Abstract: Multi-user MIMO (multi-input multi-output) system has been studied in number of papers in recent years. In this paper,a novel multi-user transceiver algorithm is proposed for the system.In the multi-user MIMO system under FDD mode for the downlink , because of the channel estimation error and feedback channel errer, it is difficult to obtain the perfect channel state information (CSI) at transmitter. With the consideration of both the line-of-sight(LOS)component and the correlation at both the transmitter and users, a minimum sum MSE criterion MIMO transceiver design algorithm based on block diagonalization is proposed. It is supposed that there is only imperfect CSI at both ends of the link. Firstly, the proposed algorithm eliminates the co-channal interference (CCI) among all the users through block diagonalization(BD), and then computes the precoding and decoding matrices by iterativing for uers’ channel with channel estimation error. This paper deduces a channel estimation error algorithm and obtains an equivalent channel model based on training sequence, which makes the CSI obtained by transmitter and receiver more practical. We provide feedback channel is perfect,so the transmitter could obtain the channel estimation matrice with estimation error only. The analytic solution of Lagrange factor is obtained by using MMSE criterion, avoiding the complicated nonlinear equation, so that the computational complexity distinctly is reduced. Compared with the existing transceiver design algorithm, the proposed not only has good bit error, but also optimizes the transceivers of the users independently and flexibly.The influence of LOS component, correlation of the channel and channel estimation error on the performance of the algorithm is investigated. The simulation results shows that the proposed algorithm has a better BER performance compared with BD algorithm, only needs 4 iterations to convergence when the SNR is 20dB and the negative effect of the channel estimation error on the performance has been suppressed effectively.