MPEG-2 isn’t going away anytime soon, giving cable myriad challenges to
ponder as it makes plans to insert advanced (and more efficient) codecs

Advanced codecs such as MPEG-4 and Windows Media 9 promise a leaner bandwidth profile and potentially better picture quality, but introducing them onto MPEG-2-dominated networks isn't going to be easy or quick.

With a massive base of legacy boxes deployed, no one sees advanced codecs usurping MPEG-2's dominance of digital television delivery anytime soon. But many see MPEG-4 and Windows Media 9 gaining footholds in advanced services such as video-on-demand, and there is some consensus building as to how to make them cohabitate with MPEG-2 in a cable network.

Transcode or simulcast?

The strategies split into two basic migration paths. One possibility is a transcoding model, where content is shipped around in a single format in the network core and then transcoded at the edge. For example, video could be stored and delivered to an edge cable modem termination system (CMTS) unit in MPEG-2, where it could be delivered to a legacy box as is or transcoded on the spot into a smaller MPEG-4 stream for delivery over coax to a newer, more advanced box.

Given the fact MPEG-4 streams are lighter, a transcoding strategy could be useful for cable operators already simulcasting analog and digital signals, according to Lou Mastrocola, RGB Networks Inc.'s vice president of business development.

"Digital simulcast in its initial form is actually a bandwidth strain on the network as opposed to relieving bandwidth," he says. "If the system goes all digital, you will recapture the analog bandwidth, but if that were to happen simultaneously with the move to an advanced codec, the ability to transcode from one format to the other would be very beneficial."

But transcoding from MPEG-2 to MPEG-4 at the edge does have its challenges, particularly if attempted in real time.

"The interesting part is they are different codecs and temporal models, so it can be awkward to take parameters that come out of one encoder destined for one type of codec and translate them into another," says Peter Monta, RGB's co-founder and chief scientist. "It seems to us most of the penalties are in rate distortion performance. You can do it, but you give up some quality or some bit rate."

So far, there hasn't been a lot of movement toward transcoding, according to David Price, vice president of business development for Harmonic Inc.'s convergent systems division.

"There are people who are saying 'we want to do transcoding,' and anyone can do transcoding by just putting a decoder next to an encoder. That's a coarse way of doing it," he notes. "As for any large, significant niche of interest in integrating that solution for transcoding that way around, not yet."

That is the situation for Comcast Corp., which sees a future for advanced codecs in on-demand services. While the proposed Next-Generation Network Architecture (NGNA) does call for a headend with the ability to transcode video, converting content there–as opposed to converting it at a video storage hub farther out in the network–doesn't offer significant benefit, according to MSO Executive Vice President and Chief Technology Officer David Fellows.

"To be honest, it's not one that we are modeling and planning on, and I've got numbers on what it will cost and what I'll save on," he says.

Simulcast is best?

The more likely strategy–particularly given the likelihood advanced codecs will pop up first in on-demand delivery–is simulcasting. That is what Comcast is eyeing for its first advanced codec rollout.

"In the video-on-demand server, you would store an advanced codec form of a movie or a standard definition MPEG-2 version of the movie and probably a high-definition version as well," Fellows explains. "Then the on-demand system can tell, set-top by set-top, whether that set-top is capable of receiving an MPEG-4 signal or another advanced codec signal, and then it would serve up that stream."

Following that model will put a storage load on video servers, but Fellows says that is more than offset by the bandwidth savings.

"The $80 billion you plowed into the ground is spectrum, and the servers are going down in cost over time," he points out. "So it is worth spending extra money on server capacity in order to preserve or free up or enable new service opportunities across the spectrum on your HFC plant."

Tony Wasilewski, chief scientist for subscriber networks at Scientific-Atlanta Inc., is seeing that duplicated content attitude gaining ground among other MSOs, as well.

"We're also hearing from inside of Time Warner that they are looking forward–they are much more amenable to duplicating content than I ever would have thought in the past," Wasilewski says. Dropping server costs may well be spurring that change, and, "I think things are happening to make that more of an easy decision to make."

In a VOD setting, simulcasting also gets around the real-time transcoding issues. "You have the complete freedom to encode from scratch each time," Monta says.

But simulcasting does face difficulties if applied to standard broadcast services, because duplicating channels in analog, digital MPEG-2 and a digital advanced codec would torpedo efforts to save bandwidth.

Given the mass of legacy boxes at work in subscriber homes "you are not gaining anything by introducing MPEG-4 in a broadcast fashion," Fellows adds. "With on-demand, you actually do save."

Others see that with the analog-to-digital conversion, enough bandwidth could be freed up to start offering MPEG-2 and advanced codec simulcasts for broadcast TV.

"You'd probably expect a trend where the higher-value services–the pay channels and certainly the HD versions of those movie channels, let's say–would be the ones that would likely get bandwidth allocated for them," Wasileweski says. "They would still stay in the MPEG-2 format but they would early [on] get the advanced codecs, as well, so the newer devices could use them."

And there is a strategy proposed to get around the simulcast bandwidth problem–enter switched broadcast, a technology BigBand Networks Inc. has championed. In switched broadcast, channels are delivered only if viewers tune to them, rather than funneling all channels down to the subscriber and letting the box do the tuning. In that way, operators can significantly slash the amount of bandwidth needed in the bandwidth constrained, coax last mile.

