03-19-2015, 06:39 PM
If you wish to get into pedantics, virtually all of optics was stolen from Carl Zeiss AG, as their scientists invented most of the math and techniqes for lens design. The difference here is that Sigma stole a specific implementation of a mechanism from Nikon. The set of possible double gauss, tessar, sonnar, etc, lens designs is infinite. The set of possible ways to precisely move an optical unit such that it stabilizes the image passing through the system is very, very finite.
Sigma's QC method is insufficient. The spacial resolution of the sensor is lower (and crucially, lower than many sensors these lenses will ultimately be placed in front of), it is not of sufficient size to see the worst manifestations of bad alignments (spherical aberration is relatively insensitive to alignment error while astigmatism and coma are very sensitive). They also do not only let lenses which are "100%" through QC. All manufactures have tolerances, at all stages of the manufacturing. The lens design itself is extensively toleranced so that the radii of curvature, aspheric coefficients, lens and airspace thicknesses, and refractive indexes of the glasses may vary somewhat. The nominal design is not the specification for pre-manufacturing.
Further than that, the assembly is toleranced so that small alignment errors are accounted for and lenses with small errors still pass QC.
For the past several weeks I have been testing nine prototypes of a single lens. This lens is approx. 6-8x higher resolution than the average (or even "good") ILC lens. So far five of the prototypes have been rebuilt more than once to attempt to get them to meet spec. Without testing the lens on an MTF bench, interferometer, or other optical metrology tool you cannot with sufficient confidence tell how close to the nominal design (and thus, how close to spec) a lens is.
One camera manufacture does lens QC by taking pictures of a bookshelf across the room. This is horribly insufficient. I do not know what most do, but I do know a very high end limited production company does their QC by actually coupling each and every lens to an interferometer. This is the best form of QC. Any method that involves sampling the lens spot with a sensor without prior enlargement of the spot is insufficient.
"Case example" - one copy of a Zeiss 135mm f/2 APO-Sonnar was dropped or otherwise damaged but was still fine. The lens normally has an MTF at 30lp/mm (a frequency of interest) of about 75%. This is well above what is necessary for current-gen FF. The dropped lens had an MTF of about 45% at 30lp/mm, 30% below the manufactured and tested average. Clearly the lens has an issue. Yet, the MTF is just under 50% at the "true nyquist" for most FF sensors, so it may look just slightly softer than your average copy.
This is the issue with camera-based QC.
Sigma's QC method is insufficient. The spacial resolution of the sensor is lower (and crucially, lower than many sensors these lenses will ultimately be placed in front of), it is not of sufficient size to see the worst manifestations of bad alignments (spherical aberration is relatively insensitive to alignment error while astigmatism and coma are very sensitive). They also do not only let lenses which are "100%" through QC. All manufactures have tolerances, at all stages of the manufacturing. The lens design itself is extensively toleranced so that the radii of curvature, aspheric coefficients, lens and airspace thicknesses, and refractive indexes of the glasses may vary somewhat. The nominal design is not the specification for pre-manufacturing.
Further than that, the assembly is toleranced so that small alignment errors are accounted for and lenses with small errors still pass QC.
For the past several weeks I have been testing nine prototypes of a single lens. This lens is approx. 6-8x higher resolution than the average (or even "good") ILC lens. So far five of the prototypes have been rebuilt more than once to attempt to get them to meet spec. Without testing the lens on an MTF bench, interferometer, or other optical metrology tool you cannot with sufficient confidence tell how close to the nominal design (and thus, how close to spec) a lens is.
One camera manufacture does lens QC by taking pictures of a bookshelf across the room. This is horribly insufficient. I do not know what most do, but I do know a very high end limited production company does their QC by actually coupling each and every lens to an interferometer. This is the best form of QC. Any method that involves sampling the lens spot with a sensor without prior enlargement of the spot is insufficient.
"Case example" - one copy of a Zeiss 135mm f/2 APO-Sonnar was dropped or otherwise damaged but was still fine. The lens normally has an MTF at 30lp/mm (a frequency of interest) of about 75%. This is well above what is necessary for current-gen FF. The dropped lens had an MTF of about 45% at 30lp/mm, 30% below the manufactured and tested average. Clearly the lens has an issue. Yet, the MTF is just under 50% at the "true nyquist" for most FF sensors, so it may look just slightly softer than your average copy.
This is the issue with camera-based QC.
Quote:What makes sigma a better company than Nikon? From my PoV they aren't better than Nikon at much of anything except cleverly cutting QC and marketing it as more strict QC and offering lower prices.
Maybe it's time for Nikon to learn from a better company?