[daip] technical detail

[Eric Greisen]

Elias Brinks writes:
 > Hi Eric,
 > 
 > I have a technical question. It relates to how spectral line channels
 > are averaged in both AVSPC and SPLIT and how the new reference
 > pixel is determined.
 > 
 > Here is an example I am struggling with. I have a dataset with 302
 > channels; reference channel is chan 128. I would expect the following
 > when asking for simple boxcar averaging of 2 channels:
 > a new dataset with 151 channels (which is what I get) and a reference
 > pixel at 63.75. However, the ref. pix. is at 64.25. Why is that? I would
 > expect that, as the number of channels is nicely even, one would average
 > chan 1 + 2, which has an average channel of 0.75 (and so on), rather  
 > than
 > what the programme seems to do, averaging chan 2+3 which gives a new
 > pixel center of 1.25, etc.
 > 
 > Also, what happens if I take 151 channels, ref.pix again at 128.
 > Same 2 pixel boxcar averaging? This seems to give the same 64.25 as
 > central pixel.
 > 
 > And a related question, what happens with the edge channels when
 > only 1 channel is left for a 2-channel boxcar?
 > 
 > This kind of detail isn't reflected in the help/explain files  
 > (understandably)
 > and short of digging through the source, I wouldn't know how to find  
 > out.
 > I'm hoping you just know the answer   ;)

I do not necessarily know the answer and would not want to swear that
all such details are handled the same everywhere.  But consider that
ref pixel 128 says that the 128'th pixel runs in frequency from 127.5
to 128.5.  It will be combined with one that runs from 126.5 to 127.5
and called pixel 64 in the new units.  The center of 64 is what was
127.5, point 128 is half again along or 64.25.

If the input ref pixel were 127 then 63.75 would be the new ref
pixel.

AVSPC requires that the same number of input channels contribute to
each output channel and so will drop the high end input channels as
needed.

Eric Greisen