A-law
The A-law method is a companding method used primarily in Europe for analog audio signals in the telecommunications sector , which is standardized in Recommendation G.711 of the International Telecommunication Union (ITU-T). In this case, the signal at the digitization with a nonlinear characteristic curve , the so-called A-characteristic curve , quantized . The aim is to achieve a higher dynamic range or a higher signal-to-noise ratio with the same resolution by resolving large signal deflections more coarse and small signal deflections more finely.
In North America and Japan, the similar µ-law method is used, which is similar to A-law, but is not compatible. For communication, e.g. B. in a telephone conversation between Europe and the USA, the digital data must be converted by appropriate converters.
The A curve
The companding of the analog signals takes place with a logarithmic quantization characteristic, the A characteristic. It is defined as a continuous function :
The inverse function is:
The following applies:
The A characteristic has a high dynamic at low signal levels and low dynamic at high levels.
The 13-segment characteristic
Recommendation G.711 also describes a linear approximation of the A characteristic curve in sections, which is much more suitable for use in digitally operating systems, the so-called 13-segment characteristic curve .
The intended PCM coding ( pulse code modulation ) with 8-bit words would lead to a level-independent resolution of 1/128 of the maximum amplitude if a linear characteristic curve was used. According to the 13-segment characteristic, small signal levels are now quantized to 1/2048 of the maximum amplitude, while large signal levels are only quantized with a resolution of 1/32. The signals are first digitized with 12-bit accuracy. The division into the segments was done in such a way that compression and expansion can be done very easily by a 12-bit to 8-bit conversion or an 8-bit to 12-bit conversion.
The audio signal is first linear 13-bit value in the form of an integer with sign digitized. This 13-bit value is represented as an 8-bit value according to the following table:
Linear input | output |
---|---|
s0000000wxyza ... | n000wxyz |
s0000001wxyza ... | n001wxyz |
s000001wxyzab ... | n010wxyz |
s00001wxyzabc ... | n011wxyz |
s0001wxyzabcd ... | n100wxyz |
s001wxyzabcde ... | n101wxyz |
s01wxyzabcdef ... | n110wxyz |
s1wxyzabcdefg ... | n111wxyz |
This coding can be viewed as a 1.3.4 floating point number with 8 bits (so-called minifloat ) and an integer value (1 bit sign, 3 bit exponent, 4 bit mantissa). The sign bit in the result is used inversely (n = not s).
In order to avoid the continuous transmission of zeros in the case of silence, the 8-bit words in the data stream are marked with 0x 55, i.e. H. 01010101 b exclusive-OR linked so that bits 2, 4, 6 and 8 (from the point of view of the ITU standard) or 6, 4, 2, 0 (from the point of view of the binary position system ) are inverted.