G.711

A-law encoding thus takes a 13-bit signed linear audio sample as input and converts it to an 8 bit value as follows:

Where s is the sign bit, s is its inverse (i.e. positive values are encoded with MSB = s = 1), and bits marked x are discarded. Note that the first column of the table uses different representation of negative values than the third column. So for example, input decimal value −21 is represented in binary after bit inversion as 1000000010100, which maps to 00001010 (according to the first row of the table). When decoding, this maps back to 1000000010101, which is interpreted as output value −21 in decimal. Input value +52 (0000000110100 in binary) maps to 10011010 (according to the second row), which maps back to 0000000110101 (+53 in decimal).

This can be seen as a floating-point number with 4 bits of mantissa m (equivalent to a 5-bit precision), 3 bits of exponent e and 1 sign bit s, formatted as seeemmmm with the decoded linear value y given by formula

${\displaystyle y=(-1)^{s}\cdot (16\cdot \min\{e,1\}+m+0.5)\cdot 2^{\max\{e,1\}},}$

which is a 13-bit signed integer in the range ±1 to ±(2¹² − 2⁶). Note that no compressed code decodes to zero due to the addition of 0.5 (half of a quantization step).

In addition, the standard specifies that all resulting even bits (LSB is even) are inverted before the octet is transmitted. This is to provide plenty of 0/1 transitions to facilitate the clock recovery process in the PCM receivers. Thus, a silent A-law encoded PCM channel has the 8 bit samples coded 0xD5 instead of 0x80 in the octets.

When data is sent over E0 (G.703), MSB (sign) is sent first and LSB is sent last.

ITU-T STL[3] defines the algorithm for decoding as follows (it puts the decoded values in the 13 most significant bits of the 16-bit output data type).

void            alaw_expand(lseg, logbuf, linbuf)
  long            lseg;
  short          *linbuf;
  short          *logbuf;
{
  short           ix, mant, iexp;
  long            n;

  for (n = 0; n < lseg; n++)
  {
    ix = logbuf[n] ^ (0x0055);	/* re-toggle toggled bits */

    ix &= (0x007F);		/* remove sign bit */
    iexp = ix >> 4;		/* extract exponent */
    mant = ix & (0x000F);	/* now get mantissa */
    if (iexp > 0)
      mant = mant + 16;		/* add leading '1', if exponent > 0 */

    mant = (mant << 4) + (0x0008);	/* now mantissa left justified and */
    /* 1/2 quantization step added */
    if (iexp > 1)		/* now left shift according exponent */
      mant = mant << (iexp - 1);

    linbuf[n] = logbuf[n] > 127	/* invert, if negative sample */
      ? mant
      : -mant;
  }
}

See also "ITU-T Software Tool Library 2009 User's manual" that can be found at.[4]

The μ-law (sometimes referred to as ulaw, G.711Mu, or G.711μ) encoding takes a 14-bit signed linear audio sample in two's complement representation as input, inverts all bits after the sign bit if the value is negative, adds 33 (binary 100001) and converts it to an 8 bit value as follows:

Where s is the sign bit, and bits marked x are discarded.

In addition, the standard specifies that the encoded bits are inverted before the octet is transmitted. Thus, a silent μ-law encoded PCM channel has the 8 bit samples transmitted 0xFF instead of 0x00 in the octets.

Adding 33 is necessary so that all values fall into a compression group and it is subtracted back when decoding.

Breaking the encoded value formatted as seeemmmm into 4 bits of mantissa m, 3 bits of exponent e and 1 sign bit s, the decoded linear value y is given by formula

${\displaystyle y=(-1)^{s}\cdot [(33+2m)\cdot 2^{e}-33],}$

which is a 14-bit signed integer in the range ±0 to ±8031.

Note that 0 is transmitted as 0xFF, and −1 is transmitted as 0x7F, but when received the result is 0 in both cases.