Just MTF Charts: Perspective Control Lenses

48 Comments

1. l_d_allan · April 29, 2019

   Thanks … for the whole MTF series.

   Seems like the chart for the
   Rokinon-Samyang 24mm f3.5 TS
   has an incorrect label for Canon 24mm.

   1. Roger Cicala · April 30, 2019

      Thank you, the computer defaulted to what it thought was best and I missed that. Corrected it now.

2. Andreas Werle · April 29, 2019

   Thanks for the TS-Series, Roger.

   Now you have done, what you denied to do, measuring Macro-Lenses. 🙂 If i remember well, Macro lenses are optimized for a distance of about 1 meter, not for infinity. Am i correct, that OLAF cannot speak MTF at short distances? Guess the new 90mm TS-E will do even better in a close up view.

   Greetings Andy

   1. lwestfall · April 29, 2019

      Hopefully this means the next batch of MTFs will be for macro lenses? 🙂

   2. Daniel Eggerstorfer · April 30, 2019

      Roger already published the MTF Data of the Canon 100mm 2.8 Macro (both), Nikon 105 2.8 Micro and Sony 90 2.8 Macro.
      Hopefully there will be a nice “Macro Round up” . Most of them are very pretty Portrait-Lenses

      1. Someone · April 30, 2019

         There was alread such a Macro round-up, including comparison of macro distance to infinity focus. https://www.lensrentals.com...

   3. Roger Cicala · April 30, 2019

      We can test at Macro distances, although it’s a PIA and we usually don’t.

3. Lee · April 29, 2019

   The text says the Schneider 28 tested is the old 2.8 PC, but the graphic says it’s the 4.5 PC-TS? I’ve heard the latter isn’t on par with the 90 PC-TS, but it’d sure be sad if it was *this* bad.

   1. Roger Cicala · April 30, 2019

      Sorry Lee, the data is for the older f2.8 28mm. Same problem as below, the computer database was guessing on proper graph labels. I corrected the label.

4. Alan · April 30, 2019

   Thanks for doing this!

   I’m also wondering which Schneider 28 this is. Info on the newer one is non-existent out there.

   1. Roger Cicala · April 30, 2019

      It’s the older f/2.8 Alan, I corrected the label.

      1. Max Manzan · April 30, 2019

         If it’s the older one, it’s not “PC-TS” but only “PC Super Angulon”, as it was only shift, not tilt-shift.

         1. Roger Cicala · April 30, 2019

            That’s the one. The Schneiders were rare birds for me; I only saw most of them the day I tested them.

5. Franz Graphstill · April 30, 2019

   The Canon TS/E 135mm f/4L shows impressively flat MTF curves.

   However the Sony 135mm f/1.8 is still sharper at f/1.8 than the Canon 135mm f/4 is at f/4, in the centre. 🙂

   1. lwestfall · April 30, 2019

      But it does look like the Canon TS-E 90mm f2.8L Macro (at f/2.8 in the center) is sharper yet.

   2. Andreas Werle · May 1, 2019

      Hi Franz,
      this is also true for the Canon 135/2 and the 200/2,8. In these lenses the MTF-Diagrams are indeed a little bit boring, perhaps due to a solid and unspectacular design. I own a Canon EF 200 2,8L II, it is a very nice lens and shines with my Canon 6D. A beautiful and pretty inexpensive combination.

      Greetings Andy

   3. preston · August 27, 2019

      Are you seriously comparing at manual focus only f/4 tilt-shift lens to an autofocus f/1.8 lens?

      1. Franz Graphstill · August 27, 2019

         Why not? Both are 135mm lenses. I could use either to shoot at f/5.6 with zero tilt and zero shift and manual focus.

         Sure, there are things that one can do which the other can’t – the tilt/shift can’t reach f/1.8; and the 135GM can’t tilt or shift (well, not intentionally…).

