Answers are grounded in this paper and cite their sources.
Handover management remains among the most critical and challenging problems in heterogeneous 4G long term evolution (HetNet-LTE) networks because fixed Reference Signal Received Power (RSRP) thresholds and static Time-to-Trigger (TTT) configurations fail to adapt to the dynamic, multivariable nature of real-world network conditions, resulting in excessive ping-pong handovers, unnecessary call drops, and degraded Quality of Service (QoS). This paper investigates a Deep Reinforcement Learning (DRL) approach for adaptive handover optimization in HetNet-LTE environments. This study is centred on a Multi-KPI Reward Function (MKRF) that simultaneously combines five network performance dimensions such as Handover Success Rate (HOSR), Channel Quality Index (CQI), User Throughput, Downlink Physical Resource Block (DL PRB) Utilization, and Call Drop Rate (CDR). A standard Proximal Policy Optimization (PPO)backbone was used, which on a discrete three-action space and a continuous KPI-derived state representation. The framework was evaluated in a simulation environment parameterized using a real-world LTE KPI dataset of 910 eNodeBs spanning urban, rural, and industrial zones of the Niger Delta region of Nigeria. The framework was not deployed in an operational LTE network; consequently, all reported performance values are simulator outputs rather than field-trial measurements. Across five independent random seeds, the DRL-MKRF agent achieves HOSR of 97.3%, reduces ping-pong handovers by 34.7% relative to the 3GPP A3 baseline, improves average user throughput by 18.2%, and reduces Call Drop Rate by 41.6%. Ablation experiments show that the multi-KPI reward formulation improves the joint KPI profile relative to single-KPI reward variants. The results support the usefulness of multi-KPI reward design for controlled LTE handover simulation, while simulator fidelity, larger action spaces, and live-network validation remain open issues.
No takes yet. Share an insight, caveat, or question.
Oisamoje et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: