f/0.38 camera lens made with oil immersion microscope objective
Вставка
- Опубліковано 7 лют 2025
- I removed the protective glass from a CMOS image sensor, and used optical immersion oil to couple the bare image sensor to a 40X NA=1.3 microscope objective.
f/0.7 Zeiss lens described in amazing detail: www.marcocavina...
Fake f/0.33 Zeiss lens: petapixel.com/...
Camera and sensor that I used: www.ids-imagin...
www.onsemi.com...
Conversion from f/# to NA: www.eckop.com/...
More discussion about the f/#: wp.optics.ariz...
Support Applied Science on Patreon: / appliedscience
I love how that this isn’t a channel for attention, it’s a channel for the science.
Thank you Ben, been watching for years. Also thanks for helping out the other tubers.
+1!
Atttention to science!
And no 'sponsored cr-ad's :ø)
I love a channel that is still like old YT instead of influencer nonsense
Well said. The hype channels are vast; every half baked joke startup Theranos wanna be has several such channels out there hawking for them, and hundreds of science “communicators” are out there keeping the tax $ flowing to weak or hyped or rehashed science - engineering. All them encourage the idea that science is simple or simplistic, or can be a secular religion for the lonely.
I don't remember HOW I came across this site years back, but I'm so glad I did, it's the most wonderful oasis of experimentation, testing, education and skill.
Don't change Ben, you inspire me constantly and I know you inspire many others :)
All the best from the UK
i dont remember either, but im pretty sure it was something related to the DIY electron microscope stuff. dude post videos that sound like clickbait but are literally just what the title says.
I swear I remember seeing Ben's blog posts featured on Hack-a-Day about 15 + years ago. One of the great at-home hackers and a true engineer
The iPhone MEMS oscillator video was one of the first for me, been hooked since!
I agree. And now i feel bad about how I reacted to Ben’s Olestra video but that’s because I did experience leakage. You see, people who have weight problems don’t eat a normal amount of chips. At the time i didnt have a weight problem I had an over eating problem which eventually caught up to me in my late 30s. TMI, anyone?
Amazing stuff as always! Really neat demo, and appreciate delving into the theory. I've definitely scratched my head over NAs and f/numbers before, but just shrugged and ignored it. Cool to see the explanation!
This made me realize there is a lot more going on in lens design, like I never thought about the need to account for the air gaps between each element inside of a lens(seems obvious now).
Also, being able to remove and replace delicate surface mount components sometimes feels like a super power.
Thats why modern coating made such a big difference in lens designs. I once had a Zeiss Planar made in 1896 from the year it was released, they were not popular then because without optical coating it just had quite poor contrast as so much of the light ended up bouncing around because the different refractive index of the glass & air.
As someone who’s recently spent a long time looking at commercially available fast lenses because my high speed camera iso only goes to iso 400, this is awesome (if far from the application I want 😂)
I need to rewatch later to appreciate the math, but the concept of immersing the sensor to allow for high convergence angles is fascinating! I’m wondering if you could cheat optics even harder by designing a nonplanar detector. Another part of me is wondering if to make that work you’d end up with a separate optic per pixel and if the bees have already optimized that 😬
If you have a spherical sensor then a lot of aberrations go away. This was actually done for Schmidt cameras (telescopes) by stretching film over a spherical form.
For your high speed
Take a look at X-Ray lenses btw, they can be had for only fractional kidneys and if your sensor is close enough to your mount you can get like F 0.75 there's a list of them on wiki and some forums dedicated to making useful lenses out of them.
"Medium division has a video"
f0.7 - Ultrafast Lenses - Legends, budget options, modding, and testing - Epic Episode #13 where they "adapt" their cine camera to use the ziess 0.7 xray lens.
Mamiya 645 medium format and speed boosters/focal reducers might be another possibly affordable option but you're probably looking at F1.5 and a muddy image depending on the sensor size you have. But I have seen people adapt them to MFT with success so possibly? The speed boosters for them on GFX sized cameras are like $500 which takes some of the shine out of the $100 lenses
Love your work in general btw. I am kinda hoping you're doing that rolling shutter combined with high speed camera to get that light in motion even slower thing with this ;-) But I have no doubt whatever you're doing it'll be cool
Nice experiment and video Ben, but I don't think that your conclusions are correct. First, fast lenses have a very short focal depth, which we don't observe in your images (quite the opposite actually). Second, if you use a microscope objective in reverse, aperture is not defined by the N.A. of the objective but more by the "physical aperture" at the other side of the objective. This aperture does in general not accept light under very wide angles and this makes the effective aperture of the objective in this configuration much lower (and for example dependent on the physical length of the objective). Which in turn has implications for the effective F-number of your optical system.
Also the gain applied for the same ISO by different sensors is quite different. Basically manufacturers make ISO numbers up, there is no correlation between different sensors (even from the same manufacturer). The specific sensitivity and gain of each of two sensors has to be calibrated to make a meaningful comparison.