         I can compare an 18 wheeler with a Tesla if I want – both can be used to take a passenger from A to B 🙂

         1. preston · August 28, 2019

            Ok, thanks for giving that other example. You’ve confirmed that sensibility of comparisons isn’t of value to you. By the way, Canon also makes a super sharp 135mm f/2.0 that is less than half the cost of the 135mm ts-e, so obviously they are different animals.

6. Kers · April 30, 2019

   May thanks for this – indeed – missing TS test.
   Clearly the new Nikon ( 19mm) and Canons are making a jump in quality.
   They are all very impressive…

   What is lacking in this test is that no owner of aTS lens uses the lens wide open- usually at f8-f11.
   Having used all Nikon PCE lenses the test confirms that the 24mm PCE is the weakest link in the line-up.
   ( a curved filed of sharpness is its main problem) The 45 and 85 PCE are still good enough on 46 MP.

   The 19mPCE is a very strong contender, also because it can handle the 12mm shift very well.

7. hendrik · April 30, 2019

   Thanks Roger for this amazing database. Yes, for the T/S Lenses there are only real world tests, and chart pic’s (from TDP).

8. Michael Tiemann · April 30, 2019

   A bit surprised/disappointed the lenses weren’t tested with larger image circles. I agree that it’s tough to really wrap one’s hands around tilt measurements, but shifts are straightforward and meaningful if you don’t have a larger sensor that can otherwise capture the larger image circle.

   1. Brandon Dube · May 1, 2019

      Our bench only does +/- 20 mm. It is not possible to do a measurement that is at all automated if we used the lens’ shift feature. Non-automated at $300/hr would value a shifted dataset at 4 to low 5 figures per sample.

9. Samuel Chia · April 30, 2019

   Hi Roger, in a previous post for the Sony GM 135mm, you mentioned “We would consider an MTF of 0.5 at 50 lp/mm to be very acceptable. This is hugely better, nearly 0.8 in the center. We’ve never seen that kind of resolution before.” Now looking at the (truly insane) MTF of the Canon TS-E 90 Macro, not to mention it wasn’t measured at macro distances, I’m wondering what you think. Is it totally unfair to compare MTF graphs across different brand lenses? P.S. I own multiple camera systems so I have no interest in making up if one brand is better than another. Would love to hear your thoughts. Thanks!

   1. Samuel Chia · April 30, 2019

      I am aware that the Canon was measured at f/2.8 and the Sony was f/1.8 which would account for quite a fair amount of the difference. Would of course be interesting to see both at f/2.8, though I know you don’t measure stopped down these days (apart from the 100lp 2.8 test).

      1. Roger Cicala · May 1, 2019

         At same aperture you can compare our MTF difference between brands; that’s the benefit of using an optical bench instead of on-camera measurements. The aperture, though, does make a difference, and that makes the measurements different. The Sony should be markedly better a stop down.

         Like everyone else I’d love to have a straight ‘at f/4’ comparison for all lenses. I just simply can’t afford to do it and while people are very kind about helping to support it the reality is it’s not reasonable. We’d probably need to raise, at minimum, half a million dollars to run every lens at f/4. And I’m not certain that would be enough.

         1. Samuel Chia · May 1, 2019

            Thank you for the insight Roger. Wanted to be sure if there were any subtle issues to bear in mind owing to different mount types and your confidence in the data out towards 20mm away from the optical axis, or sensor cover glass issues etc.

            Btw I was just then wondering about the precision your mounts are machined to in terms of flatness in microns. In shopping for lenses that would perform well for astrophotography, I have observed that a great number of lenses are tilted and/or swung in their focus plane, and not merely decentered. Recently a friend acquired 3 Sigma 40mm Art lenses and they were all tilted and swung to different degrees. We estimate the tilt to be some 50-100 microns at the mount for those lenses. This is severe enough that the depth of field at f/2.8 isn’t enough to image sharp stars at infinity. We want the flatness to be within 20 microns, ideally 10 (custom shimming work gets us there for now). We have also found that Sony a7R II sensors to be mildly to grossly tilted or swung, and little has been discussed about the appalling mounting design of the sensor assembly. (Joe has emailed you about this before). Despite our experience, we were misled to believe the lenses were all defective until we realised the cameras were instead, and we have yet to find a straight a7R II, and the III are not of a different design.

