How can lens focus ever be "off" in a way that AFMA fixes it?

[Mt Spokane Photography]

Canon cameras contain information about all the EF lenses and the commands to send to them. Third party lenses tell the camera that they are a Canon lens and then translate the command they receive to their lens. This can cause even another step that adds to inaccuracy, but it can be adjusted by AFMA as well.

Its a lot more complex that a person might think.

Again, all of that would be solved by closing the loop. Then all you'd need is a correction for sensor/vs AF array, which would be body specific and programmed by Canon at the factory.

It is sounding like the whole phase detect AF system is fully open loop, which really surprised me. Is Nikon like this too? Do they also have an AFMA type feature on their bodies?

Wish there were a Canon engineer I could speak to this about, would be a fascinating discussion!

Its not totally open loop, there is checking going on. There is information about how it works, but the fine details are not disclosed.

There are two possible types of errors.

1. The Camera tells the lens to move to 8.2 ft, but due to errors in the camera, it should actually be in focus at 8.1 ft.

2. The lens receives a command from the camera to move to 8.2 ft but due to internal tolerances and errors, it moves to 8.3 ft. It does tell the camera when it has moved, which closes the loop.

When both of these happen, there is a significant focus error. AFMA Adjusts the camera command to compensate for this. Sometimes a lens or body error compensate and the errors cancel out. In the example above, the lens could have actually been in focus at 8.1 ft even though it told the camera that it had moved to 8.2 ft as commanded. The image would have been in perfect focus and the owner boasting that his lens was perfect and needed no AFMA!


If you send your camera and lens to Canon, they check each separately and write new parameters in the internal firmware so that both are more accurate.

 

[cocopop05]

Here are my thoughts, they may be wrong

Contrast detection (Live view focusing, taken from image sensor itself) should be very accurate.

Phase detection (autofocus sensors, separate to image sensor) can have inaccuracies. This is where AFMA can resolve issues.

All I know is that I bought a Canon 5D MIII with 24-105 F/4 L, it was not pin sharp when using phase detection. I used the EOS utility with the tethered method (http://ophrysphotography.co.uk/pages/tutoriallensmicroadjustment.htm) to set my AMFA.

I found at all focal lengths of the zoom lens a +5 adjustment meant the sharpest results as possible.

After setting that value, the autofocus is almost always dead-on accurate.

Whether you think AFMA can in principle work or not, in practice with my experience, it is an invaluable feature.

 

[TrumpetPower!]

It's easy. You do need a stable tripod, and lots of light in the target (sun or tungsten/halogen, not fluorescent or LED).

Ideal and pretty easy is to put your tripod on a concrete pad like a sidewalk or your driveway and tape the target to a heavy wall like the side of your house. Do this in direct sunlight, and you've got the perfect setup for FoCal to work its magic.

Cheers,

b&

 

[jrista]

Here's a question that's been bugging me.

Why does AFMA fix anything? As I understand modern phase detect focus systems, focus is a closed loop sort of thing The camera tells the lens to change it's focus, and when the camera body sees things are in focus (the phase detect sensors say focus has been achieved) it stops moving the focus on the lens.

How can this "go wrong" in a way that AFMA can fix it? The way I see it is the only way things can go wrong is if the sensor is physically in the wrong spot, meaning an "in focus" in the phase detect sensor doesn't correspond to in focus at the sensor.

Where the lens is shouldn't matter?

So what am I missing?

Thanks!

You pretty much have it. AFMA is purely an algorithmic thing, and all it does is tell the AF drive system to adjust focus by a small amount to ensure that when the mirror lifts and the image resolves, the focus plane is centered at the sensor plane.

The reason this is necessary is simply a matter of calibration. All devices are manufactured to certain tolerances. Higher end equipment that costs more is generally manufactured to tighter tolerances, but even then, it is extremely difficult and extremely costly to keep things "perfect". There are also three separate things in play when it comes to focus...the lens, the sensor, AND the AF unit. It can be difficult to perfectly align two interacting elements of a system, let alone more than two.