The depth of field is not small, as his focal length is very short. He has a 4mm lens there as he said, that's a 4/0.38 =10mm aperture with f/0.38. So the depth of field should be about the same as a f/5 with a 50mm lens (50/5 = 10), which fits well with the observed images. Depth of field at a specific distance is basically only dependent on aperture size afaik.
It's the same reason why phone cameras have a large depth of field and need to add bokeh artificially even though they have quite small f-numbers.
@@moe.m Yes, depth of field is a property of the physical aperture size and the magnification factor.
The focal length doesn't matter at all, it's all in the ratio of aperture size and magnification factor.
People think that focal length matters because they are concerned with DoF/FoV equivalents between sensor formats, but in that case you are changing the magnification factor, and it all cancels out nicely. But if you need to think about DoF in any other context it's not that useful to do these sort of calculations, it's more useful to think in terms of aperture size and magnification factor.
But it gets even simpler, if your final display size is fixed, and you only care about the circle of confusion at that specific display size, then for a given FoV all that matters is the physical aperture size. The sensor size doesn't matter at all, it all cancels out.
So yes, a 10mm aperture for a wide angle lens is not very wide at all, quite the opposite (I estimate the FoV to be eq. to around 28mm lens on FF).
@@pdp11 Think about it from a geometric standpoint. The DoF depends on the maximum angle of the light rays from a point at the focal plane which can still hit the sensor.
This angle defines how far apart two points behind or in front of the focal plane are which can still hit the same point on the sensor with their light. So, if you imagine this in reverse, this defines how fast the "viewing rays" for each pixel spread apart around the focal plane. And this angle is only dependent on the Aperture.
Obviously his magnification factor is also very large, otherwise a 4mm lens would have an insane FoV.
@@moe.m You are absolutely right, thanks for correcting me. I wasn't really paying attention to the very short focal length.
As an optical engineer and microscopist I need to clarify a few things:
1. NA=nsin(theta), f/#=f/D thoes two are the original definition, and only at small angle NA=1/(2f/#)
2. Your data curve fitting is a bit misleading, as the red/yellow/green objectives are oil objectives with n>1, you should have another blue curve with the oil index
3. The aberration of the microscope objective you showed (used in reverse) is totally normal, as it's designed to only image a small sample area of less than 0.5mm*0.5mm (when used normally). In your case, your camera chip size is way bigger than the designed imaging area. So only the center portion of your image will be aberration-free.
Hope this helps.
1. No it is not. The issue is how you draw the focal length. If a system obeys the Abbe's sine condition then focal length is not an axial distance, but as Ben has drawn, a hypotenuse that reaches to the boundary of the spherical wavefront. Your small angle approximation is for paraxial systems. In a perfectly corrected system, free of aberrations the wavefront emerging on the image side is spherical and all the rays have to have the same length (zero optical path difference). You can only satisfy this with a spherical principle plane just as he has drawn.
Edit: In the hindsight I have phrased it badly. The hypotenuse in this case is just a marginal ray so it is a ray that defines the solid angle that in turn defines the amount of light reaching a single image point (it is a part of the Etendue or the A-Omega product) and in a well corrected system it is equal to the focal length.
@@kimjong-du3180 I agree with your objection, but just wanted to state that the zero optical path difference condition can be met on-axis without meeting "Abbe's sine condition". In fact the zero-path condition is met on-axis as long as all the rays reach the same point at the image, and in this case you have no aberrations on axis. However, an on-axis ray hitting the pupil at height y (from the axis) must have an angle theta=arcsin(y/f) at the image side in order for the off-axis fields to be without aberrations, which is another condition the should be combined with the optical path condition.
Abbe's sine condition is a condition on the rays of the on-axis field point, but it affects the aberrations of the off-axis fields.
@@kickass26306 You have just restated the Abbe's sine condition. In a well corrected system the rays will reach the same point in the image if the OPD is zero. One requirement is based on geometrical optics and the other one is physical optics (or scalar wave optics) perspective. It's the same thing but from a different perspective. Reaching the same point means the rays as emerging from the exit pupil have all the same length (i.e. 0 OPD) and this means the wavefront is a perfect sphere and this means that rays as wavefront normals intersect in the center of that sphere (converge to a common point), because definition of a sphere is that all points on a sphere are equidistant from it's center. How can this be met if the sine condition is not met?
"However, an on-axis ray hitting the pupil at height y (from the axis) must have an angle theta=arcsin(y/f)..." that's just image-side half of the Abbe's sine condition.
Ideally zero OPD condition should be met for all the points in the field of view and the Abbe's sine condition should be met everywhere where we want image to be sharp.
In general I think it all boils down to the Eikonal equation.
@@kimjong-du3180I'll try to give an example: A single thin element can concentrate all parallel rays to a single point on-axis with F# smaller than 0.5 (imagine a very high refractive index as the material of the singlet). When this happens, the phases of all rays is the same when they reach the image plane (so equal phase condition is satisfied for the on-axis rays). However, since the F# is smaller than 0.5, Abbe's sine condition cannot be met, and therefore you'll have off-axis aberrations (yet perfect on-axis PSF)
@@kickass26306 From what I understand you are talking about a paraxial f-number, in which case F#
Ben, you used the wrong olympus lens. I sold them for 22 years. That lens has a 180mm focal length to the eyepiece and this is why your image is crappy. Olympus makes the same newer lens, also oil immersion, with and infinity focus where the microscope body had a collection lens below the head set. This new version will vastly improve the image quality and also comes in an APO.
Really interesting and fun!
Regarding removing the glass from the sensor, perhaps if you put your business entity's hat on and contact the sensor manufacturer, let them know that you're working on a prototype of some equipment, they can supply you some of their sensors without the glass? When I do something like this, the response from manufacturer's varies widely between getting custom samples for free and silent dismissal. But, it never hurts to ask.
It might be 6am here, but dang, if an AS video drops, it's watching time!
5 a.m. here!
Same!
5h44 here in the Netherlands...
Same ... Poland
I'm late at 07:40, just woke up.
I appreciate you putting the math in there. Even applied scientists can learn from theory.
hi coder0xff, have you become a flat earther yet? If not I suggest watching _Level with Me (2023)_ by Hibbeler Productions
Man it's always interesting. I'll be starting to pursue an engineering degree at age 36 this year with the last ten years being a machinist, and I credit you and others like you for sparking my interest and thirst for knowledge over the years.
Thanks for the steady stream of A+ content.
Go for it! I followed a similar career path and have recently retired.
the principal plane being not flat makes so much sense! I hope this knowledge opens up new avenues of exploration.
I think Apple, Zeiss or someone has a patent for a curved CCD detector.
@@KallePihlajasaari
Everyone and their dog have one
Tho curved sensors are pretty much limited to one off scientific applications as they effectively have to be made for specific optical design
SONY at some point apparently tried using them commercially, although in low end weird selfie camera.
@@CzornyLisek Operating telescopes worn by surgeons (or your dentist) are designed to function with a curved sensor; the retina on the back of your eyeBALL. :)
Silence then out of the blue - a banger as usual. One of the more unique subscriptions I maintain.
Yes, you are right about the aplanatic system. I figured it out doing radiometry calculations for a lens. The explanations in some books are that the principle planes are curved, but I don't think that's exactly true. An ideal optical system produces spherical wavefront, that converges on the image point. On the opposite side, every point of the object place is emitting light in a form of a spherical wave, that's where the sphere comes from. That's why when doing integrals for radiometric calculations you have to consider that the focal length is a hypotenuse. Rays are just local normals of the wavefront.
Edit: The hypotenuse in this case is just a marginal ray so it is a ray that defines the solid angle that in turn defines the amount of light reaching a single image point (it is a part of the Etendue or the A-Omega product). Both numerical aperture and the f-number are just different ways of defining the light-gathering solid angle.
That last sentence is the most mind-blown I've been in a while. Thanks.
re 5:26 depth of field
the lens also has significant field curvature, which will result in a greater apparent depth of field at the cost of the center or periphery being more out-of-focus at any given time.
Yeah I could tell right away the field curvature on that lens is wild.
@@seth094978 The lens is not plan-corrected (there will be a marking for that) and given how this lens looks like a generic Chinese objective it is possible it isn't even infinity-corrected either (there is also a marking for that). I have one at home that is marked for a tube length of 170mm, though the very expensive ones in the lab I work in has infinity and plan correction.
Yes, as somebody who adapts weird (and plainly "wrong") lenses and lens like objects for cameras, it was immediately obvious that the center was not focused. There was a ring of focused light around the halfway point between the center and edge of the image. Many old movie projection lenses are like this if you try to use the entire image circle. That's how they made crazy 120mm f/1.8 lenses and other beasts back in the day - make the image circle so huge that even your aplanat lens has enough "sort of focused at the same time" area to cover a 35mm movie frame.
I understood nothing about the math, but I know my way around a bad lens :D
The world of optics is all too ironically opaque. Thanks for shedding some light on this. I appreciated the math.
Excellent video as always! Every camera lens has spherical aberration - even good brand comercial camera lens. BTW, the story behind Kubrick's cameras used to film Barry Lyndon is quite interesting: he bought rear projection projectors from the studio - projectors that at the time were not being used anymore because of "green" screen technique. After they sold him 2 or 3 projectors someone called the studio asking for them to save them as historic machines because they were the pinnacle of projector craftsmanship! Kubrick then asked a technician who worked with him to transform the projectors in cameras (an irreversible alteration, btw). Then he used the Carl Zeiss Planar 50mm f0.7 that were manufactured originally for the Apolo Program.
A curiosity: SMSL's EUV machines use mirrors to project the mask over the silicon wafer and there's only one company in the whole world that managed to manufacture them: The Carl Zeiss Company. Yep - they don't make only wet lens wipes!
A note about focal length, aperture, and depth of field: it's all about ratios! A 100mm f2 lens will have the same depth of field as a 50mm f1. Very handy when comparing lenses.
I got a cheap f0.95 35mm lens (7artisans) and even though it's not very sharp, it's so fun to use!
The only channel that I have a notification for
He posts and I excuse myself from dinner.
@@BeholdTheCraqueNetwork I have several... but this is the only one that never gets swiped away.
