[daip] aips imagr

[Eric Greisen]

Cao Hongmin wrote:
> Hi Eric
> 
> in aips, the tasks, such as, "fring", "calib", "bpass", we can provide 
> an model for the calibrator source (if so, just one calibrator should be 
> used) or choose the default  model. As far as the multiple-source files 
> concerned:
> by default,
> 1."fring" will choose a point source model at the assumed phase centre 
> (?), but how about the flux density of the model ?

FRING does not care about flux - it uses phases only.

> 2."calib" also will use a point source with a flux density given in the 
> SU table, but how about the position of the model ?
> it seems the model is also put at the phase centre. whether we can not 
> run "calib" when there is no flux information in the SU table by the 
> default setting, that is, no model provided ? (from some data processing 
> guides, "calib" is run after the task "setjy").

CALIB may be run before the flux of a calibrator is determined (most cal 
sources are time variable and the flux is not known in advance).  It 
will find gains for a 1.0 Jy source at the phase reference position for 
the calibrator

> 3. how about the "bpass" ? it seems just the flux of the calibrator from 
> the SU table is used.

BPASS also computes phase and, for calibrators, it is assumed that position
is that found in the SU table.  BPASS also uses the flux in the SU table,
although often one computes a normalized BP table, leaving the real 
amplitude calibration to CALIB.

Note that all 3 of these tasks and many more use the same software to 
compute models, whether those models are images (with or without Clean 
components) or simply calibration point sources.  If your calibration 
source is not at the phase stopping position you used for it, then you 
must correct the observations of that calibrator first before using it 
to calibrate other sources.  CLCOR can do that.

> 
> if we let the clean run freely with no any self-calibration, whether it 
> is possible to reach the state - image rms ~ 0 ?
> 

Barring numerical error issues, it is in principle possible to reduce 
the residual image to approximately 0.    When the "Clean components 
found are
restored however it will be clear that many of them are noise and the 
rms will not be zero.

Your follow-up e-mail suggests that you hit numerical error rather badly 
and I wonder if you are using an old version of AIPS (what version are 
you using?) or did not model the test source correctly.
> 
> The data average when run "split", will make the data to a small size 
> that the data processing softwares can readily handle.
> Does the average will help to enhance the base-line sensitivity (it 
> seems so from the formula of the base-line sensitivity) ?
> otherwise, it will also make the FOV much smaller than before.

Yes - you are destroying the information content in your original data set.
If the signal in each spectral channel of an IF is the same (plus noise)
then you are improving the sensitivity of the individual visibility.
However, if the signal is not the same due to the fact that each 
spectral channel is located at a different place radially in the UV 
plane, then you decrease the noise by sqrt(N) but you may reduce the 
signal by even more.
> 
> We carried out an multiple-phase centre experiment with EVN, with 8 IF 
> (16MHz of each), 2 pol, 64 channel of each IF.
> The data is averaged into 8 IF (16MHz of each), 2 pol, 1 channel of each 
> IF, and 10sec integration time after "split". So the final FOV will be ~ 
> 300mas (considering the longest baseline ~10000 km) by bandwidth 
> smearing, but all the target sources are quite weak ~ 1mJy, and the 
> position accuracy of the First sources at 1Jy ~ 1" ( the NVSS sources 
> are even worse). I feel that because we use natural weight for the weak 
> source detections, so the long baselines are weighted down caused by the 
> larger errors, so the FOV should be estimated using the relative short 
> baseline (that is the European antennas, ~ 2000km), which can give 
> enough FOV for source detections. i am not sure if it is a proper 
> argument (?).
> 
Certainly if you down-weight the longer baselines the FOV that you can 
image
is larger.  But EVN has an odd distribution of telescopes so that simple
use of NA weight may not down-weight the long spacings all that much.

> (i tried to make a map with 8192 pixels with 1mas/pixel, it is hard for 
> the software such as aips, since the data have been averaged as noted 
> above. To make several sub-maps might means a lot of work to do because 
> of the large number of the targets).

You FOV is 300 mas - an image of 512x512 covers that.  You will have no 
sensitivity beyond that.  IMAGR can do 8192x8192 or even larger if you 
have adequate computing capacity.  But multiple facets are really easy - 
a text file (see BOXFILE help) can list the RA and Dec of each desired 
image.
But you cannot feed IMAGR such heavily averaged data if you want to image
objects well away from the phase center.

If your FOV is 300 milli asec, how do you expect to find a source with a 
1 asec positional uncertainty?

Eric Greisen