A fuzzy reinforcement learning approach for cell outage compensation in radio access networks

In order to reduce cost and improve service reliability, self-organizing networks (SON) features are being introduced gradually with the arrival of new 4G systems in radio access networks. Cell outage compensation (COC) is one of the most important tasks in the context of SON. This paper demonstrates a self-organized approach for COC, which based on fuzzy Q-learning and operates in a fully autonomous manner. To improve the effectiveness of the compensation algorithm and reduce the complexity, a method of selecting the compensation cells based on knowledge of cell load and radiated power level respectively is proposed. Then the joint downtilt and transmit power compensation is described as a multi-agent reinforcement learning problem and solved using fuzzy Q-Learning (FQL). The experiment results show that the network performance degradation is minimized when a cell is in outage through compensation. Moreover, the hybrid approach which adjusts downtilt and power simultaneously delivers higher performance than the strategies varied in antenna tilt or transmit power only.