Uso de Deep Q-Learning para alocação de recursos uplink em redes heterogêneas LTE-A/5G

Abstract

The growing demand for mobile connectivity, driven by diverse applications with stringent Quality of Service (QoS) requirements, poses significant challenges to the efficient allocation of resources in heterogeneous LTE-A/5G networks. This work proposes an integrated solution for uplink scheduling and resource allocation, aiming to ensure QoS, reduce energy consumption, and mitigate intercell interference in dense and high-mobility scenarios. The proposed methodology combines Deep Q-Learning (DQL), capable of learning optimal resources allocation policies through environment interaction, and the multi-criteria decision-making method Weighted Aggregated Sum Product Assessment (WASPAS), applied to intelligent handover decisions based on multiple criteria, such as Signal to Interference plus Noise Ratio (SINR), cell load, and user mobility. The solution is structured into three main components: (i) a dual-objective scheduler that prioritizes real-time (RT) and non-real-time (NRT) traffic based on the specific requirements of each service class; (ii) an energy-constrained resource allocation mechanism that adapts to channel variations and QoS demands; and (iii) an intelligent handover strategy that promotes service continuity and reduces connection drops. System performance was evaluated through simulations using the Network Simulator 3 (ns-3) environment under various user densities and dynamic scenarios. The results show significant improvements compared to traditional algorithms such as Proportional Fair (PF) and New Courteous Algorithm (NCA), including delay reduction, increased throughput, improved fairness index in resource distribution, and lower packet loss and call drop rates. As a result, the potential of artificial intelligence and multi-criteria decision-making based approaches for optimizing resource allocation in modern cellular mobile networks is highlighted.