• Please use real names.

    Greetings to all who have registered to OPF and those guests taking a look around. Please use real names. Registrations with fictitious names will not be processed. REAL NAMES ONLY will be processed

    Firstname Lastname

    Register

    We are a courteous and supportive community. No need to hide behind an alia. If you have a genuine need for privacy/secrecy then let me know!
  • Welcome to the new site. Here's a thread about the update where you can post your feedback, ask questions or spot those nasty bugs!

Canon EOS Autofocus - BFCV and micro-adjust (MA)

Doug Kerr

Well-known member
In this note, I will talk about two related matters in the context of the Canon EOS autofocus system, the best focus correction value (BFCV) and AF micro-adjust (MA).

Phase Comparison

The Canon EOS AF system revolves around the quantitative determination of the "error" in the current focus setting through what is often called phase comparison.

Its principle is essentially the same as what is done visually in a split image focusing aid.

For each AF point, there are two little image detectors, which I will call "AF subdetectors". They both regard the same small zone of the scene, but they have their own aperture stops, much smaller than the actual aperture stop of the lens. These are located off center, on opposite sides of the lens axis.

If the current focus setting for the scene material in the field of view of these subdetectors is perfect (that is, would result in the image being formed precisely on the actual sensor during the shot), then (in a simple model of the system) the images cast on the subdetectors will exactly align (just like the two adjacent visual images on a split-image focusing aid).

If focus is not perfect, the images will not align. The amount of misalignment is a quantitative indicator of the amount by which the image would be formed in front of or "behind" the sensor (measured along the lens axis). The relationship depends on the separation of the two little "subapertures" and other geometric quantities, and is known for any body design.

But, unlike the two image areas in the visual focusing aid, the two subdetectors are not actually located adjacent on the chip that carries them. They are displaced (just to make room - there are actually a lot of them, sometimes eight for an AF point!)

But for each subdetector, a "relay lens" in the path is aimed so as to to make the two images fall on corresponding places in the two subdetectors in the case of perfect focus. But maybe not exactly (the mechanical arrangements might not be perfect).

Accordingly, a number that tells how the locations of images on the two subdetectors, observed with respect to an arbitrary reference point on each subdetector, would relate for the case of perfect focus, determined by testing after the camera is assembled, is stored in non-volatile memory in the camera.

Then, in actual operation, the locations of the two images on their subdetectors, expressed with respect to the reference point on each, are compared, and then this adjustment value is subtracted. The resulting number is taken as the focus error (again, in terms of the distance of the planned where the image would fall from the plane of the sensor).

The autofocus process proceeds based on that value, both in reckoning the amount of lens focus mechanism movement that should bring focus to "perfect", and in confirming at the end of the whole cycle that focus is indeed now "perfect" (within an established tolerance).

Spherical aberration

If a lens has truly spherical surfaces (and traditional lens design has been predicated on that - for one thing, it is "relatively easy" to accurately create such a surface) we do not get the behavior we would like, which is that for a point on the object all the rays entering the lens will converge at a single point in the image.

Rather, all the rays passing through the aperture at a certain distance from its center will converge in a point, but all the rays passing through the aperture at a different distance from the center will converge at a place that is either closer to the lens or farther from it that the first rays.

This phenomenon is described as spherical aberration. It means that there is no place we can put the sensor such that a point on the object will result in a point on the image - a "blur-less" rendering.

There is a place at which the cross-section of the set of all rays, heading toward their respective points of convergence, is the smallest in diameter, and we might think that if we arrange for that to fall at the sensor, we will get the sharpest image possible under the conditions.

But our perception of the diameter of the blur figure is not based on its overall diameter. The rays are more concentrated toward the center, and we judge its size based on that.

It turns out that for a location where the overall diameter is a bit larger than the smallest available, the distribution of the rays is such that the perceived diameter is the smallest. Placing the sensor there will thus result in the sharpest perceived image.

