[evlatests] RFI Spectral Sweep

[Ken Sowinski]

Some comments below.  Some of this was learned by running the file,
with some corrections, again and watching what happened.  This new
version if you want to look at it is rfisurvey_002 from 1902 to 1952 UY.

Ken


On Thu, 17 Mar 2011, Rick Perley wrote:

>    1) The highest seven 'tunings' (from 43 through 50 GHz) have no
> data.  (This from last evening's test run).  'No Data' means just that
> -- not 'integer zeros', but absolutely nothing at all.  On the previous
> test, only the highest three tunings were empty.  And on the first test
> (two days ago), all Q-band tunings had data.  What has changed?

The CBE is unable to keep up with the data.  Two days ago there were
fewer antennas used.  When the file finished it took the CBE more
than a minute to clear out all the still unwritten data.  If any one
wants to think about the numbers, there were two polarization products,
512 channels each, from each of eight subband pairs, for all the S band
antennas, 14 antennas, I believe.

>    2) All lower sideband tunings (all of X-band, and Ka-band from 26
> through 32 GHz) have spectra which are both mis-labeled and
> mis-ordered.  Mis-labeled means that the frequencies in the headers are
> incorrect.  Mis-ordered means that the eight subbands are incorrectly
> placed:  subbands 1 and 8 are interchanged, subbands 2 and 7 are
> interchanged, etc.  The former problem is corrected if we load in an
> alternate SDM (from /home/mctest/evla/mcaf/workspace).  The latter
> problem is not corrected by this -- but Lorant has written a special
> BDFin script which correctly re-orders these subbands upon loading into
> AIPS.   It seems to me that these new issues need to be corrected
> automatically without users having to invoke special scripts.

Prompted by Bryan's earlier email email, I describe what was done.
There was one VCI document for the entire file configuring eight 128 MHz
subband pairs to cover 1 GHz all fed to two quadrants, providing 256 lags
each.  The tuning sequence starts with L band and progresses up through
Q band, in 1 GHz steps.  When a sideband change occurs, the executor
creates a VCI "modify" document to the correlator.  That all seems to
work as expected.  The production MCAF does not see this, so creates
incorrect frequncy labels.  The test MCAF examines the "observe" document
(where the SSLO is already examined) to learn the sideband sense and
correctly label the subbands in frequency space.  When the OPT generated
VCI documents for sideband changes it must have changes the ordering of
sidebands in a way to make the result look like a natural ordering;
MCAF, I suspect, does nothing like that.


>    3) Subband #6, for the LCP correlation, for every single tuning, is
> bad.  The spectrum is 20 dB too high, and is composed entirely of
> noise.  RCP is fine.

No sign of this when I looked at the data going by just now.  Did
you apply a switched power adjustment to the data using bad switched
power values?

>    4) And, most extraordinarily, the L-band spectrum (1 -- 2 GHz
> tuning), in both polarizations, is entirely blank!  The spectral density
> and spectral shapes are correct, but there is not a single RFI spike
> visible!  (We wish ...)  All other tests of this script gave proper
> L-band spectra.  The 'blank' spectra are on all antennas.

I think this was caused by the lack of a dummy scan at the beginning of
this file.  There is now a 10 second dummy scan before the L band
measurment begins.

>    5) In the lower S-band tuning only (2 -- 3 GHz), subband 3 is
> completely corrupted in both polarizations -- the digital radio
> transmisions (which are in this subband) are weakly visible, and the
> spectral floor is 20 dB too high (like those reported in note 3, above).

No comment.  Maybe it won't happen again?