Software-only MPEG-2 encoding tools are offering more and more features, but for a full slate of professional options, hardware-based encoding is still the only way to go.
With the advent of DVD, the world is changing in a big way for today's shooters and producers. Like you, as I survey my current crop of projects, I am finding myself not as concerned with tape output anymore. DVD, satellite transmission, and a myriad of broadcast applications have made MPEG-2 the undisputed output vehicle of choice. Shooters and producers who still resist embracing the new MPEG-dominated world will find it increasingly difficult to survive in the current business environment.
For camera folks like myself concerned foremost with the integrity of our images, DVD is changing everything — the way we shoot, the way we edit, and most importantly, the ways others see and judge our work. Whether we like it or not, our meticulously lit, supremely doted-over images will be highly compressed. And that compression had better be done right. With taste — and due respect.
Effectively, this means that today's shooters must take control of the encoding process. Indeed, Hollywood cinematographers are beginning to do precisely this as studios begin to provide dailies on DVD, prompting many pros to specify not only a trusted colorist at the film lab but a favorite compressionist as well.
In years past, shooters diligently learned the nuances of every new film and tape stock. It was considered essential if a DP were to maintain strict control over his craft. In the current environment, the shooter faces the more daunting challenge of grasping the exigencies of MPEG-2. Understanding the codec and related encoding tools — hardware-based and software-based — is now vital to the shooter's survival.
MPEG-2 in a nutshell
Forget about the space shuttle or that Mercedes SLK in your garage. MPEG-2 is the technology of our time. In DVD, videogames, satellite TV, it's everywhere you want to be — in all its glory and with the occasional ugly artifact.
MPEG-2 is a clever technology, but the principle behind it is simple. If you hold up a strip of film to the light, you'd see very little change from frame to frame. Any motion or variation in color over a particular set of frames is slight. Engineers of the Moving Picture Experts Group (MPEG) built upon this simple concept.
In the world of MPEG-2 for DVD, the smallest decodable unit is the Group of Pictures (GOP) — usually (in NTSC) made up of 13 or 15 frames. The significant compression achievable in MPEG-2 is derived from the fact that each GOP contains only a single complete frame, the remaining “frames” being mathematically derived from the initial I-frame (Intraframe).
These mathematically derived frames are both predictive and interpolative, meaning that MPEG has the ability to generate motion vectors that “see” ahead in time as well as fill in any presumed action between frames.
How well your encoder (or transcoder) accomplishes this task will determine the quality of your encodes.
And therein lies the rub. As I discussed in my article “The Shooter's Guide to DVD” (Video Systems, September 2001, page 36), encoders can and do get confused in the face of complex scenes. Sharply defined, high-contrast lines or patterns may produce ugly halos or “mosquitoing.” Fast-moving scenes tend to form large blocks at low bit-rates. DV especially, with its high level of detail, often displays a hideous chiseled effect in MPEG-2.
Encoders, whether hardware- or software-based, inevitably favor one compression parameter over another — say, rapid changes in color over fast motion. This means one encoder might work well with fast-action sports such as rodeos, but would perform less robustly when confronted with a Halloween parade featuring a bevy of complex colors and costumes.
To the adept shooter, the brave new world of MPEG-2 means gaining an understanding of a range of encoding tools based on their individual strengths and weaknesses. No single tool can be expected to do it all, although the high-end hardware encoders certainly come close.
So the question is, as craft-savvy shooters and producers, can we get away without a dedicated hardware encoder?
That depends.
A software solution?
Wouldn't it be nice if software could do the job just as well? If you're encoding directly from a tape source, the software solution is of course not an option. You'll need a controllable deck with a nine-pin RS-422A connector to tie into the encoder card in your PC or Mac.
But if you're outputting your work from your NLE anyway, why bother with tape output at all? You can simply create a QuickTime movie from your timeline and drag it into one of a growing crop of inexpensive transcoders.
These software tools, available from a range of manufacturers — Apple, Heuris, Discreet, Media 100, Sonic Solutions, and others — offer many advantages over their hardware-based brethren beyond their inherently lower costs.
They are flexible, easy to master, and usually feature (as in the case of Cleaner 5) a familiar interface. The Cleaner product in particular offers users a full range of useful options, including batch encoding, noise reduction, manual cropping, and true variable bit-rate encoding.