That situation is in fact called the "best focus" situation.

Now the degree of spherical aberration can be mitigated in various ways - using lens surfaces that are not spherical, combining lens elements with different indexes or refraction, and so forth. Doing this is a major preoccupation of lens designers.

But we never can eliminate the phenomenon completely in a context where we will be using the lens to focus the camera at different distances and will use different apertures for different shots. So a compromise design is used.

The remarks above about blur figure size and "best focus" apply to lenses in which spherical aberration has been mitigated (but not eliminated).

Back to phase comparison

The AF subdetectors each work with rays through a small subaperture, essentially all rays passing through the overall aperture at a small range of distances from its center. Accordingly, these rays will very nearly converge at a point (even though there is in fact spherical aberration in the lens).

But "ideal focus" on the subdetectors (where the rays all converge at the same point) does not necessarily correspond with "ideal focus" on the sensor, which gives the best perceived sharpness when all the rays do not focus at the same point.

The best focus correction value (BFCV)

In order to deal with this, just before any phase comparison measurement is made (such as at the beginning of an AF operation), the lens delivers to the body a value called the best focus correction value (BFCV). That says, in effect:

For this lens, at its current focus mechanism position, for some assumed shooting aperture, and for the standard locations of the phase comparison subapertures in EOS bodies, the focus should be shifted from perfect focus, as indicated by the phase comparison system, by this much [the BFCV] to produce best focus.​

The body then takes the focus error as indicated by phase comparison, adjusts it by the recently-received BFCV, and uses that as the focus error for further work.

The ultimate result of this tampering with the indicated focus error is that the focus state to which the AF system seeks to place the lens will differ from where it would seek to place it, based on the actual measured focus error, by that adjustment amount.

(It is just like lowering the temperature to which a thermostat takes the room by heating the thermostat's detector a little, thus tampering with its perception of temperature error.)

Where does the lens get the BFCV it delivers?

The BFCV typically depends on, for any particular lens (model and copy):
• The zoom setting (if a zoom lens)
• The current setting of the focusing mechanism

In fact, in many case, for any combination of those, the lens delivers two BFCV's, one applicable to the spacing of the subapertures in a "standard" AF detector, and one to the (greater) spacing of the subapertures of a "long baseline" AF detector (the ones that require an aperture of f/2.8 or better).

In older Canon EF lenses (the only ones on which I have many details of this, and that only from considerable reading between the lines of the service manuals), the blue delivered is worked up this way:

• There is a "table" of the values, presumably depending on the two parameters mentioned above, that is part of the firmware for the lens model, stored in ROM.

• There is a set of adjustments to the standard table values, to precisely particularize them for this individual copy, burnt into a different area of the ROM at the factor after testing when the lens copy has been assembled.

I do not know whether this is in true ROM - write once only - or if it can be rewritten in the field.​

• There is a further set of adjustments to the standard table values (as already adjusted by the per-copy adjustments in ROM) that can be changed in the field. They are carried by electrical jumper settings (pairs of PC board pads that can be shorted together with solder drops). Often there is a single such adjustment value applicable to all the standard AF detector values and another one applicable to all the "long baseline" AF detector values.

(Yes, you have to take the lens partly apart to get at these. In some cases, this only requires removal of the "tail cone".)

The units of the BFCV

The nature of the BFCV is as an adjustment to the measured longitudinal focus error in image space. It is a distance, and can be thought of as being in mm.

I do not know the actual encoding used to deliver the BFCV or its precision or unit. However, it is fairly certain that the basic unit is the single-sided depth of focus for the maximum aperture of the lens, based on the holy Canon value of circle of confusion diameter limit (CoCDL), 0.035 mm. It turns out that this unit is, in mm:

0.035N (mm)

where N is the f-number.

The increment of the "solder-drop" adjustment is usually 1/2 or 2/5 of the single sided depth of focus.

