[Culture]

How does the AF point layout and and lens aperture setting affect focus?
As far as I have understood the AF points are sensitive to an extent. Some are sensitive to 2.8 for example.
So what will it mean if for example my aperture is set to 1.7 but my focus point is at the edges of the AF point?

What should one pay attention to then if there is a possibility of focusing error?

[UncleVanya]

The setting of your aperture shouldn't matter unless you are using an older lens that is stopped down during focusing.

But the maximum aperture of your lens does matter. I would think using the faster points would improve focus accuracy when using a lens with f2.8 or faster maximum aperture.

[photoptimist]

If you are using a fast lens at a wide-open aperture and need accurate focus, you be more like to get good results by focusing with the f/2.8.

That said, you might get perfectly good focus accuracy using the other AF points if the target has really good contrast and lighting.

[stevebrot]

How does the AF point layout and and lens aperture setting affect focus?
As far as I have understood the AF points are sensitive to an extent. Some are sensitive to 2.8 for example.
So what will it mean if for example my aperture is set to 1.7 but my focus point is at the edges of the AF point?

The points labeled "f/2.8 luminance flux" have extended focus sensitivity to f/2.8. The other points are sensitive to f/5.6. How this translates in terms of performance is a little difficult to explain, but I will give it a shot.

• The diagram labeling is somewhat inaccurate in that it is the exit pupil diameter (roughly the same as the aperture opening) rather than the luminance flux (brightness) that places the limitation for focus sensitivity with PDAF
• Focus sensitivity is the ability to detect the out-of-focus condition
• Focus sensitivity is expressed as f-number as in the following example:

  The ability of the sensor to detect out-of-focus is the same at wider apertures as at f/5.6 when focus sensitivity is f/5.6
  (sensor is not designed to take advantage of the wider aperture)

• Maximum aperture greater than f2.8 (e.g. f/1.7) will provide focus sensitivity equivalent to f/2.8 with the three center points and f/5.6 with the others in the spread
• In practical terms, focus sensitivity affects precision, the ability to assign the same plane of focus on multiple attempts. This is different than accuracy. Even a well-calibrated (accurate) detector can be wildly imprecise.

Simply put, the three f/2.8 points provide higher probability of in-focus results than the f/5.6 points when used with faster than f/5.6 lens and with no additional advantage at wider than f/2.8.

Examples:

f/5.6 lens --> equal precision with all points
f/3.5 lens --> better precision with f/2.8 points than with f/5.6 points
f/2.8 lens --> better precision with f/2.8 points than the f/3.5 lens
f/1.4 lens --> same precision with f/2.8 points as with f/2.8 lens
f/1.4 lens --> same precision with f/5.6 points as with f/5.6 lens

For a full and not too difficult description, I have found the paper by Doug Kerr to be helpful. His approach is to properly describe PDAF as a variation on split-image rangefinders. The limitations related to exit pupil diameter* are more easily understood from that perspective.

http://dougkerr.net/Pumpkin/articles/Split_Prism.pdf


Steve

* Cutting to the chase...rangefinder baseline as balanced against prism blackout as explained on the last page of the PDF.

---

Last edited by stevebrot; 11-22-2017 at 12:24 PM.

[Nicolas06]

Simply put, the three f/2.8 points provide higher probability of in-focus results than the f/5.6 points when used with faster than f/5.6 lens and with no additional advantage at wider than f/2.8.

Not really. A point see the scene at a fixed appeature by it's design. If the actual apperture is wider it doesn't change anything. If the actual apperture is narrower the AF doesn't work.

From the schema, the so called f/2.8 points are sensitive to horizontal constrast at f/5.6 and vertical constrast at f/2.8. If an f/4 or f/5.6 lens is used, then only the horizontal constrast will be detected optimally. The vertical constrast will have quite some difficulties to detect any constrast with a lens with a narrower apperture than f/2.8.

For maximum performance on theses focus points, you need a lense than is f/2.8 or wider. f/4 and f/5.6 do not benefit at all. Also the gain is only visible for constrast in some direction not all.

If you lens is f/4 only, you may as well use any other point, the performance may be even better actually as only the f/5.6 sensors can be used anyway.

[Nicolas06]

Examples:

f/5.6 lens --> work only with f/5.6 sensor, not f/2.8. Best to use only on f/5.6 sensors.
f/3.5 lens --> work only with f/5.6 sensor, not f/2.8. Best to use only on f/5.6 sensors. Do not provide better performance than an f/5.6 lens for AF.
f/2.8 lens --> work with f/5.6 and f/2.8 and benefit of added precision of f/2.8 sensors.
f/1.4 lens --> work with f/5.6 and f/2.8 and benefit of added precision of f/2.8 sensors. Do not provide better performance than an f/2.8 lens for AF

For me, this is more like that...

[Culture]

Thanks guys for the explanation.
Which bring the point to my question. How much does this affect our normally photography?
Is this so small of a problem that there is no need to pay attention to it.

I have a Tamron 28-75 f2.8 that seems to be either front or back focusing. I have seen so much missed focus shots that I am almost certain there is a problem.
The thing is when I try to do micro adjustment it tells me that everything is ok.

