The cable industry is finding itself forced to reopen the book on compression just as the primary supplier of digital technology to MSOs is preparing to deliver on its long-standing commitments.

Leading companies are rethinking earlier decisions to launch General Instrument Corp.'s proprietary DigiCipher extension of MPEG-2, with issues ranging from the costs of memory versus costs of bandwidth to the question of what impact worldwide penetration of MPEG-2 will have on cable's agenda in light of the long delay in rollout of DigiCipher. Making matters more difficult, MSOs also must adjust bandwidth assumptions to factor in new advances that squeeze far more digital channels into a 6 MHz frequency slot than earlier strategies assumed was possible.

The issue of whether or not to proceed with the DigiCipher compression technique is most pressing at Tele-Communications Inc.'s HITS (Headend in the Sky), where officials are preparing for launch of over 80 channels of digitized programming by June. HITS is hoping to make a decision on whether to use the DigiCipher syntax, or to go with full-profile MPEG-2, or a standardized extension known as "dual prime" this month, says David Beddow, senior vice president of TCI Technology Ventures Inc.

"There's no resolution of the question at this point," says Beddow, stressing that, however it is resolved, HITS' commitment to GI will stay on track. "The big issue is whether we populate the boxes with the extra megabyte of RAM (random access memory), because, one way or the other, we're going with the GI access and control (security) system."

The HITS compression evaluation is part of a larger industry discussion on the issue, now underway under the auspices of Cable Television Laboratories. Time Warner Cable, which supports the world standard approach, and TCI have agreed to try to reach consensus on compression after years of debating the issue.

Bandwidth vs. Memory

The question comes down to whether cable can use a relatively abundant resource-bandwidth-to compensate for a 50 percent reduction in the amount of DRAM (dynamic random access memory) which the full MPEG profile requires in the set-top box, says James Chiddix, senior vice president of engineering and technology at Time Warner Cable.

"The argument (for DigiCipher) is that you can achieve very high quality by going to higher bit rates without using 'B' frames," Chiddix says, in reference to the bi-directional frames that are part of the main MPEG profile, but which are eliminated from DigiCipher and the dual prime extension of MPEG.

The main profile of MPEG-2 requires relatively fewer bits than DigiCipher and dual prime for any given level of picture quality because it uses motion prediction based on the makeup of frames in front of, as well as behind, the B frame to fill in the picture for each B frame. The memory in the set-top is what allows the decoder to "look at" future frames for information that feeds back into a given B frame. The other versions only use motion prediction in a backward direction, which is to say, an algorithmic determination of how many bits are required to fill in each frame depends on information from the frames that have gone before it in the video stream.

From Time Warner's perspective, if the best approach is to avoid B frames, then it must be done in the context of using an accepted standard, either through de facto world acceptance of DigiCipher or use of the already standardized dual prime extension. "We don't want to be parties to bringing a proprietary system into the marketplace," says Chiddix.

As Beddow noted, set-top boxes to be delivered by GI for commercial deployment this year are designed to accept a full MPEG-2 or a dual prime feed as well as the DigiCipher feed. "We're not looking at any delays as a result of any new directions we might take," he adds.

Beddow puts the cost of the extra megabyte of DRAM at about $60, although just what the costs will be is part of the difficult assessment that lies ahead. Others, usually MPEG proponents, say the cost is more likely to be about $40.

Alex Balkansky, president and CEO of C-Cube Microsystems, a leading supplier of MPEG-2 chips, argues that, with the launch of digital video disks (DVD) this year, together with the recent implementation of full MPEG-2 by DirecTV, volume production of full MPEG-2 systems will quickly drive prices down, even if the memory component remains relatively high. Manufacturing costs for DVD players will start at $450 per unit and go down quickly from there, he says.

If true, this puts MPEG-2 boxes well within shooting range of GI's prices. Costs for set-tops using the DigiCipher system are much higher than originally projected, with moderately featured terminals pegged at about $400, and fully featured boxes at $450 or higher. Two years ago, the projected price for a low-end box was at or under $300 per unit.

"Forecasting the cost of memory is really the most difficult part of the digital cost equation," says CableLabs President Richard Green, adding that "the costs of memory have actually gone up a little over the past year."

Says Chiddix: "With the memory requirements associated with Windows 95 added to the other demand factors we're seeing, there's a credible argument to be made that the cost curve for memory could remain flat for the next few years. Avoiding the use of B frames could result in significant cost savings."

But, he adds, cost is also a consideration for DigiCipher, which must reach mass production scales if there is to be a cost advantage. "Unless DigiCipher can show that it can match MPEG-2 quality, which it must do if it is to become a mainstream business, I don't think we're going to see the production volume that will make it competitive with the full MPEG-2 approach taken by the consumer electronics industry," says Chiddix.

Green makes a similar point. One of cable's problems on the cost curve is that the DBS industry beat cable to the punch in launching digital service, thereby pushing the worldwide chip manufacturing community toward implementation of full MPEG-2 for encoders and decoders, he says. Thus, the burden is on cable to swing manufacturers to the simple profile approach at volume levels that will achieve the economies of scale enjoyed by full profile MPEG-2 suppliers.

A recent report in the Wall Street Journal noted DRAM prices had jumped several percentage points on product slated for delivery last fall. But it also pointed to predictions by research firms that over the next two years, supply will begin to outstrip demand, resulting in a surplus of memory chips by late '96.

