• Please use real names.

    Greetings to all who have registered to OPF and those guests taking a look around. Please use real names. Registrations with fictitious names will not be processed. REAL NAMES ONLY will be processed

    Firstname Lastname

    Register

    We are a courteous and supportive community. No need to hide behind an alia. If you have a genuine need for privacy/secrecy then let me know!
  • Welcome to the new site. Here's a thread about the update where you can post your feedback, ask questions or spot those nasty bugs!

The Canon EOS AF system - interesting observations

Doug Kerr

Well-known member
The Canon EOS autofocus (AF) system is a work of great wonderment and mystery. Canon tells us a lot about certain of its aspects - the nature of the different kinds of AF detectors, for example - but when it comes to its basic principles, they studiously decline to enlighten us. It is almost as if we knew all about different fuel injection systems for Diesel engines but we are still not sure what actually ignites the fuel-air charge, and if the cycle is two-stroke or four-stroke.

Over the years, many enthusiasts have speculated on the basic principles, resulting in the presentation of various models. The one that, at least until recently, I have favored, I will describe here (of course there are numerous details and wrinkles I leave out; the purpose here is to distinguish American football from Rugby).

The model is for the one-shot AF mode. It assumes a preselected AF point. It assumes that a focus error measurement can be made at the current setting of the lens; that is, neither AF assist (if available) nor focus search will be needed.

MY MODEL

The big plan

The overall scheme is that of a feedback control system (often called a "closed loop" control system). The thing being controlled is the position of the focusing mechanism of the lens (from here on, for conciseness, I will just say "the position of the lens"). The feedback is by observation of a quantity called the adjusted focus error, which is developed from a measurement made by the active autofocus detector. The system moves the lens focusing mechanism (from here on I will say just "moves the lens") until the adjusted focusing error is within a small band about zero (bringing it to exactly zero is no practical).

Sounds simple. The devil is of course in the details

The scenario

1. With the camera regarding the focus target, the user presses the shutter release button.

2. The adjusted focus error is determined.

2a. If it is within a certain small band about zero, the camera beeps to indicate the attainment of focus confirmation, the shutter is allowed to trip if the shutter release button is at full press, and the AF job is done.

2b. If not, proceed to step 3.​

3. The system calculates the direction and amount of change in the lens position that should bring the adjusted focus error to nearly zero, in terms of lens movement units. (The lens has provided various information needed for this, including how big its movement unit is.)

[It should be noted that, given the details of how this is done, the calculated distance is in general just an approximation of the precise distance needed.]

4. The lens is told to move by that amount.

5. When it is done moving, go to step 2.

RECENT OBSERVATIONS

Recently, my colleague "Wilba" reported to me some interesting tests he had devised, to the end of gaining further insight into the EOS AF system, and some interesting results. I performed similar tests here, and confirmed his observations.

The changing target test

In one test, the camera is set up to observe a target at a distance of, for example, 5 m, but another target on a card is held at a distance from the camera of perhaps 1 m.

I preset the lens to infinity focus. Thus, it will need to move most of its stroke to arrive at the proper focus distance for the near target.

The shutter release button is pressed, and as soon as the lens begins to move (presumably step 4 of the model has arrived), I drop the near target.

The observed result, if my timing is appropriate, is that the lens, before it has completed its movement to proper focus for the near target, reverses its motion and proceeds to proper focus for the far target.

What does this mean?

One conjecture was that the system, once having dispatched the lens on a "vector" that hopefully leads to the desired destination, continues to make measurements while the lens is moving, and uses them to update the lens' travel instructions (with the lens still moving). The thought was that perhaps this would overcome the recognized "approximate" nature of the lens' initial travel instructions, in the hope that it could "reach the hole in one stroke".

But other testing suggested that this was not so.

In one test, also devised by Wilba, the test range was dark, so that the system could not make an initial focus error measurement. I had disabled focus search. Thus the camera waits patiently hoping that the sun will rise.

The target was at a close distance, and I preset the lens to infinity focus, so the needed distance of lens movement would be substantial.​

I then manually fired a flash unit. The system made the initial focus error determination and sent the lens on its way. But "enroute" measurements were impossible the range was dark once the lens got underway. When the movement was complete, the system tried to make a new "standing still" measurement, but of course that didn't work either, so again the camera waited patiently.

But from this we could determine the distance of the lens' initial travel order. I tried this on several lenses, with several combinations of focal length and initial focus setting, until I found a situation in which the initial movement was considerably "short of the hole". I repeated the test to get a good value for the initial movement distance.

Then I put up a substantial illumination in the test range, and, again with the lens preset to infinity, did a regular AF task. Because of the modest illumination, when the lens finished its initial movement (falling quite short of "the hole), and the system took another measurement, that took a while. (It takes a while for the image to build up on the AF detector, just as for the actual image, and with modest illumination that time is considerable).

Thus I could see where the lens was at the end of the first movement. In every case, this was consistent with the first movement observed with the "flash stimulus" test.

This suggests to me that the system was in fact not updating the lens travel orders based on enroute measurements.

So then, how did the system know to change the lens' travel orders when I switched targets on it in the earlier test?

My conjecture (and I have no solid evidence for this) is that when each "enroute" measurement is made, the result is examined to see if it is wholly out of line with what would have been expected at that point. If not, nothing is done - there is really no need. The overall "closed loop" nature of the system will bring focus to the desired value whether the first shot reaches the green or not.

But if the result is dramatically unexpected ("anomalous"), the system realizes that something is going badly wrong. Perhaps the initial focus error measurement went badly. (It is quite possible that, for certain target patterns, a gravely wrong determination will be made, as we do with out eyes when we try and focus on a wall with a repetitive pattern of small holes.)

Wilba points out that many ingredients needed for this are probably already in place, need for the AI servo AF mode. Thus, he reasons, why not put them to work in overcoming an infrequent, but damaging, phenomenon.

I can't argue with that.

But we certainly don't know.

Well, that's all we know about that here for now.

Best regards,

Doug
 

Cem_Usakligil

Well-known member
A ha! The plot thickens. It is a very interesting read Doug. Thanks a lot for taking your precious time to write these nuggets of info using a very clear language which even I can follow.

Cheers,
 
Top