Reason for so many focus errors we see today?

John O'Flaherty wrote:
On Tue, 23 Jun 2009 08:56:14 -0500, Don Stauffer
wrote:

RichA wrote:
Plastic? Thermal expansion of plastic is much greater than metal and
it could very well be why we are seeing focus issues that need "lens
re-calibration" at service depots or that we see the need for in-
camera focus fine-tuning. Even cameras and lenses that appear to be
metal today may have plastic cells holding lenses, components in
cameras. The cameras are produced in a control temp environment but
that isn't real life use where temps can vary by 10's of degrees. I
don't remember all metal AF SLRs needing focus fine-tuning (or having
that facility) in the film days.

I can see that in open loop focusing, where you estimate the distance
and dial that distance on lens. However, in any closed loop operation
that source of error would not lead to a focus error. Also, there are a
some plastics that have a thermal expansion less than many metals. So
one cannot use generalities on this.

Focusing with a phase contrast system isn't closed loop with regard
to the picture sensor, since it uses a separate, simple sensor array
for focusing. Thus the focus system can get out of calibration.

I was under the impression that the OP was talking about lens cell
materials. The structure holding the image chip and the focusing chip
is in the body of the camera, so a movement of something within the lens
itself should affect both chips the same, shouldn't it?

On Wed, 24 Jun 2009 09:42:20 +0100, bugbear
wrote:

John O'Flaherty wrote:
On Tue, 23 Jun 2009 16:20:03 +0100, bugbear
wrote:

John O'Flaherty wrote:

Focusing with a phase contrast system isn't closed loop with regard
to the picture sensor, since it uses a separate, simple sensor array
for focusing. Thus the focus system can get out of calibration.
(jargon as per
http://en.wikipedia.org/wiki/Autofocus
)

Interesting - since contrast measurement can be done in software,
a camera with phase detection (which can, as you say, get out of
calibration) could use contrast measurement to self-calibrate.

That sounds like a good idea - you'd just have to take a special shot
of something with detail after each lens change.

Also, I wonder if it would be possible to make a main sensor chip with
special data paths to some small subset of the pixels, so they could
be digitized really fast, doing phase contrast

The article speaks of "phase detection" and "contrast measurement"; what do you mean
by "phase contrast", which sounds like a mixture?

Sorry, "phase contrast" is a term from microscopy which was floating
around in my head. I meant phase detection.
--
John

On Wed, 24 Jun 2009 09:04:14 -0500, Don Stauffer
wrote:

John O'Flaherty wrote:
On Tue, 23 Jun 2009 08:56:14 -0500, Don Stauffer
wrote:

RichA wrote:
Plastic? Thermal expansion of plastic is much greater than metal and
it could very well be why we are seeing focus issues that need "lens
re-calibration" at service depots or that we see the need for in-
camera focus fine-tuning. Even cameras and lenses that appear to be
metal today may have plastic cells holding lenses, components in
cameras. The cameras are produced in a control temp environment but
that isn't real life use where temps can vary by 10's of degrees. I
don't remember all metal AF SLRs needing focus fine-tuning (or having
that facility) in the film days.

I can see that in open loop focusing, where you estimate the distance
and dial that distance on lens. However, in any closed loop operation
that source of error would not lead to a focus error. Also, there are a
some plastics that have a thermal expansion less than many metals. So
one cannot use generalities on this.

Focusing with a phase contrast system isn't closed loop with regard
to the picture sensor, since it uses a separate, simple sensor array
for focusing. Thus the focus system can get out of calibration.

I was under the impression that the OP was talking about lens cell
materials. The structure holding the image chip and the focusing chip
is in the body of the camera, so a movement of something within the lens
itself should affect both chips the same, shouldn't it?

