Multichannel matching pursuit and applications to spatial audio coding

Sparse signal modeling has been considered widely in the literature. In this paper, we discuss an extension of the matching pursuit sparse modeling algorithm to the case of simultaneously approximating multiple data signals; we outline the algorithm for general and for sinusoidal dictionaries. We then apply multichannel sinusoidal pursuit (M-SP) to spatial audio coding (SAC). In most SAC schemes, multichannel audio is coded by forming a downmix signal, compressing the downmix with a legacy coder, and adding side information about spatial properties of the input audio. In the proposed M-SP system, a multichannel model of the input is used to derive the spatial information as well as a parametric model of an appropriate downmix signal. This joint spatial-parametric approach provides a different multichannel audio coding paradigm than that of previously described SAC methods.