines observations from N satellites to form double difference observation equations with N-1 double difference ambiguities. Three integer ambiguities are therefore the minimum needed to fix a solution. A fixed solution can include float ambiguities, so long as at least three ambiguities are fixed to integers. As a corollary to this definition, a float solution may have two or fewer fixed ambiguities.

TGO Float Solution

J. Anthony Cavell

In TGO, when a float solution is specified the TGO does not attempt to fix any ambiguities to integers, all of the ambiguities are real numbers.

The four-hour data set was the most problematic. This data set would not process unless the mask angle was set to 17° and the processing flow out was tinkered with. For consistency, all TGO float solutions were processed with a 17° mask angle. As with Ashtech Solutions, the IGS precise ephemeris for the previous day was required to process the baselines.

Topcon Tools

Christian Clausen

Topcon Tools returned a mixture of fixed, partial and float solutions. For the four sessions to GODE Topcon Tools used the Ionospheric Free observable with fixed for partial solutions. All the baselines to RED1 and PSU1 were processed using the Wide Lane observable with partial or float solutions. Topcon Tools was the only processor to use the Wide Lane observable. The operator is not able to choose the solution type.

Topcon Tools was the only processor that did not return a solution for the one-hour session ANP1-PSU1 baseline. One hour of data is not enough to resolve a 226-kilometer baseline reliably. Topcon Tools may have a built in protection against poor solutions.

Discussion of Results

This experiment is not a scientifically valid study. Too many variables that can affect GPS performance; four sessions for three baselines on one day is not enough to data to isolate systematic errors in the data or processing. Therefore, this discussion of results is qualitative rather than quantitative.

Graphs of the horizontal and vertical misclosures are included in this report. As discussed in the processing section, multiple solutions were performed using Ashtech Solutions, Novatel and PAGE-NT. Only one solution per session is included in this analysis. This was done to simplify the graphing and analysis. For Ashtech Solutions the solution using the precise ephemeris was selected, for Novatel the default processing solution was selected and for PAGE-NT, the partially fixed session solution was selected. These solutions were selected on the assumption that they would be the most likely to be selected in an actual survey.

Misclosures

The greatest spread of vertical misclosures occurred at the one-hour session to RED1 where the misclosures ranged from -0.106 to 0.159 meters a spread of 0.265 meters. The smallest range was for the four-hour session to GODE where the range was from -0.065 meters to 0.017 meters, a spread of 0.082 meters. A number of factors may contribute to the differences observed in the vertical results. Among these factors are treatment of tropospheric delays and antenna modeling. No investigation into how the various processors handled these factors was done during this experiment.

The greatest spread in horizontal misclosure occurred at the two-hour session to PSU1. Here the horizontal misclosure ranged from 0.016 meters to 0.505 meters for a spread of 0.489 meters. It should be noted that the 0.505-meter misclosure was for a TGO fixed solution with multiple failures. Interestingly, the TGO fixed solution with a horizontal misclosure of 0.505 meters had a vertical misclosure of -0.058 meters. The second greatest range of horizontal misclosure is for the one-hour session to PSU1 where the horizontal misclosures ranged from 0.099 meters to 0.386 meters for a spread of 0.286 meters, 40% less than the greatest spread. The smallest range of horizontal misclosures occurred for the eight-hour session to GODE. Here the spread ranged from 0.012 meters to 0.017 meters for a total range of 0.005 meters.

Ashtech Solutions

Ashtech Solutions produced consistent vertical misclosures for the GODE vector. For the longer RED1 vector the vertical misclosures decrease as session length increases but were generally greater in magnitude than the other processors. At PSU1 Ashtech Solutions vertical misclosures actually increased as the session length increased.

Horizontally, Ashtech Solutions produced consistent results for all session lengths for the vector to GODE. Results were more erratic for the vectors to RED1. For the vector to PSU1, the misclosure steadily improved from two hours to four hours.

