THE ARRISCAN

Arri's new Arriscan Film Scanner and what it does.

TELECINE, PAST AND PRESENT. The Arriscan is at heart a telecine machine, but most manufacturers these days seem to avoid that term - Philips preferring the term "Datacine" for example - once again we suppose to steer us away from any tedious notion that "Digital Cinematography" equipment is just tarted-up television equipment. Admittedly, in the past "television" telecines tended to be "real-time" machines, that is they scanned 25fps film at 25fps, whereas the advent of cheap random-access mass digital storage has meant that it is now possible to scan film "off-line" at a slower frame rate but with a higher picture quality.

In "The Good Old Days" when there was only Standard Definition TV to worry about, there were three principal approaches to telecine. The original (and some still say best) system is the "flying spot" approach, which dates from the early 1930s. (This was before there were any "serious" television cameras. In fact, people had been trying to get workable television systems using this technique since the 1890s!). Basically a special picture tube produces an extremely bright blank white raster (ie a set of ordinary TV scan lines), and this is focussed onto the film. (Originally, the scanning was done with a Nipkow disk illuminated by an arc light).

Thus a tiny and intense spot of light scans over the film from left to right and top to bottom. A set of photocells (usually three - filtered to be red-, green- and blue- sensitive) records the variations in brightness as the spot of light scans over the film surface, and that produces the video signal.

Although simple in concept, this can actually produce extremely good results (at least with SDTV), and for a long time the Rank Cintel Flying spot Telecine's were the standard by which others were measured. The flying spot technique has a lot of other advantages too, in particular the scanning geometry can be changed very easily, much the same way the height or width of a TV monitor can be adjusted. So for example, anamorphic film can can be scanned without needing special lenses.

Also, because it is only looking at one pixel at a time, it can have elaborate circuitry to handle extremely bright parts of a scene, without introducing grain into the darker parts. (Doing this with a normal television camera would necessitate duplicating the same circuitry for EACH ONE of the hundreds of thousands of pixels that make up the image - utterly impossible).

Unfortunately, the flying-spot telecine starts to run out of enthusiam once you get past standard definition line rates. It's true HD flying-spot telecines have been developed that are OK for HDTV use, but it's doubtful they will ever be good enough for true Digital Intermediate film work.

The second approach to telecine was a fairly obvious one: you more or less point a conventional TV camera at the film using an extreme closeup lens. Well, not quite as simple as that; you also need some way to synchronize the film to the video frame rate. In many such setups, the film runs continuously through the gate and is illuminated by a Xenon flashlamp during the vertical blanking periods. This "exposes" the camera sensor, and the resultant video signal is then extracted from the camera tube line by line in the normal way between flashes. Using a flashlamp "freezes" the film image, eliminating the need for an intermittent movement.

This sort of telecine was popular before the advent of color TV, since it could be easily implemented using a cheap vidicon camera tube. In those days, compared to the cost of even the cheapest broadcast-quality videotape machine, an indifferently-performing vidicon-based telecine was still a pretty good option. By the time color broadcasting started to proliferate though, VTRs had gotten a good deal cheaper and that and the emergence of microwave-linked TV networking, more or less spelled the end of the market for el-cheapo camera-tube based telecines. Although it was certainly possible to make color versions, and Sony are in fact currently championing this approach with their 1920 x1080 CCD color cameras, resolution-wise it can only really be as good as the TV camera itself, which is to say, not all that good, particularly (again) for digital intermediate work. With modern color systems, when you're after superior image quality, it's more cost-effective to build a "proper" telecine!

The third approach (used in the Spirit Datacine and others) is to run the film past three linear CCD sensors, fitted with filters to make each sensitive to one color- red, green or blue. (A "linear" sensor produces just a single line of video, so the "vertical scanning" is achieved mechanically by the running of the film past the sensor).

This also has quite a few advantages, foremost being its mechanical simplicity. Because the CCDs only have to scan and process process a single line of video, they're simpler in construction, and so can be manufactured to a much higher level of performance. Further, if you use an incandescent projector lamp to illuminate the film, you can have a fourth sensor which only responds to infrared. Movie film is normally completely transparent to infrared, so the only image that should come from that sensor will be from scratches and other defects on the film. These signals can be used to automatically "repair" such defects, usually by substituting adjacent pixels, or the corresponding pixels from the previous frame.

