5D Autofocus at F/4?

One other thing.

The fact that the AF is acting on the information from sub-apertures is the main reason why some people need their lenses and bodies AF calibrated.

In terms of the camera body, that is fairly obvious, since the optical path to the AF sensor is not the same as the optical path to the image sensor, and any difference needs to be taken into account by the AF.

For the lens, most of the aperture may be focussing at a different distance from the small sub-apertures that the AF works with. Whis is what gives rise to optical aberations, which all lenses suffer from to a degree, especially wide open.

Even at the design aperture, each AF sensor only uses around 1/8th of the lens aperture, so AF can be slightly skewed even at that optimum condition. When stopped down or opened up, it can obviously be even further off.

If the lens was geometrically perfect, you wouldn't need lens AF calibration at all. Although most of us knew it for quite a while, Canon have only just admitted they aren't there yet. ;-)

Problem arrives when you use a lens on multiple cameras. I thing the focus adjust on the 1D3 will solve most, if not all focusing problems.

Its RKM said:

One other thing.

The fact that the AF is acting on the information from
sub-apertures is the main reason why some people need their lenses
and bodies AF calibrated.

In terms of the camera body, that is fairly obvious, since the
optical path to the AF sensor is not the same as the optical path
to the image sensor, and any difference needs to be taken into
account by the AF.

For the lens, most of the aperture may be focussing at a different
distance from the small sub-apertures that the AF works with. Whis
is what gives rise to optical aberations, which all lenses suffer
from to a degree, especially wide open.

Even at the design aperture, each AF sensor only uses around 1/8th
of the lens aperture, so AF can be slightly skewed even at that
optimum condition. When stopped down or opened up, it can
obviously be even further off.

If the lens was geometrically perfect, you wouldn't need lens AF
calibration at all. Although most of us knew it for quite a while,
Canon have only just admitted they aren't there yet. ;-)

Click to expand...

RDKirk said:

Canon's autofocus system is based on depth of focus, not depth of
field, and the depth of focus of a 500mm f4 lens at any focused
distance is exactly the same as the depth of focus of a 50mm f4
lens at any focused distance. The focusing algorithm need change
only for aperture--no other factor.

--
RDKirk
'TANSTAAFL: The only unbreakable rule in photography.'

Click to expand...

How canon choses to spec and describe af is one thing - and it's fine you quote them on this.

BUT, users NEED and WANT the camera to acheive af within the depth of field . . . it's what we all know and can all check.

Honestly, canon af cameras are trying to place the point of focus within the depth of field . . . anything less would be silly. Because laymen can't measure depth of focus, or test it, it's not all that helpfull to know about canon's depth of focus claims - better to know by own tests what canons camera and lens combo achieve in a real world test for depth of field.

The only discussion (or argument) point is what standard we set ourselves . . . I feel the following is a very fair position to take, and I'd have no concern arguing this in a court of law, by supporting it with 3 years professional photography and that which other pros (those demanding high standards) have confided in me.

-

It's fair to say for consumer standards, within 1 x depth of field is reasonable focus for most.

And that for professional standards, within 1/2 x depth of field is reasonable focus for most.

-

And to relate this to the thread - a 24-105 f4 lens, comapred to a 24-70 f2.8 lens on a 20d/1dmk2 etc . . . . the f4 will be less acurate and suffer greater variation (tolerance) between shots (af). The f2.8 will be visibly superior.

Whilst a 24-70 f2.8 might need more precision to get it within the dof, the job should actually be simpler, but the benefits are definitive.

Compare these two lenses . . . same subject/distance . . .

1) A 24-70 at 70mm focussed at f2.8 (and lets say the shot taken at f4), on a 20d to within 1/2 x depth of field (which my tests on a range of 20d bodies found to be standard), distance 2.8m (x40 focal length).

Depth of Field is 17cm at af f2.8, where according to a average of users deduced long ago and as an accepted formula, the 17cm represents a camera to subject distance which will be in an acceptable level of sharpness. So pros might demand better, some consumer might be happy with less - it's an accepted thing, but also a subjective thing, consider it a great benchmark to start from.

For the camera to be accurate to within 1/2 x depth of field at f2.8 (as per my tests), then focus would be roughly + or - 4.25cm (for simplicity i'm ignoring that the whole dof is distributed 9cm behind, 8cm in front).

Which when the picture is taken with the lens stopped down to f4, then within the f4 depth of field (24cm), then accuracy is now even better than 1/2 x dof . . . it's actually within + or - 4.25cm out of 24cm dof EXCELLENT RESULT.

2) a 24-105 f4 on 20d, at 70mm, 2.8m distance. DOF is as above, 24cm. But as my tests found, the 20d with an f4 70mm achieves only af accuracy of within 1 x depth of field. THus af is within + or - 12cm (again ignoring that dof is biased slightly behind).

