Other

Sensor Stack Thickness Part III: The Summary

Published July 6, 2014

Well, I have to admit this has been a fun series. I’ve learned a whole lot. That’s what makes this so fun — I get some results I don’t understand, get some help figuring out what is going on, and before I know it, I’ve learned something that explains other things I haven’t been able to understand.

In the second part of this series, we started a database of sensor stack thickness and exit pupil distances, hoping that it would help people decide which lenses would adapt best to which cameras. (And, of course, determine which lenses would not adapt well to which cameras.) A number of people have added information to the database since it was first posted — enough to make it pretty useful.

Since the database is now large enough to be useful, I thought it would be a good idea to make a summary of what we know about lenses and sensor stacks. The best thing about all this, for me at least, is that it lets us make some generalizations about which lenses would be expected to have problems on which cameras.

General Rules

To summarize, there are three main factors that determine if a given lens will have problems when used on a different camera than it was designed for:

  1. Aperture (wider aperture has more problems)
  2. Exit pupil distance (shorter exit pupil distance has more problems)
  3. Difference between sensor stack thickness the lens is designed for, and the sensor stack thickness of the camera it’s being used on.

As always, Brian Caldwell has been kind enough to furnish graphs of the theory, which really helps in visualizing things.

Aperture and Stack Difference On-Axis

Brian’s first graph demonstrates two important points.

Graph courtesy Brian Caldwell

 

The graph demonstrates the on-axis (center) MTF of a theoretically perfect lens designed for no sensor stack filter (orange line), and the effects of adding a sensor stack equivalent to 1,2, and 4mm of glass. The first takeaway message is that for apertures of about f/2.8 and smaller (higher f number) there is really no significant effect on-axis.  (Emphasis on the ‘on-axis’ part. This doesn’t mean the corners will be great, although they may be.)

The second takeaway message is that the effect is proportional to the difference between the sensor stack thickness the lens was designed for and actual thickness of the camera it’s being used on. For example, adding 1mm of extra glass in the path doesn’t really affect things until f/1.4 or so, while a 4mm difference is quite apparent at f/1.8. Once the effect begins, though, MTF decreases exponentially with increasing aperture.

The summary, for on-axis effects, is that shooting a lens at f/1.4 on a camera with a sensor stack different by 2mm or more from what the lens was designed for will probably reduce MTF even in the center. Stopped down, though, you’re unlikely to see any difference on-axis.

Exit Pupil Distance from Sensor

Brian’s second graph shows the effect of a 2mm sensor stack on an f/2.0 lens designed for no sensor stack. If you go back to the graph above, you’ll see that at f/2.0 a 2mm sensor stack difference has really no effect on-axis. This graph shows the off-axis effects out to the edge of the sensor.

Graph courtesy Brian Caldwell

 

As you can see, the off-axis effect shows a mild decrease in MTF and increased astigmatism even with a 100mm exit pupil distance. Would you see this in real-world photography? You might, if you compared an image taken with the adapted system to an image taken with the lens native camera system. But overall, you’d probably be pretty happy with the lens adapted to a different camera.

With a 25mm exit pupil distance, the effects are very severe, even just a short distance from the center. An exit pupil distance of 50mm definitely has some effect, too, although as expected it’s not as bad as 25mm.

Some Generalizations

Sensor Stack Thickness

Our database for sensor-stack thickness remains incomplete (although I’m hoping for more contributions soon). Also remember that physical measurement of thickness and optical thickness may be different. If the glass used has a high index of refraction, it would have an optical effect greater than what is measured physically. For example, we might measure a physical thickness of 2mm on two different cameras, but if one has low-refractile glass and the other high-refractile glass, the optical measurement made might be 2mm and 2.5mm.

The bottom line is all of the sensor stack thickness measurements we have are guesstimates, probably accurate to 0.5mm or perhaps 0.25mm. However, since a difference of 1mm should have only a minimal effect that’s probably accurate enough.

Our optical bench measurements with f/1.4 lenses seem to support this. When we measure lenses with a difference of 1mm of optical glass in the path we see almost no change on-axis and usually just a mild astigmatism change off-axis. A 2mm difference usually causes significant problems, though.