Software transcoders also offer an easy upgrade path. No need to swap out a hardware I/O or PCI board, or seek the latest firmware revision. With the notable exception of Cleaner 5's Supercharger board, software transcoders typically do not require a PCI card that takes up a valuable slot inside one's CPU.
The convenience of software transcoders cannot be overstated. Their latest incarnations from Media 100 and Sonic Solutions are available as NLE plug-ins, so users now can access these tools directly from their Media 100 or Avid editing systems.
Not all peaches and cream
Even so, encoding via software is not all peaches and cream. The process itself is extremely processor-intensive — the faster the processor, the faster the process.
And, as anyone who's ever tried would tell you, encoding a long-form program via software can be an excruciating ordeal. Performance of an MPEG-2 transcoder may be as slow as 30:1, meaning a one-hour program would require 30 hours to encode.
Even with the Cleaner 5 Supercharger, my Mac G4 dual-800MHz machine still requires 12 hours or more to encode a single hour of program material. Add some moderate noise reduction and I'm easily looking at close to three days to encode a single hour of material! Whoa. This hardly makes for a pleasant (or practical) experience.
Not all software transcoders are as cavalier with your time, however. The Apple QuickTime transcoder is substantially more considerate, and while short on key features like noise reduction and inverse telecine, it is a virtual speed demon when compared to its Windows-based peers. Accessing the G4 processor directly, Apple's engineers were able to realize substantial speed gains, approaching near-realtime performance on a dual-800MHz Mac. And at bit-rates of more than 5Mbps, output quality is not half bad, either.
Still, a major drawback of using software to encode your productions is the lack of ability to “touch up” the encoded file. Objectionable artifacts or other defects apparent in the encode would subject you to another long, grueling ordeal. Faced with a real-life deadline, this could be cause enough to commit hara-kiri.
Beyond the inability to re-encode segments, another significant disadvantage of encoding via software is noticeably inferior performance, especially at lower bit-rates. If you're a DVD producer, chances are you're constantly being pushed to add “extra-value” content to your titles, the net result of which is the need to lower the overall encode rate to 4Mbps or less.
At 3Mbps, most software transcoders will produce objectionable artifacts with material of only average complexity. So under such conditions, for example, attempting to encode more than two hours of program material onto a single DVD layer, the software transcoder becomes even less of an option.
The hardware encoder
In a viable business environment, a dedicated hardware encoder may be the only practical solution to frustratingly long and futile encoding sessions. Unlike the software solutions previously discussed, a hardware unit is designed to do only one thing — produce MPEG-2 streams that are DVD-compliant.
(And let's be clear. Many MPEG-2 hardware encoders on the market are not DVD-compliant as they are intended for broadcast or satellite applications. Among other considerations, DVD's variable-bit-rate mode is not supported by these encoders, nor, frequently, is the 9.8Mbps maximum encode rate that's mandated by the DVD-Video specification.)
Shooters and producers looking at a DVD-compliant hardware encoder will find many advantages across the board. Speed is certainly an issue, but it is not an overwhelming advantage in some cases (for example, when compared to the Apple transcoder).
Using a hardware encoder, a 90-minute feature film would require approximately six hours to encode — or the equivalent of three passes of the source tape. The first is an inverse telecine pass to remove the 3:2 pulldown introduced in the film-to-tape conversion. While most software transcoders also offer an IVT feature, hardware encoders like the Sonic Solutions SD-2000 allow for varying its cadence, a feature particularly useful for producers working in animation.
The second pass of the hardware encoder is for analysis. Here the encode parameters are determined according to the complexity of scenes. In the third pass, the calculations from pass number two are applied and the actual encode is executed.
While most transcoders also offer VBR encoding, they do it with minimal analysis in a single pass in order to maintain something resembling a reasonable forward speed. The Heuris MPEG Professional 2 is an exception, allowing a discrete analysis pass prior to transcoding.
In addition, the hardware encoder offers users greater VBR efficiency, typically ramping the encode rate up and down 2Mbps or more to accommodate scenes of varying complexity. Software transcoders, working under intense pressure, do not have the luxury of ratcheting the encode rate up and down to such a degree, resulting in reduced overall efficiency and quality.
Sonic's high-end SD-2000 also permits variations in the scan order of MPEG-2's macroblocks. Nagging problems that can't be resolved any other way can sometimes be resolved by shifting the encoder's scan order from Alternate to Zigzag. This kind of control, not possible on lower-end solutions, might seldom be needed, but when it is, such a feature can save (or make) your career.