So the choice for a user in shooting the two pictures at f4 but with a f2.8 or f4 lens is as follows - the f2.8 lens gave accuracy of + or - 4.25cm, the f4 lens gave accuracy of + or - 12cm . . . . the difference to the trained eye is startling.

Whilst both fall within the accepted time honoured definition of acceptable sharpness . . . one is clearly (judged in a print) going to much sharper much more often. As a pro that likes to shoot wide open in available light I need the higher accuracy and am willing to pay (cost and weight). As a consumer hobbyist on holiday, i choose lighter cheaper options, sacrificing quality and consitency.

Regards,
Kev

http://www.dofmaster.com/dofjs.html
http://www.bobatkins.com/photography/technical/dofcalc.html

RDKirk said:

Yes, there will in a 24x30 print viewed at 18 inches. But Canon
doesn't care about that.

There won't be any difference in a 5x7 print viewed at 10 inches
(which is the system design specification standard Canon quotes in
their "Lens Work III").

Click to expand...

Did you notice that the 5x7 print size, was only a recently introduced thing (last couple of years) and that is was a lower quality standard to that which they had previously used (a bigger print size).

E.g. canon had a higher standard, users started complaining about af, canon reduced standard to a lower more easily acheived target.

Regards,
Kev

David Langworthy said:

This makes sense. Just to make sure I understand. For a good
image shot at f/8 and with a large print (24x30) viewed up close
(say 18") there will be no noticable difference auto focus between
the 24-70 f/2.8 and the 24-105 f/4, is this correct?

Click to expand...

Not necessarily, depends on nature/elements in shot. You'd spot it on a proper focus chart, and for certain commercial product photography I do it would be an issue. But for my holiday consumer stuff, f8 would cover the reduced af accuracy.

--At f4, 70mm dof 24cm (48% front, 52% back)

My post above show big benefit with f2.8 af, for those that need it on a 20d/5d/1dmk2.

--At 70mm f8 depth of field 48cm (46% front, 54% back)

--At 70mm f16 dof 98cm (now 42% front, 58% back)

24-70 f2.8, 70mm 2.8m, depth of field 17cm
AF on 20d (5d, 30d similar with f2.8) + or - 4.25cm

24-105 f4, 70mm 2.8m, dof 24cm
AF on 20d f4 + or - 12cm

I would agree, that for most, by f16 it becomes irrelevent - BUT that does again depend on subject/purpose . . . for some shots you may want three objects spread at different distances, and then the issue might concern one again.

Regards,
Kev

Dave Largent said:

Dave Largent said:

There is NO advantage in AF terms to using a lens that is faster
than the AF design f/# - no advantage in focus speed, no advantage
in low light sensitivity and no advantage in focus accuracy.

Click to expand...

I don't understand why there would be no AF advantage
to having a 1.4 vs. a 2.8. Wouldn't the extra light coming
through the 1.4 "help" the sensor? Or are you saying
the 1.4 focuses using 2.8?

Click to expand...

Most of my test were 10d's, 20d's and 1dmk2

The 1dmk2 would focus and f5.6 within 1 x depth of field, a f4 within 1 x dof, a f2.8 within 1/2 x dof, a f1.4 within 1/2 x dof . . . to explain why that makes it more accurate, I'll expand by saying af is setting a lens to a target distance.

At 2.5m distance with 50mm lens (x50), 1dmk2, dof is following for each aperture;
f1.4 - 16cm = + or - 4cm
f2.8 - 32cm = + or - 8cm
f4 - 45cm = + or - 11cm
f5.6 - 65cm = + or - 16cm

At 10m distance with 200mm lens (x50 focal length), 1dmk2;
f1.4 - 16cm
f2.8 - 32cm
f4 - 45cm
f5.6 - 64cm

Bigger native lens aperture = more accurate, if measuring the point of focus with regard to distance (cm).

If one is referring to accuracy within dof (within 1 x dof, or within 1/2 x dof) then some might argue f2.8 is same as f1.4, and f4 is same as f5.6 - but I consider this a misleading way to interprit what is really happening.

You'll notice if trying tests yourself, that a 50mm at 2.5m, and a 200mm at 10m will see the focus target cover the same area in the viewfinder, so the af sensor has virtually the same information to process, and the chart remains the same size in each shot - only perspective changes (chart looks flatter with 200mm).

I don't have a 50mm f5.6, nor a 200mm f1.4, so had to demonstrate numbers with one value fixed . . . but my tests across many examples of the following; 70-200 f4 L, 70-200 f2.8 L IS, 50mm 1.4, 50mm 1.8, 17-40 f4 L, 16-35 f2.8 L, 24-70 f2.8 L, 100 f2.8, 100-400 L -5.6, 20-35 f3.5-f4.5?, and more - all showed the same characteristics regarding af point and depth of field.

http://www.dofmaster.com/dofjs.html

10d's are within 1 x dof for any lens, assuming very well calibrated, and most weren't.

