Professional lenses with built in sensors

Topic: 

A friend (Larry P.) suggested that the time was here for serious (ie. DSLR) cameras to undertake a design revolution. The old SLR design, with a mirror that flips up and must sit between the last lens element and the sensor, creates a lot of problems in designing the lens and camera systems. Yes, being able to view directly through the lens with your eye is a very useful thing. But at what cost?

We're already seeing the disappearance of optical viewfinders, even rangefinders, from small consumer cameras, if only to save space. Few people were using them any more, since the screen display turns out to show a lot more, and is even better than the eye in low light.

Serious cameras aren't seeking (too much) to save space. We want image quality most of all, and the tools to shoot good images. Looking through the viewfinder is one of those tools, but again, at what cost?

So a proposal is put forward that now that sensors are dropping in price -- even full frame sensors -- that each lens have its own sensor, and shutter, that is part of it. There would be a body which has a digital (and mounting) connection with the lens. The body would have display, processor, controls, battery and so on. It's a pretty radical proposal. Let's look at the advantages:

  • There is much more freedom in lens design, and lenses can be smaller, less expensive (for the lens at least) and lighter.
  • Each sensor can be custom fit to the lens and its image circle. Some lenses could have small sensors and some have large ones. You could work with both super large hi-res sensors on a 28-70mm zoom, and also carry a small, dense sensor which offers you a (higher noise) super-tele in a tiny package.
  • Each sensor can be tuned to the flaws of a particular lens, ready to correct distortions and other problems. (This could be done with a protocol for communicating those distortions to the camera too, and we're finally starting to see things like the 5D's database of lens light fall-off.)
  • You would not get dust on the sensor
  • You could build special bodies and/or lens holders that could hold multiple lenses, as now there is only an electronic connection to each lens. As a result you could switch among lenses instantly!
  • It might be possible to have standarization, so you could mix and match lenses from different vendors as you choose.
  • Image stabilization designs could be done with both sensor and optics, whatever works best.
  • The lens could be some modest distance from the "body."
  • Body design can also be liberated, as the mechanical linkage with the lens can be designed without the need for a light path.

There are some downsides

  • Obviously, sensors are not yet so cheap that this isn't a more expensive approach initially. But serious lenses are often more than $1,000 and this approach might not increase their cost by more than a few hundred dollars. For cheaper lenses, putting on a high quality sensor would not make sense, cost-wise.
  • In turn, where now you might put a lot of money into your one sensor, here it must be spread.
  • Today, if you get a new body with a new sensor, you now get the better sensor with all your lenses.
  • You lose the TTL viewfinder and focusing screen.
  • You need all new equipment, and probably want new mounting hardware too.

Sensors may not be cheap enough to do this today, but they are getting cheaper, and thanks to Moore's law this will continue. We've pretty much got all the megapixels we want now, so the main focus will be in improving sensor quality and ISO speed. Until sensors get so cheap that we might buy several that we know will be obsolete in a few years, one approach would be to still have a mount, so that sensors on a lens can be change. However, this need not be a quick disconnect mount, it would be more intended so you could swap out the sensor on a lens.

And of course, there could be a "sensor" on the lens which is not a sensor, but rather a mount to go on a body with a sensor, as we have today. However, this would have to be a body without a flip-up mirror, as the focal planes of these lenses would be much closer to the last lens element than they can be with an SLR. And I could also see the potential of a super-fancy rangefinder, which uses its own lens, but is digitally tied to focus and other information from the real lens to give you a view identical to the main lens, though DOF preview and manual focus would still be best on the screen.

Aside from the option of better lens design (and thus better image quality for the money) the two most appealing features to me are the instant electronic lens switch, and the ability to use different size sensors. Much as I would like to, even if I wanted to pay $6,000 for one of those amazing super-tele fast lenses that sports photographers use, I would only carry it around rarely. On the other hand, I might very well carry a short 85mm lens with a small sensor of the sort found in P&S cameras that gave me the field of view of a 600mm lens with 10 megapixels. It's going to get me photos I would not otherwise get because I'm just not going to carry a 600mm f/2.8 in my bag. Instant lens switch might also change your desires about what zooms you want, since one of the goals of a zoom is to switch focal lengths quickly, though another goal is to have fewer lenses in the bag. If not using a mount that holds multiple lenses, lens switch could still be a very quick unsnap/snap, with no caps to remove and no seal to make.

