QR-decomposition with M algorithm is a promising technique for Multi-Input Multi-Output (MIMO) systems. When the number of candidates is insufficient, its performance is, however, degraded due to an inaccurate Log Likelihood Ratio (LLR). To address this problem, an enhanced soft LLR calculation scheme is proposed for soft-output MIMO detection. Based on QR-decomposition, the proposed algorithm can efficiently obtain a soft information of Minimum Mean-Squared Error (MMSE) equalization. With this information, an optimal weighted combining method is derived in an MSE sense. In addition, the authors compute a more reliable clipping value from the soft information of MMSE equalization.