[June 16, 2011] Decoding Digital Imagers: Part 2 [American Cinematographer, The] Tweet (American Cinematographer, The Via Acquire Media NewsEdge) An analysis of the sensors used in today's digital-imaging systems. Last month we established the two dominant forms of imaging sensors in use coday. in this article, we will examine how these sensors arc applied in specific digiral-imaging platforms, and how their image d'ita is processed and delivered to various recording mechanisms. Solutions for Color Though a photosire on a typical imager can produce a range of analog clcctric.il signal strength (representing die brightness at a given point in an object-image being projected onto the sensor), the photositc itself cannot dillcrentiate between die color wavelengths of the light being collected. Camera engineers have therefore devised some clever ways of capturing and interpreting this color information. The first of such strategies is using three separate imagers attached to a prism, with each imager filtered for scnsithity to a separate H (red), G (green) and B (blue) color exposure. Thanks to recent Advances in color demosaic processing and CMOS-imager fabrication techniques, singleimatrer systems that add color filters over the sensor's individual photosites in various patterns, or "arrays," have also emerged. Tri-Im-ager Systems Tlirce-chip digit; J -imaging systems evolved out of the 2/3"-video market of die 1970s. Those cameras used various vacuum rubes LIS dieir photosensitive image pickup devices. Because CCDs dominated in early digital-video camera systems, they were the first sensors to be incorporated into this camera architecture. With tliree-CCD cameras, an object-image passing through a lens is directed through a beam-splitting prism that separates its red, green and blue light components (through the use of various interference filters) before focusing each onto a separate 2/3" CCD sensor. Each sensor then captures full-resolution data for the corresponding co lor- filtered channel. "The 2/3-inch format was spedii - caJly designed to empower the development ot the first truly portable video cameras," notes Larry Thorpc, national marketing executive of Canon's Broadcast and Communication Division. "This image format became so ubiquitous worldwide that it continued when the CCD arrived in the mid1980s, resulting in the first 2/j-inch standard-dei CCD camera systems. In the 1990s, the first V;-ineh III) cameras appeared, and a few years later, the first 24P I ID system was developed [by Sony, with the CmeAlni HDWF900]. "A fundamental philosophy underlying the design of] early HD motion -imaging systems was the desire to have complete intere hangeability ot all available HD lenses, both cine-style and video, among all cameras that used the 2/3-inch image-format platform. However, by any definition, the 2/3-inch format is a small image. Though 2/3 inch equates to an almost 17mm-diameter circle, the actual specified image format is only llmm diagonal, which poses a unique challenge to I ID lens design." With any digital imager, as the size of the sensor is decreased, the demand on the quality of lenses - their MTF performance capabilities - is proportionally increased. Thorpc explains, "The size of an HD -inch sensor is 9.6mm x 5.4mm. The MTF issue for any system is, 'How many line pairs can we pump through Imm?' In the lens-camera system, we look separately at resolution horizontally and vertically primarily because there are different mechanisms going on m both directions; we hope the lens is balanced in both directions, but in the camera it can be different. A typical 2/.i-inch imager has 1,080 samples vertically. Nyquist says that if you have 1,080 vertical samples, you can only resolve 540 line-pairs. Now, it just so happens that a -/.i-inch sensor is 5.4mm tall. That means the lens has to transfer 100 line-pairs/mm - with high contrast - it that image is to appeal' sharp. With 35mm-sized sensors, the lens only needs to resolve 20-25 line-pairs/mm to have the [sufficient] contrast. Because 2/3inch sensors shrunk the image fomat down so much, the [lenses' optical performance had to be pumped up. "It you dice that same 2/3-inch sensor in the horizontal direction," he continues, 'You'll see that the final 33Mhz filter required in all HD cameras actually helps a hit - we only need 81 line-pairs/mm ior HD. [For SD, we need 32 line-pairs/mm.] That's 2.7 times more information that has to be transmitted rHrouiih the lens. So there is a huge difference between un HD lens ami an SD lens." Adding to this increased optical demand in the 2/3" format is tiie beamsplitting prism in the optical path. "Clearly, there is an optical complexity in this