Criteria and algorithms for blind source separation based on cumulants
Jun Wang, Zhenya He · International Journal of Electronics · 1996
We have studied the problem of blind source separation, which is to recover unknown sources from observations without knowing the characteristics of the transmission medium, by using third- and fourth-order cumulants. A class of criteria is presented, which states that for independent sources, it is sufficient for separation that there is no more than one gaussian source. Algorithms are also developed based on these criteria, whose efficiency is verified by experiments.