            Reviewers who photograph test charts instead have tended to show for the better lenses, peak center resolution is reached at ~f/2.8 rather than f/4. In your experience, have you observed this?

            1. Brandon Dube · May 1, 2019

               The flatness of the mount is important – only the pose of the lens’ boresight to that of the sensor. In a completely different unit, our mounts have a “runout” of under 10 microns from +20 to -20 mm across the image plane. This works out to parallelism of a bit better than an arcminute.
               A runout of 4 mil over the mount of a sigma lens is certainly too much. How are you measuring? At this sort of tolerance, you really need either a nice micrometer with a constant force thimble or a CMM – calipers won’t do. A good mic, good set of calibration gauge blocks, and good user will get you to about +/- 5 microns.

               If you want this to be a relatively constant thing, you will wish to replace the bayonet on your camera with a very hard metal (ex, Ti or a stainless alloy) and your lens bayonets with a softer one (ex. Brass) and re-engineer the bayonets to guarantee constancy of force so they wear evenly, or preload them such that the actual bayonet — the part normal to the sensor — takes all of the wear. Otherwise time will rob you of your precision.

               4/5-axis (and probably even 2-axis) IBIS systems will also be useless to you since they cannot maintain this sort of tolerance.

               1. Andreas Werle · May 1, 2019

                  Hi Brandon,

                  with the new generation of lenses and cameras with huge megapixel-counts, the correct orientation of bayonet, mount and sensor will be more important, simply because you can see the deficits. This raises some questions.

                  One would expect, that with a small sensor and a short flange-distance – for example in case a mirrorless micro 4/3 like the Lumix GH5 – there will be less problems and easier manufacturing of a correct bayonet-mount-sensor-orientation. With a mirror and a larger sensor, the problems should increase and my guess is, that mediumformat cameras – especially those with interchangeable data-backs – need individual adjustments to eliminate errors from a misalignement of image and sensor plane.

                  In case of consumer-cameras with IBIS it should be possible to adjust the sensor electronically. The sensor moves anyway, so it should be possible to correct the plane. As always thanks for your work and your kind responses!

                  Greetings Andy

                  1. Samuel Chia · May 2, 2019

                     Hi Brandon,

                     It seems to me that you don’t understand how IBIS systems work in these cameras, and Andreas too. No, you cannot adjust the sensor tilt or swing electronically. This is a common misunderstanding when manufacturers claim “2/4/5 axis correction”. The sensor moves parallel to the image plane at all times on three ball bearings. You may argue that ball bearings are not perfectly spherical, but I have yet to see proof that the parallelism error introduced by the bearings are in the order of 30 microns or more at the sensor edge. Post custom shimming work, the presence of IBIS has not yet shown observable detrimental effect as far as I can tell.

                     It seems that Sony uses shims in increments of 50 microns to adjust the straightness of their sensors in these alpha 7 series cameras. Roger’s teardown was one piece of vital evidence for this. Given where the shimming points are, if that’s true, then that means that adjustments are possible each way across the sensor in increments of just under half that, (the shimming points are twice the width and the height of the sensor respectively) so 22 to 23 micron steps or so, then, theoretically, the actual error could be limited to half of that at best. However, the errors we are seeing are typically 3x-6x that or even more, so the assumption is that Sony isn’t shimming cameras individually, but rather as an entire batch. Some Canon and Nikon sensors are mounted using spring-tensioned screws, meaning infinitely variable precise adjustments possible, and the entire thing during assembly is mounted on a device where they are able to adjust the positioning and verify straightness in real time. We haven’t studied enough Canon cameras (and no Nikon cameras yet) to make conclusive statements, but we’ve also not yet found one that isn’t straight.