system that is absent troni any Him camera," says Thorpe. "The lens and the prism operare as an integral optic.U system, and each must be ot the highest optical quality to ensure that the overall HD invagini; performance is optimi/ed and the optical aberrations minimi/ed. The. latter LS especially challenging within the context of the triimager's 11mm diagonal image format." Steve Mahrer, senior technologist for Panasonic's Production and Media Services Division, expands on this idea. "With a three-CCD prism assembly, there are some unique aspects that are quite important and often overlooked. First, the light coming through the lens is split optically into three wavelengths [red, green and blue] by the prism, but you're actually not losing any light; you're just channeling the appropriate wavelength through to its imager - red to the red inniger, green to the green imager and blue to the blue imager. This differs greatly from a color-filter overlay, where you are essentiallv throwing away two- thirds of the light at each [photosite]. With a three -imager system, you can actually use that light, and that makes the camera significantly more efficient and sensitive. That's quite important, especially with smaller imagers." A strong attribute of the triimager configuration is its ability to provide full-resolution bandwidth tor each primary color channel. Mahrer elaborates, "When a tri-imager camera has three native 1920x1080 CCD imagers - one each for red, green ami blue - that means it's a 4:4:4 camera all the way through from image capture to image processing and then to recording. There's no deBaycring required, no interpolation, no black magic. It's just Liking the raw RGB information from the imagers and processing it. That's quite important, especially when you're doing any sort ot color-specific work such as greenscrccn photography, where vou could get color aliasing from a Bayer pattern." An additional consideration of 2/3" tri-imagcr systems is that they produce depth -of- field characteristics similar to those ot the 16mm format, so you must use lenses that are scaled accordingly. This has been a significant factor in the push to develop singleimager 35mm-sized sensors that can use the vast array ot 35mm cine lenses. Single-Imager Systems The last decade has seen an everincreasing rise in the development anil implementation oí single-imager digital-camera systems. Tliis is due not only to increasing fabrication capabilities, but also to improved image-processing algorithms and color-filtering techniques. "The single-imager system inherently enjoys the same optical simplicity as a 35mm film camera," says Tliorpe. "Absent an optical relay system, it presents the same single tocal plane as a film camera, so it can use standard tilín lenses. Additionally, single-imager cameras employ a single, proprietary, large-format, solid-state image-sensor system - either CCD or CMOS - whose function is to implement an opto-electronic transformation of the focused object-image, [similar to] the opto-chemical transformation that takes place in the film camera." However, unlike film emulsions, current semiconductor-fabrication techniques cannot adequately allow a single point on a sensor to be sensitive to (and able to capture a value for) each of the three priman· colors. (I lowever, Foveon manufactures an imager incorporating this idea for use in singlesensor still-photography cameras.) So in order to derive RGB color information, CCD and CMOS single-imager systems require a color-filter array - applied directly to the sensor's photosites - to designate a prescribed portion ot the photosites as green-sensitive, blue-sensitive and red-sensitive. Color-Filter Arrays CFAs perform the same function as the tri-imager's prism assembly, hut instead of separating light to individual sensors, they divide incident light to specific photosites. Tliorpe elaborates, "The separation ot the incident light into die requisite RGB color representations is accomplished with a mosaic of filters that individually applies a primary color to encompass each [photositc] within the imagcr's array in a predetermined pattern. Precision interpolation - real time or non-real time, dependin"; on the manufacturer - then takes place in the digital-video-processiiig system to reconstruct the separate RGH components." John Galt, senior vice president of Panavision's Advanced Digital Imaging Group, notes, "The filters used in a CFA arc basically pigments lithographically printed onto a sensor during fabrication. They arc typically absorprion-type filters that have much less efficiency than a three-CCDs interference filters, which have about 90-percent efficiency. So when you're building a