FEATURE

Videoconferencing: Bytes, Camera, Action!

by Dave Brown

The future of proprietary algorithms for videoconferencing is as bright as the future of EBCDIC for data exchange. In the early days of two-way compressed video, manufacturers engineered specialized, and somewhat secret, encoding techniques to produce the best possible video and sound between their own equipment. Yet as the CCITT (now the International Telecommunica-tions Union) began to negotiate standards, customers demanded interoperability, and manufacturers had to start offering H.320 options.

These days, all new videoconfe rencing equipment can observe the standards. Most manufacturers have shelved work on proprietary audio/video compression algorithms and are pouring significant development into improving the way they handle the H.320 protocol suite.

While prospective buyers of video teleconferencing terminals (VTT) have seen demonstrations between systems from a single vendor running under the best available conditions, many need to communicate under ISDN basic rate interface (BRI) line conditions with multivendor equipment. With that in mind, we tested VTT codecs in standards mode over 112-Kbps connections.

Real-life conditions in the U.S. switched telephone network necessitated our choice of 112 Kbps, rather than the 128 Kbps possible over an ISDN clear BRI. Clear 64-Kbps per channel ISDN is routinely connected within Local Access Transport Areas (LATAs) in many parts of the country, but most intra-LATA t runks still use 56-Kbps channels. In our tests, all cross-count ry calls but one connected at 112 K bps. Only CLI, dialing from San Jose, Calif., came in at 128 Kbps. (To normalize our performance numbers, we adjusted CLI's channel-limited frame rates down to 112-Kbps equivalents, but let its codec-limited frame rates stand.)

You can bet that if things can look reasonably good using codecs that operate at 112 Kbps, they'll certainly look better at higher connection rates.

Testing the Videoconferencing Systems We produced a four-minute VHS video tape with visible frame numbers and shipped an identical copy to each participating manufacturer. The tape's test clips were designed to generate various audio patterns and challenge video compression performance: We depict the results as six graphs on pages 49 through 52.

All participants played the video tape through equipment of their choice via single-line ISDN to the same RSI Systems ERIS decoder in Network Computing's test laboratory at the University of Wisconsin-Madison's En gineering campus. If the sending codec could support more than one H.320 audio format, we tested using G.711 (which leaves only 56 Kbps available for video) and again at G.728 (which allows 96 Kbps to carry the video).

A VHS deck in our laboratory captured the reference codec's output, so the tapes could be analyzed and actual transmission frame rates could be calculated for each sequence in the test.

RSI Systems ERIS To obtain a uniform target for our testing, we borrowed an ERIS desktop box from RSI Systems, Edina, Minn.-and immediately loved it. ERIS is externally packaged in a 10.5-inch long, 6-inch wide, 5.5-inch deep box that connects either to a Windows PC or Macintosh via SCSI cable. Its current list price is $4,495. A new model to be announced in the next few months will cost less than $4,000 and will include a standalone option.

As a desktop system, the ERIS provides excellent audio with a freestanding, stalk-like Telex microphone and 3-inch speak er built into the front of the unit. An internal micr ophone can be used instead of the Telex (producing a less satisfactory "hollow" sound).

We recognize a great future for the ERIS as the core of a build-it-yourself room system or executive conference facility. The box has Main and Auxiliary video inputs, each of which can accept SVideo or RCA plugs. Video output also is provided on SVideo and RCA. Audio inputs are available for the external microphone, and a VCR or audio mixer. It has a separate output jack for external speakers, plus two isolated line outputs-one providing local transmitted audio, the other for received audio.

ERIS supports FCIF video resolution and all three of the current audio algorithms: G.728, G.722 and G.711. In our performance testing, this codec ranked with the best room systems.

ERIS' only limit to masquerading as a room system is its single BRI ISDN interface. Having no RS-449 and RS-366 ports for external IMUXes, current models a re intended to play only in the single-line ISDN arena at no more than 128 Kbps. Since that is the most pervasive and fastest-growing conferencing arena, this is not much of a handicap!

