[daip] SPLAT, SPLIT reference frequency problem

[Eric Greisen]

Michael Bietenholz writes:
 > I think both SPLAT and SPLIT (31DEC04) seem to get the frequency and
 > uvw scaling slightly wrong.  The effects are fairly small, but big
 > enough to worry about when you're doing precision astrometry.
 > 
 > 1)  When averaging IF's together, the output reference frequency is the
 >     the center (average) freqency of IF1 in the original multi-file.
 >     It should be the center (average) frequency over all the IFs.
 >     This causes position shifts in any images made from such data,
 >     so is bad for anyone doing astrometry (even when bandwidth
 >     smearing is acceptably small).

      Averaging IFs and doing astrometry?  I suspect you are getting
what you deserve.  The smearing cannot be small if the position shift
is visible.  Nonetheless you are right that we should fix the task one
way or another.

 > 
 > 2)  The values of u,v,w are in kilo-lambda at the reference frequency,
 >     thus it seems to me that the output values of u,v,w should scale
 >     exactly as the reference frequency.  This does not seem to be
 >     true if the input uv-file has a reference pixel along the FREQ
 >     axis not equal to 1.0.
 > 
 >     I think the output u,v,w values scale as the effective
 >     frequency of IF1,pixel=1.0 (FREQ axis).  SPLAT seems to put the
 >     output reference frequency at pixel=1.0 on the FREQ axis, and scales
 >     the frequency but not the values of u,v,w for any shift from the
 >     original reference pixel to 1.0.   If the input file has a reference
 >     frequency not at pixel=1 (which is often the case for VLBI data), the
 >     output values of u,v,w are wrong, again causing small shifts in position
 >     in any images etc. made from this data.
 > 
 > 
 > Notes:  by reference frequency I mean that in the uv-file header
 > as the "Coord value" of the FREQ axis.  I'm assuming that is the reference
 > frequency used for the calculation of u,v,w.

You might wish to look at the code.  Both SPLAT and SPLIT do exactly
the same thing.  They figure out a new reference frequency which is
the average frequency of the channels of IF1 that are included in the
average.  They scale u,v,w by the ratio of the new average frequency
to the old.

I wonder whether I should retain the IF averaging option.  All AIPS
tasks handle multiple IFs just fine.  IF-based flagging is very common
- so the effective freq or u,v,w may be very different for different
samples depending on that flagging.  Channel-based flagging is far
less common and spectral-line folks anyway understand its dangers.
Also spectral frequency diffs are often << IF frequency diffs.

Eric Greisen