Pipeline Profile: Chris Gordon

MPEG-4, VOD and SDV

Title: Senior director, product management, EGT

Broadband Background: Gordon is responsible for the design, marketing and solution engineering of EGT's digital media video processing product lines. He has been with EGT since 2001. Prior to EGT, Gordon worked at Viant as senior analyst and information architect and Media Metrix as director of product development.

In your recent CT article, you write that MPEG-4 is better suited for HD-VOD than for switched digital video (SDV). Why is that?

HD-VOD is a unicast service while SDV is, at least for a while, a multicast service. In other words, a VOD stream goes directly to a single set-top box, while an SDV stream is distributed to all set-tops downstream from a node. If the single VOD set-top box is MPEG-4 capable, the system can deliver an MPEG-4 stream. However, unless all set-tops downstream from the node are MPEG-4 capable, the system has no choice but to use a commonly decodable format, MPEG-2.

What about your idea of MPEG-4 service groups or nodes? How plausible is that?

It's an interesting idea and one that deserves some further consideration, but for a cable operator today, it's really not plausible. The ability to provision an MPEG-4 group or node has both technical and marketing challenges. Technically, the operator would have to ensure that all set-tops downstream from a node were MPEG-4 capable. This will only happen once MPEG-4 OCAP set-tops are firmly established in the market as a viable alternative and when service providers have a method for ensuring that no nonMPEG-4 set-tops have been installed. From a marketing standpoint, operators are not keen on providing advanced services (such as a hugely expanded HD tier) in only small sections or nodes in the network. It makes the marketing of that service complex and risky as neighboring subscribers could be turned off if they don't have the same option. Only when the MPEG-4 node or service group is large enough will the marketing challenges become surmountable.

One of your main points is that a more effective use of MPEG-2 data can enhance the quality of MPEG-4 transcoding. Could you elaborate?

Our advanced engineering team is the real expert on this subject, but generally there are decisions made in the original encode that can be re-used for the transcode operation or at a minimum provide a guide or sanity check on the transcoding decisions. For example, the type of original picture or frame (I, P or B frames) can and should be used in a transcoding application; MPEG-4 I-frames from MPEG-2 I-frames, or vice versa. Quantization step sizes in the input stream are clues for spatial and temporal complexity that can be referenced to both "clean" the picture with filtering prior to transcoding or to reduce the complexity of the transcode operation, thereby improving quality. It's also worth noting that if the transcode is from MPEG-4 to MPEG-2, the MPEG-4 encoding parameters can be optimized to improve the quality of the transcoded image.

How do those techniques figure in the VIPr-IPx video processor you announced at Expo?

VIPr is EGT's next generation platform for video processing. The most important aspect of the VIPr platform is the fully programmable engine. The reason that's so important is that from a research perspective, transcoding is relatively new. As new innovations, techniques and applications are discovered, we wanted a platform that enabled us to implement as quickly and cost effectively as possible.

At an HD workshop at Expo, Cox Communications VP Video Engineering James Kelso said, "There should be no tradeoff between quality and quantity." Is that fair? What about other tradeoffs, say, between cost and quality?

Mr. Kelso is right. There should be no tradeoff between quality and quantity. We do see other trade-offs, but not necessarily between cost and quality. From our perspective, the primary trade-off is between complexity and cost. The higher the complexity of the application or implementation, the higher the cost will be to support that complexity. That said, complexity should not have a direct correlation to quality. The best and most innovative quality improvements are more often elegantly low in complexity. The real value in video processing comes from providing the best quality at the lowest complexity.

At the same workshop, Comcast Media Center Director Advanced Engineering Ren Finley said, "MPEG-4 isn't quite here yet." Yet Cox's Kelso said he doesn't want to buy any set-tops next year that can't do MPEG-4. (For more on this workshop, see our Expo coverage.) Are we at an MPEG-4 inflection point?

I think we are. And that inflection point is really being driven by HDTVs and the proliferation of programmers, both professional and amateur. HD is to TV today what color was in the 1950s, while consumer digital HD cameras and desktop publishing tools are rapidly expanding the number of content sources. With so much HD programming being produced and demanded, the bandwidth efficiency of MPEG-4 is undeniable. The large telco TV providers, the satellite direct-to-home companies and now programmers like HBO have all made the decision to migrate to MPEG-4. And when MPEG-4 OCAP set-tops begin to reach a tipping point in cable, I'm certain we'll start to see cable operators also take advantage of the MPEG-4 standard.

What are the best measurements of HDTV quality? (Any thoughts on how the ongoing Comcast/DirecTV litigation might address that question?)

As much as we'd like there to be a more definitive objective metric, there remains no substitute for the human eye. There is, however, some very interesting research now underway at Georgia Tech to extend the notion of MTBF (mean time between failures) to video quality analysis. MTBF is a metric typically used for hardware component failures, but by extending it to video analysis, it provides a very useful framework for accounting for the quality that really matters - that perceived by the subscriber. The subscriber doesn't care if an issue is caused by the source, the compression, the transport, the decoder or the display. They just know that the picture didn't look right. We're paying close attention to emerging video MTBF standards as ways to help us optimize our end-to-end quality.