camera with a CFA, die color and saturation level of the filters greatly affect the sensitivity of the camera." "Careful selection ot the color filters is important," cautions GIenn Kennel, president and CEO of Arri, "because that defines the response of the sensor to color and mixed illuminants, as well as the overall color gamut of the sensor. That applies whether you're talking CCD or CMOS. However, the CiVIOS sensors now-superior sensitivity ^ives us more flexibility in working with the color filters; we can make the responses broader and at the same Urne tailor them to creating clean color rendition and gamut." Several types of CFAs arc used today. The most prevalent is the Bayer CFA, named for its inventor, Bryce E. Bayer of the Ivastman Kodak Co. In 1975, Bayer tiled for a patent for his CFA pattern. Hc described a scenario in which green (or "luminance-sensitive") elements are arranged in a checkerboard grid with their corresponding red and blue ("chrominance-sensitive") elements in a ratio of 2:1. (See illustration above.) The raw sampling output ot a Bayer- filtered sensor is often referred to as u Buyer-pattern image. To obtain a full-color image out of this anisotropic array of color information, various de mosaic ing algorithms must be used to combine and interpolate a Kill-spectrum composite of red, green and blue values for each point in the digital-output pixels. Today, many systems allow this demosaic processing to be done incamera (generating a video-output signal that can be monitored on set and recorded to tape or tupeless formats), while others save the raw, unprocessed Bayer data for later processing, demosaicing and transcoding into various delivery specs. "You have a fundamental dilemma with the Bayer CFA that you do not have with a three-chip camera," notes Thorpe. "With the Bayer pattern, you will see lots of green photosites in an odd quincunx structure, each surrounded by two reds and two blues. If you break them apart to look at them separately, you will find the resulting patterns for R, G and B [pictured above]. "VVe see that with the Bayer pattern, half the samples, both horizontally and vertically, are green, while the remaining halt are allotted to red and blue. The red and bhic patterns are arranged in a classic cardinal structure - they don't have that quincunx look - but there art bit; gaps between them. The fill tactor is very big Ln the red and the blue. The dilemma with the Bayer pattern is that depending on the total number of pliotositcs, you may h vive a good sampling resolution with the green, but only half that resolution with the red and blue. "With any sampling," he continues, "you have to contenti with aliasing, but with a single-sensor system you can only use one Optical Low-Pass Filter. So which sampling resolution do you design it for? Do you tavor the green or the red and blue? It you opt to filter for the green, you'll get good green resolution performance with a degree of green -aliasing control, but there will be -? whole lot of aliasing on the red and the blue. [See figure #1.] "The other option is to use an Optical LPF that properly handles the red and blue aliasing, lcanng you a respectable red and blue response bur really clobbering the green resolution. [See figure #2.] "I recommend a compromise: use an Optical LPF that tails somewhere in die middle and live with ? certain amount of green aliasing and a higher amount of red and blue aliasing," he concludes. [Sec figure #3. J "You will find this aliasing with a test chart verv easily, but fortunately, the real world is pretty forgiving!" Buyer-pattern single-sensor systems ore being used more and more in digital-imaging platforms today, including Arris D-20, D-21 and Alexa; Red Digital Cinema's One and F.pic, Vision Research's Phantom cameras; and Weisscams HS-2. Sony, long a purveyor of CCD technology, has also begun using Bayer-pattern CMOS sensors with such cameras as the I3MW-FS, as well as the new F65, wliich uses an 8K diagonally rotated Bayer-pattern super 35 CMOS sensor. Most DSLR cameras capable of recording 1 ID video also utilize Bayerpattern sensors. Demosaic Processing After the full-spectrum incident light projected by the lens onto the sensor has passed through the colormosaic filter structure and is collected by the photosites below, the camera's digital image-processing functions interpolate that information und assemble usable color information for each derived digital-output pixel. With a Bayer-pattern imagci; this process is called demosaicing. The process of interpolating, or "deBayering," the color-matrix information from a Bayer CFA sensor can first be looked at troni a single, rourpixcl cluster with two diagonally