DRL-Based Edge Caching for 360-Degree Video in Broadband Delivery of TV 3.0

초록

TV 3.0 integrates broadband into broadcasting systems to support demanding services such as catch-up replay and immersive 360 degrees content, which require adaptive and bandwidth-efficient delivery. However, these services create significant challenges in balancing network traffic and quality of experience (QoE), as popular contents lead to skewed access patterns and heavy bandwidth consumption. In this context, edge caching at broadband nodes is essential to absorb repeated requests, reduce backhaul load, and sustain QoE. To address this, we propose a deep reinforcement learning (DRL)-based caching framework that establishes the relationship between cache capacity, video quality, and backhaul bandwidth. In this framework, the action space determines caching decisions for tile-based video segments, the observation space captures popularity and quality features, and the reward balances QoE gains against bandwidth reduction. In addition, a greedy algorithm is developed to allocate cache space by maximizing a composite metric of segment quality and bandwidth cost. Extensive simulations show that our approach reduces backhaul usage by an average of 60.70% while maintaining QoE, and achieves 1.62% higher average quality compared with existing methods.

키워드