EF 50/1.8 AF Experiment?

Discussion in 'Digital SLR' started by Wilba, Dec 20, 2009.

  1. The mechanical backlash in the focusing system would explain that,
    since whichever direction the focus was last moving in would also be
    the last direction in which focus error was measured.

    In the case of lenses with extremely critical focus and extremely
    shallow depth of field the minimum AF motor focusing step will often
    produce such a directional bias, especially in the cheaper models with
    cheaper PD AF systems.
     
    Chris Malcolm, Mar 26, 2010
    1. Advertisements

  2. Wilba

    Wilba Guest

    Ah, it never crossed my mind that you might be talking about Nikon lenses.
    Blame my ignorance of Nikon gear and spending too much time in Canon fora
    (and the title of the thread). :- )
    I haven't tested the one shot vs. servo difference. It makes sense to me
    that it might help.
    There is a patent application by Canon that describes a hybrid system -
    phase detect for rapid response followed by contrast detect for the final
    result. Maybe we need to think about this the other way around - a great
    EVIL camera with some kind of TTL phase detection via a much smaller and
    lighter partial mirror. :- )
     
    Wilba, Mar 27, 2010
    1. Advertisements

  3. Wilba

    Wilba Guest

    Sounds like you haven't considered the evidence. :- )
    Sure, but I always get a confirmed focus. It's consistently in the right
    place if I start from the right directions, and consistently in the same
    wrong place if I start from the wrong direction.
    Except at apertures wider than about f/2.8. Taking into account the results
    of various experiments, it seems that the softness due to spherical
    aberation wide open causes the AF sensor to confirm focus over a wide range.
    Yep, that's the myth. It dies very hard.
    No, backlash is not a factor.

    Here's a thought experiment for you. Near the closest focus distance at
    f/1.8, the DOF of the 50/1.8 on a crop body is about 3mm by conventional
    calculations. If you were to move the camera in small steps through a 4mm
    range, focussing via contrast detect at each point, what pattern would you
    expect to see in the focus achieved?
     
    Wilba, Mar 27, 2010
  4. Wilba

    Wilba Guest

    Backlash, "crudeness", or any other electro-mechanical deficiency of the
    lens is proven not to be the cause. When I focus by changing the
    camera-to-subject distance until focus is confirmed (i.e. no focus ring
    motion throughout the process), I see exactly the same sidedness - focus is
    optimal from one direction and consistently front-focussed from the other.
    Identical results to autofocus, with no backlash involved.
    Have a go at the thought experiment I posted in reply to Mikewhy.
     
    Wilba, Mar 27, 2010
  5. Wilba

    Wilba Guest

    No worries.
    Once you accept that Canon PD AF is closed-loop (easily proven), and that
    focus can be confirmed over a range of camera-to-subject distances around
    the size of the minimum DOF (ditto), and no other credible explanation has
    been offered, it makes perfect sense to me that it's because of the
    spherical aberation. I'd expect any other system based on the same
    principles to exhibit similar behaviour.
    I'll see what I can do. :- )
    I guess there's something novel in the way they put those two methods
    together.
     
    Wilba, Mar 27, 2010
  6. Wilba

    MikeWhy Guest

    I stopped following this months ago. What did you conclude, after
    considering all the evidence?
    It hurts my head to have to think about uncorrected aberrations at the
    center of the image, where the sensor in question is reading, caused by the
    glass near the rim. About the best I can do is furrow my brow and mutter "I
    don't know. Maybe. How does one even do that?" But then I recall that I have
    done so, and did so quite frequently.

