5.3.4 ACELP Excitation encoder

26.2903GPPAudio codec processing functionsExtended Adaptive Multi-Rate - Wideband (AMR-WB+) codecRelease 17Transcoding functionsTS

5.3.4.1 Open‑loop pitch analysis

Same as 3GPP TS 26.190 but with the minimum and maximum pitch lags as functions of the internal sampling rate.

5.3.4.2 Impulse response computation

Same as 3GPP TS 26.190.

5.3.4.3 Target signal computation

Same as 3GPP TS 26.190

5.3.4.4 Adaptive codebook

Same as 3GPP TS 26.190 but with the minimum and maximum pitch lags as functions of the internal sampling rate.

5.3.4.5 Algebraic codebook

Same as 3GPP TS 26.190. The codebooks used in some basic core rates are the same as used in AMR-WB at 12.65, 14.25, 15.85, 18.25, 19.85, and 23.05, respectively. The codebook used in mono rate 208 bits/frame is the same as the one used in AMR-WB at 8.85 kbps (20 bit codebook). The codebook used in mono rate 240 bits/frame is a 28 bit codebook where two tracks contain one pulse each and two tracks contain two pulses each according to the following table:

Table 12: Potential positions of individual pulses in the 28-bit algebraic codebook

Track	Pulse	Positions
1	i0, i4	0, 4, 8, 12, 16, 20, 24, 28, 32 36, 40, 44, 48, 52, 56, 60
2	i1,	1, 5, 9, 13, 17, 21, 25, 29, 33, 37, 41, 45, 49, 53, 57, 61
3	i2, i5	2, 6, 10, 14, 18, 22, 26, 30, 34, 38, 42, 46, 50, 54, 58, 62
4	i3,	3, 7, 11, 15, 19, 23, 27, 31, 35, 39, 43, 47, 51, 55, 59, 63

5.3.4.5.1 Codebook structure

Same as 3GPP TS 26.190.

5.3.4.5.2 Pulse indexing

Same as 3GPP TS 26.190.

5.3.4.5.3 Codebook search

Same as 3GPP TS 26.190.

5.3.4.6 Quantization of the adaptive and fixed codebook gains

The adaptive codebook gain (pitch gain) and the fixed (algebraic) codebook gain are vector quantized using the same 7-bit codebook used in AMR-WB for modes 2 to 8. However, instead of using MA prediction to obtain the predicted gain g‘_c , it is found by directly quantizing the average innovation energy in the whole frame.

Let E_s(n) be the innovation energy (in dB) at subframe n, and given by

where N=64 is the subframe size, c(i) is the fixed codebook excitation, and E_i is the un-scaled innovation energy given by

An estimated innovation energy is computed and quantized, and used to find the estimated gain g‘_c . That is,

which is derived from the relation .

A correction factor between the gain g_c and the estimated one g‘_c is given by

The pitch gain, g_p, and correction factor  are jointly vector quantized using the same 7-bit codebook used in AMR-WB, and using the same error minimization procedure. That is, the gain codebook search is performed by minimizing the mean-square of the weighted error between original and reconstructed signal.

The estimated innovation energy is computed and quantized as follows. First, the LP residual energy is computed in each subframe n by

then the average residual energy per subframe is found by

The innovation energy is estimated from the residual energy by removing an estimate of the adaptive codebook contribution. This is done by removing an energy related to the average normalized correlation obtained from the two open-loop pitch analyses performed in the frame. That is

where is the average of the normalized pitch correlations obtain for each half-frame from the open-loop pitch analysis.

The estimated innovation energy is quantized once per frame using 2 bits, with the quantization levels: 18, 30, 42, and 54. Further, the quantized estimated innovation energy is constrained to be larger than E_max-37, where E_max is the maximum value of E_res(n) from the 4 subframes. This is done by incrementing by 12 (and the quantization index by 1) until or .

The quantized estimated innovation energy is then used to compute the estimated gain in each subframe is explained above.