Abstract: A sub-optimal channel estimation algorithm in orthogonal frequency division multiplexing (OFDM) system was proposed, to reduce the complexity of traditional minimum mean square error (MMSE) channel estimator, and to alleviate the performance deterioration of MMSE estimator due to the mismatch of the estimator-to-channel statistics. This algorithm utilizes time-average channel correlation matrix in place of exact channel correlation matrix and approximates the MMSE estimator by a low-rank structure. Based on subspace tracking, this algorithm adaptively estimates the noise power and the principal eigenspace of time-average channel correlation matrix, and asymptotically converges to the solution of optimal MMSE estimation. Further theoretical analysis indicates that delay subspace tracking algorithm is a special case of this newly proposed algorithm, and both theoretical analysis and simulation results illustrate that this algorithm can significantly improve channel estimation accuracy at low signal to noise ratio(SNR) with reduced implementation complexity.