Reflections on current optical designs
Recently I had the chance to analyse MTF measurements form some Canon lenses and a spate of others. The results will be quite surprising and will blow away some long-standing views and myths. Macro lenses.
I could compare the Canon FD Macro 3.5/50mm with the current EOS Macro 2.5/50mm. The FD has the standard six element Gaussian layout with a thinker inner element that allows for some field flattening. The EOS version has nine elements and AF.
The lenses were measured on the famous Zeiss K8 equipment and should generate very accurate results. The overall performance of the Canon 3.5/50mm is outstandingly good.
Even wide open at 3.5 we already have an average contrast across the image field of 90% for 10 lp/mm and 60% for 40 lp/mm, going from 80% on axis to 40% at the corners. At 5.6 the optimum is reached with above 90% for the 10 lp/mm and 80% contrast for the 40 lp/mm over a larger proportion in the centre, dropping to 60% over most of the rest of the image area. The EF 2.5/50mm wide open is quite disappointing, optically and mechanically because the optical performance is reduced by large assembly errors, but even neglecting these, the inherent performance is weak. Stopped down to 5.6 and 11, we get almost identical performance as we saw with the FD version. The aim of the designers is obviously to provide the same quality, but with a wider aperture for a more clear view of the subject when focusing.
The interesting point here is the constant quality of the two lenses that are separated by a full generation of lens design (FD from the mid sixties and the EF from the mid eighties). Top: FD, bottom EF
Given the fact that even today the overall performance of the EF is typical of current Macro lenses, we may state infer that the FD lens was very good, to say the least, in the FD era. A comparison with the Macro-Elmarit-R 2.8/60mm, the Micro-Nikkor 3.5/55mm, the Zeiss S-Planar 2.8/60mm do not show major differences in performance between the several designs. That is to be expected, as most follow the same design outline.
Analysing the close-up performance (1:5 to 1:10 magnification) proves the claim that these macro designs are optimized for the near focus region. To be more accurate it should be noted that the performance at close range is not as good as that at infinity (contrast drops a bit), but the loss of contrast and definition of fine detail is much less than what we get when using a standard 50mm lens at closer distances.
Medium tele lenses
I also have the new Canon EF 1.2/85 L II for test on my cameras (EOS 30 and EOS 5D).
Again the question is where do we see the progress.
The MTF measurements are quite revealing. The new 1.2/85 performs as well as the older FD 1.2/85 SSC. A surprise again: more than 30 years of design separate the old and the new versions and no optical advancement, at least not visible. One major change happened under the bonnet: the old lens used glass with a high lead content. Newer designs have to use lead-free glass and this is in many cases, especially high-speed lenses a problem. Glass with lead content is the friend of the designer as this type has many desirable characteristics. Lead-free glass brings its own range of problems and thus a new design may be as good as the previous one and still represent a good design as performance has not degraded.
Canon 1.2/85 SSC at 1.2; 2/ 2.8/5.6
Canon EF 1.2/85 L II at 1.2; 2; 2.8/5.6
A full test of the 1.2/85will be presented in an upcoming article. Let it be enough for this article to say that the image quality wide open is excellent and as good as one could possibly hope for, given the constraints in size and price. Wide open it is a bit soft, but stopped down to about 5.6 it performs in the same league as the best 85-90mm lenses, including the new Sonnar ZM 85 and the Summicron 90 asph.
Given this good performance compared to the best of the crop in this age, we may wonder how good the 1.2/85 SSC must have been thirty years ago.
Leaving aside many subtle aspects of lens performance, we may say that all very high speed lenses in the 1.4/75mm to 1.4/85mm range are a bit soft wide open and not one of them gets superior quality, much above what the others do offer. A very fine example of this type is the Pentax 1.4/85mm lens: this one really shines! And can be seen as a proof that sometimes a lens in a second-tier lens system can become a winner. On the whole however, the systems offered by Leica, Canon, Zeiss and Nikon are consistently better than the ones from Pentax, Olympus (non digital system!), Minolta or Konica.
From a historical perspective this landscape unfolds. In the sixties and seventies Nikon offered the best optical system from Japan, but since the FD series of lenses and certainly with the EF range Canon now has without any doubt the best lens system, even equalling and sometimes surpassing the Leica R range of lenses. In the age of the great mechanical cameras we must re-arrange the order: Leica had and has the best mechanical mounts, but Canon must be very close to what Leica offers in the realm of image quality. Nikon and Zeiss then land on third place.
