Abstract: This paper proposes an efficient iterative detection algorithm for MIMO systems, based on the Soft Fixed-complexity Sphere Decoder (SFSD). The algorithm updates the soft output of SFSD using the feedback of the decoder, thus obtaining substantial iterative gain. The number of the branch metric calculation is reduced by exploiting the feature of the multi-level bit mapping constellation. The steps of the detection ordering determination and the QR decomposition are combined, in order to reduce the number of matrix inversion operations in the preprocessing stage. Simulation results in the Long Term Evolution (LTE) downlink system show that, the performance of the proposed algorithm approaches very close to that of the Maximum A Posteriori (MAP) detection with much lower complexity.