Reinforcement Learning-Based Adaptive Streaming Scheme with Edge Computing Assistance

Area of Science:

• Computer Science
• Artificial Intelligence
• Network Engineering

Background:

• Dynamic Adaptive Streaming over HTTP (DASH) is crucial for video streaming QoE.
• Existing DASH schemes lack client coordination and use fixed heuristics, limiting performance.
• Edge computing offers potential for enhanced streaming services.

Purpose of the Study:

• To propose a novel adaptive streaming scheme leveraging reinforcement learning (RL) within edge computing environments.
• To improve both the overall Quality of Experience (QoE) and QoE fairness among multiple clients.
• To utilize edge computing for collecting client-side observations to inform RL-based adaptive streaming policies.

Main Methods:

• Developed a reinforcement learning algorithm for adaptive video streaming.
• Integrated edge computing to provide client-side observations to the mobile edge.
• Implemented a multi-client adaptive streaming policy generation using RL agents.
• Conducted simulation-based experiments under diverse network conditions.

Main Results:

• The proposed RL-based scheme demonstrated superior performance compared to existing adaptive streaming methods.
• Significant improvements in overall client QoE were observed.
• Enhanced QoE fairness among concurrent users was achieved.
• Edge computing effectively facilitated the RL agent's decision-making process.

Conclusions:

• The integration of reinforcement learning and edge computing presents a powerful approach for optimizing adaptive video streaming.
• The proposed scheme effectively addresses limitations of traditional DASH by enabling client coordination and dynamic adaptation.
• This research highlights the potential of edge intelligence for improving real-time streaming services.