If the evolutionary trend lines may be drawn into the future, we should accept the fact that Canon soon will have the best optical system for SLR cameras. But remind yourself of the famous Economist rule: every projection into the future based on current knowledge will be false.
Macro lenses are credited with the ability to resolve much finer detail than the normal 50mm lenses. The MTF results for 40, 80 and 160 lp/mm are quite interesting. At 5.6 the FD lens has on axis (5mm image height) a resolution of 160 lp/mm with a contrast of 20-30%, going to almost zero in the edges. But the 80 lp/mm offer a contrast of above 50%. The EF version is comparable, but the centre disc of high contrast is smaller in diameter than the FD lens. The FD lens then is the better lens, even when we consider the generation difference. The FD was also in its time more expensive than the current EF 2.5/50mm version.
left: FD for 40,80, 160 lp/mm; right: EF for 40, 80, 160 lp/mm
It is interesting to compare this high-resolution performance with the ability to resolve fine detail of some high-speed lenses of 50mm focal length. It is a solid part of the photographer's knowledge that a high-speed lens is not as good as a lower-speed lens when maximum resolution is required. I checked the following lenses: Zeiss Planar 1.4/50mm ZF, Leica Summilux 1.4/50mm ASPH, Cosina Nokton 1.5/50mm, Sonnar 1.5/50mm (historical), Zeiss Planar 2/50 ZM, Leica Summicron 2/50mm. let us look at the maximum resolution at modest apertures, like 5.6.
Below are the results: these are actual measurements, not calculated results at 1:5.6.
|lp/mm||Planar 1.4/50||Planar 2/50||Nokton 1.5/50||Sonnar 1.5/50||Summilux 1.4/50||Summicron 2/50|
At 160 lp/mm these lenses in general resolve very fine detail with a contrast of about 35%. These numbers may be viewed with some suspicion and some curiosity. They are not consistent with the knowledge of the street, where it is commonly known that high resolution and high speed do not go hand in hand. There is even a notorious discussion about the design of some of the older Summicron lenses that should have a special design deliberately leaning to high resolution with a loss of contrast wide open. The most surprising result in my view is the high quality of the older Sonnar 1.5/50mm, not a lens known for its high definition of the fine details.
We should interpret these figures with some background knowledge. First of all, the lenses have been refocused at every aperture and for every aperture the optimum location of the focal plane has been selected. And they are measured at the centre of the image area. The results do represent the best possible result, most certainly not reproducible in normal practice.
Even so, we have to admit, not for the first time, that resolution tests are futile in every respect. Most lenses are able to resolve more lines than the photographer in practice can attain. The many reports on the internet, where persons claim that they have found a resolution of 70 lp/mm with one lens and 56 lp/mm with another lens are not relevant. True resolution is invariably much higher that what can be found. And there is no direct correlation between resolution and image quality. No one would deny that the older Sonnar is not as good as the current Summilux 1.4/50 asph, even when both lenses have the same maximum resolution.
These results also imply that the current hype of magazines and websites to test and compare digital camera-systems based on resolution charts (2000 lp/image height etc) is far removed form the photographic reality (see below!). Even when the results are OK (what can be seriously doubted in a number of cases), the inference from these numbers to real photographic needs is weak to say the least.
The results presented above are the ideal values. In practice there are a number of disturbing circumstances that will lower the figure substantially: the accuracy of the mechanical and optical engineering of the system, the presence of focus differences when stopping down, the care of the photographer, the slightest movement of the camera, the linkage between the camera and the lens and the focus system in the CRF, the exposure, the film plane surface, the location of the image plane, all will degrade the result substantially.
Of more importance are the numbers for the 20 and 40 lp/mm at an aperture of 2, as we will find here the practical limits for film-based photography and 80 lp/mm when we go for solid state capture. Here the figures are widely different and do represent what the photographer can expect. For 20 lp/mm the figures vary from 90% (Summilux asph) to 70% (Planar ZF 1.4) and the rest between these extremes. For 40 lp/mm the range is from 70% (Summilux asph) to 40% (Planar ZM 2/50). For 80 lp/mm the range is from 30% (Summilux asph) to 15% (Planar ZF 1.4). The Sonnar 1.5/50 mm and the Nokton 1.5/50mm are quite surprising as they are often above average in all measurement ranges. To repeat: these figures are for optimized conditions. The focus difference and the mechanical tolerances will degrade the real performance in general photography.
