In the diagram, "Hardware Abstraction Layer (HAL)" shows where the "Audio-Video COding/DECoding" (codec) module communicates with the OS and the GUI that handles audio and video display, whiteboarding, and control of call setup, audio volume and picture quality. A few years ago, when few host PCs offered graphics acceleration and had CPUs that operated in double-digit ranges (66 MHz or less), some hardware-based VT implementations would obtain window parameters from the OS and then drive the on-screen video display directly from the codec. But in today's "multimedia" host environment, PCs come with built-in graphics acceleration, caches and plenty of buffer memory. VTs can communicate with the OS through the HAL. They observe modular design rules and tend to be very stable, especially under Windows NT.

Compression is crucial to visual telephony. Raw PCM (pulse code modulated) output from a "broadcast quality" video camera clocks at 90 Mbps: a firehose emission that would swamp even Fast Ethernet. You'd need to compress the signal by a factor of 60 to send it over a T1 circuit. At speeds typically used by videoconferencing room systems, 384 Kbps, a 240:1 compression ratio is needed. And if a 28.8-Kbps modem were to support high-quality, full-screen, full-motion video delivery, 3125:1 compression would be required in real time. That's not yet possible, and that's why Internet video streaming, most of which still is tuned to analog modems in the home, can support only slow frame rates in small screen windows.

Present-day VT systems routinely achieve video compression ratios up to 1400:1. Some (the Zydacron Z350 board is an example) accomplish this in hardware by using digital signal processor chips on a plug-in PCI board. This somewhat pricey solution is mitigated by an ability to support high-performance video with a relatively modest host PC (a 166-MHz Pentium, for example). Lower VT system price points can be achieved by placing only simple video frame-grabbing circuitry on the PCI board and implementing codec functions in host software. This strategy is used by the Intel ProShare System 500, which requires support from a 400-MHz Pentium II processor to achieve 30-fps delivery of "sharp" pictures.

Video Codec Algorithms

An important trick used in video compression succeeds because raw input contains more color and luminosity information than the human eye can discriminate. One of the processes employed by a video encoder is to average the color information in adjacent pixels and discard half of the fine color distinction. An observer doesn't miss this in the reconstructed image.