Reverse-engineering AF fine tune

Started Jul 27, 2012 | Discussions thread
ForumParentFirstPrevious
Flat view
Horshack Veteran Member • Posts: 8,005
Reverse-engineering AF fine tune
1

While I wait for my D800's Left AF issue to be fixed I've been thinking about AF fine tune on phase detect focus systems and how it's actually implemented. This may seem boring to most but for some reason it intrigues me.

My first though was that the AF tune value is simply an offset applied to the generated lens distance servo command. For example, after the AF system evaluates the phase of the split beams and determines both the direction and magnitude of the phase differential, it translates that calculation into a command on how far (and in which direction) the lens should be moved (via either the internal servo motor for lenses without AF motors or a discrete command to the lens that have motors). I was thinking the AF tune was simply a post-calculation offset to that value, so that if the needed lens movement was calculated as 'x' it would simply be changed to 'x' + a translated value derived from the user-specified AF tune value.

A quick experiment disproves this though. If I enter Live View and establish perfect focus either via CDAF or manual AF and then exit Live View, the VF confirmation dot is illuminated to confirm that the subject is indeed in focus. However if I then change the AF tune value with the focus unchanged, the confirmation dot disappears and is replaced with the rangefinder arrows indicating how far the PD system thinks the lens is out of focus.

Based on the above it must be that the AF tune value is not post-phase (output) adjustment offset but instead an input offset into the phase-detect sensing mechanism itself, ie the offset skews what the phase detect logic senses as in-phase vs out-of-phase. So for example on a body+lens needing no AF adjust, when PDAF is acquired the sensed phase differential of the split beam is zero. Contrast that with a body+lens needing AF adjust; when PDAF is acquired the sensed phase differential of the split beam would be zero but the logic would interpret that as being out-of-phase when the AF tune value is applied, so that in-focus is established when the logic detects the two beams as out-of-phase by an amount equal to the AF tune value (translated of course).

In retrospect it seems it would have to work this manner anyway in order to support fine tuning of manual focus lenses. It might also have to work if you consider that PDAF is not a single open-loop evaluation/command sequence but instead appears to involve several sensing stages, each followed by a separate command phase (there was an article recently that proved this to be true, thus PDAF is somewhere between an open and closed loop feedback system).

Assuming my observation is correct, I wonder what the implications are if any for how well the AF tune mechanism works across different focal lengths of a single lens (zooms) and across the various aberrations the AF system might encounter, either from the lens design itself or from the particular scene.

Sorry if this is all already obvious to everyone else

ForumParentFirstPrevious
Flat view
ForumParentFirstPrevious
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow