Study On The Energy Saving Technologies Based On Traffic Prediction In Cellular Net-works

Recently, with the sharp increase of the cellular users, typical cellular networks have quickly been large-scaled systems, and the cellular networks have also emerged as one of the dominate comsumers of the world energy, which brings large pressure and great challenge to the ecological environment. Besides, larger and larger energy costs have troubled all service providers, resulting in less benefit of their business. However, there exists severe wastage in the energy usage of modern cellular system, due to the conflict between the system design rule of maximum capacity and variable user demands. Thus, modern cellular systems are usually not energy-efficient at all. Therefore, it is really necessary to take into consideration the energy efficiency rule when deploying cellular networks. This paper studied the application of network traffic prediction on the design of green cellular systems, and gave one traffic prediction method based on matrix spatio-temporal compressive sensing. Furthermore, the paper also provided two energy saving schemes, which relied on the adaptation to predicted traffic demand.To begin, this paper briefly introduced network traffic prediction, including the concept of traffic matrix and conventional methods on traffic prediction.Then, we studied the traffic prediction based on compressive sensing. Existing algorithms on compressive sensing do not perform so well while applied over actual data, especially when the data loss rate is high, or when the data loss is structured. The main reason is that those actual data does not match the mathematical constraints which make existing algorithms on compressive sensing perform best. My paper gave a traffic prediction method based on spatio-temporal compressive sensing in cellular networks, which constructed a cellular traffic matrix, considered the elements in the future as missing data, and reconstructed these missing elements using approximate matrix factorization, according to the inherent low-rank characteristic and spatio-temporal relevance in cellular traffic data. Compared to traditional traffic prediction schemes, our method improves the prediction performance greatly. Meanwhile, my paper analysed the statistical characteristics of the cellular traffic data, such as the similarity or periodicity in time dimension, spatial relevance and low rank feature. According to this, we constructed the spatio-temporal relevance model of the cellular traffic data. Thus, the predicted traffic matched the intrinsic spatio-temporal features excellently.Nowadays, the energy consumption of cellular networks is usually not relevant to user traffic demand. So most base stations in cellular networks keep working even when there does not exist user demands, creating severe energy wastage. In order to solve the energy wastage problem in cellular systems, the paper studied the application of the heterogeneity of cellular traffic in time and space dimension on the energy efficiency improvement of cellular networks, and proposed one off-line dynamic base station switching-off scheme based on traffic prediction, which concentrated all the cellular traffic to some subset of the base stations, and turned off the base stations unnecessary to achieve energy saving purpose. Moreover, due to that the dynamic base station switching-off method mentioned above did not consider the additional energy consumption caused by the coverage expansion of some base stations to accept the transferred user traffic, and how each base station could adjust its coverage to accept the transferred traffic, my paper put forward one dynamic base station switching-off scheme based on traffic blocking, on the basis of the dynamic base station switching-off scheme mentioned above. Simulation results demonstrate that the dynamic base station switching-off scheme based on traffic blocking can improve the energy efficiency of cellular networks vastly.