Computation

              

This section is only relevant for self calibration of images that have not been made with multi-frequency synthesis techniques. That is, you have just one channel (usually channel 0) that you are imaging and self calibrating, not a band of channels. For self calibration of multi-frequency synthesis images, you should use MIRIAD.

The task you will need for self-calibration is CALIB. Once again, I have provided some procedures to help you do self-calibration with a reduced set of CALIB's inputs. There is one procedure for multi-source files called ATMCAL, and one for single-source files called ATSCAL. If these are insufficient for your purposes, you know what to do.

When you self-calibrate with a multi-source file, all you do is produce a new CL table for application to the data; first you run CALIB to generate the solutions, and then you interpolate them to a CL table with CLCAL in the usual way. This means that the fundamental time scale on which you can alter the gains is the CL table entry interval. ATLOD builds the CL table with 3-minute intervals. You may have something different if you have rebuilt it with INDXR (it defaults to 5-minute intervals).

On the other hand, with single-source files, each application of CALIB\ produces a new and corrected data base. This has the advantage that, because there are no CL tables with single-source files, the fundamental time scale on which you can correct the gains is the integration time of the data (usually 10 seconds). The disadvantage is that you must duplicate the data with each pass. If you are working on a large data set, this may cause difficulty according to the available disk space.

1. Let us continue now with a discussion of some of the adverbs for multi-source self-calibration with ATMCAL. Many of the adverbs have already been discussed in § 8, so I will not cover them all.
   • Select the source to self-calibrate with calsour and calcode.
   • Select the desired frequency with freqid.
   • Calibrate the data by setting docalib=1 and by selecting the correct CL table with gainuse. This will probably be version 2 if this is the first self-calibration pass following basic calibration. If this is a later pass, then gainuse will be a higher version number.
   • Select the model, a list of CLEAN components, by filling in the CLEANed image name with in2name, in2class, in2seq and in2disk (use 2name). The version number of the CLEAN component file is selected with invers (generally there is only one CC table, so leave this at 0).
   • ncomp selects the number of CLEAN components in the CC table to use as the model, starting at one; this is a very important parameter. Your goal is to provide CALIB with the best possible model of the source, so you should not include CLEAN components that are obviously wrong. One obvious example of this is a negative component, as the source must be positive. A common technique is to use all the components up to the first negative. However, if you have just one bad component in a much more extensive list of positive components, then it may prove advantageous to include more components than this simple approach suggests.

     This makes it clear why good windowing is so important during the CLEAN step; it helps exclude the possibility of including wrong components in the CLEAN model.

   • uvrange is also a fairly important parameter. Your model must match the data. Thus, as is often the case, if your model only represents a source with high spatial frequencies and your data contain a more extensive range of spatial frequencies, you must exclude the data that are not represented in the model. Two numbers are required, a minimum and maximum uv-distance, in . See also the task KALIB discussed at the end of this section.

     One good way to set this range is to make a plot of visibility amplitude versus uv-distance with UVPLT. Find the total CLEANed flux density in your model by listing the CLEAN component table, and look up the corresponding uv-distance on the plot. Set the uvrange to exclude the baselines interior to this value.

     With VLA data, this is a fairly safe recipe, because there will be many baselines in the selected uv range. However, with ATCA data, you must be careful not to exclude too many baselines, such that a good solution cannot be found. You must compromise between matching the data to the model and and having as high a level of over-determination in the problem as possible.

   • It may be that your source is a point source, or that by suitably restricting the uvrange, can be viewed as a point source (make sure you still include enough baselines to get a decent solution). In this case, rather than using a list of CLEAN components, you can specify a point-source model with the smodel array. smodel(1) is the flux density of the point source, smodel(2) is the x offset of the point source from the phase centre in arcseconds, and smodel(3) is the y offset. Note that if you leave the offsets at zero, the self-calibration procedure has the freedom to phase-shift the data so that the source appears at the phase centre (this can be a useful way to move a point source to the centre of a grid cell). Absolute position information is lost with self-calibration.

     smodel can also be used for elliptical Gaussian models. See HELP smodel for more information.

