SON conflict resolution using reinforcement learning with state aggregation

In future generation networks one of the main focuses is on automating the network optimization. This is done through so called Self Organizing Network (SON) functions. A SON instance is a realization of a SON function that governs one or several cells. Several independent SON instances of one or multiple SON functions are likely to generate conflicts. This raises the need for a SON COordinator (SONCO) meant to solve these conflicts. In this paper we consider that each SON function has one SON instance on every cell and we present the design of a SONCO function for coordinating all these instances. The SONCO solves the conflicts that appear on the update requests arbitrating (i.e. accepting/denying the requests) so that it minimizes a predefined regret. This regret takes into account the weights associated to the SON functions that rank their importance according to the operator policies. We solve the problem in a Reinforcement Learning (RL) framework as it offers the possibility to improve the decisions based on past experiences. We employ a state-aggregation technique to make the state space of our solution scale linearly with the number of cells. We provide a study case for two SON functions: Mobility Load Balancing (MLB) tuning the Cell Individual Offset(CIO) and Mobility Robustness Optimization (MRO) tuning the CIO together with the handover hysteresis. The proposed SONCO function solves the conflicts on the CIO update requests. Numerical results show how the proposed SONCO is able to favor either MLB or MRO requests according to their associated weights.