The low f via oil is an interesting pursuit. I’ve done a little surface mount IC work in the day and, well, the skill required to do what was done here removing the IC and reattaching, plus the glass removal is so substantial it dwarfs every other effort in this project. Precise heat for precise time particular to the IC size and type, solvents, optics, tools, environment. Yes people get trained to do this but it’s a LoT of training.
You really make optics an interesting matter. Your investigation of the formula for aperture, fascinating. Kudos!
Wow, look how similar f/0.38 image looks to what Riddick in Chronicles of Riddick sees with his night vision eyes.
I was just about to comment about it but here you are :) real life riddick vision
That's it
I thought I had seen something similar before
Ben i love that you take the time to puzzle through all of the supposed math to find the formula that fits the data rather than the other way around.
Your awesome.. i worked in an optical lab for 20 yrs and 10 of them maintaining the anti reflective and dip coating equipment . Then i come to your channel and learn more about optics in 20 minutes on your channel then i did 20 yrs in the industry. Given the info wasnt really required. But i like to know why and lots of things where never explained. And then here you are breaking it down and building the equipment in your garage . Thank you so much for sharing your knowledge
I wish i had the ability to give you all the money you need to explore any, and every, idea that interests you. You approach topics at a higher level than most, while also effectively communicating the topic to those of us with the base knowledge. Love your videos!
Where's @HuygensOptics when you need him
I was thinking the exact same thing. HuygensOptics has some very amazing and well explained experiments.
+1
LOL! I was totally thinking the same! He's (one of) the best with light IMHO
I came here to say this. It's almost like we all have the same UA-cam subscriptions. 😅
He actually already commented before you guys did.
Another fascinating video Ben! Thanks for being my favorite science educator here. Your work is very inspiring to a new generation of curious minds!
Vivitar made a retail "professional" 135mm f/1.5 lens back in the late 60's and early 70's. I guess it is somewhat of a rare lens that was marketed by Ponder & Best (?). A friend of mine has one with a Nikon "T" mount adapter on it and it is quite a hefty lens at over two kilos (about 4.5lbs)!
Fascinating to learn about the slight curve at the very start of it all throwing off your calculations. But once you figured out that tidbit of info it clicked and things finally made sense. I appreciate your determination over the months to keep digging and get to the bottom of the mystery!
Recently started working at a IOL manufacturer. It’s been a lot of fun learning about optics. This video came at a great time!
this is, and you are, still the best channel on youtube. by far.
now we can just wait for other 'science' channels to copy what they just saw for next couple of months
I love the stuff you do with optics! It's always a good day when I see a new video from you. Thank you for making such awesome, interesting, & intellectual videos for all of us, the internet strangers {the public} & free of charge(!), other than having internet access. I think a patreon sub to you is well deserved by now, considering i've rewatched every video you've posted multiple times. Hope you have a great day Ben! Your constant curiosity and technical knowhow is inspiring.
You are an entertainment even when I cannot follow the math or chemistry. Been enjoying you for years.
Not an optical engineer here, but I've worked in testing systems, and along people that are/were prominent optical designers for Zeiss and Hasselblad. I think I have a reasonable understanding of this... :)
You are basically correct about the aperture limits. Theoretical limits for a lens that interfaces with air is sqrt(2) over 4 = about f/0.354. Lower than this is physically impossible, no matter what the Ri of the last medium before the image is - since it would require the system to curve light back to the front surface from behind the image surface...
And yes, that's the limitation of the exit aperture as you concluded. The exit aperture is what each point on the image (projection surface) sees, and for image formation purposes it HAS to be an exact sphere (+/- a few wavelengths of the light you're capturing) if you want a low aberration image formation. A flat plane aperture only works (poorly, if you add in efficiency) for a collimation system that only works in ONE point, the axial center point.
IMO the easiest way to start understanding this is not through geometrical optics (which are at the very, very best a rough estimation unless you're working with low f/# low aspect ratio collimation systems). Geometrical optics are basically "lies for children" if that adverb exists in english. Something that's mostly correct for most "simple" applications of the theory.
To get a deeper understanding of image formation optics, the best places to start would IMO be either an astronomy forum like www.telescope-optics.net, or by starting to look into what the "Strehl ratio" actually means for an optical system. Either of these will get you very far :)
On your discussion of F number you have hit the nail on the head.
Numerical aperture is defined as 1/2F, because the approximation works.
The old saw of at low x, sin(x) = x is still being used everywhere.
So for professional lenses they keep using the same old formula even though it's wrong.
I love someone who digs into how things work, excellent work as always.
This is o3-mini-high answer. And I love your videos.
'microscope objectives are designed so that their effective f-number is approximately 1/2NA in air. That is why your measurements “follow a factor of 2” even though the simple small‑angle approximation doesn’t exactly apply'
Here’s why you still see a factor of 2 in practice:
1. **Design Convention (f-number):** In many imaging systems (including microscopy), the f-number is defined as
\[
\text{f-number} = \frac{f}{D}.
\]
For an aplanatic system in air, it turns out that
\[
\text{f-number} \approx \frac{1}{2\,\mathrm{NA}}.
