Canon's AF system vs. Nikon's AF system

hockeynut said:

These people are incredibly quick to blame the camera. In almost
every case, they do NOT blame themselves or the lenses. They will
always blame the camera. You don't see them in the lens forums,
do you?
In all seriousness, Canon sold 1.4 million DSLRs last year. If the
problems were as severe as a few people made them out to be, we'd
be in serious, serious trouble on these forums. But in reality, we
people every week who claim trouble. Rarely do they post samples.
Rarer still is EXIF data. And after all is said and done, maybe
one in 100 has a legitimate beef with Canon. And even then, it's
more than likely they've got a bad lens rather than a bad camera.

Click to expand...

hockeynut said:

Frankly, I believe that there are more 3rd party lens compatibility
issues that anyone cares to admit. If you track down all of these
"troubles", many of them are 3rd party lenses. How many "L" lenses
have troubles? How many Canon USM lenses have troubles? As you go
to the lower priced lenses, you have more ... inconsistency. Isn't
that the nature of price vs. quality? If you want the performance
of a Canon "L" lens, you aren't going to get it by paying $220 for
a Sigma. You can get close, and you might be happy 90% of the
time, but 99% of these AF claims are from people with a SINGLE
LENS! How is that proving that the camera is the problem?

Click to expand...

Bitplayer said:

I agree with everything you have said. But doesn't the camera
re-evaluate the focus after the lens has completed its action?
Won't it fine adjust, or if focus remains way off, hunt? These
following short motions would usually be impossible to
differentiate (by the user) from the initial main focus move.
However, when the predictive system is a bit off, this would be
detectable as short stutters after the primary move. I believe I
have detected this once in a while. (The fine adjustments can be
predictive too.)

Click to expand...

Bitplayer said:

I have designed motor control (not camera related) systems that
work in a similar manner. The predictive system allows most of the
slewing to take place at the highest possible rate. However,
unless the predictive system is pretty close to perfect, you have
to go into a fine adjustment phase after the main move. You might
call this a "stepped" or "punctuated" closed loop.

Click to expand...

Doug Kerr said:

The problem I have is that then it vitiates the explanation of how
lens "calibration" error can produce incorrect focus.

One possibility is that there is a dead band for the "refine" stage
so that further correction doesn't occur unless the residual error
after the (initial) predictive stage is greater than a certain
amount - which might be substantial.

Click to expand...

AK_2 said:

Every 20d, every 10d had the same major flaw with af - in that with
certain lenses (irrespective of price) the tolerance between shots
varies so much (with the same situation, body, lens) that no matter
how the body and lens is calibrated it can never be relied on to
give an in focus shot. Every 20d and 50mm f1.4 combo i have tested
had the same problem.

Click to expand...

AK_2 said:

The other side of this - is that the very same body, with certain
other lenses can give a good well focussed shot every time. So if
one buys with the knowledge of what works well, one is likely to
have very few (if any) issues.

The above is one reason not everyone sees a problem. User error is
another. And of shocking insight, is that the user error problem is
demonstrated by non other than canon's own cps technical experts -
if they don't know how to test af, and they then cause damage to
camera they wrongly adjust, how is the average customer ever to
know any better ??? Canon damaged my system twice !!! The techy
they used had no idea what he was doing when testing and adjusting
af on my system of bodies and L lenses.

My tests are 100% repeatable, and have been related to other pro's
equipment and how they use it in the real world.

For me, the only canon bodies that pass the test with a suitable
range of canon lenses are the pro spec 1 series.

In addition to tolerance issues, some bodys and lenses are out of
adjustment - thats a seperate isssue, and certainly for the 20d I'm
yet to test one myself out of spec.

The 20d also works very well with sigma f2.8 lenses, better than
the same body works with a canon L f4 for auto focus accuracy. so
cheaper doesn't mean worse.

The nikon d70 happens to focus more accurately than the canon
equivalents - i'd love to know why ???

Regards,
Kev

Click to expand...

northernlights said:

RDKirk said:

You list a set of ORs that pretty much encompass the range of
possibilities. The fact that this equipment is intended for active
use requires the ability to readjust it when the occasion arises.
No, we don't have to calibrate the hard drives on laptops, but they
also fail frequently when pushed beyond rather narrow tolerances.

I can't think offfhand of any truly versatile and high-performance
devices that are not adjustable.

However...this equipment IS actually user-calibratable (just like
my target pistols). Canon used to provide the software to do it.

dboreham said:

I've seen enough posts here with solid evidence to
believe that there is some kind of generic problem
with the 20D focus and certain lenses. I have to say
however that I have not seen a problem myself.