20d are better with f2.8 (1/2 x dof), have same level with f4/f5.6/f1.8 (1 x dof) and are worse with f1/4 (outside 1 x dof - a bug they fixed in the 5d).

5d (i haven't tested many) are better with f1.4/f1.8/f2.8 (1/2 x dof), and same level with f4/f5.6 (1 x dof). I understand 30d is the same, but haven't tried one in a proper test yet.

My chart is at http://www.canon-dslr.com/Canon_Jan05/Canon_SLR_Focus_Test.htm

If it helps, few consumer user spot a focus error up to 1 x dof - and even canon technicians of 20 years experience couldn't spot an error of 3 x dof . . . . which curiously is about the same adjustment one can make with new 1dmk3 af micro adjustments.

Regards,
kev

Well, since that contradicts what I know about how the AF system works, I did try it - maybe Canon sneaked something else into their AF system since I last examined it in detail!

Before I give you my results and conclusions, though, here is exactly what I did:

I used 3 Canon lenses on my 5D:

100mm f/2.8 macro (ie. Uses the f/2.8 AF sensor at its working aperture)
50mm f/1.4 (ie. allegedly lets more light onto the AF sensor)
24-105mm f/4 L Zoom (set at 100mm) (ie. uses the f/5.6 AF sensor)

The 100mm and 24-105mm lenses are the controls, or references, the 50mm lens is the test of your assertion that optics faster than the AF aperture give better AF focus results.

I downloaded Tim Jacksons focus test chart at http://www.sebian.pl/foto/download/focus_test.pdf and set that up at 1 metre range from the camera at the required 45deg viewing angle (so that the focus error distance can be measured directly in mm).

I then fitted each lens in turn and shot 10 frames from each, resetting the focus manually to infinity before each shot. This ensured that the AF system was recalculating the focus each and every frame.

Having my "raw data" of 30 frames, I then loaded each into Photoshop and viewed them at 300% scaling, noting the position of optimum focus and its error from the nominal position.

I entered the data into an Excel spreadsheet and calculated the standard deviation (ie. the spread) in the focus error for each lens. Using the standard lens formula (1/u=1/f - 1/v) I then calculated back to see exactly how much error in depth of focus each lens produced.

The circle of confusion due to AF errors at the image can then be calculated simply by dividing this depth of focus error by the f/# used for the image and also the CoC of the AF by dividing the focus error by the f/# of the AF that should have been used (2.8 for the 100mm & 50mm lenses and 5.6 for the 24-105mm zoom).

Here are the results:
(Lens, variation in AF distance, AF depth of focus, Image CoC, AF CoC)
100 f/2.8: 0.567mm; 7.25um; 2.59um; 2.59um
50 f/1.4: 2.87mm: 7.9um; 5.67um: 2.84um
24-105 f/4 3.12mm; 38.6um; 9.65um; 6.90um

One fairly obvious observation is that the 24-105mm f/4 zoom lens is clearly using a less accurate AF system than the other two lenses, with a distribution of focus which is almost 6x wider than the 100mm f/2.8. So there is no doubt at all that this lens is activating the lower precision AF sensor.

However, it is also clear that the f/1.4 lens offers absolutely no advantage in terms of focus accuracy over the f/2.8 lens. The depth of focus variation for the f/1.4 lens is very similar, actually a little worse, than the slower f/2.8 lens, with similar variations for the CoC when calculated at the AF working aperture.

It is likely that slight degradation is within the random variation possible with only 10 samples of each image, however it certainly rules out any "improvement" from the faster lens.

Finally, the calculated CoC at the AF apertures appear to be much lower than the specifications that RDKirk provided earlier in the thread:
f/2.8 AF - Canon say 0.01mm; 0.0028 worst case measured
f/5.6 AF - Canon say 0.035mm; 0.0069mm measured.

However it should be noted that I have measured and computed "standard deviation", ie. the average spread, whilst the Canon figures that RDKirk quotes is closer to a "worst case" or 99% probability figure, which will be at least 3 standard deviations. So although at face value there appears to be an inconsistency here, the numbers are actually consistent with the Canon specfication which RDKirk quoted earlier.

So, having taken up your suggestion of "try it and see", I conclude that the faster lens makes absolutely no difference to the AF precision. That is what the theory of the operation of the AF system predicts and it is what the measurements have demonstrated.

The only time that a faster lens improves the AF is when it enables the high precision AF sensor to be used. Beyond that, you can increase the lens speed as much as you like, it makes no difference to the AF at all.

Moment I read this I knew your test was completely different to the one I use (which has same features/methodolgy as canon inc's official tests) - and as one would expect (which I've said before), small one piece charts tested too close give a far better impression (accuracy and variation between shots) then testing at proper distances (like real world examples) or with a pukka big two piece chart (minimum A3 plus size).

The moment anyone mentions mm's and af accuracy i know they are testing far too close, or are doing something else wrong.