\]
This immediately implies
\[
D \approx 2 f\,\mathrm{NA}.
\]
So even though the full trigonometric relation for a thin lens is more complicated (especially when the angles are not small), microscope objectives are engineered so that their effective aperture and focal length satisfy this 1/(2NA) rule.
2. **High‐NA Effects:** For high numerical apertures the angle \(\theta\) becomes large and the difference between \(\tan\theta\) (used in the simple derivation) and \(\sin\theta\) (which actually defines NA) is significant. That is, the small‑angle approximation \( \sin\theta \approx \tan\theta \) no longer holds. Yet, in the design of objectives the effective aperture (or “pupil”) is chosen in such a way that the overall system performance is optimized, and one still ends up with the effective relation \( D \approx 2 f\,\mathrm{NA} \).
3. **Effective Aperture vs. Physical Aperture:** The “aperture” in a microscope objective is not always the same as the physical diameter of one of its elements. Instead, it is an effective aperture defined by the optical design (which includes several elements, refraction at multiple surfaces, and field flattening). This effective aperture is what leads to the well‐known relation between f-number and NA, with the factor of 2 built in.
I can feel a lense-grinding machine build coming on! think of the possibilities!
It would be interesting to know why fluid coupling isn't used in commercial applications, since it's higher coefficient of refraction enables much more extreme lenses to work. I guess fluid in a camera system creates many problems, but for applications like extreme low light shots it could be useful.
It is used. I once took apart a scraped CRT rear projection TV, and the tubes were attached to the first lens with some sort of optical jelly. Also, wet scanning is a thing, definitely for stills, and I think even for some telecine machines.
@johnalexander2349 very interesting, although I was thinking more of a portable stills or video camera
Glass elements if they are very very close to each other or touching usually have a layer of "cement" between them so that there is no layer of air between them.
Some of Roland Christen’s telescope designs (Astro-Physics, Starfire et al) were oil-spaced - not between the lens and image plane, but at least between the elements of the triplet objective. I assume it simplified the designs before he got his interferometer, and/or good optimization software, or something. :)
There have been other uses of oil spacing in astronomical optics. The goal here is usually sharpness at wider apertures (and color correction).
@@johnalexander2349 I took apart one of those TVs as well years ago, though it used some flavour of oil rather than a gel. Now after almost a decade I finally know the purpose of it!
We used to work at the same company a couple of years ago. If you're still there good luck, I hear things aren't great.
Great work! I can explain one confusing point: your "f/0.38" number comes from dividing an oil-side focal length by an air-side entrance pupil. Mixing oil and air gives a strange result, because oil immersion lenses have different focal lengths on the air and oil sides, If you use the air-side focal length with the air-side entrance pupil, then your computed f/number will drop below 0.5, consistent with the rest of the discussion about what's needed to get a good image.
FYI- The focal plane of a camera is that symbol of the circle with a line through it, sometimes on the hood, or the top of the camera near a dial. As a photographer, this is amazing information and science. I'm used to the smaller the number (f2.0..) the more costly the lens like a long focal length telephoto 200mm. (most affordable 200mm lenses are usually f4.0 or f5.6)
For imaging purposes, as you mentioned, you need more focal length, which also means larger entrance pupil, while still retaining a decent field of view. Therefore, microscope objectives will not yield good results.
It would be very interesting to test the transmission spheres for interferometers, there are some f/0.56 ones with entrance pupils of 4 inches. Too bad they are so expensive.
Personally, I obtained excellent results with the collimating and imaging lenses from old fluoroscopy machines. Yes, the (in)famous Rayxars and Heligons... but there are also less famous lenses like the Kowa 90mm f/1, that can be mechanically modified to focus at infinity while covering the full frame sensor and with good sharpness almost to the very edge.
This was fascinating and illuminating. I don't understand maths due to discalculia, but I *feel* maths somehow. The theoretical aspect of the video was so interesting and I would happily hear it explained in even greater depth
The raw and unpolished-ness of this content is such a big part of why it's so good. It's like being in Ben's shed with him while he shows you cool shit he's been working on.
Nice video. One thing about the film Barry Lyndon. The candlelight scenes were taken using the special low-f lenses, yes, but the film stock itself was high ISO (that is why those scenes are so grainy) and it was chemically treated during developing to enhance the impression. Also, the prints made from the negatives were overexposed as much as possible.
Squeezing, squeezing that last little bit out of the available technology!
I haven’t seen the movie, but the attempt at applying technology to art, for artistic reasons, is interesting.
Fantasticly interesting video that combines theory and practical implications. 👌
Outstanding video, really cool adaptation-always pushing the envelope over at Applied Science! All that broken glass reminded me of an experimental concave-air underwater lens I made from clear light bulb glass back in 1974.
Very interesting and thought-provoking. Great video as always, Ben!
Dof is calculated through magnification, so a low fl number just means it’s lower mag at larger distances. So with the same mag(same fov on same sensor size) the lower f number should give you the less dof. And the calculation is the square of the mag so the mag is far more important than the aperture
2:20 This isn't just glass, it's an IR filter. It appears opaque to IR but transparent to visible light, to remove thermal artifacts from the image. If you want to turn a camera into an IR camera, place a piece of overexposed color film negative (processed) over the CCD in place of where the glass was; it has the opposite effect.
