Power-Aware spatial multiplexing with unilateral Antenna cooperation

Abstract

This thesis offers a power aware solution to the problem of spatial multiplexing of independent data streams in a multi-antenna system, assuming that either the transmit antennas or the receive antennas are not allowed to cooperate and a sum power constraint is imposed. Cooperation is achieved by using spatial linear filters (beamformers), which are optimized jointly with the power allocation. To this end, the spatial channel characteristics must be known at the cooperating side. The design goal is to achieve individual signal-to-interference-plus-noise ratios (SINR) with minimal power consumption. This problem occurs, e.g., in a cellular wireless communication system, where a multi-antenna base station communicates simultaneously with several decentralized mobile terminals, each equipped with a single antenna. Downlink and uplink correspond to cooperating transmit and receive antennas, respectively. As a first approach to this optimization problem, the beamformers are assumed to be fixed. This allows for a characterization of the global optimum, which will prove useful for the following analysis. It turns out that there is a fundamental duality between uplink and downlink spatial multiplexing. In particular, the same SINR values can be achieved in both channels under the same power constraint. Hence, a solution of the downlink problem, which has a complicated analytical structure, can be found by solving the “smoother” uplink problem instead. A second important step towards a general solution is the investigation of joint beamforming in the absence of noise, as proposed by Gerlach and Paulraj [10]. This problem consists of minimizing a non-differentiable `∞ objective function. A key step in finding a global solution is to formulate an equivalent eigenvalue optimization problem and to show that each local minimum is a global minimum. Also, there is an interesting relation between optimal `∞ optimization and the alternative `1 objective proposed by [10]. The analytical results of this study lead to the development of globally convergent algorithms with strictly monotone iteration sequences. Typically, only a few iteration steps are required. The insight gained from studying interference balancing in the absence of noise is used to derive a power aware strategy for SINR balancing with minimal power consumption. This new algorithm has the same excellent convergence behavior as the one derived for the noiseless case. By choosing individual target thresholds, arbitrary SINR levels can be achieved as long as the problem is feasible. Necessary