Towards cognitive autonomous networks (5G and beyond)
* Register (or log in) to the AI4G Neural Network to add this session to your agenda or watch the replay
With Self Organizing Networks (SON), the Mobile RAN in 4G has a strong legacy in Network Management Automation. There, mostly rule-based functions are deployed to cope with the complexity of a heterogeneous network. In 5G, additional service types (critical and massive Machine Type Communications) are introduced which are realized by concepts like URLLC and multi-connectivity within a virtualized, sliced mobile network. This leads to additional operational complexity, e.g., with regard to jointly optimizing physical and virtual resources and perform intra- and inter-slice management.
To address such operational complexity we argue that a Network Management System needs to be “Cognitive”, i.e., being able to conceptualize and contextualize itself and its environment. The increased availability of network data together with the identification of suitable AI methods allows to create Cognitive Network Management (CNM) functions for 5G operational use cases. Such functions are thus able to autonomously adapt to different network deployment environments and their respective context as well as different operating points for a given deployment.
As a case study for the described characteristic, we show “Predictive Location-Aware Network Automation for Radio (PLANAR)” enabling coverage & capacity prediction in the digital twin of a sliced 5G trial network (Hamburg seaport).
Towards 6G it is anticipated that not only Network Management but also Network functions will become “Cognitive”. Hence AI will be a much more integral part of the network as well as the management architecture. Thus, e.g., ML models and their respective training need to be managed as part of network operational procedures, and their use wrt. location and time has to be orchestrated / coordinated in a multi-vendor (standardized) way.
In summary, Cognitive (AI-enabled) Network Management functions, which are able to exploit the measurement and context data generated within the network infrastructure and the corresponding deployment environment, allow to handle network infrastructure complexity and optimize performance. From the perspective of the human network operator the focus shifts from participating in the execution workflow of network operations to supervising that workflow.