Abstract: Discussion of blind signal separation problem under underdetermined case (i.e., the case of less observed signals than sources) is presented. First, a formula to calculate the normal vector of any hyperplane is given and a mixing matrix recovery algorithm based on the normal vector of any hyperplane is proposed. Second, for audio signal, k-source intervals are introduced and a method to detect them is proposed. So, the algorithms under the k-SCA condition are extended to blind non-sparse signal separation. To reconstruct the sources, a new algorithm is proposed to simplify the l1-norm solution. Several experiments demonstrate the performance of the proposed algorithm.