Pro Photo HOME
Go Back   Pro Photo HOME > Professional Digital Camera Discussion - Full Access for Basic Members > Canon 1-series Digital SLR and EOS 5D

Notices

Reply
 
LinkBack (11) Thread Tools Display Modes
  3 links from elsewhere to this Post. Click to view. #8  
Old 06-26-2010, 12:05 PM
Doug_Kerr's Avatar
Doug_Kerr Doug_Kerr is offline
Premium Lifetime Member
 
Join Date: Oct 2003
Location: Weatherford, Texas
Posts: 2,471
Doug_Kerr
Re: Canon EOS AF scheme

This morning I ran some simple tests regarding the matter at issue.

I used a Canon EOS 40D body with two different lenses:

A Canon EF 70-200mm f/2.8L IS USM, equipped with a Canon 2x focal length converter.

A Sigma 18-200mm F/3.5-6.3 OS. (not the HSM focus drive version)

The test was as follows. The camera was mounted on a tripod, and aimed at a scene about 25' distant, with good patterns for AF. In each case, image stabilization was disabled.

I would then set the initial focus position to one end of the range or another and then half press, while observing the focusing scale (on the Canon 70-200+2X) or the focusing ring (on the Sigma 18-200).

In the case of the Canon 70-200+2X, on many (but not all trials) I was able to observe the focusing scale moving decisively to essentially the final position, and after a very brief pause, there was a second very small movement.

In the case of the Sigma 18-200, in most cases I saw the focusing ring move to near the final position, after which there was a second small movement, and then an even smaller movement (and I couldn't be sure exactly how many stages of this "convergence" there were.)

I can postulate at least two explanations for this behavior:

Explanation 1:

1.1 The AF system makes a measurement of the "focus error" at the AF detector (with the focusing cam at its present position) and feeds that to the lens.

1.2 The lens, based on a function described by tables, working from the current position of the focusing cam (presumably provided by a position encoder) and the error feed from the camera, calculates the focus cam position that should result in ideal focus, and instructs the focus cam positioning system to move the focusing cam to that position.

1.3 The focusing cam positioning system does this in a multi-stage process, perhaps first moving at maximum speed to essentially the desired position (based on feedback from the position encoder) and then, at a slower speed, moving to exactly the desired position.

1.4 When this has fully played out, the lens so reports to the camera, which declares focus lock to be achieved.

Explanation 2:

2.1 The AF system makes a measurement of the "focus error" at the AF detector (with the focusing cam at its present position) and feeds that to the lens.

2.2 The lens, based on a function described by tables, working from the current position of the focusing cam (presumably provided by a position encoder) and the error feed from the camera, calculates the focus cam position that should result in ideal focus, and instructs the focus cam positing system to move the focusing cam to that position.

2.3 The focusing cam positioning system does this in a single-stage process, perhaps moving at maximum speed to essentially the desired position (based on feedback from the position encoder).

2.4 When this has played out, the lens advises the camera.

2.5 The camera makes a new determination of "focus error" on the AF detector.
--2.5a If the error is now within the established threshold, the camera declares focus lock to be achieved [and we exit this process].
--2.5b If not, the residual error is fed to the lens, and steps 2.2 through 2.5 are performed again.

By the way, I seems that in the case of the Sigma 18-200, the final result of the movement of the focusing cam, even though done in several stages, does not always produce a focus situation that the camera would adjudge "perfect".

I base this on the fact that if I release the shutter release, and freshly half press, the focusing cam moves again slightly (thus suggesting that there was a focus error at the current setting of the focusing cam, the setting that resulted from the full playout of the AF process the first time).

(In fact several years ago I suggested this as a test to ascertain if the AF process is working "perfectly" on a certain lens-body combination.)

This suggests to me that Explanation 1 is in effect.

But who knows. Perhaps ingredients of both are in play.

Best regards,

Doug





Last edited by Doug_Kerr; 06-26-2010 at 12:20 PM.
Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #9  
Old 06-28-2010, 02:54 PM
ChuckWestfall ChuckWestfall is offline
Premium Lifetime Member
 
Join Date: Jul 2000
Location: Lake Success, NY USA
Posts: 2,599
ChuckWestfall 10
Re: Canon EOS AF scheme - Chuck?

Hi, Doug:

Thanks for posting your test methods. In doing so, you may have exposed some variables that could explain your results.

I'm sure you already realize that AF measurement in the camera is a separate operation from AF drive in the lens. What may not be immediately apparent is that different AF drive motors have different stopping characteristics.

