How far to cool a CCD

[Catanonia]

I have always taken my images at -20 with the QHY9, but there was a lot more left in the unit as it hardly ran at 30% power.

So last night I set it to -40 and it does make a big difference.

Now going through a new set of dark captures to -40

Has anyone else noticed the same. It came to me when watching a video on the net about the professionals using liquid nitrogen to cool thier CCD's

Currently running at -40 and 50 - 60% cooler power on the QHY9, but it is below zero ambient here.

[FraserClarke]

Yep; cooler is generally better. I've measured the dark current on our ST8 at different temperatures, and the dark current in a 900s exposure is lower than the read-noise below -15C. So I set our cooling to -20C, so that I can run with the same temperature the whole year round -- means the calibrations are slightly easier to do

BUT, you do tend to get fewer hot pixels the colder you go, which is a good thing. I'm not sure how much the QHY9 is affected by hot pixels?

It also depends on the type of CCD you have. Most professional devices are "back-illuminated" chips (basically, the chip is turned round and the substrate material removed). This makes them more efficient (QE>95%), but very expensive! It also makes them a lot more susceptible to dark current, which means you need to cool them a lot more. Most professional back illuminated CCDs operate at 150-170K (-120C to -100C). Another effect is that the colder you go, the lower the QE gets, but it's not a very strong effect.

Most amateur CCDs are "front-illuminated", which is a bit less sensitive, but a heck of a lot cheaper to make. It also produces a lot less dark current, which means you don't need to cool them as hard to control the dark current.

[ianaiken]

I cool to -20. The manual states service temp is -20 to +30 so I stick to -20. I've read people going much lower but I don't know if this will have a long term effect on the camera health.

[riklaunim]

Compare darks at 0, -10 C... and so on. At some point the dark current at your typical exposure time will be so low that cooling it even lower won't give much. That KAF doesn't need extreme cooling.

[ollypenrice]

My Atik 4000s can just hang onto minus 20 in the summer - usually. So I have tended to run at that throughourt the year. However, in tryng to get better results from the H36 we have stopped using darks and switched to a master bias, defect map and dither guiding. This is certainly better on the H36 - much better - so I will be giving it a go on the Atik 4000s as well. This would allow more flexibility in terms of setting the temperature, for one thing.

Olly

[Zakalwe]

I guess the limiting factor will be what the camera can do below ambient- the delta T. It'll be easy to get low temperatures in winter, but what can be achieved in spring when (hopefully) the weather gets a bit warmer)?

[narrowbandpaul]

Well understood fact about dark current faling with falling temperture. When the camera is cooled, it only takes around 4-5C lower to halve the dark current...ie even a 5C change will have quite an effect on the current and hence noise. Have used the QSI583 at -32C so far (80% cooling)

fraserclarke has summed up the difference in sensor design very well. The KAF sensors are well known for being noisier than the KAI and especially the sony chips.

as for measuring dark current: very easy to do. Set the temp to the value you use at night (if you can). Take a variety of long exposures say, 1s, 30s, 1min, 2min, 5min, 10min, 15mn, 20min, 30mins. Subtract off a high quality bias frame. Measure the average count for a group of pixels (around a 100x100 box). Try to ensure there are no hot pixels in your box. Also record the std dev of that box. Thats the noise. Ideally the dark noise shouldnt exceed your read noise. The noise you measure in that box is the total noise (inclusive of read noise), so the total noise should not exceed 1.41*read noise.

you can measure the read noise also very easily. Subtract one bias frame from the other. Measure the std dev of a large group of pixels (again around 100x100). Divide this by 1.41.

You do not need to know your noise in electrons (you should though since thats all that is physically relevant) to know when your dark noise exceeds your read noise. But to know your dark current in electrons you do. To do this, assuming that your camera has been set up correctly, divide the full well by the saturation signal. This is the gain. Multiply the noise and average count in your darks to convert to electrons. Plot average count vs exp time, the gradient is the dark current. The noise vs exp time would also be interesting.

Testing cameras is a lot of fun!!

[FraserClarke]

To do this, assuming that your camera has been set up correctly, divide the full well by the saturation signal. This is the gain.

Another way to get the gain is to take a sequence of flat exposures with different exposure times, and then make a plot flux versus variance (std-dev^2) similar to the way narrowbandpaul says for the dark current. The gradient of this line is the gain.

It also helps to take pairs of frames at the same exposure time and divide them. That corrects for the variations in the flat (i.e. dust spots) and just leaves the noise inherent in the chip. Again you'll then need to divide the measured std-dev by 1.41 to account for the fact you're effectively measuring the noise twice.

That test will probably also show you how the gain changes as a function of flux level -- also known as the non-linearity of the chip. It's usually not a straight line unless you pay a lot for a chip. I have an ST8 where the variance vs flux line turns over at ~45,000 counts -- presumably because the pixels start bleeding in to each other and the noise gets correlated. Nasty stuff. Haven't used any data above 40,000 counts on that detector since I found it out :-\

Pretty irrelevant for making pictures, but important if you want to do photometry with your images.

I guess it must take a certain kind of mind to be happy spending most of the time taking images of the inside of the dome just so you can test the chip

[narrowbandpaul]

Hi Fraser

Yes, I was going to mention the photon transfer method. Its certainly the way I would do it, but it is slightly more involved. Spent a good year doing this at Sheffield Uni (or trying to!) with some CMOS sensors.

Paul

[Gina]

Oh dear, I must be getting old - all that's gone right over my head