Abstract: The performance of joint detection depends highly on the accuracy of channel estimation. In an intra-frequency network, inter-cell interfering signals are treated as white noise in conventional channel estimators, which degrades the accuracy of channel estimation. In this paper, a Minimum Mean Square Error (MMSE) based joint multi-cell channel estimation algorithm is proposed. By including the strong interfering users of adjacent cells into the channel estimation matrix, the noise power is significantly reduced which guarantees more precise channel estimation. The newly proposed algorithm can be applied to both multi-cell joint detection and single cell joint detection. Compared with conventional Steiner channel estimation, the proposed algorithm shows much higher estimation accuracy and then greatly improves the performance of the TD-SCDMA system.