Information theoretic comparison of training, blind and semi blind signal separation algorithms in MIMO systems

We consider the problem of comparing source/signal separation algorithms for MIMO wireless channels. The channel state is not known to the transmitter and the receiver. The receiver estimates the channel via a training sequence or uses blind methods to separate the signals. By comparing the 'capacity' of a composite channel, we find the optimal training size in training and semiblind methods. We also conclude that for Rician channels (with strong line of sight, LOS) (semi)blind algorithms outperform training based methods but for Rayleigh fading one should used training based methods only.