Deep learning-based caching optimization for VR 360° videos in vehicular edge computing.

Area of Science:

• Computer Science
• Electrical Engineering
• Multimedia Systems

Background:

• Virtual Reality (VR) and 360° video streaming present significant challenges in Vehicular Edge Computing (VEC) due to high bandwidth demands and ultra-low latency requirements.
• Existing caching strategies (LFU, LRU) and viewport-aware methods fail to adequately address the spatio-temporal dynamics and mobility inherent in VEC environments.
• Seamless VR experiences require advanced solutions that can dynamically adapt to user behavior and vehicular movement.

Purpose of the Study:

• To propose DeepEdge360, a novel deep learning-based framework for optimizing the caching of 360° videos within VEC systems.
• To enhance the quality of experience for VR streaming by addressing challenges related to bandwidth, latency, and user viewport dynamics.
• To develop a proactive and intelligent caching strategy that adapts to real-time vehicular mobility and user viewing patterns.

Main Methods:

• Implemented an adaptive tile-based segmentation and request mechanism using Long Short-Term Memory (LSTM) for popularity prediction.
• Developed a proactive caching strategy for vehicles and Roadside Units (RSUs) that optimizes storage based on user behavior and mobility.
• Utilized a Deep Q-Network (DQN) for an intelligent cache eviction strategy to balance performance metrics.

Main Results:

• Achieved an 82% cache hit rate, demonstrating superior caching efficiency.
• Reduced end-to-end latency to 45ms, crucial for immersive VR experiences.
• Optimized bandwidth utilization to 76% through intelligent caching and prefetching.

Conclusions:

• The DeepEdge360 framework effectively supports high-quality VR streaming in dynamic vehicular networks.
• Deep learning-based caching significantly outperforms traditional and state-of-the-art methods in VEC environments.
• The modular design of DeepEdge360 ensures practical deployability in edge-assisted systems for enhanced VR services.