[daip] AIPS image scaling effect

[Emil Lenc]

Hi Eric,

> I am perplexed by your findings.  I cannot find anything in the
> changes in 2005 and 2006 to account for a difference between those 2
> versions of UVFIX or SPLIT.  UVFIX "recent" change was in 2003 to do
> shifting nore accurately (before which your test would have failed
> badly).  The change in  the 2007 (and patched 2006) affected
> uncompressed data and made an error in frequency during position
> shifting equal to the reference channel * channel freq increment.
There was one minor difference between the attempts at processing the  
data that I neglected to mention (sorry, I didn't think it important  
but perhaps it is). In my AIPS 2005 attempt I removed edge channels  
on either side of the band before running UVFIX (By using  
BCHAN=50,ECHAN=449,NCHAN=1,CHINC=1 in SPLIT but without performing  
any averaging on the data - BTW, I had APARM=1,0 but I assumed that  
this wouldn't have an effect because of my NCHAN/CHINC settings).

In my AIPS 2006 / AIPS 2007 approach I ran UVFIX with all channels.  
After the (um,v) shift, I removed the edge channels in the averaging  
phase (By using BCHAN=50, ECHAN=449, NCHAN=400, CHINC=400, APARM=1,0  
in SPLIT).

> If I understand correctly, you use UVFIX to shift the full vis data to
> the position of the target source.  This should bring each channel to
> near 0 phase so the fact that SPLIT corrects u,v,w to the apparent
> average frequency (ignoring flagging) should not matter.
Yes, that is correct.

> Then when you image, the source is not at the expected position  
> exactly but off
> by e.g. 0.5 arc sec although it is not smeared out.
This is definitely the case when I image the phase reference from the  
target field - the phase reference is one of the few sources for  
which I have a good absolute position for. I only have mediocre  
absolute positions for the three other sources that I can see from  
both fields (from NVSS/WENSS). Even so, their relative positions as  
viewed from either field should coincide very closely with each  
other. However, I observe a significant offset in their relative  
positions that is also of order 0.5".

At this point perhaps a diagram is perhaps in order to show what I've  
been imaging and to put it all in perspective (as trying to picture  
this with words is getting a bit hard). The attached image shows the  
two fields that I've been surveying. The lower field is the phase  
reference field and the upper field is the target field - these were  
treated as two separate data sets and the sources within them imaged  
separately (even though the two fields overlap and have sources in  
common). The concentric rings around each field phase centre mark  
annuli (0-0.25, 0.25-0.5, 0.5-1.0, 1.0-1.5 and 2.0-2.5 degrees from  
the phase centre of each field). Increasingly restrictive (u,v)  
ranges were applied in each annuli as the distance from the phase  
centre increased in order to reduce the effects of bandwidth and time- 
averaging smearing. The blue and red crosses mark all of the sources  
that I targeted with IMAGR in order to check for a VLBI detection.  
Sources within rectangles indicate VLBI detections found when imaging  
sources in the target field and those within diamonds indicate VLBI  
detections found when imaging the phase reference field. Four of the  
sources were commonly detected in both fields - these sources are  
marked within both a rectangle and a diamond. It is the positions of  
these sources as measured independently from each field that exhibit  
a ~0.5" offset with respect to each other.

> UVFIX does take differential aberation into account when computing the
> new u,v,w.  But if the phases are made near 0 by the position shift
> then it should not matter so I do not understand your remark about
> aberation.  You tell UVFIX a shift in RA/cos(dec0) and in Dec and it
> does that rather blindly.  Subtleties in the new u,v,w would affect
> sources away from the new phase center not sources at the new phase
> center.
OK, so this is unlikely to be the cause - I just thought it best to  
check and be absolutely sure.

> Note that if you use IMAGR on the averaged data to make an image of a
> source way away from the image center then you should expect both
> smearing and, if there is any systematic channel-dependent flagging, a
> position error.
I limit the u-v range for the wider-field sources to reduce the  
effects of both time averaging and bandwidth smearing (this of course  
comes at the cost of reduced resolution and increased noise).

> I wish I had some better answer to explain what is going on, but I do
> not.
That's OK, we thought it best to ask for your opinion in case you  
knew of an obvious cause or had some more ideas to bounce around. It  
seems that this problem is going to continue to cause a bit more head  
scratching and hair pulling for some time yet :-(

Cheers,

Emil Lenc.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: targets.jpg
Type: application/applefile
Size: 83487 bytes
Desc: not available
URL: <http://listmgr.nrao.edu/pipermail/daip/attachments/20070531/3e2eb8b4/attachment.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: targets.jpg
Type: image/jpeg
Size: 129845 bytes
Desc: not available
URL: <http://listmgr.nrao.edu/pipermail/daip/attachments/20070531/3e2eb8b4/attachment.jpg>
-------------- next part --------------