Ergodic mutual information of full-duplex MIMO radios with residual self-interference

We study the theoretical performance of a full-duplex multiple-input multiple-output (MIMO) bi-directional communication system. We focus on the effect of the residual self-interference due to channel estimation errors and transmitter impairments. We assume that the instantaneous channel state information (CSI) at the transmitting nodes is not known and the CSI at the receiving nodes is imperfect. To maximize the system ergodic mutual information, which is a non-convex function of power allocation vectors at the nodes, a gradient projection (GP) algorithm is developed to optimize the power allocation vectors. This algorithm exploits both spatial and temporal freedoms of the source covariance matrices of the MIMO links between the nodes to achieve higher sum ergodic mutual information. It is observed through the simulations that the algorithm reduces to a full-duplex scheme when the nominal residual self-interference is low, or to a half-duplex scheme when the nominal residual self-interference is high.