I expect I'm just ignorant, but the term "calibration" gets used a lot and I'd like someone to explain what it means to "calibrate" a lens (or a camera).
If the camera says to the lens "focus at 12 feet", then the lens would need to know precisely where "12 feet" occurs, and set itself to that position. If this is the way the system works, then I understand the terms.
But that is not how it should work, it seems to me, particularly in servo modes. The camera should constantly monitor image contrast as the lens moves in and/or out, issue instructions to the lens to move forwards or backwards until the camera detects maximum contrast, and then the camera should tell the lens to stop moving. In this scenario, where is the concept of calibration? In this scenario, there is no way for manual focus and auto focus to give different results unless the focus screens for manual and auto focusing sit at different distances from the lens.
But clearly, something ~does~ go wrong, or Nikon and Canon and presumably others wouldn't put capabilities to adjust auto focusing into their cameras. In the first scenario, how does the camera know the subject is at 12 feet? There is such a thing as radar distance measuring but I don't think it's in these cameras.
So, what gives? Color me confused and thanks for any insight.
PS: I suppose the details of how all of this works are proprietary, but I would find it interesting to learn just exactly how these cameras work in this and other regards.
The main thing about AF sensors is that they are "phase detectors". You have to understand the general workings of optics if the rest of the explanation is going to make any sense... This is a good starting point: http://doug.kerr.home.att.net/pumpkin/Split_Prism.pdf
The AF sensors takes light from opposite parts of the aperture area and guide them to two line-shaped sensors. Think of them as small digital cameras, but their resolution is only one pixel in height, and ~50pixels in width of the picture. They're very small slices of the picture. Now, just as in the example with the "split-prism" in the link above, the two pictures from the line-sensors should match up in sideways placement - then the image is in focus. If "line1" is apparently more to the left than "line2", the focus is behind the intended subject. If to the right, the focus is in front of the intended subject (or the other way around depending on which line is "1" and which is "2" - not important for the explanation).
Since the angle between the two points of measurement on the aperture is known, the AF sensor can via some clever math say "move distance X closer" or "X further away". The lens knows where it's focused distance is right now, get's a command to move "X" amount of distance from the camera, calculates how much it has to turn the internal parts of the focusing ring - does that and reports back to the camera - "all done". Or at least it SHOULD be... Now for the error sources (or a few of them at least...)
The AF-sensor in the camera can be off - the lines may be incorrectly aligned or "calibrated" - this results in either constant back-focus of front-focus. This is individual for every sensor point in the system, but generally they are ok as a unit. That all are off in a certain direction is more common that just some in one direction and others in the other.
The "current focus position" as measured by and in the lens can be off, then the calculation of the amount of focus movement (this is done by the lens) will be off. The movement follows a roughly 1/x curve depending on optical construction, so this error can result in to little corrective movement up close and to much when further away - or the other way around. The movement distance from near focus to infinity focus may be off - this will give the same effect. Infinity point ("zero" for the lens) may be off, this will also result in a nonlinear error.
There's also a correction for average light temperature in the AF-sensor, since the sensor is more sensitive to infrared light than to visible light, and lenses generally don't focus IR at the same distance as f.ex green.
Micro-adjustment works on the camera part of the error, and cannot adjust for the non-linear behavior that most of the error sources in the lens show.
The servo mode is a just the same, but a little bit different... :-) It's not a continuous movement, it's a string of small adjustments as the chain has to do the full turn for every correction.... Measurement, calculation, lens movement, lens confirmation. Then it starts all over again. When you're machine-gunning the camera the readouts can only be made when the mirror is down, once between every picture. A good explanation of AI-servo is here: http://www.nikon.com/about/technology/core/software/caf/index.htm
This was the shortest, most simplified version of AF workings that I've ever written I think... :-)
Not much to add to theSuede's excellent explanation!
Just a note that the latest Canon DSLRs have the option of using the contrast detect mode in Live View. If you try it out, you'll realize why it is not the primary focusing method (too slow). The same mode is used by P&S cameras and that is why they are so slow in focusing.
However, the contrast detection mode in Canon DSLRs is very accurate. Except with some Sigma lenses, which makes me think they might have some issues in implementing the EOS lens protocol fully.