The summary regarding sensor stacks seems pretty clear. Leica cameras (with possible exception of the M240) have less than a 1mm stack. Most SLRs are around 2mm. Micro 4/3 cameras, with the exception of the Black Magic cameras, are about 4mm.

We should expect the following to give problems:

1. Using a lens designed for film on an SLR would give a 1.5 to 2.5mm stack difference, and we should notice a performance drop-off when using wide-aperture lenses with shorter exit pupil distances.

2. Using a lens designed for an SLR on a micro 4/3 camera would give a 2mm stack difference, and we may notice a performance drop-off on wide-aperture lenses with shorter exit pupil distances.

3. Using a lens designed for film on a micro 4/3 camera would give a 4mm difference, and if the other factors (exit pupil distance and wide aperture) are present we will almost certainly notice a performance drop-off.

On the other hand, using a Nikon lens on a Canon camera, or either of those on an NEX or Fuji camera shouldn’t give major problems since all of those sensor stacks are similar.

This should answer one question that several people have asked: Do third-party lens makers have to alter a lens optically when making a Canon versus a Nikon mount? I wouldn’t think so; the sensor stack thickness is very similar. It may (I’m just guessing) also answer an unasked question: Did Zeiss decide to make Tuitt lenses for Sony and Fuji, but not m4/3 because of the sensor stack difference? I don’t know the answer, but Sigma makes the same mirrorless lenses for both m4/3 and NEX. There may be compensating optics in those, but also none have wider aperture than f/2.8, which might minimize the optical difference.

Rangefinder Wide-angle Lenses

As the graph above shows, lenses with an exit pupil distances of less than 50mm are significantly affected by a sensor stack difference. Those with exit pupil distances of less than 25mm are greatly affected. Glancing at the database, it’s obvious that most SLR lenses have an exit pupil distance of 50mm or more. Many rangefinder lenses are less than 50mm and some (especially wide-angles) are around 25mm.

The lens database has gotten fairly large, so I’ve changed it to be sortable, which should help you find what you’re looking for. A couple of lenses do stand out that, according to theory at least, should make bad adapter candidates (at least when shot wide open).

The Voigtlander 15mm f/4.5 Heliar M mountZeiss ZM 15mm f/2.8 DistagonZeiss ZM 21mm f/2.8 Biogon, Leica-M 28mm f/2.8 ASPH Elmarit, and Zeiss ZM 35mm f/2.8 Biogon all have exit pupil distances of less than 30mm.

A number of other rangefinder lenses have exit pupil distances of less than 50mm, but particularly notable are the Voigtlander 35mm f/1.2 and 50mm f/1.5, both of which have wide apertures in addition to short exit pupil distances. Obviously, stopping these lenses down will help minimize the effects of sensor stack if you just have to shoot them on an m4/3s camera.

SLR Lenses

Most SLR lenses have exit pupil distances of 50mm to 100mm. One thing that might be lost in the table is that the Sigma 18-35mm f/1.8 Art lens has a nice, long exit pupil distance, especially at the side end where it’s 150mm. This should make it a superb lens to use on cameras with different sensor stack thickness. Other ‘long exit pupil distance’ lenses include the Nikon 35mm f/1.8 DX (136mm), Nikon 24-70 f/2.8 AF-S G (116mm at 24mm), and Tokina 11-16 f/2.8 (100-117mm depending on zoom distance), and Canon 50mm f/1.2 L (103mm). The Canon 17mm f/4 TS-E (91mm) and 24mm f/3.5 TS-E (86mm) also have fairly long exit pupil distances. All of these should give good performance even out to the edges (assuming a good adapter, of course).

Summary

While I don’t have a mathematical formula to predict how well a given lens will work on a given camera, it should be apparent that wide-angle rangefinder lenses designed for film are going to struggle, especially on m4/3 cameras, but to some degree on other cameras. Longer focal length rangefinder lenses tend to have longer exit pupil distances and should do better. Stopping down to f/4 or f/8 should reduce the problem significantly, although it isn’t guaranteed to eliminate it near the edges of the frame.

