The format offers an intermediate step on the way to Ultra High HD.
If you are in London during the 2012 Olympics without a ticket, there is an alternate way for you to “be present” at certain events. The BBC plans to broadcast part of the games in Ultra High HD (UHD) in select locations. UHD, which has been under development by NHK for years, delivers an “8K4K” (7680 × 4320) pixel image. With a planned commercial introduction in five years, an intermediate production step is necessary. This step, to be generally introduced at CES 2012, is “4K2K.”
For many in the broadcast industry, 4K2K will be a surprise — likely an unwanted surprise. For those in the film world, 4K2K is not new at all. Film is routinely telecined to 4K2K files for digital intermediate (DI) processing, and features are shot with 4K2K cameras.
In this first of a two-part series, we will begin our exploration of 4K2K by examining several current 4K2K cameras. Doing so will introduce the many challenges involved in this technology.
Beyond pixel count
Amazingly, unlike a decade ago when 2-megapixel CCDs were exotic, most camcorders now employ CMOS sensors, and 8-megapixel sensors are so common that many smartphones feature them.
Issues beyond pixel count are present with both full HD and 4K2K video cameras. First, how is information read from a CMOS sensor? The greater the number of photosites, the higher the frame rate and the larger the frame size, the more demanding the read-out process. Unless implemented well, a sensor's clock rate will be so high that the power it consumes will create more heat than can be dissipated, leading to limited shooting durations, especially in high ambient temperatures.
Second, many full HD, and most 4K2K cameras, have only a single sensor and thus use a Bayer RGB filter. To obtain an RGB image, sensor data must be de-Bayered. Camera engineers have a choice: Design a camera that delivers equal, or less, RGB resolution than does a three-chip camera.
Third, is the camera targeted to a cinema or video shooter? If the former, then the sensor will be large, Super 35 (23.60mm × 13.25mm), APS-C (23.6mm × 15.7mm), or Micro Four Thirds (M43) (21.6mm × 17.3mm) size, to support obtaining a shallow depth of field (DOF). A videographer may, however, require only the high resolution of 4K2K, thus allowing the use of a small 1/2in (6.4mm × 4.8mm) chip or 1/2.3in (6.16mm × 4.92mm) chip.
Consumer 4K2K cameras — yes, there will be consumer 4K2K cameras — will likely have small chips. Prosumer 4K2K cameras may use small M43, APS-C, or Super 35 chips. The use of M43 and APS sensors is natural because DSLR development is the parent of many of today's single-sensor full HD cameras and camcorders. This technology transfer will continue for 4K2K.
Red Digital Cinema's RED ONE
The Red Digital Cinema RED ONE is the 4K2K camera that most in the broadcast industry are familiar with. Equipped with the newer MYSTERIUM-X sensor, the ONE has a 5120 × 2700 photosite, Super 35, 14-megapixel CMOS chip. The sensor can be “windowed” during read-out to provide four 4K2K frame-sizes.
The largest 4K2K frame size, called “4K 16:9,” has a 4096 × 2304-pixel frame. The second frame size, called “4K 2:1” and “4K anamorphic 2:1,” has a 4096 × 2048 frame. The third frame size, called “4K HD,” has 3840 × 2160 pixels and can be used when shooting HD and 4K2K video productions. (I expect professional, prosumer and consumer 4K2K cameras will use a 3840 × 2160 frame. Projectors and monitors will likely support both 3840 × 2160 and 4096 × 2160.)
The RED ONE can shoot a range of frame rates. However, a trade-off must be made between frame rate and frame size. To shoot at high rates, read-out line width can be set to “3K” or “2K” pixels.
Once a RED ONE captures a frame, it is read out as sequential RGB. The sRGB data then take two different paths. RAW sensor data are wavelet compressed (12-bit REDCODE RAW codec) and output for recording. Data are also sent through the second path to an on-location monitor. While watching the image, a DP can adjust it to create a desired look. Settings, however, do not alter data that are recorded. Rather, the settings become metadata that are recorded with image data. During post, a colorist can use the on-location settings as well as make their own adjustments.
To obtain an RGB image, after decompressing REDCODE data, they are de-Bayered within RED applications. Until this point, frames are defined only in terms of photosite count. Camera resolution cannot be measured. Because there are many de-Bayering algorithms, the process is a critical aspect of single-sensor cameras. See the “Understanding de-Bayering” sidebar.
RED ONE, with the original 4520 × 2540 MYSTERIUM sensor, has a horizontal resolution measured at almost 3100 TVL (1550 line pairs), which means de-Bayer “efficiency” is about 78 percent. The goal of obtaining 4000 TVL resolution is brought closer by the RED EPIC, which has a 5K (5120 × 2700) sensor that, after de-Bayering, should provide about 3874 TVL (1937 line pairs).
At the opposite end of the price range, Sony offers its F65. The F65 has a CMOS sensor that Sony markets as delivering “true” 4K.
The 20-million photosite sensor, which has a diagonally oriented Bayer pattern, provides a green sample for every pixel in a 4K2K RGB image. Sony claims the chip's unique design provides full horizontal, vertical and even diagonal resolution on the green channel, plus full horizontal and vertical resolution on the blue and red channels. (This chip would have been unbelievable in 1985, when I bought Sony's first 250,000-pixel 2/3in CCD camera.)
Although different in price, the Sony and RED cameras both have a single Super 35-size CMOS sensor that can be read out fully for each capture.
Canon's EOS C300
The announced Canon EOS C300 also has a Super 35-sized sensor. It has a 3840 × 2160-photosite Bayer-pattern sensor. However, the sensor operates in unusual manner: Two rows are read out simultaneously. Thus, an “upper” and “lower” photosite pair are available at the same time. When each photosite pair from two rows is combined with another pair, delayed by one pixel-time, from the previous column, signals from four photosites are available. Each 2 × 2 photosite group (RGBG) is combined into a single RGB pixel. This process yields a 1920 × 1080-pixel frame that has, by definition, 4:4:4 color sampling. Each full HD frame of 8-bit data is then compressed using 50Mb/s, 4:2:2 long-GOP MPEG-2.
Canon has already announced that it is developing a full-frame DSLR that will have a 3840 × 2160-photosite sensor that will record 4K2K.
Steve Mullen is the owner of Digital Video Consulting.