Film at 11 We tested four codecs in "room" systems and six more that manufacturers package as desktop models. Results show that contemporary videoconferencing systems operating in standards mode are, well, pretty standard. Some, like PictureTel's 4000/ZX can crank out impressive frame rates when the picture is steady, but all were brought to their knees by the diabolical fine resolution and whirling pattern tests we put in our benchmark. In the end, the codecs in desktop systems perform as well or better than those in room systems. We've withheld top scores for the next round of improvements, where you'll see more computational power and predictive intelligence built into future video encoding processors.

If today's codecs are all pretty much the same, the finish and various op t ions are what distinguish these products. Many of these items will be detailed in our Buyer's Guide on videoconference systems in our March 15 issue, and we'll review them specifically later this year. We focus here, instead, on what we think are the most significant issues for your purchasing decision: price, audio support and video quality.

Price: What a Picture's Worth Manufacturers of the room systems we tested offer models that cluster in price around $20,000 and in the $40,000 to $60,000 range. For $20,000, you can buy a rollabout configuration with a single 27-inch or better monitor, pan-tilt-zoom camera and user-friendly control panel. The more expensive systems are usually built into executive conferencing rooms. They have two pan-tilt-zoom cameras and a document camera. Large monitors show what's coming from the other end and preview the camera shots you're about to send. Better quality cameras on the high-end room systems also contribute to perceived picture quality.

Room system manufacturers aim at specific market niches. VTEL suppo rts telemedicine with interfaces that can accept sound and video from clinical diagnostic equipment. PictureTel and CLI have been traditionally strong in the administrative conference room setting.

In the desktop arena, the tradeoffs are in price and computational load sharing. The PictureTel LIVE 100 and VTEL Enterprise Series, each listing for less than $5,000, put all the codec's computational power on circuit boards that plug into a Windows-based PC. RSI's ERIS system (priced at $4,495) is similar, but it sits externally on the desktop and can hook into either a Windows or Macintosh PC via SCSI.

The Intel ProShare ($1,999 list, but you can get it for as low as $999 with rebates from long distance and local telcos) needs computational support from its Windows PC host, which should be at least a 486 DX2 or a Pentium with 16 MB RAM. Apple PowerMacs require very little additional investment for videoconferenci ng. Our tests showed the 7100 AV performs codec functions very well with Quick Time and the Mac's inherent video processing power.

Do You Hear What I Hear? Poor audio can quickly give end users a bad taste for videoconferencing. Our testing showed that all manufacturers had excellent coder-to-decoder audio performance, even when using the G.728 audio compression algorithm. It's in the choice of microphones, noise suppression and echo canceling that designers can make their systems sound extremely good or horrible. Aggressive settings of noise cancellation, for example, can make normal voices sound gargly and choppy.

The trickiest feat for manufacturers is to design microphones that can pick up normal speaking voices from anywhere in the room, but not pick up loudspeaker sound and feed your own voice back with an annoying delay. Before buying, get whole-system demonstrations end to end that particularly test with a variety of sound sources.

You must investigat e one particularly important interaction between audio and video, especially if you inte nd to conference routinely at speeds as low as 112 Kbps. H.320 defines a choice among three audio algorithms: G.711, which requires as much as 64 Kbps bandwidth, but typically uses 56 Kbps; G.722, which requires 48 Kbps; and G.728, which gives very good sound, but uses only 16 Kbps. During a standard call set up, the audio algorithm is negotiated first. Whatever bandwidth remains is allocated to the H.261 video algorithm. G.728 audio is the best choice if the codec can support it. Wherever possible, we tested each codec in G.711 and G.728 to demonstrate the resulting differences in video throughput.

When Looks Are Everything All but one codec we tested generated output in "Full CIF," the best screen resolution (352-by-288 pixels) defined for H.320 systems. This is only about half the NTSC resolution used in the U.S., so don't expect studio monitor quality from your videoconferencing syste m.

The ProShare 200 using Room Video so ftware (Intel's H.320 offering) looks good when viewed in a window no larger than one-fifth of your computer screen. However, the ProShare image played through to a room system monitor will look coarse and jagged. This is because the ProShare currently transmits only in "Quarter CIF" (176 x 144 resolution). Intel introduced its Room Video option only in the past year to satisfy customer demand. Look for significant improvements in 1996.

The camera and monitor system(s) do the most to improve overall perceived visual quality for any video conferencing system. As we learned from our tests, it is the higher investment in controllable cameras with good optics, large monitors and enhanced sound systems that distinguish the room systems from the desktops, not the codecs.

