[daip] [JIRA] (CAS-8477) exportuvfits not honoring multiple spectral windows in FEED table

[Eric Greisen]

On 04/04/2016 09:19 AM, David Mehringer (JIRA) wrote:
>
>      [ https://bugs.nrao.edu/browse/CAS-8477?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=86238#comment-86238 ]
>
> David Mehringer commented on CAS-8477:
> --------------------------------------
>
> Hi Eric,
> Before I start implementing this, I want to be certain I'm not missing something. Part of MS FEED tables' primary key is the spectral window ID. It looks like there is no corresponding column for this in uvfits AN tables, but there is a FREQID keyword for AN tables. Therefore, it looks to me that the only way to handle this issue when writing such MSes to uvfits is to write multiple AN tables, each with its own FREQID (corresponding to spwID). Would you say that is the most reasonable way to proceed?
>
>> exportuvfits not honoring multiple spectral windows in FEED table
>> -----------------------------------------------------------------
>>
>>                  Key: CAS-8477
>>                  URL: https://bugs.nrao.edu/browse/CAS-8477
>>              Project: CASA
>>           Issue Type: Bug
>>           Components: casacore-fits
>>     Affects Versions: Release 4.6
>>             Reporter: David Mehringer
>>             Assignee: David Mehringer
>>
>> As noted in CAS-7081, exportuvfits is not writing information for multiple spectral windows in the FEED table. The only solution seems to be to write a separate AN table for each spw in the FEED table, since AN tables do not have an associated column for spwID but do have a FREQID keyword.
>
>
>
> --
> This message was sent by Atlassian JIRA
> (v6.3.1#6329)
>


UVFITS does not have a FEED table.  There is an FQ table which is what 
should be used.

UVFITS has a fundamental limitation that all spectral windows must have 
the same number of spectral channels.  If that does not apply to the 
particular data set, then the spectral windows must be separated into 
separate output files each containing only those windows with the same 
number of spectral channels.  In each of these, the FQ table should be 
used to contain the frequency offset between the reference frequency (in 
the main header) and the frequency of each spectral window at the 
reference channel.  The "IFs" in the FQ table are the spectral windows 
and appear in the data "cube" that is each visibility sample (COMPLEX x 
STOKES x FREQ x IF).  A 2 window example:

Ncol 5  Nrow 1    Sort cols:
     Table has     1 keyword-value pairs:
    NO_IF    =            2
    Table format incompatable with FITS ASCII tables

COL. NO.1   2                 3                  4                5
ROW FRQSEL IF FREQ        CH WIDTH           TOTAL BANDWID      SIDEBAND
F NUMBER                  HZ                     HZ                 HZ
1  1  0.0000000000D+00    9.765625E+02       5.000000E+05           1
1     8.0000000000D+05    9.765625E+02       5.000000E+05           1

FQ IDs are usually used to separate data that in fact should never be 
reduced together at the same time.  They were invented when the VLA 
users asked to have L and C band in the same data set.  It is one of the 
worst design features ever put into the system.  Multiple spectral 
windows in the same frequency band should be reduced and imaged together 
as much as possible - hence those are "IFs".

Eric Greisen