22
\$\begingroup\$

I have heard many times that to minimize noise, the highest ISO you want to use is ISO 1600, and that no matter what camera or sensor you use, once you have passed this (ISO 1600) threshold, the noise grows exponentially.

Some say this is math, but some say this has to do with the way the signal is processed, like how applying a huge gain in audio equipment will cause distortion.

Basically they claim that ISO 1600 is mathematically the most "economic" sensitivity for reducing noise.

Why is this?

  • What makes ISO settings higher than 1600 so bad?
  • Does the camera apply some exponential algorithm once you have passed ISO 1600?
  • Does the signal to noise ratio dramatically change once you have passed ISO 1600?

Edit:

I know ISO performance is getting better, sorry but this question is not about that. I had a 2.0 Megapixel camera, and ISO 200 was rubbish, so I know how much things have improved, and they will continue to improve.

After digesting the lengthy answers, it seems that ISO 1600 is the limit of analogue signal amplification, at least on some models. Beyond ISO 1600, depending on the brand, different digital algorithms are used to enhance the signal (a digital push). Since this digital push acts on Photon Shoot Noise and Read Noise, because this is after ADC, the overall resulting noise level is high.

This is what I understood collectively from the answers, I am still a bit confused if I understood it correctly, so please correct me.

I am still wondering, how do I know my camera's limit of it's analogue amplification? If I want to know when does a Canon/Nikon digital push occur, is such information available from their websites? Or is it a secret?

\$\endgroup\$
8
  • 5
    \$\begingroup\$ Could you site some examples of "they" making these claims? \$\endgroup\$
    – mattdm
    Nov 12, 2012 at 3:57
  • \$\begingroup\$ Well, one "they" is probably this aging forum thread: Idiot and Numpty ISO. The claims made by Browning were confirmed by some other individuals, the 1/3rd stop push/pull and loss of DR can be easily demonstrated with just about any Canon camera, and sample images of noise for ISO's above 1600 on Canon cameras clearly show different characteristics than at or below ISO 1600. I am not sure if there are any other "they"s making such claims. \$\endgroup\$
    – jrista
    Nov 12, 2012 at 5:58
  • 1
    \$\begingroup\$ The Canon EOS D60 didn't even support ISO 1600, and was top of the line ten years ago (released in early 2002). \$\endgroup\$
    – user
    Nov 12, 2012 at 9:25
  • 1
    \$\begingroup\$ ISO 1600 being somehow special is just nonsense. Different ISO values result in different levels of noise, and the tradeoff varies from camera to camera. With a Nikon D3s I find 1600 the "normal" ISO for walking around, but that's for that camera and my tradeoffs. See embedinc.com/d3s for details of measurements I took with my camera. Most other cameras will be worse, some a lot worse. \$\endgroup\$ Nov 12, 2012 at 22:43
  • \$\begingroup\$ @mattdm I would love to but I have lost all reference. Having read the various answers, I now understand what exactly I was trying to describe and I have updated my question. \$\endgroup\$
    – Gapton
    Nov 13, 2012 at 7:49

5 Answers 5

39
\$\begingroup\$

Nothing Special

There is nothing particularly special about ISO 1600, although in some cases ISO settings beyond 1600 have incurred less effective and efficient ways of amplifying the image signal. When you set ISO on a camera, that is simply instructing the camera to change the maximum saturation point of the sensor, from which the signal will be amplified. Increasing ISO will usually instruct the camera to amplify the signal before downstream electronics add more electronic and quantization noise, so using a higher in-camera ISO is usually (with caveats) better than using a low ISO and boosting exposure in post when you can't get enough light down the lens. There is no special math behind ISO 1600 specifically being the "best" high ISO setting to minimize noise in all cases, however there may be brand-specific amplification mechanics that affect the quality of noise at high ISO in some cases. Noise is a function of two key factors: electronic noise present in the circuit of the sensor & the lesser contributor of noise, and photon shot noise, sometimes called Gaussian noise, which is the primary contributor of noise. How noise exhibits (noise quality) is a factor of the method(s) of amplification.

