Shocking insight: AF is not a closed feedback loop!

Started Jul 18, 2003 | Discussions
Larry H. Smith Veteran Member • Posts: 6,270
Not solved...

DavidP wrote:

Buy fast primes, but stop them down.

Problem solved.

No problem is 'solved' by a solution that requires that I NOT use the 'fast" part of the fast lens I've bought.

But time and techmology march-on...things will get better!

Larry

This whole "what the camera can/can't do FOR you" thing is the main
reason I won't be happy till I have a pro-level camera that will
let ME do what I want, with a viewfinder that CLEARLY shows me what
is happening.

OP Mishkin™ Contributing Member • Posts: 917
you always do impossible

e.g. shooting Wilkinsons at f/1.2 1/8000s ISO3200. This feat is not possible for us mortals

DavidP wrote:
What you're telling me is that I'm attempting to do the impossible.
You know that, right? LOL.

Mishkin2 wrote:

It's a typical recommendation coming from pros to use fast primes
and shoot available light without flash. Available light forces one
to open up the lens and misfocusing (within the spec!) can bring a
lot of disappointment to newbies who heard about primes being
incredibly sharp.

-- hide signature --

Mishkin

Steve Mason Senior Member • Posts: 1,204
Quit hogging all the questions! :)

DavidP wrote:

Sometimes a lens (noteably wide angle) on a brick wall test will
exhibit this "jumping" behavior in AF. Yet other times (without
moving the lens) it will just focus (no jumping).

?

OP Mishkin™ Contributing Member • Posts: 917
use and abuse digital... it's free!

with ultrafast apertures, rely on focus bracketing by firing a lot of frames while rocking your head or rotating the focus ring. With 10D's zoom review feature, it's easy to delete the badly blurred frames, so storage capacity won't suffer a lot, too.

Larry H. Smith wrote:

DavidP wrote:

Buy fast primes, but stop them down.

Problem solved.

No problem is 'solved' by a solution that requires that I NOT use
the 'fast" part of the fast lens I've bought.

But time and techmology march-on...things will get better!

Larry

This whole "what the camera can/can't do FOR you" thing is the main
reason I won't be happy till I have a pro-level camera that will
let ME do what I want, with a viewfinder that CLEARLY shows me what
is happening.

-- hide signature --

Mishkin

Pekka Saarinen Senior Member • Posts: 1,826
Re: Hang on...

Mishkin2 wrote:

yada10d wrote:

I would think the recomendation should be to get a fast prime and
stop it down, not put a possibly less capable lens on the camera to
'hide' the problem.

It's a typical recommendation coming from pros to use fast primes
and shoot available light without flash. Available light forces one
to open up the lens and misfocusing (within the spec!) can bring a
lot of disappointment to newbies who heard about primes being
incredibly sharp.

There is only one in-focus plane whether it's f1.4 or f22. It should and it will not make any difference what aperture you use when gear is ok. Misfocusing with 1.4 lens @ 1.4 and @ 22 is equal - it just does not show as easily with 22. This babble is of course academic but this must be cleared.

But what is true is f/1.4 lens is better for AF because or 1/DoF specs of 10D. AF is always done in open aperture by the camera.

-- hide signature --
OP Mishkin™ Contributing Member • Posts: 917
nope

if this were the case, the focusing will get infinitely more accurate as f-stop goes to zero (and technically speaking, f-stop much smaller than 1 is possible with parabolloid and a subject at infinity, just place AF sensor in its focus). This is not the case simply because it's size of the pixels in AF array that limits AF "resolution".

Once you reached this limit, CoC on sensor plane when AF sensor made an error of 1 AF pixel becomes inversely proportional to F-stop. I can prove this with formulas - later.

1 DOF spec of 10D means 100% DOF at f/2.8 and CoC=0.03mm. This accuracy is determined by AF sensor spatial resolution. This is what 10D promises at f/2.8 - majority of the pics will be tack sharp, some blurred by 1 pixels, some by 2, less by 3, and small number by 4 pixels. And it's all within spec, even if camera and lenses are perfectly calibrated.

