Multiple antenna wireless systems: Capacity and user performance limits

A wireless broadcast channel is defined by a central transmitter sending independent data to multiple receivers. If the central transmitter is equipped with multiple transmit antennas, it can send data simultaneously to multiple receivers, creating the potential for a unified view of the traditionally layered tasks of scheduling and subsequent transmission of user data using signal processing. Such a "cross-layer" perspective forms the motivation for this research.;First, numerical methods to evaluate the capacity of a multiple antenna broadcast channel, using dirty paper coding, are proposed. Dirty paper coding requires solving for the optimal transmit covariance structure of the input signal. Obtaining this optimal covariance structure is computationally a complex problem. Numerical algorithms (based on gradient projection) to solve for the capacity-achieving transmit covariance structure are proposed. With this optimum transmit covariance, transmit pre-filters must be designed that allow capacity-approaching transmission over the multiple antenna broadcast channel. These transmit pre-filters, however, result in a power increase. Techniques to reduce this power increase are proposed.;Next, this dissertation establishes that opportunistic scheduling (in various forms) provides an aggregate throughput advantage that scales as O(log logK), where K is the size of the user population. This general scaling law is largely independent of the physical transmission strategy employed.;Finally, the main theoretical result in this dissertation is the development and analysis of the multi-user EXP scheduling rule (M-EXP) motivated by the physical model for multi-user transmission over a multiple antenna wireless broadcast channel. The stability region for transmission over a multiple antenna broadcast channel is a convex combination of convex polytopes. This M-EXP scheduling rule is shown to be throughput optimal (i.e. achieves the largest possible stability region) and pathwise optimal (i.e. minimizes workload and maximum queue length). The geometry of the problem does not permit traditional stability analysis in the fluid timescale directly. Hence a timescale separation argument is used to construct appropriate limiting processes on a 'finer' timescale. This finer timescale construction allows use of classical quadratic Lyapunov techniques to establish a state-space collapse. Finally, using this state-space collapse, these finer timescale optimality properties are shown to "map" to the fluid and diffusion timescales as well.