The following is a report on the stability of the dark current seen in the UH8k CCDs.

Date: Thu, 16 Jan 1997 16:12:20 -1000
From: Jean-Charles Cuillandre 


A quick information about uh8k dark counts level, linearity
and stability.

I have two median darks and one median bias I built from the 
last uh8k December run.  I'm giving here the dark counts for each 
chip for a bias, a 24mn exposure and a 40mn exposure. The 24mn 
and 40mn have been corrected from the bias level so the following 
numbers are really related to the exposure with no offset.

First let me remind you the geometry of the mosaic (chip numbers)

		2	3	5	4

		0	1	7	6

chips 0,1,2,3 are read first and 4mn later it is chips 4,5,6, 7 turn.

Levels are in ADUs, "*" means less or equal than 3 ADUs
(sorry my soft does not go lower but it has no influence here).
(all the images have been overscan corrected with a 2D modeling)

Bias:		*	*	*	18

		*	*	5	10

Dark24		44	45	49	64

		31	44	43	34

Dark40		75	79	82	115

		53	77	73	58

The ratio between dark40 and dark24 should be 1.6666 and is actually:

		1.70	1.75	1.67	1.79

		1.71	1.75	1.70	1.70

there is a slight difference (#3%) so it means that
the dark current is not linear at a level of 3% from
a 24 to a 40mn exposure. Now what is the sky level
on 20mn V and I exposures (very roughly):

		I: # 7000 ADUs (gain ranges from 1.5 to 2.2 e-/ADU ...)
		V: # 500 ADUs

So the sky background photon noise in ADU (using g=2e-/ADU) is 
on 20mn exposures: N(V filter)=16 ADUs N(I filter)=59 ADUs.

The level error introduced by rescaling the darks for let's say a
24mn to a 40mn exposure time (that won't happen that much often
I guess so this is the most catastrophic case), is 3% of 50 ADUs
(average dark level for the 8 chips on a 24 mn exposure) say
1.5 ADUs. It appears that the noise associated to the dark current
responds to gaussian distribution characteristics: the noise is about
5 ADUs for a average dark count level of 50 ADUs (still g=2 e-/ADU).
On the 40mn exposure it is 77 ADUs and the noise is 6.2 ADUs. 3%
of it is 0.2 ADUs that has to be compared to the sky photon noise
in V it is # 1% of it and # 0.4% of it in the I band. So the effect
is negligible.

Now the time stability, I made a difference between a dark
taken at the beginning of the very last uh8k run, the first
was taken on the 04-01-97 and is a 40mn exposure, the second
one is a 20mn exposure taken one week later on 11-01-97.

The flux (after overscan correction and bias correction) is:

04-01-97: 40mn	78	81	83	107

		54	79	73	57

11-01-97: 20mn	37	36	40	49

		25	36	34	27

(note that these levels are very close from the measurements
obtained from the December run).

The average ratio is a bit higher than in the previous 
example and is equal to 2.15 against a supposed factor of 2 
(error of 7%), again let me point out that 20 to 40 is a 
very rare case. As the rescaling method should be used to
jump only from 20 to 10 or 30 mn exposure times.

After rescaling and subtraction, no structure 
appears in the image (flat and level #2 ADUs).
We can conclude that the dark is stable in time both for the
dark current generation and structure among the chips.

My advice is: build a very good dark (typically 20mn) rather 
than having several poor S/N darks with different exposure time 
AND build a good bias for use with the short exposures (less
than a minute). For observing time of a few minutes (#5) I 
would recommend to do a good dark with this exposure time as 
the rescale factor from the 20mn dark would be like 4.
For any kind of run then the following exposures should be enough:

bias (B), dark5mn (D5), dark20mn (D20)

And building other darks with the simple rule: for a exposure time N mn:

D(N) = (D20-B)*(N/20) + B

(adding the bias avoid having to do a dark AND
a bias correction during data reduction, always
the battle to save disk space...)

Same comments about dome-flats. But this is another story and
the time spent for each frame is only due the readout time.

It might be good to pass this info to future observers. And 
in a few words the conclusion of this mail is that uh8k dark current 
non-linearity residual when rescaling darks is negligible compared
to the background photon noise obtained with the current filters.
(I'm not talking about the B filter but it won't be used that much,