BLIND SEPARATION OF INSTANTANEOUS MIXTURES OF NON STATIONARY SOURCES

Dinh­Tuan Pham, Jean­Franc›ois Cardoso
Dinh­Tuan.Pham@imag.fr, cardoso@tsi.enst.fr

Most ICA algorithms are based on a model of station­ ary sources. This paper considers exploiting the (possible) non­stationarity of the sources to achieve separation. We in­ troduce two objective functions based on the likelihood and on mutual information in a simple Gaussian non stationary model and we show how they can be optimized, off­line or on­line, by simple yet remarkably efficient algorithms (one is based on a novel joint diagonalization procedure, the other on a Newton­like technique). The paper also includes (limited) numerical experiments and a discussion contrast­ ing non­Gaussian and non­stationary models.