However, this approach still has its deficiencies. For one thing, it can't produce a proper still frame, it can only "fake" one by accessing the data stored in its memory. While this isn't necessarily a problem for the user, it makes maintenance (focussing etc) somewhat tedious. The earlier versions were also notorious for the "plastic-ey" color images they produced, although this has admittedly been improved in later models with better CCDs.

THE ARRISCAN Arri's new film scanner is a sort of hybrid, using the best features of these three approaches. It uses an intermittent movement basically similar to those used flying-spot scanners, but for the optical pickup it uses a "35mm-sized" CMOS sensor, similar to those used in digital stills cameras, but without any of the microscopic "pixel filters" they use to compute a color image. In other words, it's like a monochrome version of a "megapixel" stills camera sensor.

To produce the color image, each frame of film is "flashed" three times, using Red, Green and Blue Light Emitting Diode (LED) backlights in sequence. The CMOS sensor thus takes three "photographs" of each frame, one under each color. Since the LEDs are inherently monochromatic (that is, they only produce a single wavelength of light), no filters are required. The brightness level of the LEDs is also inherently stable, but simple feedback monitoring makes this even better. In their press release, Arri also mention the possibility of infrared scratch and dust scanning (using a fourth image taken with infrared LEDs) but they don't go to this in any detail.

Before suitable LEDs came on the market, there wasn't really any device capable of accurately carrying out the three-color flashing technique, without resorting to clumsy mechanical devices. Xenon flashtubes can produce short, sharp pulses of illumination, but they require filtering to get the required three colours, and that would require either a motorized filter wheel or a separately filtered flashtube for each color, greatly complicating the construction. Xenon tubes also suffer from a fairly short operational life and considerable temperature and ageing drift, all of which would have to be compensated for. They also generate a lot of heat. A tungsten projector lamp would suffer from similar problems, as well as needing something similar to a camera shutter to pulse the light on and off.

The LED light sources make everything much, much simpler. LEDs only require 2 to 4 Volts to operate and they're very efficient, meaning enough light can be produced with negligible heat generated, a great concern when scanning somebody's precious negative. With LEDs it's also much easier to accurately control the amount of illumination provided, by simply switching them on for longer or shorter periods of time.

However, perhaps the greatest innovation with the Arriscan its ability to "double expose" each frame, to give an previously-unheard-of 16 bit resolution

The problem of dynamic range with electronic image sensors is well known. All electronic devices generate some random "noise", and this is what gives you snow on a weak TV signal and hiss on a weak radio signal. It is also what gives you grain with image sensors. And in all cases, you can usually "drown out" the noise by supplying it with more signal - putting up a better TV antenna, or giving the image sensor more light.

The only problem is that if the signal suddenly gets stronger, it can cause signal overload. This is like what happens when you're watching a TV program and you turn up the volume during an annoyingly quiet bit, and then a noisy ad comes on and blows you out of your chair:-)

With TV and radio stations, there can be enormous variations in signal strength when you change channels, but all modern TV sets are designed to cope with this with what is called "Automatic Gain Control" or AGC. You can also have AGC with image sensors, and in fact most cameras provide this feature in one form or another.

The problem is that while they can cope with quite wide variations in the overall brightness of a scene, they can't do anything about the variations in brightness inside a scene. The actual brightness of the individual pixels focussed on an image sensor (or a piece of film) can vary over a range of tens of thousands of times with a contrasty scene.

The problem simply is that if you let in enough light to drown out the noise in the darker parts of the image, the sensor overloads on the brighter parts, giving the all-too-often seen white patches on video images. But if you close down the iris to stop the overload on the brighter bits, you don't get enough light to operate the sensor on the darker bits.

So what do you do? Well, you might notice that just about all TV programs shot on videotape are done in closed studios where the lighting guys can make damned sure that there are no mega-varations in image brightness! Otherwise you shoot on film, which automatically squeezes the brightness range down to a much more manageable level, like, one that an electronic sensor can cope with. (Shooting on film is really the equivalent of putting a gigantic tent over New York City and re-lighting it to television standards. No, don't sell your Kodak shares yet!)

Up until now, this was satisfactory for TV broadcasting, but not always up to the more exacting demands of Digital Intermediate post production. However the new CMOS sensor used in the Arriscan allows you to overcome this limitation with a rather neat trick, one that is impractical with the more usual CCD sensor.

