VoIP voice quality often best
When Harry Nyquist of Bell Labs developed the Pulse Code Modulation (PCM) algorithm in 1928 to translate analog voice signals into digital equivalents, his design rested on several fundamental signaling characteristics:
- He chose a frequency range of 4 kHz for voice information (and although the human voice covers a broader range of possible frequencies, this turns out to be enough range to make human conversation quite intelligible)
- To capture the proper degree of resolution, he decided to sample voice information at double the frequency range, or 8,000 times per second. Thus, PCM grabs a chunk of data every 0.125 milliseconds (8,000 * .000125 = 1.0).
- Each sample grabbed used 8 bits worth of data, so overall bandwidth required is 8000 * 8, or 64,000.
This explains why standard digital telephony requires 64 kbps of bandwidth for voice communications, in case you didn't already know.
Unfortunately, Harry didn't have access to modern digital signal processing (DSP) technology when he designed his algorithm back in 1928. In a word, compression makes it possible to ferry intelligible voice communications in narrower ranges of bandwidth in packetized VoIP streams. In fact, I've heard numerous reliable reports--and experienced for myself--that using Adaptive Differential Pulse Code Modulation (ADPCM) for voice encoding even at rates as low as 16 kbps can produce
The factors that affect intelligibility of voice communications as bandwidth shrinks come primarily from three factors:
- Delay: As packets travel from sender to receiver, the number of hops and the distances that such packets must travel contribute to the amount of time it takes, on average, for them to make the journey. The longer this takes, the longer the delay (also known as latency); the longer the delay, the more issues like echo and over talking start to affect conversation quality. Smaller bandwidths are more subject to this phenomenon because they can send less data over time, making delay more apparent.
- Jitter: Because not all packets follow the same route from sender to receiver, there's often some variability between the slowest one-way trip time and the fastest one-way trip time. To keep signals flowing smoothly, VoIP circuitry has to smooth out jitter, usually by buffering ahead long enough to compensate for the difference between the fastest and slowest one-way trip times. When this difference exceeds buffer size, drop-outs are inevitable. Smaller bandwidth sometimes means smaller buffers, so that larger jitter values are more likely to cause drop-outs.
- Packet loss: When delays cause VoIP packets to arrive out of sequence, or when packets get lost en route from sender to receiver, older packets are discarded to accommodate newer ones. Either way, data gets lost. Lose too much data and conversation becomes choppy, or suffers from outright interruptions. Smaller bandwidth uses less packets to move data, and is thus more subject to problems when losses occur.
Although these characteristics make it more difficult to get best results when higher compression ratios (and lower bandwidth) are used on voice data, voice quality on modern networks can be surprisingly good, even on IP networks where no QoS or grade of service provisioning applies to voice communications, as long as a network is not heavily utilized. The great thing about ADPCM is that bandwidth (or compression ratios) can be negotiated, so that you can use smaller channels when network conditions are favorable, but switch to larger channels when conditions deteriorate sufficiently to cause voice quality to degrade. This permits better overall utilization of bandwidth, and permits you to outperform normal digital PSTN voice signaling under most circumstances.
Thomas Alexander Lancaster IV is a consultant and author with over ten years experience in the networking industry, focused on Internet infrastructure.
This was first published in December 2001