Despite the need for more standards and affordable bandwidth, current desktop videoconferencing products work well for group collaboration
Eric Garland and Dave Rowell
Videoconferencing is not a new industry, but until recently, it has been confined to large room-based systems and cartable (roll-about) small-group systems costing $20,000 and up. Faster desktop PCs, increased network and communications bandwidth, and more-capable digital video components--all at lower prices--have brought videoconferencing on desktop systems into the realm of possibility and even affordability (see ``Video Conquers the Desktop,'' September BYTE 1993).
A broad range of videoconferencing products is now available for a variety of hardware platforms and communications channels. We look at five of them in thi
s review. Two are Unix software packages that work over networks on video-equipped Silicon Graphics, Inc., and Sun Microsystems' workstations: InPerson Desktop Conferencing from SGI and ShowMe from Sun. Two are Windows-based hardware/software kits that give a PC videoconferencing abilities over ISDN: Vistium Personal Video 1300 from AT&T Global Information Solutions and PictureTel Live PCS 100 from PictureTel. And one, InVision Desktop Video/Document Conferencing Software V.3.0, from InVision Systems, is a software product that turns a Windows PC equipped with relatively inexpensive video and audio hardware into a network videoconferencing station. It works with a variety of network operating systems and media types.
New Uses
Putting videoconferencing capabilities into desktop systems has spawned new applications like document conferencing, or whiteboarding, where people work together on an electronic document using software tools (see ``Desktop Data Conferencing,'' May BYTE). Many videoconferenc
ing programs allow application sharing, where participants collaborate on projects using the same virtual spreadsheet or word processor.
These new videoconferencing applications don't require the same video quality as do long-distance meetings, where appearances and the ability to detect nuances of expression are critical. As a result, they've created a market for even lower-priced videoconferencing systems. Although companies like Compression Labs and PictureTel have had desktop products for several years, their systems are still expensive because they rely on a hardware codec to digitize, compress, and decompress the transmitted video data.
Intel started the current trend toward more-affordable systems when it introduced its ProShare Video System 200 in January. By relying on a software codec (based on its Indeo codec) that doesn't depend so much on DSPs (digital signal processors), Intel can offer its hardware/software package for $2500. The Intel product provides the same software features a
s the more expensive PC upgrade kits, works over the same 128-Kbps ISDN connections, but produces somewhat diminished video quality. Creative Labs recently took that trend further by offering a $1500 upgrade kit that works over regular phone lines (see the text box ``Low-End Conferencing'').
Old Barriers
In spite of lower prices and mature applications, there are reasons to wait regarding desktop videoconferencing. Lack of standards is one. H.320 is an ITU-T (International Telecommunications Union-Tele-communications Standards Section) group of standards (see the text box ``Videoconferencing Standards'') that includes the H.261 video codec standard among several others (H.320 has come to represent its parent Px64 videoconferencing standard). This standard has encouraged the proliferation of videoconferencing products in the last couple years, but it isn't well suited to software processing by a CPU, nor does it support document conferencing. Companies like Intel have used their own codecs to ship
affordable products. Although Intel has promised H.320 support by year's end, the ProShare Video System 200 has been popular enough that it threatens to make the Intel codec a de facto standard.
For remote connections, ISDN is the connection of choice. Unfortunately, switched digital phone services such as ISDN aren't yet universal or fully standardized. Also, no inexpensive and simple way exists to videoconference over ISDN with more than two parties involved. To do so, you must go through a service provider with MCU (Multipoint Control Unit) capability.
In fact, AT&T plans to have a complete multipoint conferencing service available by the first quarter of 1995, using its existing digital infrastructure. As part of its WorldWorx Network Services, AT&T will offer multipoint collaboration by linking two to 22 people in a videoconferencing session. The service will also include data transfer and application sharing. Although WorldWorx will be based on H.320 standards, AT&T has agreed to support
Intel ProShare by transcoding. This strategy would allow ProShare systems to link with H.320 systems in a multipoint videoconference, but transcoding would entail some measure of delay. The AT&T announcement makes Intel's future support of H.320 a bit less critical.
