Re-running editing and group delay resolution reveals that group delays at baselines with EFLSBERG have ambiguities at X-band around 0.3 ns as it seen at the plot of post-fit residuals. These ambiguities were not resolved correctly in the original analysis. Group delay ambiguities – these famous artifact of Fourfit algorithm – usually do not pose a problem since their spacings are normally at the range of 28&mfash;200 ns. Successful resolving group delay ambiguities with spacings 3.9 ns became routine. However, ambiguities with spacings 0.3 ns are unusual and their resolution poses a challenge. The ambiguities in the database posted at the IVS are resolved incorrectly. This causes an error in estimate of EFLSBERG position.
So, out of curiosity I decided to take the visibilities data provided by the correlator and re-process experiment anew. euro112 was processed with DiFX correlator that unlike to Fourfit, outputs the data in the standard FITS-IDI format which tremendously facilitates processing.
Using standard VLBI processing pipeline: coarse fringe fitting, adjustment of final complex bandpass, fine fringe fitting, computation of total group delay, computation of small differences between theoretical group delay and delay derived from fringe fitting, LSQ adjustments, outliers elimination, reweighting, second fringe fitting of outliers with narrow window, final LSQ adjustments, outlier elimination and reweighting I got a new solution. Details of group delay computation from visibility data can be found in L. Petrov et al, AJ, 142, 35.
Post-fit residuals of the new solution look very different. Ambiguities at baselines with EFLSBERG disappears. Data at baselines with CRIMEA restored.
Statistics of euro112 solutions using the same visibility data from two
analysis chains: DiFX → difx2mark2 → Fourfit → Calc/Solve and
DiFX → difx2fits → Pima → VTD/Solve.
| Statistics | Fourfit | PIMA (pcal) | PIMA (no-pcal)|
| # obs total | 6043 | 6200 | 6200 |
| # obs recoverable | 3909 | 5637 | 5637 |
| # obs used | 3274 | 5328 | 5549 |
| wrms of resid. (ps) | 59.3 | 28.3 | 27.8 |
Station EFLSBERG had phase calibration turned off first 9 kiloseconds in euro112 experiment. Therefore, its pcal extracted by the correlator for first 2 hours was a poor noise as it seen from this plot, and therefore, useless. I computed two complex bandpasses for EFLSBERG: for the period of time without pcal, and for the period of time after pcal was on. I had to evaluate clock function break for station EFLSBERG at the epoch of tuning pcal on.
Extraction of the phase-calibration signal and its formatting to FITS-IDI format was a relatively new feature in DiFX and the latest bug was fixed in 2012. Therefore, one may think that ambiguities in the Fourfit solution may be somehow related to phase-cal. However, reprocessing the correlator output without applying phase-cal, i.e. considering phase offsets between intermediate frequencies constant during entire experiment, did not change solution significantly. In fact, no-pcal solution appeared slightly better: 4% more points in the solution and 2% smaller residuals.
Station CRIMEA lost 5 IFs at X-band. This caused an increase of residuals to a 200-400 ps level. Position formal uncertainties are 30 mm for the vertical coordinate and 3 mm for horizontal components. This is not a great result, but it is hard to justify throwing the data entirely. Comparison of EFLSBERG position estimate using the IVS database and re-processed data shows the difference 11 mm in Up component, 13 mm in East component, and 1 mm in North component.
Reprocessing the data starting from the visibilities produces a dataset of group delays without ambiguities. There are several tens zero fringe rate points like that and this, but re-fringing with a narrow search window corrects this error of the fringe fitting algorithm.
Statistics of euro113 solutions using the same visibility data from two
analysis chains: DiFX → difx2mark2 → Fourtit → Calc/Solve and
DiFX → difx2fits → Pima → VTD/Solve.
| Statistics | Fourfit | PIMA (pcal) |
| # obs total | 4749 | 4755 |
| # obs recoverable | 4404 | 4626 |
| # obs used | 3688 | 4172 |
| wrms of resid. (ps) | 39.5 | 35.9 |
Why Fourfit solution has ambiguities? Because of phase-cal? Reanalysis of euro114 without using phase-cal produces substantially inferior solution. Phase-cal phase have peak to peak variations up to 1 rad at several stations, for instance, at YEBES40M. As a results, fringe fit is poor which causes an increase of the scatter, but not ambiguities.
Statistics of euro114 solutions using the same visibility data from two
analysis chains: DiFX → difx2mark2 → Fourtit → Calc/Solve and
DiFX → difx2fits → Pima → VTD/Solve.
| Statistics | Fourfit | PIMA (pcal) | PIMA (no-pcal) |
| # obs total | 5561 | 5609 | 5609 |
| # obs recoverable | 4712 | 4755 | 5637 |
| # obs used | 3405 | 4661 | 4608 |
| wrms of resid. (ps) | 35.2 | 23.8 | 31.1 |
It is worth mentioning that ambiguities spoil not only euro experiments, but other experiments as well, for instance t2079.
Last update: 2012.02.05_01:13:34