Noise

Noise is an attribute of the image signal, and ultimately has to do with how the sensor is designed, the nature of its fabrication, signal to noise ratio (S/N), signal gain, ADC efficiency, and a number of other factors. These factors differ from brand to brand, model to model, and in some ways even from one camera to the next of the same model. Newer sensors generally tend to exhibit less noise in general than older sensors, regardless of pixel size or ISO setting used. The highest usable ISO setting can differ from two samples of the same camera (which is often touted as the case with Canon 7D DSLRs) and from cameras of differing generations (such as the Canon 400D and 650D).

Electronic noise, which can have a variety of forms such as Fixed Pattern Noise (FPN), Horizontal and Vertical Banding Noise (HVBN), is often unnaturally patterned and thus very undesirable, but only affects the lowest levels of the image signal (i.e. the deep shadows). As the ISO setting is increased, electronic noise exhibits less and less, and is ultimately entirely overpowered by photon noise.

Photon noise is the result of the random nature of light, which follows a Poisson distribution at the sensor. That means photons randomly strike the sensor, but at a high enough signal level their distribution is roughly even, and thus photon noise exhibits in the same way at any location on the sensor. Photo noise accounts for the very vast majority of noise in a digital photograph, orders of magnitude more than electronic noise (with the exception of improperly exposed photos where the signal is only a tiny fraction of the full well capacity, or FWC.)

There are several factors that affect how significantly photon noise will exhibit in a photo. Historically, the larger the pixel, the less this type of noise would exhibit. A photodiode is sensitive to light over area...photon penetration depth into the silicon of the photodiode is not a factor of pixel saturation. Larger pixels usually allow the use of higher ISO settings, as they capture more photons per unit time. More photons per unit time means more photons overall for any given exposure, which increases S/N. A higher S/N allows the use of a higher gain (more on this in a moment) when converting the analog sensor signal into a digital signal (RAW file), which is what your ultimately use to generate JPEGs and TIFF images with post processing software.

Gain

Gain is the conversion ratio of electrons (e-) to digital units (DU). A camera that converts exactly one e- to one DU has "unity gain". Most cameras achieve unity gain at some exact (but possibly non-selectable) ISO setting. More frequently, gain is fractional, such as 5.7 e- to every DU. For every stop increase in ISO, gain drops by the same factor. If you have a gain of 5.7 e-/DU at ISO 100, you would have 2.85 e-/DU at ISO 200, 1.425 e-/DU at ISO 400, .7125 e-/DU at ISO 800, and 0.35625 e-/DU at ISO 1600. As you increase ISO, you lose signal to noise ratio (S/N). A lower S/N is never really a good thing...it always means more noise due to a lesser signal being amplified. A lesser signal means less color fidelity along with less detail. The point at which those facts cause an image to degrade ultimately depends on the mechanism of amplification, which can differ from generation to generation and brand to brand.

Newer sensor technology has been increasing full well capacity generation over generation, even as pixel area shrinks by employing more effective means of directing photons to a light-sensitive surface rather than light absorbing surfaces and components. The relatively recent introduction of microlenses to CMOS sensors has helped direct photons onto the sensitive surface of the photodiode, and away from readout wiring and other non-sensitive surfaces. Lightpipe technology uses high refractive index material below specially tuned microlenses to help guide light through the channel of readout wiring above the photodiode, such that more of it reaches the diode instead of reflecting off the wiring. Backside illuminated sensors simply flip the whole structure, exposing the photodiode directly to light, eliminating the need for all of the other structures. All of these things improve "quantum efficiency" (Q.E.), or the overall conversion rate of photons to electrons in the sensor. Sensors with a higher Q.E. support higher maximum signal to noise ratios, which in turn support higher ISO 100 gain...which further supports higher gain at lower ISO. Higher gain, more electrons per digital unit, which lessens the effects of photon noise at every ISO setting.


Mechanisms of Amplification

Now onto the reasons why ISO 1600 (in the past) was the highest "useful" ISO setting in many cases. Some brands, such as Canon and in some cases Nikon, use more than one mechanism to amplify the signal coming off the sensor. In the last several years, ISO 1600 was usually the last "natively amplified" ISO setting, beyond which additional amplifiers or even digital amplification was used to achieve the next ISO setting. Canon is probably the worst offender to use alternative amplification mechanisms. In their last generation of cameras (7D, 5D II, 1D/s III, and all of the Rebel series up to the 650D) used an additional analog gain on the pixel stream...after pixel read but before ADC (analog-to-digital conversion), to achieve ISO settings above ISO 1600. Nikon used something similar in sensors they manufactured themselves (any Nikon cameras, and for that matter any other brands, that use Sony Exmor sensors use a radically different approach to signal management in general, so they don't apply here.)

