Separation of a polynomial phase signals mixture using sparsity

This paper addresses the problem of blind source separation in the under-determined case (i.e. less sensors than sources) using a sparse component analysis (SCA) approach. The sources here are considered to be real polynomial phase signals (PPS). Our algorithm consists in including to the classical sparse component analysis a thresholding step to cancel high level noise and a convenient linear transformation which makes the signals become sparse. The simulations results reveal that using a wavelet packet transform, we can separate efficiently a mixture of six polynomial phase signals with only two sensors, even for a low SNR