Enter AF micro-adjustment

Modern Canon EOS bodies have a feature called AF micro adjustment (MA).

This allows a user to store in the camera focus error correction values for individual lenses (that is, the user's copies).

I believe that separate corrections can be stored for different combinations of focus setting and (where applicable) zoom setting.

My belief is that this is just a further layer of the BFCV story. That is, at least for the older EF lenses I mention above, the actual BFCV used by the AF system is the concatenation of:

• The table value from the standard table for that lens model (for the zoom setting, if applicable, and focus setting in effect).
• The adjustment for the particular lens copy stored in the second area in ROM.
• The field adjustment stored in the solder-drop jumpers (there is always a value, although it might be the default, which is handily "no solder drops").
• The applicable MA value stored in the body for this lens (and perhaps this zoom and focus distance situation).

The unit

It has been reported by one of our colleagues here that the unit of MA setting (I understand that one an set values over the range of ±20 units) is 1/8 the single-sided depth of focus.

For an f/2 lens, that would be a distance of 0.00875 mm.

Significance in object space

Keep in mind that this value applies to a shift in focus error in image space (shift in the position of the image with respect to the sensor plane).

We of course become aware of focus error in object space; that is, the departure of the plane of best object focus from the plane containing our "target" ("front- or back-focus").

That is related to the focus error in image space by a straightforward (if not trivial) mathematical relationship, which I will not discuss here. The relationship depends on the focal length of the lens and the distance to the target.

To put it into perspective, assume a focus error of 1/8 the single-sided depth of focus (the magnitude of one unit of MA adjustment) in the direction that leads to back focus.

Assume a 50 mm f/2.0 lens and a target at 3000 mm (118.1 inches, almost 10 feet).

Then the amount of back focus resulting from that image space focus error would be about 36 mm (about 1.4 inches).

Norms

Again by way of perspective, Canon's published norms for overall AF accuracy of an EOS camera are ±1 single-sided depth of focus for "standard" AF detectors, and ±1/3 the single-sided depth of focus for "long baseline" AF detectors.

In the service manuals for some of the early EF lenses, the target for accuracy of the lens itself (presumably in a "perfectly calibrated" test body) is ±1/4 the single-sided depth of focus. Although this is not explicit, this might be for both detector types.

Acknowledgment

My thanks to my colleague Wilba from the ProPhoto Home forum, who pointed out to me recently the role spherical aberration plays in this story.

Best regards,

Doug
 
Last edited:

David Ellsworth

New member
Hi Doug,

Thanks very much for this excellent article. I've often wondered what the root cause of systematic microfocus error is, and I have a clearer idea of it now thanks to what you've written.

I recently bought a 7D, and love having the ability to calibrate microfocus myself. But I have noticed that the calibration is different at every focal length on a zoom lens, and even depends on approximate focus distance on at least one of my lenses. So it's frustrating to only have a single value to change for each lens; it means I have to change it in the field, on the fly, if I want to get the best focus for a given situation (I've added "Microfocus Adjustment" to my 7D's User menu, but it's still an inconvenience, requiring several button presses).

I would very much like to know more about how BFCV is tied to focal length and focus distance, especially in particular lens models (e.g., the 100-400mm) with information on specifically what subranges are used, and whether there is interpolation between them. AFAIK, no Canon camera body yet has firmware that allows the user to fine-tune the microfocus calibration at different focal lengths; so if a lens really does store a table of BFCVs for a range of focal lengths and distances, this means there's a big advantage to sending it in to Canon for calibration. Do you have any specific evidence that this is the case? Is it only the case for super-high end L lenses, or do consumer zooms do it as well?

