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

[Ags]

I thought I would spend today getting to the bottom of how my 1100D behaves with respect to noise at various ISO settings and temperatures.

My process was as follows: I set up my camera with a remote shutter release inside my fridge. Once it had cooled down, I started a sequence of 16 1-minute subs with no cool-down pauses. I did a sequence for every ISO setting of my camera. The sequences were started at around 10C and at the end of the sequence the camera was at 16C.

While the shutter was snapping away, I extended my image analysis software to process directories of images and spit out suitable spreadsheets. Someone recommended exiftool as a way to read camera temperature from raw files, so I used that with another program I wrote to rename the TIFF files with their ISO and temperature (which could then be read by my software while it analysed the images).

I will present my graphs below, but here are my conclusions:

1. At least with my camera, avoid ISO 800 - I get extra white noise and hot pixels at ISO 800. At room temperature this behavior shifts to ISO 400, by the way.

2. Over the six degree range I looked at, temperature only has a subtle effect on noise.

3. Even if you live in Siberia, ISO 3200 and ISO 6400 simply produce too much noise.

Firstly a graph of the total noise pixels produced at each ISO at approx 10 and approx 16C:

The noise stays relatively constant regardless of temperature, going up a bit at high ISO. But most of these noise pixels are very dim. I did see a weird effect where total noise would drop as the temperature went up. I do not know why, but I saw this repeated for several ISOs. My theory is that in the early shots the sensor is heating up rapidly and perhaps this rapid change in temperature can cause noise effects.

Now a graph of the fainter noise pixels (1-10% illuminated, pixels converted to greyscale):

The picture is clear - ISO 3200 and 6400 have enough gain to record background static going on on the sensor. As has been said elsewhere, the ISO is not making the chip more sensitive, it is only increasing the gain applied to the analog signal - in this case increasing the noise signal enough to be quantized to a non-zero pixel value.

Now a graph of the fairly bright noise pixels, much nastier customers!

Again, the highest ISOs have disproportionately more bad pixels. Interestingly, ISO 800 is also relatively nasty, and ISO 1600 gives less middle-level noise than ISO 800.

Finally the hot pixels - defined by me as anything with a greyscale value over 40%. As you can see, I won't be using ISO 800 ever ever again!

[Chris]

Wow, ISO 1600 it is then with the 1100D at that temp range, and I really hope the ISO 800 results don't apply to my 350D's. Good work! oh and again my apologies for the mix up on the other thread I didn't even register that your full name was Agnes:D

[Photosbykev]

Testing it in the fridge is probably similar to a cold clear night, neat idea. There is a significant step change 800-1600 which is interesting.

A test that might be warrented is to see if a longer exposure say 600seconds at iso400 is comparable to 150seconds at iso1600 i.e longer subs and lower iso against shorter subs and high iso

[Ags]

Yes, I have more tests lined up - including different sub lengths and cooling strategies... But it took me the whole weekend to put this together - oddly the biggest problem I had was renaming the files (I kept making stupid mistakes and scrambling the results).

The question is indeed if 4 minutes of ISO1600 are as nasty as 1 minute of ISO 6400. The way noise varies with ISO is not a simple relationship to the ISO value - the drop from ISO 800 to ISO 1600 is one example of this, but if you compare ISO 400 to 1600, you see no increase in noise, which is all sorts of strange.

These first tests were run at 1-minute subs because firstly that is the sub length I can actually achieve with my limited equipment, and because I wanted to see the effects of temperature (a longer sub would show results over a range of temperatures, whereas a short sub isolates the temperature variable).

But based on these results I think I will stick to ISO 400 and 1600.

[JamesF]

That's a very interesting (and unexpected result. The ISO800 results are so out of line with the rest that you have to wonder whether it's some sort of systemic problem with the camera design or just an unusual fault with yours. You have to wonder if the results are similar for other cameras, too. I'd not be surprised to find that a fair bit of the firmware is shared between the different camera models so if it's a fault there then I guess it's quite possible.

James

[Ags]

If it had been only one ISO800 sub that had these weird values, I would have thrown it out. But all 16 were the same. And at room temperature, I've seen ISO 400 do the same.

[Gina]

Most amazing results and unexpected as has been said There seems to be a lot more going on than one would expect. A range of dark subs at various settings just don't seem to tell the whole story. I'm sure there's noise being generated by the signal level. I seem to have found that bright objects produce more noise than faint ones in my DSO imaging. eg. the big and bright M31 has shown far more noise than my fainter images - M8, M20, M51, Cygnus Loop / Veil Nebula etc. Though that could be because it's a galaxy and most of the other things I've imaged are nebulae with narrow band emissions.