My impression of the o.p. was that it was about the camera in general,
anything that could require recalibration of focus. Anyway, I think a
phase system is open loop in general, since it moves the lens focus in
a direction and amount indicated by the comparison of two adjacent
sensor strips (at least for a first pass). So, maybe changes in the
lens itself will affect the accuracy.
--
John

In rec.photo.digital.slr-systems John O'Flaherty wrote:
On Wed, 24 Jun 2009 09:04:14 -0500, Don Stauffer
wrote:

John O'Flaherty wrote:
On Tue, 23 Jun 2009 08:56:14 -0500, Don Stauffer
wrote:

RichA wrote:
Plastic? Thermal expansion of plastic is much greater than metal and
it could very well be why we are seeing focus issues that need "lens
re-calibration" at service depots or that we see the need for in-
camera focus fine-tuning. Even cameras and lenses that appear to be
metal today may have plastic cells holding lenses, components in
cameras. The cameras are produced in a control temp environment but
that isn't real life use where temps can vary by 10's of degrees. I
don't remember all metal AF SLRs needing focus fine-tuning (or having
that facility) in the film days.

I can see that in open loop focusing, where you estimate the distance
and dial that distance on lens. However, in any closed loop operation
that source of error would not lead to a focus error. Also, there are a
some plastics that have a thermal expansion less than many metals. So
one cannot use generalities on this.

Focusing with a phase contrast system isn't closed loop with regard
to the picture sensor, since it uses a separate, simple sensor array
for focusing. Thus the focus system can get out of calibration.

I was under the impression that the OP was talking about lens cell
materials. The structure holding the image chip and the focusing chip
is in the body of the camera, so a movement of something within the lens
itself should affect both chips the same, shouldn't it?

My impression of the o.p. was that it was about the camera in general,
anything that could require recalibration of focus. Anyway, I think a
phase system is open loop in general, since it moves the lens focus in
a direction and amount indicated by the comparison of two adjacent
sensor strips (at least for a first pass). So, maybe changes in the
lens itself will affect the accuracy.

I think rather than open loop "in general" you mean "at first". It's
difficult getting at the technical details of phase detection
autofocus, since it's both highly technical, and at least in some
cases the details are regarded as a trade secret by the
manufacturer. However, having read the workshop repair manual for one
set of Canon lenses, which in passing mentioned various aspects of
autofocus as relevent to a service engineer, the impression I was left
with was that in those particular lenses, which were made for specially
fast focusing, what happens is as follows.

The AF phase detection makes a first measurement of the direction
and approximate distance to focus. The focus motor control then sets
up the parameters for the focus movement. If the distance is long
enough to warrant it this is a four phase movement.

The first phase is an acceleration of the lens up to its maximum speed
of focus travel. This is terminated automatically when the motor is
spinning at maximum speed, and the second phase begun. The second
phase travels at maximum speed until it gets to the deceleration
position, at which point it automatically starts the third phase, the
deceleration ramp down to the final focus approach speed. That final
fourth phase approach speed is slow enough to allow a closed loop
approach to focus detection by the AF sensor which given the inertia
of the lens etc.. can be stopped close enough to exact focus that
there will be no overshoot.

If the lens overshot the focus point it would have to
reverse. Reversal of direction is expensive in time, so like a car
approaching a stop light it decelerates at a speed at which it can
reliably stop close enough to the line. In the case of these lenses
"close enough" is a permissible error parameter previously read by the
camera from a table in ROM in the lens, and which is alterable by
service engineers in the lens focus calibration procedure. That
parameter may also be adjusted by the camera depending on the chosen
aperture. In that way smaller apertures can focus the lens a bit
faster. That parameter may also incorporate an early stop factor to
take account for the delays, inertia, etc. in the system, which mean
that it will always stop a little bit further on than where it was
when the software loop decided to stop the motor.