As it is now, I basically have two choices:
1) Be totally obsessive about adjusting microfocus for every shot, and thus lose a lot of opportunities for photos while I'm fiddling about with settings... and also make the whole experience less enjoyable... but be confident that when I do take a photo, it has the highest possible probability of being in optimum focus.
2) Pick a microfocus adjustment that's a compromise between extremes, possibly weighted towards the shooting situations I'm in more often (especially the ones where I shoot wide open). Rely on random errors in AI Servo to make up for the error in this compromised setting, in order to get some shots that happen to be in perfect focus.

I don't like either of these options. I'd really like to calibrate my own lenses for a range of focal lengths and distances, so that in the field I don't have to worry about it.

Best regards,
David
 
As it is now, I basically have two choices:
1) Be totally obsessive about adjusting microfocus for every shot, and thus lose a lot of opportunities for photos while I'm fiddling about with settings... and also make the whole experience less enjoyable... but be confident that when I do take a photo, it has the highest possible probability of being in optimum focus.
2) Pick a microfocus adjustment that's a compromise between extremes, possibly weighted towards the shooting situations I'm in more often (especially the ones where I shoot wide open). Rely on random errors in AI Servo to make up for the error in this compromised setting, in order to get some shots that happen to be in perfect focus.

I don't like either of these options. I'd really like to calibrate my own lenses for a range of focal lengths and distances, so that in the field I don't have to worry about it.

Hi David,

There may be 2 more choices to consider.
1. Attempt a calibration by a service center, although it might be a frustrating experience. I'm not sure, but I believe that they have the possibility to do a calibration at different focal lengths, getting them more aligned to a common/uniform offset. That would only leave a (more) uniform camera calibration which you can do with the help of AF micro-adjustment, usually for the longer FL (because the wider FL has more DOF anyway).
2. Consider the use of fixed focal length lenses. Sure they are less convenient, but they may help your photography (having to make more deliberate choices before shooting), and are possibly of higher optical quality, and have better lenshoods (not only dimensioned for the wider FL).

Cheers,
Bart
 

Doug Kerr

Well-known member
Hi, David,

I would very much like to know more about how BFCV is tied to focal length and focus distance, especially in particular lens models (e.g., the 100-400mm) with information on specifically what subranges are used, and whether there is interpolation between them.

This is a very interesting matter. I have very limited information on it. Here is some of what I have.

In three early EF-series zoom lenses (on which I have information because it was include in error in an early Canon Service Manual for some other lenses!), it seems as if the zoom setting is noted by a position encoder in 5 bits, apparently 32 ranges in fact being considered. But in these lenses, as well as for several fixed-focal length lenses of the same vintage, it does not seem as if the focus setting is noted at all.

(The lens of course knows of increments in that, but no reference point is assuredly taken when we first power up - that is, the lens does not always run to one end when powered up, for that purpose, and it is not likely that the current absolute position is maintained in non-volatile memory, at least not in older lenses.)

Both the zoom setting and focus setting potentially affect:

• The BFCV

• The focus sensitivity coefficient, a value used by the body to calculate an approximation of the needed change in focus position to overcome the focus error as measured. (This is a very complicated matter I wont deal with here. It is important to note that flaws in this value will not cause focus errors at the time of the shot; they may cause acquisition of best focus to take longer than it might otherwise.)

I have to assume that in later models, the focus setting is noted (in that sense). I do know that in one not-very-later model (but fixed focal length), the EF 50 mm f/1.4, the focus setting is note by a two-bit position encoder, four subranges being in fact delineated.

It is unclear exactly how the values of BFCV are "adjusted" when the lens is calibrated after manufacture. By reading between the lines of some (old) Canon Service Manuals, it seems that this this underlying arrangement might be possible for lenses of a certain genre and vintage:

• A base table, for the lens model, is stored in ROM, part of the ROM load for the lens (which includes its microprocessor firmware and so forth).

• Increments (perhaps separate by subrange, or perhaps global) are stored in another area of write-once ROM when the lens is calibrated during manufacture.

