A low-complexity method for direction of arrival(DOA) estimation based on estimation signal parameters via rotational invariance technique(ESPRIT) is proposed.Instead of using the cross-correlation vectors in multistage Wiener filter(MSWF),the orthogonal residual vectors obtained in conjugate gradient(CG) method span the signal subspace used by ESPRIT.The computational complexity of the proposed method is significantly reduced,since the signal subspace estimation mainly needs two matrixvector complex multiplications at the iteration of data level.Furthermore,the prior training data are not needed in the proposed method.To overcome performance degradation at low signal to noise ratio(SNR),the expanded signal subspace spanned by more basis vectors is used and simultaneously renders ESPRIT yield redundant DOAs,which can be excluded by performing ESPRIT once more using the unexpanded signal subspace.Compared with the traditional ESPRIT methods by MSWF and eigenvalue decomposition(EVD),numerical results demonstrate the satisfactory performance of the proposed method. A low-complexity method for direction of arrival (DOA) estimation based on estimation signal parameters via rotational invariance technique (ESPRIT) is proposed. Instead of using the cross-correlation vectors in multistage Wiener filter (MSWF), the orthogonal residual vectors obtained in conjugate gradient (CG) method span the signal subspace used by ESPRIT.The computational complexity of the proposed method is significantly reduced, since the signal subspace estimation primarily needs two matrix vector multiplication at the iteration of data level. Futuremore, the prior training data are not needed in the proposed method. To overcome performance degradation at low signal to noise ratio (SNR), the expanded signal subspace spanned by more basis vectors is used and simultaneously renders ESPRIT yield redundant DOAs, which can be excluded by performed ESPRIT once more using the unexpanded signal subspace. Compared with the traditional ESPRIT methods by MSWF and eigenvalue decomposition (EVD), nume rical results demonstrate the satisfactory performance of the proposed method.