Optimized data aided training in MIMO systems

We evaluate the achievable capacity of training-based MIMO systems, when the channel is allowed to be re-estimated, e.g., a decision-directed system. A lower bound on the capacity is derived for the single user block fading MIMO model for systems based on superimposed pilots, which include conventional time-multiplexed pilot based systems as special cases. The system is optimized with respect to the proportion of time and power spent on training and data, respectively, and how the power should be allocated over the block. It is found that the capacity can be very close to the fundamental capacity of the non-coherent channel, especially when the channel coherence time is short, and that a significant gain compared to conventional receivers is achieved at the expense of a linear increase in complexity.