   • solint controls the solution integration time (in minutes) over which the gain changes are assumed to be constant. If you have sufficient signal-to-noise ratios, integration times of the order of one minute are a good starting place. Depending on the phase stability of the data, this number may need to be shorter or longer. You will probably need to experiment, but remember, for weak sources you should make sure you set this long enough to ensure a signal-to-noise ratio of at least a few on all baselines. However, this may preclude any useful modification of the gains, if it is too long.
   • The conventional wisdom at the VLA has always been that you should first do a couple or a few phase-only self-calibration passes ( solmode='p') and then, once the phases are in order, attempt an amplitude and phase self-calibration (solmode='a&p'). In my experience with ATCA data, it is sometimes better to try and do phase and amplitude straight away. I believe the reason is, that once errors in the model get frozen in, they are particularly hard to remove with ATCA data owing to the small number of baselines. Thus, you are better off trying to remove both the amplitude and phase errors straight away. As always though, experiment !!
   • If the source you are trying to self-calibrate is polarized, but you have loaded the data as polarizations rather than converting to Stokes parameters, then recall that XX = I + Q and YY = I - Q, so that you must average XX (called RR) and YY (called LL) before trying to work out a self-calibration solution. Do this by setting doavg=1. If the source is unpolarized, it may still be advantageous to do this, as it will boost your signal-to-noise ratio slightly. If you have converted to Stokes parameters, then set doavg=-1.

   inname,inclass,inseq,indisk Select data base

   sources='0412-23' Select source

   qual=-1 Select all qualifiers or specify

   freqid=2 One run per freqid

   docalib=1 Calibrate data with

   gainuse=2 the correct CL table

   flagver=0 Select flagging table

   in2name,in2class,in2seq,in2disk Select CLEAN image for model

   invers=0 Select CLEAN component file version

   ncomp=250 Select number of CLEAN components to use for the model

   smodel=1.3,-34.2,54.3 Use point source model rather than CLEAN components

   uvrange=0 Specify uv range to match model

   wtuv=0 Zero weight for points outside of uv range

   refant=3 Specify the reference antenna

   solint=1 Solution time in minutes.

   soltyp=' ' Least squares algorithm. If too many failed solutions, try 'L1'

   solmode='a&p' Solve for amplitude and phase

   doavg=1 Average XX and YY together

2. After the new solutions have been generated, examine them, as always, with SNPLT. Once you are happy with the results, apply them to the current CL table and write a new CL table; remember, CL tables are cumulative.

   inname,inclass,inseq,indisk Select input data base

   sources='0412-23',' ' Select source calsour='0412-23'

   calcode=' ' Leave blank

   freqid=1 One freqid per run

   timerang=0 Select time range

   interpol='2pt' Select two point interpolation

   gainver=2 Apply solutions to current CL table

   gainuse=3 and update to next CL table

   refant=3 Select ref. antenna

3. If you are using single-source files, then use the procedure ATSCAL which has inputs very similar to ATMCAL. I refer you to the discussion above for details. The main differences are the inclusion of the output corrected data base adverbs outname, outclass, outseq, and outdisk, and the absence of flagver, gainuse and the need to run CLCAL. Note that an SN table is still generated so that you can inspect the solutions. It is attached to the input data base.

   inname,inclass,inseq,indisk Select data base

   sources='0412-23' Select source

   qual=-1 Select all qualifiers or specify

   freqid=2 One run per freqid

   docalib=1 Calibrate data with

   gainuse=2 the correct CL table

   flagver=0 Select flagging table

   in2name,in2class,in2seq,in2disk Select CLEAN image for model

   invers=0 Select CLEAN component file version

   ncomp=250 Select number of CLEAN components to use for the model

   smodel=1.3,-34.2,54.3 Use point source model rather than CLEAN components

   outname,outclass,outseq,outdisk Specify output corrected data base

   uvrange=0 Specify uv range to match model

   wtuv=0 Zero weight for points outside of uv range

   refant=3 Specify the reference antenna

   solint=1 Solution time in minutes.

   soltyp=' ' Least squares algorithm. If too many failed solutions, try 'L1'

   solmode='a&p' Solve for amplitude and phase

   doavg=1 Average XX and YY together

4. You might also consider the task KALIB. This is a clone of CALIB, but it allows you to specify a uvrange (via the different adverb uvlimit for each subarray in the file (e.g. if you have concatenated configurations together with DBCON you will have several subarrays). With ATCA data, it is often crucial to finely tune the uvrange (e.g. to exclude just one shortest spacing in each configuration) and the uvrange adverb applies to all configurations in the file so that you end up excluding or including spacings undesirably. See the EXPLAIN file for more details.
5. It may occur that you wish to apply an SN table to a single-source file and write out the corrected visibilities (we apply CL tables to multi-source files and SN tables to single-source files). For example, you may be working on spectral-line data, and there is a strong maser at one channel. You can use this maser, which is easily self-calibrated, to generate solutions that you would then like to apply to say, a continuum data base. You simply copy the SN table to the desired single-source file, and then apply it with SPLIT (although in this case there is no `splitting', just application of the SN table) by setting docalib=1.

Last update : 27/11/93

1

APPENDICES