A robust approach to the cell switch-off problem in 5G ultradense networks

Abstract

Ultra-dense networks (UDNs) are recognized as one of the key enabling technologies of the fifth generation (5G) networks, as they allow for an efficient spatial reuse of the spectrum, which is required to meet the traffic demands foreseen for the next coming years. However, the power consumption of UDNs, with potentially hundreds of small base stations (SBSs) within each macrocell, is a major concern for the cellular operators, and has to be properly addressed prior to the actual deployment of these 5G networks. Among the different existing approaches to address this issue, a widely accepted strategy lies in the selective deactivation of SBSs, but without compromising the QoS provided to the User Equipments (UEs). This is known as the Cell Switch-Off (CSO) problem. The typical formulation of this problem is based on estimations of the traffic demand of the User Equipments (UEs) within the network. But these estimations could not be met. This work approaches these uncertain scenarios by extending the CSO problem with additional objectives that account for the robustness of the solutions to disturbances in these traffic estimates. To do so, a computationally demanding Monte-Carlo sampling is used to evaluate each solution. To manage such an increasingly large computing cost, a parallel version of the NSGA-II algorithm that is able to run on a computing platform composed of more than 500 cores has been used. It is able to compute in roughly 2 hours, an accumulated execution time of more than 42 days, which is within the expected timeframe of operators to make changes in the network configuration.