TOWARDS COGNITIVE AUTONOMOUS NETWORKS IN 5G

Cell densification and new Radio Access Technologies (RATs) have been the solutions towards improving area-spectral efficiency to serve the ever-increasing traffic demand. However, these solutions lead to increased cost and complexity of network design and operation. The solution thereto is globally agreed to be further network automation, specifically, by embedding cognitive capabilities derived from Artificial Intelligence (AI) and Machine Learning (ML) into the network. The resulting Cognitive Autonomous Network(CAN) however need to be delineated. This paper develops the models underlying such a design and proposes a design that addresses the CAN requirements. As an example, we show the benefit of such an approach by evaluating models that learn the network's response to different to different mobility states and configurations.