[Marc Sabatella]

The body does the AF, not the lens. If the body works well with a number of lenses, how can one lens be off?

The body turns the screw. But the lens has to respond to that screw-turning correctly. The body does an initial check, determines how far focus is off and in which direction, then calculates how many turns of the screw it needs to fix it. If the lens is "off" in some way, it won't be in focus after the "correct" number of turns.

Even if the lens is odd in some respect, the feedback loop in the AF mechanisms should keep adjusting until the subject is in focus.

Kind of, sort of, but it doesn't really check that hard, it seems - at least, not once it gets enough info to decide how much adjustment is needed.

[Ben_Edict]

Lighting can throw it off as well. It seems, from some of my reading, that tungsten lighting does not focus the same as daylight. I suspect that colour temperature will affect focus somewhat. Lenses [used to] have focusing marks for infrared for just this problem, but at a greater difference in frequency, of course. I suspect that tungsten would require a technical correction to the focus, and the detector might be thrown off.

I don't think so. It might just be, that the AF sensor has less sensitivity in the reds and thus might not "see enough light" for fast and accurate focusing. Tungsten lighting is usually much dimmer than daylight sources and if the reds are ignored by the sensor, it is even more so.

Sorry, this is only guesswork, as I don't know the frequency response of the AF sensors.

Ben

[Finn]

I don't think so. It might just be, that the AF sensor has less sensitivity in the reds and thus might not "see enough light" for fast and accurate focusing. Tungsten lighting is usually much dimmer than daylight sources and if the reds are ignored by the sensor, it is even more so.

Sorry, this is only guesswork, as I don't know the frequency response of the AF sensors.

Ben

I'm hoping that's the case for me. I ran a test of my lenses (under tungsten light) so I could adjust their focus (thanks, K20D!). For each lens I shot five frames, unfocusing and refocusing between each frame. For each lens I had five DIFFERENT results for FF/BF, sometimes differing by A LOT, like way past the 3mm mark.

I think I'm going to have to try again in better light...

[Canada_Rockies]

I don't think so. It might just be, that the AF sensor has less sensitivity in the reds and thus might not "see enough light" for fast and accurate focusing. Tungsten lighting is usually much dimmer than daylight sources and if the reds are ignored by the sensor, it is even more so.

Sorry, this is only guesswork, as I don't know the frequency response of the AF sensors.

Ben

I was working on the concept that tungsten light balance is around 3,200 degrees while direct noon sunlight runs about twice that, circa 6,500 degrees. Infrared is just beyond the lower end of the red spectrum. I have not yet found a colour temperature range for infrared. Depending on the optical formula of the lens and the type of elements and coatings, lenses can misfocus longer wavelengths.

I am also not sure that the AF sensor in the Pentax has the IR cutoff filter that is in front of the sensor. If they do not have identical cutoff filters, it is possible for the AF system to "see" in the red range more than the sensor, and thus mis-focus.

All of the above is SPECULATION by me, the POSTER. Take with a grain of salt.

[Class A]

It seems, from some of my reading, that tungsten lighting does not focus the same as daylight.

Yes, it is a well known fact that AF depends on the colour temperature of the light. It applies to other camera brands as well, but they may have figured out some ways to better compensate for the effect.

But the lens has to respond to that screw-turning correctly.

Not really, if it is a closed feedback loop. The body turns, checks, and if it is not enough it turns further or turns a bit back.

If the body just calculated the number of turns and performed them, then a lens would never hunt back and forth. There has to be a final check after the control has been exercised. Perhaps -- pure speculation -- the test has a too high degree of tolerance. Lenses that get it right would then fare better.

I don't think so. It might just be, that the AF sensor has less sensitivity in the reds and thus might not "see enough light" for fast and accurate focusing.

That doesn't explain the consistent misfocussing under tungsten light. It is always FF. AFAIK, it determines the focus consistently, but just a bit off. I guess it has to do with the prisms used in the AF system and their (obvious) dependency on wave length. Probably one could compensate for the effect but that would require measuring the colour temperature which would probably expensive to implement.

I think I'm going to have to try again in better light...

Yes, definitely, unless you want your focus to be right in tungsten light only.

[Ben_Edict]

That doesn't explain the consistent misfocussing under tungsten light. IIRC, it is always BF. AFAIK, it determines the focus consistently, but just a bit off. I guess it has to do with the prisms used in the AF system and their (obvious) dependency on wave length. Probably one could compensate for the effect but that would require measuring the colour temperature which would probably expensive to implement.
.

