[daip] AIPS image scaling effect

[Emil Lenc]

Hi Eric,

I probably didn't explain the procedure that I followed very clearly.  
I actually performed the UVFIX on the full un-averaged data set (i.e.  
with all available frequency channels). It is only after the shift  
that I performed averaging in frequency with SPLIT - this was to  
reduce the 70Gb data set down to a more manageable(?) 0.25Gb data set  
for further calibration and imaging of the source of interest.

By the way, in the first stage of my processing I used IMAGR (DO3D  
enabled) with boxes around each of the WENSS sources to determine  
whether or not there was a VLBI detection at any of these locations  
in the first place (based on a 6 x RMS noise cut-off). Since the  
WENSS/NVSS astrometry was generally only good to ~1" and was more  
sensitive to large scale structure, I used the IMAGR results to  
determine a more accurate source position for each positive detection  
(this was subsequently used to guide UVFIX). When I imaged the u-v  
shifted and frequency averaged data-sets, the sources appeared pretty  
much in the centre of each map (as one would hope) and so this  
suggested that both IMAGR and UVFIX were at least consistent.  
However, both exhibited a significant position error between the four  
common detections that I mentioned in my previous email.

I've been having a number of discussions with my co-authors and we  
were wondering whether or not UVFIX/IMAGR took into consideration  
corrections for aberration (to account for the motion of the Earth)?  
A back of the envelope calculation seemed to suggest that the  
aberration effect could conceivably account for the magnitude of the  
offsets that we were witnessing at the edge of our fields (assuming  
it was not completely corrected for).

Cheers,

Emil.


On 30/05/2007, at 3:00 AM, Eric Greisen wrote:

> Emil Lenc writes:
>> I have been working on a 90 cm VLBA/WSRT/76m Lovell data set in an
>> effort to perform a wide-field VLBI survey around two nearby sources
>> (a phase calibrator J0226+3421 and a gravitational lens 0218+357 1.89
>> degrees away). The survey targets WENSS sources that exist around
>> these fields. After initial calibration of each field, I run UVFIX to
>> shift the data set to each target source, SPLIT the result (averaging
>> all of the frequency channels together) and create a dirty image.
>>
>     Averaging the channels is not a good idea - you have the channels
> in part to provide added accuracy when looking at large position
> shifts.
>
>
>> In my first attempt to image sources in the field with AIPS 2005 (mid
>> last year) I noticed that the positional accuracy of my detected
>> sources appeared to degrade significantly with radial distance from
>> the phase centre of each field. It appeared as if my image was scaled
>> down by a factor of 1.29x10^-3 +/- 7x10^-5 which corresponded to an
>> offset of 53+/-3 frequency channels in our data set (interestingly
>> this appeared to correspond quite closely to the 50 lower-band
>> channels that were flagged during editing - though this may just be a
>> coincidence).
>
>     Not suprising - the average u,v,w for the sample is computed
> without regard to which channels actually contributed to the sample -
> or at least I would expect that is mant cases.  That you can see a
> shift due to N channels means that you should never hjave averaged
> them in the first place.  AIPS will grid each in the correct place in
> the UV plane and then not make position errors.
>
>>
>> At the end of last year I reprocessed the data in AIPS 2006 and this
>> scaling effect appeared to largely disappear - perhaps as a result of
>> one of the bug fixes in that release. As I expanded my survey field
>> of view to just over 2 degrees from the phase centre of each field, I
>> discovered 4 common sources across both fields. However, when I
>> imaged these sources and compared the positions from each field there
>> once again appeared to be a kind of scaling effect occurring (i.e.
>> the corresponding sources from each field did not exactly coincide).
>> This time, however, the effect was much smaller ie. 7.3x10^-5 +/-
>> 1.4x10^-5 or equivalent to an offset of approximately 3 +/- 0.6
>> frequency channels.
>>
>> This morning I installed AIPS 2007 as I noticed that there was a fix
>> in UVFIX, but the offset still continues to be present and by the
>> same factor.
>
>      The correction in UVFIX was in the code that does position
> shifting within UVFIX and applied only to one of the two cases
> (compressed and not compressed).
>
>>
>> I've attached an image of one of these sources, it is in fact my
>> phase calibrator (J0226+3421) as seen from the centre of the J0226
>> +3421 field (green) and as seen 1.89 degrees away from the phase
>> centre of the 0218+357 field (the red contours are the results from
>> AIPS 2007 and the blue contours are the results from AIPS 2006 - the
>> blue contours are partially hidden because they almost exactly match
>> the AIPS 2007 results). All images have been restored with the same
>> beam. The three other common sources exhibit offsets of a similar
>> nature when compared against each other (with offsets of ~440-690  
>> mas).
>>
>> Would you know what might be causing this effect?
>>
>> In case this information is useful to you: The observation was
>> recorded at 320.49 MHz (lower frequency), has a 4 MHz IF, 512
>> channels, 0.25 s integration time and single polarisation (LL). It
>> was correlated at the EVN correlator at JIVE, Dwingeloo, the
>> Netherlands.
>
> Try imaging without averaging up all the channels.  You will get morwe
> accurate positions and they should agree.  Also - be sure that you are
> using multiple facets as needed at such long wavelengths (task SETFC
> can tell you what you should do).
>
> Eric Greisen