Real-Time Sports Event Multi-View Streaming Optimization With Large Models: A Collaborative Edge Framework in 5G/6G Wireless Networks

Abstract

Real-time multi-view sports streaming poses challenges in latency, Quality of Experience (QoE), and bandwidth efficiency under dynamic wireless conditions. Traditional centralized methods struggle to meet the demands of personalized viewing in 5G/6G environments. This paper presents CEFLM, a Collaborative Edge Framework empowered by Large Models, which integrates a transformer-based predictor for user viewpoints, a QoE-aware stream selector, and a federated multi-agent scheduler across edge nodes. A cross-layer optimization module further refines video quality and resource allocation. To evaluate CEFLM, we construct two datasets—MVSports-360 with synchronized multi-view annotations and YouTube MV-Highlights with aligned sports highlights. Experimental results show CEFLM achieves a Top-1 viewpoint accuracy of 84.6%, reduces latency by 24%, and improves QoE by 10% over strong baselines. Compared to a recent RL-based method, CEFLM increases QoE by 9.8% and lowers rebuffering. Ablation studies confirm that removing the large model or edge collaboration degrades performance, with QoE dropping by up to 7.9%. These results validate the effectiveness of CEFLM in enhancing adaptive, user-centric multimedia delivery in future wireless networks.