There is not a lot on information freely available on the characteristics of AF modules. The only I found is a paper by Fuji (http://www.fujielectric.com/company/tech_archives/pdf/46-04/FER-46-04-135-2000.pdf), which explains an AF module which they developed for compact cameras. This is interesting, because it proves my suspicion, that AF modules may have a spectral response, which deviates from the ccd/cmos imaging sensor. In fact it is a strongly narrow response, with a peak sensitivity at arround 500nm (which is near the human eye's maximum sensitivity) and a strong fall-off towards the longer spectral end (at 700nm, the deep red, it has dropped to 60%). As this is a very basic type of AF module (because it needs to be cheap for compact cameras), one can easily deduce, that a higher quality AF module, used in DSLRs has a even stronger narrow-band characteristic, because it would be possible to include optical or electronic filters to achieve this.

The reason for narrow-band response it to exclude the influence of chromatic aberration of the lens from irritating the AF processing. As basically all lenses have different chromatic characteristics, concentrating the peak sensitivity around 500nm coincides with the standard wavelength for optimum lens sharpness (app. 550nm is the standard wavelength for which an achromatic lens (= most photographic lenses) is corrected).

This would mean, that the AF sensitivity drops significantly under tungsten lighting, which has a high percentage of longer wavelengthes. (You can easily see that from Wien's displacement law - Wikipedia, the free encyclopedia). And that in turn leads to unreliable AF function. And if you search in Canon forums, you will find the exact same AF problems...

Ben

[Marc Sabatella]

Not really, if it is a closed feedback loop. The body turns, checks, and if it is not enough it turns further or turns a bit back.

If the body just calculated the number of turns and performed them, then a lens would never hunt back and forth.

It *does* perform a second check, but perhaps not a third or fourth unless it detects a big problem. Look up "phase detect", which is the kind of AF system used in all Pentax DSLR's. There is *definitely* a precalculation that does most of the work. I think much of the hunting we see if a case of the lens not even being to get a good enough reading to do this initial calculation.

Perhaps -- pure speculation -- the test has a too high degree of tolerance. Lenses that get it right would then fare better.

Yeah, that makes sense.

That doesn't explain the consistent misfocussing under tungsten light...
I guess it has to do with the prisms used in the AF system and their (obvious) dependency on wave length.

I think you answered your own question there...

[Class A]

There is not a lot on information freely available on the characteristics of AF modules. The only I found is a paper by Fuji (http://www.fujielectric.com/company/tech_archives/pdf/46-04/FER-46-04-135-2000.pdf), which explains an AF module which they developed for compact cameras.

Thanks, I'll read that with interest.

I found this resource to be very instructive: Principle of the Split Image Focusing Aid and the Phase Comparison Autofocus Detector in Single Lens Reflex Cameras.

Quickly looking at it again, confirms my hunch that if an AF system using prisms and phase detection is tuned to a particular frequency, it will misfocus on a different frequency. Unless I got something wrong in my thinking, this should be the case.

In fact it is a strongly narrow response,

I still need to read the article you pointed to, but it cannot be too narrow since it needs to respond to a fairly broad range from daylight to tungsten.

The reason for narrow-band response it to exclude the influence of chromatic aberration of the lens from irritating the AF processing.

Makes sense. But this would mean that the AF system shouldn't work with a pretty high frequency blue light only. And/or with a pretty low frequency red light only. I haven't tested this, but I'd be a bit surprised if the AF system failed to react. My hypothesis rather is that it will react but misfocus the more the further away the frequency is to which it was tuned.

This would mean, that the AF sensitivity drops significantly under tungsten lighting, which has a high percentage of longer wavelengthes.

This may be true but it doesn't explain the consistent FF under tungsten lighting. If it were simply "unreliable" both BF and FF would occur randomly. This is not the case.

And if you search in Canon forums, you will find the exact same AF problems...

Yep. Not sure about how much their AF is off under tungsten lighting but I have definitely also read about this phenomenon appearing with Canons.

@Rice High: Did you measure AF consistency of your Canon under tungsten lighting yet?

[Ben_Edict]

Thanks, I'll read that with interest.

Quickly looking at it again, confirms my hunch that if an AF system using prisms and phase detection is tuned to a particular frequency, it will misfocus on a different frequency. Unless I got something wrong in my thinking, this should be the case.


I still need to read the article you pointed to, but it cannot be too narrow since it needs to respond to a fairly broad range from daylight to tungsten.

Prisms are very comparable to ordinary lenses. They have a dispersion and spread out the light according to wavelength over a certain length. That is the same as colour aberrations in a lens. Thus narrowing the frequency repsonse of an AF modul would minimize the focusing error introduced by any prisms. Also, the prsims are very small and (without calculating that), the dispersion is very small.

