VLBI Data Analysis System for the Key Stone Project

Yasuhiro Koyama
Kashima Space Research Center
Communications Research Laboratory
893-1 Hirai, Kashima, Ibaraki 314, Japan

1. Introduction

The data analysis software package of the Key Stone Project, which was nicknamed "Takemikazuchi" (---"Takemikazuchi" is the name of a mythological god who is symbolized by thunderbolt. Kashima Shrine located near the Kashima Space Research Center is the home of this god, and the gigantic "Key Stone" was put on the head of a huge catfish by him long time ago to prevent hazardous earthquakes which had been caused by the catfish, so it was told.---), is characterized by its highly automated design and its operator friendly Graphical User Interface. The software runs on a UNIX workstation (Figure 1) at Koganei Central Station located in the headquarters of the Communications Research Laboratory.

Figure 1. HP9000 model 715/100 workstation used for the Keystone VLBI data analysis.

The software utilizes CALC (Version 8.1) and Mark-III Database Handler software developed by Goddard Space Flight Center of National Aeronautics and Space Administration (NASA). Since all the processed data from a Key Stone VLBI experiments are stored in a Mark-III database, the results can be exchanged with other institutes participating geodetic VLBI activities worldwide. Figure 2 illustrates the data flow of the data analysis software.

Figure 2. Flow chart of the Key Stone Project VLBI data analysis.

As soon as the bandwidth synthesis processing (KOMB) is finished on all the observations taken in a daily VLBI session, 'KOMB' output files are organized and two databases (X-band and S-band) are created from these files. S-band database is only used for correlated amplitude monitoring purposes, and on the other hand, X-band database is used for data analysis in the following steps. The X-band database is prepared for data analysis by the program DBUPDATE. DBUPDATE stores various a-priori information in the database. Next, (CALC) is executed on the X-band database to calculate theoretical delay and delay rates and their covariance matrices. At this point, the X-band database becomes ready to be analyzed by the least-square parameter adjustment program VLBEST. At first, VLBEST runs with only clock offsets and their rates of change at all stations except for a reference station. Kashima station is always chosen as the reference station whenever the station participated the session. From the results, REMAMB resolves delay ambiguities. Next, VLBEST runs again with the same sets of estimate parameters on the ambiguity-resolved data. From the results, MRKOBS flags out bad data points which have large residuals. Finally, VLBEST runs again and the full set of estimation parameters are adjusted. These parameters are: clock offsets and their rates of change at all stations except for the reference station, wet component of tropospheric delay (three hour intervals) at all stations, and site coordinates of all stations except for the reference station. Unless the root-mean-square of the residual delays exceeds a certain threshold (100 psec), the estimated results are soon released in a variety of data types and formats. If the RMS residual exceeds the threshold value, an e-mail message is sent to a responsible operator to notify the problem. The operator investigates the data closely and tries to resolve the problem. "Takemikazuchi" provides Graphical User Interface to support these tasks for the operator. The operator can remove or recover any data point, insert epoch points for tropospheric delay and clock parameters, and change the estimation strategy very easily.

In addition to the normal data analysis procedure just described above, results are revised automatically upon receipt of the weekly and monthly bulletins of the International Earth Rotation Service (IERS), i.e. Bulletins A and B respectively. This is necessary since Earth Orientation Parameters (EOP) available at the time of the first data analysis are, in fact, predicted values based on the data in past. The predictions have large uncertainties, especially the last Bulletin A is almost one week old. The uncertainties of the predictions are estimated as 3.8 mas for terrestrial pole positions and 1.5 msec for UT1-UTC at prediction length of 10 days (IERS, 1995). These uncertainties correspond to the site position estimation error of 2.0 mm and 12.0 mm respectively for the baselines with distance of 110 km, which is roughly the distance between Kashima and Koganei stations. All the e-mail messages received by the Takemikaduchi system are analyzed and the EOP data file is updated according to the Bulletin data if the e-mail was either Bulletin A or Bulletin B. Whenever the EOP data file is updated, all the affected databases are also updated and processed for data analysis. As such, the same database is repeatedly analyzed with updated EOP values at least four times (two times with Bulletin A values and two times with Bulletin B values).

