This paper proposes a fourth order cunmulant of tensor decomposition method to solve the problem for blind identification of underdetermined separation.First

the fourth order cumulant of covariance is constructed based on observe signals.By assuming that the source signals are independent

and are of zero mean

the hierarchical alternating least squares method is used to decompose fourth order cumulant of tensor into the nonnegative Tucker.Then

the mixture matrix is acquired by the parameters of nonnegative Tucker model.The source is also acquired by solving the pseudo-inverse of mixture matrix.Finally

by using the speech signals and the biomedical signals

two numerical examples are conducted to demonstrate the effectiveness of the proposed method.