I move my focus point around alot when using this lens at events, hence the observation.

[BrianR]

Not really. A point see the scene at a fixed appeature by it's design. If the actual apperture is wider it doesn't change anything. If the actual apperture is narrower the AF doesn't work.

Steve's correct. I've seen the tolerances of AF points described in terms of depth of focus (not depth of field) - for example the f/5.6 focus point should get you be within 1 depth of focus of the true target. Lets say this is 0.1mm. If you use an f/2.8 lens with this f/5.6 point, it will get you within 0.1mm. However, if you're shooting wide open, the actual depth of focus will be 0.05mm, and you can be right out of it (you may also get 'lucky'). Same problem if you're shooting at f/1.4 on an f/2.8 point, or even worse with an f/1.4 lens on a f/5.6 focus point.

Using a lens faster than the focus point is designed doesn't change the tolerance, but you need to be more precise with wider apertures if you want to reliably to get the focus "close enough".

[Nicolas06]

Steve's correct. I've seen the tolerances of AF points described in terms of depth of focus (not depth of field) - for example the f/5.6 focus point should get you be within 1 depth of focus of the true target. Lets say this is 0.1mm. If you use an f/2.8 lens with this f/5.6 point, it will get you within 0.1mm. However, if you're shooting wide open, the actual depth of focus will be 0.05mm, and you can be right out of it (you may also get 'lucky'). Same problem if you're shooting at f/1.4 on an f/2.8 point, or even worse with an f/1.4 lens on a f/5.6 focus point.

Using a lens faster than the focus point is designed doesn't change the tolerance, but you need to be more precise with wider apertures if you want to reliably to get the focus "close enough".

The problem for me was more that f/3.5 for example is narrower than f/2.8 and so an f/3.5 lens can't be used with the f/2.8 sensor at all not matching the min apperture requirement, so only the f/5.6 sensor can be used and that give no advantage toward an f/4 or f/5.6 lens.

[stevebrot]

How much does this affect our normally photography?

How picky are you?

The short answer is that PDAF sort of sucks for consistent fine focus and that the weakness is particularly apparent when using fast lenses at wider apertures where missed focus may be more evident. The problem is less acute when using the three f/2.8 points because they are better able to detect out-of-focus with a fast lens than the f/5.6 points.

The precision issue is readily apparent when using a fast lens with a long focus throw and the focus confirm feature (PDAF with manual focus). The green hexagon in the viewfinder will generally stay lit through several degrees of arc for the focus ring with subjects at moderate or close distance; the true point of focus being attained somewhere with that range, but not detected as such. With AF, the same thing happens with the focus motor stopping somewhere within that range.

I have three "fast" AF primes* and my practice with those lenses is:

• For general shooting at moderate apertures where fine focus is not the goal, I use the AF system
• For fine focus at wider apertures, I switch to manual focus using my Katz Eye split-image screen. If on tripod, I use magnified live view.

Steve

* FA 77/1.8 Limited, FA 35/2, DA 50/1.8

[stevebrot]

The problem for me was more that f/3.5 for example is narrower than f/2.8 and so an f/3.5 lens can't be used with the f/2.8 sensor at all not matching the min apperture requirement, so only the f/5.6 sensor can be used and that give no advantage toward an f/4 or f/5.6 lens.

I believe there is a misunderstanding. The f-number designation is not the minimum usable aperture diameter. It is the maximum aperture (exit pupil size) that can be addressed in full by the virtual aperture of the sensor. The sensor can still detect coincidence, just with less precision than what the lens may offer. The minimum aperture, OTOH, is usually one to a few stops narrower and represents the threshold for "blackout" below which the detector is "blind"*. For the f/5.6 sensors, that point is reached somewhere between f/8 and f/11**. When approaching the minimum, the detector will work, though with lower efficiency. Since the f/2.8 points incorporate both f/2.8 and f/5.6 detectors, it is difficult to know where the actual minimum is for the f/2.8 component is. I would expect it to be narrower than f/4 and that the system uses output from the detector with the best signal.


Steve

* This is strictly analogous to blackout with split-image and microprism focus aides in the traditional focus screen. FWIW, the split-image in the better screens are designed to be sensitive to f/1.4 with blackout at about f/8. A microprism focus aid is generally sensitive to somewhere between f/2.0 and f/2.8 with blackout around f/5.6.

** My K-3 is unable to AF at narrower than about f/9

[leekil]

The short answer is that PDAF sort of sucks for consistent fine focus and that the weakness is particularly apparent when using fast lenses at wider apertures where missed focus may be more evident.

How much of the focus inaccuracy is due to the PDAF, and how much due to the low precision of screw-drive lenses (which were the 3 you cited)? I recall an article about Canon AF where it needed more precise lenses to actually achieve what the AF system was cable of. So if using DC, SDM, or PLM lenses, would the focus precision (and potentially also accuracy) be improved in the same way?

[stevebrot]

How much of the focus inaccuracy is due to the PDAF, and how much due to the low precision of screw-drive lenses (which were the 3 you cited)?

