Advanced technologies for energy savings in small cell networks

Since small base station (SBS) deployment is one of key technologies in 4G and 5G to meet the explosive increasing traffic demand, the power consumption problem becomes more serious. To reduce the power consumption of cellular networks, two promising technologies were proposed: one is energy efficient SBS deployment, the other is BS sleeping. For the former one, how to identify the number and the locations of SBSs is open research worth investigation. For the latter, when and which SBS is switched on/off are the core issues. Besides, how to guarantee the Quality of Service (QoS) while BSs are switched off also needs to be considered. This thesis tries to answer these two questions by proposing novel algorithms. In Chapter 3, the SBS deployment problem is investigated, and novel data-driven methods are proposed in different scenarios and different constraints. Based on existing networks, the aim in this scenario is to uncover the blackspots, improve the coverage probability, and minimizing the power consumption. Based on the Twitter data and k-means, the optimal number of SBSs and the tradeoffs between power consumption and coverage probability is investigated and compared with existing method. For a scenario where existing network is not available or non-existent, the aim is to satisfy users traffic requirements, and minimize the power consumption. A reward function is proposed for this work, then the tradeoffs between power consumption and the percentage of served traffic is investigated compared with existing method. The results in this chapter show the superiority of the proposed methods. In Chapter 4, a joint sleeping control and bandwidth allocation problem is addressed and formulated as a mixed integer non-linear programming (MINLP) problem subject to the transmission rate requirements. The joint optimization problem is then decoupled into two sub-problems: a centralized bandwidth allocation (CBA) sub-problem that minimizes the power consumption of the system by optimizing the allocated bandwidth of the active SBSs; and a centralized sleeping control (CSC) sub-problem that finds the optimal SBS sleeping strategy among all the possible ones. To solve the CSC sub-problem, two different algorithms are proposed based on K-Nearest Neighbor (KNN) and Convolutional Neural Network (CNN), respectively. For the KNN-based algorithm, the effectiveness of the algorithm is theoretically proven. The performance of proposed algorithm is evaluated in terms of average total power consumption (APC), percentage of unserved traffic (UR), and the complexity. As to CNN-based algorithm, the CSC problem is transformed to a classification problem and solved by a CNN model. For this algorithm, the CNN model is firstly trained by training data, and evaluated by the testing data. The metrics for this algorithm includes APC, UR, complexity, and accuracy. Simulation results in this chapter show the proposed schemes have superior performance compared with existing approaches. In Chapter 5, a similar problem to Chapter 4 is considered, while a reinforcement learning based mechanism is proposed to solve CSC sub-problem. By regarding the sleeping strategies as arms, mapping the transmission rate requirements to states, and defining the optimal CBA solution corresponding to a sleeping strategy as the arm's reward function, the CSC subproblem is transformed to a multi-state multi-arm bandit (MSMAB) problem, and a modified Q-learning algorithm is proposed for solving it. The convergence of the modified Q-learning algorithm is theoretically proven, and the computational complexity of proposed algorithm is theoretically analyzed. Finally, numerical results show proposed mechanism has a low computational complexity and can significantly reduce the total energy consumption of all SBSs, subject to the transmission rate requirements compared with existing methods.