[Ventzy]

According to physics the light has dual nature – like photons and like wave. As consequence of this dual nature, the noise always appear, in every digital image – sometimes it is invisible, sometimes it is annoying and intrusive. Photographers are in constant battle with it from their first shot on.

But now I am going to write about some other noise, the one connected with the camera and the equipment with which we are working. The noises generated by different reasons in the camera are too many and too different and I will analyze their influence over the final result.

In one modern camera of nowadays, no matter what technological scheme it has (mirrorless or DSLR), the light collector is the so called digital sensor, which is composed of millions, placed one by another, light convertors. I am not going to look in details the different constructive realizations, but their common thing is that they are all analogs.

Yes, it is indeed curious, but the fact is that the so called digital cameras are in their nature analog-digital. No matter what is the specific technology of the photo-receiver, after it reaches certain potential, the digital sensor is “read” and the data is sent to Analog digital converter (ADS). This ADS measures the gathered in every single receiver potential and converts the result in digital format, e.g. in a digital number. From that point on we can call the camera “digital”, because all subsequent actions will be only with digital data and within digital environment.

The aim of this article is not to analyze the methods and the ways of the digital processing of the signals. I am going to focus basically on the noise problem and the cause of its appearance in the analog part of the process, that is to say in the sensor – before it is converted into a digital amount.

From where, after all comes that extra noise? The function of the light sensor is to capture a photon and to transform it into an electron. So far, so good - certain amount of photons are coming, they are transformed into certain potential, i.e. into a certain number of electrons. Essentially we can compare this construction on a capacitor with a certain capacity.

But the electrons can occur there at the same place for other reasons determined by the physics. The main factor is the temperature. I suppose that everybody remembers, from the first physics lessons that the temperature in nature is the average kinetic energy of the atoms and molecules in a certain system. The higher the surrounding temperature is, the bigger is the chance of random electrons to be accumulated in our hypothetical capacitor, without having come in like photons. In other words if we heat the sensor, surely the higher the temperature, the more it will increase the number of the accumulated electrons, which have not been photons. And now starts my essential thesis, aimed at cameras.

The temperature of the sensor, of course, is determined by the temperature of the entire camera as well as from the temperature of the surrounding environment. But there is another factor that is particularly important. A serious reason for the temperature rise of the sensor is its working process and that of the electronics around it. The forced movement of the free electric charges, or simply put the conduct of electricity in one chain is accompanied by the release of heat due to losses, unless the chain is made from excess wires, which are still unavailable in the cameras. And these losses, according to the laws of physics are in fact heat.

Or in shorter and in simpler way we can say that if a camera works, it inevitably produces heat and that causes the increase of the noise under the same other conditions. Actually the reason I am dealing with this problem is an analysis whether and how the noise increases and how this in practice affects the final result – the photography.

At the moment I have available the ideal composition to determine the influence of the temperature – Pentax K3II. The camera is perfect for that purpose, because it is DSLR and I can test it either in a short time working mode of the sensor or in continuous mode - the so called live view, so that to be able to see the difference.

Of course, I will surely check, how the camera behaves at temperatures below zero of the body, at normal room temperature and also at that around the maximum levels to which the manufacturer guarantees normal functioning of the camera.

It is winter outside and it is the right time for testing the camera in real freezing conditions.



The room temperature of 20 degrees Celsius is easy to provide, so I give more attention to the upper limit of 40 degree Celsius, which according to Ricoh is permissible.


I supervise the process and do not rely much on the stove thermostat, because its tolerance is too big, so I am controlling the temperature by checking it with a digital thermometer. And of course I’m waiting for 30 minutes at that temperature, to make sure that the camera is heated evenly inside.

What follows is a photo material that can visualize the results. They are obvious and notable, but I did not intend them to be in exact numbers. This will probably be the subject of a future analysis, for the moment what I see is more than enough to prove my thesis. In order to have good consistency, all images, that I compare with one another and quote here, are treated in a same RAW->JPG convertor with the same settings, such as exposure corrections and so on.

