[daip] FLATN artifacts in 31DEC10

[Eric Greisen]

Kristine Spekkens wrote:
> Hi Eric, 
> 
>  My student Karen and I have been puzzling over some artifacts in the 
> mosaic of GMRT pointings that she's making using FLATN, and so we 
> decided to check for consistency between FLATN outputs in 31DEC08 (I 
> used FLATN successfully in this version in the past) and 31DEC10 (which 
> Karen has installed on her machine). We find that the artifacts only 
> show up in the 31DEC10 version.
> 
>  More specifically, our test was to use FLATN to mosaic two GMRT 
> pointings, whose headers are at the end of this email. The FLATN inputs 
> are also at the end of this email: we're basically running in default 
> mode, specifying that the primary beam be corrected using 
> GMRT-appropriate values. 
> 
> The attached image FLATN_31DEC08.png shows the result in 31DEC08, which 
> is as I expect: two overlapping square images (the input images do not 
> go out to 0.1 of the GMRT primary beam, and we've set PBPARM(1)=0.1), 
> with "seam" between the two.
> 
> The attached image FLATN_31DEC10.png shows the result in 31DEC10: it 
> differs from the one above in the appearance of the "seam": for reasons 
> that I don't understand, there is a circular "seam" in addition to 
> remnants of a square one like in 31DEC08. The difference between the 
> datasets obtained using the two versions of FLATN is in FLATN_DIFF.png.
> 
>  Should I expect different outputs from FLATN in these two versions of 
> AIPS? What is the circular seam in FLATN_31DEC10.jpg? We have tried a 
> few other tests with this version of FLATN (such as changing PBPARM(1) 
> to 0.5), and the circular seam just won't go away; the 31DEC08 version 
> of FLATN, on the other hand, seems to be behaving as expected (ie. 
> blanking according to PBPARM(1)).
> 
>  Any insight you have into why we're getting different outputs in 
> different AIPS versions would be greatly appreciated; apologies in 
> advance if we've made a silly mistake somewhere. If you need more 
> details about the datasets, etc, let me know.
> 
>   I hope that you are well!

Doing just fine thanks.

FLATN was changed to have a circular EDGSKP which you can override by 
setting it to a negative number - it then reverts to square.  The idea
was to limit corners of FFT images which are more prone to numeric 
error.  You probably do not need that limitation.

Cheers,

Eric Greisen