The analysis of estimation error of non-causal training based on a unified error model

An important problem in channel estimation of time-varying channels is how to reduce the influence of the channels time variation on the channel estimation error. And previous works give us some hints in a time-varying Gauss-Markov Rayleigh fading channel by using both the training sequences at the each boundaries of a data sequence to train the channels in-between, while traditionally a training sequence is only intended for the training of the upcoming channels. In this paper, by analyzing the source and expression of the channel estimation, we compare our error model with the model given in [11] and we find that our strategy outperforms it in both estimation quality and capacity lower bound. And we reach the conclusion that the estimation error model in [11] is a special case of our unified error model in high SNR scenario.