The LS channel estimation of OFDM systems can be viewed as a noisy observation of the true channel frequency response, so the noise component can be compressed through subspace projecting. In this paper, the substance of the improvement of LS channel estimation is analyzed when subspace projecting is enforced, and the general framework to perform subspace projecting in the context of OFDM channel estimation is given, based on which this method is exterded to non-LS channel estimation. When signal subspace varies with time, subspace tracking should be performed to maintain a good estimation of signal subspace. At the end of this paper, a subspace tracking based parametric channel estimator is presented, and computer simulation shows that this channel estimator outperforms its nonparametric counterparts.