AI Insight
This study addresses handover optimization problems in 5G heterogeneous networks where ultra-dense small cells create mobility management challenges. The researchers compared four handover control approaches—fixed parameters, analytical adaptive control, tabular Q-learning, and deep Q-network (DQN)—that adjust handover margin and time-to-trigger based on user device velocity and handover direction. The DQN machine learning approach achieved the best overall balance across multiple performance indicators while maintaining compatibility with conventional 3GPP Event A3 handover frameworks.
Why it matters
Optimizing handover procedures in 5G networks is critical for preventing dropped calls, ping-pong effects, and unnecessary handovers as users move between cells at different speeds. This research demonstrates that machine learning methods can improve network reliability and user experience in dense urban 5G deployments without requiring fundamental changes to existing network protocols.
by Fozia Panhwer, Muhammad Mujtaba Shaikh, Beenish Ansari, Iqra Ameer, Mohamed A. Elmagzoub, Asadullah Shaikh
The use of ultra-dense small cells in 5G heterogeneous networks increases capacity, coverage and spectral efficiency, but it also complicates mobility management. In dense macro-small-cell networks, user devices often cross cells with varying coverage, transmit power, and propagation conditions, which can lead to high risks of radio-link failure, handover failure, ping-pong handover and unnecessary handover. These problems are exacerbated by using fixed handover-control settings, in which a single-value handover margin and time-to-trigger cannot provide a good trade-off across UE velocities and handover directions. In this paper, we propose a mobility-state- and handover-direction-aware HCP optimization for 5G HetNets while retaining the Event A3 handover scheme, state-based handover flow and radio-link-monitoring framework. The proposed framework is simulated with 3D geometry with user height, 3GPP TR 38.901 UMa/UMi propagation models, and a repeatable two-tier macro-small-cell deployment. Four controllers are compared with the same settings: fixed A3, an analytical adaptive controller, tabular Q-learning, and an improved deep Q-network (DQN). The analytical adaptive controller optimizes HOM and TTT based on the UE velocity and handover direction (macro to small, small to macro, and same-tier handovers), and the learning-based controllers use a multi-KPI reward to learn discrete HOM-TTT actions. Across the tested UE velocities, tabular Q-learning improves the principal mobility-robustness KPIs relative to fixed A3, while DQN provides the best observed mean multi-KPI balance among the four controllers. The DQN trend is also consistent with recent ML/DRL handover-optimization studies while retaining the conventional Event A3, finite-state handover, and radio-link-monitoring framework. Because the policy-level comparison is based on three evaluation runs, the reported uncertainty measures are interpreted descriptively and the conclusions are limited to the tested simulation conditions.