Technology preview?

With photokina 2010 near and rumours abounding about an E-3 succesor coming, I thought it might be good to take a look at some of the technology that's going to be talked a lot about in the upcoming months. Now I'm not saying any of this tech will make it in the E-3 succesor, but I feel it would be godo to make sure everybody knows what's being said. So here goes: a laymen's guide to the tech we don't have (yet).

"Hybrid" Autofocus

One of the rumours is that some time in the future, Olympus is going to get rid of the optical viewfinder. One of the reasons not to do that yet is that the autofocus system used by mirrorless cameras (like micro-four thirds) is very slow compared to that of a regular DSLR. A regular DSLR uses a seperate autofocus sensor for "Phase-detect autofocus"  (PDAF). I'm not going to go into technical details here, but suffice to say it's a lot faster than primary-sensor based "contrast-detect autofocus" (CDAF) (one reason for this is that PDAF can predict in which direction focus is off, and by roughly how much).

To the left here you see how a classical DSLR does it's focussing. Light is reflected off a semi-transmissive mirror (it reflects X% of the light and lets Y% through). One part of the light heads up to the optical viewfinder while the rest is reflected by another mirror onto the PDAF sensor.

This drawing is just a sketch, in reality the distance between the two sensors and the mirror is exactly the same.

Now, look down a bit. Here is one way of getting the best of both worlds so to speak. You still have a semi-transmissive mirror, but instead of directing light to the optical viewfinder and the AF sensor, it divides light between the main sensor and the AF one. This allows one to use both PDAF and CDAF at the same time (gaining most of the speed of PDAF and accuracy of CDAF) and gives you an EVF without losing AF speed.

Of course there's still plenty of other pros and contras to replacing the OVF with an EVF, but having fast PDAF AF will likely convince a lot of naysayers, especially as EVF technology improves greatly over time.

The downside? Well, you lose some light on the main sensor when using this trick. This can be easily boosted out for the EVF or back LCD, but when taking the actual picture you'll want to drop the mirror. This means you cannot use PDAF while recording movies (unless you're okay with half a stop of light loss or so). In the end this means that some of the advantages of a truely mirrorless system are lost: you still get a mirror slap (though the mirror can be a lot smaller and thus more quiet in this configuration) and a rather complex construction.

(gapless) Microlenses

An image sensor has a lot of pixels (or photosites). However, these are not perfect: only a small part of each pixel is actually sensitive to light. The rest is a bunch of transistors and other overhead.

This means that not the entire surface area of the sensor is being used to gather light. Obviously, that's not dieal.

One solution to this is microlenses: a tiny lens in front of each pixel that directs the light to the actual light gathering element. This would effectively boost the base sensitivity of the sensor, and reduce noise. Huzzah!

The downside? Well, you need to put a very small lens accurately on top of every pixel. These need to be very well alligned and formed because otherwise you might get uneveness of pixels (and a static noise from that). All this needs to weigh little enough to not mess up sensor-shift IS and not get in teh way of the SSWF anti-dust system.

Now recently the new trend is "gapless mirrorlenses" which eliminate overhead even further by removing the remaining gaps between the microlenses themselves. This theorethically means that all the light hitting the sensor gets recorded. panasonic is rumoured to be working on such a gapless microlens sensor.

As a final note: four thirds sensors are currently NMOS type rather than CMOS, which means they have less overhead than a traditional CMOS sensor (and thus microlenses would have less an impact).

Non-Bayer sensor

Cameras cannot see colour. No, don't show me any colour pictures you've made, those are just a result of tricking teh camera. You see, every photosite simply captures photons, it doesn't care what colour (wavelength) they are. To make a camera see colour, we filter out everything that is not that colour before it hits the photosite. Every photosite in a CMOS sensor has a little colour filter in front fo it that only lets through one of the primary colours: red, green or blue. These photosites are surrounded by photosites that see the other colours and they are combined with these differently-coloured brethren to create full colour. The pattern that these photosites are distributed is in 2x2 squares with red and green on the top row and green and blue on the bottom one (there's twice as much green because the human eye is more sensitive to green).

