The noise produced by a Canon 1100D at various ISO settings and temperatures

[Gina]

Gina: could you upload 1 each of all of your 120s subs @ iso 1600?

Can do Except that I haven't got all the temperatures yet.

https://dl.dropbox.com/u/39971951/D_2012-11-09_23-38-43_0338_ISO1600_120s__-10C.CR2

https://dl.dropbox.com/u/39971951/D_2012-11-13_20-50-01_0091_ISO1600_120s__-4C.CR2

https://dl.dropbox.com/u/39971951/D_2012-11-08_01-36-10_0123_ISO1600_120s__0C.CR2

https://dl.dropbox.com/u/39971951/D_2012-11-12_02-09-12_0114_ISO1600_120s__6C.CR2

https://dl.dropbox.com/u/39971951/D_2012-11-18_22-37-11_0321_ISO1600_120s__8C.CR2

https://dl.dropbox.com/u/39971951/D_2012-11-16_23-34-40_0166_ISO1600_120s__10C.CR2

https://dl.dropbox.com/u/39971951/D_2012-11-09_01-37-23_0150_ISO1600_120s__12C.CR2

https://dl.dropbox.com/u/39971951/D_2012-11-19_22-43-45_0306_ISO1600_120s__15C.CR2

https://dl.dropbox.com/u/39971951/D_2012-11-20_22-41-45_0354_ISO1600_120s__18C.CR2

I have done some testing myself and found that below +10 the noise is very low and below +5 there is mostly random noise. My tests were done at 120s @ iso 1600 with a 450D and 600D.

Some of the test images with the 600D at 120s @ iso 1600 were pretty much the same as bias frames...maybe there is no point in going extremely cold or to use dark frames at all since the noise is random?

Yes, I've found the noise very low at very low temperatures.

I can post some pics when i have done some sorting of all test images.

That should be interesting

[Ags]

From what I can see in the data so far, cooling below zero does not bring much additional benefit, although perhaps the 12 minute subs might change this picture.

[Gina]

Currently a -2C darks test run is under way - overnight.

[Xplode]

Here's a quick grape i made from some testing with my 600D.

As can be seen here the noise is pretty much the same under with a sensor temperature under 10C so there's isn't that much point in cooling it to very low temperatures.

I'm doing some more testing now with the sensor at -2C and the STDEV starts at 60 for 30s, 60s and 90s (bias STDEV is 56), 63 for 150s , 69 for 300s, 89 for 600s and 121 for 1200s so it doesn't scale at these low temperatures.

[Cath]

The thing is it's not just the sensor noise that limits the low noise ability, the other noise sources are the analogue amplifier components between the sensor and the ADC input, and also the ADC itself, plus the power supply's noise powering the sensors/amplifier/ADC, and also the driver electronics that drive the various pins on the sensor, plus how well they have laid out the PCB tracking etc.

[Gina]

Interesting graph but it doesn't agree well with my findings - I'll be particularly interested in what you get from my data

I think the noise reduction may be higher with longer exposures too. Maybe at higher ISO too - see my thread with results at ISO 6400 and a couple of hours.

[Gina]

Last night's run at -2C went well with an EXIF T of -2C throughout

[Ags]

I can't wait to get hold of the full set of your data, Gina!

I think the point about thermal noise is that it is cumulative with time, so it is not going to be apparent in shorter exposures.

[Gina]

I can't wait to get hold of the full set of your data, Gina!

Just 4 runs left now Agnes I can send it to you then if you like and send the odd 12m subs through Dropbox.

I think the point about thermal noise is that it is cumulative with time, so it is not going to be apparent in shorter exposures.

I agree there. With the shorter subs and particularly at low temperatures it will be readout noise that 's the problem I think. Though I would have thought that would have been temperature sensitive.

[Ags]

Yup, let's send the 12 min subs through drop box.

[Gina]

Yup, let's send the 12 min subs through drop box.

OK - will do

[Gina]

+4C dartk test run now in full swing The jolly old obsy is getting a bit of a buffeting but holding firm as the storm rages. The rain has stopped for present.