Regarding LANs, there are no standards for videoconferencing at all. LANs have plenty of bandwidth compared with ISDN, but the unpredictable and bursty nature of data transmission over shared-media networks like Ethernet and Token Ring isn't naturally suited for transmitting audio and video data, which require dependable isochronous bandwidth. LAN-based videoconferencing software must jump through a few hoops to provide smooth audio and video without unacceptable delays. H.320 does not yet fit well with LAN requirements, and although some programs provide it, LAN products use proprietary codecs that prevent inter-operability.
Switching hubs and faster LAN standards will ease data flow and reduce the traffic burden that video data pla
ces on a network. Also, ITU-T LAN videoconferencing standards are under development, as well as an IEEE standard for isochronous Ethernet. For long distance networking, ISDN is still your best bet.
ATM (Asynchronous Transfer Mode) may solve the problems involving bandwidth and standardization. At speeds ranging from 155 Mbps to over 2 Gbps, ATM can provide the necessary speed to handle videoconferences and still have plenty of bandwidth left over for other operations. ATM will also span from LAN to WAN, including communications service providers, raising hopes of videoconference standards that work at all levels.
One advantage that LAN-based videoconferencing packages have over ISDN-based products is multipoint conferencing. If your LAN can support the traffic, LAN programs like those from SGI and Sun let you conference with multiple parties.
Frame Test
All the products we evaluated provide real-time color videoconferencing. Evaluating video quality involves several factors, includ
ing resolution, image clarity, and frame rate. The first two factors are subjective; frame rate can be measured.
Testing actual frame rate at the receiving end of a videoconference is tricky. Movement in the image, for example, makes video compression less effective and increases the amount of data sent per frame. Where bandwidth is limited, the software usually drops frames or, as in the case of AT&T's Personal Video 1300 software, lessens image quality. Therefore, frame rate and image quality typically vary with the degree of movement.
All the systems we tested boast frame rates of at least 15 frames per second (full motion is usually defined as 30 fps). These frame-rate estimates are based on best-case conditions with minimal movement--basically, a still picture that looks the same whether the frame rate is 1 or 30 fps. For a more complete picture of frame rate, you need motion.
We developed the VCBench test to provide a measure of frame rate, given a great deal of motion. Using a Crea
tive TV Coder from Creative Labs, the VCBench test creates a test image at the sending end of a conference connection. The image is a white rectangle on a black background. The white rectangle fills one quadrant of the video image and moves sequentially from quadrant to quadrant at a set interval that can be adjusted with the test program. At the receiving end, four optical sensors monitor the video image, recording when the rectangle appears in each quadrant. (Thumper II, BYTE Lab's battery testing system, read the optical sensors.)
To test frame rate, we slowly increase the speed of the moving box so that it switches quadrants at a speed that matches the frame rate. The VCBench test detects dropped frames, which makes it easy to match box movement to frame rate. The test then displays actual frame rate. With all systems, we tested with the maximum supported video resolution (160 by 120 pixels for InVision, 352 by 288 pixels for AT&T and PictureTel, 320 by 240 pixels for SGI, and 640 by 480 pixels fo
r Sun. We tested the three LAN-based systems on an isolated 10Base-T segment to remove traffic as an uncontrollable variable. For the two ISDN products, we used Teleos ISDN units to make the connection. Both local and remote video windows were active in the receiving systems.
While a white rectangle moving over a black background hardly resembles a talking head, the image stresses the tested system. More important, the results matched our own subjective estimates of frame rate with high motion in real images. Also, the relative ranking of the five systems matched our own impressions. Frame rates for typical video sequences will fall somewhere between the vendors' estimates and the VCBench test results.
Vistium Personal Video 1300
AT&T has been involved in bringing videoconferencing to the desktop from early on, providing chip sets as well as its own add-in boards. The Vistium Personal Video 1300 is one of the two H.320-compatible ISDN systems we tested. It's the one we like best because of
its functional software, fluid video, and use of a speakerphone as its ISDN interface.
The $4995 kit is designed for PCs running Windows and includes two joined expansion cards, Windows software, a camera, and a speakerphone. The two cards handle voice, video, and data transmission with AT&T's AVP codec chip set and several DSPs. The two boards are joined and require adjacent ISA slots. The host system's graphics card works through the Personal Video 1300 cards via a pass-through connector. The host systems that AT&T provided for testing were two System 3333s (486DX2/66 CPUs). As with PictureTel Live, hardware video processing means the AT&T system works well on slower Windows PCs.
Rather than use an expansion card to provide an ISDN connection, the Personal Video 1300 can connect directly to an AT&T 8510T ISDN speakerphone. We really like this setup because it uses the already familiar phone interface. You can make telephone calls without having to deal with a software interface or having your
computer on. The phone has all the buttons you would find on a standard PBX phone and a few more.
This is a handy way to implement ISDN, and we hope this is the way they all will be some day. Our only complaint is that a regular incoming phone call causes the videoconferencing software to give you a ``Call ringing on invalid type'' error message that shouldn't come up. You can still answer the call as you normally would, so it isn't a major inconvenience.
A big problem with the audio portion of conferencing is echoing, which occurs when the microphone gain on the other end is too high and sends your voice back to you. This is not a problem with headsets and can be solved easily with some systems by moving the mike and speaker farther apart or adjusting the volume.
The best approach with speakerphones, however, is echo cancellation, where the audio system subtracts some manifestation of your outgoing audio signal from the incoming signal after an appropriate delay, thus removing the echo.
Both the AT&T and PictureTel systems have hardware echo cancellation for their speakerphones. When you start it up, the AT&T system also produces a series of initialization tones that allow the speakerphone to adjust its mike gain to the acoustic surroundings.
We also like the Personal Video 1300 software. It's a group of six applications managed by an overlord program called Video Manager. There's a window for the videoconferencing function and a separate whiteboard window. The package also includes file transfer, application-sharing, directory, and phone-dialing programs. The whiteboard program follows the paradigm of a draw program, so everything you draw is an object that you can manipulate. Video Manager makes it easy to manage the different pieces of the package. For instance, if you load the whiteboard program, you can shrink the videoconferencing program into an icon in the Video Manager to get it out of the way.
The Personal Video 1300 hardware uses chroma keying to superimpose a video
over an image generated by the graphics card. The graphics card generates a box filled with a key color where the video should display and sends that image to the video board via a pass-through connector. The video board then uses the chroma key to overlay the video image and sends the resulting mixed image to the monitor.
Chroma keying has several benefits. Because the graphics card controls the video area, it's easier to put video in a well-behaved window that can be overlapped by other windows. Another benefit is that the video data doesn't have to be stored in the graphics card's frame buffer, an action that requires work by the CPU--nor is the CPU involved in decompressing the incoming video. With chroma keying, the video card does that work. By not loading down the CPU, Personal Video 1300 can run gracefully on slower systems.
One hitch with overlaying the video with chroma keying is that screen capture implemented through the graphics card will show just a colored box where the video win
dow is. However, the AT&T system provides a way to capture video images directly from the video board.
Video boards with multiple input and output jacks give you flexibility. The Personal Video 1300 software lets you toggle between its two video inputs so that you can, for instance, switch to a document camera or a prerecorded video sequence.
Personal Video 1300 got an impressive VCBench test rating of 7.9 fps, but its snappy response with motion came at the expense of image quality. When there was a lot of movement, the picture got blocky. Still, it had the most fluid video next to Sun's ShowMe, and that's impressive for a system communicating over a 128-Kbps ISDN line.
Since we tested the Personal Video 1300, AT&T upgraded the quality of the camera, which should sharpen images. The company also introduced Vistium Personal Video 1200, which uses a software H.320 codec. Priced at $2500, it's intended to compete with Intel's ProShare videoconferencing system.
InVision Desktop Video/D
ocument Conferencing Software V.3.0
Invision software's strong point is its $595 price and its support of both LAN and long-distance communications. This Windows-based videoconferencing package's weak point is video quality. A software-only product, it handles several codecs and communications protocols and will support more in the future, including H.320. InVision Systems claims its package will work with almost any network operating system or architecture, including WANs, as long as there is TCP/IP or IPX support. It also will work over V.32bis or faster modems.
Because InVision is a software-only product, it relies on a third-party video capture board for digitizing a camera's video input and a Novell or TCP/IP network driver for communications. The program currently works only with Video for Windows compression cards but will support H.261 and MPEG codecs in an upcoming version. InVision Systems has found that decompressing the incoming video stream in software and displaying it on the monit
or directly through the graphics card provides the best video response on a LAN. With this approach, a faster host system with local-bus graphics makes a big difference in video response.
