[daip] GMRT and Channel numbering

[Eric Greisen]

Hans-Rainer Kloeckner wrote:
> 
> Hi Eric,
> 
>     Could you please clarify how AIPS is handling the frequency 
> calculations and in particular using non integer values for the reference
> pixel. The reason why I'm asking is that the  GMRT data generally starts 
> with a reference pixel of 0.5 rather than a full integer.  So here are 
> my questions: How does AIPS compute the frequencies, does it compute the 
> centre frequencies of a channel bin? If so than the reference pixel of 
> 0.5 makes no sense and actually produces a wrong frequency.
> 

I have been through the GMRT frequency business multiple times and aips 
has been guilty of errors when averaging channels in SPLIT and SPLAT. 
However, the actual frequency computation has always been correct. 
Reference pixel 0.5 means the edge of channel 1, so the actual frequency 
of channel 1.0 is the ref frequency + 0.5 * increment.  I think GMRT 
sometimes chooses 0.5 so that the USB and LSB both have 0.5 which was 
the exact frequency which divided the two.  They then have opposite 
frequency increments.  The frequency computation is linear

    freq = freq0 + (ch - refch) * increment

and the center of a channel is an integer channel number (as are the 
centers of all pixels in WCS/FITS).

Eric Greisen