[evla-sw-discuss] Keeping track of spectral windows at the EVLA: SDM vs. BDF

[Michael Rupen]

Hi folks --

   please find attached a portion of the SDM intro document dealing
with the association between spectral windows in the SDM, and those
in the BDF.  A previous iteration was approved by Francois so I think this
is kosher.  Note that there are subsections at the beginning & end you
probably don't care about at the moment.  Much of the rest of this
message is detailed and EVLA-centric; I've cc'd several CASA and ALMA
folks in case they have comments, since this affects the filler at least.

   Basically we have to ensure that the order in which the spectral windows
are stored in the BDF matches the order in which those spectral windows are
referenced in the dataDescriptionId array in the SDM's ConfigDescription
Table.   The order in the BDF is required to be such that swbb and sw occur
in ascending order; the combination of swbb and sw is unique for the EVLA
(for ALMA they have to worry about the sideband but that's a detail).  So
if we agree on swbb and sw and send that information to both the CBE and
MCAF we should be OK.  We are sending all other information like by sending
the VCI's subarray element to MCAF as well as the Confiruation Mapper, so I
propose using that same route here.

swbb is the "baseband ID" in the SDM sense.

   1. I propose that we make this one-to-one with the VCI's baseBand name
     (e.g., A1/C1).
     - This is currently an optional attribute; the suggested change would
       make this required.  If this does violence to the VCI (e.g., all
       names are currently optional), we should create a new attribute, maybe
         swbbName  enumeration or string

   2. The BDF and the SDM use the BasebandName enumeration:

        BB_1  BB_2  BB_3  BB_4 BB_5  BB_6  BB_7  BB_8  NOBB

     I propose adding more useful EVLA names to this enumeration:

        A1/C1_3BIT  A2/C2_3BIT   A/C_8BIT
        B1/D1_3BIT  B2/D2_3BIT   B/D_8BIT

     Part of the point here is to avoid confusion with WIDAR's baseband IDs,
     which are *not* the same (though they are closely related).

     2b. If it takes any time to add to that enumeration, I propose
       something similar to Barry's scheme:

         BB_1 ==> A1/C1_3BIT
         BB_2 ==> A2/C2_3BIT
         BB_4 ==> A/C_8BIT

         BB_5 ==> B1/D1_3BIT
         BB_6 ==> B2/D2_3BIT
         BB_8 ==> B/D_8BIT

   3. The required ascending order is then:

        BB_1  BB_2  BB_3  BB_4 BB_5  BB_6  BB_7  BB_8  NOBB
        A1/C1_3BIT  A2/C2_3BIT  A/C_8BIT  B1/D1_3BIT  B2/D2_3BIT  B/D_8BIT


sw is the spectral window index (an integer) in the BDF, which is NOT the
same as the spectral window ID (tag) in the SDM.

   1. We could use the VCI's subband element's name attribute
     (an optional ASCII string at the moment), making that mandatory; or we
     could create a new attribute like
       swName   string
     or perhaps better
       swIndex  integer

   2. The value of name/swName/swIndex would have to be an integer starting
     at 1.   Ascending order is then obvious.

The CM would pass these new attributes on to the CBE, which would sort
all the extant swbbName/swIndex pairs into the proper order and store them
in the sdmDataHeader element.

The Executor would pass these new attributes along with the rest of the
VCI subarray fragment to MCAF, which would extract the swbbName/swIndex
pairs and figure out from those the proper order for the dataDescriptionId
array.

Of course this begs the question of who comes up with swbbName and swIndex
to begin with.  This seems properly left to the OPT: there is enough
information there, and these values can be determined independently for
every scan.  It would be preferable to have a deterministic algorithm for
doing so.  For swbbName this is trivial.  For swIndex within a given
swbbName I propose:

   1- sort by the lower edge of the lowest frequency channel in each subband
     (to match the ALMA ACA correlator)
   2- for remaining matches, give the lower swIndex to the SpW with
     the narrower channels
   3- for remaining matches, give the lower swIndex to the SpW with the
     fewer total channels
   4- at this stage we have only exact matches left.  If the OPT or whatever
     is explicitly setting the StB filter to be used, the lower filter gets
     the lower swIndex.

All of this of course is negotiable.

A couple points:

* In the BDFs, all SpW in "lower" swbbName's come before all SpW in "higher"
   swbbName's, regardless of the actual frequency tunings.  A/C comes
   before B/D, full stop; similarly data from A1/C1_3BIT comes before
   A/C_8BIT.

* In the BDFs, within a given swbbName, the above scheme would put
   non-overlapping SpW in frequency order (low to high).

* None of this says ANYTHING about the ordering of the actual Spectral
   Windows seen by the observer in post-processing, or even in the SDM.
   That is a separate discussion, which I hope to initiate tomorrow or
   Thursday.



Comments?

            Michael
-------------- next part --------------
A non-text attachment was scrubbed...
Name: alf.pdf
Type: application/pdf
Size: 69215 bytes
Desc: spw_sdmbdf.pdf
URL: <http://listmgr.nrao.edu/pipermail/evla-sw-discuss/attachments/20100216/5f0213d2/attachment.pdf>