[daip] AIPS problem with fitld douvcomp +-1

[Eric Greisen]

Alexey Melnikov wrote:
> Dear Eric,
> 
> We find something looks like bug, but very strange thing in AIPS. Tried with
> different versions - 2003, 2006, 2012, everywhere is the same result.
> 
> I've correlated our VLBI experiment RU0083 with DiFX, there was 4 stations: Svetloe
> (Sv), Zelenchukskaya (Zc), Badary (Bd), and Crimea (Sm). After we processed
> fits with AIPS.
> 
> At the pictures attached you'll find autocorrelation for Sm and Zc.
> Loading file with fitld with douvcomp -1 is ok, but if douvcomp is +1 then we
> saw strange effect for Sm.
> 
> I prepare short fits file is you need it - find link in the attached text
> file, there is also list of AIPS's commands to get this effect.
> 
> If you need any additional information, please, ask.

I am not surprised that you encounter some issues with compressed data. 
  The compression looks through all the data in a record and finds a 
scaling parameter that causes the highest point to become about 32000 
and scales all the others with that parameter to 16-bit integers.  If 
there is a crazy point (at the first or last spectral channel usually), 
then all the rest of the data can even be turned into pure 0.  More 
commonly, they will show some segregation into discreet levels.  This is 
less obvious if you average the data later.

In any case, with such a small interferometer, you really should use 
DOUVCOMP = -1 - disk is cheap now and it costs noticeable cpu time to 
compress and uncompress the data.

Eric Greisen