(Note update Dec. 2004;
see also SPIE contribution of June 2004 with differing numbers)

Date: Sun, 21 Mar 2004 23:31:13 -0500 (EST)
From: John Monnier <monnier@umich.edu>
Subject: Tests of Lincoln Labs CCD (new StarTracker)

NOTE: Please send corrections to monnier@umich.edu

Basic Tests of Lincoln Lab CCD to determine dark current, electronic gain,
and readnoise.



1) Default conditions using vxworks-start-LL (2004Mar21): 4x4 binning
2) Thermo-Electric Cooler set for 43.2 kOhms (-5.3 C), Current 0.96 Amps
3) Temperature in lab: 65 F (18 C)
4) All lab lights off, cardboard blocking entrance holes of camera.
5) Using IDL script called ~iota/Idlstuff_g3.2/Startracker/st_stats.script
   to control tint, read 20 frames, take mean, calculate sigmas.

T_int	Mean	Sigma
(ms)	Cts*	Cts

2	192	12.9
4	281	14.6
8	465	17.9
12	659	20.0
20	1067	26.1
40	2141B	34.6
100	5477	51.8
200	10953	67.6
250	13641	75.0
300	16227	61.3 **** LL CCD SATURATES in 300 ms ON DARK
400	16383	0

[* mean counts in this table are not BIAS corrected.  Based on counts vs
t_int graph below, I will use a BIAS of ~100 ADU -- I do not know how to
measure this directly but this was found by extrapolating to tint=0s]

** note that the t_int estimates might not be completely accurate
due to millisecond-rounding errors in VME realtime system**

By only using the first 8 points (up to 6000 counts in the mean),
we can plot variance vs. mean to come up with the following statistics
(see attached graphs):

electronics gain: 2.1 e-/ADU
READNOISE:	  11 ADU = 23 e-

[4x4 binning]
BIAS (tint=0s):	  ~100 ADU
Dark Current: 	  54.6 ADU/millisec = 115 e-/millisec   ** 4x4 pixels **
	    : OR  3.4 e-/ms/pixel   = 7.2 e-/millisec/pixel

Comparison to Specification in Lincoln Labs Notebook

1. Readnoise
Spec sheet says:  	7+/-1 e-
We measure:		23 e-

** Our 'readnoise' is not a CDS readnoise, but a single-read #

Conclusion: CCD much noisier (3x) than expected.

2. Dark Current

Spec sheet says: 	0.15 nA/cm^2 at -5 C
We can convert this to e-/ms, since we know a pixel is 21 microns on
a side.

   Dark Current: 0.15nA   1 e-        (0.0084 cm)^2
		 ------ X --------- X ------------- =
		 cm^2     1.6e-19 C   (4x4 binning)

		= 66 e-/millisec for 4x4 binning

My measurement: 115 e-/millisec

Conclusion: Dark Current is 2x higher than expected

Note: readnoise and dark current are 2-3x higher than expected, as
if the gain was incorrect. However, mean-variance measurements clearly
indicate electronics gain at IOTA Of 2.1, less than 6.8 stated in
Lincoln Lab Documents.

1) Maybe there is a light leak, but I took great pains to shield
detector and eliminate all light sources.
2) Maybe I made some other mistake, so take this analysis as preliminary
until it can be checked.



Performance is 2-3x worse on key specifications (read noise, dark
current) than expected.

Dark current is **TOO HIGH** for observing faint sources at IOTA. The
detector (at -5 C) actually saturates on dark counts in 300 ms!! and we
used to slow-guide at this tip-tilt rate on the faintest, most obscured


1) WHY is read noise so much higher than Lincoln Lab Measurements ? Are we
introducing noise somewhere? More tests needed.

2) We *must* reduce dark current, presumably by cooling detector to a
temperature much below the current -5 C (hopefully down to -50 C).
Currently we CAN NOT observe sources as faint as possible with previous

Expected termperature dependence of readnoise
According to spec sheets (for measurements at 22C and -5C)  and general
principles, the dark current is expected to double every ~5.7 C degrees.
Using the results from above, we can calculate the temperature needed to
achieve for dark current fluctuations to be equal to read noise for
t_int(ms) integration:

  23 e- = sqrt( 115 e-/ms *2.^(DeltaTemp/5.7) * tint)
=> DeltaTemp = 18.9 * (0.66 - log10(tint) )

Hence, if we want good performance for tint=100 ms, then
   DeltaTemp = -25 degs (compared to -5 C)

This corresponds to an operating temperature of -30 degs C
(dark current of ~5 e-/ms per 4x4 binned pixel)
According to Lincoln Lab documention, this corresponds to
a thermister resistance of R~179 kOhms (compared to 43 kOhms currently).

