[evlatests] EVLA and VLA Tsys measurements

[Rick Perley]

    In reviewing last night's L-band data, some notable problems in Tsys 
values were seen.

    In general, the measured values for (backend) Tsys for EVLA antennas 
are noisier (less accurate) than those from the VLA.  In addition, some 
antennas from both arrays are showingTsys fluctuations (or errors in the 
measurement of Tsys) of ~0.5K on ~minute timescales. 

    I would expect to see an rms level in measured value of Tsys of roughly:

     sigma_T ~   2*(Tsys/Tcal)*Tsys/sqrt(B.T)

    where B is the bandwidth, and T the effective integration time.  The 
factor '2' comes from the 50% switching noise diode. 

    The ratio Tsys/Tcal is normally set to about 20 for both EVLA and 
VLA.  The characteristic averaging time is 10 seconds.    The BW was 
12.5 MHz.  From these, we should see:

    sigma_T ~ 0.0035*Tsys -- about 0.1 K for a normal Tsys. 

    Accurate measures of changes in Tsys are important, as they are 
needed to convert the correlation coefficients to visibility flux 
densites. 

    1) EVLA Antennas:  (all at L-band):

    In general, the apparent noise in the measurements is at least twice 
that expected.  This is easily seen for antennas 13, 14, 16, 21, 24, and 
26.  But the Tsys fluctuations are much better on 23 and 17.  
    Two antennas --

    2) VLA antennas:

    In general, the noise is at or better than the calculated level.  
But there are notable exceptions:
    3C, 5A, and 4 (all IFs) are quite poor.

    All antennas show some degree of Tsys fluctuations on ~minute 
timescales.  These are usually less than 0.1K, so are comparable with 
the noise in the measurements.  But some antennas are spectacularly worse:

    1(all IFs) shows ~1K variations. 
    28 C and D (in LCP) shows 1 to 2 K variations. 

    By looking at the detailed calibration solutions, we could determine 
if these Tsys variations are real (in which case the corrected 
visibilities will not show any fluctuations) are not.  The current 
dataset is not well suited for this, as the calibrator scans are only 50 
seconds long, so i haven't tried this.