Of course, to do this would require a very high-bandwidth data/control/power bus that ideally was standardized over vendors and designed to be upwards compatible with the future, faster bus that might come along. There is already a Camera Link bus specification, but the technologies behind SATA-600 (also 6gb) or 10gig ethernet might make sense.

So I suspect that as sensors get cheap enough, we might see things move this way.

Wide angle lens

Let's consider how this might help us produce a wide angle lens. Good wide angle lenses are expensive. It takes work and good design to keep them free from distortions, vignetting and to make them rectilinear with a flat focal plane. Flare is also always a problem, as is doing all this for a sensor that is far from the last element. And these things are hard to do for a big image circle, though smaller image circles require very short focal lengths.

A sensor-included wide-angle could select the right focal length and image circle to get the best price/performance at suitable low noise. The sensors's pixels could sit in distorted rows to match the distortions of the lens -- indeed, one could go all the way to a fish-eye lens and put a fish-eye sensor on it to make it rectilinear. (This could also be done in software with some loss of sharpness.) The sensors could be designed so that they are larger (or have larger covering lenses) at the corners, to perfectly account for vignetting. And of course, one could use the short-focus design common in view cameras that can't be done in SLRs because the focal plane is so close to the last lens element.

It's not out of the question that such a lens/sensor could even be cheaper than a high quality lens able to put a great image on a 36mm full-frame sensor, and take better photos.

Comments

The gain in sensors has been (and will continue) to be slower than the usual Moore's law. Moore's law was originally phrased in terms of transistors per dollar, and includes gains from the squared improvements from feature size reductions. It also gets gains from dropping cost of silicon, cost of processing, etc.

Transistor features are already much smaller than optical sensor needs. The gains that remain fr are reduced wasted area set aside for wiring lanes. There is not a lot of improvement left because most of the silicon area is already active sensor area.

The gains from improved processing and reduced silicon costs do help sensors, but those are running about 10% per year. So you will need more patience.

The current Leica M8 digital does give you a hint of what this could be like. It has eliminated the mirror, etc and has the resulting improvements in lenses. It also has micro lenses on the sensor to mitigate the vignetting effects of some lenses. This could go further if the lens to sensor geometry were fixed per lens.

I'm not sure about switching from TTL viewing to LCD viewing. I find the current DSLR LCD viewing to be much inferior to TTL viewing. I'm not sure how much of that is lack of practice, and how much is other problems inherent to the LCD viewing environment. One part that would be needed in this new camera is a properly enclosed viewer. The side glare problems need to be deal with. But an LCD viewer with enclosed eyepiece (like you find on pro video cameras) can deal with that, and proper controls could let me see the affect of aperture and focus adjustments.

Transistors/chip was the original Moore's law, but in fact it has become more about price/performance. It has been happening more in small sensors, because unlike most chips, big sensors are looking for a large feature size, and want to be error-free over a very large chip area, while CPUs and other chips are trying to get tiny.

Small sensors are indeed getting cheaper very quickly, and better too. I am surprisingly impressed by the quality of the 10MP (and up) sensors found in tiny pocket cameras costing under $200 today. I think they've been on a much better than 10% improvement curve.

One advantage of this approach is, as I say, you don't have to decide on a fixed sensor size, and do everything around that sensor size. Instead, you can design a lens and sensor size together. The reality is that for most of us, the size, weight and cost of a 600mm f/2.8 wildlife lens for full frame 35mm are just not practical. We might very well decide that for that application, a 200mm lens with a 1/9th size sensor (and image circle) would be better, even though it is not as low noise. And, since we would have it on a rack of lenses on our tripod, we could switch to it at the press of a button.