In order to avoid possible problems such as light entering around the lens cap, every tested camera was with its front cap on, at maximum closed aperture and before taking the photo was placed in well closed 100% opaque box.

At approximately-5 ° body temperature, exposure 30 sec./ISO 100/~ + 4stops

At approximately + 40 degrees, 30 seconds exponation/ISO 100/+ 4stops


At room temperature of 20 degrees Celsius.
Fully tempered camera, which was not switched on at least one hour.

Exposure 30 sec,/ISO/100 + 4stops

The quoted image was shot immediately after the camera was working almost 50 minutes in live view mode.

Exposure 30 sec,/ISO/100 + 4stops


And a little more on this subject, I did four consecutive photos with Pentax K-3II at equal conditions ISO 100, closed front caps and in a dark. The only difference is that the exposure times are respectively 1/1 100sek sec, 15 sec, 30 sec. The camera is at the room temperature of about 20 degrees Celsius and has not worked at least one hour before the shot. Adjustment of the shutter stops 5 + (the same for all frames).





In the last shot you can clearly see how exactly the sensor starts to heat up, more at the edges and in the middle...

Up till now everything was connected with a single camera Pentax K3-II. Is it possible that all that is not valid for the other cameras?

Hardly. Physics is the same for all, at least for now, but the good practice requires a check.

I asked a friend of mine (Sabin Kolarov) with Fuji X-T1 to do the same tests, as the one above. The tests were made at the same exposure time and other conditions as those made with the Pentax K3-II. We did not made the whole range of tests at all temperatures (-5C….+40C), but we focused on two tests. The one at room temperature and that in the room temperature but after 50 minutes of camera working time.

The results was treated by photograph Sabin Kolarov and for a better visualization they are graphically presented after an analysis with RAW Digger over the RAW files.

Here is the result after 50 minutes working time, in the same other conditions. I have to note that the outside body temperature rises with 6 degree Celsius. Meanwhile, the camera was not held by hand but stood on a table.



Of course, we did a series of tests with a duration of exposure of 1/100, 1 sec., 15sek., up to 30 sec., as each of the exposure is accordingly: at the base ISO, that of ISO 400 and ISO1250. I am not going to bore the audience with quotes of the results, in short, there is a direct connection between the temperature of the sensor and the appearing noise. What is more important, we find that there is a clear connection between the working duration of the camera and the final result as a parasitic noise in the image at the same temperature of the surrounding environment.

That conclusion rises many questions. It is serious reason to put under suspicion all published tests regarding the dynamic range and noise, in which the temperature of the camera and its working duration before the shot, was not taken into account and quoted.

An even bigger question is that about the comparison between the different types of construction. It is natural that one DSLR even in a long hour working regime, if it was not in live view mode, will work in a much better conditions, in comparison of the equivalent of mirrorless camera that in the meantime has heated and its electronics has worked without interruption those same hours. In that case, I ask myself the question where exactly we must measure the dynamic range of the one and of the other camera? And which of the results will be the correct one? Because as you can see, only due to one hour working timeout, the results of the same single camera can be reduced several generations back...

This article, I hope, could serve to my colleagues for specific conclusions.

Knowing the problem, every one of my colleagues can do the comparisons I quoted above on his own and specifically for the equipment he/she is working with.

The conceptual basis of this analysis are the principals by which are made the astro shots in the big telescopes, where the sensor is obligatory cooled to a very low temperatures, which brought me to the idea of the same problems into the photo equipment.

I am sure that the sceptics who read the article up to here are already smiling and asking themselves, well, has no one until now thought about the problem and how the video cameras work then? In them the sensor is permanently activated and the same is the electronics. This is a serious argument and I made a research on that question myself as I consulted a colleagues from an official video technical service of one of the world’s leading video companies. I received a very clear respond from there – yes, the problem is well known from long time and it is very serious in the video cameras. To solve the problem in all serious video camcorders, the sensor MUST BE COOLED VERY INTENSIVELY. They even send me an image showing, how the radiator on which the camera must be mounted, looks like, for a certain model of professional video cameras. Here it is:



Probably similar solutions can be applied to the photo cameras, but considering the size of the sensors, it will be impossible, because it will cause a general increase in the camera size, energy consumption, etc. In this connection, the advantage of the DSLRs for some specific tasks is obvious (when they are not in live view mode) and the owners of mirrorless cameras must consider some strategies when making photos requiring long time working of the sensor. This, of course, concerns mostly for those cameras that have bigger sensor. It is of great importance what kind is the body of the camera, from what material, what size, and also whether it could and how well to dissipate the heat.