                     This is a serious problem plaguing Sony cameras today, and they need to know that it is affecting people who are needing their excellent quality sensors to do high end work.

                     Interesting points about re-engineering the bayonet mount with new designs and materials. However, my equipment is treated with great care unlike rental gear typically, and the amount of material wear over the useful life of the camera+lens is not likely to introduce errors of 4 mils at the mount. I’ve not seen any observable shifting yet since the shimming was done to my camera. One is just as likely to argue that thermal expansion can also be large enough to introduce focus shift. So I don’t think this path is worth pursuing.

                     A micrometer is useful when measuring the flatness of a mount or a lens adapter. I have been able to achieve a measurement repeatability of under 2 microns. However, it is useless for measuring the amount of tilt and swing in a lens (or of the sensor). You’ll only be measuring the relative parallelism of the mount to the front of the lens barrel, and that has no direct bearing on the optical performance. For a lens, a giant ruler sort of target is best. Tiles of equal size and spacing over a gigantic area, or lines painted at equal distance in a field/road/running track. You want to focus at about 100+ feet. Observe the plane of focus and how it is curving on either side so you don’t confuse field curvature for swing/tilt. Use the formula to convert errors in focus distance to errors in tilt/swing in microns at the sensor edge, then figure out what that is at the shimming points on the mount.

                     Measuring the sensor tilt/swing was considerably more difficult. One might be able to do it with fancy laser interferometry, though I don’t know enough about that to make any useful comments, and it’s not likely that regular photographers would have the means to do so. My friend figured out a brilliant genius method which was shared with Roger. I can forward the email to you if you are interested.

                     1. Brandon Dube · May 2, 2019

                        The roundness spec on a typical “precision” ball bearing is 1 mil (25 microns). If you assume independent, normally distributed errors then you would expect (25^2 * 3)^0.5 = 43 um of runout from 3 of them. Compound this with the flatness of the Vs, which will be worse.

                        The IBIS systems that use 3 ball bearings place 3 balls in 3 Vs, two of which are in a line. This is not kinematic; the system is not fully constrained.

                        Interferometry isn’t fancy – it’s a 70 year old technology and you can build a twyman-green interferometer for ~$2k in parts. To measure tilt of the mount and the sensor that way, all you would do is simultaneously capture fringes on both of them, evaluate the tilt over each in angular units, and take the vector difference of them. That will be the orthogonal components of the tilt between them. The bayonette is a high reflector and the sensor is a relatively low reflector, so your phase unwrapping algorithm needs to be reasonably robust to contrast mismatches. You could do the measurements separately if you had a pellicle and a very stable cavity, eliminating the reflectance mismatch for the metal part.

                        Repeatability is also not the same thing as accuracy.

                        1. Samuel Chia · May 2, 2019

                           I think you’ve just perfectly demonstrated that regular photographers (like myself) are not likely to have the means to make such measurements. The method we used cost us next to nothing in additional equipment and we didn’t need to figure out algorithms to derive the amount of shimming required. We’ve never had this problem with Canon DSLRs in the past. I’ve also not seen any actual proof that the IBIS system in Nikon’s Z7 is causing problems with maintaining the straightness of their sensor mounting, and we now know that it’s mounted by spring-tensioned screws.

                           Also, this should not be a problem that regular photographers need to solve on their own.

                           I read that precision ceramic bearings can be as low as grade 5, or 0.127 microns. That’s way less than the error you’re suggesting. Theoretically, I follow how you on how this prevents perfect straightness in the sensor mounting, but are you actually seeing this occur in practice?

                           Our intention is only to make the mounts straight relative to the opposite side, not machined to a precise thickness. There is only so much we can do to ensure the precision of our measuring equipment without spending hundreds to thousands more.

                           1. Brandon Dube · May 2, 2019

                              I’m not making any statement about anyone’s mount construction, except that 4 mil on Sigma’s lens bayonets is way too much and there must be a measurement error.