"So any program that isn't being watched doesn't get pushed down to that service area at all, or any program that is being watched in one format and not another format only gets pushed down in the format in which it is consumed," says Seth Kenvin, BigBand's vice president of corporate development.

The first mile

But even if the bandwidth problem could be solved, advanced codecs for broadcast still face a content hurdle. While it might save on transponder space for satellite operators and program distributors to use an advanced codec to beam signals to and from a satellite link, programmers may see little benefit in converting their content to an advanced codec at the origination point.

"There is a whole ecosystem of products that need to be developed around MPEG-4, but it all starts with the Catch 22–until somebody puts it into a set-top, nobody's going to serve it up in the headend," Fellows notes. "So yes, likely at some point in order to save on satellite transponders you could pitch MPEG-4 out of the uplink, but there is no real buzz about that. HBO hasn't come and said, 'Gosh, get ready to replace all of your satellite receivers with satellite receiver transcoders'."

Despite the differing strategies and significant hurdles, Comcast is still on track in its plans to debut advanced codecs in 2006, starting with on-demand services.

"I'm optimistic about the end of next year, but I would bet my job on the end of 2006," Fellows says.


Advanced codec battle creates split in chip strategy

The ongoing horse race between MPEG-4 and Windows Media 9 is prompting a side competition over silicon strategies.

With the two in a virtual dead heat for the cable advanced codec crown, gear makers are falling into two camps to deal with the uncertainty: upstart programmable chips versus tried-and-true application-specific integrated circuits (ASICs).

It comes down to flexibility versus cost: programmable digital signal processor or field programmable gate array chips can be reprogrammed to process alternate codecs, but ASICs still hold the all-important edge in cost.

In the programmable chip camp are outfits like Atlanta-based EGT, a new player in the encoder space. EGT is arguing that with the advanced codec competition still a toss-up, cablers should not be buying gear specifically for any one codec.

The argument is bolstered by technology improvements that bring DSPs close in processing power to ASICs, according to John Hartung, EGT's vice president of engineering. "So you deploy a box that can do MPEG-2 and tomorrow it can do H.264 or Windows Media 9 and deploy that more and more as time goes on," Hartung says.

That's exactly the plan for EGT's current product, a lower-cost MPEG-2 encoder that could one day be converted into an advanced-codec box.

"There may be a board that needs to be swapped out or something like that, but it's all DSP-based, and you can take the software off, put the new software on, and then the same equipment will perform as an advanced video codec encoder," says Greg Nicholson, EGT's CEO.

RGB Networks Inc., meanwhile, also is in the programmable chip camp, but it is relying on FPGAs. FPGAs are preferable to DSP chips because they can better handle the multiple levels of control needed to transcode content, according to Peter Monta, RGB Networks' co-founder and chief scientist.

"There's a control aspect to the entropy [of] decoding and encoding; there is control for memory; there is control for the arithmetic like DCT and ICT. All of these things are poor fits to DSP because DSP has a single instruction stream–or at most, a few instruction streams," he explains. "So FPGAs give us the complete freedom of just writing custom hardware for each block–we've done the whole implementation from scratch, actually."

But there is still significant backing for ASICs. Broadcom Corp., bolstered by its recent acquisition of advanced codec specialist Sand Video, has developed an MPEG-2/ MPEG-4 Part 10 decoder chip that is now undergoing market testing and should start limited production in the fourth quarter, according to Rich Nelson, Broadcom's vice president of marketing for broadband communications.

A tri-mode chip that adds Windows Media 9 also is a distinct possibility, although Broadcom hasn't announced such a product yet. "I think that's an ideal situation if you are a cable operator," Nelson says. "That gives you the maximum choice; that gives you the maximum flexibility."

In that camp you can find Scientific-Atlanta Inc., which is aiming for such a tri-mode chipset for its future set-tops. Although S-A has looked at programmable chip options in the past, the fact that the advanced codec race has boiled down to just two players means not all that much flexibility is needed, says Tony Wasilewski, S-A's chief scientist for subscriber networks.

While it is not clear that any one codec is going to win, "there is a pretty small set to choose from," he notes. It's not as "if there were 20 advanced codecs and we didn't know which one of them was going to win, and they were all different," he notes.

Harmonic Inc., meanwhile, also is solidly in the ASIC camp. It came out last year with an encoder that can process MPEG-4, and this spring, it added Windows Media 9 capability. David Price, vice president of business development for Harmonic's convergent systems division, notes that compared to the pennies in cost to use dedicated ASIC silicon carrying both codecs, single-codec DSPs can add $20 or more to the materials cost "which by the time you roll it up is really going to hurt the operators."

Cost is the big issue for Comcast Corp. as it chooses from among advanced codec products, according to executive vice president and CTO David Fellows. Some vendors have offered completely programmable DSPs that could compress voice or video or speech recognition, "but as far as I can tell, those chips are too expensive to put into the set-top on a box chip that goes into Dave's $50 set-top box," he says, referring to the proposed low-cost digital-to-analog conversion box.