    As I said, the PDAF sensor is exactly the same as a split prism focuser. I
    spent a large part of my adult life staring at precisely that, and never
    thought I could identify defects in the image on the split prism. In other
    words, the focusing errors might provably be NOT a lot of other things, but
    that doesn't mean that it is caused by "fuzziness ... over a range of the
    order of the DOF" on the PDAF sensor. There is no DoF or CoC in the aerial
    image on the split prism. It is looking at the light through a single point.
    (But then my head begins to hurt again, thinking about how much illuminating
    flux passes through a diminishingly small point.)
    That myth isn't dead here at all. I pointed the AF dot at spring clamps
    clamped at various distances along a tape measure, and shot them front to
    back then back to front a few times, wide oopen and f/8 at each step. The
    plane of of focus is very evident by examining the markings on the measuring
    tape. Focus and mis-focus were consistent in my sample set. They were
    consistently at the same point each pass. My conclusion will differ from
    yours. My results say the focus step is too coarse at near focus for exact
    placement, but by 4 ft distant, the effect is not measurable. So, I continue
    to not use the 50/1.8 for critical focus close work. It's otherwise a fine
    lens, and I will continue to trust its AF despite the reports to the
    contrary. Even when focus was "off", the focus point was still within DoF at
    wide open aperture.

    What do I expect when I move the lens in sub-millimeter steps? I expect
    focus to slide with camera placement until the PD correlation function
    decides to step the focus. I think you've reached a point where looking at
    it from outside the camera is too limiting. Your next step is to hack the
    firmware and see what the phase detect sensors are seeing.
     
    MikeWhy, Mar 27, 2010
  7. [/QUOTE]
    Spherical aberration introduces aperture-related focus drift, because
    PDAF systems work with the lens wide open and the AF sensor imposing
    its own smaller fixed aperture. If the lens is stopped down further
    than that the focus point will drift. Where DoF is very shallow it can
    drift right out of the DoF.
     
    Chris Malcolm, Mar 27, 2010
  8. I agree: I must apologise for having omitted a "not" in the sentence
    of mine which you quote above. I was trying to explain why the
    described problem would *not* be due to backlash :)
     
    Chris Malcolm, Mar 27, 2010
  9. That's the effect I noticed in focusing closely with my Sony 50mm
    f1.4. It is possible, but tricky, to adjust focus manually on the lens
    to greater than the steps of the AF motor drive. It's also easy to
    adjust focus manually by moving the camera along some approximation to
    a focusing rail. So in those situations where AF is not accurate
    enough I use manual focus.
     
    Chris Malcolm, Mar 27, 2010
  10. Wilba

    Wilba Guest

    The central "enhanced precision" PD AF sensor is sufficiently confused by
    the fuzziness caused by spherical aberation for it to confirm focus over a
    range of camera to subject distances of the order of the minimum DOF (widest
    aperture, closest focus distance). In other words, the errors traditionally
    attributed to crudeness in the lense's focus system are caused by focus
    being confirmed at two different places, depending on from which direction
    you begin.

    The key to understanding this is accepting that the PD AF process stops when
    focus is confirmed (it's a closed-loop control system). If you have 4mm of
    DOF, and focus is confirmed over a 3mm range, and one end of the range puts
    the subject in the centre of the DOF (like it does with my calibrated gear),
    then the other end of the range puts the subject outside the DOF. That's
    what users observe this lens doing, but because they haven't compared the
    results of controlled tests using a variety of auto and manual focussing
    methods, and they believe the myth that Canon PD AF is open-loop, they just
    go with the myth that what they're seeing is a result of the "crude focus
    mechanism".
    But the sensor isn't "reading" from the centre! It needs an f/2.8 or better
    exit pupil to illuminate its rangefinder base, and the two beams it receives
    do come through the glass near the rim.
    Yes, that's one of the good reasons why all the blather about
    aperture-related focus shift is irrelevant.
    No, it's looking at light from both sides of the lens near the rim. A
    non-photographic rangefinder has two lenses, right? The one built into you
    camera uses two virtual lenses within your single objective lens.
    Not sure which myth you mean here.
    So you would expect to see a cycle of the subject going out of focus by the
    step distance, until it reaches a point where it jumps back into focus,
    right? That's exactly what I would expect, based on the crudeness theory.