• Anther increment (global) can be set with solder-bridge jumpers on the lens circuit board (someones in 2 bits, sometimes in 3). (This item is certainly so.) This is evidently used for recalibration of the lens at Canon Factory Service Centers and (when there were such things in the US) Canon Authorized Service Centers.

Another possible reading of the tea leaves is this:

• A base table, for the lens model, is stored in ROM, part of the ROM load for the lens (which includes its microprocessor firmware and so forth).

• Increments (perhaps for every range, or perhaps global) are stored in EEPROM when the lens is calibrated during manufacture. This is changed during lens recalibration at Canon Factory Service Centers

• Anther increment (global) can be set with solder-bridge jumpers on the lens circuit board (someones in 2 bits, sometimes in 3). This was evidently used for recalibration of the lens at Canon Authorized Service Centers (when there were such things in the US) .

AFAIK, no Canon camera body yet has firmware that allows the user to fine-tune the microfocus calibration at different focal lengths; so if a lens really does store a table of BFCVs for a range of focal lengths and distances, this means there's a big advantage to sending it in to Canon for calibration. Do you have any specific evidence that this is the case? Is it only the case for super-high end L lenses, or do consumer zooms do it as well?

I don't really know (as the discussion above reveals). At one end, we have the prospect that (and this in fact may vary not only with lens stratum but also vintage) the entire table, a function of two parameters, might be rewritable.

I wish I knew more about this, but I don't at this point.

Best regards,

Doug
 

Doug Kerr

Well-known member
It is important to recognize two different implications of the use of AF micro adjust (as I see it).

One is to overcome inappropriate values of the BFCV returned by a particular lens (at a particular focus setting and (if applicable) at a particular focal length).

The second is done by using an "all lenses" choice when entering the micro adjustment value. What does that do? Take care of the fact that all our lenses have suddenly developed the prank of returning an inappropriate BFCV (all by the same amount, in every situation)? Hardly.

In fact, it allows us to correct in the field an error in the "calibration" of the body.

What does that mean? Well in the secondary image registration technique (often, though not aptly, called "phase detection"), the two little (secondary) images of the part of the field of view regarded by a particular AF detector, falling on two little secondary detectors (part of that AF detector), have their relative positions on their respective secondary detectors compared. Basically, a certain value of that relationship corresponds to the main image falling precisely on the main sensor (the AF target being in "perfect focus").

But what relationship is that? For one thing, it depends on the exact position of the plane containing the secondary detectors with respect to the plane containing the main sensor. While their locations are carefully controlled in manufacture (perhaps involving the use of calibrated shim washers), the result is not precisely uniform. But no matter - one or more constants describing the exact relationship for this body are determined when the body is calibrated in manufacture and stored someplace (probably in EEPROM).

But the values may somehow be "off". Perhaps the mechanical arrangements have somehow shifted, or the camera was manufactured on a Monday.

When we set a micro adjust value "for all lenses", this in in effect makes an offset to the principal constant involved, bringing the situation in the body to what it should be.

Or at least, that's how it looks to me.

Best regards,

Doug
 

David Ellsworth

New member
Hi Doug,

Thanks for your detailed response.

it seems as if the zoom setting is noted by a position encoder in 5 bits, apparently 32 ranges in fact being considered
I tried to repair my 18-55mm IS once, and I saw inside how the focal length and focus distance are detected. Both for zoom and focus independently, there is a row of contacts (maybe 5 or 6) on the inner barrel that moves as the zoom or focus changes. On the outer barrel is something reminiscent of a binary counter, where the set of shorts between contacts changes going up and down. This is what you're talking about, right?

So not only can the focal length be detected (albeit quantized to 32 ranges), but the absolute focus distance can be detected similarly, with no need to keep track of every autofocus event (which would be inaccurate anyway, with errors piling up very quickly, and no detection of manual focus).

I haven't studied this in detail, but I do know that my 100-400mm reports focal length and focus distance, and these are stored in the EXIF and the focus distance is actually rather accurate for close distances (I haven't tested it for far distances) and beyond that, it is very precisely repeatable (the same actual distance will always result in the same reported distance range).

So what I gather from what you've written is, we're basically guessing that Canon lenses use a table of BFCV for different focal length ranges "because they can" — i.e. because the focal length data is reported, and the table is ROM is suspiciously large enough to accommodate focal length ranges — right? Is there any evidence beyond that? (Is there any data on how large the ROM table is for particular lens(es)?)

One test I can think of is to see if there are any discontinuities in the BFCV error (uncorrected BFCV) at the threshold between two focal length ranges.

• The focus sensitivity coefficient, a value used by the body to calculate an approximation of the needed change in focus position to overcome the focus error as measured. (This is a very complicated matter I wont deal with here. It is important to note that flaws in this value will not cause focus errors at the time of the shot; they may cause acquisition of best focus to take longer than it might otherwise.)
Then I must have misinterpreted what you said in the original post. In my first reply I was not talking about adapting to focus distance in that way; I was talking about actual differences in the BFCV at different focus distances. At one point my 100-400mm had a best-working microfocus adjustment of +4 or +5 at a distance of about 10 feet, but +7 to +10 at distances approaching infinity. With this lens the PSF at optimum focus does change as a function of distance at full tele. Also on this lens, moving by one focus unit has a much faster effect on focus at close distances.

Best regards,
David
 

David Ellsworth

New member
In fact, it allows us to correct in the field an error in the "calibration" of the body.

Nice follow-up post, and I'd like to express a point here:

"Best focus" really depends on how much you're magnifying an image. If you're viewing an 18 MP image at 33%, then a PSF that brings almost all the energy into a 13 micron circle will do a good job. But if you view it at 100%, then that same PSF may result in a soft image, and you want one that brings more of the energy into a 4.3 micron circle, at the cost of diffusing some energy around it. (I just know I'm getting the numbers wrong here, but I'm only trying to express a point qualitatively.)

The BFCV that results in sharp focus at one magnification may not be the same as the BFCV that results in sharp focus at the pixel level (which continues to change as megapixel counts increase). A compromise may be necessary for pixel-level sharpness, which results in sharp focus with some haze around it. I've observed this effect when fine-tuning focus using Live View on some lenses.

We really don't know what Canon's optimizing for when they calibrate microfocus. In fact I would argue that it's really up to individual taste. A pixel peeper will definitely want optimum focus at the pixel level. Some photographers may prefer to optimize focus at a larger scale.

Also, I have a hypothesis that even if Canon calibrates all your lenses for their reference body, there's no guarantee that the offset for your camera body will be the same on each lens, or even at each focal length on a particular lens, even if Canon has done the calibration perfectly according to their standards of optimum focus... and even if your standards are the same as theirs. The derivative of focus concentration with respect to perpendicular-to-focal-plane movement, at the visual detail level, may be different than that derivative at the phase-detection sensor's detail level. (Again, I'm sure I'm getting this a bit wrong, but I hope it gets my idea across.)

But if your visual standard of optimum focus differs from that of the Canon tech who did the calibration, then you're probably really up the creek, because there's probably not very much correlation between best focus at one detail level versus another across a range of lenses and focal lengths on each lens.
 

Doug Kerr

Well-known member
Hi, David,

I tried to repair my 18-55mm IS once, and I saw inside how the focal length and focus distance are detected. Both for zoom and focus independently, there is a row of contacts (maybe 5 or 6) on the inner barrel that moves as the zoom or focus changes. On the outer barrel is something reminiscent of a binary counter, where the set of shorts between contacts changes going up and down. This is what you're talking about, right?

Indeed. But for certain old lenses, the manual details suggest that there is only such an encoder for zoom position, not also for focus barrel position.