SLR lenses may also have some troubles on m4/3 cameras, since there’s nearly a 2mm difference in stack thickness. They shouldn’t have as much of a problem as rangefinder lenses, though, since they are designed for a 2mm sensor stack and they tend to have longer exit pupil distances. Older lenses designed for film cameras will tend to have more issues, of course, since they were designed for no sensor stack.

Obviously, these are generalizations and suggestions. There are going to be exceptions. But, if nothing else, hopefully we can help a few people stop trying one adapter after another, hoping to make their 15mm rangefinder lens look great on the m4/3 camera.

 

Roger Cicala and Brian Caldwell

Lensrentals.com

July, 2014

 

Author: Roger Cicala

I’m Roger and I am the founder of Lensrentals.com. Hailed as one of the optic nerds here, I enjoy shooting collimated light through 30X microscope objectives in my spare time. When I do take real pictures I like using something different: a Medium format, or Pentax K1, or a Sony RX1R.

Posted in Other
  • Roger Cicala

    Horst, I certainly think it’s possible to some degree without affecting things. We can’t ‘see’ much of a difference between 1 and 2 mm when we test lenses, so I doubt that kind of reduction would have a major effect.

  • Horst Tobart

    Thank you very much for doing these experiments and publishing the results in this very accessible article!
    For me it has had a huge impact on my initial plans for adapting lenses, it certainly does many others, too.

    Could you please make an educated guess about the statetment:
    Given that current lenses for Sony, Nikon, and Canon are designed for 2 mm filter stacks, can and will these manufacturers reduce the filter thickness in future camera models without loosing compatibility with current lenses?

  • Ted Miller

    I am a poor Nikon shooter who has some of those film lenses still around. I reazlize that the longer telephotos have a huge exit pupils, so the effects may be minimal. I have a lens or two that has the filter slot between the lens and the camera, and it brings back to my mind the manual that says to always keep “something” glass in there, as the lens is designed to include that glass. Now that there is a couple of mm of glass over the sensor, should I be leaving the glass out of the filter-holder?

  • Ilya Zakharevich

    Vivek,
    > Olympus went to extraodrinary lengths to expound on the virtues of telecentricity (and claim all their lenses were “nearly” telecentric).

    So, what can you say about their *actual* telecentricity?

    Brian,
    I have not seen it discussed: what do your converters do with the distance exit pupils? Do they improve telecentricity?

  • Ilya Zakharevich

    Brian,
    indeed, the mystery of the usefulness of the top graph is explained by MAGNIFICATION of your converter. For example, 40 lp/mm AFTER 0.58x converted corresponds just to 23 lp/mm BEFORE the converter. So while 1.6mm mismatch with f/1.0 lens hurts “only” by MTF=55% at 40 lp/mm (which is VERY sharp), BEFORE magnification this is MTF=55% at 22 lp/mm, which (at center!) is indeed not a big deal for a f/1.7 lens to achieve. Thanks for your patience in fixing my misjudged remarks!

    So my current summary of your graph is: near f/1.0 one needs an extremely sharp lens to notice the effect in the center (but with your converters, even mediocre lenses become extremely sharp in the center). Above f/1.4, the effect would probably be not noticeable without special equipment.

    (BTW, the test shots you references are behind a registration-wall.)

  • I have a question about the following excerpt from the article:
    1. Using a lens designed for film on an SLR would give a 1.5 to 2.5mm stack difference, and we should notice a performance drop-off when using wide-aperture lenses with shorter exit pupil distances.

    What differences would be noticed using a Canon lens used on their EOS 1V (film) and one of their full frame DSLRs?

  • Lee Saxon

    Does this discussion have anyone else eyeing maxmax.com? I’d been leaning toward getting something converted to OLPF-less anyway, this just makes it seem like an even better idea. I’m a manual alt-glass shooter, so (a) my lenses are for film, this is very relevant to me and (b) I don’t need autofocus to work.

  • Brian Caldwell

    Vivek: Actually, I hope you get involved in lots of discussion, particularly if you know about something that seemingly doesn’t “fit” the trend (e.g., the Voigtlander 12mm, which I want to check out as soon as I can get my hands on one). Also, if you are aware of any glaring errors in the filter stack database it would be great to find out about it.

    Ideally we would measure a filter stack by two methods: 1) first by the non-contact optical method, and 2) the physical method with a micrometer. By combining the results you can accurately determine both thickness *and* refractive index for all the component plates.

