Blind separation of sources: A nonlinear neural algorithm

International audience; In many signal processing applications, the signals provided by the sensors are mixtures of many sources. The problem of separation of sources is to extract the original signals from these mixtures. A new algorithm, based on ideas of backpropagation learning, is proposed for source separation. No a priori information on the sources themselves is required, and the algorithm can deal even with non-linear mixtures. After a short overview of previous works in that field, we will describe the proposed algorithm. Then, some experimental results will be discussed.