View Full Version : The Canon EOS AF system revisted

Doug Kerr
November 23rd, 2011, 10:25 AM
As you may know, I have been part of a small group who have been seeking to develop an improved understanding of the actual concepts of the Canon EOS autofocus system. In an earlier thread, I summarized the understanding of the scheme that had emerged at that point in the group's work.

Subsequent and recent work by several of the participants has led to a somewhat modified model of the scheme. I will describe it here.

Note that the nature of this matter is such that our confidence in our thoughts as to the specific details is "modest". We know there are many internal details not yet seen clearly - certainly many not seen at all.

In any case, this description of "our" findings is my own, and has not yet been "signed off on" by the others involved in the process.


Terminology and notation

I will first establish some terminology and notation I will use.

Focus barrel (FB): the "part" of the lens moved by the AF motor, which in turn moves one, several, or all the element groups of the lens to change its "focus state".

Focus barrel position (FBP): the position of the focus barrel (in terms of "ticks", the unit by which changes in barrel position are commanded).

Detector offset (DO): this is the offset between the positions of the images on the two sensors of an AF detector from which we can determine if the image would fall precisely on the main sensor and if not, in which direction and by how much.

Raw focus error (RFE): the amount (if any) and direction by which, based on the DO, we can expect the image to fall in front of or behind the main sensor.

Adjusted focus error (AFE): the RFE adjusted by an amount whose purpose is to compensate for the fact that for various reasons (notably spherical aberration) the focus state in which the DO is zero may not be the "best" focus state for the actual shot image.

Distance to perfect focus (DTF): The amount (and direction) by which it is projected that the FB position would have to be changed to attain perfect focus. This is based on AFE plus a parametric model of the lens+body mechano-optical behavior (in some cases, a fairly primitive one).

Acceptance tolerance (t): The amount by which the current focus error (usually stated in term of DTF) may differ from zero and still have focus considered "good enough to shoot". In terms of the "golf" metaphor used by the group in discussions, this is the 'radius of the cup'.

Approach tolerance (T): The amount by which the current focus error (usually stated in term of DTF) may differ from zero, and be greater than 't' but nevertheless small enough that we qualify for a special expedited procedure to "close". In our metaphor, this is the 'radius of the green'; the green is small, and a single well-planned putt from anywhere on it will almost always put us in the cup.

The scenario

We will assume here that focus search is not necessary; that is, at the initial focus state of the lens when AF is commenced, a reliable DO determination can be made.

Phase 1

Th operator presses the shutter release (half or full). The system determines DO, then RFE, then AFE, then DTF. The lens is instructed to move the FB DTF ticks (DTF has an algebraic sign, so the movement can be in whichever direction is needed).

Note that ideally, when the lens has done this, the residual focus error will be within 't' (the acceptance tolerance).

But the matter of calculating the DTF is very tricky, and we cannot always be certain that the value is good for a "hole in one". So we will do "fly-by-wire" on our little golf ball.

After the lens is up to speed, as frequently as possible a new determination is made ("on the fly") of DO>RFE>AFE>DTF, and a new order is given to the lens, telling it to move the FB DTF ticks from where it is now. Note that this process is quite tricky, given that the FB is moving along and it takes a while for the DO value to "develop", and the value of DTF may not be ideal for various reasons. Nevertheless, this iterative process is convergent, and thus will to a great extent at least overcome the imperfections in the initial determination of DTF.

Eventually, the error (in terms of the DTF it leads to) will fall within 'T" (that is, the ball seems to be "over the green"). The system tells the lens STOP NOW.

This is the completion of Phase 1: hopefully we are on the green (and in one stroke, thanks to "fly by wire"!). But maybe not quite on the green.

Phase 2

A new determination is made of DO>RFE>AFE>DTF. (This uses the less tricky "from a standstill" process.)

At this point, the scenario branches, depending on whether or not the shutter release is at full press (that is, does the user want to fire the shot as soon as possible or not).

Shutter release only at half press

[**]If DTF is not greater than 't' (we are "over the cup"), focus is declared "confirmed", and whenever the shutter release goes to full press, the shot is fired (AE, mirror lift, shutter released).

If DTF is greater than 't' (we are not "over the cup"), the lens is told to move the FB DTF ticks. When the lens is done, a new determination is made of DTF (I won't mention the three first stages of it any more).

We then loop to "[**]".

Shutter release at full press

[***]If DTF is not greater than 't' (we are over the cup), focus is declared "confirmed", and the shot is fired (AE, mirror lift, shutter released).

If DTF is greater than 'T' (we are not "on the green" at all), the lens is told to move the FB DTF ticks. When the lens is done moving, a new determination is made of DTF.

We then loop to "[***]".

If DTF is greater than 't' but no greater than 'T'" (we are "on the green" but not "over the cup"), then:

• AE is done and then mirror lift is commenced.

• The lens is told to move the FB DTF ticks (while the other stuff is happening).

When the lens is done moving, the shutter is released without further ado.

We are done.

Wow! This guy is so cocky he strokes the putt, tuns around, and goes right into the clubhouse!

Then object of this is that, if the user has exhibited his anxiousness to get the shot off by giving a full press at the outset, the system hustles by starting the shot preparations (AE, mirror lift) as soon as the ball is "on the green", and releases the shutter immediately after the putt, "on the come".

What if, after the lens stops moving, focus is not within the degree corresponding to the tolerance 't'? Too bad. Stuff happens. It should be "rare".

Well, that's how it looks from here, today.

Great thanks to my colleagues in the effort for their respective contributions. They include:

"Wilba" from Western Australia

Hans P Jørgensgaard from Denmark

And most recently, not really a member of the group, but making great contributions by way of correspondence with "Wilba", Rainer Hönle from Germany.

They have done the heavy lifting here.

Best regards,