Comparison of rdv-11 experiment processed by WACO+Fourfit versus and VLBA+AIPS
Contents:

Introduction

Experiment rdv-11 was processed by VLBA correlator and then AIPS (databases $98OCT01X7/$98OCT01S7), the tapes from 6 stations for this experiment, GILCREEK, KOKEE, LA-VLBA, MK-VLBA, NRAO20, SC-VLBA, were processed by Mark-3a correlator in Washington and then by Fourfit (databases $98OCT01X6/$98OCT01S6).

It was noticed long time ago that solutions of the databases processed by VLBA correlator and later AIPS have some peculiarities. Namely, it was noticed that observations of southern sources has additional noise. This effect is rather small, but irritating.

Investigation of the differences in delays and in solutions of the databases created by Mark-3 + fourfit (thereafter WACO) and VLBA + AIPS (thereafter VLBA) was aimed to disclose the reason of this peculiarity.

Comparison of solutions

First, databases were re-Calced, group ambiguities resolved once more, outliers eliminated and baseline-dependent re-weighting parameters were recomputed once more. Usual quick-mode solutions were produced: However, the databases had different number of observations, different set of suppressed observations, slightly different re-weighting parameters what made difficult direct comparison. A set of programs was developed which
  1. Computes differences of X-band and S-band group delay: VLBA-WACO and pouts them in the slot of User calibration.
  2. Computes suppression status by such a manner that only the observations which were not suppressed in both VLBA and WACO solution.
Then this program was run in user calibration mode using WACO database. Two solutions were made. The first listing was produced in the solution when user program rdv_1 ran, but user calibration was not applied. It means that WACO group delays were used, but only those observations which were not suppressed earlier in either WACO or VLBA solution were used. The second listing was produced in the solution when user mode calibration was applied. It means that parameterization, weights, suppression status was exactly the same as in the previous solution, except the fact that VLBA group delays at X- and S-band were used.

Differences in postfit residuals

WACO | VLBA | Overall fit: 31 ps | 28 ps ---------------------------|----------- GILCREEK-KOKEE 56 ps | 45 ps GILCREEK-LA-VLBA 28 ps | 24 ps GILCREEK-MK-VLBA 22 ps | 19 ps GILCREEK-NRAO20 26 ps | 24 ps GILCREEK-SC-VLBA 41 ps | 36 ps KOKEE -LA-VLBA 51 ps | 45 ps KOKEE -MK-VLBA 44 ps | 37 ps KOKEE -NRAO20 46 ps | 40 ps KOKEE -SC-VLBA 58 ps | 53 ps LA-VLBA -MK-VLBA 24 ps | 20 ps LA-VLBA -NRAO20 23 ps | 24 ps LA-VLBA -SC-VLBA 41 ps | 40 ps MK-VLBA -NRAO20 28 ps | 25 ps MK-VLBA -SC-VLBA 47 ps | 40 ps NRAO20 -SC-VLBA 41 ps | 34 ps VLBA data produced about 10% better fit for 14 out of 15 baselines.

Differences in baseline lengths (mm)

WACO | VLBA | GILCREEK KOKEE 4728081870.39 -+ 4.52 | 4728081866.19 -+ 4.05 GILCREEK LA-VLBA 4157847360.67 -+ 2.51 | 4157847359.17 -+ 2.25 GILCREEK MK-VLBA 4923018344.09 -+ 3.26 | 4923018339.74 -+ 2.92 GILCREEK NRAO20 5035085018.84 -+ 2.97 | 5035085005.38 -+ 2.66 * GILCREEK SC-VLBA 7403221566.01 -+ 5.66 | 7403221562.27 -+ 5.08 KOKEE LA-VLBA 5199804060.73 -+ 3.88 | 5199804061.12 -+ 3.48 KOKEE MK-VLBA 507886504.67 -+ 2.24 | 507886506.37 -+ 2.01 KOKEE NRAO20 7209144111.64 -+ 5.36 | 7209144093.61 -+ 4.81 * KOKEE SC-VLBA 8856977501.65 -+ 8.20 | 8856977500.93 -+ 7.36 LA-VLBA MK-VLBA 4970071720.64 -+ 2.47 | 4970071719.35 -+ 2.22 LA-VLBA NRAO20 2345872620.87 -+ 1.61 | 2345872616.41 -+ 1.44 * LA-VLBA SC-VLBA 4458991013.35 -+ 3.17 | 4458991013.28 -+ 2.84 MK-VLBA NRAO20 7029977086.98 -+ 3.66 | 7029977067.58 -+ 3.28 * MK-VLBA SC-VLBA 8611584641.52 -+ 6.61 | 8611584638.54 -+ 5.92 NRAO20 SC-VLBA 2708415909.58 -+ 2.68 | 2708415899.77 -+ 2.41 * Baseline length for 14 out of 15 baselines are shorter in VLBA solution when in WACO solution. Baselines with recently deceased station NRAO20 shows significant shrinking in VLBA solutions.

Excessive noise for southern sources

Ratios of partial weighted sum of squares of residuals over all used observation of each source to its mathematical expectation (chi/ndg) were computed for WACO and VLBA solutions. More details about this technique can be found in this memo. I don't see significant increasing noise for southern sources.

Comparison of raw differences in delay

Differences in group delay VLBA-WACO (nsec) as a function of time (time in hours elapsed from the nominal start of the session) I made also plots of differences in delays versus But I didn't find anything interesting.

Comparison of delay misclosures

Misclosures in group and phase delays for triangle LA-VLBA/MK-VLBA/NRAO using WACO and VLBA data were computed. Of course, correction for differences in reference time epochs was accurately applied. Group delay misclosure are in psec, phase delay misclosures are in phase turns. WACO data are presented by green color, VLBA data are presented by blue color. Only observations which were selected in solution for both WACO and VLBA data are shown.

Comparison of residual phases

Residual phases after group delay fitting were computed for two baselines: LA-VLBA/MK-VLBA and MK-VLBA/NRAO20, for X-band: and for S-band

Conclusion

Facts:
  1. Differences in group delay VLBA-WACO are noise-like with average more than 10 nsec and peak-to-peak variations about 100 psec.
  2. But they are not noise!! (Noise, being added or subtracted from good data, would increase wrms in both cases, but actually wrms is increased only if we subtract the differences.)
  3. VLBA data fits 10% better using the same parameterization, suppression, weights.
  4. Difference WACO-VLBA are baseline-dependent. Group delay misclosures for VLBA data are noticeably worse at S-band.
  5. Residual phases of VLBA data are wrong. (sum of residuals for each moment of time should be zero).
Differences WACO-VLBA are 100 times larger than we expected. These differences are not of geometric origin or correlator model (geometric model conserves misclosure). I found nothing that would support a hypothesis about predominant increasing noise in VLBA data for southern sources. I think that probably there is no a simple solution of finding the origin of discrepancies. I recommend tedious, step by step investigation of Fourfit/AIPS and comparison of intermediary results.

Back to Leonid Petrov's discussion page.
This page was prepared by Leonid Petrov

Last update: 14-AUG-2000 20:20:31