Most of blind channel identification algorithms cannot estimate the channel with common zeros and are sensitive to the channel order error. A Semi-Blind(SB) channel identification method is proposed in this paper. Through this algorithm, the channel matrix is decomposed into a same dimensional matrix and a unitary matrix in the form of their product based on Singular Value Decomposition(SVD). By using the received data and some known symbols to solve same dimensional matrix and the unitary matrix respectively, the closed form solution of channel response is obtained. The proposed algorithm effectively overcomes many limitations of blind channel identification algorithms. It avoids the selection of optimal weighted parameter of the traditional semi-blind methods. It is stable and robust to channel noise and channel order. Simulation results verify the effectiveness and superiority of the proposed algorithm.