Machine Learning-based SON function conflict resolution

With the advent of 5G, it is expected that the heterogeneity of networks will increase significantly. This is due to the coexistence of different radio access technologies, service-specific requirements for different slices, combined with a multitude of conflicting optimization goals including, coverage, capacity, various QoS requirements, energy savings, etc. This complexity results in an explosion of the optimization space, which turns the network management into an intractable problem. In order to solve this, we introduce cognitive capabilities into the network to discover the dynamics between different SON functions (SF) and then automatically derive rules to coordinate them without explicit human modeling. This is different from previous designs of Self-Organizing Network (SON) functions based on single objective optimization approaches, where conflicts on the best configuration arise when more than one function is acting on the same cell. A well-studied conflict happens when Mobility Load Balancing (MLB) and Mobility Robustness Optimization (MRO) are running simultaneously with their competing goals of capacity optimization and user quality. In this paper we propose a Machine Learning (ML) driven framework to automatically derive the system's model, considering the dynamics of the selected SFs, in order to ease the optimization process handling the underlying conflicts in an automated way.