This is the limit for 10D. Past this limit, larger (wider) apertures begin to progressively increase CoC's projected on sensor plane when it's slightly (but within the spec!) OOF.

E.g. at f/1.4 you can expect 0.06mm CoC in some photos and you can't complain - it's within the spec!

Solution? Focus bracketing, multiple shots.

Pekka Saarinen wrote:

But what is true is f/1.4 lens is better for AF because or 1/DoF
specs of 10D. AF is always done in open aperture by the camera.

-- hide signature --

Mishkin

yada10d Regular Member • Posts: 279
Re: nope

Mishkin2 wrote:

This is the limit for 10D. Past this limit, larger (wider)
apertures begin to progressively increase CoC's projected on sensor
plane when it's slightly (but within the spec!) OOF.

E.g. at f/1.4 you can expect 0.06mm CoC in some photos and you
can't complain - it's within the spec!

Solution? Focus bracketing, multiple shots.

Maybe, for you if you want to do that. Most will just stop down unless they NEED the DOF effect of the open aperture.

If they're newbies, even stopped down fast primes have way less DOF than their P&S digicam they had before the 10D.

Y.

Pekka Saarinen wrote:

But what is true is f/1.4 lens is better for AF because or 1/DoF
specs of 10D. AF is always done in open aperture by the camera.

-- hide signature --

Mishkin

Frank Nichols Senior Member • Posts: 1,815
hmm, AF sensor vs image sensor

1. Are you confusing the AF sensor resolution with the Sensor resolution?

2. If I follow what you are saying, then I would expect that error would be a function of subject distance and would appear a a "ripple" of error. In otherwords, what ever the stepping error is, it will repeat so that as you walk forward toward the subject, you will get more and more then less and less error. At some distances you should get "none".

3. Don't forget the "lens type specific database" in the 10D - hmm step functions?

4. This would also say non-mfr lens owners are SOL, but it doesn't explain why some 3rd party copies focus and others don't - ie some Tamron 28-75's focus and some don't.

I think there is still a bit more to this story. Otherwise, lens focus perfomance would be predictable.

Frank

Mishkin2 wrote:
if this were the case, the focusing will get infinitely more
accurate as f-stop goes to zero (and technically speaking, f-stop
much smaller than 1 is possible with parabolloid and a subject at
infinity, just place AF sensor in its focus). This is not the case
simply because it's size of the pixels in AF array that limits AF
"resolution".

Once you reached this limit, CoC on sensor plane when AF sensor
made an error of 1 AF pixel becomes inversely proportional to
F-stop. I can prove this with formulas - later.

1 DOF spec of 10D means 100% DOF at f/2.8 and CoC=0.03mm. This
accuracy is determined by AF sensor spatial resolution. This is
what 10D promises at f/2.8 - majority of the pics will be tack
sharp, some blurred by 1 pixels, some by 2, less by 3, and small
number by 4 pixels. And it's all within spec, even if camera and
lenses are perfectly calibrated.

This is the limit for 10D. Past this limit, larger (wider)
apertures begin to progressively increase CoC's projected on sensor
plane when it's slightly (but within the spec!) OOF.

E.g. at f/1.4 you can expect 0.06mm CoC in some photos and you
can't complain - it's within the spec!

Solution? Focus bracketing, multiple shots.

Pekka Saarinen wrote:

But what is true is f/1.4 lens is better for AF because or 1/DoF
specs of 10D. AF is always done in open aperture by the camera.

-- hide signature --

Someday I will take a good photograph - until then I will blame my equipment.

 Frank Nichols's gear list:Frank Nichols's gear list
Nikon 10-20mm F4.5-5.6 VR
OP Mishkin™ Contributing Member • Posts: 917
Re: hmm, AF sensor vs image sensor

Frank Nichols wrote:

1. Are you confusing the AF sensor resolution with the Sensor
resolution?

No. AF sensor doesn't know anything about what kind of sensor (or film - "what's film?") is in image plane. AF sensor resolution (pixels/mm) directly relates to the size of maximum CoC projected on sensor/film plane. For 10D's AF this is up to 0.03mm at f/2.8, for 1D/1Ds it's 0.01mm at f/2.8. This is the maximum diameter of each point point in the image blurred due to permissible [that is, to maximum ability of AF sensor] misfocus.

