Two parameters influence the signal replication quality of the resulting digital representation.
- Sample rate (take signal probes regularly at uniform time intervals)
- Quantization (map the analog probe values into n-bit signal steps (binary encoded magnitude intervals))
It was found, that voice is still sufficiently understandable when the transmitted frequency spectrum ranges from 300 Hz up to 3,4 kHz.
This is the frequency, which is targeted in the ITU-T Recommendation G.711 - titled "Pulse Code Modulation (PCM) of voice frequencies".
The original G.711 recommendation started in 1972 and was approved in November 1988. Meanwhile, there are two ammendments added:
- G.711 Appendix I ( September 1999) on "A high quality low-complexity algorithm for packet loss concealment with G.711" and
- G.711 Appendix II ( February 2000) on "A comfort noise payload definition for ITU-T G.711 use in packet-based multimedia communication systems".
The standard PCM sample rate is choosen to 8000 samples/second (often called "8kHz sample rate") with 50ppm precision. This is due to the Nyquist–Shannon sampling theorem, where the sample rate must be at least twice the highest encoded frequency component. (f_sample >= 2 * f_signal). With 3,4 kHz maximum voice frequency, the 8k sample is sufficient and allows for non-ideal low-pass filtering of the source signal spectrum.
Each analog sample is digitized into a 8-bit code word. This quantization allows for 256 magnitude values, where the original sample is mapped onto the closest digitally represented magnitude value - a "stairs" approximation of the original signal curve.
The resulting voice frequency equivalent digital signal" - called "DS0" is 64kbit/s. This is the base channel rate for PDH and SONET/SDH digital multiplex transmission hierarchy systems.
However, the stairs (magnitude steps) are not equally distributed across the magnitude range. This is due to the fact, that quiet/hushed signals encode more voice information than loud signal components.
Two encoding schemes are defined, which weight the magnitude step granularity towards the quiet amplitudes. They are called "A-law" (mostly used in Europe) and "µ-law / mu-law" (mostly used in North America and Japan). Signal exchanges between countries based on differing encoding laws are defined to use A-law encoding at the interconnection.
To achieve this encoding and recoding a 14 bit uniformly distributed PCM (uniform PCM) can be implemented. Those codes are then mapped into the A-law or µ-law 8-bit code words resulting in the fixed 64kbit/s encoding rate.