We tested InVision software with two 66-MHz 486 Dell Dimension 66Vs equipped with Intel Smart Video Recorder cards and Number Nine 64-bit VL-Bus graphics cards over 10Base-T Ethernet running NetWare. The Intel boards use Intel's Indeo software codec. The host system must also have Wave-compatible sound support. Like many current desktop videoconferencing products, it communicates only with itself.
You can vary video frame rate from 1 to 20 fps to fit the capabilities of your communications pipe. On a LAN, that sets the program's bandwidth requirement between 56 and 512 Kbps. With a modem, you must drop image size down to 100 by 100 pixels to get any kind of motion in the video. You can't count on the built-in compression abilities of V.32bis and faster modems, because the data is already compressed. In addition
, audio with a modem connection must be ``out of band,'' which means it must be a regular call over a separate phone line so you can talk while viewing the video image.
LAN Audio is unidirectional (because of the Wave sound cards), so InVision software provides settings for the threshold volume that puts a card in record mode, and a stop delay so that the card will wait out a set period of silence without stopping recording.
InVision comes bundled with Vision Graphics, a revamped version of Future Labs' Talk Show, which is a typical whiteboard program with features like a slide-tray interface for presentations. The package provides no application-sharing abilities but can do JPEG compression for data transfers. One feature that InVision software offers that most others don't is multipoint document conferencing (for up to 12 people). InVision Systems promises multipoint videoconferencing in the next version.
Despite testing on an isolated 10-Mbps Ethernet segment, InVision software gave th
e worst video performance of any system. Running on a 486DX2/66, the program's result on the VCBench test was a slow 2.9 fps. There were no problems with audio response and flow, but it tripped up on video periodically. We measured occasional delays between frames as long as 990 milliseconds.
The benchmark only confirmed our subjective impressions of the test system's video abilities. This program relies on Indeo software compression to digitize, compress, and uncompress the video while its own software spits the video out over the Ethernet. The computer simply couldn't keep up. The program should run better on a Pentium system. As systems become faster and codecs more efficient, this type of product will provide more-satisfactory videoconferencing.
PictureTel Live PCS 100
PictureTel has been in the videoconferencing business for a long time and is well known for its larger systems. PictureTel Live brings some of that technology to the desktop. Like AT&T's Personal Video 1300 system, it's
a competent two-board, ISDN add-in kit ($5995) that provides H.320 compatibility with hardware codec processing. It also has the best camera of all the systems we tested. If your company already has higher-end PictureTel equipment, PictureTel Live will work with it.
Unlike Personal Video 1300, PictureTel Live has SVGA (up to 1024- by 768-pixel resolution) on its video board. The video board and audio-communications board connect and must sit in neighboring ISA slots. (An EISA version is also available.) The audio board connects directly to an ISDN line, unlike that of the Personal Video 1300, which goes through an ISDN phone. A phone-like unit acts as the audio interface, providing a handset and speakerphone capability but no dialing pad.
The PictureTel Live system has an excellent camera with lots of features: lever adjustments for brightness, focus, and zoom, and a black, metal camera stand that lets you flip the camera straight down for shooting documents. You can also set the camera on your
monitor.
The PictureTel Live software is a modified version of Person-to-Person, IBM's Windows-based videoconferencing software. You wouldn't recognize it as such because PictureTel added its own software interface, which we found relatively easy to use. The software has a full set of features, including whiteboarding, application sharing, file transfer, meeting minutes, and a phone directory. When you pick up the audio unit's handset, a keypad comes up in a little box on the screen so you can dial. If the number isn't stored in the directory, you must dial using the mouse or keypad. We much prefer the AT&T approach where you dial on the attached phone.
One annoyance is the lack of a video-hold feature. If you want momentary privacy, you have to slide shut the camera's lens cap (the mute button on the audio unit cuts only sound).
The audio box has a handset and speakerphone, with a button to toggle the speaker on and off. A jack in the unit lets you plug in an optional headset. PictureTel
Live's audio unit stands out for the deafening volume it can produce on its speakerphone. This can be a real benefit for room conferencing.
