[daip] IMAGR dirty beam problem

[Eric Greisen]

Geoffrey Bower wrote:
> Hi Eric,
> 
> I'm encountering a weird imaging problem.  IMAGR sometimes produces a dirty beam which has its maximum in its corners.  See the attached image of the dirty beam (log scale, compressed pixel range).  The same data set doesn't do this all the time.  I also get images where the maximum is in an edge pixel.
> 
> I'm not doing anything odd with imaging parameters that I am aware of.  The results might be sensitive to the size of the image I am making.
> 
> This is new VLBA data and I am using 31Dec2010 AIPS (old, I know).  I'm running the linux version.
> 
> Do you have any ideas what might be behind this?
> 
> Cheers,
> 
> Geoff
> 

Absolutely.  You probably have lots of data in the form of many channels
and/or samples.  After all these samples are gridded and the FFT, the 
image has to be corrected for the FFT of the gridding convolution 
function.  That has very large values in the corners (10^6) so 
cumulative computational error is magnified there.  That is what is 
happening.

In more recent versions of AIPS the pseudo-AP is all double precision 
and this does not happen.  Note that it takes only a few minutes to 
install a new version of AIPS and there are many improvements (see 
aips.nrao.edu for AIPSLetters describing these).

Eric Greisen