Acoustic signal processing in shallow water environments is a challenging problem because of the presence of reverberation. Based on some models of reverberation, data from sensors array is pretreated to suppress reverberation. Considering reverberation as a sum of echoes of transmitted signal, the principal component inverse (PCI) algorithm deletes the largest singulars of data matrix, which is constructed from array data. However, estimating a threshold which is needed in PCI is difficult in practice. In this paper, two new subspace methods, Deleting Big Eigenvalues and Subspace Projection are proposed. The two novel methods, substituting automatic signal-number estimation for threshold estimation, are operated via eigendecomposition. According to a simulation which takes broadband linear modulated frequency signal as transmitted signal, these two methods show a similar performance but smaller computing quantity compared with PCI.