Contextual Bandit for Cognitive Management of Self-Organizing Networks

Self-Organizing Networks (SON) concept is a technology that aims to improve the management and operation of mobile networks, through automatic configuration of network parameters. Even though SON functions are able to change network parameters automatically, the algorithms that run inside these functions still rely on parameters and rules that are manually defined by the operator, depending on its objectives. Thus, in order to realize a network that is self-organized as a whole, there is a clear need for a higher-level management entity that automatically translates operator objectives into SON configurations. In previous works, we have already studied and proposed an intelligent integrated management solution empowered with Reinforcement Learning (RL), namely the Cognitive Policy Based SON Management (C-PBSM).The C-PBSM is able to learn optimal SON configurations through direct interaction with the network. In this paper, we address crucial aspects of the mentioned approach, namely adaptability with different and varying network environments, transferability of the knowledge and the speed of convergence. We argue that the C-PBSM has major limitations with respect to these aspects. We consequently propose a context aware C-PBSM show that it is able to overcome the limitations of the C-PBSM.