New Workshop! Lighting 3 | Advanced Off Camera Flash Preorder

Tips & Tricks

Here’s What Your Camera Does When You Change ISO & Why Higher ISO Creates Noise

By Kishore Sawh on November 12th 2014


I can’t speak for anyone but myself here, but I would imagine I’m not alone when I say I’m always fascinated by, and eager to understand, the things we all use in everyday life and really don’t quite know how they function. A basic example would be how a car works. We use them all day every day and yet many people haven’t the foggiest idea of what’s actually going on within the innards of the car, the parts you can’t see, to make it go. There’s a lot that’s going on regarding the sequence of events that occur when the gas pedal is depressed.

Similarly, in photography, so many of the basics that we use day in day out, go misunderstood. Or, if not misunderstood, there’s no real thorough comprehension of how it works, other than how it affects our ultimate exposure. ISO, in today’s age, is one such facet. Sure, most know how to manipulate ISO and balance it with shutter speed, aperture etc, to end up with a desired exposure result, and we know that raising the ISO will generate noise in our image, but do we know why?


[REWIND: Your Work Can’t Speak For Itself | Do You Know What Else You Need To Be Hired?]

Ian Gibbons of Panvista Productions put out a video a while back which does a nice simplistic overview of how ISO really works behind the scenes, and thus, how it affects exposure, what the camera is thinking and doing when you raise or lower it, and why we get ‘noise’ from digital ISO changes. It’s quite fascinating stuff, and I’m not ashamed to admit I hadn’t known before that it was electrical interference that causes noise. It’s all a somewhat simplistic overview, but a good one regardless.

Source: PetaPixel

This site contains affiliate links to products. We may receive a commission for purchases made through these links, however, this does not impact accuracy or integrity of our content.
Terms: #ISO

A photographer and writer based in Miami, he can often be found at dog parks, and airports in London and Toronto. He is also a tremendous fan of flossing and the happiest guy around when the company’s good.

Q&A Discussions

Please or register to post a comment.

  1. Mark Sheppard

    I think he must have gotten tired of getting his socks beat off and pulled it.

    | |
  2. Manuel Lopez

    The video has been removed.

    | |
  3. Adam Sheridan

    Video’s gone

    | |
  4. Dave Haynie

    That’s lots of video to say very little about what’s actually going on.

    The pixel-site sensor he’s talking about is a photodiode. Photodiodes have a very broad spectrum of sensitivity, they don’t see color, thus the Bayer pattern filters we’re mostly come to expect (with a few variations, or weird experiments like Foveon). The photodiode transfers one electron for every photon that hits it… the transfer of electronics though an electronic circuit is what we call a current, of course. So more photons, more current.

    That current is stored in the sensor array after an exposure, perhaps as a voltage. On readout, the pixel data passes through a programmable-gain amplifier — that’s your ISO. If you’re a camcorder user, you’re setting the actual gain on that amplifier, which of course makes life interesting, because +20dB doesn’t give you the same result from camera to camera. The ISO scale normalizes the effect of gain, so that within the limits of design constraints, ISO 100 is the same EV value on every camera, regardless of the actual gain employed.

    So then there’s noise. There will always be noise, not because of the activity of the electronics alone, but because of heat. Heat excites electrons in wires and silicon…. not in any useful direction, just randomly. That’s what’s called the thermal noise floor, also called Johnson-Nyquist noise. Add to that the dark current in a photodiode… every photodiode has a characteristic very small current that will flow when in total darkness. And finally, some noise do to coupling of signals in the system; not just inductive coupling (as he suggests), but electric and RF coupling are all possible in a system. Control of this latter one is the main reason digital sensors have become quieter over the years.

    It’s also interesting to note that modern sensors do dark balancing for every shot. They basically take a picture in total darkness, which is a picture of the dark current of the sensor, and subtract that from the normal exposure. This can’t eliminate pure noise, which is random, but it does eliminate differences between each photodiode.

