[evlatests] Results from Switch Power Calibration Test

[Rick Perley]

    Finally, the actual results from the calibration test run two days 
ago.   (Previous reports all dealt with problems found in the system).   
This is a long report!  Impatient readers should skip to the bottom, and 
read the 'Summary' and 'Recommended Action' sections. 
   
    The goal of the experiment was to see if simple application of the 
switched power (and the associated Tcal and antenna efficiencies) 
provided correct gain calibration for a source of know flux density. 

    To do this, I observed 3C286, at an elevation of 55 degrees, using 
the old narrowband (two subbands) OSRO mode, with a single pair of 
128-MHz-wide subbands within each of the eight cassegrain bands.  For 
each band I chose two frequencies, reasonably well separated, which are 
known to be free of RFI. 
    The target source (3C286) was observed twice at each band.  The band 
sequence followed was:  L, S, C, X, Ku, K, Ka, Q.  Prior to the X-band 
observation, referenced pointing was done to try get the main beam 
squarely on the target source, and the results applies to the subsequent 
high frequency bands. 
    The weather was clear, but rather breezy (winds 15 to 20 mph).  
Unfortunately, the time chosen was about 6PM, so the referenced pointing 
offsets were remarkably large -- over an arcminute for most antennas!  
They also changed significantly for some antennas between the two 
referenced pointing determinations, perhaps a result of the wind.  These 
problems certainly affected the two highest frequency bands, as the 
diligent reader will note in the results given below. 

    The data will filled using the latest gain curves (updated this 
morning), and using the default atmospheric opacity model. 
    The data (both visibilities and switched power values) were edited 
for the problems noted in previous reports.   The AIPS program 'TYAPL' 
then produced a corrected visibility file.  Basic calibration (delays 
and bandpasses) was done.  The antenna gains were then generated, using 
modern (and we think correct) values for the flux density of 3C286.  The 
results of this are the gain corrections needed to convert the 
visibilities to correct flux densities.  A perfectly calibrated system 
will then show gain values of 1.0.  Values greater than 1.0 mean the 
visibilities are too low, and values less than one mean the visibilities 
are too high. 

    The process of converting raw visibilities into corrected 
visibilities involves the values of the switched power noise diode 
(Tcal), and the antenna efficiency, epsilon.  The sense of the 
correction is such that:

    - If the values of Tcal that are utilized are lower than the true 
value, then the resulting visibilities will be too low, causing the 
derived gain values will be greater than 1.0
    - If the values of the antenna efficiency that are used are lower 
than the true value, then the resulting  visibilities will be too high, 
causing the derived gain values to be less than 1.0. 

    From the derived gains, I plotted histograms for each frequency at 
each band.  For each of these, I give below the median offset, and the 
fraction of the antennas which are with 5% and 10% off this median. 
    For each band, there are usually a small number of antennas which 
are truly discrepant.  These are identified in the listings below.  Some 
effort into understanding why these antennas are discrepant would be 
useful. 

    1)  L-band

    Median gain = 0.95 for all IFs.  50% of the antennas are with .05 of 
this, 85% are within 0.10. 
    Four antennas are very discrepant:
        ea04 on all IFs.  Median gain = 1.35
        ea17 on IF 'A' only, gain = 1.28.  (PDif values are clearly 
weird for this IF).
        ea19 on all IFs.  Median gain = 1.35
        ea28 on all IFs.  LCP was flagged out (PDif = 0), RCP median 
gain = 1.35. 

    2) S-band

     Five antennas are greatly discrepant.  The statistics below ignore 
these:
     Median gain = 0.92 for A&C, 0.90 for B&D.  75% of antennas are with 
.05, 100% within .10
    The five discrepant antennas are:
           ea01 on all IFs. Median gain = 0.30.  (The Tcal values must 
be way to low --  I bet it's still at 1.0)
           ea07 on all IFs.  Median gain = 0.6
           ea04 on all IFs.  Median gain = 1.5
           ea10 on all IFs.  Median gain = 1.3
          ea12 on all IFs.   Median gain = 1.3
    In addition to these, ea16 is bad on the AC side only:  median gain 
= 1.15 

    (It gets a lot easier from here).