  • Vivek

    Hi Brian,

    Apologies if my choice of words were poor.

    I shall refrain from discussing anything here from now on.

    I appreciate the chance to type a few words here to you and Roger.

  • Brian Caldwell

    Ilya: Actually the results down to f/1.0 and below are very pertinent to people using Speed Boosters because it is so easy to get down to f/1.0 or faster this way. For example, the standard m4/3 Speed Booster used on Blackmagic cameras results in noticeable undercorrected spherical aberration because the Blackmagic filter stack is only 2.4mm thick instead of the usual 4.0mm. And this is noticeable with garden variety 50mm f/1.4 and f/1.2 lenses – no need to use a Zeiss Otus to see the problem.

    So when developing custom Speed Boosters for the Blackmagic cameras we had to take the actual filter stack thickness into account. By doing this its possible to get sharp images down to f/0.74. Post #50 of this thread ( http://bmcuser.com/showthread.php?9441-Speedbooster-for-GH4-BMCC/page5 ) on BMCUSER shows a quick and dirty test with the Nikon 55mm f/1.2 lens that illustrates what I’m talking about.

    So, even with “lousy” lenses like the ancient Nikon 55/1.2 the issues with filter stack thickness can be very significant.

  • Brian Caldwell

    Vivek:
    Regarding coverglass thickness, this is fortunately very easy to measure (at least the optical thickness) by the non-contact method. On all of the cameras I’ve looked at the coverglass portion of the filter stack gives the least ambiguous optical thickness results, and the measurement accuracy is generally better than +/- 5 microns. This combined with the fact that the glass material itself is almost certain to be Schott B270 or similar means that accurate values for both physical and optical thickness of the coverglass are usually easy to obtain.

    I personally haven’t measured a camera in which the coverglass is actually a part of the AA stack, but then I’ve only measured one Canon camera.

    Regarding whether this whole effort has been useless, I’m sure that some of the data and results will be improved and refined over time. However, I think that both the filter stack and exit pupil data are already sufficiently accurate that the rule-of-thumb conclusions are valid.

  • Vivek

    “I hope this hasn’t been asked before, but why exactly do manufacturers choose differing thicknesses of optical glass over their sensor? And why so much thicker in mirrorless cameras? Does it actually help with anything if the glass is thicker? The only reason I can think of is to lock lenses from other systems out, which would be somewhat odd given that lens adaptability is a major selling point of mirrorless systems”

    That question (and EcoR1’s) stem from Olympus’ 43rds system. Yes, Olympus did that to “lock out” competition. Olympus went to extraodrinary lengths to expound on the virtues of telecentricity (and claim all their lenses were “nearly” telecentric).

    The m43rds kept those “standards” only for the cameras and not for their initial line up of the lenses. Olympus, in particular, churned out run of the mill lenses. Their 17/2.8 and the kitzoom with light leak and so on while Panasonic came up with splendid fisheye, 7-14/4 and so on. Over the years, this has changed for the better (there were abominations like the Panasonic 12.5mm stereo lens and the Olympus body cap lens- both past diffraction values).

    “Also, does the measurement of the sensor stack’s thickness include filtration elements (OLPF, UV, IR) or just the cover glass? In the latter case, how do those elements factor in lens performance (beside their obvious effects)?”

    Excellent point. Shane Elen’s data listed in an earlier post, AFAIK, applies only the UV/IR cut filter in the entire stack. While I understand the difficulties in assembling data from a vast range of sensors/cameras, making up a story based on a set of wrong data is absolutely useless.

    FWIW, it is the cover glass that is the most difficult to get to and measure. According to Dan at MaxMax, the cover glass on (many) a Canon sensor is also part of an AA stack.

  • Ilya Zakharevich

    I wrote:

    ========================
    f/1.0 lens is still not making much practical sense: I expect that this setting of the lens is not usable with the current generation of sensors, since the field of view of a sensel is much smaller than f/1.0.
    =============================

    Actually, I was very wrong here. As threads on DPReview show, sensels of MFT sensors have a very good viewing angle; for example, f/0.95 lens is sensel-vignetted only by 0.4EV. This means that the graph in the beginning of this post MAY be applicable to extremely sharp f/1.0 lenses.