We also learned that time base correction is an important concern if you plan to play videotape through your conferencing system. Tape decks have inherent fra me-to-frame time-sync instability that modern monitors easily correct. Some codecs, not as forgiving , will transmit jittery or undulating pictures from a tape source. Only one of our test participants had an obvious problem with this, and was able to correct it before running the benchmark. Another, the Matsushita-Panasonic Vision Pro KXC-M7500, appeared to transmit the most stable benchmark picture. The company reports that this is because it has "genlock" built into all of its video inputs.

Bumps On The Way To Videoconference

Like a golfer's swing, you can get many different opinions and absolutely stated recommendations, but you must ultimately decide on your own what is most comfortable for you.

One big issue is cost. Operating at 336 Kbps requires an upfront investment in six 56-Kbps digital service lines or, if you can get ISDN, three BRI lines. ISDN provides a small bonus. Each of the six channels (two per BRI) can operate at 64 Kbps, so in aggregate, ISDN can do 384 Kbps.

To get this aggregation, you also need to invest in an invers e multiplexer (IMUX). The IMUX performs "bonding." It combines the separate input channel bandwidth and delivers one sum-total channel with a clock signal to the codec. For outgoing calls, the IMUX can accept dialing instructions from the codec or look up the information from pre-stored call records. Then it places calls on each of the individual lines, aligns and bonds them before telling the codec "ready to go."

Inverse multiplexers that handle up to eight 56-Kbps DS0 circuits and deliver up to 448 Kbps on output can cost as much as $5,000 for new equipment. However, you will find very few sites with which you can communicate at those speeds. Most sites equipped to operate at "high speed" choose 336 or 384 Kbps as the upper bound. ISDN-capable IMUXes for up to four BRI inputs typically sell for less than $4,000.

Jerky movements can be notice able in conferences conducted at 112 Kbps or 128 Kbps. The degree depends on how much motion is taking place. Codecs achieve video compressions of as much as 1400:1 by analyzing each frame in the transmission image and only sending information about what actually changes in successive frames. "Talking head" pictures show up well at low speeds-most movement is only around the mouth or eyes.

The sharpness of a conference image depends on a codec parameter called the Common Intermediate Format (CIF). Only the frame presentation rate varies as a function of available transmission speed. Low speeds cause low frame rates when there is a lot of motion in the picture.

Conference system designers have found that for applications like training, the most cost effective strategy is to invest in high-quality, large-screen installations at participating sites, but to reduce transmission costs by operating at only 112 or 128 Kbps between the sites. For applications like telemedicine, the more effective strategy may be to use 336- or 384-Kbps transmission speeds.

The cost differe nces between low -speed and high-speed conferences become most apparent in multipoint situations, when a bridging multipoint control unit links more than two sites. In a recent conference arranged by the author, four meeting sites within Wisconsin were bridged for two hours using an MCU-operated by an in-state bridging service, Access Wisconsin. During the meeting, all sites operated at 336 Kbps. Transmission and bridging charges totaled $996. Had the same meeting been conducted at 112 Kbps, transmission and bridging charges would have totaled $572.

Standards: Above And Beyond H.320

H.320 is often mistakenly applied to the entire suite of standards that define how video conferencing systems from different manufacturers can interoperate. There actually are five overall standards suites (see chart, below).

The figure on the next page shows how the H.320 suite applies to a standard video teleconferencing terminal (VTT) that communicates via narrowband ISDN. N-ISDN covers bandwidths u p to 1.44 Mbps, but it is most commonly installed as one or more 128-Kbps basic rate interface (BRI) lines.

H.261: The Video Encoding Standard All but the new H.324 VTT, which will operate over 28.8-Kbps public switched telephone network circuits, use H.261 as the mandatory video encoding standard. A neat thing about H.261 is that it rigidly defines what the decoder must do with the digital video bit stream, but it leaves many options open in the encoder.

Most video compression algorithms, including H.261, use Predictive Coding, Discrete Cosine Transform (DCT), Motion Compensation and Variable Length Coding techniques to transmit TV pictures of acceptable quality at very low bit rates. Encoders are highly sophisticated image processors designed to find redundancy in successive video frames.

