Optimal configuration determination in Cognitive Autonomous Networks

Cognitive Autonomous Networks (CAN) promises to raise the level of operational autonomy in mobile networks through the introduction of Artificial Intelligence (AI) and Machine Learning (ML) in the network processes. In CAN, learning based functions, called Cognitive Functions (CF), adjust network control parameters to optimize their objectives which are different Key Performance Indicator (KPI). As the CFs work in parallel, there is often an overlap among their activities regarding control parameter adjustment, i.e., at one point of time, multiple CFs may want to change a single control parameter albeit by different degrees or to different values depending on their respective levels of interest in that parameter. To resolve this dispute, a coordination mechanism is required for sharing the parameter among the independent CFs according to their individual interest levels. In this paper we provide the design of such a Controller in CAN to determine the optimal control parameter value. The Controller first quantifies the impact of that parameter on the objective of each CF, based on which the Controller determines the optimal value using Eisenberg-Gale solution. A numerical evaluation shows that compared to state-of-the-art, the proposed Controller can improve performance by up to 7.7%, while implementation in a simulation environment shows that the proposed Controller is feasible for use in a real life scenario.