If current one-stage chiller cannot cool detector to desired temperature,
then perhaps the existing glycol chiller at IOTA can be used as a first


A TEST to reduce detector temperature and measure dark current change.

I increased the cooling so that the thermister resistance is 70.2 kOhms
(1.49 Amps current -- near 1.50 amps limit). According to chart, this
is -14.2 C.

New Test:

tint	mean	sigma
ms	cts*	(cts)

2	141	11.5
10	271	14.7
50	1015	24.4
100	2014	34.6
200	4076	45.8
300	6140	51.8
500	10203	66.7
750	15125	78.7

* Again, will use Bias of ~100 cts (for tint=0)


Temp	Dark Current
deg C	ADU/ms/ (4x4 binned pixel)

-5.3 	54.5
-14.2   20.2

For 5.7 deg C doubling rate, we would have expected ~18 ADU/ms, a bit
lower than the 20 ADU/ms observed but close..

Possible Explanations for why dark current did not go down as much as
 a) 5.7 deg formula not accurate over full range (-14C -> +22 C)
  ** FAVORED **
 b) Small Light leak (not likely given precautions)
 c) Thermister -> temperature conversion incorrect (using Lincoln Labs
 d) CCD not at thermister temperature.

I favor (a), therefore dark current is reduced with temperature almost in
accordance with expectations.

Dark current does indeed decrease as the detector temperature is lowered,
nearly as much as expected.  Might need to cool CCD lower than -30C
(see Conclusion/ACTION from above memo) to achieve desired performance.
Furthermore, if readnoise is reduced from 23e- to the expected 7e-
readnoise then we will need to chill detector even further to avoid being
limited by dark current fluctuations. Of course, we should try to cool the
detector as cold as possible within reasonable means (maybe -50 C).


Date: Mon, 22 Mar 2004 20:43:27 -0500 (EST)
From: John Monnier <monnier@umich.edu>
Subject: pixel-based stdev

Wes wanted me to check that the stdev across the frame was equivalent to
the stdev for a single pixel (successive reads).

I measured 100 frames at tint 2ms. For each pixel in the (9,8) frame, I
calculated the sigma.

The mean counts in frame was 211 ADU.
The mean counts vary across the frame from about 180 to 250 ADU, depending
on which pixel.

The frame-based sigma yields stdev 13.3 ADU.
The median pixel-based stdev was   13.7 ADU.
The stdev varied from 11-15 ADU depending on which pixel you look at.
There is no correlation between stdev and mean counts when looking at the

Conclusion: The "large" 23 e- readnoise seems real, not effect of
averaging over frame.



Date: Mon, 13 Dec 2004 07:41:55 -0500 (EST)
From: John Monnier <monnier@umich.edu>
Subject: ccd memo: update

hi [...],

I just ran a quick re-test of CCD to document the improvements due to the
extra cooling stage.

In summary, the new cooling has drastically improved the CCD performance,
as expected and reported.  The new dark current is 3.8 e-/millisec instead
of 115 e-/millisec.  The readnoise also is lower for some reason (13e-
compared to 23 e- in March -- compare to 7 e- from spec sheet). For some
reason there is large (~1000 ADU) bias level now, but this doesn't matter
I suppose.

See measurements below.



Basic Tests of Lincoln Lab CCD to determine dark current, electronic gain,
and readnoise.

Original: 2004Mar21
Updated:  2004Dec13


1) Default conditions using vxworks-start-LL (2004Dec13): 4x4 binning
2) Thermo-Electric Cooler thermister set for 196 kOhms (<~-30 C)
3) Temperature in lab: 62 F (16 C)
4) All lab lights off, cardboard blocking entrance holes of camera.
5) Using IDL script called
   to control tint, read 20 frames, take mean, calculate sigmas.

T_int   Mean    Sigma	Mean	Sigma
(ms)    Cts*    Cts

2       192     12.9	1080	5.6
4       281     14.6	1081	5.7
8       465     17.9	1085	6.0
12      659     20.0	1090	6.0
20      1067    26.1	1099	6.3
40      2141    34.6	1121	7.0
100     5477    51.8	1194	9.0
200     10953   67.6	1327	11.1
250     13641   75.0	1396	12.3
300     16227   61.3 	1466	13.6
400     16383   0	1609	14.9
1000			2511B	22.3

* mean counts in this table are not BIAS corrected.

>From variance vs mean graph,
[using bias 100ADU in 2003Mar, 1058 ADU in 2004Dec]

		2004March	2004December
electronics	2.1 e-/ADU	2.7 e-/ADU 	[prob same within noise]

READNOISE	11 ADU=23 e-	5.1 ADU= 13.8 e-