If you find time to study the Fuji article, you will find, that the AF sensor's sensitivity is narrow but sure not what I would call "narrow-band". So it would not completely blend out short or longer wavenlengthes. Instead the lower light level at these wavenlengthes would induce a longer focusing time. Also, it could be, if you have a stronger monochrome light (like the blue light you mention or low wattage household bulbs), that the AF sensor "uses" the bits of "green" light (where it exhibits most sensitivity) this light still contains (as this is usually temperature radiation and not line emission), whereas the majority of the image forming light is of another (blue or red) wavelength. This would naturally induce a misfocused image (with or without prism dispersion taken into account).

Ben

[Class A]

Thus narrowing the frequency repsonse of an AF modul would minimize the focusing error introduced by any prisms.

Only in the sense that it would stop working before it got (too) inaccurate.

If I understand the split-prism system correctly (whose principle is also used in AF systems) then any deviation from a "tuned" frequency, will result in an error. By narrowing the frequency you only make it stop earlier and allow for better phase matching as there isn't as much overlap of signals (assuming the AF sensors themselves cannot distinguish between colours).

Also, the prsims are very small and (without calculating that), the dispersion is very small.

Everything is small here, isn't it? I'm sure the difference in phase will be very, very small even for considerable misfocus.

Instead the lower light level at these wavenlengthes would induce a longer focusing time.

I haven't heard anyone complaining about slow focusing under tungsten light. Under low light, yes, but the common problem with tungsten light is BF, not speed or inconsistency.

this light still contains (as this is usually temperature radiation and not line emission), whereas the majority of the image forming light is of another (blue or red) wavelength.

Fine, but the usable portion of the spectrum would still be far enough away from the tuned frequency to cause an error. At least that's a possiblitiy.

This would naturally induce a misfocused image (with or without prism dispersion taken into account).

Why would it without taking prism dispersion into account? What else explains a consistent FF? I don't see how a loss of intensity (usable light for AF purposes) does that.

[Class A]

(assuming the AF sensors themselves cannot distinguish between colours).

Excuse me for quoting myself here, but this just sparked an idea. If the AF sensors were colour sensitive (say using the foveon sensor technique) then the AF system could see what colour temperature the signal has and the phase matching could be taken it into account.

Of course, such a solution would be more expensive, but if a) my cause analysis is correct and b) the solution works, then this could explain why other brands may have less problems with AF under tungsten lighting. While some Canon users are affected (to what extent?) I haven't heard complaints from Nikon users?

Don't Nikon users have a problem? Do their AF works differently? Or does Nikon use a slightly more expensive AF implementation which allows to compensate for the dependency on colour temperature?

[wildman]

Something I have always wondered about:

I hardly ever shoot AF. Most all of my shooting is through 500mm and more scopes using manual focus.

Considering that the light path when focusing through the viewfinder is different from the light path to the sensor when the shutter is released has there been any reports of mis-calibration, however slight, between the two light paths in the past?

I wonder with age and bouncing around out in the field if such a thing could happen. After all focus could LOOK spot on in the viewfinder but be slightly off with respect to actual sensor focus. The eye focusing through the viewfinder and the focus on the sensor are not necessarily the same thing.

I ask because I have not been on this forum very long so I'm not aware of this issue ever coming up. I do know Canon did have such an issue with one of their full frame cameras in the past.

[Class A]

Considering that the light path when focusing through the viewfinder is different from the light path to the sensor when the shutter is released has there been any reports of mis-calibration, however slight, between the two light paths in the past?

You are right, the optical path lengths to the sensor and to the focussing screen (for MF) and to the AF system (for AF) have to be calibrated with each other.

I've read about people reporting about a misalignment between focusing screen and sensor. To what extent this can happen to a perfect camera after some rough times, I don't know.

It probably won't apply to you with your long lenses but in general one also needs to be aware that the stock screen won't show differences in micro focus adjustments for lenses with a (focusing) aperture wider than ~f/2.8 very well (at all?), hence making it very difficult to manually focus such lenses.

I do know Canon did have such an issue with one of their full frame cameras in the past.

Perhaps not quite what your are looking for but after reading these user reports about Canon EOS 1D Mark III focusing troubles
I'm not sure I'll ever buy a Canon.

[wildman]

I think the Canon issue did not involve AF at all.

It was the result of the prism not being perfectly perpendicular to the axis of the light path. It turned out that someone on the assembly line took it upon him/herself to assemble the prism assembly slightly out of the order of assembly required by Canon.

Still, on such a high end body, one would think Canon's QC would have caught such a glaring defect.

[Class A]

I think the Canon issue did not involve AF at all.

Did you intend to write "MF" instead of "AF"? Surely it was an AF problem.

It turned out that someone on the assembly line took it upon him/herself to assemble the prism assembly slightly out of the order of assembly mandated by Canon.

Really? Can I read that somewhere? That sounds too simplistic for an fault that persisted embarrassingly long. Have you read about the 10/10 cameras that didn't work for a reviewer? Then another hand selected 2/2?