The fact that the body and lenses can be
adjusted at_all is to me a red flag : it means
that the system design is bad. There's no way
that a control system that requires callibration
should be designed into a consumer item,
IMHO as an engineer. It's just bad bad bad bad.
The focus system should either auto-callibrate,
OR be designed such that it stays within
tolerances for the available manufacturing tolerances,
OR it should support in-field callibration.
All this sending bodies and lenses to Canon is
just wrong. It has to cost Canon far more
to have a person callibrate a body than all the
profit they would ever make from the sale
of that body.

Imagine if you had to callibrate your hard drive
before the head would track the data tracks
properly...

Click to expand...

--
RDKirk
'TANSTAAFL: The only unbreakable rule in photography.'

Click to expand...

Click to expand...

dboreham said:

bibi0012 said:

But some of issues maybe due to lens mechanic because folks who
sent the lenes back to Canon for re-calibration eventually got back
a good focusing lens or camera.

Click to expand...

This may be placebo effect.

It's been a few years since I had to study control theory,
but it seems to me that the AF control system always
slews the focus ring in one direction. It's optimally
damped, or whatever the term is : there's no overshoot
and no ringing. I think this might be the root of the
problems because you always get a non-zero error
in control systems of this type. Ideally the offset
error should be very small, but perhaps there are
things that can happen with the lens that give
rise to a larger error.

The other thing that seems to happen is that the input
to the control system is the AF sensors. Those sensors are
not at the focal plane. The body needs to be callibrated
for the difference in optical path length between the
AF sensors and the focal plane. If that distance is incorrect
then the body will consistently focus with a constant
error , either front or back of correct focus.

Click to expand...

Bitplayer said:

Yes, the question is why would a camera report to the user that a
focus had been achieved when that wasn't true -- given that the
camera can detect, with good contrast, etc., proper focus.

There may be a simple explanation. If the lens focusing mechanism
doesn't insure that the moving lens(es) remains in proper alignment
to the non-moving ones, and the focus sensor, then the camera could
be fooled about where focus would be and whether or not it had been
achieved. An error in the calibration of the physical alignment of
the lenses could explain the behavior we see.

Click to expand...

Bitplayer said:

Just speculating. I wish Canon would release some details.

Click to expand...

charles schuler said:

For Doug and BitPlayer:

So, what's your conclusion here? Does the servo run open-loop,
re-evaluate and perhaps make a second (or additonal) correction if
the sensor calls for it?

Click to expand...

Steve H said:

I understand the concept of an open loop system in this context, in
the sense that the camera doesn't re-check or fine-tune focus after
the initial focus command is completed. But suppose you then lift
off the shutter button and force the camera to do another iteration
of the process. Doesn't this then effectively become a sort of
limited "closed loop" system, admittedly with manual intervention.

Click to expand...

Doug Kerr said:

charles schuler said:

For Doug and BitPlayer:

So, what's your conclusion here? Does the servo run open-loop,
re-evaluate and perhaps make a second (or additonal) correction if
the sensor calls for it?

Click to expand...

Yes, I think that is the most credible conjecture, although there
are of course some things that don't seem to fit perfectly.

Click to expand...

Doug Kerr said:

Hi, Thomas,

Generally speaking, a "closed loop" system is one that depends on
feedback to achieve the desired result. For example, in my home,
the thermostat causes the furnace burner to operate until the
thermostat determines that the temperature is that which you have
set, at which point the burner is shut down.

An "open-loop" system operates without benefit of feeback. For
example, I have a timer that allows water to flow through a hose
into my swimming pool for a preset period of time. I set the timer
based on how much I want the level of water in the pool raised and
the known area of the surface. The timer lets water in for that
period, and hopefully the right new level is attained. The timer,
however, has no way of knowing where the water level has gotten to:
there is no "feedback" used in the system.

Some camera autofocus systems operate on a classical closed-loop
basis. They move the focus cam while observing some indication of
the state of focus. When the right value is attained (or, in some
cases, the "best value"), the movement is stopped. This is a
classical closed-loop operation.

The Canon system operates on a different basis It determines how
far the image is out of focus (in terms of some scale that relates
to the geometry of the focus detectors) with the focus cam in the
lens wherever it happens to be when focsu is determined.

Then the camera instructs the lens to shift its focus setting in a
certain direction by a cerain amount (again, presumably in the
basis of a scale relating to focus detector geometry).

If we stop here for a moment, and don't inquire how the lens is
able to do that, then we can see that the overall system (treating
the lens as an obedient black box) is open loop. The camera does
not, for example, tell the lens, "shift your focus cam and I'll let
you know when you have gotten it to the place that produces correct
focus". It says, "make this change - I know the result will be what
we want."