2. If I follow what you are saying, then I would expect that error
would be a function of subject distance and would appear a a
"ripple" of error. In otherwords, what ever the stepping error is,
it will repeat so that as you walk forward toward the subject, you
will get more and more then less and less error. At some distances
you should get "none".

Possible. But even if you're at such distance, a jump of 1 pixel in AF sensor is still possible intermittently (due to noise, imperfect matching of the signals from half-pairs, etc.) and that will cause jump in AF.

3. Don't forget the "lens type specific database" in the 10D - hmm
step functions?

Who knows...

4. This would also say non-mfr lens owners are SOL, but it doesn't
explain why some 3rd party copies focus and others don't - ie some
Tamron 28-75's focus and some don't.

I don't think so. As long as lens is made to report a matching signal about distance back to camera, it doesn't matter who is manufacturer.

I think there is still a bit more to this story. Otherwise, lens
focus perfomance would be predictable.

I think 99% is explained already. Thinking outside the box of closed feedback loop solved the puzzle why some lenses focus perfectly, yet some do not.

-- hide signature --

Mishkin

Bob Watt Contributing Member • Posts: 532
Look at the detailed schematic, focus control line missing

Miskin,

Look at the detailed schematic. All the lines with one
exception (going to a box called the "display") from the AF calculation
module near the bottom, are shown as inputs. To achieve control of the
lens focus there MUST be a control line going from
the AF calculator to the lens uprocessor. This missing
control line (or maybe the arrow is on the wrong end) is in
fact by necessity an error signal telling the lens AF motor
to drive + or - from the present position. It does look like
this feedback loop is a virtual one, coming from a uprocessor
after a comparison, and thus subject, for instance, to a mismatch
between the step size of the "drive quanity detector" and the
control signal, which could cause discrete jumps. The triangle
however is NOT a comparator, it is the "drive actuator", according
to the image in the writeup.

So while there may be discrete mismatch issues wrt size of
the error signal bits and the drive quantity detector bits, that the
lens uprocessor can then miss by + - one discrete unit (the bigger one),
the system is indeed a feedback loop. It must be. A one directional
system like you discribe ("step 20 units, ok all done, nothing more")
would be prone to the problem you are talking about, but can not
be the system really used. If you doubt that, then explain how the
lens hunts back and forth before settling down. A unidirectional
system would move and stop until you once again pulled the
trigger. Only a real time feedback loop (hardwire or virtual thru
a uprocessor) has the inherent lag that results in hunting. A
servo loop of this type will have a damping constant that may
be adjustable to allow over, under or critical damping during the
feedback process. Critical damping will cause it to slew to the point where
the error signal is 0 and stop. Under damping will overshoot and
hunt (one of the dread 28-135IS focus issues). If the damping constant
is out of adjustment, particularly if it is overdamped so it never quite gets
where it needs to be before the error signal becomes too small
to act on, the focus will never be quite right.

I do not think your conjecture holds up under scrutiny. I suggest
everybody visit this page and watch the replay and mouse around
in the image on the page to find what the boxes are
called (like the triangle which is called the drive actuator, i.e.
the USM motor).

Bob Watt

OP Mishkin™ Contributing Member • Posts: 917
doesn't change a thing

Yes, you're right, the triangle is USM motor. The comparator is within lens microprocessor (the white box with "Data communication control" label).

This doesn't change a thing in my conjecture.

A feedback loop doesn't need any data from the lens. The lens just needs to respond to the signal "move closer/move farther" from the body in this case. And this would achieve perfect focus anytime, with any lens (assuming body is perfectly calibrated).

Hunting is simply searching for better data when AF sensor doesn't have enough light or contrast.

-- hide signature --

Mishkin

MarkG Regular Member • Posts: 209
Aha! and why???

You are right, this does explain everything. In control theory this system is called "feed forward error correction". The advantage is that there are no stability problems, the disadvantage is that the error (misfocus amount) is not reducable as much as with negative feedback system. My guess is that due to variation in motor parameters of different lenses the decision has been made early on not to worry about stability problems, and rather calibrate lenses to work with feed forward system. Oh, well... It is worse than I thought. Thank you for digging this out. MarkG

Mishkin2 wrote:

Yes, you're right, the triangle is USM motor. The comparator is
within lens microprocessor (the white box with "Data communication
control" label).