One big issue is that you'd want to swap the sensor and analogue electronics and the ADCs otherwise you'd be adding a bucketload of noise. Then you'd need TTL focus and metering or some whizzy rangefinder type tech to eliminate the need for focus, then meter off the final image (as video does - correct exposure for the next frame based on the current one). So there would be quite a cost involved. Enough that I'm not sure it would ever be cheaper - basically you'd have a camera body that was a CPU and storage with a UI on the back. One tiny step up from the PHD I carry... so perhaps it might work the other way - build a camera that has no UI, just a network port, and is designed to be remote controlled (much like CCTV and web cameras).

The other thing is that many of us don't actually want what you're suggesting - I want a 300/2.8 not for the long reach per se, but for the astonishing sharpness and DoF control it gives me. Throwing that away to give me a more expensive lens instead doesn't make huge sense. If I wanted a cheap, light lens with huge DoF I'd buy a generic 100-300-ish zoom, or a superzoom compact.

I think it's significant that small sensors rarely if ever go shorter than 24mm (35mm film FoV equiv), and I've not seen one shorter than 20mm. So while it might be technically feasible to make a non-planar sensor and extreme wide-angle lens in a tiny format, I'm not convinced it will happen. For starters, the photolithography would be exciting. You'd want to guarantee a few billion in sales before committing to a fab that could do it, I think. Either that or go for a multiplane sensor and fresnel lens so you could assemble the sensor out of multiple flat bits. But again, not something that's going to be cheaper than a 35mm DLSR even with expensive lenses.

I'd also suggest that we're seeing a variant on what you're talking about now with stuff like the Canon G series - "high end" compacts that sacrifice speed and the fine details of SLR to give a cheap, lightweight camera instead. There's also the X series video cameras with still capture that use standard EF mounts.

So more likely we'll see convergence at that end. That way you can buy a DSLR+DV body and swap lenses as required. Like the Red consumer videowhizzkit that is currently rumorware.

Of course a 300mm f/2.8 is a glorious lens. And if you want to pay $3,500 for it, and carry it around, then adding a sensor to it so it is a permanently sealed unit probably adds $200 or so, even today. But I bet there will be times when you are not carrying it, because it is so heavy, and you wish you could get a shot that's not quite as good, but still pretty good, and thus would pull out the 75mm f/2.8 lens with 9mm sensor which cost $300 and is small, and light and more easily carried.

Note I'm predicting that this 75mm prime with small sensor actually delivers a much better image than the superzoom or cheap zoom. It's designed together with its smaller sensor, for that image circle. It's a prime, not a zoom -- look how good Canon's $79 50mm f/1.8 prime is.

Of course the "sensor" does include some electronics, including A2D, and some basic control modes (including lower res video feed mode for preview, metering, DOF preview) combined with full-res focus-point-only video feed for autofocus, and presumably full-res VGA crop feed for manual focus. However, it doesn't do more than the most basic raw compression, or UI, or display or image processing.

As for wide-angle lenses, I added a section on that. Indeed, for a quality wide-angle you probably will use a wider frame. Don't know what size, but it will be the "sweet spot" between noise, complexity of lens design and cost. 36x24mm is a compromise for various focal lengths. The real sweet spot will always be somewhere else, with different sweet spots based on your budget.

Another way of looking at this approach would be to look at a camera like the Canon G series. Imagine being able to buy a series of G cameras with different focal length lenses on them. Now take away all the duplication but the sensor (ie. screens, UI, processor, battery etc.)

I think you would end up with a lot more primes and short-range zooms. These lenses give you a lot more lens quality for your dollars. Again look at that astounding 50mm f/1.8 for $79, which shoots as well or better than any of the expensive "L" quality zooms at the 50mm setting.

Now I will agree there is something more likely to happen -- indeed already happening -- before this happens, namely digital multi-lens cameras that do not have a mirror viewfinder, but instead use digital. That can thus have the sensor as close as desired to the final lens element.

However, eventually I think the above sealed lens approach could become more popular, if the sensor chips drop below $100 it makes a lot of sense.