                              Silicon Nitride bearings at the tightest tolerance are likely too expensive for Sony to choose – the last time I ordered some of those they were about $10/ea. I can’t imagine that that number will go below $1/pc at volume which is probably way outside the cost target Sony places on those components.

                              1. Samuel Chia · May 2, 2019

                                 Well, not too long ago I had to put in 4 mils worth of correction for my EOS 35mm f/1.4 II. I’m not at all surprised that a Sigma might need the same. If this is out of Canon and Sigma’s spec, people can come to their own conclusions 🙂

                  2. Samuel Chia · May 2, 2019

                     Hi Andy,

                     Actually it would be more difficult to make precise adjustments when the sensor sizes and mount diameters get smaller, not easier.

10. Andreas Werle · May 1, 2019

    Thanks for the link, i missed that.

11. Maksim Medushkin · May 3, 2019

    Thanks for your charts Roger.
    Do you consider measuring MTF of Cine lenses? I understand that it can’t be 10 copies, but it would be very interesting to have even a basic idea of how Angineaux, Fujinon ot Tokina Cinema actually perform.

    1. Roger Cicala · May 3, 2019

       Actually for most of the prime Cine lenses we’ll have 10 copies, although not all. Micro 4/3 MTFs are next, then the Cine’s will start.

       1. Maksim Medushkin · May 3, 2019

          That is great! But please start from those that have no still analogs. Everyone knows what to expect from Canons, Zeiss and Sigmas. But Angineaux for instance doesn’t even publish computer simulated MTF as well as Tokina and Fujinon.

          1. Roger Cicala · May 4, 2019

             I’ll bet you $1 you’re surprised by some of the Canon, Zeiss, and Sigmas. In general, I start with most commonly used, try to set the standard sort of.

             1. Maksim Medushkin · May 4, 2019

                I was under impression that Cine Sigmas should be kinda like still Sigmas, Zeiss CP.3 should be like Milvus and Canon is Canon. So, even they are most commonly used, we all have expectations from them, like what you said yourself for Rokinons. But for Tokinas who knows what to expect at all? It is the most interesting! Just a point of view, no insistance meant or anything.

             2. GuyWith · May 6, 2019

                Having read and listened to lots of cinematographer interviews, very few have said that sharpness is a priority, and here in 2019, many of them still use diffusion filters to buff off some of that edge. They’re more interested in color, beautiful flare behavior, and overall “look.”

                Do you guys have any secret tests programmed in OLAF that might reveal why optics from Angenieux. Cooke, and Zeiss, for instance, are so beloved? I have read that it might be the way aberrations are balanced against one another which gives them their cinematic quality.

12. Maksim Medushkin · May 3, 2019

    Thanks for your charts Roger.
    Do you consider measuring MTF of Cine lenses? I understand that it can't be 10 copies, but it would be very interesting to have even a basic idea of how Angineaux, Fujinon ot Tokina Cinema actually perform.

13. GuyWith · May 6, 2019

    After I shot with a 19mm Nikkor from LensRentals I had to go home and lie down till my head stopped spinning because the images were so perfectly sharp and lacking in geometric distortion. All my buildings looked like cardboard cutouts made by a 300-foot tall child.

    1. Hunter45 · July 17, 2019

       Was that a “PC-E Nikkor 19mm f4E ED”?

       1. GuyWith · July 17, 2019

          Yes that Is theone

14. Andreas Martin · December 22, 2019

    Excellent test. Highly appreciated. Leica just recently added more APO lenses for their SL2 lineup. If I did get it right they tested them up to 60 lp/mm and they were even then in a 0,9 range. That would justify almost any price… Did you ever test any of the APO Leicas? The chief optics developer claims the 35mm is their best/masterpiece.

15. MSC · April 29, 2021

    For most users who are using these lenses at small apertures, I am not sure how useful are mtf charts measuring performance at maximum apertures. What matters to me is how they perform at f8 or f11, which is where I would mostly be using them for landscape photography.