Sum-power minimization beamforming for dense networks

Abstract

The employment of dense networks is a promising solution for the upcoming 5G systems. The use of a larger number of base stations (BSs) per unit area provides a reduction in transmission distance and can significantly improve spatial multiplexing. However, the densification also brings worries mainly related to higher interference due to the reduced distances.

This master thesis aims to present ways to manage interference in dense scenarios by using sum- power minimization beamforming. More specifically, we focus in two aspects of dense networks: The solution of large-scale problems and the management of the cross-link interferences in dense networks that employ dynamic time division duplex (TDD).

For the first aspect, we present an performance analysis for a alternating direction method of multipliers (ADMM)-based solution for the beamforming,which is considered to be well adapted to large-scale optimization. In the simulations we compare the ADMM solution to a well known semidefinite programming (SDP) solution in several network configurations. The results indicate that the ADMM approach has faster convergence for large-scale scenarios when modest accuracy is required.

For dynamic TDD scenarios, we propose solutions for different beamforming problems. In the first case, we aim to protect the uplink (UL) communication by forcing a constraint on the BS to BS interference power while guaranteeing downlink (DL) signal-to-interference-plus-noise ratio (SINR). We propose a centralized and a primal decomposition based distributed solution. The simulation results show that UL performance is improved and DL SINR targets are guaranteed, and that the distributed solution iterates towards the centralized one, while feasible beamformers can be obtained at intermediate iterations at the cost of suboptimal power.

In the second dynamic TDD problem, we aim to guarantee a minimum SINR for UL and DL users. We propose a centralized and an ADMM-based distributed solution. The simulation results show that both approaches achieve good performance and the distributed solution iterates towards the centralized one, while the signaling load can be controlled by fixing the number of iterations at the cost of close to optimal power and SINR performance.