Diagnostic plots of various data sets
This space is meant for various plots and their interpretation.
L45786
Calibration solutions
AIPS plots [ produced by widemap:~wucknitz/astro_wm/lofar_obs_wm/L2012_45786/do_something.py ]
This file shows amplitude solutions (high numbers meaning low sensitivity!) on short timescale to check station performance (pages 1-12), delay solutions (pages 13-24), rate solutions (pages 25-36) and R-L differences of phase solutions (pages 37-42). The latter corresponds to differential Faraday rotation. Note the varying scales and some outliers.
Colours are for the four IFs (each consisting of 31 subbands). Red is low, blue is high frequency. Reference antenna is RS106. Model is a point source with 20 Jy and spectral index -3. Only international baselines were used.
Discussion
Using the best superterp stations as standard, we find amplitude gains between 45 and 60. Given the number of elements, this would correspond to 23-30 for RS and 12-15 for international stations. Strong deviations are found for the following stations: CS026HBA0 (80-150), CS026HBA1 (85-110), CS201HBA1 (100-170), CS501HBA1 (130-160), DE601 (40-170), DE604 (up to > 1000 and very erratic in general), RS306 (45-120), RS307 (45-100), RS503 (45-100). Significant DFR is found in all international station, particularly in FR606, SE607 and UK608.
The station calibration is obviously far from perfect for many stations.
Phase jumps
The jumps I saw are around 20:12:35, and I see them for stations DE602,DE603,UK608, e.g. relative to RS106 (which is consistent with most other stations). The jumps are frequency dependent, corresponding to a delay of a few nsec. The plot shows correlation XX on baseline RS106-UK608, subbands 200-207 averaged.
I analysed this further some time later. The plots in this PDF file are used to find phase jumps for all stations relative to RS106,RS208,RS406 and for some relative to UK608. Here are examples for three baselines:
The upper panels show phase differences between 4sec integrations. This is mostly noise, but sometimes we see “real” jumps. The lower panels show the (group) delays fitted over all 244 subbands. This is slightly affected by dispersion, but this does not cause jumps. We see that the “real” phase jumps come with jumps in delay by a few nsec.
Note that all plausible jumps happen at the same time, around 3500sec (3455 sec after start, UTC ca. 20:12:35).
Analyses of the jumps at this time are collected in this PDF file. The plots corresponding to the figures above are reproduced here:
These plots show phase jumps (measured near the mean frequency) as function of frequency. The straight lines are linear fits (delays). Note that the fit only uses delays, there is no additional constant phase offset. In other words the curve would go through the origin (0 deg at 0 MHz). The fact that these curves fit very well tells us that we are really seeing delay jumps. Note that on the baseline DE602-UK608 the delay jump is so large that the phase jump wraps around and is very small.
Conclusion
There are delay jumps of several nsec in a number of international stations relative to the Dutch stations (which are consistent with each other). The jumps happen at exactly the same time, which speaks against the hypothesis that they are directly related to the 5 nsec problem.
Approximate numbers relative to RS106 (measured in baseline XXX-RS106):
DE601 ca 1 nsec DE602 3.750 nsec DE603 2.95 nsec DE604 has other problems DE605 0.45 nsec FR606 -0.42 nsec SE607 0.6 nsec UK608 -3.48 nsec some other combinations: DE601-UK608 4.5 nsec DE602-UK608 7.62 (makes a full turn!) DE603-UK608 6.9 nsec DE605-UK608 4.01 nsec
The numbers seem to be related to the geographical position: Highest positive values are in the south-east, highest negative values in the west. Could this be a problem of the delay model in the correlator? Does it introduce jumps of the order of 1 metre (several nsec) over long baselines?
[ widemap:~wucknitz/astro_wm/
lofar_obs_wm/L2012_45786/JUMPS/ ]
Problem reported to sciencesupport@astron.nl.
