hiepphotog
Offline
• • • •
Upload & Sell: On
|
Brandon Dube wrote:
Strictly speaking, all of the MTF/variation stuff is done by Olaf Optical Testing, not LR.
The K8, ImageMaster, LensCheck, and Wells Research MTF benches all measure in the same way with minor differences in the light sources used and targets imaged. The LensCheck supports 6 different targets to calculate MTF from - different sized pinholes and slit targets if one wishes to measure only one orientation with the utmost accuracy. The ImageMaster is limited to a set of perpendicular slits (i.e a cross) because of patents filed by Optikos on videoMTF from a pinhole. The K8 is an "oldschool" machine to the best of my knowledge and has what is essentially a rotating target with slits of increasingly fine frequency printed in it which are imaged by a collimator and re-imaged by the lens under test.
The LensCheck system is horizontal, as is the Wells system, both the ImageMaster and K8 are vertical - though the K9 is horizontal I believe. The ImageMaster rotates the object about the Lens Under Test (LUT) as does the K8, but the Wells and Optikos machines rotate the LUT.
In terms of numbers, if you measure the same exact sample on two machines in the same conditions with the same spectral filtering, you get the same result to the limit of the accuracy of the equipment manufacture's calculations. These machines essentially take an image of a known target, subtract the known, and FFT the results. There is nothing different happening in any of them, and you can find a very small bit of the math in This paper from Optikos. The LensCheck manual has the complete math, but I am not sure Optikos would like me to share the manual here, so I will withhold.
Regarding the manufacture charts - I believe Canon's charts are the "as-fabricated" option from the design software they use. This is because sometimes the measurements exceed those numbers on axis in the average by a fair margin, and a copy would need to approach the nominal design to exceed that. The off-axis differences are all down to build aberrations and to some extent measurement reality. A big part of the reality is that CAD is unaffected by vignetting, while the machine must increase exposure to counteract vignetting. This causes the visible size of the spot to increase, lowering MTF. It is a fundamental problem of MTF benches, but I know for a fact both the LensCheck and ImageMaster machines do corrections for this.
Zeiss' results are indeed measured, but I believe they are at best focus for each point. E.g the ZE 21mm has simply impossible MTF for the model in their charts if you take best center focus and that's it. The most nominal copy of the ZE 21mm I've seen (of about 30 copies on the bench, mind you) has an on-axis MTF at 40lp/mm of 0.82 and at 7/10 field where the astigmatism inverts it is about 0.42. Zeiss' measurements show strong astigmatism there, but if you did MTF vs Field vs Focus you would see that what is happening is the two focal planes are curved differently and their curvatures intersect there, providing no astigmatism. Zeiss' measurements having astigmatism says to me that they are focusing for the highest absolute MTF there (absolute meaning it looks for MAX(Tan+Sag)).
Either that or they find a very well assembled but tilted as shit copy and measure the better side. Tilt + Field Curvature brings one side more into focus and one side more out of focus, raising MTF on one side and lowering on the other.
Please please please do not take Roger's word on tolerancing as gospel. I do not mean this to disrespect Roger, he is very intelligent and highly educated, but tolerancing is the most advanced topic in lens design, and he is no lens designer. It is beyond prime design, zoom design, aspheric surface design, all of these things are easier than tolerancing. I think Canon's newest designs are very intelligently toleranced, but the spots from all of the copies of the exceptionally consistent lenses are seriously clean. If the tolerances were wide the MTF would be high but the spots would be messy with many misalignment aberrations on axis. This is not the case, and the only thing I can think of is that canon is building their lenses on robotized alignment equipment. Jenoptik has a few systems in Florida that do this - using an interferometer and servo feedback loop to automate alignment, but the cost is about $1-1.5M USD per machine and is much slower than high speed serial manufacture. I do not think canon is doing this but is instead using very accurate robotic arm based manufacturing to place all pieces in the same position to a very tight tolerance, and spends a long time tuning that position with a skilled operator before initiating production, but this is only my own speculation.
I think complexity is part of it but also that the manufacturing plays a key roll. If Cosina has 25 employees producing the ZE/ZF line and each employee could build 15 pieces per day - about a whopping two per hour, that's about 100,000 ZE/ZF lenses per year. I do not think Zeiss is selling a volume hugely different than that of these lenses. The idea that the Otus are built by some uber master technicians with 35 years experience is a bit silly. Maybe they are, but there absolutely are not special employees who build the otus lenses and nothing else. Manufacturing and manufacturing line metrology determine the consistency more than the tolerancing does - if the Otus lenses are not built by different people on a different line with different (and better) equipment, they really cannot make them better. I do not believe they are built in a different environment, simply because it is not economical.
I do wonder why you think this. OLAF has two very intelligent full time employees and this summer had an optics student ( I wonder who that could be  ) working for them and actually churning out this data.
There are two big contradictions I can think of - one being imatest vs MTF bench, which is quite silly. Imatest is testing about 10 variables at once (target, lighting, support equipment/environment, alignment, focusing accuracy, camera processing, sensor, lens, sensor stack, raw converter...) while the MTF bench uses many calibrations to isolate the lens. This is not possible with imatest. The other is Roger's old data on the 11-24 vs my newer data - the "old" data method was to align the point of rotation at the front element of the lens, but this is incorrect and the lens should be aligned with its front principal plane (approx. the entrance pupil) at the point of rotation. Otherwise one indices vignetting and the results are not too correct. In teles this matters much less, but in something as wide as the 11-24 there is little margin for alignment error.
The ~1300 samples I measured over the summer had great care to be done as consistently as possible, with concern for the speed of measurement and I must admit I let myself down when I realized when we published the 50mm set that the ZE 50mm MP had "Zeiss" in its title instead of "ZE," which pushed me to increase the automation of the software I made for OLAF to prevent inconsistencies like that. If you mean testing at different F numbers, the only rebuttal I can really offer is that the f/1.4 lenses are designed at f/1.4, and when the pupil coordinates are normalized in the design software, it is done at maximum aperture and not some lowest common denominator.
As a result, any test for the alignment quality must be done at full aperture; as the aperture-dependent aberrations (spherical aberration, coma mostly) are significantly removed by stopping down. Coma is caused by two spherical wavefronts that are not centered on each other; so when you decenter an element that isn't a double plano, flat lens, you cause coma. If you stop down two stops you have removed 75% of the third-order coma, as it is field linear. The higher order coma is less generalize-able.
Here are some papers on alignment aberrations from some of the leaders in freeform optics:
http://www.creol.ucf.edu/research/publications/2011.pdf
http://www.ora-blogs.com/files/thompson_usingnodalaberration_spie_2009-2-1.pdf <- good translation to understand the full-field display in here
http://etd.fcla.edu/CF/CFE0003339/Schmid_Tobias_201008_PhD.pdf <- a past student of Jannick's Ph.D.
Here is a snippet from my own CV covering some of the projects I have worked on in the last year: http://i.imgur.com/9XbM22D.png
Best regards,
Brandon...Show more →
Thank you Brandon for the clarifications. This whole Zeiss business of reporting the best Tan+Sat value on their MTF charts is interesting. That would explain why most Zeiss MTF has much better field curvature behavior than Leica. However, under real-world testing, it still holds true but probably not to the extent of what the MTF chart would suggest.
I did mix up between Optikos OLAF and that Imagemaster. I thought whenever you guys publish MTF charts, it would be from the Imagemaster. Anyway, with you going back to school after summer, I guess it would be slow from now on. Have you tested any Leica lens?
|