I look forward to furter test results from you Agnes

[Ags]

That is a strange anecdotal observation Gina. Intuitively, you would expect a bright object to need less stretching so the noise should be easier to control. I will have to figure out a way of simulating a dim light source and see if that influences the noise levels!

[Gina]

That is a strange anecdotal observation Gina. Intuitively, you would expect a bright object to need less stretching so the noise should be easier to control. I will have to figure out a way of simulating a dim light source and see if that influences the noise levels!

Yes indeed, that's what I'd have thought.

[GlassWalker]

Gina, is the the bright object = more noise you see due to the type of subject? For instance, I find narrowband a lot easier to deal with since it has better contrast. A wideband subject (even if brighter) is lower contrast due to light pollution, and it puts the fainter parts into the noise floor.

On the original point of this thread, those charts are interesting but I'm struggling to think how or why that may occur. Only that possibly the camera changes its internal operating mode in some way at some settings to optimise the output.

[Photosbykev]

Ags, maybe a stupid thought but it might need some adjustment if you haven't done it all ready. I assume you have all noise reduction turned off (High iso and Long exposure) and the Auto lighting Optimising turned off ? On my 5D MkII I turn these off for all my astro work, I don't know what similar functions the 1100D has.

[Ags]

Glasswalker, I agree there is something complex going on in the way the camera handles ISO. I had expected something much more straightforward.

[Gina]

Gina, is the the bright object = more noise you see due to the type of subject? For instance, I find narrowband a lot easier to deal with since it has better contrast. A wideband subject (even if brighter) is lower contrast due to light pollution, and it puts the fainter parts into the noise floor.

I was talking about using a wideband camera for objects with narrow band emissions - ie. not using NB filters. The only filter I currently use with my scope camera is an Astronomik CLS CCD filter to block light pollution.

On the original point of this thread, those charts are interesting but I'm struggling to think how or why that may occur.

Same here!

Only that possibly the camera changes its internal operating mode in some way at some settings to optimise the output.

Yikes - you could be right

[Ags]

Kev, I made sure all noise-related options were off on the camera.

[Gina]

Kev, I made sure all noise-related options were off on the camera.

I did with mine too. Went through all the options.

[steppenwolf]

That is a fascinating set of results and the poor ISO 800 result is a great surprise - I'd love to know what is going on there.

[Gina]

I'd like to know if it's peculiar to that sample or whether it applies to all 1100Ds. I think I'll have to see if I can reproduce some of Agnes's results. I have 4 1100Ds altogether - 3 on the imaging rigs and one unmodded spare.

Let me add that I'm not detracting from her results in any way, just that some results might just vary from one particular camera to another of the same model. I would also like to say that I am very grateful for all her work in this. Thank you Agnes

[VigdisVZ]

Great job, I'm very interested to see what comes up when you add more data with different sub lengths and strategies.

If you're at it maybe also determining that there is no difference between starting at the lower ISO end vs the higher end, just for the sake of thoroughness.

[GlassWalker]

Out of interest, are the noisy pixels random in location or are they fixed?

[Ags]

If anyone wants to add to the test results, you can send me the images and I will process them. I could send you the tool, but I am still developing it, so that would get rapidly unmanageable. I had a look at some results from a D60 shot by a non-astro work colleague, but I could not use the data unortunately.

[Ags]

I am adding extra code to determine if the noise is random or fixed, and report the two populations separately.

[dph1nm]

My Canon 1000D has fairly awful non-random read-noise structure at ISO800 which is not present at ISO1600. I then read somewhere that Canon use different circuitry on the camera to generate ISO1600 than for their lower ISOs - so this might explain the differing noise behaviour.

NigelM

[davew]

Hi Agnes,

Have you read this ? ........... http://www.stark-labs.com/craig/resources/Articles-&-Reviews/CanonLinearity.pdf

Dave.

[Shibby]

The way noise varies with ISO is not a simple relationship to the ISO value - the drop from ISO 800 to ISO 1600 is one example of this, but if you compare ISO 400 to 1600, you see no increase in noise, which is all sorts of strange.

I remember reading something somewhere that suggested some DSLR cameras might actually adjust the black point dependent on ISO... (or was it exposure time...) Which was a somewhat worrying thing to read, if only I could find it...

[Ags]

Shibby, you probably read it in the link posted by davew above :-)

Davew that is an interesting read, although it seems a bit like a sales pitch for Nebulosity! It goes a long way to explaining many of the strange effects I see in the data.

I did not know about dcraw - that seems like a good way of converting RAWs without dependency on DPP with less black-box software post-processing.

I think the claim that anything over 400 ISO is a waste was taken back after lots of discussion on CN, but I don't recall the details.