At the time of the survey I and my colleague noticed some signs, that suggest us that there are some manufacturers who adjust noise reduction according to the sensor temperature. This article does not aim to analyze this phenomenon, but surely in the future we will pay attention on that problem, all the more it was visible in the RAW files – something that cannot be easily avoided.

And after all, are the damages, from the published in that article effects, on the photos, so serious?

It depends a lot on the specific situation but even at 1/100 sec. are completely capable, in my opinion, to ruin one otherwise successful shot. And as a whole, the added noise and the reduced dynamic according to the temperature conditions, are more then what have been achieved from generations of technology development.

(Sorry for my English, it is not my mother language)

VPetrov-photography.com

[LensBeginner]

Holy... I was actually expecting something of that kind to happen, but the amount of variation is impressive (even taking into account the +4 or +5 push)! :-|

[Ventzy]

+4 and +5 push is to easy show visually difference. And in reality 4 stops comparing with full dynamic range of camera are not to much. If you can see difference when picture is pushed to +4 stops what about 14 stops dynamic range...? And big question, what about all test of cameras? In which temperature are they made it... And how much time before test camera was work? Because difference is very, very much...

[quh86]

wow! a truly interesting read.
Your testing gives rise to many questions.

for one, if a cooler sensor is always better than a warmer sensor why is there noise showing in the test image at -5 C° ?

the first two 30 sec exposure frames show comparably little noise when looking at the next testing you did. Does the push from 4 stops to 5 stops really increase noise to this level?
(i find it hard to compare frames here as they are not numbered) in the picture of 30sec exposure with 50min life-view action is so much less noise than when simply taking 4 pictures consecutively.

does this mean i should or i should not take 2 consecutive shots of the same scene at the same settings (given a long exposure time, perhaps shooting a night scene of some sort) to average out noise in post?

sorry for the dumb question, but what are the graphs exactly showing, could you give info about the x and y axis? (i might have missed this info, im not fully awake yet -.-")

cheers,

Sebastian

[Bewatek]

Very interesting article, thank you!

This is an important subject in astrophotography where exposures can be several hundred or even thousand seconds long.
Stargazerslounge.com for example has several discussions about this topic.

I made very similar observations like yours with my cooled canon 1100D sensor (watercooled peltier cooling to approx. 20 degrees C below ambient during 600s exposures) but i am puzzled by the high amount of red channel noise in the first -5C exposure.
Noise reduction in the red channel is what all the sophisticated cooling is for. I guess the sensor heated up very fast during the 30seconds exposure without active cooling but why is there lower red channel noise in the following 20C exposure?
Maybe Sony and Canon sensors behave different in this regard?

The rule of thump says noise doubles every temperature rise of 7C and the sensor heats up to 35C above ambient during long exposures. But this is from years ago and applies to canon sensors which were used almost exclusively in the DIY astrophotography community.

[Ventzy]

wow! a truly interesting read.
Your testing gives rise to many questions.

for one, if a cooler sensor is always better than a warmer sensor why is there noise showing in the test image at -5 C° ?

the first two 30 sec exposure frames show comparably little noise when looking at the next testing you did. Does the push from 4 stops to 5 stops really increase noise to this level?
(i find it hard to compare frames here as they are not numbered) in the picture of 30sec exposure with 50min life-view action is so much less noise than when simply taking 4 pictures consecutively.

does this mean i should or i should not take 2 consecutive shots of the same scene at the same settings (given a long exposure time, perhaps shooting a night scene of some sort) to average out noise in post?

sorry for the dumb question, but what are the graphs exactly showing, could you give info about the x and y axis? (i might have missed this info, im not fully awake yet -.-")

cheers,

Sebastian

Thank you about your questions, it is very important to me, and hope other peoples will join too.