This doesn't change a thing in my conjecture.

A feedback loop doesn't need any data from the lens. The lens just
needs to respond to the signal "move closer/move farther" from the
body in this case. And this would achieve perfect focus anytime,
with any lens (assuming body is perfectly calibrated).

Hunting is simply searching for better data when AF sensor doesn't
have enough light or contrast.

 MarkG's gear list:MarkG's gear list
Ricoh GR Canon EOS 20D Sony a7S Sony a6000 Sony E 35mm F1.8 OSS +5 more
OP Mishkin™ Contributing Member • Posts: 917
thanks for feeding me forward ;)

It's great to know all the technical terms. I may print this one "feed forward error correction" at 11x17, frame it and hang it on my wall ;))

Yes, it's how AF works. With closed feedback loop, there would be no front/back focusing lenses. They will all focus within the accuracy of AF sensor. Practice shows that this is not the case.

It's all calibration. Got misfocusing camera/lenses? Calibrate!

Too bad this calibration can only be done by Canon. I believe it's should be doable by user via software. Even without "reference lens" it would be possible to optimize your system (camera + all lenses) just by shooting a set of pictures with all lenses and analyzing the sharpness of final images under AF point.

MarkG wrote:
You are right, this does explain everything. In control theory this
system is called "feed forward error correction". The advantage is
that there are no stability problems, the disadvantage is that the
error (misfocus amount) is not reducable as much as with negative
feedback system. My guess is that due to variation in motor
parameters of different lenses the decision has been made early on
not to worry about stability problems, and rather calibrate lenses
to work with feed forward system. Oh, well... It is worse than I
thought. Thank you for digging this out. MarkG

Mishkin2 wrote:

Yes, you're right, the triangle is USM motor. The comparator is
within lens microprocessor (the white box with "Data communication
control" label).

This doesn't change a thing in my conjecture.

A feedback loop doesn't need any data from the lens. The lens just
needs to respond to the signal "move closer/move farther" from the
body in this case. And this would achieve perfect focus anytime,
with any lens (assuming body is perfectly calibrated).

Hunting is simply searching for better data when AF sensor doesn't
have enough light or contrast.

-- hide signature --

Mishkin

Joe Grass New Member • Posts: 19
I sent in 10D & 3 lenses (pics)

My 35mm F2 and 17-40mm F4 focus PERFECTLY, but when I later tried THREE copies of a 70-200mm F4, they all back focused.

So I kept one 70-200 F4, and sent it with all my lenses and 10D to Irvine last week.

I only noticed the problem on the 70-200 F4 when I got blurry photos, and started noticing the background was in focus. Then I tried shooting some focus charts, and there it was, a very consistent back focus problem.

Here are some test chart photos:

http://www.pbase.com/jgrass/focus

Got the test chart from Francis Poon (www.khdotcom.net). Thanks Francis!

I'm hoping Canon can adjust this all out, without messing the lenses that already focus correctly. I think I will also rent a different body (1DS? Elan 7?) and check out my lenses when they come back. I'm hoping they don't adjust the 70-200 F4 to match my 10D, and nothing else!

Thanks VERY much Mishkin for your very informative post(s), and I'll let you know how it all turns out for me.

-Joe

Mishkin2 wrote:

it's just very exhausting that I went through FIVE 24-70's and they
all can't consistently shoot sharp pictures at 24mm f/2.8 - with
camera which doesn't have any significant issues with my other 4
lenses.

I understand now that I need to send my camera and lenses to Canon,
but I'm afraid to get screwed up and then to beg for weeks for
special attention from Canon.

-- hide signature --
(unknown member) Veteran Member • Posts: 9,762
Re: Shocking insight: AF is not a closed feedback loop!

The real question is, "is this a one time issue or is the AF inherently flawed?" Please Keep in mind, I'm barely following you here and would probably need the slower paced class at first to keep up.