2. Comparison of Results with Conventional Systems


            Table 1. List of experiments used for the comparison.
-------------------------------------------------------------------------------
Date of      (# of data actually used in the analysis) / (# of data correlated) 
Experiments    (1) KSP Automatic     (2) KSP Manual       (3) K-3 Correlator 
                   Analysis              Analysis             + SOLVE
-------------------------------------------------------------------------------
1995.12.3           58/86                 78/86                97/101 
1995.12.4           61/82                 74/82                97/101 
1995.12.5           79/89                 80/89                94/100 
1995.12.6           78/84                 80/84                95/98 
1995.12.8           71/77                 72/77                94/99 
-------------------------------------------------------------------------------

To validate the results obtained by the Key Stone Project VLBI data analysis system, the estimated site position of the Koganei VLBI station was compared with the results obtained by other conventional methods. Table 1 gives the list of experiments with the number of correlated observations and actually used data for analysis. In the table, (1) and (2) are the results obtained using the KSP correlator, whereas (3) are the results obtained using the K-3 correlator. For the least-squares parameter adjustment, VLBEST was used in (1) and in (2), whereas SOLVE was used in (3). Differences between (1) and (2) are due to whether bad data are removed (1) automatically by the program MRKOBS or (2) interactively by using the Graphical User Interface.

The number of correlated observations in columns (1) and (2) are same since they are based on the same databases. These numbers are always less than the numbers in column (3) because there were still miscellaneous problems during the automatic correlation processing. The numbers of data actually used in the analysis in the column (1) are always less than those in column (2). This is, perhaps, because the threshold of delay residuals in the program MRKOBS is too small, and hence some valid data are also removed from the later analysis. Figure 3 shows the estimated site positions of Koganei VLBI station for five experiments in the horizontal plane. Uncertainties are expressed as one-sigma standard deviation error ellipses. Figure 4 shows the similar comparison for the vertical axis. These comparisons do not indicate statistically significant differences between the different approaches of data analysis, as expected.

Figure 3. Estimated site position of Koganei VLBI station compared in the horizontal plane.



Figure 4. Estimated site position of Koganei VLBI station compared in the vertical axis.

3. Effects of EOP Prediction Uncertainties

To evaluate the effects of prediction uncertainties of EOP values, the results were compared with different IERS bulletins. The experiment done on November 7, 1995 was chosen for the comparison. The experiment was first analysed on November 8, 1995 by using EOP predictions released in the Bulletin A issued on November 2, 1995. The EOP values are revised three times, first by Bulletin A issued on November 9, 1995, and hen two Bulletin B issued on December 8, 1995 and on January 3, 1996. Figure 5 shows the difference of EOP values according to the final value in the Bulletin B issued on January 3, 1996.

Figure 5. Comparizon of Earth Orientation Parameters (wobble of Earth's rotation pole and UT1$-$UTC) in the different IERS bulletins.

As seen in the Figure 5, the differences between predicted and revised values increase, as the prediction is made for many days in the future. As the EOP values are revised in the following IERS bulletins, the values converge toward the final values. Figure 6 shows the four different site position estimates of Koganei VLBI station compared in the horizontal plane. Each ellipse represents a one-sigma standard deviation uncertainty estimated using different EOP data sets. Figure 7 compares the same results on the vertical axis.

Figure 6. Estimated site position of Koganei VLBI station compared in the horizontal plane.



Figure 7. Estimated site position of Koganei VLBI station compared in the vertical axis.

All the results except for the first one estimated using the predicted EOP values agree with each other. The agreement is so good that the error ellipses in the horizontal plane and error bars in the vertical axis are hardly distinguishable. On the other hand, the first results differ from the other three by about 12 mm. Since the final target of the Key Stone Project is an accuracy of 2 mm for horizontal plane and 5 mm for vertical axis, the difference of 12 mm is quite serious. Therefore, it is important to note that the site coordinates estimated right after the experiment is finished may have large systematic error due to the insufficient prediction accuracy. In the first analysis, baseline results should be used mainly. Three dimensional site coordinates are reliable from the second analysis results. It is also important to note that the results of the second analysis are in fact good enough, and such results can be obtained within a week.

4. Summary

Daily VLBI observations with three stations, i.e. Kashima, Koganei, and Miura stations, started on December 1, 1995. As a result of automation-oriented system design, the data processing and analysis are almost automatically done without much effort. Daily observations begin at 1:00 UT and end at 6:30 UT. The recorded data tapes at Kashima and Miura stations are transported to Koganei Central station at 8:00 UT, and arrive the next morning. Correlation processing finishes for all the observations by 6:00 UT, and the analysis results become available by 6:30 UT. Thus, all the process tasks are finished within 24 hours from the time of the last observation.

The comparison of the results with other conventional results showed no significant differences. From this result, the data analysis system of the Key Stone Project is confirmed in its functionality. To prevent unnecessarily removal of valid data, the algorithm of MRKOBS should be improved.

The effects of prediction uncertainties of Earth Orientation Parameters were also examined. The results indicated that the three dimensional site coordinates estimated in the first data analysis needs to be used cautiously. When high accuracy and reliability of the results are required, as in the case of the Key Stone Project, only the baseline lengths after the initial analysis, or the complete results after the second analysis should be used.

Reference

IERS Annual Report for 1994, Central Bureau of IERS - Observatory of Paris, 1995




Updated on June 25, 1995. Return to CONTENTS