A Hybrid Approach of Learning and Model-Based Channel Prediction for Communication Relay UAVs in Dynamic Urban Environments

This letter presents the trajectory planning of small unmanned aerial vehicles (UAVs) for a communication relay mission in an urban environment. In particular, we focus on predicting the communication strength between air and ground nodes accurately to allow relay UAVs to maximize the communication performance improvement of networked nodes. In urban environments, this prediction is not easily achievable even with good mathematical models as each model is characterized by a series of parameters which are not trivial to obtain or estimate apriori and can vary during the mission. To address the difficulty, this work proposes to integrate a learning-based measurement technique with a probabilistic communication channel model. This hybrid approach is able to predict communication model parameters based on signal strength data that UAVs observe during the mission online, thus achieving better performance compared with the model-based approach in an urban environment. The predicted parameters are based on four discrete urban environment types. Numerical simulations validate the performance and benefit of the proposed approach.