Its RKM said:

I downloaded Tim Jacksons focus test chart at
http://www.sebian.pl/foto/download/focus_test.pdf and set that up
at 1 metre range from the camera at the required 45deg viewing
angle (so that the focus error distance can be measured directly in
mm).

Click to expand...

WRONG CHART
This is a one piece chart, too small, used too close.

Granted Tim makes suggestions about the one piece element - but try for yourself testing horizantal lines, verstical lines, crosses, news print etc . . . they'll all give different results, as does for example putting a filter over the lens (even a plain old UV super hmc). You'll see the focus target pattern I settled on, even at x25 distance the one piece chart introduces af error, although at nearest distances it's less relevent.

If you don't beleive me that a two piece chart with seperate focus target parallel to sensor plane and angled measurement chart is required, then just read canon inc service manual, it's all in their. If trying to prove their specs and your theory suggest at least entertaining their test.

They have specific part numbers to order the correct official af test rig (lights, chart, software, tool lenses etc) - or you can print mine which is good for x40, make it bigger (a2 is enough) to test f5.6 at x50. Tim's chart completely ignores all of canon inc's methodology and reasoning behind af testing (which no doubt is VERY FAIR for a nikon user, although nikon are more open in the uk about af testing with show reps carrying details/examples in folders with them to discuss issue) - but tim's method is easier and does make the result appear better than in real life - not everyone can test a 200mm lens at 10 meters, nor does everyone have an A3 plus printer. So i can see why the small a4 one piece charts are so popular . . . but having been victim to canon uk messing up a complete system with an unofficial chart and method I am not one to have pleasant feelings towards them :-(.

An A4 chart is of no use at a x50 focal length, hence why tests with them are limited to around 1m and typically done at even shorter distances.

Accuracy due to higher consistency is far better at short distance (close to nearest point of focus) and gives a misleading indication as to how the lens/body will perfrom in more everyday situations such as portrait photography.

My A3 plus two piece chart reaches it's limits with an f4 lens at x50, hence why I test at x40 - not ideal I know, but I have done enough x50 to x40 comparisons to know differences for my type of work (portraits, wedding, hotel interiors), and most of my lenses are f1.4 to f2.8.

If I was mainly photographing on track motor sports, i'd probably test at x50 (or more) as a more typical situation rather than x40 - buy my portrait work is mostly 2.5 to 7.5m distance, which suits the x40 an A3 chart is capable of.

DISTANCE

a 50mm lens should be tested at half the distance of a 100mm lens - NOT at the same distance. If you do test different focal length at same physical distance it's not comparing af performance in the same way as canon or me.

Canon's advice is x50 focal length, so 50mm at 2.5, 100mm at 5m . . . you'll notice in this scenario the chart fills the shot (an a2 chart, as an a4 will be miniscule) in the same scale on each, with the af sensor getting the same target, and dof for a given aperture being the same.

Part I of II

Its RKM said:

I then fitted each lens in turn and shot 10 frames from each,
resetting the focus manually to infinity before each shot. This
ensured that the AF system was recalculating the focus each and
every frame.

Click to expand...

VARYING SHOTS - focus/lens start point

By testing from the same lens start position you will get a much improved tolerance, and a misleading af result for judging any af correction adjustment - which I know is what canon were doing even with 3 shots, they may be changing this in future, but haven't yet to my knowledge.

When i showed them how a body/lens tolerance is so much bigger when warying the lens start position (rather than all tests from infinity start point) their first response was that the body was faulty and offered to replace ! they now realise they all do this, just pro focus means less variance consumer means more, different apertures etc

As I said, I test 2 shots from infinity, and 2 shots from point of nearest focus on the lens focus ring. I am considering adding 2 more shots in future, both with the lens already in focus . . . but my archive of test shots doesn't have this to compare to.

I also give guide about lighting - as I mainly use f1.4 to f2.8 lenses I am rarely the type to shoot in bright daylight (i guess most who buy an f1.4 lens aren't typically bright light shooters, yet most af guru's tell them to test in bright light) . . . but then bright daylight is where the af works best. I myself compromised at 8-10ev as a sensible test lighting level - I have tested from 6ev to 12ev . . but suggest range of 8-10.

I think most of this is at http://www.canon-dslr.com/Canon_Jan05/Canon_SLR_Focus_Test.htm - so if you wish to test my claims you need to follow this . . . but you have already claimed to have tested my work (and disagreed with me) when you actually tested something else entirely, so please forgive if I don't take what might be your expected next non-agreeing post (done with my chart maybe) as gospel without seeing how you did them in case of an oversight in your method.

I'm a working pro and somewhat of a techie so am confident of my findings based on real life examples and making a test rig with the AIM of replicating my typical real life shooting situations (distance, lighting, varying lens start position) and canons proper test equipment - i spent a lot of time testing actual af performance, rather than justifying af theory. You've tried to use depth of field to reverse engineer depth of focus measurements, with the wrong chart, and at the wrong distances, yet feel you've proved your theory.

