AI-Enabled Automated and Closed-Loop Optimization Algorithms for Delay-Aware Network

• Published in: 2022 IEEE Wireless Communications and Networking Conference (WCNC) pp. 806 - 811
• Main Authors: Xiao, Da, Ni, Wei, Zhang, J. Andrew, Liu, Renping, Chen, Shuo, Qu, Yiwen
• Format: Conference Proceeding
• Language: English
• Published: IEEE 10-04-2022
• Subjects: 6G mobile communication; Conferences; Deep Q-Network; EDuRLLC; Hardware; Markov processes; Network slicing; over-provisioning of resources; Transfer learning; Ultra reliable low latency communication; VNF placement
• ISSN: 1558-2612
• DOI: 10.1109/WCNC51071.2022.9771970

Network slicing is one of the core techniques of the current 5G networks. To accommodate as many network slices as possible with limited hardware resources, service providers need to avoid over-provisioning of resources. In this paper, we first propose a Deep Q-Network (DQN) based network slicing algorithm to maximize the acceptance ratio and ensure prior placement of higher-priority requests for Ultra-Reliable Low-Latency Communication (URLLC) services. Specifically, we model the network slicing as a Markov Decision Process (MDP), where we consider Virtual Network Function (VNF) placements to be the actions of the MDP, and define a reward function based on service priority. For every service request, we use the DQN to choose an MDP action for performing the VNF placement. The placement results in an MDP reward that we can use to train the DQN. Once trained, the DQN approximates the optimal solution of the MDP. Considering the over-provisioning of resources, we then propose a Binary Search Assisted Transfer Learning algorithm (BSATL), in which the available hardware resources are scaled down/up and the knowledge learned from the source task is transferred to the target task in each iteration, to achieve automated and closed-loop optimization for the ever changing infrastructure, a scenario of 6G Event Defined uRLLC (EDuRLLC). Numerical evaluations show that our proposed scheme can significantly improve cost-utility while maintaining the optimal acceptance ratio.