So not only can the focal length be detected (albeit quantized to 32 ranges), but the absolute focus distance can be detected similarly, with no need to keep track of every autofocus event (which would be inaccurate anyway, with errors piling up very quickly, and no detection of manual focus).
Indeed, and I assume that most modern lenses have benefit of this.

I haven't studied this in detail, but I do know that my 100-400mm reports focal length and focus distance, and these are stored in the EXIF and the focus distance is actually rather accurate for close distances (I haven't tested it for far distances) and beyond that, it is very precisely repeatable (the same actual distance will always result in the same reported distance range).
Sure.

So what I gather from what you've written is, we're basically guessing that Canon lenses use a table of BFCV for different focal length ranges "because they can" — i.e. because the focal length data is reported, and the table is ROM is suspiciously large enough to accommodate focal length ranges — right?
Yes, that is my outlook.

Is there any evidence beyond that? (Is there any data on how large the ROM table is for particular lens(es)?)

I have no information on that - I wish I did.

One test I can think of is to see if there are any discontinuities in the BFCV error (uncorrected BFCV) at the threshold between two focal length ranges.
Sure. I'm not sure how to do that. It might manifest in discontinuities in the final error.

We can also get some insight by seeing the limits of focus confirmation in manual focus, using a micrometer slide to move the camera precisely. My colleague Wilba has done some work in that way. He had drawn some conclusions from discontinuities in the data, but it has been a few weeks since I saw that and I will have to review his reports to be certain what the observations and conclusions were. (I'm an old guy, you know, and this stuff slips into the archives pretty quickly!)


Then I must have misinterpreted what you said in the original post. In my first reply I was not talking about adapting to focus distance in that way; I was talking about actual differences in the BFCV at different focus distances. At one point my 100-400mm had a best-working microfocus adjustment of +4 or +5 at a distance of about 10 feet, but +7 to +10 at distances approaching infinity. With this lens the PSF at optimum focus does change as a function of distance at full tele. Also on this lens, moving by one focus unit has a much faster effect on focus at close distances.

I'm not sure what I said in my "original post" or what it seems to say. But in fact:

• It seems quite credible that the "optimal" BFCV would differ with focus distance. This is particularly likely with lenses not using "all group" focusing (and most don't), as we can well imagine that moving the elements with respect to each other could "screw with" the spherical aberration correction. It is also quote possible that longitudinal magnification considerations could change the SA.I'll see if there is any reference to that in the textbooks.

• If so, it seems likely that the lens would provide different values for different focus settings (probably by range, but of course I can't rule out the possibility of interpolation between table values in the lens).

Best regards,

Doug
 

Doug Kerr

Well-known member
Hi, DAvid,
Nice follow-up post, and I'd like to express a point here:

"Best focus" really depends on how much you're magnifying an image. If you're viewing an 18 MP image at 33%, then a PSF that brings almost all the energy into a 13 micron circle will do a good job. But if you view it at 100%, then that same PSF may result in a soft image, and you want one that brings more of the energy into a 4.3 micron circle, at the cost of diffusing some energy around it. (I just know I'm getting the numbers wrong here, but I'm only trying to express a point qualitatively.)
A very good point, and at first blush that outlook seems reasonable to me.

The BFCV that results in sharp focus at one magnification may not be the same as the BFCV that results in sharp focus at the pixel level (which continues to change as megapixel counts increase). A compromise may be necessary for pixel-level sharpness, which results in sharp focus with some haze around it. I've observed this effect when fine-tuning focus using Live View on some lenses.

I assume you are speaking of "image-on-sensor to imag-as-seen" magnification, not image magnification. And then viewing distance gets into the act.

We really don't know what Canon's optimizing for when they calibrate microfocus.
I'm not sure Canon "calibrates microfocus"; that is, what they change in the lens in depot may not be the same variable that we can adjust in the body ourselves.

For example, it may well be that in calibrating a lens, they can change multiple table values.

