Community Detection in Directed Graphs for Cell-to-Sector Mapping Using Handover Statistics in Mobile Communication Networks

Abstract: Many use cases in mobile communication networks are dependent on knowledge of which cells serve the same sector. These use cases include network planning, optimization and troubleshooting. Current methods of obtaining knowledge regarding this cell-to-sector mapping are based on naming conventions of cells. These methods contain many manual steps and are prone to errors, in part because naming conventions vary between different network operators. Because of this, it would be very desirable for Information and Communication Technology (ICT) companies, e.g. Ericsson, and their customers to have the ability of performing cell-to-sector mapping in a way which is independent of naming conventions. In this thesis, a method for automatically inferring this cell-to-sector mapping using handover statistics was developed. This was done using cell relation data originating from a European network and two methods for community detection in directed graphs. The mapping was evaluated with binary classification of the pair-wise cell relations, using classification accuracy as the main performance metric. The results show that cell-to-sector mapping using this method can be made with high accuracy for the cell relation data that was used in this thesis. The classification accuracy for this method was close to 99% if knowledge regarding which cells belong to which sites was available. If not, the accuracy was found to be just below 97%. After the evaluation performed in this thesis, the developed method could be used to provide ICT companies and network operators with insights regarding their networks. One example is how node configurations affect handover dynamics. Further testing on other data sets is needed before any general conclusions can be drawn regarding the method’s efficacy on cell-to-sector mapping.