Improved fuzzy reinforcement learning for self-optimisation of heterogeneous wireless networks

In this paper, a novel scheme to improve learning mechanism for future self-organising networks´ functionalities is presented using a combination of fuzzy logic and reinforcement learning. Although the two frameworks compliment each other well, an efficient reward distribution mechanism needs to be deployed or otherwise the learning performance may be degraded. This study introduces an improved reward distribution (IRD) scheme in that the action space is abstracted to represent only the actions that are most relevant to the final crisp executed action after defuzzification. As a case study, coverage and capacity optimisation of heterogeneous networks consisting of dense deployment of small cells is considered. Using the proposed method, simulation results confirm considerable performance enhancment in terms of learning efficiency and convergence time.