Mishkin2 wrote:

And THAT and only that can explain why some lenses consistently
front/back focus even if camera is well-calibrated.

Take a look at this illustration from Canon Camera Museum
( http://www.canon.com/camera-museum/tech/room/f_index.html ):

There is a device in the lens named "Drive quantity detector". As
far as I understand from the diagram, the signals from this
detector and from AF processor are sent to some digital comparator
(triangle).

Why would this detector be needed in closed feedback loop? In
closed loop, the AF processor will send the signal to the lens
until the image is perfectly in focus. There's no need to measure
lens displacement and compare it to the signal from AF processor.

Now let's turn our measurbating brains to U.S. Patent 4,078,171 by
Honeywell (thanks to tawen mei who pointed me to it).

http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO1&Sect2=HITOFF&d=PALL&p=1&u=/netahtml/srchnum.htm&r=1&f=G&l=50&s1=4078171.WKU.&OS=PN/4078171&RS=PN/4078171
For those having trouble with QuickTime, here's the screenshot of
the illustrations:

Here's an exerpt from the patent:
"FIG. 3 shows a digital automatic focus system which may be used in
photographic equipment such as a still or movie camera. The system
includes a digital focus module 50, which preferably is apparatus
like that shown in FIG. 1. The output of digital focus module 50 is
a first digital word which is used to control a position of a
primary optical means 52 such as the taking lens of a still or
movie camera. A lens position encoder 54 provides a second digital
word indicative of the position of optical means 52 with respect to
film 56. Digital comparator 58 compares the first and second
digital words. The position of optical element 52 is determined by
motive means 60, which is controlled by motive means control 62.
The output of digital comparator 58 is connected to motive means
control 62 so that the position of optical means 52 with respect to
film 56 is controlled by the comparison of the first and second
digital words.

In the system shown in FIG. 3, the first digital word from digital
focus module 50 represents the desired position of optical means
52. When the position of optical element 52, as represented by the
second digital word, is identical to the first digital word, the
system is in focus. In the case of a still camera, the motion of
optical element 52 is stopped at this point."

IOW, digital comparator (the triangle in Canon's diagram) compares
"the desired position" of lens (which is determined by AF processor
by measuring the shift between signals on AF array half-pairs as I
described in my 1st post) to the actual position of the lens as
reported by "Drive quantity detector" (or "Lens position encoder",
in Honeywell's terms). When they match, the lens stops.

Bingo!

No closed feedback loop. AF just tells the lens: move 10 units
closer and STOP. (what these units mean, is programmed into lens'
own processor; AF processor in camera doesn't have to be concerned
with these low-level lens-specific details). Actually, this
comparison of the signals can be performed several or many times
during focusing (to rectify the signal and make sure the first
command was not a mistake), but it's still NOT A CLOSED FEEDBACK
LOOP.

This explains everything. Some lenses have miscalibrated "Drive
quantity detector", that makes them stop not there where they
should (while camera&lens happy thinking the focus is perfect).

That also explains why some lenses jump so much when trying AF many
times on the same subject. Their "Drive quantity detector" does not
have steps fine enough to match the signal from AF processor
precisely and consistently. When two steps are equidistant from the
desired position, the lens will focus one time at one position,
another time at another position. That explains why several of my
24-70's are focusing so inconsistently especially in the 24-35mm
range. It's focusing group moves too little in this range to have
fine steps. A small miscalibration of my camera (not noticeable
with other 4 lenses, which have different optical construction)
might be "helping" this, too.

That also explains the words heard many times from Canon tech guys
that lenses can be "tightened up for 1.6x crop". They adjust the
"Drive quantity detector" digitizer to output finer steps.

It's all by design, folks. "Learn not to worry and love the Bomb."

Ari Riutto Contributing Member • Posts: 886
Also 16-35 L might be the one to be tightened (nt)

Mishkin2 wrote:

And THAT and only that can explain why some lenses consistently
front/back focus even if camera is well-calibrated.

Take a look at this illustration from Canon Camera Museum
( http://www.canon.com/camera-museum/tech/room/f_index.html ):

There is a device in the lens named "Drive quantity detector". As
far as I understand from the diagram, the signals from this
detector and from AF processor are sent to some digital comparator
(triangle).

