Network Computingwww.networkcomputing.com

Standard voice on digital lines utilizes the G.711 standard, which uses PCM (pulse code modulation) to sample the incoming analog voice and place it on the digital line. Typically, this uses 56 Kbps or 64 Kbps of bandwidth per call, depending on the line signaling used (channel associated or clear channel). At this rate, a full ATM DS-1 can be filled up after only about 20 calls, with no room left for data traffic.

The G.726 compression standard uses an Adaptive PCM (adpcm) method to sample the analog signal. By sampling less often, an analog signal can be reduced to between 16 Kbps and 40 Kbps, depending on the quality you desire. Obviously, the closer you get to the sample rate of G.711 (8,000 times per second), the better the quality. Voice quality can seriously degrade as the sample rate gets lower. Because the IAD is sampling the analog waveform less often, sounds can be completely left out if they occur between any two samples, giving signals a muddied quality.

The G.729 standard was created to combat this problem by using ACELP (algebraic code-excited linear prediction) compression techniques. Instead of sampling the voice signal at even intervals, ACELP looks at the waveform as a whole and tries to describe the shape of the waveform using algebraic information. This allows better sound quality with the advantage of a smaller data stream--only 8 Kbps.

While G.729 sounds great, offering good quality at low bandwidth usage, there are trade-offs. G.729 requires a much higher CPU utilization for each call than G.726. As IADs scale in voice capacity, higher-powered CPUs are required to compress the higher number of voice calls, resulting in an increased price per port for the device. Bottom line: You be the final judge.