Achieving ISO 3200 in the past would employ standard per-pixel analog gain at the time of pixel readout for all full-stop ISO settings up through ISO 1600, then an additional analog gain to the pixel stream coming off the sensor. In some cameras, ISO 6400 would use the same alternative analog gain post-read. ISO settings above ISO 6400 usually used a metadata digital gain hint to instruct post processing tools to apply an additional digital gain to achieve higher ISO settings. Such settings were usually called "Expanded" or "High" ISO settings, and could only be used in full stop increments above the cameras "native" ISO setting. (Note: The only reason one would ever really need to use an expanded ISO setting in-camera would be if they absolutely required a higher shutter speed than could be achieved with a lower native ISO setting. In many cases, still choosing the lower ISO setting at the required shutter speed, which would under expose, could still be desirable as manually fixing the exposure in post will usually produce a better result than in-camera expanded ISO.)

The information above can be applied to cameras from the last couple generations, not including the current generation. Older cameras from Canon definitely employ native analog gain as well as additional post-read analog gain, as well as a potential hard-coded (i.e. non-configurable) +/- 1/3 stop push or pull executed by the camera via behind-the-scenes exposure adjustments. This push/pull tended to cost you a loss of about 1/3rd stop of dynamic range. Past generation Nikon cameras that used sensors designed by Nikon also employed similar high ISO gain, although they seem to use analog gain for all ISO settings (including third-stops), which usually resulted in better IQ at high ISO third stop settings relative to Canon, without any DR loss. The current generation of cameras from Canon seem to employ a better analog gain approach to ISO settings up to the new maximums (25600 in the case of their non-1D lines, and 51200 in the case of the 1D X), with the additional post-read gain only being employed for the highest ISO setting (at least, as seems to be indicated by testing of these cameras so far).

That would mean that the highest usable ISO for Canon cameras has jumped from ISO 1600 to ISO 12800 at least, and possibly even ISO 25600 for the 1D X. Sony Exmor sensors, which are now used by the majority of Canon's competition including Nikon, employ a very different kind of sensor design and processing architecture. Exmor sensors top out at a native ISO 12800, and all further ISO settings are expanded mode ISO. Up through ISO 12800, Sony Exmor sensors perform quite well, on par with the 5D III and 1D X. Beyond that, noise quality starts to fall apart pretty quickly, and doesn't generally hold up to Canon ISOs 16000, 20000, 25600, 32000, 40000, and 51200. On the flip side, Sony Exmor has barely any read noise, and performs considerably better at ISO 100, 200, and to some degree even 400 in terms of dynamic range. Exmor DR has quickly become a thing of legend, and the sensor really shines for types of photography that eat dynamic range for breakfast (such as Landscapes).

Usable ISO 3200...6400...16000?

The constant stream of new and improved technology is constantly changing things. Barely four years ago, the Canon 450D and 40D could barely do ISO 800, with ISO 1600 being largely unusable. A generation later, ISO 1600 became more usable, and in the case of the 5D II and 1Ds III, ISO 3200 was even "usable" in some circumstances. Today, I regularly hear, particularly from sports photographers and photojournalists, that ISO up to 16000, 20000, and at times even 25600 on the 1D X is "entirely usable", "even printable with some post-processing work!" From a mathematic standpoint, electrons and digital units and gain and all that, I wouldn't necessarily say there is anything in particular that calls out ISO 1600 as the magic ISO number. The highest usable ISO has been increasing generation over generation, usually by around one stop, but recently with Canon's newest sensors, it has increased by as much as three, possibly even four stops.