There's a couple of problems with this.

First, you're putting a colour filter in front of every photosite and thus you're losing two thirds of your light for each photosite (that's about 1.5 stops of light).

Secondly, you need to use a process called demosaicing to get your real colour. This is a form of interpolation: you "guess" the colour of each photosite based on the data of its neighbours. You end up with only about a quarter of the colour resolution and might get artifacts.

So let's take a  look at some alternatives.

The first one here is the classic bayer pattern I just discussed. The second is sigma's foveon system. Here every photosite is made up of three photodiodes that are stacked vertically. Instead of a filter it uses the properties of light traversing silicon to sepearet the primary colours: different wavelengths can penetrate deeper into silicon (and get absorbed at different depths).  the advantage is that you don't need colour filters or demosaicing. The disadvantage (one of ) is an increased cost per pixel. Generally foveon sensors have a smaller pixel count but gerater light sensitivity and higher or equal colour resolution than higher resolution bayer pattern sensors.

The third one that's going to get a lot of talk over the enxt few weeks is fuji's superccd EXR. Here the bayer 2x2 pattern is mirrored horizontally and vertically to create a 4x4 base block. This puts photosites of the same colour next to eachother. An EXR sensor can be run in two configurations: high resolution and high dynamic-range/low noise. In the latter mode, the adjescant pixels are "binned" together into a superpixel with four times the light gathering capability of its smaller brethren. This gives an increased dynamic range and a reduction of noise.

However, there's a rather big "but". Remember that per-pixel overhead from the microlenses? Well, you get that here too and because of this the 4 combined smaller photosites do not have the light gathering capability of a pixel that is four times bigger. That means you can't simply say "create a 48 megapixel camera that can run as a 12 megapixel equivalent of current cameras" because this overhead would make each of the superpixels gather significantly less light than the existing photosites in a 12 MP sesor. Unless you use microelnses, of course (adn aer very good at them).

Also, it says "superCCD" for a reason, not superCMOS.

Another alternative, which is not depicted here, is a 3CCD type system. here there's actually three sensors spread about the mirror chamber and a central "mirror" that reflects light in different directions depending on the wavelength. The advantage is that you don't need a colour filter so you lose less light, and you don't need demosaicing so you get fewer artifacts. The disadvantage is that you get the cost of three sensors, allignment is very very critical (and hard) and there's no feasible way of providing in-body IS.

Anti-Alias filter

Recent olympus cameras have a reputation of having pretty heavy Anti-alias (AA) filters, which badly influences their sharpness. The micro-fourthirds cameras, however, have a lighter one thanks to an improved image processor that can handle with AA-filter's arch-nemesis "moiré". Moiré is the effect when a small pattern in an image (say a square pattern on a tie) resonates with itself because some of the detail falls between the gaps of sensor photosites. Say you have a tiny pattern of black and white like BWBWBWBWBW but when this gets projected onto the sensor all the W's fall on the photodiode and all the B's fall on the transistors and overhead (see why I started with the microlenses). The resulting image will have WWWWWWW and the pattern is lost. An AA filter turns all the tiny BW's into GG's (gray-gray). You still don't quite have the pattern, but at least you don't get any fake bright or dark blobs.

A good example of moiré is teh tie example above. Ever seen a TV interview and when one of the guests moved their tie seemed to pulse? That's moiré.

A related problem that AA filters deal with is "jaggies".  Jaggies are when diagonal lines seem to be sharp and blocky rather than smooth. The culprit here is a combination of the gaps mentioned above and an artifact of analog-todigital conversion.

So: stronger AA filter: less sharpness, moire and jaggies. Weaker AA filter: more sharpness, moire and jaggies.

Microlenses help against both jaggies and moiré, by the way.

More?

That's it for now. If you want more explanations about anything, just ask and I'll do my best to give a comprehensible rundown.