Multimedia
Blogs
Forum
Affordable HD
Whitepapers
Newsletters
Events
Advertisers
Blogcast
MillimeterFilm-to-Tape Goes Digital: Tackling Telecine Issues for DTV and HDTV (Part II)
Mar 1, 1998 12:00 PM, D.W. Leitner
This is the second part of a two-part series on film-to-tape trends and technologies. Part I (February 1998) encapsulated the historical developments in film-to-tape that brought production and postproduction industries to the threshold of HDTV and DTV. This article analyzes the issues surrounding the shift to digital TV. The author, D.W. Leitner, a contributing editor to Millimeter, is also an editor, DP and director, with credits in film, TV, and documentaries.
In a nutshell, both the Philips Spirit DataCine and the Cintel C-Reality high-rez telecine seek to make possible what Philips cutely calls "film-2-data." The idea is simple on its surface: the telecine as a high-res 2K film scanner, with 2K x 2K film data frame buffers and 14-bit internal signal processing.
Typically, a 2K scanner samples 2048 pixels per scan line across the width of a 35mm film frame. Although Kodak says a 4K scan (4096 pixels) is needed to capture full 35mm detail, 2K scans result in data files one-fourth the size of 4K scans. Since smaller 2K files are considerably easier to process and store, they're popular for motion picture effects.
In the case of Spirit and C-Reality, the basic idea is to capture 2K scans as scaleable high-resolution .DPX (SMPTE 268M) or .TIFF data files, then channel them along high-speed data networks to computer workstations for effects, editing, or "off line" downconversion to fixed video formats: analog 525-line/59.94 Hz or 625/50; 10-bit serial digital 8:4:4, 4:4:4, or 4:2:2, with alpha channel if desired; or DTV 1080-I (shorthand for SMPTE 274M, 1125-line interlace DTV with 1080 active lines). Your choice of 4:3 or full 16:9 aspect ratio, of course.
As far as TV is concerned, 2K represents high oversampling, so downconverting 2K files to 525/625-line images eliminates all telecine noise, aliasing, beating, moire patterns, even CCIR 601 4:2:2 matrix artifacts for a noticeably more "filmic" look.
Since horizontal and vertical aperture correction are carried out internally at 2K "film resolution," final images have a tangibly sharper feel. Anyone who has seen Tape House VP and telecine director John Dowdell's Spirit DataCine demos in New York will readily confirm this.
Dowdell's New York SMPTE presentation last June of HDTV 1080-I transfers from 16mm and Super-16mm projected by 1125-line Barco projectors at the "Show Me New York" theater on East 58th Street (the first commercial HDTV theater) upstaged, if not trounced, a notorious CBS/Sony technical presentation purporting to demonstrate the inferiority of Super 16mm for HDTV production. [The CBS test was shot on the Seinfeld lot, but CBS wasn't joking: a CBS corporate directive mandated teleproduction in the DTV era be 35mm or HDTV video. Many who witnessed the test results, however, had a problem with that.] Dowdell's video-projected images were shockingly film-like, of clear-cut commercial theatrical quality.
Dowdell, who in mid-'96 inaugurated the first U.S. commercial Spirit facility, channels his .DPX files next door (to Tape House Digital) to a Discrete Logic Inferno supported by an SGI Onyx 2 and RAID disk array.
"The magic of scanning," Dowdell has said, "is that you can stay in computer files [instead of video], because the Spirit is a big computer too."
But big computers, even as they solve problems, seem to underlie them too: take the hair-pulling incompatibilities between computer RGB color space and video YUV color space, or computer square pixels and video rectangular pixels.
But one alleged problem is a tempest in a teapot. Spirit's critics deride its output as "pseudo 2K" since its actual Academy aperture 35mm scan is 1920 x 1445 pixels (not 2048) and color is subsampled by half (960). Spirit instead uses its processing muscle to transform 960 pixels/color to twice resolution, fusing them with the sharper luminance pixels and interpolating the composite to a full 2K RGB scan.
>From there, Spirit employs a real-time 2K data reformatter and framestore called SPOT (Spatially Processed Optical Transformations) to process more than seven gigabits/sec., enabling pans, zooms, resizings, rotations and video downconversions.
While horizontal resolution is limited to 1920 samples, Spirit can sample up to 1792 vertically, naturally all square pixels since it's a computer. A tall Cinemascope frame, for instance, requires a vertical resolution of 1665 pixels.
Cintel, in response, contrasts C-Reality's Academy 35mm sampling of 2048 x 1536 pixels per color for as "true" 2K RGB. Of course, no one's seen C-Reality's output, at least commercially speaking; and while it's hard to imagine how to gild the lily of Spirit's breakthrough quality, C-Reality's output will no doubt impress.
Still, it's hard to avoid the feeling that the 2K controversy has more to do with politics of the decimal system than psychophysics of perception.
Cintel also makes a point of the classic advantages of flying-spot scans over fixed linear-array CCDs: in this case, the need for less digital processing. Spirit with its linear array can't scan a standing still frame, while C-Reality can easily park a frame in its skid plate and scan continuously. Further, by electronically changing the size and rate of its raster scan, C-Reality pans, zooms, resizes and rotates-even changes image line-structure or resolution-without resorting to any digital interpolation.
While Spirit can only scan a moving frame and then play it back from a still store, however, its transport is so sensitive that it offers a single step mode with forward and reverse.
Scanning flexibility has other advantages. One is speed. C-Reality can scan, forwards and backwards, 1125-line HD up to 30 film frames/sec. and 525/625-line formats up to 125 fps, plus double and quadruple speeds.
C-Reality also offers a 720 x 576 data output mode in which film can be scanned and transferred to video faster than normal, perhaps at the 4x rate of MPEG-2 or 50 Mb/ sec 4:2:2 DV. Spirit can run 2 to 57 fps, forwards and backwards, with a high speed search mode that features a full-frame, correctly oriented image up to 150 fps in 16mm and 75 fps in 35mm.
What about 2K transfers? Big computers have big appetites, but sucking a thick datashake through a reed-thin straw takes a while. Real-time 2K transfers require moving and storing data at almost a gigabyte a second, but no practical network or storage handles that kind of volume.
The Spirit itself can output 320 megabytes/sec., but current fast HIPPI interfaces and Fibre Channel limit 2K transfers to 90 mB/sec. or about 6 frames/sec. for Academy 35mm. That limit pertains to C-Reality equally.
Still, that's 30 to 90 times faster than typical 2K film scanner speeds, whether laser or CCD, which take from 6 to 16 seconds per frame. And Philips at IBC '97 in Amsterdam announced a partnership with Silicon Graphics to develop a high-speed data transfer architecture, the Phantom Transfer Engine, for eventual real-time film-resolution playout.
As a hybrid telecine/scanner, both Spirit and C-Reality feature SDTV (standard definition) and HDTV as well as 2K data output modes. While Spirit and C-Reality operate in 2K mode internally at all times, SDTV and HDTV outputs in real time are available for all 16mm and 35mm film formats (if data networks could handle higher data rates, Spirit's maximum 2K transfer rate would be is 23.1 fps, not quite real time due to the density of scanning required).
Multiple modes hopefully promise multiple income streams for these expensive machines: Spirit, at well over $1 million, approximately twice the price of an Ursa Diamond, as compared to C-Reality, about 30 percent more than a Diamond.
C-Reality's price tag amazingly includes an integrated Stealth controller and real-time color processor for SDTV, HDTV and 2K, although Cintel continues to share interface protocols with manufacturers of high-end da Vinci, Pogle, and Copernicus systems.
But at the Tape House in New York, premium hourly rates haven't been a show-stopper. John Dowdell likes to cite the recent experience of documentary director Ken Burns, who transferred his entire PBS series on Lewis and Clark on the Spirit for the same overall cost as an Ursa transfer, since the Spirit's twice-rate was handily undercut by its speed of set-up and halved transfer time, due to the Spirit's near-perfect match to negative and interpositive film, requiring minimal tweaking and almost no secondary correction.
To sweeten the bargain, Burns got an HDTV master from which to downconvert superior NTSC and PAL submasters needing no additional telecine or color correction.
Hold the presses: the term "film chain" may not be dead, after all. After a year of rumors, Sony tossed its own hat into the multiresolution, 2K datacine ring at IBC in Amsterdam.
And guess what: it's an intermittent, sprocket-driven projector pointed into a camera. In this case, the camera is an imaging block containing three 1920 x 1035 area array 16:9 CCD chips for RGB, plucked straight from their new HD camcorder. (1035 active lines hails from the analog 1125-line interlace HD standard with rectangular pixels; 1080 versions with square pixels are imminent, Sony says.)
Illumination is xenon, flashed in bursts of "exposure" by solid state light valves between pulldowns. Exposure control and primary color correction are handled by trimming light levels rather than electronic color correction.
Small capacitance sensors in the gate track 35mm perfs and optically on-the-fly reposition each frame by tilting a pair of glass surfaces that act as prisms-a telecine version of Sony's optical Steadishot from consumer camcorders. Image rotation is accomplished by rotating the imaging block, zooms by an actual zoom lens. Pricing and availability are not available yet.
What will they think of next? NAB 1998 should provide some eye-opening possibilities.
Continue the discussion on “Crosstalk” the Millimeter Forum.