\$\endgroup\$
17
  • \$\begingroup\$ I can attest that 1600 is very usable on 40D \$\endgroup\$ Nov 12, 2012 at 7:12
  • 2
    \$\begingroup\$ @MichaelNielsen - It is a personal preference Michael. I do not find 1600 acceptable on the 40D :) I only shoot it if it is a last resort to either capture the image or not. \$\endgroup\$
    – dpollitt
    Nov 12, 2012 at 14:32
  • 11
    \$\begingroup\$ TL;DR I just looked at the bold titles and guessed about what you wrote: "Nothing Special [since] Noise [and] Gain [are caused by] Mechanisms of Amplification [yet still] Usable [with] ISO 3200...6400...[and even] 16000 [on some cameras]." \$\endgroup\$
    – Xeoncross
    Nov 12, 2012 at 17:19
  • 1
    \$\begingroup\$ @MichaelNielsen: I also have a 450D, which is VERY similar technology to the 40D. I never found ISO 1600 usable, even when it was absolutely necessary. Dpollitt is correct that there is some subjectivity, but I think if you troll through the annals of the internet, you'll find that 6+ years ago, ISO 800 was the top dog, up to around 5 years ago ISO 1600 became top dog, and this year, ISO 6400 and 12800 have become solidly usable. Basically, once a major generation, ISO seems to improve by a stop, maybe two, such that the average photographer would find it "usable". \$\endgroup\$
    – jrista
    Nov 13, 2012 at 0:25
  • 2
    \$\begingroup\$ @jrista I think you missed my point: whether an image is "usable" depends entirely on the specific usage you have in mind, therefore you cannot say that a particular camera setting renders any potential image unusable. If you witnessed a crime and snapped the offender's license plate at ISO12800 would you say to the police "I got a picture, but despite the fact you can clearly read the license plate, I'm afraid I used ISO12800, which as we all know is unusable on anything but a 1DX so unfortunately you wont be able to make an arrest". \$\endgroup\$
    – Matt Grum
    Nov 15, 2012 at 10:10
15
\$\begingroup\$

1600 is not a magical number, but with today's current technology many DSLR's commonly produce poor results above ISO 1600. On the other hand, you could argue that many still produce fine results at 3200 and 6400 - it depends on the audience and what technology they are familiar with. If you are reading this on a forum with many users of consumer level DSLR's then 1600 is probably a safe maximum for the current technology. If you were among professionals using the top of the line equipment such as a Nikon D800 or a Canon 5DmkIII, they certainly will not stop at ISO 1600.

Take a look at some of the high ISO images from the Canon 5DmkIII here. ISO 12800 doesn't look bad at all to me. Compare that to ISO 800 on the Canon Rebel XT that I used to shoot in years past, and they are in the same ballpark. It is subjective, but from my perspective there is no way I would stop at ISO 1600 on the 5DmkIII!

To directly answer your question - nothing is special about ISO 1600. I would guess you are just reading many things that happen to mention that figure at this point in time. The figure will change over time, and it already is with professional series DSLRs.

\$\endgroup\$
1
  • \$\begingroup\$ Yes I know ISO is constantly improving. I am actually after the amplification mechanism. Sorry I didn't understand the signal processing well enough to clearly describe my question. Thanks for offering your help though! \$\endgroup\$
    – Gapton
    Nov 13, 2012 at 7:59
4
\$\begingroup\$

This is a bogus claim as far as I know. There is nothing constant about ISO 1600 other than it is ISO 1600. Technology changes constantly and ISO 1600 today is much better than five years ago.

ISO performance is also strongly correlated to sensor-size. Take a look at this comparison between recent 2X, 1.5X and Full-frame sensors from recent cameras. I've been shooting with a Nikon D4 for a month now and would not hesitate to use ISO 6400. On the Pentax K-5, I usually stop at ISO 1600 and am setting the K-5 IIs to automatically reach ISO 3200 but it is about one stop better.

On a different aspect of your question, there is no such thing as you want to use without particular circumstances. I have wanted to use very noisy settings to get a shot I would otherwise not be able to at all. These images can have value to me, despite that they would not make a great poster-size prints.