Also, I have a hypothesis that even if Canon calibrates all your lenses for their reference body, there's no guarantee that the offset for your camera body will be the same on each lens, or even at each focal length on a particular lens, even if Canon has done the calibration perfectly according to their standards of optimum focus... and even if your standards are the same as theirs. The derivative of focus concentration with respect to perpendicular-to-focal-plane movement, at the visual detail level, may be different than that derivative at the phase-detection sensor's detail level. (Again, I'm sure I'm getting this a bit wrong, but I hope it gets my idea across.)

I understand, and the presence of these subtleties may in fact be part of the overall situation.

But if your visual standard of optimum focus differs from that of the Canon tech who did the calibration, then you're probably really up the creek, because there's probably not very much correlation between best focus at one detail level versus another across a range of lenses and focal lengths on each lens.

We cannot assume that the adjustment is based on human (or even technician!) interpretation of the image on the main sensor of the "tool" body. It may well be done by software.

Another very importance wrinkle is that the visually-optimum "offset" from paraxial focus (the perfect focus for an infinitesimal aperture, where SA approaches zero) with regard to SA will depend on the taking aperture. How if at all that is taken into account I don't know.

Best regards,

Doug
 

Zenon Char

New member
Hi I am a new member. I found this thread via WilbaW on Fred Miranda. There was a link to - Busted! The Myth of Open-loop Phase-detection Autofocus on DPreview and the link to this thread was in that article.

I know this is a pretty old thread but I found it fascinating. I must admit it got a little technical for me. Long ago gave up on MA. Forget zooms. Even with prime lenses I get different results at different focal lengths. Even the camera manuals state for best results adjust MA at the location.

What really got to me was there is no real official procedure for this. There must be 2 dozen methods out there. Even Chuck Westfall states at the end of if his blurb on it "if you can do better go for it" ??? $5000 worth of precision equipment and we are kinda of going for it. I hope the technician working on the engine of the plane I'm about to fly in does not ignore the technical manual goes for a better home made adjustment.

This is from Canon and on page 6 they say the same thing.

http://www.learn.usa.canon.com/resources/articles/2011/af_microadjustment_article.shtml

I decided as much of a pain as it is to send my gear to Canon to have it serviced by trained technicians who have the proper equipment. Not been easy with the MA devotees out there. May happen here but that is fine by me.

I was just wondering what your take on my thoughts about MA?

Thanks in advance.
 

Cem_Usakligil

Well-known member
Hi Zenon,

Welcome to OPF, glad you have found us. Please feel free to join the ongoing threads or create new ones, we are a very friendly folk. In any case, I hope to be seeing you around.

Coming back to MA, Bart van der Wolf has created two excellent targets which can be used to easily micro adjust your camera/lens. The whole discussion can be read here, if you haven't done so already.

Cheers,



Hi I am a new member. I found this thread via WilbaW on Fred Miranda. There was a link to - Busted! The Myth of Open-loop Phase-detection Autofocus on DPreview and the link to this thread was in that article.

I know this is a pretty old thread but I found it fascinating. I must admit it got a little technical for me. Long ago gave up on MA. Forget zooms. Even with prime lenses I get different results at different focal lengths. Even the camera manuals state for best results adjust MA at the location.

What really got to me was there is no real official procedure for this. There must be 2 dozen methods out there. Even Chuck Westfall states at the end of if his blurb on it "if you can do better go for it" ??? $5000 worth of precision equipment and we are kinda of going for it. I hope the technician working on the engine of the plane I'm about to fly in does not ignore the technical manual goes for a better home made adjustment.

This is from Canon and on page 6 they say the same thing.

http://www.learn.usa.canon.com/resources/articles/2011/af_microadjustment_article.shtml

I decided as much of a pain as it is to send my gear to Canon to have it serviced by trained technicians who have the proper equipment. Not been easy with the MA devotees out there. May happen here but that is fine by me.

I was just wondering what your take on my thoughts about MA?

Thanks in advance.
 
Top