Why would this detector be needed in closed feedback loop? In
closed loop, the AF processor will send the signal to the lens
until the image is perfectly in focus. There's no need to measure
lens displacement and compare it to the signal from AF processor.

Now let's turn our measurbating brains to U.S. Patent 4,078,171 by
Honeywell (thanks to tawen mei who pointed me to it).

http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO1&Sect2=HITOFF&d=PALL&p=1&u=/netahtml/srchnum.htm&r=1&f=G&l=50&s1=4078171.WKU.&OS=PN/4078171&RS=PN/4078171
For those having trouble with QuickTime, here's the screenshot of
the illustrations:

Here's an exerpt from the patent:
"FIG. 3 shows a digital automatic focus system which may be used in
photographic equipment such as a still or movie camera. The system
includes a digital focus module 50, which preferably is apparatus
like that shown in FIG. 1. The output of digital focus module 50 is
a first digital word which is used to control a position of a
primary optical means 52 such as the taking lens of a still or
movie camera. A lens position encoder 54 provides a second digital
word indicative of the position of optical means 52 with respect to
film 56. Digital comparator 58 compares the first and second
digital words. The position of optical element 52 is determined by
motive means 60, which is controlled by motive means control 62.
The output of digital comparator 58 is connected to motive means
control 62 so that the position of optical means 52 with respect to
film 56 is controlled by the comparison of the first and second
digital words.

In the system shown in FIG. 3, the first digital word from digital
focus module 50 represents the desired position of optical means
52. When the position of optical element 52, as represented by the
second digital word, is identical to the first digital word, the
system is in focus. In the case of a still camera, the motion of
optical element 52 is stopped at this point."

IOW, digital comparator (the triangle in Canon's diagram) compares
"the desired position" of lens (which is determined by AF processor
by measuring the shift between signals on AF array half-pairs as I
described in my 1st post) to the actual position of the lens as
reported by "Drive quantity detector" (or "Lens position encoder",
in Honeywell's terms). When they match, the lens stops.

Bingo!

No closed feedback loop. AF just tells the lens: move 10 units
closer and STOP. (what these units mean, is programmed into lens'
own processor; AF processor in camera doesn't have to be concerned
with these low-level lens-specific details). Actually, this
comparison of the signals can be performed several or many times
during focusing (to rectify the signal and make sure the first
command was not a mistake), but it's still NOT A CLOSED FEEDBACK
LOOP.

This explains everything. Some lenses have miscalibrated "Drive
quantity detector", that makes them stop not there where they
should (while camera&lens happy thinking the focus is perfect).

That also explains why some lenses jump so much when trying AF many
times on the same subject. Their "Drive quantity detector" does not
have steps fine enough to match the signal from AF processor
precisely and consistently. When two steps are equidistant from the
desired position, the lens will focus one time at one position,
another time at another position. That explains why several of my
24-70's are focusing so inconsistently especially in the 24-35mm
range. It's focusing group moves too little in this range to have
fine steps. A small miscalibration of my camera (not noticeable
with other 4 lenses, which have different optical construction)
might be "helping" this, too.

That also explains the words heard many times from Canon tech guys
that lenses can be "tightened up for 1.6x crop". They adjust the
"Drive quantity detector" digitizer to output finer steps.

It's all by design, folks. "Learn not to worry and love the Bomb."

-- hide signature --

-taking photos and flyfishing, that's all what matters -

 Ari Riutto's gear list:Ari Riutto's gear list
Canon EOS 5D Canon EOS-1D Mark II N Panasonic Lumix DMC-GH1 Panasonic Lumix DMC-GH2 Canon EF 14mm f/2.8L II USM +27 more
Bob Watt Contributing Member • Posts: 532
all true, but...

All true Mishkin,

1) There seems to be no signal from
the body to the lens in the schematic.
That can not be if the AF calculator is in
the body as shown.

2) As I read your post you conjectured the
system was open loop. I doubt that
despite the apparently missing signal
line returning to the lens uprocessor from
the AF calculator. Without that line the
lens has no signal telling it to move.
With that line, the system is not open loop.