— Olaf Wucknitz 05-Apr-2012 13:40
Comparison with delay model
Jan David Mol provided a file with model delays that are applied in the correlator. I subtracted linear (in time) slopes and plotted the residuals. Results for a selection of stations can be found here. There are indeed significant delay jumps at the green line (20:12:37.5) for international stations. Very rough estimates from the plots:
RS106 0.07 DE601 1 DE602 4 DE603 3 DE604 2.5 DE605 0.5 FR606 ~0 SE607 ~0 UK608 -3.5
These numbers are at least similar to the ones found above.
The additional saw-tooth pattern for some of the stations is not understood yet. When plotting vs. sample number instead of time, the saw-tooth pattern is reduced a lot for most stations.
It really looks as if there is a problem with the correlator delay model.
— Olaf Wucknitz 06-Jun-2012 12:50
Superterp
Determination of phase offsets between superterp stations with astro_wm/lofar_obs_wm/check_phases_superterp.py. This works by measuring phase differences between the individual station and the combined superterp as measured on baselines to a distant station (DE602 in this case). The same code also fits the phases to maximise the total signal of the combined superterp. The plots and numbers are for subbands 110-125, but they are not much different for other bands.
Values (in degrees): offsets = dict ( CS002HBA0=[10.026,4.894], CS002HBA1=[-11.652,-15.422], CS003HBA0=[-13.620,-12.170], CS003HBA1=[3.603,-1.051], CS004HBA0=[38.064,24.855], CS004HBA1=[25.266,5.727], CS005HBA0=[-10.189,-7.179], CS005HBA1=[-22.856,-14.301], CS006HBA0=[9.370,24.458], CS006HBA1=[-61.738,-41.777], CS007HBA0=[1.100,-3.310], CS007HBA1=[22.888,28.733], )
These are inputs for phase_up3.py. (The final numbers are slightly different.)
— Olaf Wucknitz 25-Mar-2012 20:49
Direct fits of offsets: phases and delays
Later I corrected for delays and phases between superterp stations. These are now direct fits over the entire observation. [ widemap:~wucknitz/astro_wm/lofar_obs_wm/L2012_45786/SUPERTERP ]
Fits via external station DE602:
The full-resolution version of this plot is available together with similar fits for reference stations DE605,SE607,UK608,RS508,RS509 in a PDF file. For international stations the results are very similar. For Dutch stations, we still see some effects of the source structure.
After correcting for the offsets, I made another fit as a cross-check:
(Or more plots in full resolution as PDF file.)
Here is some additional info for the fits:
# fit for delays and phases
# good data points: 31071412, parameters: implicit 2622894 + explicit 22
# chi^2 no correction 3.212521e+07 with correction 2.329588e+07
# diff 8.829331e+06 diff/22 4.013332e+05
# diff relative to phases 9.848838e+04 diff/11 8.953489e+03
# reduced chi^2 no correction 1.129241e+00 with correction 8.188790e-01
# (error bars assuming reduced chi^2 of unity)
# phases in deg and delays in nsec (phases XX and YY, delays XX and YY):
# reference frequency in MHz (fitted over 114.966-162.573 MHz)
phase_delay_offsets= dict (
reffreq= 138.769531,
CS002HBA0= [ 10.728372 , 5.595891 , 0.203229 , 0.363467 ],
CS002HBA1= [ -12.226986 , -15.731574 , -0.356111 , 0.240983 ],
CS003HBA0= [ -12.776805 , -11.840582 , -0.205941 , -0.038600 ],
CS003HBA1= [ 5.296623 , -0.581472 , -0.108467 , 0.346829 ],
CS004HBA0= [ 39.052095 , 25.414495 , 0.674259 , -0.060649 ],
CS004HBA1= [ 26.677695 , 6.676455 , 0.253863 , -0.183799 ],
CS005HBA0= [ -9.663662 , -6.778751 , -0.095212 , -0.247486 ],
CS005HBA1= [ -21.756341 , -13.219949 , -0.141094 , -0.617608 ],
CS006HBA0= [ 9.597167 , 25.344461 , 0.354160 , 0.702026 ],
CS006HBA1= [ -61.627535 , -41.753241 , -0.890461 , -0.866817 ],
CS007HBA0= [ 2.359800 , -2.205662 , -0.222073 , -0.003485 ],
CS007HBA1= [ 24.339576 , 29.079928 , 0.533848 , 0.365139 ],
)
