From: Doug McDonald on
Wilba wrote:

>
> From previous tests for other purposes, I can say yes, it does exactly that.
> But the problem is, that behaviour is just as well explained by the "soft at
> f/1.8" theory - if the AF sensor doesn't have sharp fine detail to compare,
> I would expect it to show that small-scale variability.
>

Another test, which I did.

You take pictures using manual focus (at f/1.8) with the focus
indicator (the little dot at lower right in my 30D) as the criterion
of correct focus. Try to get it centered in the middle of
the "in focus is indicated" range. Make several tries. If this
gives better focus than real autofocus, something is wrong with the
"focus movement prediction" system.

My results say that the indicator is better than real autofocus,
but not by a lot. This is using only the center spot on text
7 feet away.

In focus this lens is quite sharp at f/1.8, in the center that is.

Doug McDonald
From: Wilba on
Doug McDonald wrote:
> Wilba wrote:
>>
>> From previous tests for other purposes, I can say yes, it does exactly
>> that. But the problem is, that behaviour is just as well explained by the
>> "soft at f/1.8" theory - if the AF sensor doesn't have sharp fine detail
>> to compare, I would expect it to show that small-scale variability.
>
> Another test, which I did.
>
> You take pictures using manual focus (at f/1.8) with the focus
> indicator (the little dot at lower right in my 30D) as the criterion
> of correct focus. Try to get it centered in the middle of
> the "in focus is indicated" range. Make several tries. If this
> gives better focus than real autofocus, something is wrong with the
> "focus movement prediction" system.
>
> My results say that the indicator is better than real autofocus,
> but not by a lot. This is using only the center spot on text
> 7 feet away.

Ah, this is exactly the kind of idea I need! That experiment attempts to
separate the detection (AF sensor) and actuation (lens motor) functions
within the system. I'll have a play with it today. Thanks a bunch.

> In focus this lens is quite sharp at f/1.8, in the center that is.

Not in my experience. If the lens was as sharp in the centre at f/1.8 as it
is at f/2.8, then I would see at f/1.8 a narrow band of decent sharpness
_somewhere_ in images of slanted subjects with fine textured detail, but I
don't, and not just in my own images.


From: Wilba on
Wilba wrote:
> Doug McDonald wrote:
>> Wilba wrote:
>>>
>>> From previous tests for other purposes, I can say yes, it does exactly
>>> that. But the problem is, that behaviour is just as well explained by
>>> the "soft at f/1.8" theory - if the AF sensor doesn't have sharp fine
>>> detail to compare, I would expect it to show that small-scale
>>> variability.
>>
>> Another test, which I did.
>>
>> You take pictures using manual focus (at f/1.8) with the focus
>> indicator (the little dot at lower right in my 30D) as the criterion
>> of correct focus. Try to get it centered in the middle of
>> the "in focus is indicated" range. Make several tries. If this
>> gives better focus than real autofocus, something is wrong with the
>> "focus movement prediction" system.
>>
>> My results say that the indicator is better than real autofocus,
>> but not by a lot. This is using only the center spot on text
>> 7 feet away.
>
> Ah, this is exactly the kind of idea I need! That experiment attempts to
> separate the detection (AF sensor) and actuation (lens motor) functions
> within the system. I'll have a play with it today. Thanks a bunch.
>
>> In focus this lens is quite sharp at f/1.8, in the center that is.
>
> Not in my experience. If the lens was as sharp in the centre at f/1.8 as
> it is at f/2.8, then I would see at f/1.8 a narrow band of decent
> sharpness _somewhere_ in images of slanted subjects with fine textured
> detail, but I don't, and not just in my own images.

Well, it sure has been an interesting day. :- )

Here are some preliminary results with a 45 degree target. I need to repeat
with a different target before I would state the following with confidence,
and it could be quite different for other units depending on their
calibration. But anyway...

Sharpness in the very centre at f/1.8 appeared fine, practically
indistinguishable from f/2.8 or f/4. In previous tests I might not have
looked hard enough at the very centre of the image, and may have been misled
by other sources of variation. So that's suggests a disproof of my "soft at
f/1.8" theory.

I found that my phase detect AF sensor has sidedness. If I start with the
lens focussed closer than the subject, the results are uniformly excellent,
whether autofocussing or manually focussing using the AF confirmation (as
Doug described above).

If I start with the lens focussed behind the subject, and I manually focus
using the AF confirmation, focus is always off by the same tiny amount (one
click towards infinity in the EOS Utility will bring it into optimal focus).

With initial focus behind the subject and PD autofocus, about seven shots
out of ten are out by the same one click as the manual focus, and the rest
are optimal, like when starting from the nearside. I assume that the good
ones come about from the lens overshooting and then the system corrects
towards infinity (so it ultimately approaches focus from the nearside).

