On many occasions, vendors have told me that their voice product for the WAN provides toll-quality voice. Every time I hear that phrase, I ask the same question: what does it mean?
The term toll-quality came into use about 20 years ago when T1 multiplexers first started transporting voice over private T1 lines. The original idea was that a private WAN could provide voice quality equal to that of the long-distance public switched network, which charged a toll for each minute of use.
Until recently, the telephone network was based on an analog infrastructure. Analog transmission is not particularly robust or efficient at recovering from line noise. Because analog signals degrade over distance, the public networks had to periodically amplify them. Amplification boosts both the voice signal and ambient line noise, resulting in degradation of the quality of the signal.
In response to the limitations of analog transmission, the carriers migrated to digital transmission using pulse code modulation (PCM) or adaptive differential pulse code modulation (ADPCM). In both cases, a codec converts the analog sound into digital form by sampling the analog sound 8,000 times per second and converting each sample into a numeric code.
PCM and ADPCM are examples of waveform coder/decoder (codec) techniques. Waveform codecs are compression techniques that exploit the redundant characteristics of the waveform itself. In addition to waveform codecs, there are source codecs that compress speech by sending only simplified parametric information about voice transmission (as opposed to a compressed version of the voice transmission); these codecs require less bandwidth. Examples of source codecs include linear predicative coding (LPC), code-excited linear prediction (CELP), and multipulse, multilevel quantization (MP-MLQ). The ITU-T standardizes coding techniques in its G-series recommendations.
These aren't the only forms of digital voice, however. For a number of years, vendors have been supplying voice compression equipment that uses other digital coding schemes. These schemes can compress voice call traffic down to as little as 8 Kbps -- but there's a price to pay, both in compression time and in quality. Over the years, the increasing speed of digital signal processors (DSPs), which are used to compress the voice, has allowed the use of better compression techniques and greatly improved voice quality at lower speeds.
This conversion of the long-distance network from analog to digital has improved the phone network to the point that it provides voice quality as good as, if not better than, most local networks, making the term toll-quality ambiguous. The problem is that this term is completely subjective. As we start to deploy packet voice that uses the same codecs as circuit-switched voice on the WAN, we need a common measurement of voice quality. The good news is that a measuring system already exists.
Each type of codec provides a certain quality of speech, which is a subjective judgment on the part of the listener. A common benchmark carriers use to determine the quality of sound produced by specific codecs is the mean opinion score (MOS). With MOS, a wide range of listeners judge the quality of a voice sample (corresponding to a particular codec) on a scale of one (bad) to five (excellent). The scores are averaged to provide the mean opinion score for that sample. The following table shows the MOSs for some standard codecs.
|Compression method||Bit rate (Kbps)||Framing size||MOS score|
While this table may help you determine the voice quality of a particular type of codec, it is not the final answer. Each vendor's implementation of these codecs may be different, resulting in higher or lower voice quality.
When you are evaluating packet or circuit voice solutions for your WAN, refer to this chart for a general idea of the voice quality of each codec. Ask your vendor for the results of any MOS testing done on its product. If the vendor has none, make sure you test voice quality in a real-world environment, on a network running other traffic. Generate some line errors to see how voice quality holds up under pressure.
In my next column, we'll talk about the impact of jitter and delay on voice quality.