Training sequence based channel estimation for two-way MIMO relay systems

Abstract

This thesis studies two issues for the two-way (TW) the multiple-input multiple-output (MIMO) relay systems (TWMRS), namely channel estimation techniques and the training sequence design.

The training based two-step channel estimation scheme has been proposed for two-way MIMO relay systems over uncorrelated Rayleigh channels. For a TWMRS with multiple relay antennas, the structure of the channel matrices is analyzed and can be estimated by one by one deriving the channel estimators related to each relay antenna in the equivalent one-way MIMO relay systems (OWMRS). For the proposed channel estimation scheme, we estimate the forward channel during the first phase by using linear minimum mean square error (LMMSE) method. During the second phase for the composite channel estimation, singular value decomposition (SVD) based maximum likelihood (ML) method is applied in two-way relay systems, which is extended from OWMRS. Therefore, the backward channel estimator is derived with the given forward channel and composite channel estimators. For the training sequence and relay gain design, due to the nonlinearity of the estimators and the closed form of mean square error (MSE) is unavailable, we propose a novel criterion based on Bayesian Cramér-rao Lower Bound (BCRLB) since BCRLB has been shown to tightly lower bound on the estimation MSE. The proposed novel criterion scheme aims to minimizes the trace of the asymptote BCRLB with respect to the individual channels. The BCRLB based training sequences and relay gain are though suboptimal, they are more tractable and amenable to handle, where the optimization process is efficiently performed under scalar variables. The numerical results have shown that the proposed channel estimation scheme with the BCRLB relay gain and training sequences has improved the MSE performance. Moreover, though the BCRLB based training sequences and relay gain are proposed for individual channels, the MSE performance for the composite channel estimator has also been improved. For correlated TWMRS, we consider two cases under white noise and colored disturbance respectively. For the first case, the channel estimators are derived by using a joint method, which is based on LMMSE and maximum a posteriori (MAP) methods. The training sequences for the second stage are proposed based on a novel approximated BCRLB criterion. For the second case, a novel two stage channel estimation scheme has been proposed. At the first stage, the backward channels are estimated with employment of LMMSE method. Subsequently, the optimal training sequence is straightforwardly derived with MSE criterion. For forward channel estimation, which is performed at the second stage, the estimators are obtained with the known estimated backward channels and the estimation error viewed as random variable. For training signal design at the second stage, a novel criterion has been proposed which is based on the asymptotic MSE conditioned on the estimated backward channels. Subsequently, the novel structure of the training sequences has been proposed. Finally, the numerical results show that the proposed training sequences can improve the MSE performance under both high and low spatial correlation channels.