Single Channel Blind Source Separation of Polyphonic Signals in Sub-Gaussian Condition

An extension of blindly separating disjointed polyphonic signals by single channel independent component analysis (SCICA) in Sub-Gaussian condition is proposed. Nowadays single channel ICA can only be applied in the condition of mixed signal who has disjointed power spectrum density and source signals are sparse. It makes polyphonic signal presents part Sub-Gaussian distribution and is hard to blindly separate from Sub-Gaussian environment by single channel ICA. The distribution features (including probability density, kurtosis, power spectrum and signal interference ratio) of source signals, mixed matrix, mixed signals and separated signals are analyzed. When the kurtosis of Sub-Gaussian setting decreases, the SIR of polyphonic signal who exposes Sub-Gaussian distribution reduces sharply whereas the SIR of polyphonic signal who exposes Super-Gaussian distribution changes smoothly. More specifically, when mixed signals only present Gaussian distribution or Sub-Gaussian distribution in Sub-Gaussian condition, the polyphonic signal that shows Sub-Gaussian distribution cannot be separated by single channel ICA.