Diffraction f-stop limits of IQ4150, IQ3100 and H6D-100c backs

[George_Cleansman]

Has a user of these backs any experiences with the maximum f-stop values depending on sensor size and pixel stitch due to diffraction issues? The latest Phase One back has the enormous resolution of 150 MP on the biggest medium format sensor with a pixel stitch of 3,76 µm. The other above 100 MP backs have a pixel stitch of 4,6 µm.

The airy circle radius is calculated by the formula r = 1,22 * Wavelength * f-stop. This was mentioned in an older LL thread by Erik Kaffehr in 2011. For example, the calculated airy circles with an average wavelength of 0,55 µm for the f-stops 5,6 and 6,9 are: 

r = 1,22 * 0,55 µm * 5,6 = 3,76 µm (this is the pixel stitch of the IQ4150)

r = 1,22 * 0,55 µm * 6,9 = 4,6 µm (this is the pixel stitch of the IQ3100 and the H6D-100c)

How must I handle these calculations when I shot these medium format camera backs to avoid diffraction? Do I have to avoid f-stops higher 5,6 (for 150 MP) or 6,9 (for 100 MP) for getting the best image quality? Or are these calculations more theoretically? Can I see diffraction only on my MacBook and not on a larger print?

What are your recommendations, please? 

[Jim Kasson]

Has a user of these backs any experiences with the maximum f-stop values depending on sensor size and pixel stitch due to diffraction issues? The latest Phase One back has the enormous resolution of 150 MP on the biggest medium format sensor with a pixel stitch of 3,76 µm. The other above 100 MP backs have a pixel stitch of 4,6 µm.

The airy circle radius is calculated by the formula r = 1,22 * Wavelength * f-stop. This was mentioned in an older LL thread by Erik Kaffehr in 2011. For example, the calculated airy circles with an average wavelength of 0,55 µm for the f-stops 5,6 and 6,9 are: 

r = 1,22 * 0,55 µm * 5,6 = 3,76 µm (this is the pixel stitch of the IQ4150)

r = 1,22 * 0,55 µm * 6,9 = 4,6 µm (this is the pixel stitch of the IQ3100 and the H6D-100c)

How must I handle these calculations when I shot these medium format camera backs to avoid diffraction? Do I have to avoid f-stops higher 5,6 (for 150 MP) or 6,9 (for 100 MP) for getting the best image quality? Or are these calculations more theoretically? Can I see diffraction only on my MacBook and not on a larger print?

What are your recommendations, please? 

I would recommend using not the distance between the first zeros of the Airy solid, but rather the diameter that has an encircled energy of 70%.

https://blog.kasson.com/the-last-word/blur-circle-size-estimation-with-encircled-energy/

Here are some visuals of diffraction at various f-stops with a 3.76 um sensor:

https://blog.kasson.com/gfx-100/a-visual-look-at-gfx-100-diffraction-blur/

Jim

[Doug Peterson]

I strongly suggest against a purely mathematical approach to this question. If you own the camera, simply shoot relevant subject matter in a typical way and do a bracket of 1/3rd stops between f/5.6 and f/16, then use the images as you intend to (web? small prints? large prints?) handling each raw file to its individual best (e.g. sharpening, noise reduction, and especially diffraction correction, based on making the best final image out of each individual different-aperture raw file) and see where you find the loss of sharpness offensive. If you don't own the camera, then make sure you're buying from a dealer that does this kind of testing on your behalf as part of your presales research (we've shot hundreds of gigs of such testing over the years that I've been head of R+D at DT).

You'll usually be quite surprised compared to the pure-match approach.

[BAB]

Unfortunately due to variables in lens samples, sharpening methods, sensor flatness, light direction, raw processing techniques, vibrations, and camera ISO I don’t believe in anyone’s testing results as absolutes. Whether handheld or on legs the best is only a indication of average results. This to me is satisfactory though as one would only have perfect results in a small sampling of many captured images. Then from capture thru print with.out having a written checklist of the required steps to be preformed in the correct order with each software and applying that differently to different images. Even if all is written down I personally doubt if I could manage a perfect 100 score on every attempt. I have learned that sharp is relative with landscapes, product or portrait images extreme sharpness is a figment of ones imagination. There is no perfect in photography!

[EricWHiss]

then make sure you're buying from a dealer that does this kind of testing on your behalf as part of your presales research

Doug, I assume you've done the testing. Since you never miss a chance to sell Phase in your posts (I know it's your job so don't take my comment as a slight), shall I assume your results weren't enough to make one buy?   What aperture did the real resolution drop at?   f/8 before and f/11 after diffraction correction?  or was it bigger? 