When the deceleration phase finishes the lens is now not quite at the
required focus point, and moving slowly enough to operate in closed
loop manner, closed on good enough focus detection by the AF
sensor. That'a the fourth final closed loop phase. It drives on at
that speed reading the sensor, and stops when it detects near enough
focus. Of course sometimes it won't detect focus, because the squirrel
may have moved away or the light changed enough to lose the necessary
contrast in the sensor while the lens was moving. So there is also a
timing watchdog to terminate this loop if it goes on for too
long. Other lenses/AF systems don't use a watchdog and simply keep
running the lens until it hits the end stop. Simpler, but slower.
Termination on the final focus approach loop on a focus find failure
condition will automatically initiate another complete focus from
scratch loop. This is how "hunting" happens, and whether it runs from
end to end of complete focus travel, or dithers about the focus point,
depends on whether the system uses timer watchdog or end stop failure,
and what kind of outer loop control system is operating.

Of course if the camera is doing a more complicated kind of focus such
as predictive focus of a moving object, or approaching a compromise
focus between the readings from a number of previously selected AF
sensors, a more complicated procedure is used. I'm describing here the
simplest long distance high speed AF procedure. If the focus point is
close to where the lens is of course it will only use the final
approach phase.

The manual also made it clear that not all Canon lenses were
sophisticated enough to use this sophisticated long distance high
speed focus method. Simpler modelds have only a single speed focus
system.

I also get the impression that there is another two speed AF system
used by some lens/body combinations, where the high speed checks the
AF sensor only every several motor steps, and when it gets near slows
down to checking every step.

I think it's an oversimplification to talk of THE phase detection AF
system of DSLRs, since in the search for highest speed and highest
accuracy the camera makers have kept improving the sophistication of
both their AF sensors and the focus control loops using them. In some
cases we've seen a camera maker offer a camera model software upgrade
which has improved focus performance not by fixing a bug but by
introducing a more sophisticated control system.

There are a number of different phase detection AF systems out there,
of different levels of sophistication and accuracy. Some, I suspect
these days at least most, always operate a closed loop approach of
monitoring the AF sensor during the final approach to the focus point.

Some camera makers have caused confusion on this question of whether
their DSLR AF systems are open or closed loop by describing them as
open loop because the closed loop they operate is closed around the AF
sensor and not the image sensor, whereas their compact cameras operate
a loop closed around the image sensor. In terms of control theory it's
technically correct to describe DSLR phase detection autofocus systems
as open loop, because they don't close the loop by measuring the
achievement of the purpose of the system, which is an image focused on
the image sensor. Instead they close the loop around the proxy of the
AF sensor. Hence all the focus calibration problems of DSLR autofocus.

Unfortunately that technically correct description of DSLR autofocus
as "open loop" without qualification has misled a lot of camera users
unfamiliar with the technicalities of control systems theory into
supposing that that "open loop" means that no DSLRs use the AF sensor
in a closed loop fashion.

--
Chris Malcolm

Chris Malcolm wrote:

I think rather than open loop "in general" you mean "at first". It's
difficult getting at the technical details of phase detection
autofocus, since it's both highly technical, and at least in some
cases the details are regarded as a trade secret by the
manufacturer. However, having read the workshop repair manual for one
set of Canon lenses, which in passing mentioned various aspects of
autofocus as relevent to a service engineer, the impression I was left
with was that in those particular lenses, which were made for specially
fast focusing, what happens is as follows.

The AF phase detection makes a first measurement of the direction
and approximate distance to focus. The focus motor control then sets
up the parameters for the focus movement. If the distance is long
enough to warrant it this is a four phase movement.

The first phase is an acceleration of the lens up to its maximum speed
of focus travel. This is terminated automatically when the motor is
spinning at maximum speed, and the second phase begun. The second
phase travels at maximum speed until it gets to the deceleration
position, at which point it automatically starts the third phase, the
deceleration ramp down to the final focus approach speed. That final
fourth phase approach speed is slow enough to allow a closed loop
approach to focus detection by the AF sensor which given the inertia
of the lens etc.. can be stopped close enough to exact focus that
there will be no overshoot.

