[evlatests] Stacking Baseline Boards: 784 MHz dual pol'n w/ 125 kHz channels

[Michael Rupen]

I did a first wideband test of stacking hardware (Baseline Boards [BlBs]) for
higher spectral resolution.

Using 96 BlBs to cover 6 subband pairs (16 BlB per subband pair) gave
1024 channels in each of RR and LL over 128 MHz bandwidth per subband, for
a total of 2 x 6144 channels over 784 MHz.   The attached figures show

   1- The spectrum for one baseline (10s integration time), RR on the left
     and LL on the right.  12288 channels per baseline per dump.

   2- The total Baseline Board usage (mostly for fun, for WIDAR geeks).

This is:
   * the first multi-subband use of Baseline Board stacking

   * the most BlBs we've ever used at once (the previous max was 32 I
     believe) -- here we're using 3/4 of the Baseline Boards (most of
     the other 1/4 were unavailable for hardware and connectivity reasons)

It worked amazingly well.   Kudos to Sonja in particular for getting
this going; and to Martin as well -- the CBE just _worked_ for this, which
was pretty cool.  Actually, the whole WIDAR team deserves lots of credit for
this, as this mode exercises almost all aspects of the real-time correlator
software differently than we've used them before.

The one "hiccup" is the missing IF 2 for RCP.  If any one of the 8 BlBs
stacked up to produce a pol'n product for a subband fails, that entire 
pol'n product for that subband vanishes.  Here we unexpectedly lost
b103-t-5 to the "effing" problem.  There are risks here, but the payoff
is pretty nice!

Other tidbits:
* I averaged to 10s to bring the data rate down to a modest ~3 MB/s.
   Initially I dumped every second (33 MB/s) but the CBE couldn't keep
   up :)

* A quick look shows no lost data, apart from that one subband.

* It takes ~90s to set up the correlator, dominated by the time required to
   send messages to each of the BlBs in turn.  If we can send out signals
   to all the BlBs at once this will speed up enormously (should take only a
   second or two).

* The CBE figured out the appropriate correlator chip/LTA configuration
   in 1.5 seconds.  This is only slightly slower than simple configurations
   like OSRO1.  Martin's move to smarter (and compiled!) code has certainly
   paid off.

* TelCal swallows these data fairly happily, but fails for maybe
   1/8 of the subband/pol'n pairs.   It's pretty much all or nothing -- you
   get all antennas, or you get none.  The amp, delays, etc. when they are
   computed look reasonable.
   - TelCal has occasional problems with 1024 channels per pol'n product with
     1sec dumps for a single subband pair.  2048 channels per pol'n product
     (16 BlB pairs per sb pair) makes TelCal quite unhappy, with 1sec dumps.
     I think this is not surprising but perhaps Keith can weigh in on the
     subject.

* The Configuration Mapper correctly complained when I tried to ignore
   crossbar board limitations in assigning BlBs to subbands.  And then it
   was appropriately happy when I fixed this up.  Kinda nifty, given that we
   haven't tried this on the telescope before.

The next steps for this particular mode are:

   1- do a more careful data reduction, over the weekend (check stability,
     bandpasses, SNR, etc.)

   2- try a line source, early Monday morning (to avoid the possibility
     of mucking up the correlator in the middle of weekend science)

   3- do a real continuum experiment (including flux cal, blank sky)

   4- try using all 96 BlB on a single subband (!), which should give
     the full 12288 channels over 128 MHz, for 10 kHz channels.  Ok,
     this one is just for fun, but doesn't it sound neat?  Of course
     we'd better hope that every single one of those BlBs is working...

   5- check narrower subband bandwidths


All in all a very fun day.

                    Michael
-------------- next part --------------
A non-text attachment was scrubbed...
Name: C_osro8x6_12nov10.pdf
Type: application/pdf
Size: 274081 bytes
Desc: C_osro8x6_12nov10.pdf
URL: <http://listmgr.nrao.edu/pipermail/evlatests/attachments/20101112/a9eef8b3/attachment.pdf>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: C_osro8x6_blb12nov10.gif
Type: image/gif
Size: 48513 bytes
Desc: C_osro8x6_blb12nov10.gif
URL: <http://listmgr.nrao.edu/pipermail/evlatests/attachments/20101112/a9eef8b3/attachment.gif>