Deep Reinforcement Learning for Resource Management on Network Slicing: A Survey.

Area of Science:

• Telecommunications Engineering
• Artificial Intelligence
• Computer Networks

Background:

• 5G and 6G networks rely on network slicing for customized services.
• Efficient resource management is crucial for meeting Quality of Service (QoS) and Quality of Experience (QoE) demands.
• Current resource management lacks the dynamic and predictive capabilities required for complex network slices.

Purpose of the Study:

• To identify key phases in network slice resource management.
• To analyze the application of Reinforcement Learning (RL) and Deep Reinforcement Learning (DRL) for autonomous resource management.
• To provide research directions for RL/DRL-based network slice resource management.

Main Methods:

• Systematic analysis of RL/DRL approaches for each resource management phase.
• Evaluation based on optimization objectives, network scope (core, radio access, edge, end-to-end), state/action spaces, algorithms, DNN structures, and exploration strategies.
• Identification of relevant use cases and vertical applications.

Main Results:

• RL and DRL offer viable autonomous solutions for network slice resource management phases.
• Analysis provides a framework for comparing different RL/DRL techniques across various network domains.
• Identified specific RL/DRL algorithms and configurations suitable for different network slicing scenarios.

Conclusions:

• DRL is a powerful tool for intelligent and autonomous resource management in 5G/6G network slicing.
• Further research is needed to address specific challenges in RL/DRL implementation for network slicing.
• This work guides future development of efficient and adaptive network slice resource management.