MPEG encoding, DV to MPEG transcoding, and he doesn't even mention Cube.
[WARNING: High technical content]
--------------------------------------
MPEG -2 vs. DV: An Entertaining Show
Jeff Sauer
08/30/1999
Video Systems
Copyright 1999 by Intertec Publishing Corporation, a PRIMEDIA Company. All rights reserved.
The video industry and format wars seem to have a special relationship. Once again this year, post-NAB pundits warned of the looming battle between the Sony-led Pro- MPEG interoperability forum and Panasonic's DV50-based interoperability forum. With partner companies lining up on both sides, forming alliances, and molding strategies in their respective NAB pavilions, this might seem like the next big winner-take-all, knock-down, drag-out battle for control of the digital future.
Fortunately, unlike past analog tape format wars that left wrong-side early adopters bloodied and nursing obsolete equipment, this one likely will see fewer civilians caught in the fray. More like spectators at a heavyweight prizefight, video equipment purchasers and users handicapping this battle should be able to pick a "Broadcast and Professional" company division to root for -- and more important, pick tools that suits their needs -- and let most of the heavy hitting stay in the ring.
In the corner to your left in the MPEG -2, 4:2:2 color-subsampled trunks, you have Sony. And, in the corner to your right, in the DV-50, well, 4:2:2 color-subsampled trunks -- is Panasonic. Both contenders are I-frame only, 50Mbits/s streams, and both compression formats are based on the Discrete Cosine Transform (DCT). In fact, it might be a little tough to tell the two apart if not for the usual pre-fight grandstanding.
As much as some of the hype begs for a format war, and though it might make good press to call it one, prizefighters and promoters have a knack for embellishment. Where past format war veterans, such as Betacam SP, are used throughout the video production process -- from acquisition and production to post-production to distribution -- this MPEG and DV debate primarily is, although not exclusively, about just the middle piece. While it's simplistically stated, it's largely about outgrowing Motion-JPEG.
"Hold on a minute," say fight fans. " MPEG -2 is the format of DVD and DTV, as well as the logical choice for archival, given its excellent temporal compression." That is absolutely correct. The role of MPEG -2 in the digital distribution future is not in doubt, nor a point of contention between the two apparently warring camps. Outputting compliant MPEG -2 bit streams is a must for almost any finishing tool, including DV-based equipment.
That reality introduces the possibility of working with MPEG from start to finish -- production to distribution -- with a comfortable and appealing air of neatness. Unfortunately, the math is a little trickier than that.
It's easy to miss the distinction between the MPEG50 (50Mbits/s, I-frame only, MPEG -2) proposed by the Pro- MPEG proposed by the Pro- MPEG forum and the MPEG that will be sent out over the air or burned to DVD. Further, the current MPEG solution for acquisition, Sony's Betacam SX, is yet another type of MPEG (18Mbit, with an IBIB frame order). While all are similar in I-frame coding, different bit rates and different GOP (group of pictures) structures require conversions along the way. And, that puts the marketing machines in high gear on both sides.
An MPEG for everything? MPEG -2 is a scaleable standard. Unfortunately, that is a tough concept to pin down and ultimately means different things to different people, including those in the Pro- MPEG forum. Easiest to appreciate is that, like M-JPEG, MPEG can encode at a variety of compression bit rates, making it an attractive M-JPEG replacement at the prosumer level for digitizing analog footage. The MPEG -2 standard also is robust enough to handle a variety of GOP structures -- I-frame only, IBIB, and IBBP -- while still using the same basic macroblock-based coding.
Yet, MPEG 's main appeal remains its wonderfully effective use of temporal compression to greatly reduce bit rates for efficient distribution. By removing temporal compression, MPEG50 becomes a DCT-based compression method that fundamentally is similar to DV-50, and even M-JPEG. And, it is different from the 4:2:0 IBBP MPEG -2 used for distribution.
"But surely," shouts the MPEG corner, "sharing an acronym means an easier transcode from one MPEG to another." Mathematically, they're on to something. MPEG scalability also has been shown to extend to transcoding an existing MPEG stream into different bit rates without a full decompression to ITU-R BT.601. And, the MPEG fans stand and cheer for their hero, who appears to have scored a knockout blow.
MPEG mathematicians are tapping into MPEG 's ability to carry extra reference data, a "transcoding data set," in addition to the simple video and audio information. By including history information about how the video was originally encoded, MPEG transcodes can avoid some instances of digital generation or statistical loss. For example, by understanding what macroblock boundaries were originally used during encode, a transcode might avoid concatenating artifacts from DCT rounding errors.
That's some impressive footwork with the math, deserving of the ringside cheers. However, the crowd in the other corner says that the punches never landed and that their favorite is still ahead on points. After all, the ability to effectively use the history information is far from categorical.
