[evlatests] Correctness of Switched Power Corrections

[Rick Perley]

    At the ECSV meeting this morning, diverse opinions were again 
expressed (without evidence) that application of the switched power 
does, and does not, put the resulting amplitudes on a correct scale. 

    On the one side, Vivek and I have long claimed that the resulting 
amplitudes are correct. 
    On the other side, others say that the resulting values are often 
significantly in error. 

    To settle this controversy, I have taken the 'flux density' dataset, 
and examined the gain solutions, using sources whose fluxes (I claim) 
are accurate to 1 or 2 percent.  In this note, I report only on the 
L-band results -- later this afternoon (if no other meetings intervene), 
I'll report on the other bands.

    The flux density data were taken in 'wide-band' mode (yes, huge 
overkill, but you never know when it might be useful).  For this 
purpose, I extracted the two 64-MHz-wide subbands centered on 1465 and 
1865 MHz.  I applied the delays, bandpasses, opacity, and the switched 
power values (after suitable editing of variant points), then calculated 
the antenna-based gains, using my three northern sources J0217+7349, 
J1153+8058, J1800+7828, with fluxes determined from my full analysis 
against 3C286.  These are chosen to minimize elevation dependencies, 
which are very important at the high frequencies. 

    Results:

    Both sides of the argument are correct! 

    1)  At 1465 MHz, the mean amplitude gains are 0.97 and 0.98 in RCP 
and LCP.  The median values are 0.93 for both.   At 1865 MHz, the means 
are 0.98 and .96, while the medians are about 0.94.   The significant 
difference between mean and median tells us there are significant 
outrider points ...

    2) 85% of the individual antenna values are clustered with 10% of 
each other.  But we have some very discrepant antennas:  ea19, ea25, and 
ea28 are all at least 30% in error (this is in amplitude -- so the power 
is the square of this).   The next worst antennas are ea14 and ea13.  It 
turns out that the largest errors are on the 'positive' side, which 
skews the averages to be above the medians. 

    The median values of ~0.94 look pretty good, but remember that this 
is amplitude, so the 'median baseline' will show a flux density error of 
the square of this -- or about 12% in error -- too high.  The cause of 
this offset is I think most likely due to an error in the assumed 
antenna efficiency, but could also be due to a bias in the determination 
of the Tcal values.