With an external aperture which gives similar exposures to f/2.8 (with the
lens at f/1.8), focus is still excellent starting from the near side, and it
improves the performance when starting from the far side (but still not
optimal).

Those results, and the fact that the lens is able to focus perfectly via
contrast detect, suggest to me that the "crude and sloppy mechanism" theory
is bogus. I suspect the truth is about the performance of the PD AF sensor
with this lens, specifically how much latitude it has for confirming a
focus. I noticed when manually focussing with a macro rail that I could move
the camera quite some (micro) distance and still get a focus confirmation
beep.


From: Wilba on
Wilba wrote:
> Wilba wrote:
>> Doug McDonald wrote:
>>>
>>> You take pictures using manual focus (at f/1.8) with the focus
>>> indicator (the little dot at lower right in my 30D) as the criterion
>>> of correct focus. Try to get it centered in the middle of
>>> the "in focus is indicated" range. Make several tries. If this
>>> gives better focus than real autofocus, something is wrong with the
>>> "focus movement prediction" system.
>>>
>>> My results say that the indicator is better than real autofocus,
>>> but not by a lot. This is using only the center spot on text
>>> 7 feet away.
>>
>> Ah, this is exactly the kind of idea I need! That experiment attempts to
>> separate the detection (AF sensor) and actuation (lens motor) functions
>> within the system. I'll have a play with it today. Thanks a bunch.
>
> Well, it sure has been an interesting day. :- )
>
> Here are some preliminary results with a 45 degree target. I need to
> repeat with a different target before I would state the following with
> confidence, and it could be quite different for other units depending on
> their calibration. But anyway...
>
> Sharpness in the very centre at f/1.8 appeared fine, practically
> indistinguishable from f/2.8 or f/4. In previous tests I might not have
> looked hard enough at the very centre of the image, and may have been
> misled by other sources of variation. So that's suggests a disproof of my
> "soft at f/1.8" theory.
>
> I found that my phase detect AF sensor has sidedness. If I start with the
> lens focussed closer than the subject, the results are uniformly
> excellent, whether autofocussing or manually focussing using the AF
> confirmation (as Doug described above).
>
> If I start with the lens focussed behind the subject, and I manually focus
> using the AF confirmation, focus is always off by the same tiny amount
> (one click towards infinity in the EOS Utility will bring it into optimal
> focus).
>
> With initial focus behind the subject and PD autofocus, about seven shots
> out of ten are out by the same one click as the manual focus, and the rest
> are optimal, like when starting from the nearside. I assume that the good
> ones come about from the lens overshooting and then the system corrects
> towards infinity (so it ultimately approaches focus from the nearside).
>
> With an external aperture which gives similar exposures to f/2.8 (with the
> lens at f/1.8), focus is still excellent starting from the near side, and
> it improves the performance when starting from the far side (but still not
> optimal).
>
> Those results, and the fact that the lens is able to focus perfectly via
> contrast detect, suggest to me that the "crude and sloppy mechanism"
> theory is bogus. I suspect the truth is about the performance of the PD AF
> sensor with this lens, specifically how much latitude it has for
> confirming a focus. I noticed when manually focussing with a macro rail
> that I could move the camera quite some (micro) distance and still get a
> focus confirmation beep.

I've been thinking a lot about the width of the "beep band" and what it
means, so I measured it with a macro rail. With the lens at its closest
focus (430mm from the sensor), the far beep is 3.0mm from the near beep. A
conventional DOF calculation gives around 4.3mm, and a focus from the
nearside appears to put the subject bang in the middle of that. So that
means a farside focus would put the subject something like 0.8mm outside the
DOF, which sounds about right.

I also tried it at that distance with an EF-S 18-55 at about 50mm (f/5.6).
The beep band is 12mm, the DOF is 13mm, and where you start from makes very
little difference to the focus achieved (although from the nearside is best
again). I wish I had another wide-aperture FFL lens to compare.

Any thoughts about why the 50/1.8 figures make sense (farside focus... 0.8mm
outside the DOF), but it doesn't work like that for the 18-55?


From: David J Taylor on
"Wilba" <usenet(a)CUTTHISimago.com.au> wrote in message
news:00c4cc60$0$15581$c3e8da3(a)news.astraweb.com...
[]
> Any thoughts about why the 50/1.8 figures make sense (farside focus...
> 0.8mm outside the DOF), but it doesn't work like that for the 18-55?

As others have said, the focal point will shift with f/number, and the
focus sensors may only accept the smaller cone of rays rather than the
full f/1.8 cone.

Cheers,
David