5.5.2 Stereo Signal Mid-Band Processing

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

5.5.2.1 Principle

The encoder takes the mid band mono and right channel signals, and , and inverse filters it with the core codec LPC filters derived from the mono signal. In the residual domain a shape constrained FIR filter is computed for approximating the side signal. The filter is computed by means of the covariance method using a novel spectral shape constraint. A new filter is computed for each Ldiv=256 sample frame with an analysis frame of 320 samples. The energy of the filter is smoothed to avoid sudden energy changes.The smoothed filter is quantized with a multistage predictive vector quantizer (MSPVQ).

The mono residual signal is filtered with the quantized filter and gain factors are computed for the left and right channels respectively.

5.5.2.2 Residual computation

The residual signal is computed according to

The quantized and interpolated LPC coefficients from the core codec are used in the inverse filter operation, in addition an extra subframe is computed for the overlapped analysis section. The residual side signal is computed as

5.5.2.3 Filter computation, smoothing and quantization

The filter is computed that minimizes the expression under the constraint of a spectral null at 0Hz. The filter coefficients for the filter are computed with the well know covariance method using a modified cholesky algorithm taking into account the shape constraint.

To avoid to fast changes in the filter energy, the filter energy is smoothed over time. The filter energy is first computed as

the new filter energy is computed such that the filter energy is saturated to 16 and that transitions between frames are limited within a +/- 1.5 dB interval.

After the smoothing operation the filter coefficients are quantized using a predictive multistage vector quantizer.

5.5.2.4 Channel energy matching

The quantized filter coefficients are used to filter the mono signal excitation in order to get an initial estimate of the left and the right channel excitation signals. These estimates are computed as

The energy matching for the left and right channels is computed as

The computed gains are adjusted in case of anti-correlation by computing a correlation gain

if < 0dB the gain matching factors are adjusted according to

and

The energy matching factors are quantized using a two-dimensional vector-quantizer.