The ring USM used in most current Canon EF lenses and some EF-S lenses as well is essentially a direct drive actuator. It uses an encoder with a discrete number of segments. When the lens CPU issues a drive command, the rotor portion of the USM moves in the direction specified by the CPU for a specific number of segments. There is a monitor in the lens that counts the number of segments that the rotor has moved, and effectively stops the movement by cutting off electrical power when the correct number of segments has been reached. The nature of this drive mechanism is such that its stopping motion is immediate; there is no need for the motor to slow down before it reaches the end of its "assignment." The stopping position is also quite precise because of the segmented design of the encoder and the efficiency of the "braking" mechanism, which is a type of leaf spring.

On the other hand, the rotary lens drive motors used in most non-ring-USM designs feature an indirect drive mechanism. In other words, they have an output shaft that connects to a gear-reduction mechanism that ultimately connects to the moving parts of the lens' focusing components. The drive shaft of this motor type rotates at extremely high speed during operation, which contributes to overall focusing speed. However, the rotary motor cannot stop as quickly or efficiently as a ring USM. Instead, it typically slows down in a series of steps such as full power --> half power --> quarter power, etc. until it reaches a complete stop.

I think this behavior is what you observed with the Sigma lens. If you had tested a Canon lens with a rotary focusing motor, such as the EF-S 55-250mm IS lens for instance, you probably would have observed similar results.

As you mentioned in your tests with the Canon EF 70-200L lens plus 2X extender, AF drive sometimes stopped immediately, while at other times it seemed to move twice, for a very brief moment the second time. I think this behavior shows that the camera's initial AF measurement was sufficiently accurate most of the time, but occasionally the lens drive may have overshot the originally specified pulse count by a segment or two and was therefore readjusted. I would suspect that your use of the 2X extender may have contributed to this performance, because it significantly reduces the depth of focus compared to the same lens without the extender in place. Since the encoder is primarily designed for use of the lens "as is," the demands for AF precision are increased when an extender is used.

I don't have access to any information that definitively states whether the offset for a user-input AF microadjustment is applied before or after the camera's standard AF calculation, but I suspect that it is applied before the command for lens drive is issued from the camera body to the lens CPU. Therefore, it should have no bearing on the stopping performance of the lens motor.
__________________
Chuck Westfall
Technical Advisor
Professional Engineering & Solutions Division
ITC Group/Canon U.S.A., Inc.

Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #10  
Old 06-28-2010, 03:20 PM
Doug_Kerr's Avatar
Doug_Kerr Doug_Kerr is offline
Premium Lifetime Member
 
Join Date: Oct 2003
Location: Weatherford, Texas
Posts: 2,471
Doug_Kerr
Re: Canon EOS AF scheme - Chuck?

Hi, Chuck,

Thanks you so much for that extended discussion. It is very helpful.
Quote:
Originally Posted by ChuckWestfall View Post
. . .On the other hand, the rotary lens drive motors used in most non-ring-USM designs feature an indirect drive mechanism. . . .However, the rotary motor cannot stop as quickly or efficiently as a ring USM. Instead, it typically slows down in a series of steps such as full power --> half power --> quarter power, etc. until it reaches a complete stop.

I think this behavior is what you observed with the Sigma lens.
That makes complete sense - it completely fits what I observe. (Incidentally, that particular Sigma lens has the older type of drive - a "DC motor".)
Quote:
If you had tested a Canon lens with a rotary focusing motor, such as the EF-S 55-250mm IS lens for instance, you probably would have observed similar results.
Yes, and I will look into just that with some of my non-USM EF and EF-S lenses.

Quote:
As you mentioned in your tests with the Canon EF 70-200L lens plus 2X extender, AF drive sometimes stopped immediately, while at other times it seemed to move twice, for a very brief moment the second time. I think this behavior shows that the camera's initial AF measurement was sufficiently accurate most of the time, but occasionally the lens drive may have overshot the originally specified pulse count by a segment or two and was therefore readjusted.
Ah, now we get to the precise crux of my interest. Do you mean by this that if the lens drive did "overshoot":

a. The lens drive system would notice this, based on the position encoder, and refine the position to what was originally determined as the desired position (the body not being involved), or

or

b. The camera body would perceive, via the AF detector, that ideal focus has not been attained and instruct the lens that a further "adjustment" was needed?

Quote:
I don't have access to any information that definitively states whether the offset for a user-input AF microadjustment is applied before or after the camera's standard AF calculation, but I suspect that it is applied before the command for lens drive is issued from the camera body to the lens CPU. Therefore, it should have no bearing on the stopping performance of the lens motor.
I would have conjectured that as well. In any case, it doesn't pertain to my testing, since my body does not support that feature.

Thanks for staying with me on this. We're almost there!

