ON LOCAL CONVERGENCE OF A CLASS OF BLIND SEPARATION ALGORITHMS

A class of recursive stochastic gradient algorithms for blind separati on of dynamically mixed independent source signals are analyzed. The s tudied methods utilize correlations and high-order moments in order to enforce statistical independence of the separated signals. The local convergence properties of the schemes are investigated, and it is demo nstrated that local convergence is tied to positive realness of certai n mixing transfer functions.