Decentralized strategies for beamforming optimization in wireles mobile networks

Abstract

The ever-growing demand increase of wireless mobile networks drives the evolution and shapes future communication technologies. Next network generations must deal with increasingly stringent requirements regarding the number of connected devices, data throughput, latency, and reliability. Among strategies to enhance network efficiency, we highlight beamforming, which aims to focus a signal beam in a specific direction while avoiding others, and user scheduling, which optimizes the selection of users to transmit using a given resource. Network-wide efficient algorithms must consider cooperative optimization in order to coordinate network entities into a common global objective. Due to the distributed nature of networked systems, centralized coordination is in many cases not suitable, and decentralized solutions are desired. Based on that, the main objective of this thesis is to design decentralized strategies for optimizing beamforming and user scheduling in scenarios of next-generation mobile systems. We divide this thesis contents into two independent parts: In the first, we focus on a multi-cell, multiuser, multi-stream, multiple-input multiple-output (MIMO), dynamic time division duplexing (TDD) scenario, where we consider bidirectional beamforming optimization with the objective of minimizing network sum-power expenditure with per-stream signal-to-interference-plus-noise ratio (SINR) constraints. A first approach to solve this problem assumes centralized processing and requires the availability of global channel state information (CSI). A second approach is performed in an iterative decentralized manner, based on the alternating direction method of multipliers (ADMM), which requires local CSI and has a reduced signaling load. Both approaches converge to a minimum network power expenditure, while close-to-optimum performance can be obtained when limiting the number of iterations. In the second part, we focus on the downlink (DL) of a multi-cell, multi-user multiple-input single-output (MISO) network in the presence of CSI errors, where we consider the development of two distributed-execution user scheduling schemes based on multi-agent deep Q-learning (DQL) to solve the beamforming problem of sum-rate maximization with per base station (BS) power constraints. Simulation results show that in the presence of CSI errors the proposed schemes outperform state-of-the-art algorithms both in terms of average spectral efficiency and execution time.