| CSIRO Australia Telescope National Facility ATCA Users Guide |
 |
|
|
This section will describe how to get started on reducing CABB data using the MIRIAD software suite. More information can be found in the MIRIAD manual (http://www.atnf.csiro.au/computing/software/miriad). The manual’s cookbook should be read by all users before starting their first data reduction. This section will explain how reduction differs for CABB data. The MIRIAD manual is being updated to incorporate information on CABB data reduction, but is not yet ready.
The MIRIAD reduction suite is undergoing numerous changes to ensure it continues to be capable of reducing data from the ATCA. Most of the recent changes it has undergone relate to CABB processing. Most importantly, it now deals with the much larger memory requirements of CABB data processing during calibration and visualisation. However it is still being actively upgraded.
It is therefore important that before beginning your data reduction that you obtain the very latest MIRIAD distribution, as new features may have been implemented that will make your reduction easier or more successful, or may solve an issue that you were experiencing with an older version of MIRIAD.
To get the latest MIRIAD distribution, please follow the instructions here: ftp://ftp.atnf.csiro.au/pub/software/miriad/INSTALL.html.
It is also important to report any problems you may have with your reduction to the MIRIAD maintenance team so that any issues can be resolved. If you experience any problems, please use the MIRBUG routine, or send an email describing your problem to miriad@atnf.csiro.au.
Data files from CABB are much larger than those generated by the old correlator, so it is recommended that you process them on a machine with plenty of physical memory (RAM). If you rely on virtual memory to work with CABB data files, then you will experience very poor performance.
To load CABB data, you will use the ATLOD task. The inputs to this task
should, in general, be the same as for pre-CABB data. However there are two
options that should be used for all CABB data to make subsequent reduction
more pleasant. These two options are birdie and rfiflag.
The birdie option was always recommended for pre-CABB data, but
for a different reason. For CABB data, a number of channels are affected
by self-generated interference from the 640 MHz CABB clock, and the
birdie option flags these channels during load. It also flags out
100 channels on each band edge and the unusable parts of the spectrum in
the 20cm and 13cm bands. These unusable parts are due to the fact that these
receivers do not respond to all frequencies across the 2 GHz bandwidth of
CABB. This option still works for pre-CABB data as expected.
The rfiflag option can be used now to automatically flag out
frequency bands that are known to be heavily affected by RFI.
This option is most useful in the 20cm and 13cm bands, where the ATCA
receivers are routinely affected by strong RFI. It is completely safe to use at
any band however, and should be included in the default list of options when you
use ATLOD. The file it uses to determine the flagging it performs can be
customised to include extra flagging (or less flagging) if so desired. Note that the
rfiflag option will have no effect if the birdie option
is not also specified.
The other ATLOD options are usually frequency dependent. Examples of ATLOD use are given below for all ATCA bands.
Task: atlod in = *.C007 out = c007.uv ifsel = 1 restfreq = options = birdie,xycorr,rfiflag,noauto nfiles = nscans =
For these bands, you need only extract 1 IF as both IFs will contain
identical data, and keeping both may lead to confusion later on. At present,
IF 1 has more usable channels than IF 2, so you should select IF 1 unless
you know that IF 1 had a problem during observations. The noauto
option is included here as most users will not require the autocorrelations,
and keeping them will only take up more space and possibly lead to confusion.
Leave off this option if you require the autocorrelation spectra.
Task: atlod in = *.C007 out = c007.uv ifsel = restfreq = options = birdie,xycorr,rfiflag,noauto nfiles = nscans =
If you chose to use the same centre frequency for both IFs then you should
extract only 1 IF using the ifsel parameter as described above.
Task: atlod in = *.C007 out = c007.uv ifsel = restfreq = options = birdie,xycorr,rfiflag,opcorr,noauto nfiles = nscans =
At these mm bands, you should include the opcorr option to
correct for atmospheric opacity.
Task: atlod in = *.C007 out = c007.uv ifsel = restfreq = options = birdie,xycorr,rfiflag,noauto nfiles = nscans =
After loading the data into a MIRIAD dataset, it is usually most convenient to split this into many datasets, each containing an individual source at a single central frequency. This is done using the UVSPLIT task.
At its simplest, it can be run with only the vis parameter set to
the parent dataset:
Task: uvsplit vis = c007.uv select = options = maxwidth =
The select keyword can be used to split out only certain sources,
or to effectively flag out certain antenna. Read about the select
keyword here: http://www.atnf.csiro.au/computing/software/miriad/doc/select.html.
There are two options that are useful here. Specify the
mosaic option if you are splitting a dataset that contains a mosaic
observation, and specify the clobber option if you want to overwrite
any previously split-out datasets in the current directory.
The maxwidth parameter is new to the CABB era, and allows you to
break up the 2 GHz CABB bandwidth into smaller chunks. This will be most useful
at lower frequencies where 2 GHz represents a large fractional bandwidth. At
such frequencies, calibration and imaging is far more challenging than at higher
frequencies. The bandwidth specified here should be specified in GHz. For example,
to mimic the maximum bandwidth available from the old correlator, you could
specify maxwidth=0.128 here.
In most cases, if the birdie and rfiflag options are given
to ATLOD, the data will be mostly free from RFI at this point. However,
you should always examine your data with the UVPLT and UVSPEC tasks
to make sure.
Using UVPLT, examining amplitude and phase versus time is a good way to
find time ranges that have been affected by RFI. Ensure that you specify
options=nofqav to prevent UVPLT from averaging your data over
frequency before plotting. Any RFI present in your observations will probably
be restricted to only a few channels, and these channels can be identified
by using UVSPEC.
To flag this data, you can use the following tasks.
Use the select and line keywords to select the bad data
and set flagval=flag to flag the data as bad. UVFLAG can be
used to flag any set of bad data, but can be tediously repetitive.
This task can replicate many of the plots that UVPLT generates, which may allow you to directly select the outliers representing bad data. However BLFLAG will always average your data over frequency before plotting, meaning that it may be impossible to do most flagging with this task.
This new task replaces the old TVFLAG task that no longer works on displays with more than an 8-bit colour depth. PGFLAG displays data as a channel-time waterfall plot and allows direct selection of ranges of bad data and flexible flagging options.
The calibration strategies used for pre-CABB data reduction are still valid for CABB data. These strategies are described best in the MIRIAD cookbook at http://www.atnf.csiro.au/computing/software/miriad/userguide/node84.html.
To summarise, you should use MFCAL and GPCAL on your primary flux calibrator and phase calibrators (as CABB always supplies XX,YY,XY and YX polarisation products), scale your phase calibrators to the primary flux scale using GPBOOT (or MFBOOT if your flux calibrator is a planet) and then copy your calibration to your target sources using GPCOPY.
After calibration, you should examine your solutions using GPPLT, UVPLT and CLOSURE at the very least. Calibration and flagging should be done in an iterative fashion until you are satisfied that your calibration solutions are of excellent quality.
Imaging of CABB data can be performed in exactly the same way as pre-CABB data, however you may get better results with a slightly different method.
For high-frequency observations, where a bandwidth of 2 GHz does not represent a very large fraction of the sky frequency, then the imaging strategy used for pre-CABB data will most likely be adequate. For lower frequencies the multi-frequency strategies are more likely to produce good results. In particular, the use of the MFCLEAN task should be considered. Read the excellent description of why MFCLEAN is necessary (http://www.atnf.csiro.au/computing/software/miriad/userguide/node110.html) to see if you will need to proceed with this strategy.
If you don’t need MFCLEAN then you should proceed with the usual INVERT, CLEAN and RESTOR method of creating an image, and combine this with SELFCAL if appropriate. In this case, nothing should be different from pre-CABB reduction.
If you do need MFCLEAN then the inputs to INVERT will need to be
slightly different as well. Firstly, both the mfs and sdb
options will need to be passed to INVERT. MFCLEAN also requires
that the beam be three times the size of the region being cleaned, which
cannot be achieved using a simple option, but rather requires that you set the
imsize parameter to be three times the size of the ATCA primary beam
(if you plan to clean the entire primary beam region). You should therefore
set cell and imsize together to produce the appropriate
dirty map. You may use MIRIAD to get a rough idea of these numbers by
running through the reduction without MFCLEAN and use RESTOR’s
estimate of the synthesised beamsize.
Example inputs to INVERT are shown below for a theoretical observation at 5.5 GHz for 12 hours in a 6km configuration.
Task: invert vis = obs.uv map = obs.map beam = obs.beam imsize = 1887,949 cell = 0.83,1.65 offset = fwhm = sup = robust = 0.5 line = ref = select = stokes = i options = mfs,sdb mode = slop =
This will create an image that is 26.1 arcmins on each side, and assumes a synthesised beamsize of 2.48 by 4.95 arcsec.
The inputs to MFCLEAN might then be:
Task: mfclean map = obs.map beam = obs.beam model = out = obs.clean gain = cutoff = niters = region = relcenter,boxes(-314,-158,314,158) minpatch = speed = mode = log =
Here we are selecting the central third of the image (the primary beam
area) with the region keyword. This also satisfies the
MFCLEAN requirement that a guard band be present around the
region being cleaned. The other parameters are very similar to those
of CLEAN and should be set as usual to prevent over-cleaning.
After MFCLEAN has run, RESTOR should be used in the normal way to produce the cleaned image. SELFCAL can also be used in the usual fashion in this process.