You also have to realize that even at its extremes of +/- 20, AFMA is making MICROSCOPIC adjustments to the focal plane. Reality is compressed within the mirrorbox, so a significant AFMA setting is likely only adjusting the focus plane by MICROMETERS. We aren't even talking about millimeters here (changing the focus plane by a millimeter would likely result in radical defocus)...we are talking about an incredibly FINE adjustment! It is all just a matter of calibration, and it only actually changes FIRMWARE CONFIGURATION in the body. Making an AFMA change does not actually change any physical attribute of the camera body or lens. Making AFMA adjustments on multiple different bodies for the same lens simply reconfigures each body, so the lens can still be used on each one and it should work ideally on each one once properly "calibrated."

 

[al2]

It's easy. You do need a stable tripod, and lots of light in the target (sun or tungsten/halogen, not fluorescent or LED).

Neuro's comment about fluorescent and LED light sources is true for those that are equipped with old-style magnetic ballasts that operate at 60Hz. The light output fluctuates at 120 Hz and at any shutter speed faster than 1/120 there is a high probability that the image will be captured between peak light levels and the exposure will be incorrect. Fluorescent and LED lights with electronic ballasts operate at approximately 20,000 Hz, far in excess of any shutter speed you will be using.

 

[jrista]

Canon cameras contain information about all the EF lenses and the commands to send to them. Third party lenses tell the camera that they are a Canon lens and then translate the command they receive to their lens. This can cause even another step that adds to inaccuracy, but it can be adjusted by AFMA as well.

Its a lot more complex that a person might think.

Again, all of that would be solved by closing the loop. Then all you'd need is a correction for sensor/vs AF array, which would be body specific and programmed by Canon at the factory.

It is sounding like the whole phase detect AF system is fully open loop, which really surprised me. Is Nikon like this too? Do they also have an AFMA type feature on their bodies?

Wish there were a Canon engineer I could speak to this about, would be a fascinating discussion!

You should read Roger Cicala's LensRentals blog. He does extensive testing with lenses, and in his assessment, modern AF systems, including Canon's, ARE closed loop! Here's a good read:

http://www.lensrentals.com/blog/2012/08/autofocus-reality-part-3b-canon-cameras

And to quote, from Canon's own patent:

In order to achieve this objective, this invention provides a camera system comprising: a first focus detection unit, a second focus detection unit, a stepping motor that drives a focusing lens, . . . or a rotation detector, which detects the rotation . . . of the motor. . . The control circuit performs closed-loop control, based on the output of the rotation detector to control the motor.

 

[skitron]

I've often pondered the same question as OP, so dug around and found this post:

http://photographylife.com/how-phase-detection-autofocus-works

Easy read, but enough info to understand the technical.

Basically there are known and expected phase detection errors at the individual phase detect cell level and those are dealt with in code when the camera is manufactured. AFMA allows you to "mop up" at a global level.

The phase detection process as a whole is a hybrid of open and closed loops. Basically if a phase error is detected beyond what has already been documented in code during manufacturing, instructions are sent to move the lens to a specific focus. It keeps checking the results and sending further instructions to the lens until it confirms. So in this respect it is closed loop.

However, the Phase Detection sensor and the Image Detection sensor are two entirely different physical entities operating on two distinct and separate light paths. By definition, using the PD sensor to derive an ID sensor focus state is open loop in nature. An "in focus" state on the PD sensor results in an "unknown focus" state for the ID sensor evey single time. And it is only by making assumptions about the physical relationship between the PD and ID light paths and sensors that the ID sensor focus state is derived and in fact *assumed*.

LOL, after seeing how this works I'm surprised it functions as well as it does.

The way to close all process loops is to place the PD cells in the same physical light path and focal plane as the ID cells. Seems like I heard something along those lines recently.

 

[Mt Spokane Photography]

.

You also have to realize that even at its extremes of +/- 20, AFMA is making MICROSCOPIC adjustments to the focal plane. Reality is compressed within the mirrorbox, so a significant AFMA setting is likely only adjusting the focus plane by MICROMETERS. We aren't even talking about millimeters here (changing the focus plane by a millimeter would likely result in radical defocus)...we are talking about an incredibly FINE adjustment! It is all just a matter of calibration, and it only actually changes FIRMWARE CONFIGURATION in the body. Making an AFMA change does not actually change any physical attribute of the camera body or lens. Making AFMA adjustments on multiple different bodies for the same lens simply reconfigures each body, so the lens can still be used on each one and it should work ideally on each one once properly "calibrated."

