PA-AUF-D2D: A QoE-Driven, Prediction-Assisted, and D2D-Enabled Uplink Scheduling Framework for LTE-A/5G Heterogeneous Networks

Abstract

The explosion of uplink-intensive applications, such as video conferencing, cloud-based applications, Internet of Things (IoT) services, and ultrahigh-definition streaming, has led to additional congestion, interference, and energy inefficiency within LTE-Advanced (LTE-A) and 5G (LTE-A/5G) Heterogeneous Networks (HetNets). Solution: Prediction-Assisted Adaptive Uplink Framework with D2D Offloading (PA-AUF-D2D) is developed to incorporate QoE-based scheduling, prediction-augmented resource allocation, and Device-to-Device (D2D) offloading traffic for opportunistic, scalable, and fair uplink resource management. Methodology: PA-AUF-D2D consists of a channel, traffic, and mobility predictor based on exponential smoothing, a QoE-oriented utility function for resource block allocation, a Tabular Q-Learning (TQL) agent for resource allocation, and a D2D offloading module for congestion alleviation, all within a system-level simulation environment in Python. A 20-100 user equipment (UE) system-level simulation is used to compare the proposed PA-AUF-D2D to work by Garbazza et al. (2025). Results: Compared to work by Garbazza et al., with a system-level simulation in Python with 20-100 UEs, PAAUF-D2D achieves: 81% delay reduction for real-time traffic; maximum non-real-time throughput improvements of 955% (average 540-680%) across scenarios; 4.8-5.0 mean opinion score (MOS); 23.4% average gain of user equipment (UE) throughput at the network edge thanks to D2D offloading; and 452 kbps/mW peak energy efficiency. Conclusions: PA-AUF-D2D shows promise as a low-complexity technique for managing uplink resources in LTE-A/5G HetNets, and potential future directions are discussed in light of simulation results.

DOI: https://doi.org/10.5281/zenodo.23138505