A Study about Heterogeneous Network Issues Management based on Enhanced Inter-cell Interference Coordination and Machine Learning Algorithms

Under the circumstance of fast growing demands for mobile data, Heterogeneous Networks (HetNets) has been considered as one of the key technologies to solve 1000 times mobile data challenge in the coming decade. Although the unique multi-tier topology of HetNets has achieved high spectrum efficiency and enhanced Quality of Service (QoS), it also brings a series of critical issues. In this thesis, we present an investigation on understanding the cause of HetNets challenges and provide a research on state of arts techniques to solve three major issues: interference, offloading and handover. The first issue addressed in the thesis is the cross-tier interference of HetNets. We introduce Almost Blank Subframes (ABS) to free small cell UEs from cross-tier interference, which is the key technique of enhanced Inter-Cell Interference Coordination (eICIC). Nash Bargain Solution (NBS) is applied to optimize ABS ratio and UE partition. Furthermore, we propose a power based multi-layer NBS Algorithm to obtain optimal parameters of Further enhanced Inter-cell Interference Coordination (FeICIC), which significantly improve macrocell efficiency compared to eICIC. This algorithm not only introduces dynamic power ratio but also defined opportunity cost for each layer instead of conventional zero-cost partial fairness. Simulation results show the performance of proposed algorithm may achieve up to 31.4% user throughput gain compared to eICIC and fixed power ratio FeICIC. This thesis’ second focusing issue is offloading problem of HetNets. This includes (1) UE offloading from macro cell and (2) small cell backhaul offloading. For first aspect, we have discussed the capability of machine learning algorithms tackling this challenge and propose the User-Based K-means Algorithm (UBKCA). The proposed algorithm establishes a closed loop Self-Organization system on our HetNets scenario to maintain desired offloading factor of 50%, with cell edge user factor 17.5% and CRE bias of 8dB. For second part, we further apply machine learning clustering method to establish cache system, which may achieve up to 70.27% hit-ratio and reduce request latency by 60.21% for Youtube scenario. K-Nearest Neighbouring (KNN) is then applied to predict new users’ content preference and prove our cache system’s suitability. Besides that, we have also proposed a system to predict users’ content preference even if the collected data is not complete. The third part focuses on offloading phase within HetNets. This part detailed discusses CRE’s positive effect on mitigating ping-pong handover during UE offloading, and CRE’s negative effect on increasing cross-tier interference. And then a modified Markov Chain Process is established to map the handover phases for UE to offload from macro cell to small cell and vice versa. The transition probability of MCP has considered both effects of CRE so that the optimal CRE value for HetNets can be achieved, and result for our scenario is 7dB. The combination of CRE and Handover Margin is also discussed