[evlatests] Tcal temperature sensitivity at Ku-band

[Rick Perley]

    Calibration of the Ku-band 'flux densities' data has been completed, 
and has revealed clear evidence of strong temperature sensitivity of the 
switched power calibration system. 

    As noted in an earlier message, the run was taken over a 30 hour 
period, within which a large (but not extraordinary) air temperature 
swing occurred -- from -5C at 5AM to +16C at 5PM. 

    Raw switched power values show a large apparent change in gain 
during the run for the Ku-band receivers -- typically 20%, and inversely 
correlated with temperature -- highest power at lowest temperature.  
This raw switched power variation is a combination of two systems -- the 
noise diode's injected power, and the subsequent post-amp amplifier gain 
-- with different temperature coefficients.   These two could be 
separated by monitoring the correlation cross-products (visibilities) 
directly -- however, for this run, the repeated correlator crashes meant 
that 'set and remember' failed, so that the T304 attenuator values were 
changing many times (dozens) throughout the run.   But we can apply the 
switched power corrections, and look at residual gain variation as a 
function of time.  This residual should reflect that variation of the 
switched power calibration system.  The difference between the residual 
gain and the raw switched power values should reflect the variation by 
the amplifier chain. 

    I point out here that a competing important gain effect -- the 
variation of gain with elevation -- must first be removed.  At Ku-band, 
this effect is fairly small (typically 5% in power, or less, over the 
elevation range, although for two antennas, (15 and 18) the effect is 
much larger).  Probably more importantly, the observing methodology was 
such that there is no correlation of elevation with time for timescales 
longer than an hour.   

    The calibration method first applied the switched power, then 
measured and removed the elevation gain dependence.  (Good agreement 
with the previously determined gain functions was found).  The residual 
apparent gain, as a function of time was then determined, using 
approximately 25 sources well distributed in both elevation and time. 

    The result of this is that for nearly every antenna, a clear 24-hour 
gain periodicity is seen.  The power amplitude of this variation is 
significant, 6 to 8 %, (peak-peak).  The 'corrected' gain is lowest in 
the early morning, and highest in the afternoon -- indicating that the 
noise diode corrections are *over-correcting* the actual gain variations 
from the amplifiers.  Virtually every antenna shows the effect equally 
-- most of those whose gain effect is not so easily visible are either 
bad pointers (gain residuals higher than normal), or have clearly been 
mis-calibrated by the switched power system (most notably ea13). 

    One antenna stands out from the others, however, and gives us a clue 
on how to fix this problem.  ea15 has only a ~2% effect at most, which 
is apparently out of phase with with the others.  It turns out that this 
antenna has its noise diode completely insulated.  (Thanks to Bob Stupak 
for doing this, and reminding us of this).   This supports the 
conjecture (by Bob Hayward, after last year's 'flux density' effort) 
that the problem arises due to a 'heat pipe' effect, whereby the 
exterior temperature changes is conducted to the noise diode and 
post-amps by conduction down the feed horn. 

    I met with engineering a short time ago -- they are laying plans for 
thermally isolating various components on a few antennas at Ku-band.  If 
we can repeat this run (or at least a significant portion of it), during 
weather with a large temperature change (and light winds), we should be 
able to solidly establish the origin of the temperature-sensitive 
post-correction gain, and assure ourselves that the changes put in place 
to prevent the multiple correlator failures are actually working.