Depending on how much computational ene rgy the manufacturer wishes to invest in a codec's encoder, its apparent intelligence or predictive capability has no specified limits. Still , any H.261 device installed today can decode and present the improved image.

Common Intermediate Format (CIF) is a parameter that has a significant effect on picture quality as well as the ultimate cost of an encoder. Most manufacturers of room systems use "Full CIF," which provides a screen resolution equivalent to 352 x 288 pixels. Many desktop system manufacturers can get away with "Quarter CIF" (QCIF) if they choose to present the video images in small 176 x 144 screen windows.

G.711, G.722, G.728 Define Audio Encoding A modern VTT's audio codec recognizes three ITU standards that define how the sound is encoded: G711, G.722 and G.728. While arranging to connect videoconferencing systems that operate in standards mode, presetting both units to the same audio protocol is important, as is being sure that the c alled unit can detect and adjust to the caller's algorithm.

In the early days of H.320, particularly when channel bit rates of 112 Kbps w ere achieved by installing two Switched 56 circuits, a codec's 56-Kbps Port A would be used by the audio codec in G.711 mode, and the 56-Kbps Port B would be used by the H.261 video codec. G.711 uses a simple pulse code modulation technique to achieve audio bandwidth up to 3 KHz.

G.722 was introduced to afford better audio quality (up to 7 KHz) with a more sophisticated encoding technique: sub-band adaptive differential pulse code modulation. This requires a bit rate no greater than 48 Kbps. Some VTT designs that have Port A devoted to audio will give all of the available speed, either 56 Kbps or 64 Kbps, to the G.722 encoder. Others will always set up at 48 Kbps, and waste the remaining Port A capacity.

Modern H.320 VTTs are designed for the ISDN world. Most of those configured for use as room systems can operate over communication ch annels running up to 768 Kbps, a few up to T1 speeds. The figure shows how these VTTs can be connected to an inverse multiplexer (IMUX)-a device that "bonds" multiple ISDN BRI channels and presents one synchronous clocked signal to the VTT's Port A or Port B. If an IMUX is available to provide higher communication speeds (typically 384 Kbps), G.711 or G.722 audio algorithms can be employed without significant effect. You still have 320 Kbps available to the video codec, and that provides very good picture quality.

However, the most rapidly growing segment of the videoconference system installed base throughout the world involves VTTs designed for basic rate interface (BRI) ISDN at no more than 128 Kbps. When these have to operate in G.711 or G.722, audio ties up half the available speed. The video codec has to work very hard with the remaining 64 Kbps (or in many situations, 56 Kbps).

The recent G.728 audio standard is a major development that alleviates this crowding. Us ing a code-excited linear prediction algorithm, G.728 can provide audio bandwidth up to 3 KHz using only 16 Kbps of the bonded communications channel. Over clea r channel ISDN, 112 Kbps remains for the video codec.

Dave Brown is retired from the University of Wisconsin and is presently a consultant on videoconferencing. He can be reached at dave@dbec.com. We appreciate the cooperation of Bob Perras, Craig Bluschke and UW-Madison College of Engineering staff, who helped us with our evaluation.

The T.120 Sideshow: Electronic Document Conferencing

An important selling point for PC-based desktop videoconferencing systems is that many include document conferencing or "shared whiteboard" software. These packages are real productivity enhancers. For example, it would allow two engineers separated by great distance to view and discuss a product mock-up, while also having a shared screen window on one of their CAD application programs and files.

Some videoconferencing systems we tested have whiteboard software as a side offering when comunicating in proprietary mode with other systems from the same manufactur er. None can interoperate w ith other manufacturers' systems today. Tomorrow is a different story.

The generic video teleconferencing terminal (VTT) has a data communications protocol module that will use the new T.120 overall standard for collaborative applications, including desktop data conferencing, multiuser applications and multiplayer gaming. T.126 is a specific standard for still image exchange and annotation. T.122 and T.125 are for multipoint communication services.

T.124, generic conference control, combined with token passing, will handle remote camera control. Today, one of the few good reasons to get all of your videoconferencing equipment from the same manufacturer is this ability to control "near side" and "far side" cameras from one location.

All major manufacturers have pledged support for the T.120 standard. Many have already implemented elements of T.120 under the wraps of their current data communications control modules.

February 27, 1996