  • Ilya Zakharevich

    @Brian:

    Thinking yet more: your example of f/1.0 lens is still not making much practical sense: I expect that this setting of the lens is not usable with the current generation of sensors, since the field of view of a sensel is much smaller than f/1.0. And with f/1.7 (which is reasonable to expect the vignetting-by-sensel to reduce this lens to), the center performance could be dominated by lens aberrations, not by the sensor glass aberrations. (Needs extra analysis!)

  • Ilya Zakharevich

    @Brian:
    according to the measured MTF in the first part of this series, the performance of Otus+x1.5converter at f/1.0 is (extrapolated) about MTF=0.6 at center at 40 lp/mm?—?which, as I said, is significantly worse than a perfect lens at f/11.
    On the other hand, you are right that the measured performance with 4mm thickness mismatch is (extrapolated) about MTF=0.28, which almost exactly matches the performance of a perfect f/1.0 lens with the same mismatch. So the imperfection of Otus+converter is **not** hiding the aberrations of the flat glass, as I (wrongly) presumed.

  • Sebastian

    I hope this hasn’t been asked before, but why exactly do manufacturers choose differing thicknesses of optical glass over their sensor? And why so much thicker in mirrorless cameras? Does it actually help with anything if the glass is thicker? The only reason I can think of is to lock lenses from other systems out, which would be somewhat odd given that lens adaptability is a major selling point of mirrorless systems.

    Also, does the measurement of the sensor stack’s thickness include filtration elements (OLPF, UV, IR) or just the cover glass? In the latter case, how do those elements factor in lens performance (beside their obvious effects)?

  • Brian Caldwell

    David Cockey: A plane parallel plate will introduce zero aberration in a collimated beam (parallel light). So, an ideal filter in front of a lens focused to infinity will have no aberrations. As you focus closer the aberrations become non-zero, but you have to get into the macro/micro range before they become significant.

  • Brian Caldwell

    Ilya: You’re right that optical perfection is rare in the photographic world once you go below ~f/5.6, although lenses like the new Sigma 50mm and Zeiss Otus are really pushing that envelope.

    Nevertheless, its reasonable to show results for a perfect lens simply because it eliminates variables and shows the effects that are directly of interest.

    And remember, in the first article of this series you can find real world measurements where f/1 and f/1.4 lenses suffered significant degradation when used with a non-optimal filter thickness. So the assertion that the red curve corresponding to a 4mm filter stack error would be completely hidden by aberrations just isn’t true in practice.

  • Ilya Zakharevich

    Oups, I should qualify my post. I meant only refraction lenses (I forgot what is f-number of KH-12; and LSST has an impressively bright f-number), I did not mean lenses for tiny (1/1000”) sensors (any Blu-ray player has an f/0.65 diffraction-free aplanat), and lenses with very narrow field of view (I suspect the collimators which Roger mentions all the time are diffraction-free lenses with such narrow field of view; several meters 35mm equivalent?).

  • David Cockey

    At the other end of the lens, what is the effect of high quality filter on image quality? Would an ideal filter have any effect?

  • Ilya Zakharevich

    One should take into account that the first graph of this page makes very little sense altogether. I doubt many have seen a “theoretically perfect lens” brighter than f/8 (and probably nobody saw it brighter than f/5.6). Resolution-wise, a typical below-$20000 lens would perform worse than a perfect f/11 (even at the center) at any f-stop. So the orange curve has no bearing on the real world, and the red curve would probably be completely hidden by the aberrations of the lens. (All price numbers are just figments of my imagination.)

  • Tony

    Nikon has been saying that the AA filter components are entirely absent in the D810. It will be interesting to see if that changes the total sensor stack optical distance. You guys wouldn’t mind dissecting the first one you receive, would you? (kidding)

  • mrc4nl

    The sensor stack disadvantage has not stopped me from adapting lenses.
    My favorite is the minolta f2, i did some comparisons to modern lenses like the nikon 50mm f1.4 afs.The minolta @f2 vs nikon @f1.4 my results showed the minolta dispite its age, is a serious competitor.

