Speed-Aware Asynchronous Deep Reinforcement Learning for Joint Handover Parameter Adaptation in 5G and Beyond Networks

In response to the explosive growth of data rates and traffic in 5G and beyond, network densification has emerged as a key solution. However, this approach presents significant challenges for mobility management, including high frequency of handovers (HOs), ping-pong effects, and HO failures. To address these problems, this letter proposes the speed-aware asynchronous reinforcement learning (RL) HO algorithm to dynamically adjust two HO control parameters in Self-Organizing Networks (SON), i.e., HO Margin and Time-to-Trigger. Simulation results demonstrate that the speed-aware asynchronous RL approach achieves superior performance compared to speed-unaware one, Reference Signal Received Power (RSRP)-based and other RL-based approaches.