If the lens overshot the focus point it would have to
reverse. Reversal of direction is expensive in time, so like a car
approaching a stop light it decelerates at a speed at which it can
reliably stop close enough to the line. In the case of these lenses
"close enough" is a permissible error parameter previously read by the
camera from a table in ROM in the lens, and which is alterable by
service engineers in the lens focus calibration procedure. That
parameter may also be adjusted by the camera depending on the chosen
aperture. In that way smaller apertures can focus the lens a bit
faster. That parameter may also incorporate an early stop factor to
take account for the delays, inertia, etc. in the system, which mean
that it will always stop a little bit further on than where it was
when the software loop decided to stop the motor.

When the deceleration phase finishes the lens is now not quite at the
required focus point, and moving slowly enough to operate in closed
loop manner, closed on good enough focus detection by the AF
sensor. That'a the fourth final closed loop phase. It drives on at
that speed reading the sensor, and stops when it detects near enough
focus. Of course sometimes it won't detect focus, because the squirrel
may have moved away or the light changed enough to lose the necessary
contrast in the sensor while the lens was moving. So there is also a
timing watchdog to terminate this loop if it goes on for too
long. Other lenses/AF systems don't use a watchdog and simply keep
running the lens until it hits the end stop. Simpler, but slower.
Termination on the final focus approach loop on a focus find failure
condition will automatically initiate another complete focus from
scratch loop. This is how "hunting" happens, and whether it runs from
end to end of complete focus travel, or dithers about the focus point,
depends on whether the system uses timer watchdog or end stop failure,
and what kind of outer loop control system is operating.

Of course if the camera is doing a more complicated kind of focus such
as predictive focus of a moving object, or approaching a compromise
focus between the readings from a number of previously selected AF
sensors, a more complicated procedure is used. I'm describing here the
simplest long distance high speed AF procedure. If the focus point is
close to where the lens is of course it will only use the final
approach phase.

The manual also made it clear that not all Canon lenses were
sophisticated enough to use this sophisticated long distance high
speed focus method. Simpler modelds have only a single speed focus
system.

I also get the impression that there is another two speed AF system
used by some lens/body combinations, where the high speed checks the
AF sensor only every several motor steps, and when it gets near slows
down to checking every step.

I think it's an oversimplification to talk of THE phase detection AF
system of DSLRs, since in the search for highest speed and highest
accuracy the camera makers have kept improving the sophistication of
both their AF sensors and the focus control loops using them. In some
cases we've seen a camera maker offer a camera model software upgrade
which has improved focus performance not by fixing a bug but by
introducing a more sophisticated control system.

There are a number of different phase detection AF systems out there,
of different levels of sophistication and accuracy. Some, I suspect
these days at least most, always operate a closed loop approach of
monitoring the AF sensor during the final approach to the focus point.

Some camera makers have caused confusion on this question of whether
their DSLR AF systems are open or closed loop by describing them as
open loop because the closed loop they operate is closed around the AF
sensor and not the image sensor, whereas their compact cameras operate
a loop closed around the image sensor. In terms of control theory it's
technically correct to describe DSLR phase detection autofocus systems
as open loop, because they don't close the loop by measuring the
achievement of the purpose of the system, which is an image focused on
the image sensor. Instead they close the loop around the proxy of the
AF sensor. Hence all the focus calibration problems of DSLR autofocus.

Unfortunately that technically correct description of DSLR autofocus
as "open loop" without qualification has misled a lot of camera users
unfamiliar with the technicalities of control systems theory into
supposing that that "open loop" means that no DSLRs use the AF sensor
in a closed loop fashion.

Thanks for that Chris. There's a lot of stuff there I didn't know.

Is that "workshop repair manual" available online? I have a copy of the "EF
50mm 1.8, 28mm 2.8, and 15mm 2.8 Service Manual", but AFAIK that doesn't go
into the operation like you described.