    So noise is always there, but you can’t always see it. The practical noise floor for any exposure is based on digital zero for the ADC that’s digitizing the shot. Let’s take a 14-bit sensor, and define ISO 100 as the variable amplifier set at 0dB, and define the peak within a frame as 0dB. So that means the dynamic range is 84dB, and so, the noise floor is -84dB. Lets put our noise peaks at 100dB. So at ISO 100, no noise ever gets into the ADC… there’s still a signal below digital zero, but it’s not important.

    So now, to double a signal, I amplify it by 6dB (technically 6.02dB, but close enough). So I dial in 6dB for ISO200, 12dB for ISO400, 16dB for ISO800, etc. This is just an example, and you’d find different values for different kinds of cameras. Here’s what’s important… each 6dB lowers that visible window, and thus, the effective noise floor, by 6dB. So in my model, that’s -90dB at ISO200, -96dB at ISO400, -102dB at ISO800… here, we’ve just crossed the noise threshold. So at ISO800, the loudest noise, at -100dB, is now 2dB into the signal. That’s not enough to affect a whole bit yet, but it’s enough to occasionally flip a xxxxxxxxxxxx01 to a xxxxxxxxxxxx10 (or 00 to 01, etc), as it’ll be added to the analog value of that pixel. So that’s a very small effect on the image, but it’s the first real effect of noise. And you can see that now dialing in ISO1600, I’ll have a noise floor of -108dB. So the noise is now able to add more than a bit’s worth of disturbance to the captured image… basically, the signal we’re processing is low enough to now be merging with the noise, rather than rising above it.

    | |
    • Mark Sheppard

      Nicely put Dave.

      I think we have successfully cleared it up… The guy in the video doesn’t know what he is talking about and wasted somewhere between zero and 9 minutes, 33 seconds of each of our lives. ;)

      | |
  5. Ralph Hightower

    I am knowledgeable about the exposure triangle of ISO, Aperture, and Shutter Speed having used film since 1980 and reading photography books from that era. I have shot film from ISO 25 (Kodak Kodachrome 25) up through ISO 3200 (Kodak TMax 3200, Ilford Delta 3200) and pushed Kodak TMax 3200 two stops to ISO 12800. In pushing Kodak TMax 3200 up to 12800, the film grain absolutely exploded; the grain and images were noisy, but it was what was necessary for the rock concert. B&W and color negative film have a fairly wide latitude of exposure whereas slide film is less forgiving about exposure.

    When he introduced magnetism and cross-wires, the diagram he drew looks like the core-memory used in computers in the early days.

    | |
  6. Blake Weber

    That video was PAINFUL.

    I couldn’t even watch it. I skipped ahead several times to get the gist. DUDE… you are not that interesting to listen to so that we will sit through a ten minute explanation of a simple concept. AHHHGG. And there is a reason people use LARGE white boards when they are going to do live drawings on video… it doesn’t make your hand look gigantic.

    That video could have been 60 seconds long with the same information and MUCH more effective.

    | |
  7. Mark Sheppard

    DOY! Again? 9:32 to say that noise is false color created by the wiring in the camera creating magnetic fields that cause interference with data transmission? My understanding of noise does not agree with his explanation, although there may be some truth to what he said.

    I believe that the primary cause of noise is false readings of the sensor due to it operating at a higher sensitivity via increased voltage.

    | |
    • Joseph Wu

      I agree. Partially true, but that specific? I’ll believe it, when I see some data to back that statement

      Increasing sensitivity ( gain ) of the sensor = picking up more data. Naturally any electronic will have noise, the higher the sensitivity, the more of that noise/interference is picked up.

      Most of the money spent on higher end electronics, or sensors in this case, is the filtering at the hardware level ( better routing, more intricate circuits, firmware ingenuity ).

      Exactly why some people spend hundreds of thousands on audio gear, so they can have the purist listening experience.

      | |