Aero5GBS—Deep Learning-Empowered Ground Users Handover in Aerial 5G and Beyond Systems

Abstract

Recently, Unmanned Aerial Vehicles (UAVs) have been envisioned as aerial base stations, UAV-BS, serving ground users independently of the traditional cellular infrastructure. It is expected that in current 5G and Beyond (B5G) such UAV-BSs can form Aerial Networks to provide ubiquitous connectivity in remote, underserved, or rural areas. Keeping service continuity for ground users assisted by a network of UAV-BSs, particularly, the handover procedure, is even more challenging than the support required by the ground cellular networks since the cell coverage is smaller which may increase the ping-pong effect, and neighboring UAV-BSs may interfere on the UE’s communication with the serving UAV-BS. This paper provides an analysis of various Deep Learning (DL) algorithms to address the mobility issue and proposes intelligent handover strategies for a network of UAV-BSs. Firstly, a 5G Air-to-Ground radio channel is modeled. Next, the paper proposes DL techniques for handover management based on Recurrent Neural Network for trajectory and signal predictions. The UAV-BS network and related mechanisms were implemented by adding new modules and extending the 5G Standalone (SA) libraries of the OMNeT++ simulator. The results indicate the effectiveness of the proposal compared to the baseline 5G handover procedure and related work on UAV-BS systems in terms of improved Quality of Service metrics. The GRU based on Signal Prediction presented the best results, reducing the delay in 4.91%, and the packet’s loss at 78.95%.

Keywords

• 5G and beyond
• deep learning
• handover
• recurrent neural network and unmanned aerial vehicles