Dynamic Network Slicing Control Framework in AI-Native Hierarchical Open-RAN Architecture

Network slicing is a promising technology in next-generation wireless networks that enables the division of a physical network infrastructure into multiple virtual networks, each of which is tailored for specific service requirements. This approach enables a more flexible allocation of network resources such as beamforming vector, bandwidth and transmit power; thereby effectively supporting services that require high data transmission rates. However, in dynamic network environments where multiple users dynamically move around; hence the interference relationships are dynamically varying, traditional static network slicing solution has critical drawbacks. To this end, for the effective implementation and performance improvement in practical and dynamic network environments, we first propose a dynamic network slicing control framework in AI-native hierarchical Open-RAN (Radio Access Network) architecture where mobility prediction and network controls are designed by multiple timescale decomposition. The proposed framework can facilitate effective network controls, enabling the generation of finely tuned QoS management decisions (power/bandwidth allocation, user scheduling, beam activation) in different timescales. On top of this framework, we compare the performance of a simple dynamic network slicing algorithm and an existing static network slicing scheme via simulations.