[evlatests] EVLA Sensitivity

[Rick Perley]

    As reported earlier today, I've have taken much test data in a quest
to derive the system sensitivity. 

    Data were taken in spectral line mode in all bands and IFs. 

    Some overall conclusions:

    1) A bit more than 10% of the observations were bad -- as reported
earlier.  The characteristics are simple:  Both antennas give no fringes. 

    2) The phase changes for the EVLA antennas (only) between scans.
The data were taken by cycling through the bands: 
       L -> C -> X -> K -> Q

    for IFs A, then B, then C, then D.  After these 20 observations,
the cycle repeats.  Each re-observation at L, C, etc. for each IF had
a different phase. 
    This is a serious problem which will prevent normal (non-self)
calibration (for those observations were the user changes band between
source and cal). 

    3) All other system characteristics are excellent:

    * Bandpass shape
    * Phase stability (within a scan)
    * Amplitude stability
    * Sensitivity  (see below). 

    No drops, spins, or other bizarre phenomena were observed. 

    4) Sensitivity. 

    I first worked to see if a sensible relation between system 
sensitivity,
as measured by the rms fluctions in amplitude and/or phase, and the
easily accessed 'aips weight' existed.  I derived the amplitude rms across
the 3.125 MHz bandwidth using 100 central, well calibrated channels,
over the 1.5 minute integration, and plotted this against the estimate
of the system sensitivity provide by AIPS  -- the 'Aips Weight'.  Plotting
the measured rms against the inverse  square root of the aips weight
(which is nominally equal to the rms noise times some constant), a
satisfying linear relation was found.  There are issues remaining
with the slope and intercept, but I defer this to another day.  (Besides,
it's Friday afternoon at 5:30 PM now, and there are other things to do ...).
    I thus use the 'aips weight' as a proxy for system sensitivity, and
give the results below for IFs A and B. 

    L-band:

    IF A:  Antennas 14 and 16 rank 25th and 20th in sensitivity.  Not so
good.   We do not expect them to be better -- any gain in sensitivity
due to reduced Tsys is probably more than offset by the loss of
efficiency.   Their ranking clearly depends on the elevation -- all measures
of sensitivity show the VLA antenna rapidly degrading as elevation
decreases, while the EVLA antennas degrade far less.  This is expected,
due to reduction of spillover. 

    IF B:   Antennas 14 and 16 rank 11th and 22nd in sensitivity. 

    C-band:

    IF A:  Antennas 14 and 16 rank 9th and 3rd. 

    IF B:  They are 2nd and 1st -- and ahead of the other antennas
by a significant factor.  16B has the downconverter filter which
rejects the aliased signal, so we expect to see this antenna-IF to
properly reflect the much lower Tsys.  the 14B result is a bit
surprising, -- is there some anti-alias filter in it too?

    X-Band:

    IF A:  Antennas 14 and 16 are 9th and 1st.
    IF B:  They are 25th and 3rd. 
     

    K-Band:

    IF A:   They are 19th and 13th. 

    IF B:   They are 22nd and 18th. 

    In general, we expect K-band performance for the EVLA to
be the same as the VLA -- which they are, more or less.
    It was clear in these data that we have a pointing error for antenna 14
which increased the system noise (and decreased the gain) as the source
dropped in elevation. 

    Q-Band:

    Antenna 16 gave no fringes. 

    Antenna 14 was weak, and the 10% gain drop seen in antenna 14
over the course of the run was matched by a near factor of two drop
at Q-band (supporting the notion of a pointing error).  Even for the
scan giving the strong fringes, the system sensitivity (as measured
by the aips wt.) was 25th out of 25. 

    Meaningful system sensitivity measurements at Q-band cannot
be done without referenced pointing.  Even at K-band, the results
will be questionable without referenced pointing. 

    Overall:

    I'm generally satisfied with the sensitivities in IF channel B, which
includes the downconverter filter to prevent aliasing, in all bands
except Q -- for which it seems probable the poor results are due
to pointing errors.