New AF MA micro adjust method: DotTune

Started Feb 15, 2013 | Discussions thread
Horshack
Veteran MemberPosts: 4,141
Like?
Re: Didn't work for me
In reply to WilbaW, Feb 28, 2013

WilbaW wrote:

Horshack wrote:

Your experience dovetails with my theories on PDAF operation. MFA fixes only static optical path alignment/tolerance issues, which includes the lens, mount, mirror, AF point, etc; basically, everything in the AF system short of the the AF motor/servo.

No, it contradicts what you're saying, it's an optical/detection effect, not an alignment/tolerance issue. If you eliminate driving the focus ring (as DotTune does, or very careful manual focus, or using a macro slide...), you (generally) get the same acceptance zone as when the focus ring is driven.

Have you forgotten already what we shared in the Fred Miranda thread?

We're saying the same thing, I'm just using broader terminology for the possible causes behind PD sensing errors. By alignment/tolerance issues of the optical path I'm including everything from lens optical abberations, decentering, mount misalignments, etc...basically everything that would affect the alignment and thus detection of the split phase from the lens down to the AF sensors.

AF motor/servo based issues are supposed to be accounted for by the closed-loop iterations of the PDAF system, although that system as you observed is designed to arrive at focus within a given range (for speed), which means lenses with imprecise positioning servos/detectors or nonlinearities in their movements will still produce inconsistent focus even with a perfectly tuned MFA.

That doesn't sound right to me. AF confirms focus at the ends of the range, not between them, so I'm guessing that's the "inconsistent focus" you're seeing. Then you're attributing that to mechanical imprecision and non-linearity. DotTune shows the same acceptance zone, so it's nothing to do with focus ring drive.

Are you saying you believe the PD phase is not being sample for the duration of a PDAF acquisition (ie, throughout the multiple iterations of the loop)? My experiments show the contrary. My point about motor positioning errors is that they can put the final movement at the edge of what you call the acceptance zone, which would be OOF for large-aperture situations. I can demonstrate this on my EF 50mm f/1.4

The reason multiple PDAF acquisitions must be used in most MFA techniques (including LensAlign/Focal) is to work around the shot-to-shot mechanical variability of the AF motor/servo, so that the (static) AF tune value can calculated in spite of this variability, rather than because of it.

There is a little variability with autofocus, but it's tiny compared to the width of the acceptance zone. The notorious EF 50/1.8 can position the focus ring with accuracy tighter than 0.1mm, which is 5% of the thinnest depth of field, 3% of the acceptance zone, and 0.02% of the MFD. It's not a "crude mechanism" problem.

I used to believe there was little variability in the PD sensing phase until I started experiments with the alpha implementation of the automated DotTune in Magic Lantern. Under constant lighting conditions the PD sensing sometimes shows individual outliers of up to 5 AF tune points, ie focus confirmations at +/- 5 beyond the normal confirmed range. It doesn't happen very often but I have seen it; luckily both a well-executed manual DotTune procedure and the automated ML procedure handle this variability and still produce the right midpoint. Btw, the ML DotTune is actually a great platform for experimenting with PD phase detection, like the consistency note I made above but also for comparing how PD phase detection responds to different lighting levels and temperatures.

Reply   Reply with quote   Complain
Post (hide subjects)Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow