poynton@vector.Sun.COM (Charles Poynton) (03/10/89)
Pardon me, I'm a little behind on reading news, and just recently caught the discussion in Comp.graphics about film data rates. Sony recently introduced their digital HDTV videotape recorder. It records one hour of studio-quality 1125-line HDTV on one-inch open-reel tape, at a data rate of about 1.188 Gb/s [Poynton nomenclature, little-b means bit]. The capacity of the tape reel is about 534 GB [Poynton nomenclature, big-B means byte]. The video standard conforms to the ANSI/SMPTE 240M HDTV Production Standard adopted in the U.S. three weeks ago. [Many disclaimers distance 240M from transmission standards, currently a wide open issue.] The analog standard has 1125 total lines, 2:1 interlace, 60.000 Hz field rate, a picture aspect ratio of 16:9, and 1035 picture lines. The fraction 48/55 of each (total) line contains picture; the rest is horizontal blanking overhead. The standard specifies the new SMPTE "C" colourimetry, has a new "tri-level" sync signal, and tightly defines the "luminance transfer characteristic" (since the term "gamma" is now officially deprecated in the television community). Sony of course needed to digitize the HDTV signal for their HD-DVTR, but there is currently no standard sampling frequency for HDTV. Sony picked 2200 times the line rate (giving 74.25 MHz). The signal is recorded with full bandwidth luminance in eight bits (i.e. 74.25 MB/s), and half bandwidth U (scaled B-Y) and V (scaled R-Y) colour components in eight bits each (i.e. 37.125 MB/s each). No further data compression is necessary, and the aggregate data rate of the recorder is 148.5 MB/s. Data is not guaranteed to be recorded absolutely without error, of course. Forward error correction codes are added prior to recording to enable detection and correction of short error bursts. The data is shuffled prior to recording so that if a severe error burst (such as a tape dropout) causes massive data errors, the offending samples will be widely disbursed spatially across the picture. Each isolated error sample is then "concealed", that is, replaced by a synthetic value which is interpolated from surrounding samples by a reasonably sophisticated spatial filter. About fifty generations (record/playback) can be recorded before any artifact becomes noticeable to even a highly-trained observer. Charles A. Poynton <poynton@sun.COM> Sun Microsystems, Inc. 415-336-7846