oriented green photosites bracketed by a single value for red and a single value for blue. Using this information to derive full RGB values for each ot the photosites, however, cannot yield 4:4:4 sampling. (For more details on color sampling, see "The Color-Space Conundnnn.'V/CJan. ?5 and April ?5, or visit www.theasc.com/magazine/ ianO.S/conundrum/index.html.) In this worst-case scenario, these four pixels would represent a color sampling of 4:2:0. One red pixel would be used to determine che red tor all tour points, and one blue pixel would, determine the blue values for all four outputted digita.! pixels. Fortunately, with today's powerful image -processine; capabilities and sophisticated demosaicing algorithms, this is not how typical deßaycring is accomplished. Instead ol looking at a grouping of only tour photosites to determine the color values falling in that general area ot the sensor, today's Buyer-processing techniques utilize a much larger grid of photositc data to synthesize the hnal unage, and can deliver efficiencies of at least 80 percent. "With Baycr-pattcrn-imaging cameras, you have to look at all ot the values around a given [photosite | area, typically a group of around seven pixels, and then interpret and calculate from those values what one [digital-output pixel] value will he," says Galt. "Bayers invention goes back to the 1970s, so there have been almost 40 years of development in deBayering algorithms. In fact, today's demosaic processing is contextual; it can understand it it is looking at a sky, greenery or graphics and then processes that information accordingly." Because the single-sensor i mager must divide its photosite count among green-sensing, red-sensing and bluesensing elements, tidl-resoiution information rrom that sensor is not possible. Kennel explains, "In dealing with a Baver filter arrav, you need to reconstruct a clean, high-quality signal. If you reconstruct that Bayer data hack to its RdI resolution, you're going to have some color artifacts because you're sampling the red and the blue at half the resolution ot the green - the color is sub-sampled and has to he interpolated hack up. Conversely, when you look at an over-sampled image, such as a 2K image derived from a 3 K or 4K sensor, vou get a cleaner, sharper image than when reconstructing the direct resolution ot the sensor." Thorpe expands, "With die green pixels that had gaps between them when vou looked at them horizontally and vertically, if you look ;it them diagonally, you can see that there is a continuous stream of pixels. 1 marvel at the cleverness of the Bayer pattern, because what it's effectively doing is diagonal sampling. Now, this is not intuitive, but what it adds up to is that it gives JOU the 'beans' in terms of the grcen-photosite sampling - you get the full benefit horizontally and vertically - but you lose resolution diagonally. That's, not intuitive. You would think that diagonal sampling would favor resolution diagonally." Galt explains, "When you look around you in die real world, almost all the lines in our environment are either horizontal or vertical, so sampling in a diagonal direction is a good idea. You're less likely to get aliases that way. In fact, it's an old trick that was used in halftone photographic reproduction in newspapers. The early lithographers realized that if they rotated their [halftone] 'screen' by 45 degrees, die screen became less visible. That same concept applies to the Bayer pattern." Alternative CFAs There are several alternative methods ot color filtering a singleimager system to capture color information. Some of these alternatives are Variations on the Bayer RGGB pattern, such as the RGBE [red, green, blue, emerald] filter arrangement, which splits half ui the green-filtered photosites into alternating green and emerald [blue-green] filter overlays within the Bayer mosaic structure. The CYGM [cyan, yellow, green, magenta] CFA, however, introduces a wholly different color-pattern arrangement. Next to Bayer, perhaps the most popular CFA pattern in single-sensor digital motion -picture-imaging systems today is the striped-pattern CFA, which attacks the color-filtering concept in a different manner. With this CFA structure, the red, green and blue filters are arranged in vertical columns, wirb each featuring one continuous "stripe" of color. Red, green and blue are adjacent to one another, and this RGB pattern is repeated across the sensor. Panavision's Genesis was the first camera to incorporate a Super 35 -sized CCD imager with a striped CFA. With photosite dimensions measuring 5760 (H) ? 2160 (V), the system uses "macro cell" groupings of two rows by three column stripes - one of each color - ot its photosites to derive a tull-bandwidth 1920 ? 1080 RGB output. With this over-sampling structure, a single macro cell features two red, two green and two blue photosites to derive a single RGB digital-output pixel value. (See illustration below.) Global and Rolling Shutters Unlike film cameras, digital imagers do not require a physical shutter. However, all digital- im aging sensors perform some form of electronic shuttering - a prescribed interval of light gathering - by means of die sensor's timing circuitry, !here are two forms of electronic shuttering m today's digital imagers: global and rolling. A global shutter exposes and captures the entire imagers photosite array simultaneously. At the beginning ot a set timing cycle, the entire frame of the sensor's imaging area is exposed to the object-image being projected by the lens, and tliis image information is collected tor a set period. When that predetermined period has elapsed, the sensor stops gathering light information and outputs its data before resetting itself. This is analogous to die way a film camera exposes a single frame of film. With a rolling shutter, the interval of information gathering does not occur simultaneously. A rolling shutter actually "exposes" different portions of die sensor's frame at different points in time, rolling its scan progressively from top to bottom through the frame. Because a rolling shutter collects its exposure information row by row at different moments to create a single image frame, it can cause problems if you're photographing certain types of motion or working in rapidly changing light levels. "With a rolling shutter, motion can be a problem both horizontally and vertically," states Mahrer. "You can get sloping or tilting [distortion known as 'skew'] when panning, or when you have fast-moving objects in frame. A worst-case scenario would be taking a CMOS camera with a slow rolling shutter into a badly vibrating helicopter. The whole image would 'wobble' and look like you're using a lens made of JeIl-O. "Global shuttering certainly is much more advantageous," continues Mahrer. "Although global shutters are available in CMOS imagers, they are much more complicated devices unii more expensive to make." Partial-trame exposures can also occur with rolling shutters when there is a fast-changing lighting event in the scene - a flash ot lightning, for instance. With the rolling shutter, this fast lighting change might be captured between two partial trames oí exposure, with half of the bright, light-Hashed frame appearing at the bottom ot one frame and the second half of the flash captured at the top ot the next trame. An experienced eye can detect rhese partial frames. Why not use global shutters for oil imagers?1 The fabrication techniques required to include global shuttering in a sensor's function is fundamentally different between CCDs and CMOS imagers. By design, all CCD sensors are able to deliver global electronic shuttering with little impact to the imagers fill factor. To implement the same electronic-shuttering scenario in a CMOS imager, von must introduce several more transistors per photosite, and this greatly impacts the fill factor and, in turn, the size of the photosensitive area of the photositc. (The additional shuttering transistors must be placed in what would otherwise he an optically sensitive area ot each photosite.) "Rolling shutters [on CMOS sensors] cause the integration period for each row of photosites to happen at a slightly different period in time," says Jeff Zarnowski, chief technology officer ot Panavision Imaiiing. "Row O is read and reset to start integration for reading the next frame. Row 1 is then read and reset to integrate tor reading the next frame, and die process is repeated for all rows. Global shuttering allows the collected image-signal charge tor all pixels to be stored at once, but this requires an extra transistor to independently reset the pixel separately from the sense node, and this separate reset also causes an increase in noise of at least 3dB." Color Balancing By design, a digital -imaging sensor captures light differently than a film negative does. Film stocks are manufactured to produce a neutral color response tor a specifically balanced, illumination source (davlii^ht or tungsten), whereas a digital imager can be adjusted electronically to match its spectralresponse output characteristics to a variety ot illumination sources. For example, in order tor a digital imager to display a gray card as gray, it must balance the gain applied to the red, green and blue outjnil signals from the sensor. Depending on the color balance ot the light illuminating the scene, the amount of gain applied to each signal can vary greatly. The spectral response, sensitivity ami dvnamic ranime ot a sensor are also greatly influenced by the color properties/performance of the various filters, (IR, Optical Low-Pass und ITV) placed in front of it, as well as any CFA mask pigments that might be utilized. In fact, each of these color-affecting elements is chosen and optimized by manufacturers to meet individual requirements; the net result is a general colo r- temperature range in which the lowest overall gain to the red, green and blue signals is applied, yielding the least possible noise in the image. It is this sweet spot mat is generally regarded as the "native" color temperature of the sensor. "I like to think ot a digital camera's gains as a balance beam with the green value in the middle, and the blue and red on cither side oí it/ says Gait. "So, without changing the gain of the green, in order to color balance a camera to a certain illuminant, we li ave to pivot around the green value, increasing or decreasing the blue or red. If the camera in question is balanced to a tungsten illuminant, you will have very little blue light [collected from the sensor], and in order to balance that camera back to equal outputs, you'll need to bosst that blue signal, adding about 6dB of gain to the blue channel for any CCD or CMOS camera. That will give you a surieit of red and infrared light, which means you'll have to reduce gain in the red signal. "If that same camera is balanced for daylight, however," he continues, "you'll have plenty of blue signal, making the [image] appear less noisy, and will actually have to decrease the blue gain by about 3dB while increasing the red gain by about the same amount." Digital Processing/Image Handling The transformation of the captured image miormAtion m to a digital stgnal requires several sampling steps and A/D conversions, comprising what is otherwise known as image processing. The details of this processing depend on the vinous approaches utilized by individual camera-system manufacturers and the options incorporated into their sensor/camera designs. Galt notes, "With a CMOS device, which has an amplifier with a number of transistors, diodes and capacitors, you can get much greater variation in output than with a simpler CCD imager. One of the terms ior that variation is Fixed Pattern Noise. "Now, no sensors pixels are entirely flawless," he continues. "In fact, there are different classifications of imperfections in a sensor: Class 1, Class 2, etc. But even on a Class 1 sensor, there are bad pixels. Bad pixels, however, are very rarely completely 'dead.' They usually have a slightly higher or lower output. And when you look at signal-to-noise ratios, that's really what you're looking at. Noise is seen as a variation in the output. In digital image processing, however, you can deal with FPN. When you turn on the camera and do a 'black balance,' the camera is sending a signal through every one of these pixels, creating a map of the sensor and men storing a scries of corrections for the variations." A system's digital-processing and image-handling specs greatly influence the performance and characteristics of a given camera. Bit depth, color gamut, color sampling, sharpening, raw output vs. other signal-transmission interfaces, and data compression all affect the imaging characteristics of a parucuiar systcm. A separate article would be required to address these factors in detail, but their importance in the image-performance chain should be noted. The idea of a "raw" signal output vs. a processed one has become increasingly common in digital moti on -picture cameras. Am s digital cameras, Red's cameras, Thomson's Viper and Vision Research's Phantom line all feature raw-output options as part of their recommended post pipelines. Each manufacturer has its own definition of "raw," however. Panavision and Sony approach this idea differently, offering extended contrast-curve characteristics with their Panalog and S-log gamma curves, respectively. "The Alexa offers the ArriRaw format, which is essentially everything that's et im m g off the sensor in an uncompressed raw signal," says Kennel. "By storing that onto S. two or Codex recorders and taking it into post, you can resize or process that uncompressed data to get every last bit of detail out of it. The promise of a raw format is mat the post algorithms and codecs can improve over time." Digital-Output Standards Part of a digital-camera systems image processing is to structure output signals that can be delivered to and interface with the outside world. To help standardize digital -image specs, the major Hollywood studios joined together in 2002 to form the Digititi Cinema Initiatives. As