The Cell Switch-Off Approach for Energy Efficient Cellular Networks

Network densification is a key enabler for providing high data rates and ubiquitous coverage. Although it enables the ambitious target of 1,000-fold gains in capacity, installing more base stations (BSs) challenges the energy efficiency targets of future networks. Cell switch-off (CSO) approaches are proposed to reduce energy consumption in off-peak periods by switching off some BSs. In this thesis, we define an energy-efficient cellular network as one in which as few BSs as possible are switched on while still satisfying all the users demand and quality of service. This thesis contributes to the current state of knowledge by directing the CSO research towards a more realistic, feasible, and practical implementation. We do this by arguing for employing offline (static) CSO and for considering spatially irregular BS deployments. First, we propose a dynamic CSO algorithm based on the well-known set cover problem. Our algorithm outperforms a benchmark algorithm in terms of the total number of switched-off BSs. While dynamic CSO algorithms are designed to adapt to fast changes in demand distribution, proper interference modelling is very challenging. To overcome this challenge, we next study regular static CSO patterns and describe them systematically. We propose sector-based patterns, where not only entire BSs could be switched off (site-based), but their individual sectors too. We compare the performance of different CSO patterns in terms of their energy efficiency and the number of supported users. CSO patterns are advantageous for modelling interference properly, reducing coverage holes, and making the uplink transmissions more energy-efficient for users. Nevertheless, the underlying assumption is that the BSs are deployed according to a regular grid. Finally, we consider spatially irregular BS deployments as a more realistic network model; therefore, we study applying CSO to irregular network layouts with the objective of making the active BS locations as regular as possible. We test the suitability of several algorithms from the p-dispersion problem literature for networks with BSs deployed with variable amounts of regularity. We also demonstrate some of these algorithms on real BS locations.