We took data at 5 bias points (V=5.9 -7.1 V, Delta_V=0.3 V) and 4 temperature points (T=5.5 -7.0 K, Delta_T=0.5 K). For each pixel, two spectra were taken at each point. One was taken with no light to obtain the pedestal value, and one with a light pulse corresponding to an average of ca. 2 photoelectrons at 6.5 K and 6.5 V. Each spectrum contained 20000 events; the integration time in the ADC for each event was 90 ns.
Run to run errors due to the mechanical (rather than soldered) connections, handling of the light guide, and general changes in the cassette due to thermal cycling were kept at 2%. Run to run light levels were monitored by keeping a single chip in the same slot throughout the calibration. The channel to channel gain variation of the preamp was measured to be ca. 3%. Each VLPC channel has a 100 kOhm current limiting resistor. Since each VLPC, under normal operating conditions, draws a significant amount of current ( ca. 1 microAmp/channel at 6.5 V and 6.5 K), there is a VLPC bias offset of ca. 0.1V depending upon the current. The cumulative contribution of these effects on the gain and the observed number of photoelectrons results in a 5% systematic uncertainty.
There was also a channel to channel light level variation in the
cassette.
We measured the variation by cycling a single hybrid (the same hybrid
that was used throughout the characterization to monitor the
run to run light levels) through each of the positions in the cassette
and
measuring the number of photoelectrons. Since the run to run
variations were small, we assumed that all of the variations observed
were due to light level changes
from channel to channel in the cassette and from intrinsic QE
variations across the calibration chip. In extracting the data on the
relative quantum efficiency, we further assumed that the variation
observed was due primarily to light level variation in the
system and not due to QE variations across the hybrid. To verify this
assumption, we iterated three hybrids (including the
previously
mentioned `calibration chip') through each of the slots in the
cassette and found that the observed light levels were the same within
a given slot to better than 5%.
The final light level correction
factor was
determined by referencing the light level in all 64 channels
(determined from the iteration of a single hybrid through the 8 slots)
to the first channel. The total systematic error on the relative
quantum efficiency is then
about 7%.