Scholarlink

<p>In LTE and upcoming 5G networks, supporting latency-sensitive and bandwidth-intensive services like Voice over IP (VoIP) and real-time video streaming continues to be a major problem, especially under dense user populations and diverse Quality of Service (QoS) needs. The static priority processes used by traditional channel-aware schedulers, such as exponential-based techniques, restrict their capacity to adjust to quickly changing traffic and channel conditions. An Adaptive Reinforcement Learning–Based Weighted Exponential–Logarithmic (ARL-WEL) scheduling system for dynamic radio resource allocation in LTE and 5G network slicing settings is presented in this study. The suggested scheduler uses a reinforcement learning agent to continuously adjust their relative influence based on real-time network states, combining the delay responsiveness of exponential metrics with the fairness-aware features of logarithmic functions. Without predetermined thresholds or fixed weighting factors, ARL-WEL automatically learns optimal scheduling strategies by monitoring queue lengths, head-of-line delays, channel quality indicators, and traffic arrival rates. ARL-WEL works noticeably better than traditional scheduling techniques, according to simulation findings acquired using the LTE-Sim platform under high-load scenarios (up to 80 active users). In comparison to the EXP RULE scheduler, the suggested approach improves throughput by 7–10% while lowering the packet loss rate by up to 18–22% and the average latency by roughly 20–25% for video traffic. ARL-WEL ensures compliance with strict real-time QoS restrictions for VoIP services by achieving delay reductions of up to 23% and packet loss rates below 1%. While overall spectral efficiency stays within acceptable ranges, fairness among slices is also improved, as evidenced by a discernible decrease in the Gini index.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Adaptive reinforcement learning–based weighted exponential–logarithmic scheduling for QoS optimization in LTE and 5G networks

  • 期刊论文

  • 发表日期：

    2026年7月21日

查看全文
  • Yudong Wei,
  • Chengjun Zhang,
  • Xiaofeng Nong,
  • Xiaobo Liang

摘要

In LTE and upcoming 5G networks, supporting latency-sensitive and bandwidth-intensive services like Voice over IP (VoIP) and real-time video streaming continues to be a major problem, especially under dense user populations and diverse Quality of Service (QoS) needs. The static priority processes used by traditional channel-aware schedulers, such as exponential-based techniques, restrict their capacity to adjust to quickly changing traffic and channel conditions. An Adaptive Reinforcement Learning–Based Weighted Exponential–Logarithmic (ARL-WEL) scheduling system for dynamic radio resource allocation in LTE and 5G network slicing settings is presented in this study. The suggested scheduler uses a reinforcement learning agent to continuously adjust their relative influence based on real-time network states, combining the delay responsiveness of exponential metrics with the fairness-aware features of logarithmic functions. Without predetermined thresholds or fixed weighting factors, ARL-WEL automatically learns optimal scheduling strategies by monitoring queue lengths, head-of-line delays, channel quality indicators, and traffic arrival rates. ARL-WEL works noticeably better than traditional scheduling techniques, according to simulation findings acquired using the LTE-Sim platform under high-load scenarios (up to 80 active users). In comparison to the EXP RULE scheduler, the suggested approach improves throughput by 7–10% while lowering the packet loss rate by up to 18–22% and the average latency by roughly 20–25% for video traffic. ARL-WEL ensures compliance with strict real-time QoS restrictions for VoIP services by achieving delay reductions of up to 23% and packet loss rates below 1%. While overall spectral efficiency stays within acceptable ranges, fairness among slices is also improved, as evidenced by a discernible decrease in the Gini index.