One can still get the
effect you are talking about in a digital
feedback system if the least significant
error bit or detector precision bit is too large. This
can produce what you are talking about even with
a closed loop feedback system.

Bob Watt

Mishkin2 wrote:

Yes, you're right, the triangle is USM motor. The comparator is
within lens microprocessor (the white box with "Data communication
control" label).

This doesn't change a thing in my conjecture.

A feedback loop doesn't need any data from the lens. The lens just
needs to respond to the signal "move closer/move farther" from the
body in this case. And this would achieve perfect focus anytime,
with any lens (assuming body is perfectly calibrated).

Hunting is simply searching for better data when AF sensor doesn't
have enough light or contrast.

Bob Watt Contributing Member • Posts: 532
Re: Aha! and why???

Mark,

What in the diagram leads you to the
conclusion they used feed forward error
correction. Without a line from the body to
the lens telling it to move, which seems lacking in
the diagram, this is a "do not move" control
system. Clearly the diagram must be wrong.

All outputs and no inputs from the lens simply
can not work.

Nothing in the diagram or text at Canon suggest
feed forward or feedback. The subject is completely
undiscussed.

Bob Watt

MarkG wrote:
You are right, this does explain everything. In control theory this
system is called "feed forward error correction". The advantage is
that there are no stability problems, the disadvantage is that the
error (misfocus amount) is not reducable as much as with negative
feedback system. My guess is that due to variation in motor
parameters of different lenses the decision has been made early on
not to worry about stability problems, and rather calibrate lenses
to work with feed forward system. Oh, well... It is worse than I
thought. Thank you for digging this out. MarkG

Mishkin2 wrote:

Yes, you're right, the triangle is USM motor. The comparator is
within lens microprocessor (the white box with "Data communication
control" label).

This doesn't change a thing in my conjecture.

A feedback loop doesn't need any data from the lens. The lens just
needs to respond to the signal "move closer/move farther" from the
body in this case. And this would achieve perfect focus anytime,
with any lens (assuming body is perfectly calibrated).

Hunting is simply searching for better data when AF sensor doesn't
have enough light or contrast.

OP Mishkin™ Contributing Member • Posts: 917
Re: Aha! and why???

The arrow from electronic mount to camera microprocessor should be bi-directional. It is an error in diagram.

Bob Watt wrote:
Mark,

What in the diagram leads you to the
conclusion they used feed forward error
correction. Without a line from the body to
the lens telling it to move, which seems lacking in
the diagram, this is a "do not move" control
system. Clearly the diagram must be wrong.

All outputs and no inputs from the lens simply
can not work.

Nothing in the diagram or text at Canon suggest
feed forward or feedback. The subject is completely
undiscussed.

Bob Watt

MarkG wrote:
You are right, this does explain everything. In control theory this
system is called "feed forward error correction". The advantage is
that there are no stability problems, the disadvantage is that the
error (misfocus amount) is not reducable as much as with negative
feedback system. My guess is that due to variation in motor
parameters of different lenses the decision has been made early on
not to worry about stability problems, and rather calibrate lenses
to work with feed forward system. Oh, well... It is worse than I
thought. Thank you for digging this out. MarkG

Mishkin2 wrote:

Yes, you're right, the triangle is USM motor. The comparator is
within lens microprocessor (the white box with "Data communication
control" label).

This doesn't change a thing in my conjecture.

A feedback loop doesn't need any data from the lens. The lens just
needs to respond to the signal "move closer/move farther" from the
body in this case. And this would achieve perfect focus anytime,
with any lens (assuming body is perfectly calibrated).

Hunting is simply searching for better data when AF sensor doesn't
have enough light or contrast.

-- hide signature --

Mishkin

Bob Watt Contributing Member • Posts: 532
Re: Aha! and why???

Mishkin2 wrote:

The arrow from electronic mount to camera microprocessor should be
bi-directional. It is an error in diagram.

That much we can agree on. Now, without further
information is not possible to determine, open/closed
loop or positive or negative feedback. But the
digital nature of the communication can still
lead to digitizer step size errors causing an erratic
jump + - of the largest steps involved.

Bob Watt

Keyboard shortcuts:
FForum MMy threads