First, while the transcode works on I-frames without a complete decode to 601, inserting the important bit-rate-reducing B- and P-frames does require full decompression and recompression. MPEG -to- MPEG transcoding is most effective when you start and finish with IBBP streams, as would be the case when resizing an 8Mbits IBBP stream to, for example, 3Mbits/s to fit a broadcast bandwidth restriction, or to fit onto a single DVD using variable bit-rate encoding. While quality, in that case, would degrade from the greater compression, history information could limit quality loss by avoiding concatenation artifacts. Going from I-frame only MPEG50 to IBBP MPEG , while still sharing some of the value, loses an important level of effectiveness in the B- and P-frames.
Moreover, if the editing process adds compositing layers or non-hard-cut transitions, new media gets created, overwriting the original encoding history information. Further still, some video material will simply encode better going the full recode route; for example, if the reduced bit budget would calculate a vector rather than a color region.
Ultimately, the proof of MPEG transcode has yet to be realized. While it's worth exploiting any advantage to maintain quality, leaving those tricky math decisions to inexpensive and often-inflexible silicon is unexplored territory. MPEG50 technology has yet to produce a real product and no MPEG50 acquisition products have even been announced.
For current acquisition, Sony's Betacam SX uses MPEG -2, but its IBIB structure produces an 18Mbit/s stream. To use 18Mbits Betacam SX in an I-frame only MPEG50 environment, Sony simply converts SX's B-frames to I-frames, bringing aggregate video bit rate up to roughly 30Mbit to 40Mbit, depending on video content. Sony has shown excellent video images at that bit rate, but it is well shy of the 50Mbits milestone both companies offer as a quality standard. The DV crowd can make a solid argument that whatever ground might be gained by an MPEG -to- MPEG transcode, will be lost in the reduced original bit rate of Betacam SX as compared to DV-50, yet SX is likely to be better than DV-25.
Currently, there is no standard for the sharing of that potentially valuable history information between different equipment and different companies, though two have been proposed. While MPEG 's clout as a distribution format should breed great interest in creating such a standard, with no standards body stepping into the action, this prizefight has a long way to go.
Many companies are agnostic and participate in both the Pro- MPEG forum and the DV world with products that support both formats natively. It's also not out of the question that someone from the ranks will learn to exploit DV compression history and similarly reduce DV to MPEG -2 potential concatenation artifacts. Admittedly, DV's macroblock structure differs from MPEG 's, making the process more difficult; however, Heuris does a history analysis when converting Avid and Media 100 M-JPEG files into MPEG with their MPEG Power Professional. M-JPEG is not as standardized as DV.
The tools for the job What does any of this mean for equipment purchasers? Only good news, of course, because in many instances, the conversion from either DV or MPEG will be a straight decode-recode cycle. The differences will be hard to see, even if you mix and match, because simply staying digital is a big step forward from past analog practices.
In fact, the old standby, M-JPEG, stands up pretty well if you give it a high-quality digital source. While MPEG and DV have plans to support higher bit rates, current implementations stop at 50Mbits, or about 6.25MB/s. More mature M-JPEG products can achieve 15MB/s and higher, though M-JPEG's fixed quantization over an entire frame is less agile and less efficient than MPEG or DV's variable quantization. M-JPEG also has no method for compressing audio information or system level data and really falls down with interoperability.
Anything but standardized, almost every company implements M-JPEG slightly differently, making file interchange impractical, if not impossible. Worse, larger companies have multiple versions on M-JPEG. Avid alone has seven, according to estimates, and none are interchangeable. With compositing and other collaboration becoming commonplace, lack of workplace flexibility gives the industry a strong reason to move forward.
Interoperability, rather than the desire to spark a format war, is the primary focus of what the Sony- and Panasonic-led MPEG and DV camps were trying to show at NAB. Having different companies with different agendas and different technical implementations sharing compressed files is a reason to take notice to both initiatives.
Handicapping this particular prizefight ultimately depends on individual acquisition and post-production needs. Even Sony agrees that if you shoot DVCAM and don't add analog or non-DV footage, you should edit natively in DV and save the MPEG encoding until distribution. DV-50 is the obvious choice for DVCPro50 source material, as MPEG50 will be for Betacam SX acquisition, because source footage will stay native. 601 inputs to both allow for mixing from other sources. Where compositing is integral, many professionals will accept nothing shy of uncompressed editing equipment, avoiding compressed formats altogether during editing. If you can afford it, that's always the best choice.
Regardless, fight fans are in for a treat. Sit back and enjoy the show, because this one's not even close to being over. But, don't get too caught up in it if you have to make a buying decision. The technologies are similar. Buy the tools that solve a problem and fit your budget. Let the contestants in the ring do all the sweating.
|