    I took a series of 40 CD AF shots (closest focus distance, widest aperture),
    moving the camera by 0.1mm steps through a 4mm range. The variation in focus
    between successive shots was of the same order as the variation between
    successive shots at the same camera to subject distance. IOW, if I were to
    show you two shots any arbitrary number of steps apart, and two successive
    shots focussed at the same distance, you would not be able to reliably
    determine which set was stepped and which set was static.

    When I animated the 40 shots, there was no cycle apparent, no shifting focus
    and snapping back. So the crudest step that the lense's focus mechanism can
    provide is significantly smaller than 2.5% of the minimum DOF.
    No need. Everything I need to know to get the best out of my gear is
    apparent from the black-box behaviour. With my particular calibration, all I
    have to do with that lens to get optimal PD AF is start with the plane of
    focus in front of the subject. Simple, easy, quick, 100% reliable. Starting
    with the plane of focus behind the subject gives something like a 70%
    failure rate.

    If you were playing a game in which your chance of success was 65%, and I
    offered you a simple legal method of getting a 100% success rate, would you
    take it? :- )
     
    Wilba, Mar 28, 2010
  11. The fuzziness caused by spherical aberration is caused by the fact the
    perfect lens shape is only nearly spherical. So in a spherical lens
    the rays from the lens edge focus at a different point than those from
    the centre. But this won't cause any fuzziness in a PDAF sensor
    because that works by taking two small samples of rays from two spaced
    apart little apertures equally spaced about the centre. The distance
    between them is in effect like the baseline of a rangefinder. That
    distance apart also corresponds to an aperture of that diameter. So
    the PDAF can't work with lenses of smaller aperture than the width
    apart of the AF sensor pair.

    Therefore in a spherical lens the AF sensor finds exact focus at the
    point where the rays from that particular part of the spherical lens
    surface focuses. But the image is taken at the specified aperture. So
    if it is stopped down from the effective AF aperture the focus will
    drift one way. That probably won't matter because typically the DoF is
    increased a lot more than the drift.

    If on the other hand the image taking aperture is larger than the
    effective AF aperture, as it can be with wide aperture lenses, then
    the focus drifts the other way, while at the same time the DoF is
    narrowing. In this case the resultant focus drift can take the point
    of focus outside the DoF.
     
    Chris Malcolm, Mar 28, 2010
  12. Wilba

    Wilba Guest

    Yes, that's as I've understood it since before I began this thread.

    Can you use the above to explain why an initial gross front-focus results in
    optimal focus, and an initial gross back-focus results in front-focus (with
    my calibrated 450D and 50/1.8)?

    Everything is identical in both cases except for the spatial relationship of
    the subject and the plane of focus prior to PD AF being invoked. It doesn't
    matter whether I use f/1.8 (the widest aperture of the lens), or f/2.8 (the
    rangefinder base aperture of the PD AF sensor), throughout an individual set
    of front and back-focussed test exposures, I still get the same pattern of
    optimal focus or front-focus.

    How does aperture-related focus shift explain that phenomenon?
     
    Wilba, Mar 28, 2010
  13. Wilba

    MikeWhy Guest

    I don't share your opinion that the stepper motor and focus mechanism on a
    $100 consumer grade lens can hold micrometer repeatability. The randomness
    you saw is more easily attributable to mechanical slop than differences
    arising from optical or digital computation. The lens shape doesn't change
    between each test shot. The electronics -- sensor array, phase detect and
    difference computations -- did not change. The inherent mechanical slop of
    any moving mechanism is always present.
    If it were critical to what I do, I would grab the 100/2.8 macro and focus
    manually in LiveView. I'll keep what you say in mind, despite the
    impossibility of taking advantage of it. The AF-On button on my camera is
    mapped to FEL, because the button that originally was FEL is the only
    available button for the in-VF electronic level. I get along fine by
    stopping down for critical focus and sharpness.

    But really, since directionality is involved and mechanical backlash is
    present, on the Canon lens noted most for its low price point, how exactly
    did you rule out mechanical slop? Walk me through the thinking if you still
    have the patience.
     
    MikeWhy, Mar 28, 2010
  14. Wilba

    Wilba Guest

    Whatever you mean by "micrometer repeatability" it doesn't sound like what
    I'm talking about, which is, there are no steps aparent in the motion of the
    focus ring with 0.1mm changes in camera to subject distance.

    If the focus ring moved in steps big enough to put one shot in the centre of
    a 4mm DOF and the next shot just outside that DOF, a focus "slide and jump"
    would be clearly apparent in the series of 40 shots. There is no evidence of
    sliding or jumping.
    Exactly! The difference in focus between any two shots (whether they are at
    the same distance or any number of sub-millimeter steps apart), is _not_ due
    to the system "deciding to step the focus".
    That's easy. With my gear at widest aperture and closest focus distance, PD
    gives only two focuses - either optimal, or front-focussed to the same
    degree each time. Never back-focus. I get the same two identical-looking
    sets of images whether I use auto-focus, or set the lens to manual focus and
    move the camera until focus is confirmed (the beep test).

    Take particular note of that last point - when the lens is not instructed to
    move the focus ring between shots and the focus ring does not move between
    shots, i.e. no backlash possible, I get the same two sets of results as AF.
    The initial placement of the plane of focus relative to the subject
    determines whether I get optimal focus or front-focus.

    If backlash was the cause, I would expect to see greater variability within
    the optimal set (and when I say optimal, I mean pretty much
    indistinguishable from what CD AF does), I would expect to see greater
    variability with the front-focussed set (the error is as reliable as the
    accuracy of the optimal set), and I would expect to see back-focus at
    something like the frequency of the front-focus.

    The only conclusion I can make is that the sidedness comes from the PD
    sensor, and that steps or backlash in the focus ring motion are not
    significant.
     
    Wilba, Mar 29, 2010
  15. Wilba

    MikeWhy Guest

    I mean that mechanisms have clearances between mating parts to allow
    movement; mass produced parts have tolerances; stepper motors, even ring USM
    focus drives, have discrete rotation step sizes; all of which conspire to
    prevent infinitely fine adjustment of focus, and even if could theoretically
    accomplish that, tolerances between mating parts will prevent perfect
    repeatability. I don't expect that the lens should be able to focus exactly
    on the desired point. Close enough has to suffice, simply because the
    smallest focus step will more often than not prevent exact focus.
    I recall now that you wrote earlier about not adjusting focus, and tested by
    moving the camera and watching for the in-focus indicator. Fine, but see
    above. The discrete, finite size focus steps requires that the in-focus
    indicator activates when it tests within some acceptable tolerance, which
    might well not be exact focus. Your earlier mentioned fuzziness to some
    order of DoF is that tolerance, not because the sensor is incapable of
    resolving the difference, but because it has no expectation that it can
    achieve perfect focus. We're still no closer to knowing what the sensor can
    resolve, or how closely the mechanical slop can focus. We only know that the
    smallest focus step is smaller than the DoF of the lens at its widest
    aperture (or we hope so, anyway). Closed loop final focus takes care of the
    rest.

    I expect you'll resist because you already have so much thought invested. I
    know only that the worst case in my focus tests were always within half the
    DoF at wide open aperture. That's sufficient to make a usable image. As an
    engineer myself, I expect that the AF system has to have been designed with
    performance targets which define "close enough", because exact focus will
    not always be achieved, for all the reasons pointed out earlier.
     
    MikeWhy, Mar 29, 2010
  16. That baffles me. It doesn't happen with my 50mm f1.4 lens.
     
    Chris Malcolm, Mar 30, 2010
  17. Wilba

    Wilba Guest

    Sure, but we know that the motor step is less than 2.5% of the DOF. So why
    does only PD AF and initial gross back-focus put the subject behind the DOF
    when in every other case the subject ends up in the middle of the DOF?
    We're as close as we need to be. CD AF puts the subject bang in the middle
    of the DOF every time with adequate illumination. PD AF does so to if we
    start with gross front-focus. PD AF gives a consistent front-focus if we
    start with gross back-focus.
    The best data we have tells us that the smallest focus step is smaller than
    2.5% of the DOF.
    I expect you'll resist because you already have so much thought invested in
    not accept the evidence. :- ) (I'm not assuming you don't understand the
    evidence, but you sure wouldn't be the first.)
    What lens and body?
    And more. But the evidence shows that the PD AF initial gross back-focus
    error is there because the AF sensor confirms focus over a wide range (of
    the order of the DOF), with my 50/1.8.

