Machine Learning Based Angle Estimation for Air-to-Ground Channel Measurements

Angle parameters are important for channel modeling and performance evaluation of communication systems. In this paper, a channel measurement system is developed for air-to-ground (A2G) scenarios. Moreover, a new angle estimation method based on machine learning for measured channel data is proposed, which includes channel impulse response (CIR) extraction and backpropagation neural network (BPNN) based angle estimation. Based on the extracted CIRs and conventional spatial alternate generalized expectation maximization (SAGE) algorithm, the training dataset is generated to train the BPNN. A measurement campaign is carried out under campus scenario and the analysis results show that the proposed angle estimation method can predict the angle parameter of different multipath accurately and efficiently. The proposed angle estimation method can assist the system design of unmanned aerial vehicle (UAV) beamforming communication, radiation source searching, etc.