Novatel

With the exception of the one-hour session to PSU1, misclosures, both vertical and horizontal misclosures improved with longer session lengths. The biggest improvement occurred from the one-hour session to the two-hour session. At PSU1, the vertical misclosure increased from the one-hour session to the two-hour session. The horizontal misclosure, after dropping from the one-hour session to the two-hour session increased again to the eight-hour session.

PAGE-NT

In general, PAGE-NT needs two or more hours of data to solve for tropospheric parameters reliably. This accounts bouncing of the misclosures in the one and two-hour sessions. There is consistent improvement of the misclosures from the two-hour session to the four-hour session. Horizontally, for the one and two-hour sessions Ashtech Solutions, TGO Fixed and Topcon Tools all produced better misclosures at GODE and RED1

TGO (Float and Fixed)

Vertically, the fixed solution misclosures were smaller vector to GODE, while at PSU1 the float solution misclosures were smaller. At RED1, the solutions were roughly equivalent. Interestingly at RED1, the vertical misclosures for the fixed and float solutions converged toward zero from opposite sides as the session lengths increased. The ellipsoid heights derived from the TGO float solution are consistently higher than the published height for all CORS for every session. There is not enough data available to determine if this is an anomaly or a genuine bias in the processor.

Horizontally, TGO float solutions misclosures consistently decreased at all CORS as session lengths increased from two to four-hours and then increased slightly for the eight-hour session.

The trends shown for the fixed and float solutions indicate that for long baselines the TGO float solution is a more reliable solution while for shorter baselines the fixed solution is better. There is not enough data to determine where the cut off between fixed and float solutions should be. The vertical misclosures at RED1 seem to indicate that the two solutions are equivalent at 109 kilometers. However, the sudden spike in horizontal misclosure for the eight-hour session to RED1 contrasts with the nominal increase in the float solution for the same session.

Topcon Tools

At GODE, the vertical misclosures were consistently greater than the other processors. At RED1, the vertical misclosures decreased with session length while at PSU1the vertical misclosures increased with session length. The ellipsoid heights derived from the Topcon Tools baseline solution are consistently lower than the published ellipsoid heights for all CORS for all sessions. Once again, there is not enough data to determine if this is an anomaly or a genuine bias.

Horizontally Topcon Tools produced results consistent with the other processors. In general the misclosures decreased as session length increased.

Conclusion

Although this was not a scientifically significant experiment, the participants learned much. Baseline processors using the exact same data produced results different by as much as 0.50 meters horizontally and 0.26 meters vertically. No effort was made to determine why such disparities occurred.

Appendix (For addition of Leica SkiPro 3.0)

November 30, 2003

Contributed by Shelby Griggs

SKI-Pro has a default limit of 80 km for fixed baselines. In the interest of testing the default parameters, no attempt was made to modify the default settings. Fixed solutions were returned for all data between ANP1 and GODE. The other two vectors were over 80 km and thus returned float solutions. No tuning of the data was attempted, such as deleting satellites, windowing data, changing mask angles, etc. As was true with Ashtech Solutions and TGO, the IGS precise ephemeris for the previous day was required to process the baselines if the precise ephemeris option was selected. With dual frequency data, SKI-Pro also produces an L3 (ionosphere free linear combination) solution for all lines over 15 km in length, regardless of a fixed or float solution.

Initial results did not conform well to the published ellipsoid heights. This was a consequence of using incorrect antenna modeling. SKI-Pro uses the default NGS names for the various antennas. Unfortunately, if there is not an exact match in the antenna database (as imported from NGS) a close match is selected (i.e.: SNOW vs. a non SNOW dome). In the case of no match, a new antenna is created with the name as in the RINEX file. The new antenna will have zero offsets. The key to correct and automatic antenna recognition is having the antenna name in the RINEX file exactly match the defined antennas in the NGS database. If the RINEX header is correct at the time of import than all works fine, otherwise the antenna needs to be edited after import, which will work too. The lesson learned from this experiment is that the antenna models do matter, and additionally exact naming is important.