    All you have to do is explain how backlash and/or motor steps only causes an
    error in the PD AF initial gross back-focus case, and is not evident in any
    way in the other three AF cases. If you can do that, we'll all be sorted.
     
    Wilba, Mar 30, 2010
  18. Wilba

    MikeWhy Guest

    Cite please, for the 2.5% DoF. It's far from common knowledge.
    If anything, I'm trying to avoid thoughts like those. I found it already
    focuses no worse than I would with a split prism, and went on to other
    puzzles.
    Canon 7D, 50/1.8 II. Always the same front focus, each time making a usable
    image of the focus target.
    I didn't see that behavior.

    Why this lens, of all lenses? I lost track of what point you were making.
    Was it about the nifty-fifty? Or just AF in general?
     
    MikeWhy, Mar 30, 2010
  19. Wilba

    Wilba Guest

    -

    "I took a series of 40 CD AF shots (closest focus distance, widest
    aperture),
    moving the camera by 0.1mm steps through a 4mm range. The variation in
    focus between successive shots was of the same order as the variation
    between successive shots at the same camera to subject distance. IOW,
    if I were to show you two shots any arbitrary number of steps apart, and
    two successive shots focussed at the same distance, you would not be able
    to reliably determine which set was stepped and which set was static.

    "When I animated the 40 shots, there was no cycle apparent, no shifting
    focus
    and snapping back. So the crudest step that the lense's focus mechanism can
    provide is significantly smaller than 2.5% of the minimum DOF."

    So why does only PD AF and initial gross back-focus put the subject behind
    the DOF when in every other case the subject ends up in the middle of the
    DOF?

    I'd be very interested to see what result anyone else gets from the same
    experiment on the same or different camera models.
    Okay, so you won't accept my evidence because it doesn't fit with your
    experience with a different camera. I'd be very surprised if those two
    bodies behaved identically.
    I'll take that to mean, "no, I can't". :- )

    We're not talking about /your/ camera! There's no way that I'd use the
    performance of the PD AF system on my 450D to make specific predictions
    about the performance of any other model of camera, or vice versa.

    Once you prove that the camera can put the focus effectively wherever it
    wants (finer than 2.5% of the DOF), that indicates that the PD AF error is
    not caused by the "crude focus mechanism" of the lens, and then you prove
    that the error occurs because focus is confirmed over a wide band (of the
    order of the DOF with my gear), then the question you're left with is what
    is going on optically with that lens to trick the sensor into confirming
    focus over a wide band.
    The original question was -
    A lot of interesting stuff has come from that. The 40x0.1mm shot test
    disproves the crudeness theory. The beep test (changing the camera to
    subject distance until focus is confirmed), proves that the PD AF
    front-focus error is a property of the interaction between the lens optics
    and the PD AF sensor. Starting with the plane of focus in front of the
    subject completely eliminates that error. I can't find a way in which
    backlash and/or motor steps explains anything of significance to do with all
    that.

    I'd love to know what you get from doing this - camera on a tripod, remote
    release, etc., 50/1.8 at widest aperture and near minimum focus distance,
    high-contrast detailed well-illuminated target with a slanted scale for
    evaluating the position of the plane of focus, ten PD AF shots from an
    initial gross front-focus, ten PD AF shots from an initial gross back-focus,
    and ten from each side using CD AF.
     
    Wilba, Apr 1, 2010
  20. Wilba

    Wilba Guest

    Which bit? _That_ it happens with the 50/1.8, or _why_ it happens?
     
    Wilba, Apr 1, 2010
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.