As I alluded to earlier, one major advantage of hardware encoding is the ability in some models to perform “segment re-encodes.” After encoding, some footage may exhibit a number of problem areas — wafting smoke, for example — that cover only a few frames.
In software, such minor artifacts would necessitate re-encoding the entire movie. In higher-end hardware models, only those problem frames need to be re-encoded, providing a significant time advantage over lower-priced encoder solutions.
Then there's the workflow advantage. Most hardware models, including the Sonic Solutions SD-1000/2000 and Zapex ZP-330 DD, allow for simultaneous encoding of audio to Dolby Digital AC-3. With most transcoders this requires a separate pass, if it is possible at all.
One great advantage of Sonic's SD-2000 is the incredibly useful thumbnail file it generates during encoding. This file, imported into the authoring environment, provides a frame-accurate reference by which to set chapter marks and player commands easily. Lesser encoders and software transcoders produce a file that is imported as a solid gray bar, making head-and-tail trimming of assets awkward and time-consuming.
Most critically, hardware encoders are invariably more efficient at lower bit-rates, providing superior encodes at rates as low as 3Mbps. In the case of the SD-2000, the encoder can be set to exert various degrees of “effort,” meaning the encoder will revisit the same problem scene multiple times to fine-tune the key parameters prior to encoding. This kind of precision is simply not possible in any of the available software solutions.
Another little-recognized advantage of the hardware encoder is the superior decoding ability that comes with it. Software-based decoders are inherently unreliable as proofing tools — a matter of no small insignificance. An investment in a good hardware encoder will markedly improve your ability to proof your projects — a major consideration given the normal frustrations associated with any DVD project.
But hardware encoders have their drawbacks. Very basically, these tools can be rather expensive: Sonic Solutions' top-end SD-2000 retails for $24,999 as a standalone encoder. The Zapex ZP-330 DD — check for latest availability — lists for $11,500. The venerable Mediapress Pro with the SDI option from Media 100 seems like a steal at $3,895.
There is also the requirement of an expensive tape deck and RS-422A interface. In my own work, I use the Sony DSR-40 DVCAM deck with the Sonic SD-2000 — a reliable combination that beautifully melds the DV and DVD worlds.
Integrated encoding solutions
Owing to the complexity of encoding MPEG-2 and the lack of standards with respect to GOP size and structure, it makes eminent sense that your choice of encoder be linked to your choice of authoring tool. As mentioned earlier, Sonic Solutions offers just such an elegant approach that seamlessly links its high-end encoder/decoder to its authoring environment. The result is a system that works extremely well, combining reliable, high-quality encoding with a superb workflow.
The integrated workstation appears to be a growing trend in hardware-based MPEG-2 encoding. Pinnacle Systems has introduced its DC2000 product. It's an unusual combination of hardware and software that enables not only high-level encoding of assets, but — for the first time anywhere — frame-accurate editing of MPEG-2 files via Adobe Premiere. The Windows-based DC2000 system, starting at $2,200, features a comprehensive array of A/V input options, from Betacam component to DV — plus authoring via Impression DVD SE.
By placing all this capability under a single umbrella, Pinnacle assures maximum compatibility of encoded files throughout the premastering process. Encoding for DVD is an inherently complex task, fraught with perils that can torpedo even the best-prepared project. Integrated solutions from Pinnacle and Sonic Solutions would appear to reduce, if not eliminate, many of the frustrations associated with DVD projects.
The bottom line
In the Land of MPEG-2, high bit-rates eliminate or reduce most objectionable artifacts. The more bits assigned to describe a scene, the more accurately that scene will be represented in compressed form.
Practically speaking, maintaining a high rate of 7Mbps to 8Mbps for your encoded material may eliminate the need for a hardware solution, as most modern software transcoders perform adequately at such a rate. Of course, this dramatically reduces the amount of video on a DVD to less than an hour, but this will be acceptable for some projects.
On the other hand, if you're facing a tough deadline and the quality of your show is a prime concern, then a hardware encoder would appear to be the inevitable choice. Currently, only hardware offers a full range of truly professional options, from adaptive inverse telecine and segment re-encoding to variable “effort” settings and thumbnail files that serve to facilitate the DVD process — and give you some hard-to-find peace of mind.
This article is available in 