[Gina]

Last night's +4C darks test run completed sucessfully AFAICS

[Gina]

Planning to do the -8C darks test run tonight as it looks like being quite cold. That'll spread out the 12m subs nicely too. Set up cooling ready for when it gets dark.

[Gina]

-8C darks test run is well under way now with the temperature holding steady

[Gina]

One thing I've found with doing these tests is that my Mk.2 Peltier cooled camera box is a great improvemnet on the Mk.1. The sealing is much better and I'm maintaining much drier air in the box. I'm still getting below 20% RH with a dew point of -15C. This will enable me to use much lower temperatures if it turns out that this would be beneficial. And this is where Agnes's analysis of my data will be very helpful

This also means that I could take the data runs below the current -10C after completing the current data runs. I have tried zipping one complete run and get a compressed file of 1.8GB. Just checked and the limit on Dropbox is 2GB for the total file space. A file of this size would take a long time to upload of course, but I do leave my computer on 27/7. I would have to clear out current data files amounting to about 500MB but I'll warn everyone before I do that to allow anyone who wants them to download.

[Ags]

I have run my program over Gina's 480sec ISO 1600 data, and my findings are similar to Xplode's. Below 10C the noise is quite constant, above 10C the noise soars.

Here is a graph of counts of the dimmest noise pixels (above 0.3% illuminated, so they would be visible when converted to 8-bit color):

And here a graph of the noise up to 16% illuminated, showing the same curve (two separate graphs as thee are a couple of orders of magnitude between these two sets of data):

There is one kind of noise that shows sensitivity to very low temperature, and that is brightly illuminated pixels (17%+ illuminated):

This last noise type is only a few hundred pixels at most, but bright enough to affect the image. If the noise is random, it will not be reduced very well by stacking (at least following the Average methodology).

@Xplode, how did you get your camera to 36C? Did you put it in a frying pan? :-)

[Gina]

That's very interesting Agnes Low level noise actually seems to get worse below zero! I'm dying to see the full results - should be extremely interesting. I think I might have the USB stick in the post Monday if all goes well.

I think I'll hold off doing any tests below -10C until we have the results of the current data set. Looks like going much below 0C doesn't gain anything.

There seems to be some significant difference between cooled astro cameras and DSLRs even where both are CMOS.

Wish we could think of a way of determining what happens to the signal - I'm by no means convinced that the gain is purely a function of ISO.

[Photosbykev]

Try Mediafire Gina http://www.mediafire.com/ it gives you 50Gb of free storage

[Gina]

Thank you Kev I'll take a look at that tomorrow

[Ags]

I don't think the low level noise gets worse below zero - we need the full data set so we can exclude random variations between shots, which I think explains the upward curve towards -10.

An experiement I would like to try would be to see if the temperature influences the dynamic range of the camera - I expect it does. According to an article posted earlier in this thread, the camera subtracts a constant from the real raw signal to hide the dark current from the RAW file. If that happens, then I expect that a hot camera has less dynamic range than a -10 camera, which would be an advantage for cooling even if noise seems unchanged.

[Xplode]

Ags: these newer cameras gets a lot hotter than the older 1100D, 1000D and 450D. Before i started my exposures i also turned on liveview for a few minutes, liveview heats up the sensor a lot compared to taking shots.

In the freezer i reached a stable temp of -2c with the 600D, but was down to -8c with the 450D with the same settings.

I think the 600D insulates better than the older cameras so it won't be able to get rid of the heat that easy.

[Ags]

I didn't know the 1100D was old - it only came out last year...?

[Xplode]

The 1100D of course isn't old...i should have wrote sensor since the 1000D, 1100D and 450D probably share the same sensor with some minor differences when it comes to image processing and possibly the AA filter has been upgraded to a better one.

[Gina]

Last night's -8C run was satisfactory with all but a small number of 12m samples recording -8C. 10 out of the 28 12m samples the EXIF T recorded -7C. The complete run is now on my main PC and also on the USB stick for sending to Agnes.

Here's the progress table up to date :-