On 25 Jun 2009 09:40:09 GMT, Chris Malcolm
wrote:

In rec.photo.digital.slr-systems John O'Flaherty wrote:
On Wed, 24 Jun 2009 09:04:14 -0500, Don Stauffer
wrote:

John O'Flaherty wrote:
On Tue, 23 Jun 2009 08:56:14 -0500, Don Stauffer
wrote:

RichA wrote:
Plastic? Thermal expansion of plastic is much greater than metal and
it could very well be why we are seeing focus issues that need "lens
re-calibration" at service depots or that we see the need for in-
camera focus fine-tuning. Even cameras and lenses that appear to be
metal today may have plastic cells holding lenses, components in
cameras. The cameras are produced in a control temp environment but
that isn't real life use where temps can vary by 10's of degrees. I
don't remember all metal AF SLRs needing focus fine-tuning (or having
that facility) in the film days.

I can see that in open loop focusing, where you estimate the distance
and dial that distance on lens. However, in any closed loop operation
that source of error would not lead to a focus error. Also, there are a
some plastics that have a thermal expansion less than many metals. So
one cannot use generalities on this.

Focusing with a phase contrast system isn't closed loop with regard
to the picture sensor, since it uses a separate, simple sensor array
for focusing. Thus the focus system can get out of calibration.

I was under the impression that the OP was talking about lens cell
materials. The structure holding the image chip and the focusing chip
is in the body of the camera, so a movement of something within the lens
itself should affect both chips the same, shouldn't it?

My impression of the o.p. was that it was about the camera in general,
anything that could require recalibration of focus. Anyway, I think a
phase system is open loop in general, since it moves the lens focus in
a direction and amount indicated by the comparison of two adjacent
sensor strips (at least for a first pass). So, maybe changes in the
lens itself will affect the accuracy.

I think rather than open loop "in general" you mean "at first". It's
difficult getting at the technical details of phase detection
autofocus, since it's both highly technical, and at least in some
cases the details are regarded as a trade secret by the
manufacturer. However, having read the workshop repair manual for one
set of Canon lenses, which in passing mentioned various aspects of
autofocus as relevent to a service engineer, the impression I was left
with was that in those particular lenses, which were made for specially
fast focusing, what happens is as follows.

The AF phase detection makes a first measurement of the direction
and approximate distance to focus. The focus motor control then sets
up the parameters for the focus movement. If the distance is long
enough to warrant it this is a four phase movement.

The first phase is an acceleration of the lens up to its maximum speed
of focus travel. This is terminated automatically when the motor is
spinning at maximum speed, and the second phase begun. The second
phase travels at maximum speed until it gets to the deceleration
position, at which point it automatically starts the third phase, the
deceleration ramp down to the final focus approach speed. That final
fourth phase approach speed is slow enough to allow a closed loop
approach to focus detection by the AF sensor which given the inertia
of the lens etc.. can be stopped close enough to exact focus that
there will be no overshoot.

If the lens overshot the focus point it would have to
reverse. Reversal of direction is expensive in time, so like a car
approaching a stop light it decelerates at a speed at which it can
reliably stop close enough to the line. In the case of these lenses
"close enough" is a permissible error parameter previously read by the
camera from a table in ROM in the lens, and which is alterable by
service engineers in the lens focus calibration procedure. That
parameter may also be adjusted by the camera depending on the chosen
aperture. In that way smaller apertures can focus the lens a bit
faster. That parameter may also incorporate an early stop factor to
take account for the delays, inertia, etc. in the system, which mean
that it will always stop a little bit further on than where it was
when the software loop decided to stop the motor.

When the deceleration phase finishes the lens is now not quite at the
required focus point, and moving slowly enough to operate in closed
loop manner, closed on good enough focus detection by the AF
sensor. That'a the fourth final closed loop phase. It drives on at
that speed reading the sensor, and stops when it detects near enough
focus. Of course sometimes it won't detect focus, because the squirrel
may have moved away or the light changed enough to lose the necessary
contrast in the sensor while the lens was moving. So there is also a
timing watchdog to terminate this loop if it goes on for too
long. Other lenses/AF systems don't use a watchdog and simply keep
running the lens until it hits the end stop. Simpler, but slower.
Termination on the final focus approach loop on a focus find failure
condition will automatically initiate another complete focus from
scratch loop. This is how "hunting" happens, and whether it runs from
end to end of complete focus travel, or dithers about the focus point,
depends on whether the system uses timer watchdog or end stop failure,
and what kind of outer loop control system is operating.