[Doug Peterson]

Doug, I assume you've done the testing. Since you never miss a chance to sell Phase in your posts (I know it's your job so don't take my comment as a slight), shall I assume your results weren't enough to make one buy?   What aperture did the real resolution drop at?   f/8 before and f/11 after diffraction correction?  or was it bigger?

A few dozen times over the last ten years yes.

You seem to have missed the point of my post.

The results vary based on subject matter, method of capture (handheld? tripod? MUP?), lens, shutter speed, print size, sharpening method, and most importantly: the user. One persons's "unacceptable loss of sharpness" is another person's "you're kidding! that's barely noticeable".

The point being there isn't one answer, because the (relevant to the end user) question isn't "at which point does the overlap of airy disks encircle 70% of the energy*" but rather "when does the choice of a higher aperture unacceptably affect my final result" with the words "my" and "final result" being especially important.

We are glad to work with any photographer to provide any of our executed aperture-ramps specific to their genre, lens choices, shooting conditions, etc. If we don't have them in our archive of previous tests we're glad to help them generate such ramps specific to them.

If you really want a specific single number I can tell you that Doug-the-photographer will be using the IQ4 150mp most often around f/14 when capturing work for my series Terra Nudum. If I were shooting different subject matter, or had needs/preferences different than my current needs/preferences, or placed a different priority on micro-detail versus focus-depth, or when the shot is more conducive to focus stacking or Scheimpflug, I might make very different decisions regarding aperture.

*This is not meant to be derogatory toward Jim whose work I respect a great deal. The pure-math method is the best option if you don't have access to the camera in question (because it's the only option) and can also serve as a good sanity check of the practical results. But if you have the camera, taking pictures takes into account the entire image chain, not just the physics of the aperture hole and the size of the pixels.

[bjanes]

If you really want a specific single number I can tell you that Doug-the-photographer will be using the IQ4 150mp most often around f/14 when capturing work for my series Terra Nudum. If I were shooting different subject matter, or had needs/preferences different than my current needs/preferences, or placed a different priority on micro-detail versus focus-depth, or when the shot is more conducive to focus stacking or Scheimpflug, I might make very different decisions regarding aperture.

It is very nice to have 150 mp, but an important question is how many mp one can capture under field conditions. Table 3 of the Osuna and Garcia article on LuLa shows the effective mp for various sensor formats and apertures. For a medium format sensor of 36 x 48 mm the effective mp is 31 mp at f/11 and 14 mp at f/16. This takes into account diffraction blur, but with a 3 dimensional landscape defocus blur will also be present in many areas of the image.

The IQ4 sensor at 40 x 53.3 mm is slightly larger than 36 x 48 but 31 mp at f/14 would be within the ballpark range for the IQ4. My question is would the Fuji GFX 100 body at US$ 10k be more cost effective than the $51k IQ4?

Regards,

Bill

[Christopher]

Sorry but these numbers are BS. My IQ4150 outperforms the IQ3100 and IQ180 and p65 @ f11 without any discussion. Even @f16 or 22.... diffraction isn’t a hard number...

[Doug Peterson]

I agree that those numbers do not match up with my experience, nor the experience of our clients.

[bjanes]

I agree that those numbers do not match up with my experience, nor the experience of our clients.

Exaggerated claims are not new to this forum. Remember when the P1 CCD  backs had 10 stops more DR than 135 full frame sensors or the better color of CCD vs CMOS (when CMOS was not available in MF)?

As Jim Kasson points out here, with modern lenses and sensors, landscape photographic sharpness is often limited by diffraction and defocus blur rather than lens quality or sensor resolution. Where is the sweet spot?

Bill

[Jim Kasson]

Exaggerated claims are not new to this forum. Remember when the P1 CCD  backs had 10 stops more DR than 135 full frame sensors or the better color of CCD vs CMOS (when CMOS was not available in MF)?

As Jim Kasson points out here, with modern lenses and sensors, landscape photographic sharpness is often limited by diffraction and defocus blur rather than lens quality or sensor resolution. Where is the sweet spot?

Here is one approach for balancing diffraction, aliasing, and sensor aperture blur:

https://blog.kasson.com/the-last-word/whats-your-q/

https://blog.kasson.com/the-last-word/interpreting-q-in-the-real-world/

https://blog.kasson.com/the-last-word/example-q-calculations-for-bayer-cfas/

This line of argument means we want really fine-pitch sensors.

