QoE Modeling in 5G/6G Video Streaming: A Comparative Study of Accuracy, Complexity, and Energy Trade-offs

The growing demand for uplink video streaming, live broadcasting, and immersive media upload over 5G and emerging 6G networks necessitates accurate Quality of Experience (QoE) models. A careful balance among prediction accuracy, computational complexity, and energy efficiency is required for resource-constrained User Equipment (UE). In this paper, a systematic comparative analysis of five representative QoE modeling paradigms is presented. These paradigms encompass the IQX exponential hypothesis, the Weber-Fechner logarithmic law, the linear additive penalty model, Random Forest regression, and a Reinforcement Learning-based reward formulation. A comprehensive three-component energy consumption model is developed for uplink transmission scenarios. By this model, radio transmission energy, computational encoding energy, and buffering energy expenditures at the UE are captured. Through a numerical experiment with 10,000 simulated upload sessions over a 5G New Radio network, each paradigm is evaluated across multiple dimensions. These dimensions include prediction accuracy, sensitivity to network parameters, computational cost, and energy efficiency. The inherent trade-offs among analytical, machine learning, and reinforcement learning approaches are quantified. Actionable guidelines for network operators and device manufacturers on QoE model selection in dynamic 5G/6G uplink environments are provided.