Self-calibration of CABB data

Got a calibration problem? Discuss it here.

Moderator: Mark.Wieringa

Post Reply
nseymour
Posts: 6
Joined: Wed Feb 29, 2012 3:29 pm

Self-calibration of CABB data

Post by nseymour »

Hello,

I'm imaging a bright radio galaxy (~0.8Jy at 21cm) with the data from LX and CS receivers. The aim is to achieve high dynamic range to observe relative faint sources and structure within a few arcminutes. I wish to use self-cal to improve the calibration of the radio galaxy. Obviously I'm using option=mfs and the input model is a clean component image clipped at high flux so that a only 3 or 4 pixels represent the entire structure (this seems a reasonable model). The source also has a steep spectrum so my questions are:

(1) does selcal with option=mfs just take the CC from the 1st plane above my clip level and the corresponding 'flux*alpha' points from the second plane (i.e. it presumable does not read in pixels in the 2nd plane above the clip level?)

(2) The User Guide says that the default 'line' parameter (for dealing with the spectral index) can be rather inadequate, but it does not say what the default is, how to find it out or what a sensible value should be. I guess a sensible value would be something like:

line=channel,x,1,1

where x = the channel number at/close to the reference frequency?

(3) anyone got any general tips on high dynamic range imaging/selfcalibration. I'm thinking of restricting the uv-range (select=-uvrange(0,10) so that I don't calibrate visibilities dominated by any potential extended structure.

thanks,

Nick
Mark.Wieringa
ATCA Expert
Posts: 297
Joined: Mon Feb 08, 2010 1:37 pm

Re: Self-calibration of CABB data

Post by Mark.Wieringa »

Hi Nick,

(1) Looking at the code it appears that selfcal clips the I plane with I<level and the I*alpha plane with -level<I*alpha<level. It then FFT's both planes and reads off the values for each visibility. So yes, if there are large values in the second plane where the first plane is clipped they still get used (|alpha|would need to be > 1).

(2) The default line parameter for selfcal with mfs is all channels. If you used all (unflagged) channels to create your mfs model image, then this is probably fine. I'm not sure what the 'rather inadequate' warning is based on.

(3) Others may chime in here, but your suggestion to exclude the shortest baselines, at least for the initial iterations, sounds good. Just make sure you still have enough baselines left for a well determined solution - selfcal with less than 10 baselines doesn't converge as well.

Cheers,

Mark
nseymour
Posts: 6
Joined: Wed Feb 29, 2012 3:29 pm

Re: Self-calibration of CABB data

Post by nseymour »

Hi Mark,

Thanks for your reply.

(1) OK, I see. So this default behaviour is not what you would want for a good model. You would want to select the same pixels in both planes by clipping in the I plane and then masking in the I*alpha plane using that clipping. How does miriad deal with an image with two planes? i.e. how would you select each plane to manipulate them separately with 'mask' or 'math'?

(2) rereading the manual it seems the comment 'the default parameter can be rather inadequate' refers to when your model is only appropriate to some channels in the visibility dataset (so the default of all channels seems fine if you can select the correct clean components from both planes).

(3) good point about not ending up with too few baselines! I'll be careful if I do use uvrange.

cheers,

Nick
Mark.Wieringa
ATCA Expert
Posts: 297
Joined: Mon Feb 08, 2010 1:37 pm

Re: Self-calibration of CABB data

Post by Mark.Wieringa »

Hi Nick,

Yes, I tend to agree that for models this may not be what you would prefer. I'm not sure if this was a deliberate or just a convenient way to handle this. I suspect that in practice it matters little if you are just trying to get a new overall gain solution (excess flux at one end of the band will be compensated by a deficit on the other end). It should only affect pixels close to the cutoff level and these would have little weight in the solution.

Apart from me trying to change this in the code (which would take me longer than typing this message by the looks of it), you could attempt to do the masking yourself using imsub to separate out the two planes, maths to apply the mask to plane 1, cp to copy it to plane2, and imcat to glue the planes back together.

If you attempt this and do happen to notice a significant improvement let us know here.

Cheers,

Mark
nseymour
Posts: 6
Joined: Wed Feb 29, 2012 3:29 pm

Re: Self-calibration of CABB data

Post by nseymour »

Hi Mark,

Thanks for that. I will implement the selfcalibration manipulating the two planes separately as you suggest. I suspect that this is important for bright, moderately complex sources with steep spectra, alpha<-1, and relatively large delta(nu)/nu (as in my case). I'll let you know how it goes.

cheers,

Nick
dominics
Posts: 17
Joined: Fri Jul 23, 2010 2:16 pm

Re: Self-calibration of CABB data

Post by dominics »

Hi Nick and Mark,

I'm also interested in self-calibrating my 16 cm data, and based on what I know about the various miriad tasks that could be used for this goal I come to the conclusion that at the moment one can only uses divide-and-conquer in self calibration. I'll explain how I come to this conclusion below, and I'd be interested to hear how you think about this.

I think there are three important issues that need to be address: one is the large fractional bandwidth in the 16 cm band and the variation in complex gains across this band, two is the inclusion of a clean component model when self calibrating, and three is the variation in primary beam attenuation as a function of frequency for off-axis sources. (These are the important ones for my reasoning below) As far as I can tell there are three miriad tasks that one can use for self calibration: gpcal itself, selfcal, and gpscal. I've compiled a number of pros and cons of the different programs:

gpcal assumes that the field is dominated by a single bright source at the field centre (I assume that this is the reason for the following warning in the ATCA calibrator data base "Warning for 16cm observers: A typical 20cm field will contain a 100 mJy confusing source. In very compact arrays, 300 mJy of confusing flux is common on the shortest spacings. At 16cm, calibrators stronger than 5 Jy are recommended for ATCA arrays shorter than 375m." - is this correct?)
+ the parameter nfbin lets one solve for a variation in the complex gain as a function of frequency
+ options=qusolve, nopol
- assumes a single source at the phase centre, no clean component model for complex fields

selfcal uses a clean component model as input, allows options=mfs
- no frequency dependence

gpscal uses a clean component model as input, allows options=mfs
+ solves also for polarization leakage, although I would calculate leakages from 1934-638
(these are accurate enough to get P/I <~ 0.01% at the beam centre, and stable over at least 8 hours)
- no nfbin, so the program does not solve for how complex gains vary as a function of frequency ; no options=qusolve, nopol

Ideally, a single program should correct for the variation in primary beam attenuation across the frequency band, include a clean component model, and solve for complex gains as a function of frequency. If my list is of miriad tasks complete, and you agree with my summary of strengths and weaknesses of the miriad tasks, then it would seem that there is no miriad task that simultaneously performs these three tasks: gpcal does not allow for a clean component model, and the tasks that do allow for a clean component model (selfcal, gpscal) do not solve for frequency-dependent gains. Since it's essential to run a phase selfcal on my data (S-PASS point sources), only divide-and-conquer would work as a selfcal scheme.

For completeness: I have been using gpcal: Revision 1.13, 2012/02/20 // GpsCal: version 1.0 12-feb-02 // selfcal: Revision 1.9, 2012/01/17

cheers,
Dominic
vincemcintyre
Posts: 2
Joined: Wed Aug 21, 2013 4:06 pm

Re: Self-calibration of CABB data

Post by vincemcintyre »

To follow up on this:

selfcal now supports the nfbin option. It looks like (for CABB 16cm) one should give:

Code: Select all

nfbin=4 options=phase,mmfs
.

The help text is a little confusing -
mmfs Same as mfs, but with multiple input models,
each applying to a subset of the bandwidth.
For use with nfbin>1: models will be allocated to
bins based on their centre frequency.
This raises the following questions in my mind:
* does

Code: Select all

options=mmfs
imply you must give multiple distinct files eg

Code: Select all

model=c.1200,c.1600,c.2000,c.2400
,
or is it enough to use a clean component model from mfclean?
* is this option only to be used with

Code: Select all

nfbin>1
or are there circumstances where you could use it when

Code: Select all

nfbin=1
?
Mark.Wieringa
ATCA Expert
Posts: 297
Joined: Mon Feb 08, 2010 1:37 pm

Re: Self-calibration of CABB data

Post by Mark.Wieringa »

Hi Vince,

yes, the mmfs option was added in response to Dominic's post. He specifically wanted to use multiple models across the bandwidth with nfbin>1 so the gains are interpolated across the band. (Divide and conquer will just give one gain per sub band).

To answer your questions:
1. yes - you need multiple model files with mmfs and you need nfbin>1, ideally one model per bin. Each model could itself be an mfclean model.

2. if you have a single mfclean model you would use options=mfs, with nfbin>=1

The option mmfs implies mfs, and mmfs with nfbin=1 is the same as mfs

Feel free to suggest a better help text (or update it)

Cheers,

Mark
Post Reply