Jim

[EricWHiss]

This line of argument means we want really fine-pitch sensors.
Jim

Certainly interesting, thanks Jim.  I would think if we had sensors with so fine a pitch as to be able to characterize the diffraction cone then it would be much easier to correct it? 

But of course with so small a pixel then we'd have less signal and then less SNR ...

[Bart_van_der_Wolf]

Certainly interesting, thanks Jim.  I would think if we had sensors with so fine a pitch as to be able to characterize the diffraction cone then it would be much easier to correct it?

Yes, that's correct.

But of course with so small a pixel then we'd have less signal and then less SNR ...

Yes, dynamic range will suffer, but we can then use computational photography which can e.g. combine multiple exposures so that we can achieve both goals. Of course, subject motion may present some challenges. Artificial Intelligence and machine learning can also play a role in boosting or even replacing detail with higher quality detail.

Cheers,
Bart

[Jim Kasson]

Certainly interesting, thanks Jim.  I would think if we had sensors with so fine a pitch as to be able to characterize the diffraction cone then it would be much easier to correct it? 

But of course with so small a pixel then we'd have less signal and then less SNR ...

Both of those are true at the pixel level, but I have not found a strong correlation between pixel pitch and SNR per unit area.

[George_Cleansman]

Thank you very much for your recommendations about diffraction aspects. Some of the posts point out, that not only mathematical approach is sufficient, because there are many other reasons (defocusing, shutter vibration, etc.) for decrease of image sharpness. And sometimes higher f-stops might not influence the image quality very badly. Additional, few post processing software programs are able to reduce diffraction.

That is surely right, but when I’m in the field with my camera I like to know which maximum f-stop I generally have to respect to be on the safe side. So, for myself I like to summarize the information’s for my personally better understanding and for that I like to follow only the mathematical approach. My camera is a Hasselblad H6D-100c with a pixel stitch of 4,6 µm.

I noticed that diffraction first becomes visible for 100% crop if the relation between Airy disk and pixel stitch is not balanced. The Airy disk A is calculated by the formula

A = 2,44 * lambda * N

N is the lens f-stop and lambda the average light wavelength (yellow-green lambda = 0,55 µm). The diffraction calculator e.g. of the website photopills (https://www.photopills.com/calculators/diffraction) points out, that diffraction becomes visible, when the product of 2,5 and sensor stitch p is larger than the Airy disk

2,5 * p > A = 2,44 * lambda * N 

Due to this relation the maximal N (f-stop) for a lens before diffraction may become visible must be approximately (2,44 ~ 2,5)

N < p / lambda

Bill is calculating for the diffraction limit with the factor 2,0 instead of 2,5 for the maximum N (f-stop), which leads to smaller N (f-stop) values

N < 2/2,44 * p / lambda = 0,82 * p / lambda

Jim points out in his blog about Q (https://blog.kasson.com/the-last-word/whats-your-q/) that the factor Q relating diffraction is defined as

Q = 2 * Fcsensor / Fclens = N * lambda / p   =>   N = Q * p / lambda

For the maximum N (f-stop) this leads with Q = 1 to the same relation as above mentioned

N = p / lambda
 
For a poor amateur Hasselblad photographer with a 100 MP H6D camera (pixel stitch 4,6 µm) this means that the maximal N (f-stop) due to diffraction is

N = 4,6 µm / 0,55 µm = 8,4

The Airy disk A is

A = 2,44 * lambda * N = 2,44 * p = 2,44 * 4,6 µm = 11,2 µm

So, I am on the safe side with my H6D camera with the f-stop of N = 8. For the higher sophisticated Phase One 150 MP XF camera with a smaller pixel stitch of 3,76 µm the maximal N (f-stop) is corresponding to the above formula

N = 3,76 µm / 0,55 µm = 6,8

How in reality diffraction is visible when we use higher f-stops than the maximal calculated N can be discussed as mentioned before. This is only the mathematical approach. For prints the circle of confusion (CoC) with higher possible f-stops may be more important.

But generally, more and more pixels on a same sensor area size leads to smaller pixel stitches and lower maximum N (f-stop). This means at my point of view that newer lenses must deliver the best performance at smaller N values (f-stop). Due to smaller N (f-stop) the Depth of Field (DoF) is also decreasing.

On the other hand, the sensor quality increased in the last years (e.g. BSI sensors), so the image quality today is excellent. Also, we have with higher amounts of sensor pixels a better oversampling, that results in reduction of artefacts and in that way in a better image quality.

