[evlatests] Weights

[Rick Perley]

    Following today's discussion in the 'Transitions' meeting concerning 
the weights calculated by AIPS, some investigation was done. 

    I reviewed some old data, taken in early August, for which there 
were no 'Tcal' value entered in the parameters database.  It was quickly 
confirmed that the post-calibration weights for these antennas were 
incorrect, in the sense that they were very discrepant from working VLA 
or EVLA antennas. 

    This confirmation of reports from Ed on this matter generated some 
code review and tests.  Eric soon found that when AIPS sees a 'peculiar 
gain' (the value needed to convert correlation coefficients to flux 
density) equal to zero, an artificial value was assumed, and a weight 
which was not correctly computed from that artificial value, was 
derived.  This is sufficient to explain the observation noted above.  
The error in the weight calculation is by a factor of two or four 
(depending on whether one is looking at baseline weights or antenna 
weights as computed by ANBPL), and this will propagate through the 
calibration process.  The error will only occur when the peculiar gain 
is discovered by AIPS to be zero.  Users can check for this by listing 
the 'TY' table, and looking at 'Tsys' (which is the peculiar gain), or 
'Tant', which is actually the computed Tsys.  If either of these is 
zero, then the computed weights are incorrect, and will not be corrected 
through calibration.  Users will need to run WTMOD to input 'reasonable' 
values. 
    Eric reports this bug is fixed (along with another, related one, 
dealing with data filled without application of any peculiar gains). 
    Ken reports that the parameters database will always have values for 
Tcal, for all antennas, for all bands.  However, it cannot be guaranteed 
that the lack of a Tcal in that database is the *only* mechanism for 
causing the computed 'peculiar gain' to be equal to zero (and hence 
triggering the bug in AIPS). 

    To ensure that the Tcal values are being correctly applied 
throughout the observation and calibration process, we then executed a 
short observation, whereby two antennas had their Tcals modified by 
factors of two.  One was a VLA antenna, the other an EVLA antennas.  If, 
for example, the Tcal value is low by a factor of two, the expected 
effects are:
          1) The observed visibility amplitude from that antenna to an 
unchanged one will drop by sqrt(2).
          2) The (uncalibrated) computed weight will be too high by a 
factor of two. 
          3) The CALIB solution for that antenna will rise by a factor 
of sqrt(2). 
          4) The calibrated amplitudes and weights should be restored to 
normal value through the calibration process.

    The results are in perfect agreement with the expectations, for both 
the VLA and EVLA antennas.