From: Wilba on
Paul Furman wrote:
> Wilba wrote:
>> Paul Furman wrote:
>>> Chris Malcolm wrote:
>>>>
>>>> "Back and front focus errors are common with very fast lenses (and
>>>> others). The AF sensors in all DSLRs operate at effective apertures
>>>> between f/2.8 and f/7.1, and 'see' the focus point as if the lens was
>>>> stopped down to the exact virtual f-stop the sensor in use
>>>> imposes. When spherical aberration causes a shift in focus on
>>>> stop-down, both wider and smaller apertures will no longer focus on
>>>> the targeted plane. This is corrected in camera calibration, general
>>>> or lens-specific."
>>>>
>>>> This quote comes from the section "Focus Accuracy" in this article in
>>>> the British Journal of Photography.
>>>>
>>>> http://www.bjp-online.com/public/showPage.html?page=837772
>>>
>>> That makes sense. The AF system only sees things at the designed
>>> aperture.
>>
>> I can see we're not going to agree on this, but I'd say it like, the AF
>> system only sees things through an effective aperture, whose size is a
>> consequence of the design of elements such as the prisms.
>
> OK. The important thing is the difference between an f/5.6 and f/2.8 AF
> setup and that's what he calls the 'baseine'. It's just a matter of
> geometry being set in the angle of the prisms so that the rays actually
> make it to the AF sensor. In my home made example I suppose that would be
> the separation between the holes? Apparently there are focusing screens
> for various focal lengths and speeds with different prism angles.
>
> page 17
> http://doug.kerr.home.att.net/pumpkin/Split_Prism.pdf
> "Autofocus precision
> There will always be some uncertainty in the determination of the relative
> positions of the two images, leading ultimately to a residual focus error
> after the AF system has made the correction indicated by the detector
> system. Because of geometric considerations, the greater the separation of
> the two virtual AF apertures in the exit pupil, the greater will be the
> displacement of the two images for a given focusing error; conversely, the
> greater that separation, the smaller will be the residual error for any
> given uncertainty in determination of the relative positions of the two
> images. The distance of this separation between the two virtual AF
> apertures is often spoken of as the �baseline� of the detector system, by
> analogy with the somewhat-corresponding parameter of a rangefinder.
>
> ...the focus detectors are usually designed to have virtual AF apertures
> near the edges of the exit pupil for a modest camera aperture�an aperture
> likely to be available on most lenses to be used on the camera. Of course,
> this compromise (a �reduced baseline�) results in a reduction in the
> precision of the focus detection system."
>
> Few lenses are slower than f/5.6... quite a few are that slow at the long
> end. So they have to accommodate that. F/3.5 lenses will be a little off
> wide open and the camera won't make full use of the wider aperture in
> focusing.
>
> Then you add a precision f/2.8 AF system on top of that and it absolutely
> must accommodate f/2.8 lenses because there's a bunch of pro zooms with
> that exact spec. Faster lenses don't add to the accuracy of that system,
> in fact they will tend to front-focus wide open, and back-focus shut down
> from this designed aperture baseline of f/2.8.

Yes. :- )

> I think that explains what you've experienced and the focus shift dude's
> free intro - using the logic in the split-prism article. What do you
> think?

When I read all that and consider everything else we've talked about, I
can't find anything whatsoever that can possibly explain my results. If you
can, you're going to have to put the dots a lot closer together for me.
:- )

Let me set it up for you. Let's keep it as simple as possible and consider
just three data points. Using the beep test (focus by changing the camera to
subject distance with fixed lens focus) -

(1) if the initial focus is on the near side of the subject, as the camera
to subject distance reduces focus is first confirmed at Xmm between the
sensor and the subject,

(2) if the initial focus is on the far side of the subject, as the camera to
subject distance increases focus first is confirmed at X+Ymm between the
sensor and the subject (Y is greater than 1/2 the DOF),

(3) if the initial focus has the subject within that Ymm "beep band", focus
is immediately confirmed.

(Note the cross-over in those distances - you would probably expect that for
point 2, focus would be confirmed at X-Ymm. Don't go further until you have
your head around that.)

Let's assume that there is an uncorrected error due to the mismatch between
the lens maximum aperture (f/1.8) and the effective aperture of the AF
system ("f/2.8"), and that the lens is not stopped down for any exposures
required for the explanation.

How do the ideas we've been discussing account for those specific
observations?

> I'm pretty sure I've nailed it here.

