Learning-Based Self-Optimization for Mobile Heterogeneous Networks

With the emerging of 5G heterogeneous networks, the need of coordination between cells is ever intense. Asa result, cloud based SON (self-organizing network) has been proposed as the framework for advancedinterference coordination and load balancing. However, designing effective SON mechanisms to take fulladvantage of available data remains a challenging research topic. In this proposal, we focus on theself-optimization function of SON and adopt machine learning based approach for its potential intelligence.The research will deal with how to properly deploy small or even green cells to improve energy consumptionwhile provide sufficient service quality? and how to jointly consider interference coordination schemes forbetter overall data scheduling results? We plan to work on following three specific issues:1) To achieve load balancing by turning small cells on/off using real data from a telecommunicationcompany via learning based algorithms. The function operates in hours using only call detail records.2) To further deploy green small cells and dynamically access available spectrum. Spectrum informationis additionally needed.3) To perform interference coordination and data scheduling under a smaller time scale. Channelinformation is required in this case.The topics will take increasing sets of data and operate under decreasing time scale along three years. Thegoal is to fundamentally understand the benefit and feasibility of using learning based approaches.