Long Term Drift

[Tom Droege <tdroege2@earthlink.net>]

I just looked at some current darks and bias frames and compared them to 
those of Data Set 20.

The mean bias value changes a few counts over 56 days.  When I compare the 
frame VCO temperatures, they are different (by 9C) .  When I work it out, 
the change is 50 ppm per C (of order 50 ADC counts).  This is well within 
the range I expect (+/- 100 parts per million from the parts used).  I 
think there is a good chance that the VCO temperature can be used to 
predict the mean bias frame value.  This is just the ADC electronics 
drifting around by the expected amount.

There was no significant change over time for dark current.   It remained 
constant at about .9 e- per second for the temperature of operation (-16.8 C).

The dark current noise is much larger than can be expected from the mean 
number of electrons.  This is because  I was not running cold enough (-16.8 
C) to get away from the non-linear dark current effects.  Still, the noise 
from the dark current was 20 counts vs a typical sky noise of 50 
counts.  Down in the noise, I think.

Those of you that are offended that the ADC drifts around with temperature 
should consider that there is an offset current into the ADC provided by a 
resistor from a reference supply of half full scale = 32,000 counts.  If 
this one resistor has a change of 100 ppm per C, then this will produce a 
change of the mean bias frame of 3 counts per C.  There are a dozen or so 
parts with similar effects.  It is thus not surprising that we observe 6 
counts of ADC shift per C.  While I can think of ways to greatly reduce 
this drift, I doubt that they are used by designers at the big 
observatories.  I suspect that they just calibrate the shift with dark and 
bias frames - what say experts?  Once one gets below 100 ppm, it gets 
sticky everywhere.  One has to worry about precision resistor aging, 
thermoelectric effects, and the like.  It is almost easier to put the whole 
of the circuit in an oven.

To study this, one might take some of the recent data and look at the mean 
value of the covered pixels compared to the VCO temperature.

I suspect that with a little effort, one can predict the bias value from 
the VCO temperature and the CCD temperature.  This can then be compared 
with the covered pixels as a check.

In summary, one has to worry about the shift of a bias frame with 
temperature.  But we knew that.

Tom Droege