To determine how thes e systems would perform over public networks in a real-world environment, we tested the systems between our San Mateo lab and University of Wisconsin lab over a 512-Kbps frame relay connection hosted by CompuServe. Taking it a step further, we tested the systems over a 128-Kbps connection to a public ISP from Wisconsin and conferenced via the Internet to our San Mateo lab. Out of fairness to the vendors, we did not place these wide area numbers in the performance chart because the public nature of these networks can cause significant delays that cannot be replicated for each test. The results were mainly used to let us identify where the hurdles actually lie: in the videoconferencing itself, the codec, the hardware or the actual medium used (See "A Big Hurdle: Network Latency," on page 108).

The systems were tested for interoperability by connecting each unit with the other participant's conference systems, initiating calls first from one vendor and

then in reverse. Collaboration was tested in the same

manner and in a multipoint configuration with all vendors participating in data collaboration simultaneously. We used a Network General Corp. Sniffer running version 4.52 software to analyze traffic between the connection and troubleshoot interoperability.

We measured audio and video quality using Network Computing's video benchmark, a VHS videotape consisting of a series of time-stamped video patterns. These include both still- and motion-based patterns that expose alias-related problems, video blocking and frame jitter. The benchmark also has a series of active conversations, people moving and speaking at various speeds, and color patterns and audio patterns. After establishing a videoconferencing session, the benchmark video was transmitted into the video capture card as if it were receiving the signal from the camera. The image was then captured and recorded on the receiving end by a Antec TVator Pro external video capture device. This tool intercepts the VGA signal to the monitor and transmits the s ignal to a composite line for VCR recording.

We used a Mitsubishi HU580 HiFi VCR to capture the received signal and analyze the videoconferencing recordings on a frame-by-frame basis. This method let us accurately count frame rates based on the video time stamp. Audio was determined partly by perception and partly by easily measurable qualities, such as obvious glitches in signal and echo. The audio results were reported on a scale from one to five, with one representing no audio reception and five for toll-quality, continuous audio signal.