These results also indicate that the study of MTF graphs does not make much sense unless we know the conditions of measurement.
The main issue is the performance at the non-infinity position. Most lenses will degrade more or less substantially when you focus in the close range, from 0.5 meters to 3 meters. This is never taken into account when doing lens tests. As noted for the macro designs, here the degrading of close focusing is commendably small and with the high-speed 50mm lenses the degrading is much worse, with only the lenses with floating elements staying OK at least within reasonable ranges.
Where is the progress nowadays in lens construction?
I also had the chance to study additional measurements of the performance of the Canon EOS EF 24-105mm. The MTF values for all focus positions from 24mm to 105mm are outstandingly good with contrast values for the 40 lp/mm above 60% and even reaching 80% at the 24mm position. Most designs however break down after the image height of 15mm. It seems that designers neglect the outer part of the negative area (when using 24x36mm size capture area) and assume that the smaller sensors of APS-C and H format do not require this performance in the outer zones.
As a matter of interest I can say that at the focal length of 90mm the 24-105 lens needs to be stopped down to f/11 to deliver the same quality as the 1.2/85 has at f/2.8 and at f/4 the 1.2/85 becomes better, especially in the outer zones of the image. Good primes are still better than some modern zoomlenses!
Overall however, the quality of the 24-105 is quite fine and one will not be disappointed using this one. In the past, designers were very careful when creating zoomlenses that the focus shift when you change the focal length would be very small. This phenomenon of focus constancy is to be explained like this: when you focus at the focal length of 24 and shift the focal length to 105, the focus must stay spot on. In reality this is not often the case. The EF 24-105 has a focus shift of more than a full millimetre when changing form 24 to 105 mm. That is an awful amount and a Leica designer would faint if his lens would have this focus shift. At Leica focus shift is measured in fractions of millimetres.
Canon is of course aware of this aspect and they do obviously not care about it because the AF feature of the camera will take care of the focus shift. Here we have an instance where the design of lenses is being adapted to modern demands and conditions: one does not use mechanical lenses with Canon cameras and so design requirements of the apst are not required any more. Indeed one does not see it in normal photography as the AF takes care of the shift, but if you forget to refocus the AF after shifting the focal length of the zoomlens, you will have a problem. Depth of field at the smaller apertures will also compensate a bit, but basically you must live with the phenomenon.
On the other hand we may get a better understanding why Leica lenses are so expensive: the engineering care and assembly adjustments to hold the focus shift in very small dimensions is a costly affair.
Modern lenses for the AF systems and the D-SLR systems will be much different form the classical mechanical designs. Whether we like it or not: the purist approach to optical and mechanical design will become the exception (presumably holding its own only in the CRF scene and the coming D-CRF designs).
Money is always an issue. The EF 24-105 is an excellent lens, if we accept the design aspects mentioned above and in the previous report.
Checking the cheaper EF 24-85mm I find a lens with a generally lower overall contrast and a substantial amount of asymmetry (as I found in the 2.5/50mm). At the 24 and 50mm position the lens has a quite uneven performance at both image halves (left and right). Decentring is a problem of many cheaper lenses and we will see this phenomenon in many instances in practical photography.
Here we encounter another issue of modern lens construction and manufacture: statistical quality control: the Japanese manufacturers adopt the current approach where production aims for a certain quality level and then let quality control go for the Gaussian error distribution, neglecting the extreme 3% or 5%. It is much cheaper to replace a defective unit after a complaint than to strive for a100% quality control at the end of the assembly line. It is an illusion to assume that even Leica can hold such a stringent control system that not one defective unit will happen to pass through the control. There is always a sliding scale of what is just acceptable and just unacceptable. Ad humans are indeed humans and so make errors.
When the extreme criteria of the optical bench or MTF equipment are the norm, many lenses will fail the test. But is this fair? The test results of the magazines and the websites do neglect many important issues and hardly find a lens that is defect or faulty in some respects.
It is not only a philosophical issue to reflect on the question whether the level of quality consciousness has been lowered during the last twenty years. Even a simple $100+ bread and butter lens as the EF 28-105mm or the current so-called kit lenses that are offered for a price below $100 can simply not be good! The design is quite complicated with seven and more lenses, fully glued together in plastic mounts by robots and with a low level of quality control. Still most persons love the results and are quite happy and if that is the case, why should we adopt more stringent criteria. Most often we do not even the difference between a simple SLR lens and a high-grade Canon or Leica lens.
This is a line of thinking for another article.