Like AT&T's Personal Video 1300 system, the Personal Live system uses chroma keying to place the video image onto the screen. Unlike the AT&T product, PictureTel's system doesn't give you a way to capture a still video image, one of the most useful side functions of having a videoconferencing system on your desk.
The PictureTel Live achieved a respectable 4.9 fps on the VCBench test. The number is lower than that of the AT&T system because the PictureTel system maintains a higher image resolution during high motion (i.e., a clearer but jumpier video sequence). Both systems transmit reasonably good images when there is little motion, but we prefer AT&T's approach when there is motion (e.g., gesticulation)--it looks more real to us.
Because both systems are H.320-compatible, we called one from the other. Unlike almost everything else in this review, it
worked on the first try. Of course, none of the document or applications features worked because H.320 doesn't yet deal with these functions.
InPerson Desktop Conferencing
SGI's InPerson is one of the two LAN-based Unix products we tested. The $495 program works on SGI's video-equipped workstations. We tested it over a 10Base-T Ethernet using two SGI Indy workstations equipped with 20-inch displays. Like Sun's ShowMe, InPerson supports multipoint videoconferencing and makes use of the built-in video capabilities of the host workstation.
The Indy comes equipped with a camera and a microphone and has strong graphics-handling capabilities for its price (see ``Digital-Media Power,'' January BYTE). The nice feature of the Indy Cam digital cameras is that you can focus them. The Indy also has a clip-on mike, which looked unobtrusive and performed well.
InPerson communicates using TCP/IP over Ethernet with a proprietary compression standard, which limits communications to other SGI workst
ations. SGI is working on an ISDN version using H.320-compatible hardware compression that will talk with the outside world. The program already supports H.320 audio codecs.
The whiteboard function for document sharing integrates well with the rest of the program. You click a button to bring it up, and it appears as a part of the window that holds the live video. InPerson doesn't give you application sharing, although you can get the same function through a third-party shared X server. Support for file transfers is also clever: You drag a file icon from elsewhere in the user interface and drop it onto InPerson's Shared Shelf graphic to distribute the file to all connected parties.
Appropriate to a LAN-based videoconferencing package, InPerson provides a bandwidth meter so that you can gauge the effect of videoconferencing on LAN traffic. Displaying a full-size 320- by 240-pixel image at 20 fps requires about 2500 Kbps of bandwidth.
The InPerson package uses software video and audio codecs
, relying on the processing power of the host system's RISC CPU and graphics subsystem.
On the Indy, video is definitely adequate, coming up with a reading of 5.3 fps in the VCBench test. Although this is in the same ballpark as the two ISDN systems we tested, you get a much clearer image with InPerson. This is due mostly to the greater bandwidth available on a LAN, although the quality of codec algorithms and processing speed available from DSPs or the CPU also have a big affect on frame rate and image quality.
ShowMe
ShowMe provides full LAN videoconferencing abilities to Sun Microsystems' M series of workstations under Sun's Solaris operating system. The $1280 software package provides a full set of conferencing functions, including application sharing and whiteboarding. We tested the package on an isolated 10Base-T Ethernet segment using SparcStation models 5M and 20M (see ``SparcStation Overhaul,'' September BYTE). The M models of Sun's workstations come equipped for videoconferencing
; all you need is ShowMe.
The SunVideo card in a SparcStation M model does hardware video encoding but not decoding. The result is excellent video performance. Sun's microphone fell short in our estimation, however. Although designed to sit on your monitor, Sun provides no means of attachment. Because the weight of its cord frequently pulled it off the monitor, we fashioned our own Velcro attachment.
Compared with InPerson, the ShowMe interface is functional but without grace. It works, but it's not very much fun. ShowMe, however, has functions that InPerson lacks, such as application sharing. Its snapshot capability lets you send uncompressed still video images at 640- by 480-pixel resolution. This allows good detail in something like a close-up shot of a new product design. ShowMe also provides video that's virtually unaffected by motion in the image.
ShowMe's main window houses buttons for starting the video, audio, and whiteboard programs as well as the address book you use to call ot
her people. The whiteboard software uses the paint model for annotating documents. To start a conference, you open the address book, select the people who are participating, and then return to the ShowMe window and click on the call-conference button. You can have more than two parties in a conference.
