An unconstrained single stage criterion for blind source separation
Adriana Dapena, Luis Castedo, Carlos J. Escudero · 2002
This paper addresses the problem of separating an unknown linear mixture of signals. A new algorithm is presented to adapt the coefficients of the linear combiners that perform the signal separation. The method is based on the well-known Shalvi and Weinstein (1990) criteria for blind equalization to which a cross-term has been added to prevent the same signal be extracted by several linear combiners simultaneously. The unconstrained criterion stationary points are analysed, the particular choices of the criteria are shown, a stochastic gradient algorithm is derived to compute the optimum separating parameters, and simulation results are presented.