Scholarlink

<p>Handover management is a critical performance determinant in fifth-generation (5G) heterogeneous cellular networks (HetNets) with ultra-dense small cell deployments, high user mobility, and signal fluctuations, which render conventional signal-strength-reactive mechanisms governed by the 3GPP A3 event inadequate. This paper presents, and tests, the Location-Based Handover with Reinforcement Learning (LBHO-RL) algorithm that combines trajectory prediction from GPS data with an online Expected-SARSA reinforcement learning (RL) agent to proactively trigger the handover preparation before signal degrades. QualNet 7.4 network simulation implements and evaluates the algorithm across a 5G HetNet, including seven macro base stations and twenty-eight small base stations in an urban 3&#xa0;km × 2&#xa0;km terrain. In a large-scale Monte Carlo evaluation (50 independent runs), LBHO-RL achieves a 21.8 percentage-point improvement in handover success rate (HSR) at 120&#xa0;km/h (92.0% vs. 70.2%), a 62.7% reduction in handover interruption time (HIT) at 120&#xa0;km/h (33.7&#xa0;ms vs. 90.4&#xa0;ms), a 76.0% reduction in ping-pong handover rate (PPHR) at HOM = 3&#xa0;dB, a 34.9% reduction in per-UE energy consumption at 120&#xa0;km/h, and a 46.2% reduction in handover-related signaling overhead at 80 concurrent UEs—all statistically significant at p &lt; 0.001 (Welch t-test / two-proportion z-test) versus conventional A3-event handover. Sensitivity analysis confirms that LBHO-RL shows superior performance for the GPS position errors (σ) from 0 to 15&#xa0;m and the TTT configurations from 40 to 640&#xa0;ms, showing substantial robustness to parameter variation in practice. The study further exemplifies that location-based anticipation handover lead times of ≥ 300&#xa0;ms are fundamental for viable handover management concerning 6G millimeter-wave deployments with beam widths below 5°. Comparison against 13 Scopus-indexed open access publications (2022–2025) confirms that LBHO-RL delivers superior or competitive performance on all reported metrics, and provides the first QualNet-validated LBHO evaluation framework in the literature under review.</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

LBHO-RL: GPS Trajectory Prediction and Reinforcement Learning for Proactive Handover Management in 5G Heterogeneous Networks

  • 期刊论文

  • 发表日期：

    2026年6月30日

查看全文
  • Kalpesh Popat,
  • Divyakant Meva

摘要

Handover management is a critical performance determinant in fifth-generation (5G) heterogeneous cellular networks (HetNets) with ultra-dense small cell deployments, high user mobility, and signal fluctuations, which render conventional signal-strength-reactive mechanisms governed by the 3GPP A3 event inadequate. This paper presents, and tests, the Location-Based Handover with Reinforcement Learning (LBHO-RL) algorithm that combines trajectory prediction from GPS data with an online Expected-SARSA reinforcement learning (RL) agent to proactively trigger the handover preparation before signal degrades. QualNet 7.4 network simulation implements and evaluates the algorithm across a 5G HetNet, including seven macro base stations and twenty-eight small base stations in an urban 3 km × 2 km terrain. In a large-scale Monte Carlo evaluation (50 independent runs), LBHO-RL achieves a 21.8 percentage-point improvement in handover success rate (HSR) at 120 km/h (92.0% vs. 70.2%), a 62.7% reduction in handover interruption time (HIT) at 120 km/h (33.7 ms vs. 90.4 ms), a 76.0% reduction in ping-pong handover rate (PPHR) at HOM = 3 dB, a 34.9% reduction in per-UE energy consumption at 120 km/h, and a 46.2% reduction in handover-related signaling overhead at 80 concurrent UEs—all statistically significant at p < 0.001 (Welch t-test / two-proportion z-test) versus conventional A3-event handover. Sensitivity analysis confirms that LBHO-RL shows superior performance for the GPS position errors (σ) from 0 to 15 m and the TTT configurations from 40 to 640 ms, showing substantial robustness to parameter variation in practice. The study further exemplifies that location-based anticipation handover lead times of ≥ 300 ms are fundamental for viable handover management concerning 6G millimeter-wave deployments with beam widths below 5°. Comparison against 13 Scopus-indexed open access publications (2022–2025) confirms that LBHO-RL delivers superior or competitive performance on all reported metrics, and provides the first QualNet-validated LBHO evaluation framework in the literature under review.