Abstract: Channel estimation has a great impact on the performance of OFDMA systems. In order to improve the accuracy and accelerate the multi-user’s channel estimation, a fast blind channel estimation algorithm is proposed. The received data from multi-antennas in the frequency domain is piled up, and the subspace is built whose orthogonality is exploited to blindly estimate the frequency channel response. The accuracy and convergence rate are further improved through a sampling method. The algorithm is not limited to the subcarrier allocation scheme, and independent to the virtual subcarriers. Moreover, it has a low requirement on the timing of multi-user. Simulations show that the algorithm effectively improves the convergence rate while keeping a considerable performance compared with that of existed one, and it’s insensitive to the channel order.