I've drawn the line at seeing what happens to the depth of field, as in all practicalaty that's all that matter in a real situation - as it's a real tangiable quantifiable thing everyone can see and relate to with just a little explanation.

BUT on an area we both might agree - 50mm f1.4

I have tested numerous 20d, 10d and 1dmk2 . . . i myself bought 5 1dmk2's to find to good ones . . . I found the 20d (every body i tested, older and newer) had a bug when used with f1.4 lenses, this was confirmed by the test and by some well established award winning uk pros (those using 20d with f2.8L's loved and recommended camera, those using with f1.8/f1.8 didn't think so highly).

My 5d experience is far less, I found the worse af with an f1.4 to be fixed . . . your results hint that it might still be their . . . . next 5d I get my hands on I will test across a range of distances to find out more.

Regards,
Kev

btw 100mm f2.8 USM Macro, is most accurate lens I've tested - when measured in terms of positioning af within the depth of field.

The 100mm f2.8 USM Macro was much more consistent between shots, than a f1.4 50mm . . . . although as the 50mm at 1.4 is a narrower depth of field it more accurate in absolute terms, even though it wanders about within the dof more.

To see all this one really needs a big chart, preferrably A2 . . . it's a different world than is testing observations with an A4 chart.

Regards,
Kev

Many thanks for taking the time to do this test, analyze the data and post the results!

From all I have read I was fairly certain that even though an f/4 lens can activate one or more cross-sensors (depending on camera body), using an f/4 lens did not increase focus accuracy similar to the increased focus accuracy that comes from using a f/2.8 lens. Your test confirmed this to be true. (There are various mis-informed posts here and there claiming otherwise.)

I’m planning to go from near the low end of the camera food chain (S60 - but hey, I shoot raw) to near the high end (Mark III?). Your writeup has convinced me that the first lens I will buy will be the EF 28-70 f/2.8L USM instead of the EF 24-105 f/4L IS.

Thanks again.
--
Jelf

Jelf said:

Many thanks for taking the time to do this test, analyze the data
and post the results!

From all I have read I was fairly certain that even though an f/4
lens can activate one or more cross-sensors (depending on camera
body), using an f/4 lens did not increase focus accuracy similar
to the increased focus accuracy that comes from using a f/2.8 lens.
Your test confirmed this to be true. (There are various
mis-informed posts here and there claiming otherwise.)

I’m planning to go from near the low end of the camera food chain
(S60 - but hey, I shoot raw) to near the high end (Mark III?).
Your writeup has convinced me that the first lens I will buy will
be the EF 28-70 f/2.8L USM instead of the EF 24-105 f/4L IS.

Thanks again.
--
Jelf

Click to expand...

Not sure if Phil was pulling our legs, but he says he got two 1dMk2n for £1700 inc each at focus NEC. Stunning price, and a great camera, some say mk3 won't be buyable till June, and likely at a premium for a month or two. My 1dmk2 i think were announce feb, i went straight onto 3 or 4 lists, and saw the first one june or july if i remember right, the last one september.

Not knowing what you photograph or reasons (commercial, hobby, holiday etc) . . . .

If taking the weight and quality as key factors, choices might be like . . .

A 1dmk2 or mk3 with a 24-70 f2.8 makes a lot of sense - but is heavy, no fun for holiday snaps, but great for commercial work suiting that focal range.

A 30d or new 400 with a 24-105 f4 L and a 50mm f1.4 would be ok for commercial work, but perhaps the better travel option.

I myself use a pocket camera on holiday day, for things like walking round sydney or miami during the day . . . although for getting up 5am to catch sun rise of sydney opera house, or at the grand canyon then I take the 1dmk2 and a 17-40 f4 L with tripod.

To put it another way purpose, weight, quality, price are all trade offs depending on purpose . . . . the pictures at this link http://www.kevinboulden.com/U2_2005/u2_London_2005.htm were taken on a canon ixus 500, and later ones a panasonic fx9 IS - my 1dmk2 wouldn't have got past the gate, and if they had would have been jumped on the moment I lifted one to shoot ;-) At the same site, the press pass to goodwood shots were all 1dmk2's.

Most of the other work on my sites was from 10d's (all the portraits, and most of the commercial and wedding shots), they are all in desperate need of an update as the sites are but i've been distracted with nhbc and builders specs for the past 9 months . . .

Regards,
Kev

There are some circumstances in which a faster lens will make a clear difference in focusing accuracy.

Remember that focusing calculations are SLOWED when the level of light decreases toward the minimum necessary for AF system operation (-1EV for the 5D, for instance). To the extent that a faster lens transmits a greater EV at lower absolute light levels than a slower lens, it forstalls THAT particular focusing factor a bit longer.

