Blind Separation of Ill-Condition Mixed Sources Based on Generalized Eigenvalue
Hai‐Lin Liu · Dianzi xuebao · 2006
It is proposed a new blind separation algorithm of ill-condition mixed sources.Observed signals are pre-processed through eliminating redundancy signals so that mixed matrix A is row full rank.Further,we propose a cost function that is a logarithm of a ratio of the covariance of a part sum of recovered signals and the covariance of recovered signals so that optimizing this function transform to solve a generalized eigenvalue problem.Under a loose condition,it is proved that any source signals theoretically satisfying the condition can be separated.The computer simulation shows its outstanding performance on blind source separation approach.