You would not, except for very high end, design a custom sensor for a chip. Rather there would be a range of different sensor sizes on the market, ranging from the tiny ones for P&S cameras up to 24x36mm and beyond. You would find the one that was closest to your sweet spot, and use that, and use digital correction for distortion.

For vignetting, you could make a custom lens cover for the chip with more light transmission in the corners, and do the rest of the vignetting correction in software. Though mind you, I think lenses today should contain within them a ROM describing their distortion and vignetting characteristics at all focal lengths and f/stops and have the camera or later software correct for it. There is no excuse for barrel, pincushion or vigneetting in a modern digital environment.

Looking at it the other way, this idea is chasing the "willing to change lenses but not wanting the bulk of an SLR" market, and willing to pay extra for quality. So you're into the Leica/rangefinder types, and I suspect that the micro 4/3rds and Leica cameras are there.

I would still like a tiny camera with decent quality, but I'm finding that in practice my compact waterproof camera that only really works at 200ISO (too much noise) is a good complement to the 1D. I fear that I'll end up with a 1Ds + xxD combo to get high res+crop mode combination, because I'm just not that much into changing lenses. Or more accurately, not that much into digging gunk out of the inside of my camera. Last weekend I shot an HPV race using mostly the 1DIII + 70-200/2.8IS and a 30D + 16-35. In retrospect I should have taken the Sigma 18-50/2.8 instead, then used the 24-70 on the 1D at night. Learning...

OK, new idea: ultrasonic shaking of the front element of a lens as a self-cleaning aid. Now *that* I would buy :)

So, concrete thinking hat: a 50D sensor cut in half to give 8MP and a 2x crop on 35mm lenses, and build an even smaller Sigma 18-50/2.8 which would be a 36-100/2.8 equiv FoV. Would I buy it? No, because the camera body would still be 50D sized, otherwise the 16-35 with 35mm sensor wouldn't fit into it, and the 18-50 would be skinnier but not shorter. The 8MP@100-300/2.8 (in 35mm FoV terms) would be great (50mm-150mm real range), but I keep remembering holding a 4/3 camera and wondering about longevity.

One interesting thing about this approach is it might be possible to have a system like this up and down the cost scale. Which is to say that even consumers, interested in a modest point and shoot, might end up getting a camera that consists of a consumer lens sealed to a consumer sensor, and then a consumer "body" which actually consists mostly of a big hole in the front where the lens mounts into, flush with, or sticking slightly out of the body.

By some arguments this would cost a little more than putting it all in one case, but the ability to use higher volume components should mitigate that, except for the highest volume cameras. Today the camera vendors tend to buy their sensors from a handful of chip makers, in the future they might by chip and lens assemblies, or make them on their own. This does require that the glue circuits that create the data link from sensor to body be cheap and standardized.

Now, no serious photographer is going to want the cheap lens/sensor that would come with a $150 P&S. But the reverse, where people experiment up with their lenses makes sense. Even a all-in-one camera could have the bus connector to allow it to use a fancier lens.

I imagine a mount design that uses a series of concentric mounts which fit within one another. A small lens can plug into the middle (where the data connection also is) and heavier lenses can attach to stronger, outer rings. Small cameras would not have a mounting ring able to receive a large lens, but large cameras would always be able to receive a small one.

Bodies of course will be cheap -- their most expensive part will be the screen, I suspect. I could even see the production of super-cheap bodies with a tiny (viewfinder sized) screen, so that every lens can be used to take a picture in an emergency. Pick up the lens, look in the viewfinder on the back and shoot. For more serious shooting, mount it on a real body. I presume a nicely designed, easy on/off physical mount.

Of course one configuration I would like would be a triple-mount that holds 3 lenses over a wide range. A flick of the dial on my body and I'm looking through a different lens. You could even do a UI when you spin the zoom dial on one lens and as you go below the range you feel some resistance but then are on the next lens, but still controlling it from the zoom dial of the first lens. Like a super-zoom but made of primes. That would be great for sports, wildlife and even some landscape work.

