This is a very detailed explanation of ISO from Tony (Tannin) and goes somewhat beyond the scope of NTP - but is here as you learn.
This post also explains the extended (L and H) ISO settings that are available on some cameras.
Q: What's ISO? (Or ASA)
A: ISO is a measure of the sensitivity of the sensor to light.
Q: But I have lots of different ISO settings on my digital camera, yet I know there is only one sensor - how can one sensor have several different sensitivities?
A: The sensor itself does have a single, fixed sensitivity. Some people call this the "base ISO". However the sensor system includes some very sophisticated support electronics. You actually have three parts to the sensor system. (At least three parts that are relevant to our discussion here: these are (i) the sensor itself, (ii) the amplifier, and (iii) the analogue to digital converter.
The sensor gathers light. Each element of the sensor collects photons while the shutter is open, then, when the shutter closes, sends a "summary report" off to the amplifier. This "report" is a voltage. The more photons the sensor element collected, the higher the output voltage.
Q: But even I know that a variable voltage isn't a digital signal - it's analogue! I thought we were talking about digital cameras!
A: We are. The digital part comes later. But first I better cover the way the sensor works in a bit more detail.
Q: Oh. OK.
A: You can think of each sensor element as being like a bucket to catch photons. As we already learned a moment ago, after each exposure, the sensor element reports how full it is by sending a voltage to the amplifier. Almost full (lots of photons), it sends a high voltage. Almost empty (not many photons) it sends a low voltage. In between, it sends a middling voltage.
Q: And if it doesn't collect any photons, it sends a zero voltage, right?
A: Correct! Well, almost correct - there is also a noise factor, which I'll talk about in a minute.
Q: But what happens if the bucket is filled to overflowing?
A: It just sends the maximum voltage, which corresponds to "bucket full, lots of photons!" This is why every photographer hates blown highlights - there is no detail in (e.g.,) the bride's white wedding dress because every pixel bucket covering that area is either full or overflowing, and the camera has no way to tell the difference.
Q: You were going to tell me about noise. Where does noise fit in?
A: All electronic circuits produce some noise, which you can think of as random output variations which cannot be explained by the input variations. The best-designed and most difficult to manufacture circuits produce less noise than cheap, less well designed circuits, but no circuit is noise free. There will always be some inaccuracy in any circuit, and it can only be reduced so far.
As it happens, the amount of noise a sensor element produces (the random component to the output) is pretty much fixed and unchangeable - there is always a certain amount, and this amount is much the same in any number of different sensor designs. (You could easily make one with more noise, but nobody does these days, at least not in DSLR-standard equipment.
But the signal generated by the sensor element is not the same. Bigger pixel buckets produce a bigger signal.
Q: So the noise is always the same, but the signal varies.
Q: And bigger pixel elements have a bigger signal, so in proportion they have less noise?
A: Exactly. This is why the cameras with the biggest pixels have the lowest noise. Try shooting in really bad light with a very high ISO on a Canon 450D and then a Canon 1D III - the ID III is miles better. Similarly, compare a Nikon D200 with a Nikon D3 - again, the camera with the big pixels can still deliver a clear, more-or-less noise-free picture long after the the small-pixel camera starts putting out pretty ordinary results.
Q: But don't smaller sensor elements provide more detail?
A: Yes. There is a trade-off. The smaller the pixels (up to a certain point) the more detail you get, but you pay for that with higher noise. Choose your poison.
Q: I thought you were talking about ISO. Can we get back to ISO? You might start by telling me how a camera can have several different ISO settings - 100, 200, 400, 800, and so on.
A: Now we get to the second part of the sensor system, the amplifier. The signal generated by each sensor element is very weak and it needs to be amplified to make it usable by the rest of the system. At the camera's base ISO], let's say ISO 100, it is amplified by a certain fixed amount. At each higher ISO, it is amplified by a greater amount, for example, twice as much for ISO 200, and four times as much for ISO 400.
Q: But doesn't that amplify the noise as well?
A: Yes. That's why picture quality at ISO 1600 is not as good as it is at ISO 200.
Q: So all the different ISO settings are just different amounts of amplifier gain?
A: No! Just the "main" settings. The amplifier only has a certain number of possible gain settings. On a camera with a base ISO of 100, these might correspond to 200, 400, 800, and 1600.
