Sparse channel parameter estimation based on superimposed training

A new channel estimation method based on superimposed training (ST) is proposed for the sparse frequency-selective multi-path block transmission channel, which can gain accurate channel parameters without any loss of bandwidth. With this method, first the LS solution of time domain channel impulse response (CIR) is obtained by using the superimposed training; then the channel length and the time delay of each path can be determined with the generalized Akaike information criterion (GAIC); finally the values of non multi-path position in CIR are set to zero. The new method reduces the effect of additive white noise on channel estimation and improves the precision of channel estimation. Through a series of simulations, we showed that, compared with the pure channel estimation based on ST, the proposed method reduces channel estimation error dramatically and improves the system performance.