Digital Video

Digital video started in the late 1980s as a means of improving recording quality in an analog world, so all the basic details of digital formats, at least for standard-definition television, descend directly from their analog forebears.

Digital composite recording digitized the existing analog composite waveform directly. The signal was sampled at four times the subcarrier frequency (4fsc), 14.32 MHz for NTSC signals. Digital composite was designed as a drop-in replacement for analog composite recorders andeven though they worked superbly for that purposethey suffered because they recorded a composite signal in a world that was rapidly moving to component video for postproduction purposes. Except for certain niche uses, such as tape-based time-delay, you won't see many digital composite recorders these days.

Digital component recording followed the Betacam and MII model of recording a luma signal and two color-difference signals separately, keeping them pure for postproduction manipulation. Almost all standard-definition digital component recorders conform to some variant of the ITU-R BT.601 specification (formerly called the CCIR-601 specification, back when the International Telecommunication Union was still the International Consultative Committee for Radioand Television).

Sampling

The "601 spec" calls for a sampling frequency of 13.5 MHz, yielding 720 luma samples per scanline. (There's an optional 18 MHz sampling rate in the spec, giving 960 samples per scanline, but very few recorders implemented it.) At 13.5 MHz, the limiting horizontal resolution is 540 TVl/ph (assuming a normal 4x3 picture aspect ratio), although in practice 500 TVl/ph is a better figure.

More Info

There's always some low-pass filtering of the signal prior to recording to prevent aliasing artifacts. Charles Poynton's books A Technical Introduction to Digital Video and Digital Video and HDTV Algorithms and Interfaces are great references for this sort of detail.

Note

Both NTSC and PAL-compatible signals have scanlines taking about the same amount of time to display, so both formats are captured with 720 pixels per scanline in 601 digital video.

Tip

The active picture area in the NTSC standard spans only about 704 pixels in the 720-pixel 601 line. Some digital cameras respect the 704-pixel active area and record thin black bars on either side of the picture in the remaining pixels; others record an active picture across the entire line. If you shrink a picture down in FCP, you may need to crop the left and right edges slightly to hide the black bars.

That 720 samples or 500 TVl/ph figure is an upper limit for what 601 will resolve; it doesn't mean that anything recorded on a 601 format will show that resolution. A low-end DV camera or an off-air analog signal transcoded to 601 digital may yield considerably lower resolution, and a 35mm film transfer or an HD downconvert will have a hard limit of 720 samples but may appear sharper than the numbers allow; the perception of sharpness is more complex than a simple pixel count can capture.

More Info

See www.adamwilt.com/TechDiffs/Sharpness.html for details (pun intended).

Pixel Aspect Ratio

The 601 spec, you'll notice, results in nonsquare pixels. Were the pixels square, a 4x3 720-pixel-wide picture would be 540 pixels (scanlines) tall, but "NTSC" 601 has 486 scanlines captured, whereas "PAL" 601 has 576. "NTSC" 601's pixels are slightly wider than they are tall; their pixel aspect ratio or PAR is 0.9. "PAL" pixels are slightly taller than they are wide, with a PAR of about 1.07. Keep in mind that PAR is measured as the ratio of height to width, whereas picture aspect ratio is width to heightyet another odd detail to trip up the unwary.

Tip

FCP avoids confusing PAR numbers, instead providing descriptive terms: "Square," "NTSC CCIR 601 / DV," and so on. PAR is normally set up for you when you select a capture preset or a sequence preset.

Color Subsampling

The color-difference signals C_R and C_B are normally subsampled at half the rate of luma; the resulting ratio of luma to chroma samples is called 4:2:2, apparently in a holdover from the 4fsc nomenclature.

There are variants to the 4:2:2 sampling structure. Some formats, like DVCPRO25, record 4:1:1 colorchroma samples at one-quarter the rate of luma. This may not sound like much, but it's still enough for about 135 TVl/ph chroma resolutionbetter than NTSC allows.

4:2:0 is used in PAL-compatible DV and DVCAM, and in MPEG-2 as seen in DVB (digital video broadcasting), DVD, and HDV. In 4:2:0, C_R is sampled at half the luma rate on one line, C_B on the next, and vertical filtering is used to reconstruct the image. 4:2:0 is problematic in interlaced video since the filtering has to happen on a field basis; unpleasant sawtooth-like artifacts are often seen on brightly-colored diagonals; and it's hard to perform color-oriented postproduction steps such as chroma-keying using 4:2:0that's why "PAL" DVCPRO uses 4:1:1 even though other "PAL" DV25 formats use 4:2:0.