    3)  C-Band

    Median gain = 0.95 for all IFs.  75% are within .05, 100% are within .10
    There are no seriously discrepant antennas. 

    4) X-Band (Wideband systems only)

    Median gain = 0.97 at 8.2 GHz, and 0.92 at 11.3 GHz.  50% are within 
.05, 90% within 0.10. 
    There are no seriously discrepant antennas. 

    5) Ku-Band

    Median gain = 0.95 at 16.8 GHz.  50% within .05, 90% within 0.10.  
Because of the phase wrapping problem noted yesterday, no useable data 
from the other IF were gained.
    There wee no discrepant antennas.  However, the gain distributions 
show that antennas low (or high) in one polarization are also low (or 
high) in the other -- suggesting an efficiency origin.  For example, 
ea22 and ea23 are at the low end of the distribution in both 
polarizations, (tabulated efficiency probably too low) while ea11 and 
ea20 are both on the high end (tabulated efficiency probably too high). 

    6) K-Band

    At 19 GHz:  Median = 0.92; 70% within .05, 85% within .10
    At 25 GHz:  Median = 0.90;  50% within .05, 80% within .10
    ea10 is quite discrepant on all IFs:  median gain = 1.15.   
    ea23 gain = 1.2 in IF 'D',
    ea14 gain = 1.15 in IF 'D'

    7) Ka-Band

    At this band, the troubles with pointing stability begin to show up, 
causing the much wider distribution, with the tail stretching to the 
high gain end (the visibilities are too low due to the pointing, making 
the gain corrections too high).  For both this band and Q-band, I've 
used the smaller of the two gain solutions, since this one is much more 
likely to be closer to the truth. 

    At 29 GHz:  Median = 0.90.  65% within .05, 90% within .10
    At 36 GHz:  Median = 0.87.  60% are within .05, 75% within .10
    Serious Discrepant Antennas:
        ea17 on all IFs:  gain = 1.25
        ea08 on all IFs:  gain = 0.72 at 29 GHz, and 0.62 at 36 GHz.
        ea14 on all IFs:  gain = 0.72 at 29 GHz, and 0.62 at 36 GHz. 

    8 Q-Band

    At 41 GHz:  Median = 0.80.  40% within .05, 60% within .10
    At  48 GHz:  Median = 0.85.  30% within .05, 50% within .10.
    But in fact, the median is probably lower than 0.80 at 48 GHz, as 
the distribution shows a long tail extending to higher values, doubtless 
due to the pointing offset. 
    The only antenna identifiably bad (separate from pointing, unless 
its pointing is truly abysmal) is ea03, whose gain is 1.10 at 41 GHz, 
and 1.25 at 48 GHz. 

    Summary

    At all bands -- ignoring the few discrepant antennas, and presuming 
the spread at the high frequencies is pointing-induced -- the agreement 
in the gains is very good.  The only significant issue other is the 
median offset.  Correcting for this should enable quite accurate 
amplitude gain calibration to be made -- good to 5% for sure -- without 
use of a standard source. 

    Recommended Actions

    I think the most likely cause of the median offset is an error in 
the tabulated antenna efficiency.  I suggest that a special version of 
TYAPL be generated in which the antennas gain value is modified by the 
appropriate values, to produce gains with no median offset.  Another 
test observation of 3C286 should then be made, but this time late at 
night and under clear calm conditions to confirm the sense and magnitude 
of the changes. 
    If this proves good, then a more extended observation, utilizing a 
wider range of elevation, should next be made, to check the robustness 
of the opacity and elevation correction procedures.   (This can be part 
of a real science observation, simply by adding in 3C286 once an hour, 
for example).