    But maybe its because of the stack that newer lenses dont do verry well.What make me wonder are the sigma lenses for nex (sony e whatever) and m43.They are designed for aps-c (and 2mm stack) so the results would be sub optimal.The exit pupil is quite close the the stack.

  • Vivek

    Apologies for the typo, Brian. Can not edit that now.

  • Vivek

    Hi Brain,

    As of today, no one who owns a CV-15 is going to try to mount it on a m43 camera with 2X crop to get the digital use out of it. There are plenty of other choices even up to FF sensored cameras. Vignetting was always a problem with the CV15, even on film. Only the color shifts is a digital problem. It is there- big time- on a Sony A7r, less so on the A7 and according to samples on the web, virtually non existent on the A7s. I do not have the data (but hope to have it very soon for the A7) but I suspect the sensor stack thickness actually goes up from A7r (~1.8mm – NOT what the data base claims!)to A7 to A7s. What increases here also are the active pixel area and may be the pixel geometry are also different.

    Would I choose a CV15 and look for a camera that suits over a humongous distagon 15? Yes, any time and every time. Size and weight being the main criteria. Do people shooting medium format digital backs and wide angles (with very short back focus) complain about color shifts and such? No. That is all taken care of in the post (in addition to using center filters during the capture).

    This is not some odd ball thinking. Consider the “perfect lens”, the OTUS. It is just too huge for what it covers. Zeiss are threatening to come up with a 85mm OTUS. Filter sizes of 100mm and up are not just for movie camera super zooms anymore.

    What about that FE 55/1.8? Considered to be “near perfect” by many. Will transmit in very very narrow range of the light spectrum and produces stunningly contrasty images. Is that a criterion? 😉

    Given the current developments, I have come to the conclusion that there is only one lens maker in the world now who makes the best performing and the most compact lenses and they are beginning to look like a bargain, especially when considering the OTUS line. 🙂

  • http://xkcd.com/54/

    And also: I have bought an a7s (I shoot mostly video) and had planned to look for a small lens (maybe even pancake) to go with it when I want it nearly-pocketable. This is going to rule out most of my candidates. Maybe I should just go with the Sony FE 35mm f/2.8.

    Could we get a Sony FE 35mm and 55mm vs Sigma Art 35mm and 50mm shootout?

  • Brian Caldwell

    Hi Vivek: 2 items that are relevant but extremely difficult to quantify in a rule of thumb manner are vignetting and aberration correction. Both of these are essentially infinitely variable among different lens designs. Aberration correction in particular can make the net effect of the filter stack worse or better depending on the magnitude and sign of spherical aberration, astigmatism, etc. .

    Regarding the 15mm Voigtlander, it has the shortest exit pupil distance currently in the database, so difficulties with digital cameras aren’t unexpected if we assume it was designed for film. A set of MTF measurements taken with various filter thicknesses would reveal a lot here. We have no exit pupil data yet for the 12mm Voigtlander, so there’s not much to say about it for now.

  • EcoR1

    Interesting question is why m4/3-cameras use thicker sensor glass than other manufacturers? Did they also use ~4mm thick glass with 4/3-cameras?

  • Vivek

    “But, if nothing else, hopefully we can help a few people stop trying one adapter after another, hoping to make their 15mm rangefinder lens look great on the m4/3 camera.”

    There is no digital camera that makes the C-V 15/4.5 lens look great with a possible exception of the Sony A7s. This pokes a serious hole in your story and your underlying criteria. It is sad that you pick a 15mm lens to end the story. Other than that may be (just may be) it could be OK.

    The wider C-V 12/5.6 poses no problems like the C-V 15/4.5. This also does not make this story strong.

    Let me pose this question: Are there other criteria that you have not considered and are very relevant?

  • Thanks once again to both you and Brian for doing this work and organizing the information in a useful way. This is really great stuff.

    One thing to consider about m43 is that it will use a much smaller, central part of the lens’s image circle than a full-frame sensor, so that may mitigate some of the bad effects. For example, compare the 50mm lines in the 2nd figure at 8.5mm (half the m43 sensor width) and 18mm (half the sensor width of full-frame). It’s not great at 8.5, but it’s far worse at 18.

Follow on Feedly