stated on their charter, "DCIs primary purpose is to establish and document voluntary specifications tor an open architecture for digital cinema that ensures a uniform and high leve! of technical performance, reliability and quality control." Notes Thorpe, "DCI came up with a term I quite like: they describe the structure of a digital image format as a 'container.' There can he all sorts of things going on in the imaging section of a given camera system, and within this container there can be a variety of analog content that can be variable in sharpness and dynamic range, depending on the number of photosites, etc., but each manufacturer will have to hand over one or more ot DCI's established standards, which very specifically describe the digital structure to be delivered." Recording The last decade has seen the introduction of a number of new options for recording and storing video-signal-output information from digital motion-picture cameras. Historically, such recording and storage was tapebased, but today many tapeless, filebased recording mechanisms exist, including hard drives and Compact Flash cards. "Logically, using a very thin piece of plastic film and writing to tiny tracks that move over a cylindrical, spinning transport has become a mature technology," says Mahrer. "About eight years ago, Panasonic started the P2 line of cameras, which utilize solid-state memory cards as the recording mechanism. P2 cards have always been very fast, but initially they were only recording about 100Mb/s of data. The largest P2 card we have today is 64GB, and that can hold almost 90 minutes of DS-level 4:2:2 HD, recording using the AVC-Intra codec. However, the toolkits for encoding are much more advanced now, and the memory has become bigger and cheaper. We can do today with 100 Mb/s what DS did with 235 Mb/s about 10 years ago." Today there are almost as many solid-state recording media as there are digital-camera manufacturers. With che Alexa, Arri has adopted Sony V SxS memory-card format, recording Pro Re s 4:4:4, 4:2:2 I IQ1 4:2:2 Lite or Proxy files deBavered internally. Red recently unveiled its SSI) memory cards, which can store up to 256GB of RedCode raw-image data on a single card. (Red will continue to support the use of CF memory as well, but with the release of the Epic, it will phase out older, harddrive-based RedDrives in favor of SSDs.) Meanwhile, Panasonic continues its successful P2 line of memory cards and has recently introduced the higher-resolution codec AVC-Ultra, which can record H.264/AVC data rates of more than 400Mb/s in some of its 12-bit 4:4:4 1080p/2K/4K implementations. And with the recent announcement of its F65 4K camera platform, Sony also introduced its 1TB SR memory-card format, a RAID 5 format that can achieve data-transfer rates of up to 5Gb/sec. "That's one of the beautiful things about a non-physical format: you can change the file format, the codec and the hit rate, and the cards don't care," summarizes Mahrer. "There is no fragility of tape, no impact sensitivity, and you don't need a $100,000 deck to play the material back; you just plug it into your computer. Memory-based recording and high-efficiency codecs have really revolutionized the industry." Looking Forward With the current push for more single-imager CMOS sensors, and given Moore's Law's prediction for technological advances, it seems we have only scratched the surface of the digital imager's potential. Arri has gained strong support for its large-pixelsize and wide-dynamic-range Alev-III sensor in the Alexa; Red has a working 5K Bayer-pattern imager packed in a very small body with its Epic; and the new SK sensor used in Sony's F65 4K camera system is just around the corner. The future for digital imagers looks brighter, sharper and faster with each passing day. The first commercially produced CCD camera was developed by RCA and released in 1984. The camera featured three Vi" Frame-Transfer CCDs, each 403 ? 485 pixels. RCA won an Emmy Award for its development. Below: A sample of some of the tri-imager camera systems on the market today and their respective photosite structures. "The promise of a raw format is that the post algorithms and codecs can improve over time." In addition to inteviews by the author, both installments of this article incorporated material written by the interviewees, including John Galt and Larry Thorpe's presentation "Demystifying Digital Camera Specifications" and internal white papers by Galt, Thorpe and Jeff Zarnowski. Christopher Probst is a cinematographer and the magazine's technical editor ("Decoding Digital Imagers: Part 2, "p. 68). (c) 2011 American Society of Cinematographers [ Back To TMCnet.com's Homepage ]