Older DSLR's like the Canon XTi had a small set screw that let you change the position of the sub mirror. This was a mechanical adjustment for AF that was easy to do. I judt turned the screw 2 clicks with a tiny allen wrench and adjusted my AF system to be accurate. After the XTi, Canon went to all electronic adjustments on the Rebel line.

http://www.astrosurf.com/buil/autofocus/adjust.htm

 

[Loswr]

Canon cameras contain information about all the EF lenses and the commands to send to them. Third party lenses tell the camera that they are a Canon lens and then translate the command they receive to their lens. This can cause even another step that adds to inaccuracy, but it can be adjusted by AFMA as well.

Its a lot more complex that a person might think.

Again, all of that would be solved by closing the loop. Then all you'd need is a correction for sensor/vs AF array, which would be body specific and programmed by Canon at the factory.

It is sounding like the whole phase detect AF system is fully open loop, which really surprised me. Is Nikon like this too? Do they also have an AFMA type feature on their bodies?

Wish there were a Canon engineer I could speak to this about, would be a fascinating discussion!

You should read Roger Cicala's LensRentals blog. He does extensive testing with lenses, and in his assessment, modern AF systems, including Canon's, ARE closed loop! Here's a good read:

http://www.lensrentals.com/blog/2012/08/autofocus-reality-part-3b-canon-cameras

And to quote, from Canon's own patent:

In order to achieve this objective, this invention provides a camera system comprising: a first focus detection unit, a second focus detection unit, a stepping motor that drives a focusing lens, . . . or a rotation detector, which detects the rotation . . . of the motor. . . The control circuit performs closed-loop control, based on the output of the rotation detector to control the motor.

Yes, Canon patented that. But...did they actually implement it, and if so, in which lenses and bodies?

 

[jrista]

Canon cameras contain information about all the EF lenses and the commands to send to them. Third party lenses tell the camera that they are a Canon lens and then translate the command they receive to their lens. This can cause even another step that adds to inaccuracy, but it can be adjusted by AFMA as well.

Its a lot more complex that a person might think.

Again, all of that would be solved by closing the loop. Then all you'd need is a correction for sensor/vs AF array, which would be body specific and programmed by Canon at the factory.

It is sounding like the whole phase detect AF system is fully open loop, which really surprised me. Is Nikon like this too? Do they also have an AFMA type feature on their bodies?

Wish there were a Canon engineer I could speak to this about, would be a fascinating discussion!

You should read Roger Cicala's LensRentals blog. He does extensive testing with lenses, and in his assessment, modern AF systems, including Canon's, ARE closed loop! Here's a good read:

http://www.lensrentals.com/blog/2012/08/autofocus-reality-part-3b-canon-cameras

And to quote, from Canon's own patent:

In order to achieve this objective, this invention provides a camera system comprising: a first focus detection unit, a second focus detection unit, a stepping motor that drives a focusing lens, . . . or a rotation detector, which detects the rotation . . . of the motor. . . The control circuit performs closed-loop control, based on the output of the rotation detector to control the motor.

Yes, Canon patented that. But...did they actually implement it, and if so, in which lenses and bodies?

Read the article I linked. According to Roger, all of Canon's newer lenses, dating as far back as the 70-200 f/2.8 L II I think, have rotation detectors. The new generation of Canon bodies, at least the single-digit ones, seem to have improved firmware to support closed-loop operation. That includes the 1D X and 5D III at the very least, both of which score PDAF marks as high as CDAF. I am not sure if the 7D has the necessary drive firmware...there was a recent firmware update, but I don't know if that included better drive. In my experience, the 7D with the new firmware focused WAY better with the 300, 500, and 600mm Mark II lenses I rented than the 100-400mm lens. That is largely anecdotal, and it could just be the relatively poor quality of the 100-400mm lens. I couldn't say whether the 6D had closed loop operation, although I don't see why it wouldn't.

