Know Your Formats, Part 2

Figure 1 redux: Formats and features.
Click here for a larger image.

This month, we''re tackling high-definition video formats. In the first segment, I discussed what aspect ratio is and why you care about it. In this issue, I''ll discuss the remaining items of the features table—which, through the marvels of modern digital technology, I''ve duplicated here.

Let me breeze through the next few definitions then circle back and discuss the individual formats. Of course, "codec" refers to the technology used to compress the video. All affordable HD acquisition formats use compression; otherwise the digital files would be too large for convenient on-camera storage. As you can see, there are three codecs used in this class of camera: DCT-based frame compression, MPEG-2, and AVC/H.264.

"GOP structure" refers to techniques used by the codec during compression. Technologies such as MPEG-2 and AVC/H.264 have three frame types to work with: intraframes (also called I-frames), bidirectional (B) frames, and predictive (P) frames. However, as technologies such as AVC-Intra illustrate, camera vendors don''t have to use all three frame types when choosing their formats; they can choose to go I-frame only.

When a vendor uses all three frame types in their format, it''s called a "long-GOP" format, with "GOP" standing for Group of Pictures. Each group of pictures starts with an I-frame and has multiple P and B frames within.

Data rate is the amount of data per second used to store the video stream. All compression technologies are lossy, so the more you squeeze, the more you lose. Higher data rates are generally better, but you can''t discuss data rate without considering codec and GOP structure because they interrelate so significantly. We''ll explore this in more detail below in the XDCAM HD-vs.-DVCPRO HD comparison.

Understand that there are few absolutes in the realm of compression. For example, AVC/H.264 is without question a more efficient compression algorithm than the older MPEG-2 format, so in theory, that makes it better. However, long-GOP AVC/H.264 is currently not supported by the Adobe family of editing products, so you can''t edit it in Premiere Pro or After Effects.

In addition, because long-GOP AVC/H.264 is a more advanced technology, it''s tougher to decompress during editing, so even in the few editors that can edit the format natively—such as prosumer editor Pinnacle Studio—you need a very powerful computer to achieve responsive editing. The other approach, used by Apple in Final Cut Pro, is to convert the AVCHD video into ProRes, which is easy to edit, but requires lots of disk space.

In addition, although the AVCHD spec enables data rates of up to 24Mbps, the highest data rate I''ve seen implemented is about 14Mbps, which lets vendors boost the recording time on their camcorders at the expense of quality. I also haven''t reviewed an AVCHD camcorder that had sufficiently large imagers for pixel to pixel capture, or lenses of anywhere near the quality of the average $4,000 consumer camcorder.

All told, although AVCHD is undoubtedly a better compression technology than MPEG-2, as implemented, it has delivered consistently poorer quality video with significantly fewer editing options. When the first true professional AVCHD camcorders come to market, you''ll have to look at them with a totally fresh eye.

Panasonic released the AVC-Intra-based Varicam 3700, the AJ-HPX3700, at NAB this month.

The lack of absolutes is also evident in the debates regarding the virtues of Panasonic''s DVCPRO HD and Sony''s XDCAM HD. I thought Panasonic used I-frame only MPEG-2 in its high-definition DVCPRO format, but here''s what a Panasonic white paper had to say:

In essence, DVCPRO HD uses the same compression technology as DV at data rates of up to 100Mbps, according to the same white paper. In contrast, according to Sony''s white paper, XDCAM HD uses long-GOP MPEG-2 at up to 35Mbps. (In February, Sony announced a new implementation of XDCAM HD, which employs 4:2:2 long-GOP MPEG-2 at 50Mbps. This technology makes its debut in the PDW-700 camcorder, available this month.)

Panasonic proponents argue that long-GOP formats can''t deliver the quality of their DCT-based algorithm, and are much less responsive during editing. Shooters drinking the Sony Kool-Aid argue that long-GOP MPEG-2 is so much more efficient than I-frame-only DCT that it delivers the same or better quality at one-third the data rate, and that today''s multicore computers are easily sufficiently powerful to work edit responsively with MPEG-2 (and if you need more speed, use an intermediate format). They also criticize DVCPRO HD for requiring three times the storage space, which is a particular issue for camcorders capturing to solid-state memory.

Who''s right? I dunno; the video produced by both camcorders looks wonderful to me.
The point of this newsletter is to define the terms and identify talking points, not pick a winner.

That said, looking back at NAB, it''s tough not to get excited about the AVC-Intra based camcorders that Panasonic showed, with full-raster high-definition recording—1920x1080 pixels for a 1920x1080 display image, and the more efficient AVC codec replacing DCT. It''s also not hard to see the ultimate (although not imminent) demise of HDV, which doesn''t have the resolution, codec, or data rate to compete in the long term.