ShowMe then calls the other people's systems. If you are one of the selected participants and are logged in and running ShowMe, you see a box pop up with the choice of either accepting or rejecting the conference request. Upon acceptance, appropriate connections are made. You then open the video and audio tools to see small windows pop up showing each of the other members of the conference. ShowMe's ``picture in a picture'' feature lets you nest the view of yourself into the corner of the view of one of the other participants, shrinking it down to do so. This saves screen real estate when you're conferencing with a group of people.
You can set frame rate as high as 30 fps and image size as high as
640 by 480 pixels--way beyond what the other tested packages can do. You wouldn't want to use these settings routinely because of the LAN traffic it would generate. You can also set audio for half- or full-duplex. With half-duplex, audio goes in only one direction at a time, triggered by who is speaking. Using half-duplex reduces bandwidth requirements. Many systems work only at half-duplex.
ShowMe provides a bandwidth meter so you can see the impact of your settings on LAN traffic. We measured LAN traffic independently and found that a typical 15-fps session with 320- by 480-pixel resolution created from 10 percent to 20 percent LAN utilization on an isolated 10-Mbps segment. You can't have too many sessions going on with any of these LAN-based systems unless you go to switching or a faster Ethernet.
The most outstanding thing about Sun's videoconferencing system is its amazing ability to maintain frame rate in spite of motion in the video. When running the software at 30 fps, the video coming
across the LAN connection was virtually identical to the video being generated.
What the Future Holds
The desktop videoconferencing market changes fast, with new products being announced almost every week. The current crop of products we tested all work well for video-assisted collaboration, despite the need for more standards and more-affordable bandwidth. The ISDN systems can also serve for long-distance meetings.
The technology is improving at a rapid pace, creating less expensive systems that can work better with less bandwidth. Within a year, you should see single-board videoconferencing products priced under $700 that work on standard phone lines. We can also hope for the quick arrival of standards for less expensive systems, LAN systems, and document conferencing.
As ATM services become widely implemented, there will be new standards constructed and new systems for multipoint videoconferencing. These high-end business systems will support 30-fps high-resolution video at a mo
derate cost. The market is still young and dynamic.
The Facts
AT&T Global Information Solutions
(Vistium Personal Video 1300)
1700 S. Patterson Blvd.
Dayton, OH 45479
(513) 445-5000
(800) 225-5627
InVision Systems Corp.
(InVision Desktop Video/Document Conferencing Software V.3.0)
8500 Leesburg Pike, Suite 300
Vienna, VA 22182
(703) 506-0094
fax: (703) 506-0098
PictureTel Corp.
(PictureTel Live PCS 100)
222 Rosewood Dr.
Danvers, MA 01923
(508) 762-5000
(800) 716-6000
fax: (508) 762-5245
Silicon Graphics, Inc.
(InPerson Desktop Conferencing)
2011 N. Shoreline Blvd.
Mountain View, CA 94043
(415) 390-1445
(800) 800-7441, Dept. D150
fax: (415) 390-6218
Sun Microsystems Computer Co.
(ShowMe)
2550 Garcia Ave.
Mountain View, CA 94043
(415) 960-1300
(800) 821-4643
fax: (415) 969-9131
Table: DESKTOP VIDEOCONFERENCING--COMPARING FEATURES (This table is not available electronically. Please see November, 1994, issue.)
The AT&T Personal Video 1300 and PictureTel Live prices are high because they include the necessary hardware to turn a Windows PC into an ISDN videoconferencing node. Also, both kits have hardware codecs for better video quality. InVision's Windows software provides videoconferencing over networks, near and far, but requires that the host PC have video hardware. SGI's InPerson and Sun's ShowMe are Unix programs that provide LAN videoconferencing for video-equipped versions of each company's workstations.
Photograph: Vistium
Photograph: InVision
Photograph: Vistium
Photograph: InVision
Photograph: PictureTel
Photograph: InPerson
Photograph: PictureTel
Photograph: InPerson
Photograph: ShowMe
Photograph: ShowMe
Eric Garland designed and wrote the BYTE Lab's videoconferencing frame-rate benc
hmark. You can reach him on the Internet at
ejg@christa.unh.edu
or on BIX c/o ``editors.''
Dave Rowell is a BYTE technical editor. You can reach him on the Internet or BIX at
drowell@bix.com
.
Face-to-Face Collaboration