I doubt it. :- ) It's clear to me since Chris finally defined his
"effective aperture" (the "f/2.8" required exit pupil one) as distinct from
Kerr's virtual AF aperture, that we (all) have frequently been talking about
one when the listener is thinking about the other. It's pretty safe to
assume that most of what I've said so far has been referring to the virtual
AF aperture, unless I specifically identified the other. I'm happy to
continue using those terms to for the two distinct concepts.

> The angle of the prism sets the working aperture of the AF system.

Is that "effective aperture" or "virtual AF aperture"? :-D

There's a lot more to it than that, but the specific point I was making was
that the virtual AF aperture is not AFAICT primarily determined by stops in
the path, and for that matter, neither is the effective aperture. And I've
made that point often enough to shut up about it now. :- )

> A more sophisticated system might have an adjustable baseline. That would
> require changing the prism angles with mirrors or sliding holes or
> something.

Yep.


From: Wilba on
Paul Furman wrote:
> Wilba wrote:
>> Paul Furman wrote:
>>> Wilba wrote:
>>>> Paul Furman wrote:
>>>>> Wilba wrote:
>>>>>> Paul Furman wrote:
>>>>>>> Wilba wrote:
>>>>>>>>
>>>>>>>> What do you see in this arrangement, back-focus or front-focus?
>>>>>>>>
>>>>>>>> Camera Subject Plane of focus
>>>>>>>
>>>>>>> Back focus.
>>>>>>>
>>>>>>>> And therefore this is...?
>>>>>>>>
>>>>>>>> Camera Plane of focus Subject
>>>>>>>
>>>>>>> Front focus.
>>>>>>
>>>>>> Excellent. So here we go.
>>>>>>
>>>>>> C = Camera,
>>>>>> POF = Plane of Focus,
>>>>>> DOF = Depth of Field,
>>>>>> S = Subject.
>>>>>>
>>>>>> When the plane of focus is a long way (i.e. much greater than the
>>>>>> DOF, >>DOF) in front of the subject -
>>>>>>
>>>>>> C POF<------ >>DOF ------>S
>>>>>>
>>>>>> I call that "near focus". You can think of it as initial gross
>>>>>> front-focus.
>>>>>>
>>>>>> If I start like that, autofocus and beep focus both put the plane of
>>>>>> focus coincident with the subject. Lovely.
>>>>>>
>>>>>> When the plane of focus is a long way behind the subject -
>>>>>>
>>>>>> C S<------ >>DOF ------>POF
>>>>>>
>>>>>> I call that "far focus". You can think of it as initial gross
>>>>>> back-focus.
>>>>>>
>>>>>> If I start like that, autofocus and beep focus both put the plane of
>>>>>> focus just outside the DOF (>DOF/2) on the _front_ side of the
>>>>>> subject -
>>>>>>
>>>>>> C POF<-- >DOF/2 -->S
>>>>>>
>>>>>> Summary - initial gross front-focus results in optimal focus, and
>>>>>> initial gross back-focus results in _front-focus_.
>>>>>>
>>>>>> This outcome -
>>>>>>
>>>>>> C S POF
>>>>>>
>>>>>> never occurs with my gear in my tests using autofocus or beep-focus.
>>>>>>
>>>>>> Let me know if that doesn't make sense. :- )
>>>>>
>>>>> That makes sense but doesn't match your previous diagram:
>>>>
>>>> Yeah, but that's _your_ diagram, which is wrong. :- )
>>>
>>> Well hell.
>>> Maybe it is focus shift then.
>>
>> No! No! No! No! :- )
>>
>> It's all about the beep band (see below). From one direction focus is
>> confirmed in the right place, and from the other it's confirmed in the
>> wrong place. Where is the focus shift in that?
>
> Could be miscalibrated.
> That's where my averaging idea has relevance.

Okay, let's assume you're right, that there is a miscalibration which
results in a consistent offset in the focus (I've accommodated that
possibility in my other reply today). How does what we've been talking about
account for the difference between the two focuses? One is consistently
_here_, and other is consistently _there_. That's the significant thing. How
do you account for the width of the beep band?