# phase_delay_errors= dict (
# CS002HBA0= [ 0.016306 , 0.015814 , 0.003141 , 0.003049 ],
# CS002HBA1= [ 0.016509 , 0.016449 , 0.003195 , 0.003187 ],
# CS003HBA0= [ 0.016401 , 0.015892 , 0.003164 , 0.003069 ],
# CS003HBA1= [ 0.016504 , 0.016414 , 0.003178 , 0.003168 ],
# CS004HBA0= [ 0.016383 , 0.015891 , 0.003161 , 0.003074 ],
# CS004HBA1= [ 0.016100 , 0.016332 , 0.003108 , 0.003155 ],
# CS005HBA0= [ 0.017283 , 0.017326 , 0.003352 , 0.003364 ],
# CS005HBA1= [ 0.016460 , 0.015729 , 0.003176 , 0.003041 ],
# CS006HBA0= [ 0.016386 , 0.016543 , 0.003169 , 0.003199 ],
# CS006HBA1= [ 0.016185 , 0.015573 , 0.003118 , 0.003004 ],
# CS007HBA0= [ 0.019326 , 0.019308 , 0.003809 , 0.003791 ],
# CS007HBA1= [ 0.016553 , 0.015859 , 0.003194 , 0.003069 ],
# )
— Olaf Wucknitz 05-Apr-2012 16:40
In fits for X-Y offsets I found that the weights produced by phase_up4.py are overestimated. It turned out that this is due to correlations of the noise between superterp stations. This means that a real source in the superterp phased-up beam contributes significantly to the noise. This noise is then obviously correlated, because the intra-superterp correlations are significant. In phase_up5.py this is now taken into account. This version also computes the autocorrelation of the phased-up superterp and uses them (actually the contributions from auto and cross correlations) to correct the weights of all baselines with the superterp. Originally these weights were computed from the weights of the combined baselines alone, which relies on uncorrelated noise.
In the case of L45786 the weights are scaled down by a factor of 0.33=1/3.03 with respect to the previous version. This implies that the coherently added superterp is not 12 times more sensitive than the individual station but only 12/3.03 or 4 times.
This effect is so strong, because TauA itself contributes to the noise. In the case of the phased-up superterp this “wave noise” does actually dominate.
— Olaf Wucknitz 12-Apr-2012 22:05
L62844
This is a short (~ 1min) observation of 3C295 to test phasing up the superterp, taken by George Heald.
I ran it through lbw_fit_phases_superterp.py in order to determine offsets
between the stations. Here are the
plots for all RS+IS external stations. Note that the international stations
are a bit too weak to serve as reference stations. RS307 seems to be quite
good. All numbers of the fits can be found
here.
Here are values derived from the RS307 fits (page 11 of the pdf) derived for 126 MHz (about the frequency of SB019), approximately relative to CS002HBA0:
CS002HBA0 X const -0.046494 linear 0.158208 George 0 CS002HBA0 Y const -0.099445 linear -0.216914 0 CS002HBA1 X const -21.826880 linear -19.941349 -23.15983494 CS002HBA1 Y const -21.800689 linear -20.110180 -23.070956 CS003HBA0 X const 45.107360 linear 41.949859 29.51807725 CS003HBA0 Y const 38.681503 linear 35.436110 23.11166696 CS003HBA1 X const 39.702733 linear 37.475449 26.37202499 CS003HBA1 Y const 33.785169 linear 30.763441 19.73965005 CS004HBA0 X const 69.704557 linear 64.021061 64.85530977 CS004HBA0 Y const 52.530494 linear 50.514153 51.71035217 CS004HBA1 X const 48.239319 linear 45.339133 43.11659232 CS004HBA1 Y const 33.763469 linear 33.831065 32.38522569 CS005HBA0 X const 29.657276 linear 28.820131 36.55736923 CS005HBA0 Y const 34.439021 linear 33.707089 4177547539 (41.77?) CS005HBA1 X const 29.327615 linear 27.234263 25.51752009 CS005HBA1 Y const 32.069381 linear 30.122757 29.24410287 CS006HBA0 X const -23.800379 linear -21.769515 -23.78864541 CS006HBA0 Y const -2.679169 linear -2.277020 -4.23214965 CS006HBA1 X const 8.658124 linear 8.714773 5.05917215 CS006HBA1 Y const 28.552542 linear 26.751832 24.10707984 CS007HBA0 X const 52.668600 linear 52.902507 39.11122934 CS007HBA0 Y const 51.811068 linear 50.283647 35.96414392 CS007HBA1 X const 65.808169 linear 61.495488 49.0431901 CS007HBA1 Y const 62.146657 linear 57.594019 45.86857076
const is fitted as constant offset over all frequencies, linear is the
linear fit at 126 MHz. The last column shows George Heald's BBS fits for
SB019.
