Unbiased blind separation using the threshold nonlinearity
H Mathis, M. Joho · 2002
The stability analysis of blind signal separation algorithms using the threshold nonlinearity is extended to discrete distributions, such as used in digital communication systems. Conditions for the threshold parameter are derived. An extension to the standard algorithm mitigates separation coefficient bias, which is introduced by additive noise at the sensors. Simulations quantify the improvement in terms of steady-state interchannel interference.