• 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 - a "new" wrinkle

Doug Kerr

Well-known member
The actual workings of the Canon EOS autofocus (AF) system remain rather mysterious. Although Canon has nicely described many aspects of it, some very fundamental things remain clouded (and not by accident, evidently).

One outlook that has been widely accepted (including by this office until just recently) as to the basic system concept is this one. (I will of course leave out numerous details.)

For compactness, I will speak of movement or position of the lens focus mechanism as "movement (or position) of the lens".

We will assume that the camera, when we half press, is not in proper focus on our "AF target".

So here we go:

• We half press.
• The system makes a focus error measurement.
• A lens movement that ideally should bring focus to "perfect" is calculated.
• The lens is told to make that movement.
• At the end of the movement, another focus error measurement is made.
•• If the error is within an established band about zero (the acceptance band), focus confirmation is declared: we get a beep and a steady green light, and the lift of the mirror and subsequent release of the shutter are enabled (and will happen if we are at full press).
•• if not, a new lens movement (that ideally should now bring focus to perfect) is calculated, and the lens told to make that movement.

And so on.

This figure summarizes this behavior in a compact way:

canon_eos_af_fc41.gif


For our purposes here, an important tenet of this algorithm is that the shot is never enabled until it has been determined, by a measurement taken at the lens position that would be used for the shot, that the focus at that position is within the acceptance band.

But, not so fast

However, intensive analysis of recent test observations here and by our colleague Hans Jørgensgaard suggest a somewhat different situation. I will describe only the case where the difference shows up; other cases follow what was described above.

• We half press.
• The system makes a focus error measurement.
• A lens movement that ideally should bring focus to "perfect" is calculated.
• The lens is told to make that movement.
• At the end of the movement, another focus error measurement is made.
• • If the error is not within the acceptance band, but is within a slightly wider band (the approach band), then:
• • • We get a beep and a steady green light.
• • • Lift of the mirror is enabled (and will happen if we are at full press).
• • • A new lens movement (that ideally should now bring focus to perfect) is calculated, and the lens told to make that movement. The movement is "tagged" so that another focus error measurement will not be required at its conclusion.
•At the end of this movement, release of the shutter is enabled (and will happen if we are at full press).

Now, how can this work? Well, when in the approach band ("focus nearly acceptable but not quite"), the calculated movement (very small), even if not perfect, will assuredly take focus to within the acceptance band. (To use a golfing metaphor, this is a putt that can certainly be "tapped in".)

What is the advantage?

Time saved. For one thing, if the target luminance is low, it can take an appreciable time to make the focus error measurement. Time is required for secondary images to "build up" on the sensors in the AF detector. (This is no different from the time required to get a proper exposure on the main sensor for our "taken" image.)

Thus, with no final measurement needed in the situation of landing in the approach band, we speed up the overall time to shot by that amount.

In addition, with the mirror lift underway before performance of the final movement, we expedite the actual shot by the time required for the final movement.

This figure summarizes this behavior in a compact way for the most illustrative cases:

canon_eos_af_fc42.gif


Shortly after the observations that led to this "theory" of system behavior, Hans found a 1988 Canon patent (US Pat. 4,792,821 to Akashi) in which exactly this behavior and rationale is described. (There is some lack of clarity regarding the mirror and shutter details, but "focus confirmation" is clearly shown to occur as I describe and, in a later patent, presenting the same scheme, the linkage of shutter release to "focus confirmation" is articulated.)

Ah, the mind of Canon.

Much thanks to Hans and to our colleague, "Wilba", for their work on this matter, part of a much broader attempt to gain further understanding of many aspects of the Canon EOS AF system (Will Thompson calls it the "Canon antifocus project"). It runs around the clock in three shifts (helped by time zone matters).

I do not in any way wish to suggest that this outlook was discovered here - it may have been widely discussed elsewhere. But it's new to me.

Best regards,

Doug
 
Last edited:
Top