 

[Mt Spokane Photography]

Canon cameras contain information about all the EF lenses and the commands to send to them. Third party lenses tell the camera that they are a Canon lens and then translate the command they receive to their lens. This can cause even another step that adds to inaccuracy, but it can be adjusted by AFMA as well.

Its a lot more complex that a person might think.

Again, all of that would be solved by closing the loop. Then all you'd need is a correction for sensor/vs AF array, which would be body specific and programmed by Canon at the factory.

It is sounding like the whole phase detect AF system is fully open loop, which really surprised me. Is Nikon like this too? Do they also have an AFMA type feature on their bodies?

Wish there were a Canon engineer I could speak to this about, would be a fascinating discussion!

You should read Roger Cicala's LensRentals blog. He does extensive testing with lenses, and in his assessment, modern AF systems, including Canon's, ARE closed loop! Here's a good read:

http://www.lensrentals.com/blog/2012/08/autofocus-reality-part-3b-canon-cameras

And to quote, from Canon's own patent:

In order to achieve this objective, this invention provides a camera system comprising: a first focus detection unit, a second focus detection unit, a stepping motor that drives a focusing lens, . . . or a rotation detector, which detects the rotation . . . of the motor. . . The control circuit performs closed-loop control, based on the output of the rotation detector to control the motor.

Yes, Canon patented that. But...did they actually implement it, and if so, in which lenses and bodies?

I think it may be used in the STM lens. However, that only means that the internal lens stepping motor operates as a closed loop.

A Phase Detect System for all manufacturers operates as a limited closed loop, but that only assures that the lens tells the camera that it has moved where it was told to go.

A lens cannot measure the actual distance to the subject, its calibrated and the resulting values are stored in the lens. The camera determines the distance. As others have pointed out, the thickness of the lens mount, position of the sub mirror, position of the sensor all have tolerances. The camera is calibrated at the factory to compensate for tolerance buildup as are the lenses.

However, things do go wrong and AFMA merely adjusts the offset of the focus + or -. On the 5D MK III, 1D X it allows two adjustments for zooms, wide and telephoto, and then interpolates the values between the extremes.
It the best setting for a 24-70mm lens is -2 at 24mm, and +2 at 70mm, it will set AFMA to zero at approx 47mm.

 

[Loswr]

Read the article I linked. According to Roger, all of Canon's newer lenses, dating as far back as the 70-200 f/2.8 L II I think, have rotation detectors.

Oh, I've read it - including the part where you seem to have missed the 10 year difference between when you suggest rotation detectors were incorporated, vs. when Roger does, or maybe that's a typo on Roger's part...but it makes me wonder what was going on for those 10 years...

A Phase Detect System for all manufacturers operates as a limited closed loop, but that only assures that the lens tells the camera that it has moved where it was told to go.

This was my point earlier - the closed loop is 'look-move-confirm' but not 'look-move-confirm-look'.

 

[LetTheRightLensIn]

Canon cameras contain information about all the EF lenses and the commands to send to them. Third party lenses tell the camera that they are a Canon lens and then translate the command they receive to their lens. This can cause even another step that adds to inaccuracy, but it can be adjusted by AFMA as well.

Its a lot more complex that a person might think.

Again, all of that would be solved by closing the loop. Then all you'd need is a correction for sensor/vs AF array, which would be body specific and programmed by Canon at the factory.

It is sounding like the whole phase detect AF system is fully open loop, which really surprised me. Is Nikon like this too? Do they also have an AFMA type feature on their bodies?

Wish there were a Canon engineer I could speak to this about, would be a fascinating discussion!

You should read Roger Cicala's LensRentals blog. He does extensive testing with lenses, and in his assessment, modern AF systems, including Canon's, ARE closed loop! Here's a good read:

http://www.lensrentals.com/blog/2012/08/autofocus-reality-part-3b-canon-cameras

And to quote, from Canon's own patent:

In order to achieve this objective, this invention provides a camera system comprising: a first focus detection unit, a second focus detection unit, a stepping motor that drives a focusing lens, . . . or a rotation detector, which detects the rotation . . . of the motor. . . The control circuit performs closed-loop control, based on the output of the rotation detector to control the motor.

