Skip to Main Content
Practical MIMO channels often exhibit structure in both space and time, i.e. a spatio-temporal structure. The potential of exploiting this structure in training based schemes is studied using a common ray-based channel model that captures parts of the structure observed in measurements. A lower bound, the Cramer-Rao lower bound (CRB), on the channel estimation error and a lower bound on the capacity are used to study the potential gain in exploiting channel structure. It is found that the training based capacity may be substantially increased since a more parsimonious channel model with fewer parameters to estimate can be used. Numerical evaluations indicate that the capacity grows with the number of antennas, similar to the case of a known channel if the structure is exploited. If it is not exploited, the training-based capacity reaches a maximum after which it decreases with the number of antennas. Furthermore, the temporal structure can be used to interpolate or predict the channel between training instants and it is found that prediction can improve performance for training based schemes.