However, I suspect there is a quickly achieved upper limit above which more light cannot speed the focusing calculation any further.

--
RDKirk
'TANSTAAFL: The only unbreakable rule in photography.'

Don't forget the proper specification of terms. When Canon says "Within 1/3 of DoF," they're talking about the Depth of Focus at the sensor plane, NOT depth of field in the image. To the extent that Depth of Focus can be correlated with Depth of Field, it ONLY applies to a 5x7 image viewed at 10 inches--the standard specified by Canon.

Remember that the depth of field varies inversely with the degree of enlargement of the final image display. A camera/lens combination that successfully places the actual focused plane within the depth of field on a 5x7 image viewed at 10 inches might not do so when the image is viewed at 40x60 at ten inches (the typical situation when we peer at a 5D image at 100% of the native pixel resolution on a monitor).

Most significantly, also remember that the tolerances allow the system to place the actual plane of focus ANYWHERE within the specified tolerances, and not necessarily in the same place every time. A perfectly operating system may place the actual plane of focus within the depth of field every time on a 5x7 image viewed at 10 inches--and yet be all over the place in a 40x60 image viewed at 10 inches.

--
RDKirk
'TANSTAAFL: The only unbreakable rule in photography.'

Whilst I understand your concern about the accuracy of the AF being most significant near infinity, that is only the case in OBJECT space. In image space, which is after all where the image is formed and where the AF occurs, it makes absolutely no difference. This is why Canon specify their AF system in terms of CoC, not DoF, which varies wildly with focus distance and focal length. It is also why I converted the actual focus error distribution from object space into image space, as depth of focus.

I maintain that your critique of my test is irrelevant once the data is converted to image space (it is all down to physical lens displacement relative to the focal plane anyway, so focus distance is irrelevant when the comparison is made in image space).

Nevertheless, despite your concerns about these tests being at short range, the objective of the test was to determine if the faster lens actually improved the accuracy of the AF system. In that respect, it does not matter one iota what the absolute accuracy was, the fact remains that the test demonstrates that there is absolutely no AF accuracy advantage to a lens faster than the design aperture of the AF.

Incidentally, I just repeated the test with 50 shots of each lens using your test pattern at a range of 7.5m (it is dark here and that is almost as long an unbroken line of sight as I can get inside my home). The results were virtually the same - no more than 5% variation from the originals. So I don't think your questions about the test stand up to being tested either. ;-)

So, the bottom line is that a faster lens does NOT make the AF more accurate. In addition, I conclude that the Canon specification for the AF CoC is a fairly high probability that the AF will be within that CoC. 50 shots showed a fairly bell shaped distribution of the focus error. In both the original and this second test, I used the standard deviation to detemine the spread of focus error, which includes around 68.3% of a random distribution, 3xSD includes 99.73% of the distribution. I don't have the time or the inclination (nor am I prepared to dedicate the wear on my 5D shutter!) to making enough test exposures to validate that level of distribution. Nevertheless, I plan a 3rd with an easier test pattern to make the exact position of optimum focus point easier to discern visually, based on aliased patterns off to the side of the reference focus mark. It shouldn't change the results, but it will make discerning actual focus position easier.

Its RKM said:

blackhawk13 said:

--The AF will always use the Maximum native aperture to focus
regardless of the AV setting.

Click to expand...

NO IT WON'T!
That isn't how Phase Difference AF works!

The PD AF hardware includes micro-optics which re-focus pencil ray
bundles from opposite sides of the lens aperture that the AF is
designed to work with onto linear sensors. The difference in
position of the images on these two sensors, the phase difference,
determines the amount and direction that the lens is out of focus
and the software then drives the lens motor to correct the error
until it is within specification.

The critical issue here is that the micro-optics are fixed units.
They must be designed to redirect light from the periphery of a
cone of light corresponding to a specific f/#. Once designed and
manufactured, the AF optics do not change. Putting a faster lens
on the camera DOES NOT allow any more light to reach the AF sensor,
because the AF optics only capture light from one particular area
of the lens aperture.

Think of the AF sensor as two small circular apertures, on the
opposite inner edges of the much larger optical aperture of the
lens. The AF apertures are a fixed distance apart - the aperture
that just includes them at that separation is their design f/#.
Making the lens aperture larger lets more light onto the image
sensor, but it doesn't get any more light into the AF sensor,
because that only sees through those small apertures. Neither does
a larger lens aperture make the AF any more accurate or faster,
since that is fixed by the distance between the small apertures.

Either the lens is big enough to allow light through the AF
apertures, and the AF works, or it isn't, and the AF fails.
Obviously there is a transition area where the AF sensor collects
light that is partially from the lens barrel and partially from the
lens aperture, so AF can operate irregularly at slightly higher
than the f/# they are designed for, but not by much. Search for
threads on tricks to make teleconvertors work and you will see many
posts there about how erratic the AF becomes - that is why. At the
design aperture of the AF system, all of the light reaching the AF
sensor comes through the lens aperture.

