To satisfy the ever increasing demand for high data rates, multiple antenna transmission techniques have been widely adopted in recent mobile communication standards since they promise a linear capacity increase in the number of antennas. However, to leverage this potential capacity gains, the propagation channel has to be well-conditioned and the interference has to be efficiently mitigated. It is well-known that simultaneous transmissions to multiple spatially scattered receivers create a channel with higher capacity than transmission to a single receiver due to the increased spatial diversity. The latter transmission scheme is called single-user MIMO, while the former scheme is usually referred to as multi-user MIMO and is the topic of this dissertation. In MU-MIMO, the signals are precoded such that the interference at the receivers is minimized. Efficient precoding techniques require knowledge of the downlink channel state which is typically only available at the receivers. Therefore, the downlink channel state has to be made available to the transmitter through feedback from the receivers over a limited rate channel resulting in more or less accurate channel state information at the transmitter (CSIT).
In this thesis, we study MU systems where each receiver has a single antenna, called MU-MISO. We focus on linear precoding techniques, since they offer a good performance/complexity trade-off. More precisely, we consider optimal (maximizing the weighted sum rate) linear precoding, matched-filter precoding, zero-forcing (ZF) precoding, regularized ZF precoding and a practical scheme coined CUBF, where the precoding matrix is unitary with constant modulus entries. Due to the random propagation channel it is difficult to gain insight into the system behavior which is necessary to solve important problems such as the optimal amount of feedback. Typically, the random signal-to-interference plus noise ratio (SINR) can be bounded by a deterministic quantity considering the average system behavior and a specific precoding scheme. However, another way to render the SINR independent of the channel realizations is to assume that the number of transmit antennas M and receivers K is large while their ratio remains bounded. This assumption is well motivated since the current LTE-A standard already defines up to eight transmit antennas. In this dissertation, we develop a general and consistent framework for the study of linear precoding techniques in large MU-MISO systems under a wide range of channel propagation environments and imperfect CSIT. We provide the necessary tools from large dimensional random matrix theory to derive deterministic equivalents of the random SINR, i.e., SINR approximations independent of the channel realizations that are almost surely exact as M and K go to infinity. These SINR approximations can be applied to solve a variety of practical optimization problems of which several are presented in this work. Simulation results show that these solutions are close-to-optimal even for small system dimensions. Our framework forms the basis for the study of more complex systems like MU-MIMO and multi-cell MU-MIMO and opens up new ways to analyze limited feedback systems.