Apparently the rotation detector measures how much it spins and where it actually lands since the breaking of the rotation may not be perfect and it might overshoot a trace or not and then it applies a correction for how much over or undershoot it measured. Apparently the newest Canon lenses such as the 24-70 II, 24 IS, 28 IS and a few others have a higher resolution rotation detector. Apparently the lens is not able to use its own rotation detector to apply corrections internally but must send the info back to the body which then sends out a new commands to rotate a bit (and hopefully with this smaller command the new amount doesn't overshoot or undershoot to as great a degree as during the initial rotation) so apparently you also need a new body that has higher precision to be able to take advantage of the higher resolution rotation detectors and apparently that is why only the 5D3 and 1DX make use of the new ultra-precision detectors in some of the new lenses.

To be honest the whole system sounds crazily over complex and you'd think some sort of much more direct closed loop would be less prone to all sorts of issues, at least for non-AI Servo AF, but I assume they know what they are doing so there are most likely very good reasons for it all (although perhaps some are only due to the tech avail back in the 80s??).

 

[LetTheRightLensIn]

Canon cameras contain information about all the EF lenses and the commands to send to them. Third party lenses tell the camera that they are a Canon lens and then translate the command they receive to their lens. This can cause even another step that adds to inaccuracy, but it can be adjusted by AFMA as well.

Its a lot more complex that a person might think.

Again, all of that would be solved by closing the loop. Then all you'd need is a correction for sensor/vs AF array, which would be body specific and programmed by Canon at the factory.

It is sounding like the whole phase detect AF system is fully open loop, which really surprised me. Is Nikon like this too? Do they also have an AFMA type feature on their bodies?

Wish there were a Canon engineer I could speak to this about, would be a fascinating discussion!

You should read Roger Cicala's LensRentals blog. He does extensive testing with lenses, and in his assessment, modern AF systems, including Canon's, ARE closed loop! Here's a good read:

http://www.lensrentals.com/blog/2012/08/autofocus-reality-part-3b-canon-cameras

And to quote, from Canon's own patent:

In order to achieve this objective, this invention provides a camera system comprising: a first focus detection unit, a second focus detection unit, a stepping motor that drives a focusing lens, . . . or a rotation detector, which detects the rotation . . . of the motor. . . The control circuit performs closed-loop control, based on the output of the rotation detector to control the motor.

Yes, Canon patented that. But...did they actually implement it, and if so, in which lenses and bodies?

My understanding is that it goes back quite a ways at the least.
However, my understanding is that only very recently did they up the precision for the rotation detector+camera body system and that only the 5D3/1DX are capable of sending back micro-precision rotation adjustments and that only lenses starting around the time of 70-300L or so release have the ultra-precision rotation detector in them. Lenses before that have a lower precision rotation detector and pre-5D3/1DX and even current but lower end bodies don't understand the extra bits of precision. I'm not sure if the lenses have always had rotation detectors or not.

 

[hgraf]

Thanks all for the info, lots over very good hints and links in this thread.

So, to summarize, in case I've got something wrong:

There are basically 2 major types of error when PD focusing:
1) Misalignment between the PD array and the sensor plane
2) Error in the lens with regards to responding to a position request, i.e. command "focus to 8.1ft" results in actual focus to 8.2ft

The first error is mostly lens independent, and should be fully accounted for during manufacture, but can result from physical changes in the body, i.e. dropping, wear, etc. Hence AFMA helps here.

The second error is almost exclusively lens dependent, and can result from a variety of factors, including error in position sensor in the lens, and physical changes in the lens, i.e. wear, dropping, etc. Here AFMA has the most important role in fixing the problem. It also explains how the 3rd party systems can correct for focus issues just by modifying firmware in the lens.

Being an engineer and having taken courses in control theory had me thinking that by "closed loop" that meant the PD array was used to close the loop, camera checks PD array, moves lens, checks PD array, moves lens, etc. until PD array reports focus.

I didn't consider that another type of closed loop is camera checks PD array, tells lens to move to a position and lens reports done. While technically closed loop, it isn't as simple as the first case, and exposes the system to more sources of error.

Engineering is all about compromise, so the fact that Canon does it the way it does points to there being a benefit to the more complicated method, my guess would be speed.

