Blind equalisation using approximate maximum likelihood source separation

Abstract

Blind equalisation of single input multiple output FIR channels can be reformulated as the problem of blind source separation. It has been shown that the natural gradient-based source separation method could recover the source successfully for ill-conditioned channels since it has the equivariant property. However, the effect of additive noise was not considered. A new approximate maximum likelihood source separation (AMLSS) method using the natural gradient is developed and applied to the task of blind equalisation. It is shown that the proposed method outperforms the BSBE in the presence of Gaussian noise.