"It's amazing how seeming simple questions are difficult to answer", this is the top of the most important science rabbit holes in history.
Asianometry told story about high NA process used in photolithography.
Each times when light goes thru two materials with different optical density part of the light is reflected back. Oil immersion reduce the impedance missmatching between cmos sensor and camera lenses. Each times when light pass thru by border between two different materials it can absorb (of reflect) 5% of light. 3 lenses on optical path means 6 times when light change propagation medium and it can delivery only 72 % of initial light.
Super interesting. I feel like everyone on UA-cam who does high speed is looking at that f number like 😮
One way that you can preserve wide angles of view and have shallow depth of field on a small sensor is by using a large format camera and then focusing another lens on the image produced on a frosted glass plate at the focal plane of the large format system. It doesn't give you good light gathering, but it does give a very shallow depth of field.
For example, using the Kodak wide field Ektar 10" f/6.3 on an 8x10 camera, then capturing the image off the back with your 20mm/1.4 lens would give you roughly the equivalent of a 17mm/0.42 in terms of depth of field, as far as the μ4/3 camera is concerned. If you can find something like an f/4 or f/4.5 lens that will cover 8x10 at a similar focal length, then you could be looking at effective f stops of 0.3 or less.
I built a large format camera with a f/0.28 equivalent to full frame aperture for the shallowest of DoF.
Because the 400x400mm projection is so large, it has a crop factor 1/13 compared to 35x24mm film.
Using a Pentacon 420mm f/3.6 episcope lens, it effectively becomes a 32mm f/0.28 lens.
If you want to see an image, look up bramstolk on blue sky.
Great video, btw!
Really glad to hear you intend to scale this up, just gotta get those yields up before moving to micro 4/3 sensors
Probabiliy the best video you have made.
In case you haven't seen it, Media Division on UA-cam built an f0.3 lens (by using a diffuser sheet in the middle as a hack).
They did not, they did something that "looks" like what a f0.3 lens would look like. Which is interesting in its own right, but not an answer to the question being asked here by Ben.
@@gu4xinim It depends if you’re describing it with the optics f0.3 or the photography f0.3, because with the former the technicalities make a difference but with the latter, if it looks 0.3 it _is_ 0.3.
as far as depth of field goes, it looks like f/0.3. but as far as light collection goes, it’s not because the light doesn’t go direct to the sensor. that’s what makes a true ultra fast optic like this so special, it collects stupid amounts of light and shoves it right onto the sensor!
Where did you get this idea from? There is no dual definition of F number, it has a clear definition (focal ratio). Half of the video above is clearing that up. The Media Division video is careful to use the "equivalent" qualifier to make clear they're emulating one aspect of what a f0.3 would look like, depth of field in this case.
Best way to measure effective lens-speed imo is reflectively-to shoot a grey card illuminated at x footcandles, in a known colorspace/gamma. Linearizing the picture should, in theory, allow you to work out the true T-stop of the lens accounting for any light loss.
As a camera nerd, this is great. I've indeed been interested in fast lenses and the physics/math behind them
See everyone comes at the depth of field question from a different angle. You say that the 4mm focal length is responsible for the huge depth of field, but I say it's due to the tiny sensor size.
In the end it's about multiple variables being changed at once, put a 4mm lens on your micro 4/3rds sensor, and you distance to subject shrinks so much that you lose all your depth of field in order to maintain the composition, exactly as you said at the end.
I just find it interesting which of the two variables people chose as dependent.
Then to really through the cat amongst the pigeons, Tony Northrup starts talking about the equivalent ISO of different sensor sizes...
These videos are very educational and fun. Please keep making them.
Dude, I love your channel and all your work! This is amazing! I learn so much from you.
Hi, these immersion microscope lenses are often correct for a certain thickness of cover glass slip. Since you have it optically bounded directly to the sensor, it might actually induce abberation!
I’m here for the custom use of a cordless tool battery. I feel you showed me the way out of the box 🤩
The faster the lens (lower F stop) the shallower the DOF (depth of field), however the aperture blade count has a direct effect on how the DOF resolves an image. In laymen’s terms the blurry area in high contrasting areas will not be round (spherical). They will look flat sides, so the higher count of aperture blades the better, better yet don’t use an aperture and use iso and shutter speed to control exposure.
What a super interesting video. Very convincing argument about the relationship between NA and F#.
OLD 3 TUBE red blue green big screens use cooling oil that couples the image of the tubs to the lens also There may be some use full info in such old big screens
This was absolutely fascinating. I feel like I understand so much more about lenses now!
You amaze me every time. Wish i lived next door and could volunteer in your lab every day.
Very cool! Nice discussion. And the effect of using oil is the reason you can't see a glass sinked down in another bigger glass filled with oil.
Always happy to see a video from Ben!
If I remember correctly, in Barry Lyndon they used special candles that gives twice the bright of normal candles. Thanks for sharing and, Cheers !