My purpose is to point to this problem. I thing every one is possible to made such comparing for it own camera and temperature and working conditions. And hope of course to see somebody to invest much more efforts in details...

Why -5 degree visually is more noisy I don't know. Yes I see this, but post as is... somebody will find why hope. And of course both test as I write are not 100% comparable, because RAW development are not in one time... I'am not sure about how exactly work RAW converter.

My point in the tests was to compare results in concrete situations. It mean, that photos taken to compare are ABSOLUTELY equally shot + ABSOLUTELY equally RAW developed. But this is not try for all of the test together. They are not comparable each test to other. Yes, will be very good if all the test are made in same manner and after that noise is measured not visually but in RAW digger with data. But is very time consuming and a lot of work. And other very important question is that I suppose that some manufacturers have temperature noise corrections.

[LensBeginner]

*snip*

sorry for the dumb question, but what are the graphs exactly showing, could you give info about the x and y axis? (i might have missed this info, im not fully awake yet -.-")

cheers,

Sebastian

RAWDigger graphs should be
x-axis = channel luminance
y-axis = no. of pixel
IOW, your average multi-channel histogram (4 in that case because there are two green channels since there is a double amount of green pixels compared to R and B)

[Ratcheteer]

although the camera sensor is a semiconductor in the absence of light it acts like a normal conductor, so when you cool the sensor to -5c your dropping its resistance so when you apply the carrier voltage (what ever the voltage is at iso 100) the red channel will start fluctuating because red requires less energy than green and blue to change state under long exposures

unfortunately when you heat a semiconductor beyond its normal operating temp the resistance drops even lower than cooling it to -5c

[bertwert]

This is really interesting!
Thanks for putting the time into it

[MikeD]

So let me see if I get this right
If the camera sensor is hotter due to ambient temperature or to extended use such as in long term exposures or or a lot of use in burst mode with no cool down period there will be more noise. Anybody correct me if I am wrong.
I assume this is true as I have seen similar showing more noise from a hot sensor, many times referring to a smaller body like the Sony A7 series (big sensor, small body more heat trapped).

[Ventzy]

I thing that problem with noise, is may be one of the main advantage of DSLR cameras...

[Ratcheteer]

So let me see if I get this right
If the camera sensor is hotter due to ambient temperature or to extended use such as in long term exposures or or a lot of use in burst mode with no cool down period there will be more noise.

correct

I've seen the same behavior on canon powershot A710 IS running the CHDK hack (allows saving to raw, bulb exposure and scripting)
canon 5d mark 2
and the first generation sony rx100 (in the later models a overheat warning pops up and wont let you shoot, mark 2 a7's have the same protection, if you manage to overheat it before the battery runs out)

[npc]

although the camera sensor is a semiconductor in the absence of light it acts like a normal conductor, so when you cool the sensor to -5c your dropping its resistance so when you apply the carrier voltage (what ever the voltage is at iso 100) the red channel will start fluctuating because red requires less energy than green and blue to change state under long exposures

unfortunately when you heat a semiconductor beyond its normal operating temp the resistance drops even lower than cooling it to -5c

When you refer to "red channel" does it mean that in the existing/current sensors the individual pixes are wired in a way that makes it possible to amplify them differently based on what color from the bayer filter happens to be on top of them ?
Or do you mean that the values for reds are boosted after conversion to digital data and during image processing?

I thought current sensors are basically black and white if the bayer filter is removed (or if no light passes though it like when the cap is on during exposure like in this test)...

[clackers]

Worse in a mirrorless because the CPU is always going to update the EVF or LCD. Video can be very problematic, hence external recorders.

[biz-engineer]

Yes, very good thread, and yes well known thermal effect on sensor noise. That's a downside of mirrorless and sensors mounted on a floating plate typically used by camera having in camera image stabilization , versus sensor fixed on on a heat sink with a thermal bridge to camera chassis.