Bahlke, Florian (2019)
Optimization Methods for Heterogeneous Wireless Communication Networks: Planning, Configuration and Operation.
Technische Universität Darmstadt
Dissertation, Erstveröffentlichung

Kurzbeschreibung (Abstract)

With the fourth generation of wireless radio communication networks reaching maturity, the upcoming fifth generation (5G) is a major subject of current research. 5G networks are designed to achieve a multitude of performance gains and the ability to provide services dedicated to various application scenarios. These applications include those that require increased network throughput, low latency, high reliability and support for a very high number of connected devices. Since the achieved throughput on a single point-to-point transmission is already close to the theoretical optimum, more efforts need to be invested to enable further performance gains in 5G. Technology candidates for future wireless networks include using very large antenna arrays with hundreds of antenna elements or expanding the bandwidth used for transmission to the millimeter-wave spectrum. Both these and other envisioned approaches require significant changes to the network architecture and a high economic commitment from the network operator. An already well established technology for expanding the throughput of a wireless communication network is a densification of the cellular layout. This is achieved by supplementing the existing, usually high-power, macro cells with a larger number of low-power small cells, resulting in a so-called heterogeneous network (HetNet). This approach builds upon the existing network infrastructure and has been shown to support the aforementioned technologies requiring more sophisticated hardware. Network densification using small cells can therefore be considered a suitable bridging technology to path the way for 5G and subsequent generations of mobile communication networks. The most significant challenge associated with HetNets is that the densification is only beneficial for the overall network performance up to a certain density, and can be harmful beyond that point. The network throughput is limited by the additional interferences caused by the close proximity of cells, and the economic operability of the network is limited by the vastly increased energy consumption and hardware cost associated with dense cell deployment. This dissertation addresses the challenge of enabling reliable performance gains through network densification while guaranteeing quality-of-service conditions and economic operability. The proposed approach is to address the underlying problem vertically over multiple layers, which differ in the time horizon on which network optimization measures are initiated, necessary information is gathered, and an optimized solutions are found. These time horizons are classified as network planning phase, network configuration phase, and network operation phase. Optimization schemes are developed for optimizing the resource- and energy consumption that operate mostly in the network configuration phase. Since these approaches require a load-balanced network, schemes to achieve and maintain load balancing between cells are introduced for the network planning phase and operation phase, respectively. For the network planning phase, an approach is proposed for optimizing the locations of additional small cells in an existing wireless network architecture, and to schedule their activity phases in advance according to data demand forecasts. Optimizing the locations of multiple cells jointly is shown to be superior to deploying them one-by-one based on greedy heuristic approaches. Furthermore, the cell activity scheduling obtains the highest load balancing performance if the time-schedule and the durations of activity periods is jointly optimized, which is an approach originating from process engineering. Simulation results show that the load levels of overloaded cells can be effectively decreased in the network planning phase by choosing optimized deployment locations and cell activity periods. Operating the network with a high resource efficiency while ensuring quality-of-service constraints is addressed using resource optimization in the network configuration phase. An optimization problem to minimize the resource consumption of the network by operating multiple separated resource slices is designed. The originally problem, which is computationally intractable for large networks, is reformulated with a linear inner approximation, that is shown to achieve close to optimal performance. The interference is approximated with a dynamic model that achieves a closer approximation of the actual cell load than the static worst-case model established in comparable state-ot-the art approaches. In order to mitigate the increase in energy consumption associated with the increase in cell density, an energy minimization problem is proposed that jointly optimizes the transmit power and activity status of all cells in the network. An original problem formulation is designed and an inner approximation with better computational tractability is proposed. Energy consumption levels of a HetNet are simulated for multiple energy minimization approaches. The proposed method achieves lower energy consumption levels than approaches based on an exhaustive search over all cell activity configurations or heuristic power scaling. Additionally, in simulations, the likelihood of finding an energy minimized solution that satisfies quality-of-service constraints is shown to be significantly higher for the proposed approach. Finally, the problem of maintaining load balancing while the network is in operation is addressed with a decentralized scheme based on a learning system using multi-class support vector machines. Established methods often require significant information exchange between network entities and a centralized optimization of the network to achieve load balancing. In this dissertation, a decentralized learning system is proposed that globally balance the load levels close to the optimal solution while only requiring limited local information exchange.

Frage zum Eintrag

Redaktionelle Details anzeigen