Putting a larger aperture lens on the camera only improves the AF
speed or accuracy IF the lens enables a more precise AF sensor to
be switched in, eg. going from f/5.6 to f/2.8. Since Canon does
not have AF sensors which are optimised for faster than f/2.8 in
any of their cameras, the AF will not improve in either speed or
accuracy for lenses which are faster than f/2.8.

This is exactly the same effect as with the old split image focus
screens that film cameras used. It was weill known that using a
lens slower than the f/# that the split image was designed for
caused the split image to darken and become unuseable. However it
was less well known, but just as true, that putting a lens that was
faster than the split image design f/# did not make the split image
any brighter - it made the ground glass screen around it brighter,
but not the split image or the microprism collar. The reason is
simple - the optics creating the split image and the microprism
collar were designed for a particular f/# of lens - they didn't
"see" anything that the lens produced outside of that aperture.
Phase difference AF sensors use similar, in fact almost identical,
optics.

There is NO advantage in AF terms to using a lens that is faster
than the AF design f/# - no advantage in focus speed, no advantage
in low light sensitivity and no advantage in focus accuracy. That
doesn't mean there are no advantages to faster lenses - there are
plenty, including lower light photography, shallower DOF and so on,
but no AF advantages.

If you want to learn how phase difference AF (a derivation of the
original split prism viewfinder that used to be common on film
SLRs) works, read this document which explains the principles in
very simple terms from early basics:
http://doug.kerr.home.att.net/pumpkin/Split_Prism.pdf

The original PD AF system was invented by N. Stauffer and D.
Wilwerding at Honeywell in the 1970s and several fixed and scaning
optical implementations of that were developed. It is a derivative
of the fixed optical solution that we have in dSLRs today.

Click to expand...

--The AF on the 20, 30, and 5D uses the lenses largest native aperture to focus. if that is f/1.2 the camera will open to that, focus, and stop down the the set aperture. Obviously a fast lenser cast the same size image as a slow lense on the sensor or the AF sensor. RIGHT?!

Therefore a f1.2 lense provides more light for the AF sensor to use than a f/2.8 lense, so a faster lock is obtained. The faster the lense, the faster the AF will lock up.

This info is buried somewhere in the Canon white papers.

-nothing beats a fast lense, except a fast girl-

AK_2 said:

RDKirk said:

Yes, there will in a 24x30 print viewed at 18 inches. But Canon
doesn't care about that.

There won't be any difference in a 5x7 print viewed at 10 inches
(which is the system design specification standard Canon quotes in
their "Lens Work III").

Click to expand...

Did you notice that the 5x7 print size, was only a recently
introduced thing (last couple of years) and that is was a lower
quality standard to that which they had previously used (a bigger
print size).

E.g. canon had a higher standard, users started complaining about
af, canon reduced standard to a lower more easily acheived target.

Click to expand...

Actually, the "5x7 at 10 inches" is roughly the same standard as the "8x10 at 18 inches" that has been the universal standard for at least 100 years.

However, there was a loosening of the specification when Canon released their EOS lenses. When I compare the depth of field scale on my old 50mm f1.8 Canon FD lens (design circa 1973) with the scale on my old 50mm f1.8 Mk I Canon EF lens (design circa 1985), the older lens clearly has a tighter depth of field tolerance than the newer lens.

--
RDKirk
'TANSTAAFL: The only unbreakable rule in photography.'

...is that Depth of Field exists only the final desplay medium (print or monitor) and is constantly variable according to the size of the enlargement and the viewing distance.

So they start with an arbitrary final display size and viewing distance (5x7 at 10 inches), the average resolution capability of the human eye (300 dpi), and work backward from there to determine the permissible circle of confusion.

That tells them the necessary resolution of lenses, resolution of sensor, and accuracy of focusing system.

--
RDKirk
'TANSTAAFL: The only unbreakable rule in photography.'

RDKirk said:

There are some circumstances in which a faster lens will make a
clear difference in focusing accuracy.

Remember that focusing calculations are SLOWED when the level of
light decreases toward the minimum necessary for AF system
operation (-1EV for the 5D, for instance). To the extent that a
faster lens transmits a greater EV at lower absolute light levels
than a slower lens, it forstalls THAT particular focusing factor a
bit longer.

Click to expand...

Sorry RD, but it doesn't. Just as with my analogy with your eyes looking through a porthole, the speed of the faster lens does NOT let any more light into the AF sensor. It lets more light onto the image sensor, certainly. However the AF sensors are working at relatively high f/#s (f/8-16) for the sub-apertures within the aperture of the prime lens. No matter how big you make that primary aperture, no more light reaches the AF sensor.

Do you realise that this is almost the same effect as for the viewfinder?

