TY - GEN
T1 - Candidate-Aware Long Short Term Memory Handover Control in a Simulated LTE Scenario With Open RAN
AU - Talipov, Rashid
AU - Gaysin, Artur
AU - Galieva, Emiliia
AU - Cheremisova, Anastasiia
AU - Iuzhakov, Maksim
AU - Xia, Zhaoqiang
N1 - Publisher Copyright:
© 2026 IEEE.
PY - 2026
Y1 - 2026
N2 - Mobility management in multi cell cellular systems remains difficult because poorly tuned handover logic can increase signaling overhead, repeated returns to recently visited cells, and service degradation. This paper studies a candidate-aware recurrent neural approach for handover control in a simulated cellular scenario with open radio access network compatible control abstractions. The method utilizes a compact observation space, leakage safe dataset construction, joint prediction of handover initiation and target cell selection, constrained decoding over valid non serving candidates, replay based policy screening, and controller side safety checks for online execution. Offline evaluation shows that restricting target selection to realistic candidates substantially improves target choice quality while keeping initiation quality competitive. Replay based analysis further shows that strong predictive outputs alone are not sufficient for operationally reasonable behavior without conservative execution rules. Final matched closed loop experiments over long runs show a clear trade off across recurrent-only, baseline-assisted recurrent, and validation-selected multilayer perceptron modes. Stability-first learned modes reduce handovers and rapid returns but can increase delay, whereas the hybrid mode preserves service quality more effectively. The simulation results show that strong offline target prediction does not automatically translate into superior end to end mobility control.
AB - Mobility management in multi cell cellular systems remains difficult because poorly tuned handover logic can increase signaling overhead, repeated returns to recently visited cells, and service degradation. This paper studies a candidate-aware recurrent neural approach for handover control in a simulated cellular scenario with open radio access network compatible control abstractions. The method utilizes a compact observation space, leakage safe dataset construction, joint prediction of handover initiation and target cell selection, constrained decoding over valid non serving candidates, replay based policy screening, and controller side safety checks for online execution. Offline evaluation shows that restricting target selection to realistic candidates substantially improves target choice quality while keeping initiation quality competitive. Replay based analysis further shows that strong predictive outputs alone are not sufficient for operationally reasonable behavior without conservative execution rules. Final matched closed loop experiments over long runs show a clear trade off across recurrent-only, baseline-assisted recurrent, and validation-selected multilayer perceptron modes. Stability-first learned modes reduce handovers and rapid returns but can increase delay, whereas the hybrid mode preserves service quality more effectively. The simulation results show that strong offline target prediction does not automatically translate into superior end to end mobility control.
KW - closed-loop control
KW - handover control
KW - LSTM
KW - MLP
KW - mobility management
KW - ns-3
KW - O-RAN
UR - https://www.scopus.com/pages/publications/105048085202
U2 - 10.1109/EDM69524.2026.11632090
DO - 10.1109/EDM69524.2026.11632090
M3 - 会议稿件
AN - SCOPUS:105048085202
T3 - International Conference of Young Specialists on Micro/Nanotechnologies and Electron Devices, EDM
BT - Proceedings of the 2026 IEEE 27th International Conference of Young Professionals in Electron Devices and Materials, EDM 2026
PB - IEEE Computer Society
T2 - 27th International Conference of Young Professionals in Electron Devices and Materials, EDM 2026
Y2 - 27 June 2026 through 1 July 2026
ER -