The other ISO settings are handled later on by a different part of the camera. To understand that, we had better have a think about the third main part of the sensor system, the analogue / digital converter.
Q: OK. Tell me about the A/D (analogue to digital) converter.
A: This part is easy: the A/D converter simply take a voltage as its input, and outputs a number. The higher the voltage, the higher the number. The smallest number it can output is (of course) zero, and the maximum size is fixed by the design of the A/D converter. Most current cameras have a 12-bit A/D converter (i.e., can go from 0 to 4095); some newer models have 14-bit converters which go to 16383, and (if I remember correctly) some Pentax models have 16-bit converters which go up to 65535 - though no-one knows why as they don't seem to actually do anything with the extra information!
But for simplicity let's imagine we have an old camera with an 8-bit converter, which outputs numbers between zero and 255. If you have a full photon well in the sensor element, the A/D converter will output the number 255, if it's half full, it will output about 128, and so on.
Now we get to the other ISO settings. If the camera has the ability to do "ISO expansion", then the camera electronics do some sums with the output of the A/D converter.
Q: So the "native ISOs" are taken care of by the amplifier, but the "other ISOs" are not?
A: Correct! All the non-native ISO settings are simply mathematical trickery. For example, on my 20D, if I set ISO 3200, the sensor system simply carries right on shooting at ISO 1600 (because that is as high as the amplifier can go) but the camera electronics set one stop of negative exposure compensation and double whatever number the A/D converter puts out.
Q: So what is the difference between using "H" (ISO 3200) on a Canon 20D, and simply under-exposing by one stop at ISO 1600, and turning the brightness up later when you post-process?
A: None. No difference whatsoever.
Q: Doesn't deliberately under-exposing and then pushing it up a stop in PP add a lot of noise?
A: Yes. So does using "H". Exactly the same amount, in fact.
Q: So why does anybody do it?
A: Laziness, I guess. Hell, I used ISO 6400 on the Canon 1D mk III myself the other week, shooting indoors at night without flash.
Q: Were the results any good?
A: Well, "barely usable" would be a fair description - but then they wouldn't have been any better if I'd under-exposed and pushed, and in that dreadful light I couldn't have got anything at all if I hadn't done one or the other.
Q: Hey - I just thought of something. Your Canon 40D has lots of other ISO settings besides the "main" ones (100, 200, 400, etc.), such as ISO 320 and ISO 640. Where do they come into the picture?
A: They are a bit like the "H" and "L" ISO settings. They take the nearest "real" ISO setting and multiply or divide to get the desired setting. For example, ISO 320 might be a "real" ISO 400 divided by 1.25, and ISO 500 is also a "real" ISO 400 multiplied by 1.25. (Actually, I forget which ISO settings on the 40D are the "real" ones. I could look it up if you are interested. But in any case, the principal is the same.)
Q: One last thing I don't understand. You seem to be saying that the "real" ISO settings like 200 are good, and the "fake" ones like "H" and 320 are bad, or at least not so good. Why does it matter where the multiplication takes place? What is the difference between turning up the gain on the amplifier to increase the output, and simply multiplying the answer to do the same thing?
A: The amplifier, being analogue, and also closer to the source of things, is more accurate. It is multiplying an exact voltage. The camera electronics, when they simply double the number output by the A/D converter, are multiplying an approximation. They are also multiplying the total noise generated by the sensor system, including the A/D converter which, like all circuits, makes some noise of its own. In contrast, the amplifier is only multiplying the noise generated by the sensor element itself. So, although the A/D converter will still take the output of the amplifier and turn it into a number, and in doing so generate a certain amount of noise itself, because we are only taking the output number and not multiplying it afterwards, we are getting 1 * the A/D converter noise, not 2 * that noise.
Q: I'm starting to get the impression that the "non-native" ISO settings are kind of like digital zoom - they don't actually achieve anything you couldn't achieve just as easily and every bit as well in post-processing.
The information in the following schematic diagram explains what is going on inside your camera when you take a photograph and the difference between raw and JPEG images. Note where the ISO gain (amplification) occurs.
The following image shows the primary light paths in a DSLR camera. The above diagram shows what happens after the light is captured by the sensor.
Cross-section view of SLR system.
1 - Multi-element lens
2 - Reflex mirror
3 - Focal-plane shutter
4 - Sensor
5 - Matte focusing screen
6 - Condenser lens
7 - Pentaprism
8 - Eyepiece