Both CCD and CMOS sensors basically consist of a thin slice of silicon covered with hundreds of thousands (or millions) of tiny photo-electric cells (basically microscopic solar cells). Focussing an image onto this array produces an analog of the light image in the form of electrostatic charges stored in each of the cells. Such devices are exquisitely efficient at collecting photons; the problem has always been getting the charges off the sensor and into the image processing circuitry, without introducing too much noise.

With CCDs the actual accumulated charges themselves are shunted off the chip, while with CMOS the circuitry only reads the voltage stored at each pixel. In other words, the CCD readout is "destructive" wheras with the CMOS sensor you can read it out as many times as you like without affecting it.

This makes it easy to "double expose" each frame. A longer exposure "brightens up" the darker parts of the film and allows the CMOS sensor to work with minimum noise ("electronic grain") but at the expense of overloading it on the brightest parts of the image. However with a CMOS sensor this problem can be overcome by taking two "pictures": one early in the exposure cycle before the brighter parts of the image have had a chance to overload and then a second one later in the cycle when the darker parts have had a chance to expose properly.

It's possible to discard the underexposed ("noisy") parts from the first exposure and remove the overexposed ("bleached") parts from the second exposure and then combine what's left. This allows a previously unheard-of 16-bit resolution (in other words, 65,536 possible video levels per color). Although double exposure could theoretically also be done with a CCD sensor, it would be a far more complex process, and probably not practical. Some people say 16-bit resolution is a bit of an overkill, and so the usefulness of this will be determined by user feedback, according to Arri. The great thing is, this feature can be provided entirely by software changes, so if you don't need it, you just don't use it.

Another intriguiging aspect is that the Arriscan software is "open source" based on the Linux Operating system rather than the more usual Windows or Mac OS. That means that anybody is free to modify the operating software or write their own, a refreshing change.

The Arriscan can produce 2K, 4K or 6K scans, depending on how much time (and "time" = "money":-) you want to spend. The CMOS sensor was a "native" resolution of about 3,000 x 2000 pixels but higher resolution scans are available, by physically moving the sensor chip with piezoelectric actuators. (This is essentially how ordinary document scanners work, by the way). Basically, the lower the resolution you output at, the faster the scanning goes. The fastest "2K" scanning speed is about one frame per second, a limitation imposed by the CMOS sensor itself.

As we have pointed out numerous times here, there is a definite technological "brick wall" you run into when you try to increase the resolution of television camera sensors. The six megapixel CMOS sensor used here can only run at a maximum of three frames per second, which means one fps if you're sequentially scanning the Red, Green and Blue images. There certainly are higher resolution sensors available, but they take up to twenty seconds to download a single picture! At the moment it's much more practical to simply move the sensor around a bit:-)

CMOS sensors have numerous practical advantages compared to CCDs. A little-appreciated fact is that they can be fabricated on standard modern CMOS digital chip production lines. CCDs on the other hand are based on an ancient chip technology called PMOS, which isn't used for anything other than CCD sensors any more, so it's a bit like someone wanting to make bread rolls in a cookie factory - the basic manufacturing processes may be similar, but they're different enough to be a real headache in a mass production situation.

Once a chip manufacturer has got his CMOS production line running smoothly - he's got his temperatures and gas mixtures and so on adjusted "just so" - he's not about to go changing everything just because somebody wants to do a smallish run of antediluvian PMOS chips! It's rather like film processing: once the processor has got his chemical mix stabilized for standard color film, he's not going to be particularly interested in emptying it all out because somebody comes in with two reels of Black and White!

Thus CCD sensors have to have their own separate fabrication plants, and nobody is going to want to sink too much money into improving the PMOS process for such a small-market item like the CCD sensor. At the moment, CCDs still have the edge over CMOS in low-light sensitivity and overall picture quality, although the gap is narrowing. Still, this isn't such a concern with a telecine, since there is plenty of illumination available.

So the Arriscan is really a very clever application of some already-existing technologies, (several patents are pending) put together in a way that nobody else seems to have thought of, and perfectly complements the Arrilaser! But this is typical. Panavision used to have clever engineers who could think up innovative ideas like this, but they're all long gone, mostly replaced by the classic IBSs ("Idiots in business suits":-)

So what does it all mean? It means that while Arri were developing the Arricam, Arrilaser and Arriscan, Panavision were pissing around with "Panavising" Sony's half-assed HDTV cameras. For ONE customer who took his subsequent business elsewhere anyway... Nice one JF!