A Complex Network Model Based on Radio Map for Navigation of Cellular-Connected UAV

Currently, the primary method for investigating the navigation of Cellular-connected Unmanned Aerial Vehicles (UAVs) is centered around machine learning, e.g. the utilization of Deep Reinforcement Learning (DRL). However, this approach inherits common limitations associated with machine learning, notably prolonged training periods and limited model adaptability. In the context of addressing path planning for cellular-connected UAVs with the restriction of three-dimensional radio maps, this study initially formulates a problem model to contextualize the challenge and define solution objectives. Subsequently, aligned with this model, a navigation model construction method based on Complex Network theory is introduced for path planning of cellular-connected UAVs. Evaluation results demonstrate that within the same conditions, employing the traditional A* algorithm within our Complex Network based model leads to a significant reduction in signal outage probability compared to the DRL approach. This improvement comes at the cost of a slight increase in path length, while the time required for our model construction is notably shorter than that of DRL.