I was referring to the ability (sensitivity) of PDAF to consistently detect out-of-focus.

Mechanical lash is a separate concern, though the system is not supposed to signal focus attained until it no longer detects out-of-focus. Higher mechanical precision should improve the system's ability to avoid overrun and subsequent hunting, but will not change whether the plane of focus when the motor stops is consistently adequate for fine focus. I don't have the tools, but would love some day to compare focus methods using something like the LensAlign FocusTune which includes statistical analysis and graphical plotting as part of its tool set.

As for the three fast AF primes on my shelf (all screw drive)...my pockets are not deep enough to dabble in available K-mount motor-in-lens options at f/2 and wider maximum aperture. I should probably add an edit indicating that I avoid using PDAF focus confirm for any of my dozen or so manual focus lenses unless light is too dim for viewfinder focus or live view.


Steve

[Nicolas06]

I believe there is a misunderstanding. The f-number designation is not the minimum usable aperture diameter. It is the maximum aperture (exit pupil size) that can be addressed in full by the virtual aperture of the sensor. The sensor can still detect coincidence, just with less precision than what the lens may offer. The minimum aperture, OTOH, is usually one to a few stops narrower and represents the threshold for "blackout" below which the detector is "blind"*. For the f/5.6 sensors, that point is reached somewhere between f/8 and f/11**. When approaching the minimum, the detector will work, though with lower efficiency. Since the f/2.8 points incorporate both f/2.8 and f/5.6 detectors, it is difficult to know where the actual minimum is for the f/2.8 component is. I would expect it to be narrower than f/4 and that the system uses output from the detector with the best signal.


Steve

* This is strictly analogous to blackout with split-image and microprism focus aides in the traditional focus screen. FWIW, the split-image in the better screens are designed to be sensitive to f/1.4 with blackout at about f/8. A microprism focus aid is generally sensitive to somewhere between f/2.0 and f/2.8 with blackout around f/5.6.

** My K-3 is unable to AF at narrower than about f/9

Here for a refresher of PDAF: How Phase Detection Autofocus Works

The thing is basically, the sensor is composed of 2 sensitive parts and measure the amount of dephasing between the 2 parts. The key question is the distance you put between the 2 parts of the sensor. If the distance is quite narrow, the difference in signal will be small or not visible at all. But it will work with a quite narrow lens apperture. If the distance is quite large, the difference in signal will be large but if the lens has too narrow apperture, the defocussed light ray will not be visible at all as the lens simply do not transmit them.

It is not by random that f/5.6 is the basic AF point apperture and that the slow lenses are also f/5.6 (Pentax has actually and f/5.8 lens but that's near enough f/5.6). If is true that you may make the AF to work up to f/8 or f/9, but then the performance is quite unreliable and bad and work only in some case. Lot of light, huge constrast and so on.

You bet that at f/3.5 the f/2.8 sensor that is then mostly blind due to a too narrow apperture would perform better than the f/5.6 sensor that is less precise but has then a perfect signal to analyze. I'am sure that you right, there likely an apperture narrower than f/2.8 where the f/2.8 sensor would still perform better than the f/5.6 sensor. For me this may be f/2.9 or f/3. For you that may extends to f/3.5 or maybe f/4 or f/5.

In the official documentation the f/2.8 sensor if for lenses with f/2.8 as max apperture or faster and f/5.6 for lenses with f/5.6 max apperture or faster. They no official support for anything slower and this show for example on AF compatibilities when using the 1.4 TC. Pentax consider that using the TC on any lens that make the system having a max apperture smaller than f/5.6 will not AF properly. I take the same assumptions.

[beholder3]

I believe there is a misunderstanding. The f-number designation is not the minimum usable aperture diameter. It is the maximum aperture (exit pupil size) that can be addressed in full by the virtual aperture of the sensor. The sensor can still detect coincidence, just with less precision than what the lens may offer. The minimum aperture, OTOH, is usually one to a few stops narrower and represents the threshold for "blackout" below which the detector is "blind"*. For the f/5.6 sensors, that point is reached somewhere between f/8 and f/11**. When approaching the minimum, the detector will work, though with lower efficiency. Since the f/2.8 points incorporate both f/2.8 and f/5.6 detectors, it is difficult to know where the actual minimum is for the f/2.8 component is. I would expect it to be narrower than f/4 and that the system uses output from the detector with the best signal.

One line "AF point" is made up from two pixel lines. There are two sets of pixel lines: widely spaced, so they are only fully unshaded with lenses F2.8-F1.2 and more narrowly spaced so they are fully unshaded even down to F5.6.

Using an F8 open aperture lens means that F2.8 sensor lines are fully shaded and do not work at all. And from the F5.6 pixel lines only a very short piece is getting image information (remember that the left pixel line is being cut away on the left side and the right pixel line is being shaded from the right edge, so only the center bit remains functional for comparison), which makes it quite difficult to reliably find focus. There simply is only a fraction of contrast information left.
Basically at F8 your center AF point is very, very small.