The Scene Recognition Technique In Wireless Networks Based On Spatial Clustering

With the rapid development of society, the city is evolving rapidly and new types of geographic scenes in coverage areas of wireless networks are challenging the network construction of operators. In order to adapt to the new development and improve the user experiences and customer satisfaction, network optimization oriented to various types of geographic scenarios in network is a very necessary work.There are different geographic scenarios in network coverage, such as high-rise buildings, high speed railways, tunnels, green fields and waters. In different coverage scenarios, the wireless propagation environments, user distribution, user behaviors, traffic and service types are all different. For example, user behavior in business district is significantly different from that in school. The wireless signal traveling in building areas attenuates greater than in green fields. So the geography scenario is an important factor that affects the quality of services of networks. To improve the quality of network operation, it is needed to configure and optimize the network parameters for different geographical scenes.By making use of a mesh density based spatial clustering method, this paper presents a method to recognize different geography scenes in the coverage areas of the wireless networks. At first, the coverage area is divided into a grid of cells. The map layers of landforms are overlayed on the cells and geographical and special attributes of geographical elements are saved in corresponding cells. Then all cells are grouped into clusters by means of a density based spatial clustering algorithm. Each cluster is a collection of cells that reach a certain density on the geographical distribution and the special attributes of cells in a cluster are same or similar with each other. The boundary of each cluster is smoothed, and then the convex hull of the cluster is calculated, so the cluster is reperesented as a polygon.Based on the above mentioned approaches, the geography scene recognition system is built. The system is developed by making use of C++, SQL Server 2005 and MapXtreme that is a popular GIS platform. The system takes the landform map layers as its inputs, and then recognizes point scenes, line scenes and surface scenes in the coverage areas. The recognition results are provided for configuring and optimizing network parameters.We test and verify the method and the system by geographical data in the GSM networks in several cities, such as Chengdu, Yingkou, Jinzhou and Shenyang. The results of scene recognition are consistent with the actual geographical scenes and the efficiency of the methods we presented is proved. This system is helpful to network optimization.