QUESTIONS ON SELF-CALIBRATION
- There are 13 questions below.
- Quick links to casaguides:
Q1: Self calibration: it is stated here and in other guides how high S/N is needed to do self-calibration. How high?
- A: The answer depends on the solution time you want to use. Also, note that the snr constraint is not for a single baseline, it is an antenna-based value. For example, if you set minblperant=4 antennas as the criterion for a successful solution, then the number of baselines in a successful solution is only guaranteed to be 6. Also, note that the sensitivity calculator alone will not do the right thing for you. Say you have 25 antennas, and you want to explore whether you can use solint=10sec, in sensitivity calculator you can put 10 sec and 25 antennas and 2 pol and channel width, then you must multiply the number you get by the sqrt(antennas) in your data. If this rms value and signal of the source corresponds to a S/N not greater than 3, then you need a longer solint. What signal to use? If it's a single point source, it is the peak flux density. If there are 2 point sources, it is the sum. If there is one point source plus some fainter extended, then use the point source. If the source is complex, then it's a more complicated answer because not all baselines will see all the flux. You can't just use the integrated flux in the image in the calculation. * Ed: I would run self-cal and examine the solution, such as using the X/Y polarization ratio. Also, the smoothness of the solution vs. time is a good indicator. * Steve: If you lose too low of an snr cutoff, the solver will find a solution that takes noise and puts it into the model.
Q2: When plotting the calibration solutions (seen in Fig 9b):
What exactly are we looking for?
- A: You are looking for smoothly-varying phase solutions that you can connect as a function of time on each antenna. If it is semi-random then it is just tracking noise. If you are not sure, just go ahead and apply it and make a new image. If the peak does not go up significantly and the rms go down, then it is not a good solution. Note that it is easy to generate 10-20 sigma point sources at the phase center even if there is no real signal in the data. In cases with few antennas and thus low S/N for antenna-based solutions, if you self-cal the continuum it may look like the image improves, but when you transfer the solution to the spectral line data you may find no improvement. This is in part because self-cal is good at giving you what you asked for.
- Ed: Only put a box on the emission you know is real. In the next iteration, you can box lower level emission that still appears real. This is necessary in VLBI. For large-N arrays like ALMA, only a few iterations of self-cal are necessary.
- What if one antenna solutions look different from the rest (especially in amplitude self-cal solutions)? The NOT antenna selection works in clean, so you can make an image without that antenna and see if it will help to flag it entirely.
Q3: self-calibrate the uv-data. The continuum is far too weak? How do we determine that? A rule of thumb
I used to use was a few sigma in a minute or so but Holdaway thought that was too strict (as I recall). Of course,
the continuum is only in one mosaic field. If it were stronger, would it still be a bad idea to use as it falls
in fewer of the mosaic fields than the line emission?
Q4: In working through the guides, I can clearly see doing phase self-calibration in each case did in fact improve the quality of the data. However, I am still not clear on even performing the self-cal. Is this something to always consider even if you are not sure it will help? Why consider phase-only for the data? What evidence in the data might help the user decide that they should consider phase-only or even try try amplitude calibration using this method? Or when would it be obvious that self-calibration is not recommended if there are examples of such a case (other than not having sufficient S/N)?
Q5: I am relatively new to selfcal, spectral line selfcal in particular. While there are some explanation already on the casaguide, more detailed explanation would help us a lot, especially for those who are not very familiar; e.g. tips on which field to chose? what kind of s/n is needed to do reliable selfcal? Is there a criteria for the number of antennas? is it safe to apply selfcal for a source with complex structure like the overlap region of the antennae?
- A: Be careful to specify the channel number in gaincal for the UVDATA, not for the channel in your CLEAN map, because you may not have imaged the whole cube, only the strongest few channels to produce your model.
Q6: Does selfcal force the science target to be at 0,0?
- A: Absolutely not. If you are using the model, it forces it to be at the centroid where the model is. Only if there is nothing in the model column (i.e. 1's in the amplitude and zeros in the phase) will the target be placed at the phase center.
: Is it dangerous for astrometry?
- Ed: Your astrometry is how it is tied to the calibrator.
- Crystal: If you have a sufficiently detected source, and you have placed the clean box corectly, the worse case error is that you can move the source by 1 pixel, which is another reason to oversample the synthesized beam by a significant factor (>=4).
- Steve: You can get an inflated impression of your absolute astrometry after selfcal, because it will inflate the source flux density. You should do the source position accuracy calculation (beam/SNR) before self-cal.
Q7: What if the source is (slightly) extended - can you legitimately use selfcal?
Q8: What "solint" is recommended to use in self-cal? Why do we use here solint='int' for self-cal? It was not the same in the other guides..
Q9: Why is calwt=F for self-calibration?
- A: Because setting it true, the behavior is not well understood. For example for extended sources, calwt=T will upweight short spacings and downweight long spacings and lead to a strange solution.
Q10: NGC3256: Why do you ask for 30m solutions but your scans are only 10m. Since you don't set combine='scan' you get one solution per scan (probably is what you want) which has S/N about sqrt(3) times worse that what you calculated.
- A: If you are trying to specify a long solint (i.e. longer than a single scan), then you must set combine='scan'.
Q11: If you have data from two different configurations (like some ES proposals requested), and the source is strong enough to selfcal, would you self-cal the two datasets individually, or wait until they are concatenated, or both? (Presumably in the larger configuration, the large-scale emission will be resolved out, so the mask would need to be different in the individual datasets.)
- A: We are still exploring the best path for this still. Even if you do them individually, it is probably best to also do a combined solution after combining them.
Q12: If the PI has specified a goal sensitivity which is unattainable due to dynamic range considerations, when is it time to stop trying self-cal iterations (or, put another way, what's the maximum dynamic range achievable in practice in ES likely to be?).
- A: Unfortunately, you don't really know what the dynamic range limit is, or what is actually limiting it. If you can identify where it is coming from, there may be ways to increase it somewhat (such as using nterms=2 instead of 1), unless it is simply a uv-coverage limit. There is nothing you can do about uv-coverage limit. In TW Hya SV data, we reached ~500 which is not near the sensitivity limit, even though it is essentially a point source. Similar result on Band 3 quasar observations. It's not clear where this limit is coming from yet.
Q13: The self-cal section of the Antennae CASAguide states: "We will also use gaintype='T' to tell CASA to average the polarizations before deriving solutions". The help guide specifies for gaintype: " 'T' obtains one solution for both polarizations; Hence. their phase offset must be first removed using a prior G."
- A: Using T (which means Troposphere) will combine the polarizations before computing a solution. It is safe to do in selfcal because the data have already been calibrated. However, for the initial calibration gaincal step, you should not use T. You must use the G solution. At some level, the dynamic range will be limited by polarization leakage, in which case it might be a good idea to run selfcal with G with a long solint.
- Steve: If you have a source with significant linear polarization, then you might introduce offsets in a G amplitude solution (probably not in a phase-only solution).