Of course if the camera is doing a more complicated kind of focus such
as predictive focus of a moving object, or approaching a compromise
focus between the readings from a number of previously selected AF
sensors, a more complicated procedure is used. I'm describing here the
simplest long distance high speed AF procedure. If the focus point is
close to where the lens is of course it will only use the final
approach phase.

The manual also made it clear that not all Canon lenses were
sophisticated enough to use this sophisticated long distance high
speed focus method. Simpler modelds have only a single speed focus
system.

I also get the impression that there is another two speed AF system
used by some lens/body combinations, where the high speed checks the
AF sensor only every several motor steps, and when it gets near slows
down to checking every step.

I think it's an oversimplification to talk of THE phase detection AF
system of DSLRs, since in the search for highest speed and highest
accuracy the camera makers have kept improving the sophistication of
both their AF sensors and the focus control loops using them. In some
cases we've seen a camera maker offer a camera model software upgrade
which has improved focus performance not by fixing a bug but by
introducing a more sophisticated control system.

There are a number of different phase detection AF systems out there,
of different levels of sophistication and accuracy. Some, I suspect
these days at least most, always operate a closed loop approach of
monitoring the AF sensor during the final approach to the focus point.

Some camera makers have caused confusion on this question of whether
their DSLR AF systems are open or closed loop by describing them as
open loop because the closed loop they operate is closed around the AF
sensor and not the image sensor, whereas their compact cameras operate
a loop closed around the image sensor. In terms of control theory it's
technically correct to describe DSLR phase detection autofocus systems
as open loop, because they don't close the loop by measuring the
achievement of the purpose of the system, which is an image focused on
the image sensor. Instead they close the loop around the proxy of the
AF sensor. Hence all the focus calibration problems of DSLR autofocus.

Unfortunately that technically correct description of DSLR autofocus
as "open loop" without qualification has misled a lot of camera users
unfamiliar with the technicalities of control systems theory into
supposing that that "open loop" means that no DSLRs use the AF sensor
in a closed loop fashion.

That's a lot more detailed information than I've ever read before -
thanks for that. So, they are already decelerating short of the target
point, then closing the loop. Do you think it would be possible to get
quick end-point contrast readings from the image sensor itself, or
from small parts of it, to make a hybrid system that would have the
advantage of both speed and accuracy?
--
John

Chris Malcolm wrote:


I think rather than open loop "in general" you mean "at first". It's
difficult getting at the technical details of phase detection
autofocus, since it's both highly technical, and at least in some
cases the details are regarded as a trade secret by the
manufacturer. However, having read the workshop repair manual for one
set of Canon lenses, which in passing mentioned various aspects of
autofocus as relevent to a service engineer, the impression I was left
with was that in those particular lenses, which were made for specially
fast focusing, what happens is as follows.

There was a big patent fight between Honeywell and someone (I forget who
right at the moment. I followed the case because I worked for Honeywell
at the time. The lawyers even gave a noontime seminar one day for folks
interested in the case). It was over the details of that particular
autofocus method. The material disclosed in court is voluminous, but is
at least part of public record.

In article , Don Stauffer
wrote:

There was a big patent fight between Honeywell and someone (I forget who
right at the moment. I followed the case because I worked for Honeywell
at the time. The lawyers even gave a noontime seminar one day for folks
interested in the case). It was over the details of that particular
autofocus method. The material disclosed in court is voluminous, but is
at least part of public record.

minolta.

In article , Don Stauffer
writes

I was under the impression that the OP was talking about lens cell
materials.

No, the OP was continuing a decade long rant about something he has
little knowledge of but finds it a convenient subject to troll with.
--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's ****ed.
Python Philosophers (replace 'nospam' with 'kennedym' when replying)