Research On Cell Sleeping Technique In Heterogeneous Cellular Network

As the wireless network traffic increase, wireless network demand of the system capacity become more and more serious. The traditional single-layer cellular networks can not satisfy the growing traffic demands. The concept of heterogeneous cellular networks is put forward, which attracts widespread attention for its flexible networking, high spectrum efficiency characteristics. In heterogeneous cellular networks, the network energy consumption has become an important issue in network optimization. As an effective method to reduce energy consumption, cell sleeping can increase network energy efficiency, while the quality of service is guaranteed. In this paper, we analyze the cell sleeping strategies, aiming at improve system reward reducing energy consumption and meanwhile ensuring the communication quality of network users.In our heterogeneous cellular network model, there exists a control center, which can be scheduled users to access macro cell base stations or small cell base stations aiming at saving energy. As for the user model, we consider two cases: complete information model and incomplete information model. In complete information case, the control center knows full-time information of the users. In incomplete information case, the system only knows the history information of the users. In this paper, two threshold-based policies are given for these two cases, and the performance of these policies are analyzed. In addition, the impact of future information on the algorithm performance is also evaluated. Simulation results show that the energy consumption of the proposed dynamic cell sleeping scheme is lower than the static policy, and the upper bound of the energy consumption utilizing the algorithm is derived.Not only is the energy consumption problem considered in this paper, but we also take network latency and communication reward into account. A new reward function is formulated by combining these three aspects. the formation of new revenue function. Using this system model, we study the cell sleeping problem where the user information is unknown, and the cell sleeping procedure is reformatted as a Markov decision Process. With reinforcement learning method we obtain a suboptimal cell sleeping solution. In this solution, the cell sleeping procedure is a dynamic learning process. Finally, we analyze the convergence of the algorithm.