\$\endgroup\$
6
  • \$\begingroup\$ ISO performance is strongly correlated to price range, not sensor size. Even though I haven't tested them myself, I can guarantee You'll get best results from the $15K at high ISO than the $2k from this list here. And soon high end 4/3 will have similar results with full frame (in the same price range). Infact, a lot of reviews are comparing them, like yourself. \$\endgroup\$
    – BBking
    Nov 12, 2012 at 4:22
  • \$\begingroup\$ That depends how you compare. DxOMark, for example, normalizes to a standard print size and has $2K cameras beat $20K ones. If you see a high-ISO file from a Medium-Format back, it actually looks pretty bad despite a larger sensor and larger price tag. \$\endgroup\$
    – Itai
    Nov 12, 2012 at 4:39
  • \$\begingroup\$ The problem with theoretical vs practical performance is that all else is never equal. If it were, then I sensor-size would clearly dominate the performance simply because it has more light to work with. In other words, if you built sensors which are identical except for size, the larger one would have lower noise and higher dynamic-range simply because of physics. \$\endgroup\$
    – Itai
    Nov 12, 2012 at 4:42
  • \$\begingroup\$ Yes, a noisy photo is better than none. You always have the option to make it into B&W ;) . I have shoot for 5-6 years and I understand that. Thanks for the help, it was my fault that I couldn't describe my question very clearly. \$\endgroup\$
    – Gapton
    Nov 13, 2012 at 8:01
  • \$\begingroup\$ @Itai If it were, then I sensor-size would clearly dominate the performance simply because it has more light to work with. If by I you mean one, that's the point I'm trying to make. A FF only captures more light because it has more surface area to cover. All I'm saying is that you could build a cheap $1K FF camera and have poor high ISO performance. You said yourself If you see a high-ISO file from a Medium-Format back, it actually looks pretty bad which goes against your ISO performance is also strongly correlated to sensor-size argument. \$\endgroup\$
    – BBking
    Nov 13, 2012 at 22:47
3
\$\begingroup\$

ISO is a metaphor from film photography that is really controlled by the sensor gain before the analogue to digital conversion, until a certain point. This type of gain is good, because it only affects some parts of the noise, which look good and are not that bad usually. Then to achieve higher ISO they "cheat" and just do digital shift, i.e. for each bit they shift the "ISO" doubles, but this enhances ugly read noise, the same as if you underexpose the image and boost it on your computer. On the generation of cameras where we had in late 2000's the analogue gain was up to 1600 and then we could achieve 3200 through user setting "high" (3200) - but this is digital boost and you can see on sensorgen.info the direct effect this has on read noise and DR:

http://sensorgen.info/CanonEOS_40D.html

http://sensorgen.info/CanonEOS_5D.html

I know a guy who says that some cameras are even more complicated than that - using a mixture of analogue and digital gain before they switch to using only digital. Maybe that is what is going on on the nikon here, where 100-200 goes wrong:

http://sensorgen.info/NikonD3100.html

But on the new generation of cameras it seems ISO 3200 is the new "1600":

http://sensorgen.info/CanonEOS_5D_MkIII.html

\$\endgroup\$
2
  • \$\begingroup\$ The difference between analogue and digital gain is what I am after! Thanks! When you mentioned generations of cameras, I wonder, how do we find out our camera's analogue gain ceiling? Is this a piece of specification that is publicly available? What technological limitation has created such ceiling? What kind of advance in technology will push this ceiling higher? \$\endgroup\$
    – Gapton
    Nov 13, 2012 at 8:09
  • \$\begingroup\$ I think that they use digital gain to achieve some of the 1/3 stops between the mail ISOs (which is weird as gain is dB-discrete in ~0.3db, not 6db which is a stop, steps in my machine vision cameras), but when the relative noise curves turn around and the DR curve turns south in the sensorgen charts is when they maxed out the analogue gain. \$\endgroup\$ Nov 13, 2012 at 15:02
3
\$\begingroup\$

Using a maximum of ISO 1600 is a terrible rule, IMO. Yes, higher ISOs do have much more noticeable noise. However, depending upon the camera/sensor, the scene, and your intended use I bet 3200, 6400, and maybe even higher are perfectly acceptable. Sharing a photo on the web, for example, I'm sure you can have a fine result at a much higher ISO. Similarly, printing a small 4x6, I am confident you'll get an acceptable result. Printing a 16x20 is a different story, of course.

Whenever this is mentioned, the other thing I always think is that the person saying this obviously doesn't have much history with photography to have used film. Today's ISO 1600 is clearly better than ISO 400 film. ISO 400 film was certainly usable.

One last point: is it better to have a noisy photo shot at high ISO or no photo at all? I know which I would choose -- and have chosen, many times over.

\$\endgroup\$
1
  • \$\begingroup\$ I agree with everything you said and I have used ISO 6400 or higher before (because I simply must capture it, period). I understand all this, unfortunately this is not the answer I am searching for. \$\endgroup\$
    – Gapton
    Nov 13, 2012 at 8:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.