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

>>>> Why would you expect fuzziness in a hysteresis band? I've seen
>>>> plenty of systems with hysteresis with very precise boundaries to
>>>> the hysteresis band.
>>>
>>> I don't, but David seems to, unless I'm reading him wrong.
>>
>> I don't. I don't think I have ever used the word "fuzziness".

> But you did say, "you will likely stop at a different position, always
> within that dead-band", which doesn't happen with this system (it always
> stops at the ends of the beep band).

>> However, given that there is noise in any electronic or optical system,
>> and given that there will be tolerances in any mechanical system, I would
>> not be surprised if given a fixed starting point, the AF system did not
>> stop at /precisely/ the same finishing point every time.

> Correct. Even CDAF has shot-to-shot variation. PDAF is a touch more
> variable. But we're not talking about micrometric precision, just not
> focussing between the end points of the beep band. The beep test shots
> match, so it seems that PDAF triggers focus confirmation on the transition,
> on entering or leaving the beep band.

Then my guess is that this is the programmed behaviour of that
camera's AF control algorithm. As someone suggested earlier, while not
the best behaviour for accuracy, it's probably the best behaviour for
speed, and most people don't seem to want more AF accuracy if it slows
things down.

Have there been any updates to the camera's OS? Does the camera maker
boast of superior AF in more expensive or later models in terms which
give any clues? Have any details (or educated speculations) been
published in technical forums about how that camera's AF is
programmed?

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

>>> David J Taylor wrote:

>>> But you did say, "you will likely stop at a different position, always
>>> within that dead-band", which doesn't happen with this system (it always
>>> stops at the ends of the beep band).
>>
>> So your system doesn't behave in the way which I would consider "likely".
>> Maybe the software is very clever and works out the absolute minimum
>> distance to move the focus to be /just/ within that dead band? Without
>> the design details it's difficult to be sure.
>>
>>>> However, given that there is noise in any electronic or optical system,
>>>> and given that there will be tolerances in any mechanical system, I
>>>> would not be surprised if given a fixed starting point, the AF system
>>>> did not stop at /precisely/ the same finishing point every time.
>>>
>>> Correct. Even CDAF has shot-to-shot variation. PDAF is a touch more
>>> variable. But we're not talking about micrometric precision, just not
>>> focussing between the end points of the beep band. The beep test shots
>>> match, so it seems that PDAF triggers focus confirmation on the
>>> transition, on entering or leaving the beep band.
>>
>> Consistent with moving the minimum possible amount - i.e. getting the
>> fastest AF.

> Except it does the exact opposite - it makes front-focus out of gross
> back-focus.

That sounds like more movement, which sounds like it wouldn't be the
fastest AF, but if it's the result of an estimated dead reckoning
movement from a distance which involves that kind of error, then
improving it would involve another iteration through the focus
measurement process, which would slow the AF down. So that behaviour
could still be the result of going for AF speed.

AF speed seems to be what most people want, and the camera makers seem
to prefer to offer manual focus, or manual focus trimming, for best
focus precision, rather than offering two modes of AF, one approximate
and fast, one precise and slow.

--
Chris Malcolm
From: David J Taylor on
"Chris Malcolm" <cam(a)holyrood.ed.ac.uk> wrote in message
news:7s2o8tFaqaU2(a)mid.individual.net...
[]
> AF speed seems to be what most people want, and the camera makers seem
> to prefer to offer manual focus, or manual focus trimming, for best
> focus precision, rather than offering two modes of AF, one approximate
> and fast, one precise and slow.
>
> --
> Chris Malcolm

It's some time ago now, but with the Nikon 990 one thing which happened
when you switched to macro mode was not only an extension of the close
focus range, but a slowing down of the auto-focus speed. Mind you, this
is on a compact camera with contrast-detect AF.

Cheers,
David

From: Wilba on
Chris Malcolm wrote:
> Wilba wrote:
>> David J Taylor wrote:
>>> Wilba wrote:
>>>> Chris Malcolm wrote:
>>>>>
>>>>> Why would you expect fuzziness in a hysteresis band? I've seen
>>>>> plenty of systems with hysteresis with very precise boundaries to
>>>>> the hysteresis band.
>>>>
>>>> I don't, but David seems to, unless I'm reading him wrong.
>>>
>>> I don't. I don't think I have ever used the word "fuzziness".
>>
>> But you did say, "you will likely stop at a different position, always
>> within that dead-band", which doesn't happen with this system
>> (it always stops at the ends of the beep band).
>>
>>> However, given that there is noise in any electronic or optical system,
>>> and given that there will be tolerances in any mechanical system, I
>>> would
>>> not be surprised if given a fixed starting point, the AF system did not
>>> stop at /precisely/ the same finishing point every time.
>>
>> Correct. Even CDAF has shot-to-shot variation. PDAF is a touch more
>> variable. But we're not talking about micrometric precision, just not
>> focussing between the end points of the beep band. The beep test shots
>> match, so it seems that PDAF triggers focus confirmation on the
>> transition,
>> on entering or leaving the beep band.
>
> Then my guess is that this is the programmed behaviour of that
> camera's AF control algorithm. As someone suggested earlier, while not
> the best behaviour for accuracy, it's probably the best behaviour for
> speed, and most people don't seem to want more AF accuracy if it slows
> things down.

Yes.

> Have there been any updates to the camera's OS? Does the camera maker
> boast of superior AF in more expensive or later models in terms which
> give any clues? Have any details (or educated speculations) been
> published in technical forums about how that camera's AF is
> programmed?

Not AFAIK.


From: Wilba on
Chris Malcolm wrote:
> Wilba wrote:
>> David J Taylor wrote:
>>> Wilba wrote:
>>>> David J Taylor wrote:

>>>>> However, given that there is noise in any electronic or optical
>>>>> system,
>>>>> and given that there will be tolerances in any mechanical system, I
>>>>> would not be surprised if given a fixed starting point, the AF system
>>>>> did not stop at /precisely/ the same finishing point every time.
>>>>
>>>> Correct. Even CDAF has shot-to-shot variation. PDAF is a touch more
>>>> variable. But we're not talking about micrometric precision, just not
>>>> focussing between the end points of the beep band. The beep test shots
>>>> match, so it seems that PDAF triggers focus confirmation on the
>>>> transition, on entering or leaving the beep band.
>>>
>>> Consistent with moving the minimum possible amount - i.e. getting the
>>> fastest AF.
>>
>> Except it does the exact opposite - it makes front-focus out of gross
>> back-focus.
>
> That sounds like more movement, which sounds like it wouldn't be the
> fastest AF, but if it's the result of an estimated dead reckoning
> movement from a distance which involves that kind of error, then
> improving it would involve another iteration through the focus
> measurement process, which would slow the AF down. So that behaviour
> could still be the result of going for AF speed.

I don't think so, because when _I_ do the focussing movement at very low
speed (the beep test), I get the same two focus points as AF does. So it's
however focus is confirmed that determines where the focus ends up. One part
of that is how the light from the lens is turned into two signals by the AF
sensor, and another is how the phase difference between the two signals is
deemed to be close enough. Somehow that comes out as two distinct points,
and the one you get can be predicted from the initial conditions.