Deep Learning-Based Handover Management to Steer Traffic in the 6G Intelligent Networks

Handover (HO) User Equipments (UEs) among base stations (BSs) intelligently has become one of the important approaches for traffic steering. Performance degradation is inevitable in conventional HO designs because HO is triggered after the monitored performance metrics have deteriorated. Although prediction-based schemes can be applied to make HO decisions before performance degrades, the challenges of accurately predicting performance in high-dimensional environments and managing large-scale data make it more difficult to achieve precise HO decision prediction. Therefore, this paper proposes a deep learning-based HO design to steer traffic in the sixth generation (6G) intelligent networks. By using the labeled datasets generated by the implemented emulator, a deep neural network (DNN) model is trained with the training loss and validation loss are 0.0917 and 0.1762, while the training accuracy and validation accuracy are 96% and 93%, respectively. Targeting at maximizing the average throughput of the UEs under the constraints of ping-pong rate and HO failure rate, the trained model infers the HO decision for all the UEs by extracting the features of the performance measurements from the UEs and BSs. Therefore, the HO decisions provided by the trained model not only avoid performance degradation but also achieve the optimum performance. Simulation results show that outperformed downlink throughput is achieved compared to the existing A3 event HO scheme in the fifth generation (5G) New Radio (NR) network.