The moving target

At presstime, CableLabs was slated to soon begin side-by-side showings of digitally compressed material that would allow its members' engineers to subjectively weigh whether the simple profile of MPEG-2 with dual prime or DigiCipher extensions can achieve the picture and sound quality delivered by the main level profile of MPEG-2. CableLabs is compiling tapes made with full MPEG-2 encoders to compare with tapes using material generated from simple profile encoders developed by General Instrument Corp. and AT&T, whose dual prime system is now in deployment by Cablevision Systems Corp.

"Dual prime is the equivalent to DigiCipher in terms of employing compression without B frames, but it happens to be part of the standard, whereas Digicipher isn't," Green notes. "So we have to be careful that we get the maximum bang for our buck as we search for cost-effective solutions.

"We're getting much closer to a consensus within the cable industry on all aspects of digital compression," Green adds. "So now we're trying to attract the attention of the manufacturing community to get them to focus on the approach we want to take."

The viewing tests are crucial to this process, not only with regard to achieving consensus within cable, but also in persuading the manufacturing community to go cable's way, Chiddix notes. "If it convinces us, it should convince others," he says.

But the testing process is tricky, given the fact that encoders, which determine how good the quality is, vary in performance from manufacturer to manufacturer even within a given MPEG category, Green notes. In fact, the latest developments in techniques surrounding the implementation of the basic MPEG algorithms in the encoder have made the whole issue a moving target which CableLabs will be hard pressed to keep up with in any evaluation process.

For example, iMedia Corp., a new company, demonstrated encoding and multiplexing technology at the Western Show last month that can deliver as many as 24 movie channels in a single 6 MHz TV channel using full MPEG-2 compression. And Compression Laboratories Inc., supplier of encoders to DirecTV using chips made by C-Cube, showed off statistical multiplexing and other techniques in real-time compression that can support up to 18 movie channels in an MPEG-2 stream.

The levels of compression achieved by CLI and Imedia assume use of B frames, although CLI has conducted tests to determine what the bandwidth tradeoffs are if B frames aren't used, says James Lakin, vice president for marketing and business development for the Broadcast Products Group.

"Based on two types of tests we've developed, we think that for any given picture quality, the bit rate goes up 15–20 percent without B frames," he says.

These new levels of video throughput are attained without any alterations in the basic MPEG algorithms and without sacrificing video quality, proponents say. Instead, they employ a variety of techniques in the encoding and post-encoding processes which eliminate any bits not necessary for a given quality threshold and assign bandwidth on the fly so that no one video stream is using any more space than it needs at any split second in time.

"In constant bit rate encoding, MPEG allocates too many bits to simple frame changes and too few to complex changes," says Edward Krause, vice president of research and development at Imedia. "So we start with a variable bit rate encoder, where we use an algorithm that determines exactly how many bits are necessary to sustain a constant picture quality."

This algorithm can be applied to any VBR encoder, Krause notes, adding that Imedia does not make the encoders but is strictly a software company. At the output of the encoder, in the multiplexing process, the Imedia system uses a statistical multiplexing process which varies bandwidth allocations to fit the exact needs of each channel's bit stream.

The Imedia 24-channel video count assumes a data rate of one megabit per second per digital channel, with modulation running at the 64 QAM (quadrature amplitude modulation) rate of 27 Mbps per 6 MHz channel. Krause says the system would require 1.5 Mbps to run live sports action.

These rates represent dramatic reductions from the bit rates commonly assigned to video, with movies typically running at between 2 and 3 Mbps, and sports consuming 5 Mbps or more. Imedia's demo at the Western Show operated at baseband using multiple clips of the same video running at 1 Mbps rather than putting 24 separate channels together over a 64 QAM stream, but officials say the full suite of software components will be ready for commercial deployment by the second quarter of '96.

CLI's capabilities are reflected in actual commercial operation, although the DirecTV system is not pushing the limits of the technology. CLI's Western Show demonstration showed how VBR encoding and statistical multiplexing result in constantly varying bit rates per channel in a live feed from DirecTV transponders. Alphanumeric windows displayed the bit rates on each TV screen, with rates increasing or falling as the scenes shifted on each channel.

"DirecTV is operating at an average of 3–5 Mbps per channel, but, if you were using MPEG without adding these techniques, the hockey game you see here would be assigned 6 or 8 Mbps," noted Gary Trimm, president of CLI's Broadcast Products Group, during the demonstration.

Along with its own patented version of statistical multiplexing, CLI uses a proprietary technique known as "detelecine," which involves more efficient use of the redundant frames in film sequences. Because film operates at 24 frames per second, and NTSC video operates at 30, a film channel, to be compatible with television receivers, carries six frames per second of "padding," which can be used for other information in the transport stream, says Trimm.

Statistical multiplexing and detelecine each expand bandwidth efficiency by about 20 percent, Trimm adds.

A momentous decision

The stakes are high as cable prepares to launch digital services, playing catchup rather than leading the charge as had been originally intended. C-Cube's Balkanski offers a sobering analysis of what the cost impact of widescale adoption of MPEG-2 could mean to the cost curve by citing what his firm has experienced with MPEG-1 CD players in the Far East, which were introduced over a year ago as a playback system for karaoke, multimedia games and other full-motion video formats.

The CDs hold only 74 minutes of video, but have proven extremely popular, with player sales in the millions of units, says Balkanski. "The manufacturer's cost for video CD players a year ago was $300 per unit," he notes. "Today, it is $130."

With over one million subscribers now taking services employing full MPEG-2 rather than high-speed MPEG-1 via the three satellites operated by Hughes Communications, advocates of the worldwide standard can be expected to ratchet up efforts to persuade terrestrial network operators to swing behind this approach to digital transmission, rather than waiting for other options.

"We believe this will open the eyes of people in the U.S. who are weighing which way to go," says Trimm. "The cable industry can't afford to make the wrong choice."