DV vs. MPEG-2


What is the best conversion method for analog video?  DV or MPEG-2?

This is a question that has been debated by video professionals for the last 15 years.

DV was developed in 1995 as a codec that would allow video to be shot in and transferred to digital quite easily.  TV Networks, such as NBC and Fox in the US and the Miracle Channel in Canada, have been using one of the DV for their Standard Definition newscasts and some programming since the late-1990's.  A number of community access channels across the U. S. and Canada adopted the DV format as well.  DV is a broadcasting standard.  There have been three DV variants released: DV (Mini-DV just uses a smaller tape shell for this), DVCam and DVCPRO, aside from some minor technical details, all three variants used the exact same DV codec.

MPEG-2 was also developed in 1995 as an improvement of the MPEG-1 codec.  MPEG-1 had been developed to encode video onto a CD, however the quality of the video was about that of VHS. MPEG-1 was also used at higher encoding levels by some digital satellite/cable channels, however, this encoding scheme has been discontinued in favor of MPEG-2.  Sony used MPEG-2 compression on it's professional Betacam SX and MPEG IMX lines, as well it's consumer level Micro MV format.  However, unlike Micro MV, which sampled it's luminance and chrominance at a 4:2:0 level, just like DVD, Betacam SX and MPEG IMX sampled their color at 4:2:2, plus their data rate (Mbps) is much higher than Micro MV or DVD.  For this article these professional formats will not be mentioned anymore, as this discussion is about MPEG-2 at the 4:2:0 level, which is used for DVD, Micro MV and digital television broadcast. 

4:1:1 or 4:2:0?

In all digital video, the luminance channel (a.k.a the black and white channel) is represented by a 4.  This allows for the ratio's to be more easily remembered, otherwise we'd have ratio's of 1:0.5:0.5, etc.  The next two numbers represent the compression of the chroma channels (the color); 1 means that the color is sampled at a quarter of the resolution that the luminance is sampled at, while a 4 means that the chroma is sampled at the same level as the luminance.

DV encodes its luminance and color at a 4:1:1 ratio, which means that the red and blue channels are sampled at only a quarter of the luminance channel.  As DV's luminance is, in analog terms, 13.5 MHz; this means that both of it's chroma channels are sampled at a 3.375 MHz level.  This provides better resolution than NTSC or PAL broadcast signals, as combined DV's chrominance is 6.75 MHz.

At 4:2:0, MPEG-2 also encodes it's luminance at 13.5 MHz and chrominance at 6.75 MHz, but unlike DV, MPEG-2 co-sites, or averages it's chrominance channels into one stream.  To us an analogy, this is similar to how S-VHS played back it's video in the analog realm, where the luminance was in it's own channel/signal, but the choma channels were combined into one signal.  For MPEG-2 this means that the vertical resolution of the chroma is greatly reduced; and for analog to digital conversions this is not good, as the interlace video format of, VHS for example, has already reduced it's vertical resolution.  With interlaced video it is not clear whether 4:2:0 samples 2 side-by-side lines in a frame, or if it samples two side-by-side lines in a field, thereby skipping over lines (interlace is an old compression scheme from the 1940's, originally designed for analog television transmission, that split a frame of video into two, and sent one "field" ahead of the other). 

Now then, there is one exception for DV, since European PAL DV/DVCam uses 4:2:0.  4:2:0 seems to work better with PAL and SECAM's already reduced chroma resolution, however, no one really seems to know why, since PAL DVCPRO25 uses the 4:1:1.  As a result, PAL DVCPRO25 decks have the dubious distinction of having to convert from 4:2:0 to 4:1:1 for playback.  (For more information I would recommend checking out Adam Wilt's DV FAQ.)

Compression Ratio

For comparison's sake, uncompressed digital video is usually stored at a compression ratio of 2.1 or 3.1; most sources give 2.1, but some indicate that it is as low as 3.1.  Digital Betacam (Standard Definition) and XDCAM (High Definition) tend to shoot at this level, and some editing, especially for theatrical release, occurs at this level, although uncompressed uses a ton of memory

By comparison, DV compresses its video at a 5.1 ratio, whereas MPEG-2 compresses at a 10:1 ratio.  While 10:1 is okay for final delivery, for editing this requires a ton of computer power, and causes generational loss.

What Does This Mean For DVD Transfers?

When you put this together, if you capture your video just in MPEG-2 4:2:0 (which is what most people do with set top DVD recorders and the $20 dollar department store capture devices), you are transferring in low quality.  By capturing in DV, then going to MPEG-2, I am able to capture in a higher quality, and then put the video onto DVD.  This is analogous to, in the 90's, shooting your family's home movies on VHS-C and then making copies for family members onto VHS from the VHS-C.  You would not use the VHS-C to make a good copy for Grandma, since you would be losing quality: the better way would be to transfer the VHS-C to a Hi8 tape, and then use the Hi8 tape to make another VHS copy.  So you end up with a video that looks just as good, if not better than the original tape.

Software or Hardware Encoding

Also, for formats such as U Matic, Betamax, VHS, S-VHS, Video8 and Hi8, the video is recorded on the tape in a composite format, which is already a compressed analog signal.  Most set-top DVD players only capture analog video by the yellow composite RCA or RF inputs, and this leads to video that does not show off its full potential.  While set top DVD recorders do use hardware encoding, it is usually the cheapest hard ware out there (usually equivalent to a camera that has a 1 CCD for capturing video).  Unless you plan on watching your DVD via your DVD player's composite connection, you'll find that DVD's made in this fashion have horrible chroma problems when played back by S-Video, Component or HDMI.

The various USB converters that are sold in department stores for around $20 dollars tend to use software encoding more often than hardware encoding.  Software encoding tends to soften an image and never gives you the quality that you are looking for.  These types of converters are better suited for just capturing TV shows from analog/digital cable than for converting analog video to digital for your precious memories.

Here at Trevor Thurlow Productions one of my main converters is the Grass Valley Canopus ADVC-300.  It is a high-quality, broadcast level analog-to-digital DV converter, that uses hardware to convert the signal to digital before sending it to the computer for burning to a DVD.  I know that there is one company out there that tries to say that it is a poor converter, however, I have tried several converters over the years (including set top DVD recorders) using the DV and MPEG-2 codecs, and I have never found a better converter than the ADVC-300.  The chroma's very stable (especially with video from VHS and U Matic) with no chroma bleed, and the converter delivers a very sharp picture. 

Even for editing I find with my own projects that I get a better quality image when I'm using footage (even footage from DVD) that's been captured and imported with DV.  With DVD, if I just copy the VOB or use another program to convert the MPEG-2 4:2:0 image, I find that I end up with very choppy video that also looks like it came from a VHS recorded in the SLP mode.  This is due to MPEG-2 recording in the inter-frame GOP mode, where it only records (for example) frame 1, and then frame 5, and then it relies on information from both those frames to create the missing frames.  So everytime that I apply an effect to an MPEG frame, the computer has to uncompress the video, fill in the missing frames, apply the effect, and then recompress the image: this leads to a large loss of visual information and sharpness.  DV, on the other hand, records each frame and compresses it (sort of like how a JPEG is compressed) in intra-frame compression, so it has access to every frame and the rendering does not cause a major quality drop to its image.