>>>> Here's mine again and we'll go through it line by line. Constant width
>>>> font required for scale. Disregard any previous misunderstandings -
>>>> these are the results. :- )
>>>>
>>>> --------------------------------------
>>>>
>>>> Beep Test (focus by changing the sensor to subject distance until the
>>>> system beeps to confirm focus)
>>>>
>>>> Sensor to plane of focus distance -
>>>> |< 430mm >|
>>>> (The lens remains focussed at 430mm through the beep test.)
>>>>
>>>> Theoretical DOF -
>>>> | |< 2.2mm >|< 2.2mm >|
>>>> (The calculated DOF goes from 427.8 to 432.2mm.)
>>>>
>>>> Sensor to subject distance you get with an initial near-side focus -
>>>> |< 430mm >|
>>>> (Start with the subject 440mm from the sensor and decrease that
>>>> distance. When the system beeps the subject is perfectly in focus.)
>>>>
>>>> Sensor to subject distance you get with an initial far-side focus -
>>>> |< 433mm >|
>>>> (Start with the subject 420mm from the sensor and increase that
>>>> distance. When the system beeps the subject is 3mm further away than
>>>> the plane of focus, i.e. front-focus.)
>>>>
>>>> The "beep band" -
>>>> | |< 3.0mm >|
>>>> (The system will confirm focus over this range of sensor to subject
>>>> distances, for a plane of focus 430mm from the sensor.)
>>>>
>>>> Outside the DOF -
>>>> | |<0.8mm>|
>>>> (In the initial far-side focus case, the subject ends up 0.8mm further
>>>> from the sensor than the far limit of the theoretical DOF.)
>>>>
>>>> --------------------------------------
>>>>
>>>> AF Test
>>>>
>>>> Sensor to subject distance -
>>>> |< 430mm >|
>>>> (The sensor to subject distance remains the same throughout the AF
>>>> test.)
>>>>
>>>> Sensor to plane of focus distance you get with an initial near-side
>>>> focus -
>>>> |< 430mm >|
>>>> (Start with the lens focussed at 420mm from the sensor and invoke AF.
>>>> When the system beeps the subject is perfectly in focus.)
>>>>
>>>> Sensor to plane of focus distance you get with an initial far-side
>>>> focus -
>>>> |< 427mm >|
>>>> (Start with the lens focussed at 440mm from the sensor and invoke AF.
>>>> When the system beeps the plane of focus is 3mm short of the subject,
>>>> i.e. front-focus.)
>>>>
>>>> Theoretical DOF for an initial far-side focus -
>>>> | |< 2.2mm >|< 2.2mm >|
>>>> (The calculated DOF goes from 424.8 to 429.2mm.)
>>>>
>>>> The "beep band" -
>>>> | |< 3.0mm >|
>>>> (The system will confirm focus over this range of sensor to plane of
>>>> focus distances, for a subject 430mm from the sensor.)
>>>>
>>>> Outside the DOF -
>>>> | |<0.8mm>|
>>>> (In the initial far-side focus case, the subject ends up 0.8mm further
>>>> from the sensor than the far limit of the theoretical DOF.)


From: Wilba on
Chris Malcolm wrote:
> Wilba wrote:
>> Chris Malcolm wrote:
>>>
>>> "Back and front focus errors are common with very fast lenses (and
>>> others). The AF sensors in all DSLRs operate at effective apertures
>>> between f/2.8 and f/7.1, and 'see' the focus point as if the lens was
>>> stopped down to the exact virtual f-stop the sensor in use
>>> imposes. When spherical aberration causes a shift in focus on
>>> stop-down, both wider and smaller apertures will no longer focus on
>>> the targeted plane. This is corrected in camera calibration, general
>>> or lens-specific."
>>>
>>> This quote comes from the section "Focus Accuracy" in this article in
>>> the British Journal of Photography.
>>>
>>> http://www.bjp-online.com/public/showPage.html?page=837772
>>
>> Excellent! So, now that you have some corroboration for that idea,
>> please remind me what you're claiming as a consequence of it.
>
> In the quote above it says that there are misfocusing problems at
> effective apertures different from the effective focus aperture of the
> AF sensors, defined by the width between them, and that these are
> corrected by calibration. But that calibration can never be
> perfect. When focus is not highly critical that doesn't matter because
> the residual uncorrected imperfections are swallowed in the DoF. But
> when focus is highly critical with very shallow DoF, such as wide
> apertures on high quality lenses, the inevitable imperfection of
> general lens and camera calibrations matters. That's why the very top
> end DSLRs, even though they have the very best AF systems,
> nevertheless allow the user to trim the calibration for each lens, and
> those users who have used it report that it gives useful improvements
> in focus accuracy.
>
> Of course if your camera has been calibrated for best focus with one
> specific lens, you're happy with the results, and you will never
> acquire another critical focus lens, then you won't ever need to know
> any of this. But since we're having this discussion in public, and
> some of our audience does have more than one lens of critical focus,
> it's worth mentioning these general issues.

That's it?! Okay, since none of that has ever been contested, and it's never
been shown to be relevant to my test results, unless you have something that
is, you can retire from the discussion satisfied that you have made your
point. Thank you. :- )


From: Chris Malcolm on
Wilba <usenet(a)cutthisimago.com.au> wrote:
> Chris Malcolm wrote:

>> Since the lens aberrations which affect focus increase with the width
>> of a lens (focal length being constant), an uncorrected autofocus will
>> only perfectly focus an image taken at an aperture which has the same
>> width as the distance between the two AF sampling "holes". In fact
>> because exact focus of a lens is a best compromise between the various
>> focus affecting imperfections, an uncorrected phase detection AF
>> system will only accurately focus images taken at a slightly larger
>> aperture than the aperture defined by the distance between the two AF
>> sensor apertures, probably around one stop larger.
>>
>> That's why the "effective AF aperture" in the sense of the aperture
>> defined by the width apart on the lens of the two AF samples is an
>> important parameter in knowing how an AF system will focus different
>> lenses.

> So that's your definition of "effective AF aperture"? That would have been
> handy two weeks ago when you first started talking about it. :- )

I explained it in the same terms over a week ago. It would have been
handy if you'd paid attention :-)

>> Of course if you're only concerned with the behaviour of a DSLR which
>> has been optimised for best autofocus with one specific lens you can
>> ignore all this. Those of us with more than one lens with critical
>> focus can't ignore it.

> What do you do about it?

Assuming your camera can't be adjusted for each lens individually,
since you can't get them all to autofocus accurately, you have to
decide which are the most important to get right, and if there's more
than one, how to best compromise between them. Then you get the camera
adjusted to the specific lens or compromise you have chosen, and when
necessary use manual adjustmants of one kind or another to cope with
the others when in conditions you know will be wrongly focused.

In my case I adjusted the camera for perfect autofocus with the 500mm
f8, which gave me perfect focus with everything else except apertures
wider than f2.8 on my 50mm. So when I need apertures wider than f2.8
on the 50mm I either focus it manually or lock autofocus and move
forward the appropriate centimetre or so.

Doing that has improved the sharpness of focus on the 500mm and 50mm
by enough to raise the maximum size of sharp print by at least 40%.

--
Chris Malcolm
From: Chris Malcolm on
Wilba <usenet(a)cutthisimago.com.au> wrote:
> Chris Malcolm wrote:
>> Wilba wrote:
>>> Chris Malcolm wrote:
>>>>
>>>> "Back and front focus errors are common with very fast lenses (and
>>>> others). The AF sensors in all DSLRs operate at effective apertures
>>>> between f/2.8 and f/7.1, and 'see' the focus point as if the lens was
>>>> stopped down to the exact virtual f-stop the sensor in use
>>>> imposes. When spherical aberration causes a shift in focus on
>>>> stop-down, both wider and smaller apertures will no longer focus on
>>>> the targeted plane. This is corrected in camera calibration, general
>>>> or lens-specific."
>>>>
>>>> This quote comes from the section "Focus Accuracy" in this article in
>>>> the British Journal of Photography.
>>>>
>>>> http://www.bjp-online.com/public/showPage.html?page=837772
>>>
>>> Excellent! So, now that you have some corroboration for that idea,
>>> please remind me what you're claiming as a consequence of it.
>>
>> In the quote above it says that there are misfocusing problems at
>> effective apertures different from the effective focus aperture of the
>> AF sensors, defined by the width between them, and that these are
>> corrected by calibration. But that calibration can never be
>> perfect. When focus is not highly critical that doesn't matter because
>> the residual uncorrected imperfections are swallowed in the DoF. But
>> when focus is highly critical with very shallow DoF, such as wide
>> apertures on high quality lenses, the inevitable imperfection of
>> general lens and camera calibrations matters. That's why the very top
>> end DSLRs, even though they have the very best AF systems,
>> nevertheless allow the user to trim the calibration for each lens, and
>> those users who have used it report that it gives useful improvements
>> in focus accuracy.
>>
>> Of course if your camera has been calibrated for best focus with one
>> specific lens, you're happy with the results, and you will never
>> acquire another critical focus lens, then you won't ever need to know
>> any of this. But since we're having this discussion in public, and
>> some of our audience does have more than one lens of critical focus,
>> it's worth mentioning these general issues.

> That's it?! Okay, since none of that has ever been contested, and it's never
> been shown to be relevant to my test results, unless you have something that
> is, you can retire from the discussion satisfied that you have made your
> point. Thank you. :- )

Since your test results remain by your own admission completely
inexplicable it can't be claimed that none of that will turn out to be
relevant when the explanation is found :-)

--
Chris Malcolm