For most stations the numbers are very similar, but not for CS003HBA0, CS003HBA1, CS005HBA0, CS007HBA0, CS007HBA1, as found by George already in an earlier version.
Phasing up with my corrections results in a weight factor of 0.762383 due to correlated noise.
— Olaf Wucknitz 13-Aug-2012 15:05
Addendum
George Heald made some additional tests confirming his numbers. What could cause the discrepancies? Source structure can probably ruled out. Shifting a point source by 4arcsec will only account for about 1deg phase change over 300m baseline change. Because we have two components, the total phase change may be larger, but only if we are close to a null. But the first null is only at ~50km baseline, quite far from the 20km for RS307. And if it is really source structure, my numbers should differ a lot from reference station to reference station, which is not the case.
Could it be the source position? The data set lists the phase centre as 14:11:20.88, 52:13:55.2. NED, on the other hand, says 14:11:20.519, 52:12:09.97, about 100arcsec away! This corresponds to a phase change of about 25deg over 300m, about what we observe! And this difference will then only depend on the position of the superterp station but not on the reference station.
I used NDPPP to shift the data to the NED position and re-ran the fits. Here are the new plots and numbers.
The comparison with George's numbers is here:
CS002HBA0 X const 0.011893 linear 0.100410 George 0 CS002HBA0 Y const -0.047678 linear -0.274723 0 CS002HBA1 X const -25.146677 linear -23.067633 -23.15983494 CS002HBA1 Y const -25.123805 linear -23.236499 -23.070956 CS003HBA0 X const 31.653246 linear 29.609743 29.51807725 CS003HBA0 Y const 25.243859 linear 23.095959 23.11166696 CS003HBA1 X const 29.074315 linear 27.713709 26.37202499 CS003HBA1 Y const 23.162073 linear 21.001680 19.73965005 CS004HBA0 X const 71.767420 linear 65.784373 64.85530977 CS004HBA0 Y const 54.582886 linear 52.277415 51.71035217 CS004HBA1 X const 47.748556 linear 44.782236 43.11659232 CS004HBA1 Y const 33.266612 linear 33.274148 32.38522569 CS005HBA0 X const 37.364905 linear 35.725580 36.55736923 CS005HBA0 Y const 42.119161 linear 40.612561 4177547539 (41.77?) CS005HBA1 X const 29.478235 linear 27.251557 25.51752009 CS005HBA1 Y const 32.208099 linear 30.140095 29.24410287 CS006HBA0 X const -26.053688 linear -23.930044 -23.78864541 CS006HBA0 Y const -4.938545 linear -4.437464 -4.23214965 CS006HBA1 X const 4.844294 linear 5.140505 5.05917215 CS006HBA1 Y const 24.738455 linear 23.177545 24.10707984 CS007HBA0 X const 36.647708 linear 38.227115 39.11122934 CS007HBA0 Y const 35.815216 linear 35.608245 35.96414392 CS007HBA1 X const 51.809793 linear 48.662453 49.0431901 CS007HBA1 Y const 48.173666 linear 44.761026 45.86857076
Now it seems to be consistent within less than 2deg or so. Problem solved! — Olaf Wucknitz 14-Aug-2012 15:18
L71716
This is a 12h observation of TauA in HBA low, see data set description.