Makes sense we dont see oil based lenses in production since it would only add a minor improvement over one with air gaps and probably cost significantly more to make it sealed properly
Probably the best way to get a sharp image from a "low f ratio lens" is to use an array of lenses, either on one big sensor or each with its own sensor, and to stack the resulting images. The downside would be that the bokeh would have gaps in it resulting from the gaps between the lenses.
very cool and now thank you for venting my frustration: Yes, “various different” is redundant. “Various” already implies diversity, so “different” is unnecessary. Use just “various” or “different” depending on the context.
The optical guys I used to work with at Battelle were wizards, using their knowledge of the dark arts to bend light to their nefarious purposes!
One project required a perfectly collimated beam from halogen lamp. This is especially difficult because the lamp has a wide spectral output, so all the individual wavelengths had to be focused together. This needed a multi-element lens assembly where each lens element not only varied in focal length, but to create the optical properties needed, also varied in refractive index and coating. If I remember it was a 7-element lens.
I had another job at a start-up working on a binocular system that needed to have peripheral vision. Each lens had a FOV of 220 degrees - actually greater than 180! It was an amazing 11-element lens assembly.
This is very close to a type of system I've wondered about where the focal plain is *at* the surface of the lens. Rather than use oil, you would glue the sensor to the glass. Of course that would result in a fixed focus system which would likely only be useful for limited case like astronomy (or anything else where only a focus at infinity is of interest).
Very interesting! I can see something like this useful for very high speed cameras where you need to gather as much light as possible.
It's fascinating that there has been hundreds of years of lens development and research, we know that low f stop lenses are possible, we made them, yet they are still not readily available or very usable.
I appreciate the math Ben. I'm going to be doing college classes for my apprenticeship and that includes mathematics and lots of it. Thanks!
Why doesn't this video have more likes? Great channel.
Thanks for this. I'll definitely have to rewatch this but the intuition based approach helps me understand it more
You need to get your hands on a stepper lens, used for making ICs. These monsters are super high NA and corrected for every possible geometric aberration (but not chromatic aberration). You’ll need two or more people to pick one up!
Ben you did it again. Good video. Looking forward to the next one.
You can get the IDS cams on their website without the protective glass.
This is science! Love your work Ben, keep at it!
Sir, you could single-handed restart industrial civilization of something happened. Improved.
Hi Ben,
Couple of thoughts about the video.
As numerical aperture can be defined as NA = n*sin(θ) or NA = n/(2*f/#), from this focal ratio can be defined as f/# = n/(2*NA).
Subsituting NA with n*sin(θ) indecies of refraction cancel out and we are left with f/# = 1/(2*sin(θ)). Thus regardless of the medium between the rear lens element and the focal plane the focal ratio is determined by the sine of the half-angle of the light cone.
For a planar detector the maximum angle that light can physically reach the detector is 180 degrees, and thus the maximum half-angle θ is 90 degrees. As sin of 90 degrees is 1, the maximum f-number for such a system is thus f0.5 and maximum numerical aperture in air NA = 1.
From the definion of numerical aperture we can observe that the only way to increase it above 1 is to replace the medium between the objective and detector with a medium which has a refrective index of more than 1. Commonly used mediums for microscopy are either water (n=1.33) or immersion oil (n=1.51).
But does using an oil immersion objective with a numerical aperture greater than 1 such as the NA 1.30 one you use actually admit more light to the detector?
The half-angle θ can be calculated derived from the numerical aperture definition as θ = arcsin(NA/n). Now plugin in the numbers we get arcsin(1.3/1.51) ≈ 59.42 degrees, with this angle one can calculate the f-number to be around f0.581.
There are microscope objectives that are designed to be operated in air without a coverslip with a numerical aperture of NA 0.95. Such an objective would have a half angle of arcsin(0.95/1) ≈ 71.8 degrees, which correspondes to an f-number of around f0.526.
One case where the oil immersion objective does collect more light is when the detector has a cover glass optically coupled to it, this is effectively the scenario shown in the video at 15:00. Since if one uses an objective designed to be operated in air with an numerical aperture 0.95, the NA is conserved as per Snell's law when transitioning to the new medium resulting in a half-angle of about 40 degrees as opposed to the oil objectives 68 degree half angle if an objective with NA 1.40 is used.
If the high NA oil immersion objective does not collect more light then how does it manage to obtain higher resolution, when it is used to observe samples under a microscope?
The answer lies in the fact that when light travels in a medium with a refractive index of more than one it slows down. Since the frequency of the light does not change its wavelength must reduce and since resolving power is proportional to the wavelength this results in a higher resolution.
The objective being used the Olympus UVFL 40x/1.30 Oil appears not be corrected for field flatness on the edges of the image circle and as isn't an infinity corrected objective as others have also pointed out. This results in an image that times weirdly in focus, with the edges being sharp and the center blurry and vice versa.
A 4mm f0.581 lens would have an entrance pupil diameter of approximately 6,9mm, which would be equal to 20mm f2.9 on a micro four thirds camera or a 40mm f5.8 on full frame sensor or 35mm film. Such lenses would not be considered having shallow depth of field at normal subject distances of 2-3 meters.
