Abstract: The noise distribution of the array signal may be inconsistent due to the complexity of the underwater acoustic environment. When the noise powers of the array elements are different, there will be an error between the eigen subspace of the real target and the eigen subspace obtained by the eigen decomposition of the array covariance matrix, thus resulting in the degradation of the performance of the beamforming algorithm utilizing the eigen subspace. In this paper, a new eigen-subspace estimation method, realized by setting the diagonal of the array covariance matrix as zero, is proposed to overcome the discrepancy of the eigen-subspace for eigen-decomposition estimation under non-uniform noise conditions. By applying this method to the characteristic space beamforming algorithm, the azimuth resolution ability of the characteristic space beamforming algorithm can be highly improved under inconsistent noise conditions. Simulation and experimental results verify the feasibility and effectiveness of the method proposed in this paper.