But now lets see how the lens does its part of the job. It takes
the command from the camera, and based on knowledge of the present
position of the focsuing cam (derived from a position encoder on
it) and tables describing the effect on the amount of misfocus (on
the scale used by the camera) at different positions of the
focsuing cam, determines what new position of the focusing cam is
needed to fulfill the camera's directive. (There is of course a
microprocessor in the lens that does the heavy lifting here.)

Having decided where it wishes the focsuing cam to be, the
processor starts the motor and , watching the position of the cam
as indicated by the position encoder, runs the motor until the cam
is in the desired positon.

This latter phase of the deal is a classical closed-loop servo
operation.

Inicidentally, if the tables in the lens do not precisely reflect
the relationship between focusing cam posiiton and the resulting
state of focus, the behavior of the whole process will be faulty.
Correcting this is what is usually involved in "focus
recalibrating" an individual lens.

There are a few further complications, but that gives the basic
ideas involved.

Best regards,

Doug

Click to expand...

Daniella said:

I am skeptical about that. I woudl think that the calibration is a
hardware issue, not software. hmm are you sure?

Click to expand...

Daniella said:

bibi0012 said:

Why Canon's AF has caused so much grief for many people here on
this forum? Why we don't hear Nikon owners wining about Nikon's AF
system as much?

Click to expand...

you hear a lot of complain about the AF on the d70. just do a
search on that forum and you'll see..lots of problems and also back
focusing too.

the AF on the nikon is also dependant on the lens. Take a look at
all the complain about the slug slow 80-400 VR adn you will
see..you can start to AF that lens and go have a coffe...

Click to expand...

Devendra said:

by the time 300D starts up?

its funny how you and a few others immediately pull out D70s
problems while the talk was about Canon and Nikon AF.

Click to expand...

Doug Kerr said:

I don't think so. The focus detectors should be able to deternmine
if a well-focued image is in place, and nothing in the lens could
mess that up (unlesss of course the lens generated a completely
corrupt image for both the focsuing detectors and the imaging
system).

Click to expand...

Doug Kerr said:

Doug Kerr said:

Just speculating. I wish Canon would release some details.

Click to expand...

Click to expand...

RDKirk said:

Canon camera bodies interrogate each lens for its operating
characteristics when the lens is first mounted. The body
determines (independingly of the state of the lens) the intended
aperture. It calculates the Depth of Focus at the virtual sensor
plane based on the intended aperture, thus the physical amount of
tolerance it must achieve in terms of focused plane movement
distance.

Click to expand...

RDKirk said:

circle) of the light at the virtual sensor plane and determines how
far and in what direction the actual focused plane needs to move to
achieve the required tolerance (either merely within the Depth of
Focus at the virtual sensor plane or within 1/3rd of the Depth of
Focus at the virtual sensor plane). It gives the lens that
command: Move forward/backward x millimeters. It does not check
the state of focus at the virtual film plane again to determine if
its command was successfully carried out (this is the "open loop"
aspect). (Also, note that this entire process is totally distinct
from the question of whether the system is focusing on the point of
the scene that the photographer intended.)

Click to expand...

RDKirk said:

How does the inertia of the lens components factor in to this?
That would depend on what the command consists of. I'm presuming
it's a direction/distance of movement command vice a
direction/duration of moment command. If so, the component
inertial would only matter in considerations of focusing speed, not
accuracy.

bibi0012 said:

Why Canon's AF has caused so much grief for many people here on
this forum? Why we don't hear Nikon owners wining about Nikon's AF
system as much?

Here is how I see it...

All Nikon's AF motors are built inside the camera bodies. (except
for a small number of Nikon telephotos lenses that have built-in AF
motors.) It's easier for Nikon cameras to achieve precise focus
because it knows the torque of its built-in AF motors.

All Canon's AF motors are built inside the lenses. Different lenses
use different motors, with varing torques, or maybe different
generation of in-the-lens microprocessors. And there are different
types and different generation of USM motors.

I am sure that Canon's cameras can tell which type of motors is
used for which lens and adjust it's AF signal timing accordingly.
The problem is that any small amount of variation in AF motor's
torques will throw the focus off.

Because there are so many permutation of lenses and cameras, AF is
always a hit-or-miss thing with Canon.

As a Canon owner, you need to have your lens or camera
re-calibrated if you find your camera has trouble focusing
correctly.

This are just my personal opinions. Comments are welcome.

Click to expand...

--
RDKirk
'TANSTAAFL: The only unbreakable rule in photography.'

Click to expand...

RDKirk said:

Devendra said:

by the time 300D starts up?

its funny how you and a few others immediately pull out D70s
problems while the talk was about Canon and Nikon AF.

Click to expand...

--
RDKirk
'TANSTAAFL: The only unbreakable rule in photography.'

Click to expand...