It shoud be pointed out you had measured entrance pupil diameter to be 11mm. I don't have no good explanation for this.
Comparing the brightnesses of two images presented, the one from the camera with the microscope objective is certainly brighter I would estimate in the range of 2-3 stops, although this is a bit hard to judge.
There are however a few things that come to mind, which might have an effect, these being:
- The differing transmittance of the lenses
- The differing sensitivities of the sensors at the same ISO, this could be 20% or more e.g. (ISO 100 of one sensor actually closer to ISO 120 and anothers to ISO 80). This can be avoided by (ideally) using the same sensor, but other ways exist as well
- Pixel shading at high admittance angles/focal ratios, where some of the light rays miss the light sensitive part and hit other features on the detector, thus not contributing to the amount of collected light
- Optical density of the infrared-cut filter above 700nm. As some other also pointed out the image looks like it has some ir contamination, which would increase image brightness. Cameras used for photo- and videography feature a hot mirror filter which increasingly attenuates red light above 580nm until it cuts it out almost completely at 700nm. Using a normal IR-cut filter will usually include these wavelengths making the image appear redder and brighter.
I would like to conclude by saying thank you for all the weird and wonderful experiments and topics you've covered over the years, that have always kept me entertained and coming back wanting to learn more.
I really enjoyed your video and your experiment with the microscope lens. But there's something that you may not have considered.
You used the oil immersion feature of the microscope lens to couple the lens with the photo sensor. Then you were able to make photos using the lens. However, in that configuration, the lens was backwards. Thus, the indicated lens specifications would have been different from the specifications as used.
In normal usage, the oil immersion feature is on the side of the lens into which light enters. The specifications apply when the lens is used in that manner since that is how the lens was designed to be used. However in your setup, light exited from the side of the lens with the oil immersion feature. In essence, you were looking down the wrong end of a scope.
You can, of course, mount a lens backwards. It's a common method used for taking macro photos without a macro lens. But if you do mount a lens backwards, you cannot assume that the lens specs would be the same.
I believe that there might also be some added confusion regarding NA for single elements vs multi element systems, for a single element the NA calculation are easier, and you can have lenses with very low NA on their own. For multi element systems, I would say the fact that the definition of an f stop is related to the entrance pupil, and not, the geometric size of the aperture itself is important. What you need to measure is the size of the aperture as viewed through the lens (often magnified) , and not it’s actual aperture size itself. If you measure the actual aperture diameter of highspeed lenses, which often have quite a few positive front groups that magnify it, it will often be smaller than the geometric calculation would imply it to be! I think this is related to the curvature of the principle plane, which you can usually see with your own naked eyes if you just look at the aperture of a lens from the front
One thing to remember is that a theoretical f/0.0 lens would gather only about 5 times more light dan a f/1.0 lens. You can not make such a lens, but you can make a mirror that effectively works at f/0.0.
The way to do that is to take a strongly curved parabolic mirror and have a small sensor in front of it, facing the mirror. It will focus only at the middle of the sensor, and only to one distance, rendering it pretty useless.
Also, to win one stop over f/1.0, you would need a f/0.62 lens instead of f/0.7
The microscope lens might be infinity-conjugate corrected. This could also contribute to the strong aberations you see when focusing on "nearby" objects (PC keyboard).
An image of the night sky would be interesting to see.
Very cool! I wonder if you tried asking any AI chat bots to speed up your search for the answer. I've found they can give surprising insight sometimes.
For instance, I just asked Meta's Llama AI why the 1/2f relationship continues even to low f stops, and at first it thought I was talking about the aperture blades of a camera, but when I said I was using bare lenses, it seems to have correctly identified the reason. This is what it returned:
In that case, the reason why the aperture number of a lens is approximated as 1 / 2f, even at low f-numbers, is due to the way the lens is designed to correct for spherical aberration.
In an idealized lens, the aperture diameter would indeed follow the trigonometric identity I mentioned earlier. However, real lenses are designed to minimize spherical aberration, which occurs when light rays entering the lens at different angles are focused at different points.
To correct for spherical aberration, lens designers use a variety of techniques, including curvature of the lens surfaces, aspheric lens shapes, and lens combinations. These techniques effectively reduce the angle of incidence of light rays on the lens, making the sine of the angle approximately equal to the angle itself (in radians).
As a result, even at low f-numbers, the aperture diameter of a real lens is still approximately proportional to the focal length, following the 1 / 2f approximation. This is a fundamental property of lens design, and it's what allows lenses to produce sharp images over a wide range of apertures.
My favorite part of the video - you use "raises the question" instead of "begs..."
yes!
Fresnel lens does not have to have spherical aberration - it can be manufactured "aspheric", if shape of each groove is optimized. It has unfixable issues with off-axis rays, but on-axis it can be perfect. So nothing prevents it from having proper max aperture limited by total internal reflection.
This camera concept could have great potential as micro robot's optic for working in low light, it could benefited from extreme low light performance as it could use weaker infrared LED than conventional lens, saving energy and get a bit more of range out of it.
you know it's a good day when AS drops a video.