QoE-Aware Adaptive Resource Allocation for 5G Video Streaming Using Deep Reinforcement Learning
Contributors
AJEYPRASAATH KB
Sai Kiran Oruganti
Keywords
Proceeding
Track
General Track
License
Copyright (c) 2026 Sustainable Global Societies Initiative

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Abstract
Quality of Experience (QoE) is also important in 5G video streaming, which is influenced by the variations of channel conditions, traffic demands, and radio resources. Existing approaches for resource allocation are usually biased towards channel condition, fairness, or traffic demand and none of them considers QoE maximization. In this work, we propose a QoE-based adaptive resource allocation scheme for 5G video streaming using Proximal Policy Optimization (PPO), a deep reinforcement learning based model. The scheme is to treat the resource allocation as a sequential decision-making problem, where the state is composed of CQI, traffic demand, buffer state, latency, packet loss and QoE score. Continuous allocation priorities are generated by the agent for the user equipment (UE) and then translated into integer allocations of PRBs, keeping the resource constraint intact. A reward function that combines QoE, throughput, buffer occupancy, fairness, delay, packet loss, and stalling is considered. The model is assessed through a mathematical abstraction of the system-level 5G network with ten UEs and benchmarked with Round Robin, Proportional Fair, Best-CQI, and demand-based allocation schemes. The preliminary results of the experiments on the same scenario indicate that Best-CQI outperforms all other approaches in terms of average QoE score (0.5352) and average effective throughput (6.1826 Mbps), while PPO reaches the average QoE score of 0.4929 and throughput of 5.4020 Mbps with five seeds. Such results can serve as a baseline to assess the adaptive learning approach and call for further research on the issue under difficult traffic and network scenarios.