Best regards,

Doug

Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #11  
Old 06-28-2010, 05:46 PM
ChuckWestfall ChuckWestfall is offline
Premium Lifetime Member
 
Join Date: Jul 2000
Location: Lake Success, NY USA
Posts: 2,599
ChuckWestfall 10
Re: Canon EOS AF scheme - Chuck?

Hi, Doug:

I'm glad this information was helpful to you.

I understand your "crux" question quite clearly, and realize that it is at the heart of an ongoing debate about "closed loop" vs. "open loop" focusing algorithms. Sorry to tell you, but I won't be drawn into making a definitive statement on that point. Suffice to say it's one of the specific topics that Canon does not publicly disclose, as is their prerogative.

I would only point out that in the specific case under discussion, i.e., use of a large-aperture telephoto zoom lens with a 2X extender, the precision of the USM's encoder is being taxed to its utmost, due to a significant reduction in depth of focus compared to the use of the same lens without an extender.

For the sake of discussion, it would be interesting to perform the same tests that you did with the 70-200/2.8L and 2X vs. a fixed focal length EF 400mm lens. If we are to hypothesize that the AF system is remeasuring AF after initial lens drive is complete, then the stop-and-start behavior that you observed with the 70-200/2.8L and 2X should be present with the EF 400/5.6L. I'll bet that it isn't.
__________________
Chuck Westfall
Technical Advisor
Professional Engineering & Solutions Division
ITC Group/Canon U.S.A., Inc.

Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #12  
Old 06-28-2010, 06:09 PM
Doug_Kerr's Avatar
Doug_Kerr Doug_Kerr is offline
Premium Lifetime Member
 
Join Date: Oct 2003
Location: Weatherford, Texas
Posts: 2,471
Doug_Kerr
Re: Canon EOS AF scheme - Chuck?

Hi, Chuck,

Quote:
Originally Posted by ChuckWestfall View Post
I'm glad this information was helpful to you.

I understand your "crux" question quite clearly, and realize that it is at the heart of an ongoing debate about "closed loop" vs. "open loop" focusing algorithms. Sorry to tell you, but I won't be drawn into making a definitive statement on that point. Suffice to say it's one of the specific topics that Canon does not publicly disclose, as is their prerogative.
I certainly understand. I would much rather hear that than have to believe that I had not clearly enough formulated my question!

Quote:
I would only point out that in the specific case under discussion, i.e., use of a large-aperture telephoto zoom lens with a 2X extender, the precision of the USM's encoder is being taxed to its utmost, due to a significant reduction in depth of focus compared to the use of the same lens without an extender.
I understand.

Quote:
For the sake of discussion, it would be interesting to perform the same tests that you did with the 70-200/2.8L and 2X vs. a fixed focal length EF 400mm lens. If we are to hypothesize that the AF system is remeasuring AF after initial lens drive is complete, then the stop-and-start behavior that you observed with the 70-200/2.8L and 2X should be present with the EF 400/5.6L. I'll bet that it isn't.
Thanks for that observation, and thanks for all your help in this matter.

Best regards,

Doug

Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #13  
Old 06-29-2010, 04:58 PM
Doug_Kerr's Avatar
Doug_Kerr Doug_Kerr is offline
Premium Lifetime Member
 
Join Date: Oct 2003
Location: Weatherford, Texas
Posts: 2,471
Doug_Kerr
Re: Canon EOS AF scheme - Chuck?

"As though I had nothing better to do" - Part n+1

I just conducted a little round of tests that casts further light on this whole area.

The tests were conducted using an EOS 40D body. The center AF point was preselected. The camera was aimed at a good focus "target". One Shot AF was in effect. Focus Search was turned off (just in case that might introduce any complications). At the beginning of each test shot, the lens focus was preset to a setting from which the AF system was prepared to proceed without benefit of Focus Search.

Three lenses were tested:

EF 70-200 F2.8L IS USM e/w EF 2X focal length converter. (The actual interest in the converter in this case is that with it aboard, the lens slew speed is reduced, which facilitated the maneuver to be described.)

EF 28-105 F4L IS USM

Signal 180200 f.3,5-6.3 OS

The test maneuver was this: with the lens focus preset as described above, I would full press and then quickly move my hand in front of the lens. The object of course was to prevent the camera from doing anything useful with the AF detectors after lens focus movement was completed.

In each case, after the lens movement was completed, the camera did not fire, and the finder showed a flashing green focus confirmation dot ("no confirmation"). (Yes, Carla was watching in the finder - I don't have the dexterity to do all of that myself anymore!)

Then, with full press still held, I moved my hand away.

In many cases, the lens focus ring/scale then moved slightly, and in every case, (after that slight movement, if present), the camera fired.

