The hybrid schemes for channel estimation are considered as the flexible structure satisfying demand of high data rate transmission in the multiple-input multiple-output (MIMO) technique. Generally, the training-based least square (LS) algorithm is chosen in wide applications from the first generation wireless network to the newest one of MIMO. For resolving the bandwidth efficiency problem, we can combine the LS estimation with the pure blind techniques such as second-order statistics (SOS). In this paper, we attempt to find the hybrid scheme which is so-called semi-blind channel estimation and outperforms than the conventional LS based approaches. Simulation results confirm our modified scheme and illustrate that the subspace based new semi-blind channel estimation is capable of improving the performance of the overall system.