[rodeknyt]

I need some help understanding something about modern autofocus.

Back in the good old days of manual focus, SLRs used split-image and/or micro prism aids for achieving focus. Get the two halves lined up, or make the shimmering disappear and you were in focus. Worked the same all the time, regardless of what lens you used. Simple and effective.

Now, with autofocus we have phase-detection which is, from what I understand, the electronic equivalent of the split-image rangefinder. The lens is adjusted by the system until the subject is “lined up” and focus confirmation is given.

If that is the case, then why would there be the need to make an adjustment (other than because of manufacturing slop, in which case one global adjustment would take care of it) and why would there be a difference from one lens to the next?

After many years of hiatus following sale of a Pentax ME Super, my foray into the autofocus digital world started with the K-10D. That camera didn’t have focus adjustment capabilities, yet I never saw any problem with focus. Through successive iterations of K-7, K-5 and now K-3, the same lenses have all needed different adjustments on each camera.

Somebody please shed some light on this for me.

[jatrax]

other than because of manufacturing slop, in which case one global adjustment would take care of it) and why would there be a difference from one lens to the next?

You have answered your own question. Manufacturing slop, or tolerances are not precise and vary from lens to lens (even in the same exact lens).

Modern cameras have three separate focusing systems:
1) The focusing screen, same as what you used in your ME Super, though not as good. Used for manual focusing. This can be adjusted by adding or subtracting shims above the actual focusing screen in the camera. Normally this is done in the factory and should be fairly precise but it can be off. And if you add a third party screen adjustment is sometimes needed to get what your eye sees in sync with what the lens is presenting to the sensor

2) PDAF this is the standard viewfinder AF system and out of the box is usually fairly good with slow lenses like the kit lens. But put a fast (f/2.8 or faster) lens on and the manufacturing tolerances may be enough to cause front or back focus. Not enough to be noticeable on a slow lens but easily seen on a fast one. As it was explained to me if both camera and lens are at technical zero then all is wonderful. But if the camera is off say -5 in one direction and the lens is off -5 in the opposite direction now you have a -10 error. That is why some lenses work fine on one camera but not on another. This system is adjusted using the camera's built in AF fine tuning controls

3) CDAF this is what is used in Live view and uses the sensor itself to determine the focus. As I understand it this is always correct and there is no adjustment mechanism provided. The downside, at least currently, is that this system is slower than the PDAF system though it is theoretically more accurate. Focus peaking on the Live View screen is the visual equivalent in CDAF of the old micro mirror prism screen on film cameras.

Hope that helps, and hope I got all the terms correct

[seventhdr]

In the "good old" film days the focusing screen was positioned directly under the viewfinder pentaprisim such that the distance from the mirror to the focusing screen and the distance from the mirror to the film plane were the same. This ensured that what you saw in focus in the viewfinder was what you got on the film.

When auto focus was first introduced a secondary mirror was installed behind the main, now partly transparent, main mirror to reflect a small portion of the incoming light to the auto focus metering system in the bottom of the camera. This introduced additional elements and components into the system, all of which could be partly out of optimal adjustment. In the film days we seemed to be less focused on whether our focus system was totally in alignment, but now with digital sensors, and especially high megapixel sensors, we pixel peep more and can detect and be upset by slighly mis-aligned auto focus systems. The ability the adjust the auto focus system also allows us to accont for sample variations in the lenses we use.

This is why we need to at times optimise our auto focus system to achieve optimal focus and minimise front or back focus issues.

With sensors now able to determine auto focus directly off the sensor we are back to having auto focus being able to be determined at the sensor plane the same way it was done, equivalently, at the focusing screen.

Regards

Chris

[stevebrot]

You have answered your own question. Manufacturing slop, or tolerances are not precise and vary from lens to lens (even in the same exact lens).

Mechanical lash is part of the issue, but focus confirm (manual focus using the PDAF detector) is also affected. The answer lies with how PDAF works. Yes, it uses the same principle as a split image, but not in the same way. There are a couple of real good articles on the web that explain how it works in easily assimilated terms and which make the calibration aspect more obvious. The optical characteristics of the lens come into play. If I find the links, I will post them here. One includes a Javascript simulation of showing the light paths at various points of focus and out-of-focus. The other is a long PDF document describing PDAF as well as split-image and microprism rangefinders.


Steve

[cfraz]

Mechanical lash is part of the issue, but focus confirm (manual focus using the PDAF detector) is also affected. The answer lies with how PDAF works. Yes, it uses the same principle as a split image, but not in the same way. There are a couple of real good articles on the web that explain how it works in easily assimilated terms and which make the calibration aspect more obvious. The optical characteristics of the lens come into play. If I find the links, I will post them here. One includes a Javascript simulation of showing the light paths at various points of focus and out-of-focus. The other is a long PDF document describing PDAF as well as split-image and microprism rangefinders.


Steve

I've seen the web site showing the light paths as the lens element moves closer to and further from the sensor, going in and out of focus, and how that movement affects what the PDAF sensors "see."

The idea in the focus animation on that web site is that if the distance from the lens to the main sensor is identical to the light path distance (through the semi-transparent mirror and reflecting off the secondary mirror) from the lens to the PDAF sensor, then autofocus is well calibrated.

But like the OP, I still don't get why individual lens adjustments are needed. I understand there are manufacturing tolerances in both camera bodies and lenses. It makes sense to me that a global adjustment is needed to correct small positioning errors of the PDAF optics so that when the image is in focus on the main sensor, it signals focus as well.

Once that is achieved, how is the behavior of one lens different than another? It must imply that lens A projects an image which is focused on the main sensor and the PDAF sensors as well, while lens B projects an image which shows in focus on the PDAF sensor but not the main sensor.

There are subtleties which the web animation doesn't seem to address. And for which I've not yet heard (or at least understood) an explanation.

[cfraz]

Here is the animation I was refering to: Autofocus: phase detection

[stevebrot]

Here is the link to the PDF I mentioned above:

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


Steve

---------- Post added 08-27-14 at 06:14 PM ----------

But like the OP, I still don't get why individual lens adjustments are needed.

It is my understanding that optical aberrations may bias the PDAF system.


Steve