This strongly suggests that, after the focusing cam has been moved to the position calculated by the lens, the camera does look at the state of focus, as indicated by the AF detector.

If the indication is indeterminate, AF lock is not declared. If there is an indication, but it is not "right on", the lens is caused to move again and then (I assume if there is then a "right on" focus indication, ) AF lock is declared.

Interesting.

Best regards,

Doug

Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
  #14  
Old 06-30-2010, 10:15 AM
Doug_Kerr's Avatar
Doug_Kerr Doug_Kerr is offline
Premium Lifetime Member
 
Join Date: Oct 2003
Location: Weatherford, Texas
Posts: 2,471
Doug_Kerr
Re: Canon EOS AF scheme - Chuck?

In the last chapter of this saga, I reported on some simplistic tests which, on the surface, seemed to suggest that the AF scheme in my EOS 40D was what we might describe as "iterative open-loop".

That is, when we press the shutter release:

a. The body (if it can) makes a determination of the "focus error" (with the lens focusing cam in its current position) by observing "split-image" alignment on the AF detector(s).

b. The focus error is reported to the lens which, using a function described by internal tables, calculates, from the error and the current focusing cam position (read from a position encoder), the focusing cam position that should result in ideal focus for the current focus target.

c. The lens then drives the focusing cam to that position. It might do that "open loop" (in effect, counting motor "steps"), it might do it "closed loop" (controlling the motor while observing the position encoder), or it might use a combination of the two (this likely being different for the different drive mechanism types).[Note that this is a wholly-separate distinction from whether the overall AF process is "open loop" or "closed loop" or something in between.]

d. When this has concluded, the body observes the residual focus error (via the AF detector).
--1. If the error is within some established tolerance, focus lock is declared (and the camera will fire if we are in full press).
--2. If the error is not within the established tolerance, the cycle is repeated from step b.

********
The problem with this conjecture is that, were it true, then the accuracy of focus would not be affected by imperfections in the tables in the lens. If the function relating focus error and focusing cam position were not "exact", the result would be that the "open-loop" cycle might have to repeat (and perhaps then focusing would take longer than was desirable). But the result would nevertheless be focus that was correct "within the established tolerance".

Yet it is well recognized that a particular copy of a lens may exhibit systematic focus error, which can be minimized by having the lens "calibrated", which we have always assumed means to "perfect" its internal function by adjusting its tables to more accurately reflect its specific behavior.

So I remain baffled.

Canon's "secret" is safe for now - at least here.

The thought does occur to me that a different strategy may be followed by different bodies.

Best regards,

Doug

Digg this Post!Add Post to del.icio.usBookmark Post in TechnoratiFurl this Post!
Reply With Quote
Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is On
Trackbacks are On
Pingbacks are On
Refbacks are On


LinkBacks (?)
LinkBack to this Thread: http://www.prophotohome.com/forum/canon-1-series-digital-slr-eos-5d/91885-canon-eos-af-scheme-chuck.html
Posted By For Type Date
Re: I think I know why shims help some, and don't others: Canon Rebel (EOS 1100D-300D) Talk Forum: Digital Photography Review This thread Refback 07-09-2013 07:41 AM
Busted! The Myth of Open-loop Phase-detection Autofocus: Digital Photography Review This thread Refback 06-27-2013 06:58 PM
autofocus - When a camera auto-focuses, is the current focus distance reported from the lens to the camera? - Photography This thread Refback 01-11-2013 08:04 AM
An attempt to adjust focus of Sigma lens.: Canon Rebel (EOS 1100D-300D) Talk Forum: Digital Photography Review This thread Refback 01-02-2013 06:10 PM
Busted! The Myth of Open-loop Phase-detection Autofocus: Digital Photography Review This thread Refback 08-09-2012 03:14 AM
autofocus - When a camera auto-focuses, is the current focus distance reported from the lens to the camera? - Photography This thread Refback 07-02-2012 11:37 PM
Busted! The Myth of Open-loop Phase-detection Autofocus: Digital Photography Review This thread Refback 10-14-2011 03:14 AM
How to Buy Snowshoes This thread Trackback 02-01-2011 07:28 AM
Canon EOS focusing strategy - Open Photography Forums This thread Refback 07-08-2010 09:18 PM
Canon EOS focusing strategy - Open Photography Forums This thread Refback 07-01-2010 02:38 PM
Canon EOS focusing strategy - Open Photography Forums This thread Refback 06-30-2010 10:06 PM


All times are GMT -4. The time now is 09:45 AM.


Advertise Here! - Contact Us!


Content Relevant URLs by vBSEO 3.1.0