Do it yourself autofocus adjustment for Canon 350D

In article .com,
says...
Bart van der Wolf wrote:

The initial focusing data (based on phase differences) will be
improved by camera calibration. The following info exchange between
camera and lens will determine how much that particular lens will need
to move the optics.

And when the lens doesn't move enough, the camera sees this lack of
focus, and asks it to move some more.

http://en.wikipedia.org/wiki/Closed_loop


AF on Canon (and probably other) DSLRs is open loop - it is definitely
NOT closed loop. The camera says to the lens "focus by amount X" and
trusts the lens to do what it is told. There is no check that it has
actually done so. Presumably this is because a closed loop AF system
would be much slower.

"Graeme Cogger" wrote in message
...
In article .com,
says...
SNIP
And when the lens doesn't move enough, the camera sees
this lack of focus, and asks it to move some more.

http://en.wikipedia.org/wiki/Closed_loop


AF on Canon (and probably other) DSLRs is open loop - it is
definitely NOT closed loop.

From what I've read, it apparently seems to be open-loop indeed
(http://en.wikipedia.org/wiki/Open_loop).

The camera says to the lens "focus by amount X" and trusts
the lens to do what it is told. There is no check that it has
actually done so. Presumably this is because a closed loop
AF system would be much slower.

Yes and, because of the relatively wide EF mount, the phase detection
may provide enough information to accurately determine the direction
and amount of movement to succeed in one go. It would partly explain
the focus speed.

Bart

wrote in message
oups.com...
Bart van der Wolf wrote:

You are assuming it's a Closed Loop system, any evidence?

Well, you can engage the "AI Servo" mode if you want ;-)

But even in "one shot": pick up EF 500/4 (or probably any f/4
lens),
and connect to a stacked pair of teleconverters: 1.4x + 2x. Mount
the
mess to 1DMkII, point at object, and hit the AF. Watch, listen and
even feel as the focus motors slowly, but surely, bisect onto the
correct focus. Even with "lesser" lenses, you can occasionally feel
the focus motors tweak more than once in "one shot" mode.

Failure to achieve focus, and trying/failing again, is IMHO not the
same as a closed-loop system with adequate light levels. It also
doesn't explain the difference between Pro and non-Pro body focus
accuracy (caused by smaller/more accurate phase detection elements).

I'm fairly amazed it would be an "open issue"; AF wouldn't work
well
if it was based on a single observation, single command, and no
follow
up.

Canon's historical backwards-compatibility break going from the FD to
EF mount was not only about the introduction of Electronic Focus. It
was also about mount size, thus allowing more accurate Phase detection
for closed-loop AF. The compatibility break was inevitable and, in
Europe anyway, it started a landslide type of brand switch from Nikon
to Canon for sports and journalism.

http://www.pat2pdf.org/patents/pat6603929.pdf describes "a" method of
closed-loop AF, which requires multiple measurements (of phase *and*
contrast) to achieve both speed *and* accuracy, but there is no
evidence that this (Aug-2003) patent is used in EF-mount designs
(which pre-date the patent). It instead seem likely that this patent
is a new method, not (yet) implemented in the older EF-mount/Camera
intelligence. It might even just apply to P/S cameras which have
life-sensor-review feedback.

Bart

wrote:
You are assuming it's a Closed Loop system, any evidence?
Well, you can engage the "AI Servo" mode if you want ;-)

That's constant refocusing.

I'm fairly amazed it would be an "open issue"; AF wouldn't work well
if it was based on a single observation, single command, and no follow
up.

Sometimes it doesn't work that well. Ever run into a lens with focusing
problems? Typical problems like constant slight misfocusing should not
ever be an issue with a properly designed closed-loop focusing.
--
harri

Bart van der Wolf wrote:

Failure to achieve focus, and trying/failing again, is IMHO not the
same as a closed-loop system with adequate light levels.

Any system that examines what it is doing in order to behave properly
and/or better is, by any reasonable definition of the phrase, "closed
loop". If you wish to argue otherwise, you'll have to do it on your
own.

It also
doesn't explain the difference between Pro and non-Pro body focus
accuracy (caused by smaller/more accurate phase detection elements).

One presumes that the "one shot" mode stops as soon as the focus
estimate is supposedly optimal. If the variance in the estimates is
larger for non-Pro bodies, we can expect the non-Pro bodies to have
inferior AF properties. And indeed, this is what we _do_ observe.
Also note that this would be observed whether or not the AF system was
open or closed.

With evidence in hand, let's speculate:

The Canon AF system is closed, and works something like this:

1. user engages AF function
2. camera makes an AF measurement
3. if in focus (by whatever measure)
and we are in "one shot" mode, then we are done.
3. camera sends lens a command
4. camera waits for lens to say "done"
5. goto 2

Minor and major details omitted -- e.g., it's clear that when step (2)
fails, there is a sub-mode that commences a focus search by racking the
lens in and out, and probably the AF sensor is tickled continuously
while this happens, looking for any hook to lock onto, and so on.

But the above algorithm explains _all_ the behaviour I have observed
with my old Canon EOS 5, a Canon 10D, and a Canon 1DMkII. In
particular, it is completely consistent with the observations I posted
back on Friday. Has anyone repeated my experiment?

Harri Suomalainen wrote:

wrote:
You are assuming it's a Closed Loop system, any evidence?
Well, you can engage the "AI Servo" mode if you want ;-)

That's constant refocusing.

Yes, you see that ";-)" I used? Are you aware of its meaning? Did you
even read the subsequent paragraph I wrote?

I'm fairly amazed it would be an "open issue"; AF wouldn't work well
if it was based on a single observation, single command, and no follow
up.

Sometimes it doesn't work that well. Ever run into a lens with focusing
problems? Typical problems like constant slight misfocusing should not
ever be an issue with a properly designed closed-loop focusing.

You need to re-read this thread. From the beginning.