The task is to separate N unknown sources S(t)
from M measured linear mixtures of sources X(t), that
contain unknown additive noise also.
The added noise vector is modeled as an convolutive mixture
of one noise signal n(t).
Fig. The model of source mixing with additive and convolutional noise.
In general the problem is very difficult to solve
without some apriori knowledge about the noise.
We have proposed a method for
simultaneous source separation and noise cancellation,
in case of an additive, convolutional noise
and when a reference noise is available.
For details see the paper in Neural Computing and Applications, 1997.
An earlier paper at the conference ICSP 1996 also addresses this approach.
Blind source separation with noise cancellation
June 1996