I agree with a lot of what's behind this post. I see two main problems with the dominant DSLR design: the mirror and the interchangeable lenses. The mirror is noisy and gets in the way of wide-angle lens design.

The standards necessary to change lenses force a physical dimension which makes these cameras really bulky. There is certainly a market segment for whom size is irrelevant compared to image quality, but this segment is very small compared with the segment where size, quality and cost all must be balanced against each other. This is the segment I'm in.

Before reading this article, I wrote a similar article generally decrying DSLR's as dinocams: http://www.embracingchaos.com/2009/03/dinocams---the.html Here I propose single-purpose cameras for extreme shooting situations and big-ratio zooms for everything else. This again shows my bias towards "good enough" picture quality. I spent enough of my 20s lugging around a 10 pound camera to care that much any more.

But for those who really do care about image quality, the idea of moving the interchangeable interface past the sensor is very interesting. As in software design, one trick in choosing the right place to draw an interface line is knowing what's going to stay standard. Ancient SLR lenses still work today because photons haven't changed and we're still using flat imaging planes that are about finger sized. If we move the interface to an electronic part of the stack, we need to think carefully about the evolution of electronics. If the A2D is on the camera then that means the interface is digital, and we know how long electronic digital interconnects last as we've watched the evolution from RS-232 through USB 2. Maybe it should use DVI or good old ethernet? Whatever we choose it's got maybe 5-10 years before it's obsolete. If the interface is before the A2D it could last a lot longer, and get benefits of improved analog electronics, but it means you need to be really careful in the design of the interconnect not to introduce noise. I don't know enough about analog electronics to know if this is practical but if it is, I could see a well-designed standard here lasting a long time. Something akin to an XLR cable in audio.

The other question to ask is how much value is there in the back parts of this chain? Would it be better just to put all that into a lens too? I guess we can just look at the price of high-end SLRs and ask if that's an efficient market. Given that Nikon and Canon are fierce competitors, it probably is, unless you want to claim that there's some giffen good effect at the very high end whereby they charge more for the best just to look cool. I doubt that though. I suspect these top-end cameras are really expensive because Canon/Nikon is doing everything they possibly can with their best research and production to make a camera that's slightly better than Nikon/Canon's best.

It's a very interesting idea. I hope it goes somewhere.

I fear it would be difficult to have the A2D be in the camera as I don't think you want to send the signal from each charge well that far before measuring it. In low light shots this can mean wanting accuracy of just a few electrons. I don't know whether that is possible or not.

It does, however, make sense to design each expensive lens so that the sensor unit and electronics can be replaced. The glass can then live on. I see two levels of this:

  • One is an interface which can be changed in the field. This would allow you to have an expensive sensor you share among expensive lenses. You then risk dust on the sensor. However, there is no mirror to worry about. This might be a quick mount like today's bayonets, or something more solid and slower when switches are rare.
  • Another approach would be something only switchable in the studio or in the service lab. It might involve unscrewing things, a need for a clean environment or even some alignment. Mainly to be done either before a shooting trip, or only when upgrading the sensor in a lens.

The mechanical mount might well stay non-obsolete for a long time. Because that mount does not have to be a ring, it can be designed to be simple, strong and long lasting. However, it needs to be designed to handle both small lenses and large ones, including lenses so long that the tripod attaches to the lens and the lens holds the camera.

Then you can swap out your electronics. I have more faith in the protocols and digital interfaces if they are kept general. I can still plug in to 10baseT today, though not so easily to coax. The protocol should run on an IP layer of course, so that what's under that can readily change.

You underestimate the value of the interchangeable lens. There are better zooms today but they don't compete with prime lenses because they are more complex. A zoom has many lens elements -- it has to. Each lens element is a chance for internal reflection, which means loss of sharpness and contrast. You can work hard to make your zoom better, with lots of coatings and baffles and groups, but you can't beat the simplicity (and better contrast) of a prime. Especially a prime aimed at a custom sensor designed for it. The high quality superzooms are not bad, and have been available for SLRs for some time. But you don't see people carrying them that much. Trust me, it's not because they love the extra weight of carrying lots of lenses.

Add new comment