But what is the limit for the amount of sensor pixels on the big 54x40 full medium format sensor area in the future? Do we need more and more pixels or do we need pixels with more or other features?

Thanks for any response, George

[JaapD]

But what is the limit for the amount of sensor pixels on the big 54x40 full medium format sensor area in the future? Do we need more and more pixels or do we need pixels with more or other features?

Good question. For me the answer will be: a bayer-less sensor, a ‘real’ color sensitive sensor with no interpolation anymore trying to figure out what the value between the samples should be. But no Foveon. I see the Organic sensor as the next major step in image sensors.

Regards,
Jaap.

[BJL]

Both of those are true at the pixel level, but I have not found a strong correlation between pixel pitch and SNR per unit area.

Thanks Jim; and it also seems that with new technologies like BSI,  a sensor having higher resolution(*) does not suffer much or any loss of "well capacity per square mm". If so, there is very little downside to "excess resolution" a.k.a. "over-sampling" except when sensor read-out speed is an issue.

So for all but high frame-rate photography (including video) it seems best to push towards sensors that out-resolving all lenses (both current and future ones likely within the lifespan of the sensor), so as to get the best possible out of any given lens, and then choose aperture based on the quality of the image delivered by the lens. Trade-offs of DOF vs diffraction vs usable shutter speed etc. can all be decided "at the lens", not distracted by the red-herring of avoiding any perceptible diffraction softening. Which in most situations is only slightly less silly than the opposing goal of avoiding any perceptible OOF effects; the pin-hole camera extreme.

(*) Meaning smaller photo-sites — I prefer the positive phrasing about resolution, and the argument would be the same with film of increasing resolution showing diffraction earlier.

[Jim Kasson]

Thanks Jim; and it also seems that with new technologies like BSI,  a sensor having higher resolution(*) does not suffer much or any loss of "well capacity per square mm". If so, there is very little downside to "excess resolution" a.k.a. "over-sampling" except when sensor read-out speed is an issue.

Yes. What you wrote is the reason that what I wrote is true.

[Bart_van_der_Wolf]

Both of those are true at the pixel level, but I have not found a strong correlation between pixel pitch and SNR per unit area.

"Per unit area" being the 'issue', especially for those who do large format output. The "per pixel" SNR will degrade as the photosites become more densely placed. The ability to capture converted electrons is rather well related to photosite surface area (both from the point of Poisson statistics as well as physical doped silicon properties). This will affect the smoothness of gradients, and the robustness of files against post-processing.

However, as diffraction reduces resolution (or rather the 'per pixel' MTF), the relevance of 'per area' SNR increases, especially when downsampling.

Cheers,
Bart

[Shrev94412]

I hope I do not regret posting this but, here goes. I owned the IQ180 and now own the IQ3100. I am not a mathematician so that conversation is out for me. I have found that shooting both these back between f/5.6 and f/11 yields the optimal results with no defraction (My personal f stop “safe zone”). That is based on the IQ180 with an Alpa Body and Alpa Rodenstock glass and the IQ3100 with both the XF and Alpa body and both Alpa Rodenstock Glass and Phase One Schneider glass. Further, I believe the Alpa Rodenstock glass to be the sharpest of ALL glass I have ever used and the Phase One Schneider glass to be a VERY close 2nd.

As to the FUJI GFX100 (someone above mentioned FUJI), I own that one too, and just took it to France and shot over 5,000 clicks with it. The files are extremely good but, NOT as overwhelmingly impressive as the Phase One Files. Every time I open a Phase One file shot below its ISO tipping point I am amazed at the quality. IMO nothing beats it.

BUT......Now is my non-post related rant......Phase needs to build in a muti point (100 points minimum) focus system with a very accurate phase detect focus system with usable continuous focus tracking and higher usable ISO before it can be a single camera solution. A single point auto focus system is not usable for action subjects. They need to get away from “studio” mentality and embrace a broader range of photography genres. Hence, why I took Fuji and Nikon to France for shooting a Sailing  Regatta and the white horses at Camargue.

Is anybody listening? So the answer is stay between 5.6 and 11 and use focus stacking if you need more DOF.  If you think you need to open up more or stop down more you need to go out into the field with YOUR camera and lenses and shoot a full aperture sweep and see where the defraction begins and ends. See....no math needed....just go out to the field and shoot. Besides, when you spend $10,000 to go to that awesome location to shoot are you going to rely on math or known data based on YOUR gear??

Just my opinion.