A simple adaptive matrixing scheme for efficient coding of stereo sound

A generalized scheme for mapping stereo audio channels onto coding channels is considered in the paper. An improved joint stereo transform-based perceptual audio coder is proposed that employs this scheme instead of MS coding. The discussed frequency-dependent matrixing scheme uses a free rotation of the two-dimensional signal space adapted to the dominant direction of arrival within particular frequency bands. This direction is determined using 2-D principal component analysis (PCA) in subbands or groups of transform coefficients. Experimental results show a significantly increased compression efficiency compared to traditional MS coding based on a static matrix. The proposed technique is also more compatible with mono mode since the new M channel represents signal energy that is maximally focused within each band, and the residual energy in S channel is minimized.