It's similar to how the contrast detect focusing in live view is "slow", contrast detect doesn't deliver as reliable a "your focus is off by this much" sort of error signal, so the loop is closed by the contrast detect, hence the hunting you see while in live mode.

So thank you all for clearing this up for me!

One last question: how does Nikon do it? Does it also have an AFMA type tool?

Thanks!

 

[friedmud]

Nikon is similar. AFMA is called "AF Fine Tune".

I found this thread informative... my Nikon D600 is spot on wih my 50mm 1.8G and my 70-300... but is quite off at the wide end of my 24-70 f/2.8. I'm sending it in for adjustment soon (after the next round of trips) but I could never work out how the focus could be perfect for all of my other lenses and be off for one. This thread definitely gave some insight into that!

 

[Rienzphotoz]

It's easy. You do need a stable tripod, and lots of light in the target (sun or tungsten/halogen, not fluorescent or LED).

Ideal and pretty easy is to put your tripod on a concrete pad like a sidewalk or your driveway and tape the target to a heavy wall like the side of your house. Do this in direct sunlight, and you've got the perfect setup for FoCal to work its magic.

Cheers,

b&

neuroanatomist & TrumpetPower, Thanks

 

[jrista]

Read the article I linked. According to Roger, all of Canon's newer lenses, dating as far back as the 70-200 f/2.8 L II I think, have rotation detectors.

Oh, I've read it - including the part where you seem to have missed the 10 year difference between when you suggest rotation detectors were incorporated, vs. when Roger does, or maybe that's a typo on Roger's part...but it makes me wonder what was going on for those 10 years...

A Phase Detect System for all manufacturers operates as a limited closed loop, but that only assures that the lens tells the camera that it has moved where it was told to go.

This was my point earlier - the closed loop is 'look-move-confirm' but not 'look-move-confirm-look'.

Hmm, that is not my experience. And in one of the articles in that series Roger posted, he clearly stated that the AF system seems to fully adjust the lens as necessary in a single press of the AF button in single-shot mode to fully lock AF. His expectation was that it the AF system only did "look-move-confirm" rather than "look-move-confirm-move-confirm" that multiple presses of the AF button would further refine focus. That was not the case. AF was fully and precisely locked the FIRST time, and all subsequent attempts to AF resulted in no change...but only with the latest gear...i.e. a 5D III with one of the new STM lenses, or one of the new Mark II Great Whites. Older cameras still seem to operate on a partially open loop.

 

[Loswr]

I think the situation may be a bit more complex than that. One observation I made when doing manual AFMA (LensAlign Pro) with the 85L II (a lens with AF slow enough that one can actually observe the focusing with the focus distance window) was that the 7D seemed to lock on faster (or at least in a different manner, I didn't time it) than the 5DII. Specifically, when focusing from infinity or MFD, the 7D would move the focusing group in one direction then stop, whereas the 5DII would consistently overshoot by a little bit, then move slightly in the opposite direction.

 

[jrista]

I think the situation may be a bit more complex than that. One observation I made when doing manual AFMA (LensAlign Pro) with the 85L II (a lens with AF slow enough that one can actually observe the focusing with the focus distance window) was that the 7D seemed to lock on faster (or at least in a different manner, I didn't time it) than the 5DII. Specifically, when focusing from infinity or MFD, the 7D would move the focusing group in one direction then stop, whereas the 5DII would consistently overshoot by a little bit, then move slightly in the opposite direction.

Usually my 7D will focus in a single movement, but that is not always the case. There are times when it will make on big move, then one or two small moves, then issue AF lock confirmation in the viewfinder. It is usually VERY quick at it, but you can just barely feel the lag and see those last couple of movements. I have my AF button set to a rear button on the body, so it is separate from the shutter button, so there is no confusion about what is causing the movements.

If Canon still used an open/partial open AF system, I don't believe the secondary movements would occur. I have not used a 1D X, so I am not sure if it might do the same thing at times. It has a vastly superior AF system as well, so the chances that you even need a second or third movement are probably so low as to be practically unheard of...especially with newer lenses that have updated electronics and firmware to support a more accurate AF system overall.