The viewfinder comprises a lightly ground glass screen which scatters a little of the light incident on it. "In the old days", focus screens were coarser and scattered much more than modern screens. The reason for this is that a coarser screen gives a better representation of the actual focus position and depth of field. However only a small amount of the light scattered by the screen actually reaches the eyepiece, and is subsequently available to view. So with a coarse screen you get a good representation of depth of field, critical to manual focus, but a dim viewfinder. With a finer screen more light passes directly through to the eyepiece and less is scattered - a bighter viewfinder but less visibility of the critical focus point, which is needed less with AF cameras anyway. So brighter SLR viewfinders have pretty much appeared as a consequence of AF cameras - but they do have downsides: they are next to useless for proper DoF assessment.

Now, if you look at the eyepiece in a typical SLR, Canon are no different in this respect to other manufacturers, it is fairly comfortably optimised for your eye with some lateral and longitudinal eye relief - you can move your eye in the eyepiece reasonably comfortably without the viewfinder darkening. That means that the effective aperture of the eyeiece is determined primarily by the aperture of your eye which, depending on illumination level is around f/10 - part way between the aperture of the two AF sensors.

So stick your 50mm f/1.4 lens on your 5d with its default focus screen and press the DoF preview button. The aperture ranges from f/1.4 to f/22 - that is 7 stops. Does the viewfinder image look like it has darkened by 7 stops going from f/1.4 to f/22? I don't think so! More signifcantly, you can go from f/1.4 to f/4 withough seeing much change in viewfinder brightness at all.

The reason is pretty simple - the fine bright ground glass screen doesn't scatter much at all, so most of what you are seeing as viewfinder image is aerial - direct unscattered light transmitted by the prime lens. Your prime lens might be f/1.4, but what actually limits how bright that image is on your retina is your eye pupil, and at roughly f/10 it doesn't see anything coming in from wider than that.

This is exactly the same (without any scattering of course) as using a fast lens on the AF sensors - no more light reaches the AF sensor - it can't, because the AF is working through much smaller portholes within that prime optic aperture.

Of course, with the viewfinder you will see a small amount of image darkening, at a rate which gradually increases as the f/# increases, but it doesn't really kick in until around f/5.6-f/8. The limited viewfinder darkening that you see between f/1.4 and f/4 is almost all lost scattering of the reduced light from the primary lens. Since most of the light reaching the eye is unscattered, you see limited darkening until the lens f/# approaches that of your eye.

If you happen to have an old film SLR, eg. Nikon F, Pentax Spotmatic, Olympus OM-1/2/3/4 or Canon A(E)-1 for example, with an original coarsely ground screen, pop an f/1.4 lens on that and see just what a difference in brightness you see between f/1.4 and f/22. If you happen to have a modern bright focus screen that fits those cameras, pop that in and do the same - it will be somewhere between the original, great for manual focus, and the modern, great for AF, viewfinder - both in brightness and visibility of stopping down.

Just like the AF sensor, it is all down to apertures within apertures. ;-)

I know some folks have real difficulty getting to grips with this - the concept that a faster lens transmits more light is something that is ingrained into most photographers before the reach infant school. That is why I keep referring to Doug Kerr's excellent description of how the phase difference AF system works, and also to Canon's rather idealised diagram which first appeared on one of their AF patents years ago but which is still being reproduced on the recent white paper describing the 1DMk3. Study both of those documents and these issues should be a lot clearer.

blackhawk13 said:

--The AF on the 20, 30, and 5D uses the lenses largest native
aperture to focus. if that is f/1.2 the camera will open to that,
focus, and stop down the the set aperture. Obviously a fast lenser
cast the same size image as a slow lense on the sensor or the AF
sensor. RIGHT?!

Click to expand...

WRONG !

Focussing is implemented with the lens wide open. That is the same in all cases, but completely irrelevant to either the speed, the accuracy or the low light capability of the AF.

The AF sensor only uses PART of the lens aperture. This is inherent in the Phase Difference AF system which compares the light arriving at the AF sensor from different sides of the lens. The AF sensor specifically looks through two small sub-apertures in the primary lens aperture. These are separate by angles corresponding to f/5.6 and f/2.8, and each of the small apertures is in the region of f/8 -f/16 in itself. Once the lens is fast enough to enable a particular AF sensor (either f/5.6 or f/2.8) then it doesn't matter how much faster the lens is, no more light can reach the AF sensor.

In optical terms, no optical system can be faster than its limiting aperture. The limiting aperture in this case is the AF sensor - not the primary lens!

blackhawk13 said:

This info is buried somewhere in the Canon white papers.

Click to expand...

Indeed it is - take a look at the diagram on Page 17 of the recent white paper on the 1DMk3. You will find the same diagram on many Canon publications, right back to some of their early AF patents.

When you have looked at that diagram, come back and tell us how you think the faster lens will get more light into those AF apertures that Canon have drawn nicely projected onto the prime aperture to make it obvious!