Monday, October 31, 2005

Eyes on the 1080p Prize

In the best of all possible HDTV worlds, 1080p would be a firm reality.

We would, that is, have TVs to which true 1080p source material could be input and displayed as such: video frames having 1,080 rows of pixels, each horizontal row containing 16:9 x 1,080 = 1,920 pixels.

A video frame? It's what you see when you hit pause on a DVD player, a single frozen image, with no movement. Moving video is really a succession of such frames, displayed one after another so fast that your eye perceives motion.

For 1080p reception, each 1,920 x 1,080 input frame of video would be received and presented fully intact, which is what the "p" (for "progressive scan") means. As transmitted, it would not be subdivided into two "i"-for-"interlaced" fields, with only the odd-numbered pixel rows present in field #1 and only the even-numbered rows, 1/60 second later, in field #2.

A pixel row is also called a "scan line." When the latter type of 1080-line scanning is done, with each video field containing only alternate scan lines, not the whole image "raster" at once, the transmission method is 1080i, not 1080p. 1080i scanning typically produces 60 fields per second, which comes to 30 frames per second.

Over-the-air high-definition television broadcasts use either 1080i or 720p: 720 rows of pixels, 16:9 x 720 = 1,280 pixels per row, progressive scan, 30 non-subdivided frames per second. 1080p is not supported over the air; it doubles the number of pixels transmitted per second by 1080i, requiring TV channels to hog unacceptably high bandwidth.

Unless ... a more efficient digital video compression scheme is used. At the time the standards for HDTV broadcasting were set, MPEG-2 was the best compression method around. Now we have MPEG-4, which squeezes digital video down far more compactly. We also have VC-1, based on Windows Media Video (version 9). Both are supported by the two emerging competitors for high-definition blue-laser DVDs: Blu-ray and HD DVD. Blu-ray and HD DVD could offer us true 1080p discs without paying a huge price in data rate or storage capacity.

The blue-laser DVD gurus, however, have as yet committed to no more than 1080i. For one thing, there are few if any TVs that can input 1080p — even though a number of new models can display it. (See, for example, Sony's Groundbreaking New SXRD RPTVs.) There seems to be no agreement on how to copy-protect 1080p, for another thing.

There are yet subtler problems. For example, true 1080p promises greater visual detail than merely "deinterlacing" 1080i would imply. But how to get there?

Most newfangled TVs need to deinterlace 1080i to some progressively scanned form — unless they are based on old-fashioned cathode-ray tube technology with actual scan lines traced out by an electron beam, that is. One conceptually simple form of deinterlacing is to "borrow" the pixel rows that are not present in a given field from the other field in the same frame. Another is to interpolate the pixels between those which are actually present: to "guess" what they would contain, if they were really there.

There are more sophisticated approaches. One of them is to notice that the video program originated in film, and use so-called 3:2 pulldown compensation. The TV figures out which other fields of which other frames really came from the same film frame — owing to the fact that film unwinds at a mere 24 frames per second, not 30 — so it knows how to fill in the missing pixel rows in a given field with total precision.

Unless deinterlacing can leverage such special knowledge as that, the interpolated or borrowed or calculated pixels constitute "false" information. What's worse, they don't really count as "new" information at all, for they don't contain any extra detail. For that reason alone, true 1080p can improve on the "false" detail present in deinterlaced 1080i quite considerably.

But there's another reason as well, yet more subtle, why we ought to prefer 1080p. In interlaced scanning, vertical resolution — the number of horizontal lines actually visible in the image — is reduced to avoid interline flicker.

So if there are 1,080 rows of pixels, there may be only 70% of that number of visibly distinguishable lines of image content: say, 756 lines of vertical resolution.

"When adjacent lines in the frame (which are transmitted and displayed one field-time apart) are not identical," says this web page, interlaced scanning will produce a flickering effect. If the field-time is 1/60 second, you can get a tiny detail (say, white) in the first 1/60 second that disappears (turns black) in the second 1/60 second. Enough of those details coming and going, and the display as a whole will have an annoying 30-Hz flicker.

That's bad, so the camera simply filters out the smallest details present in the vertical dimension of the image. 1,080 rows of pixels then produce only, say, 756 lines of vertical resolution.

But if you know the image is going to be captured, recorded, transmitted, received, and displayed at 1080p, you don't have to filter to avoid interline flicker.

Another thing 1080p offers is a way to avoid the 3:2-pulldown problem entirely.

When film at 24 frames per second is transferred to video at 30 frames per second, the film is "pulled down" — advanced — in a herky-jerky way such that some film frames contribute their visual information to three video fields, while other film frames contribute to only two video fields. That's 3:2 pulldown, also called 2:3 pulldown.

If such film-transferred video is not deinterlaced with totally accurate 3:2-pulldown compensation — TV user menus often call it "film mode" — then those video frames whose two fields come from two different film frames can produce "jaggies," or worse, on the screen. Fie!

1080p can avoid that problem entirely by recording video at the same frame rate as film uses: 24 fps. That way, there's a 1:1 correspondence between film frames and video frames. (To avoid the flicker sometimes associated with such a low frame rate, the TV can display each video frame three times in each 1/24 second frame time, for an effective 72 frames per second.)

In the real world, we don't even know as of now whether there will ever be any true 1080p DVDs, much less whether they will be made with all the vertical resolution 1080p "ought to have." They might be filtered in deference to CRT-based interlaced HDTV displays.

But we can always hope true 1080p DVDs are on the way, with no vertical filtering!

No comments: