Fast Beam-Pointing Control Algorithms Based on Deep Learning Models for Moving Target in MWPT

Microwave wireless power transmission (MWPT) over long distances holds the significant value and broad application prospects. The current studies mainly target fixed objects, but few focus on moving targets such as drones. Technical challenges in MWPT for moving targets lie in meeting the real-time and accuracy requirements of beam pointing control while also meeting the quality requirements of the radiation pattern when moving targets are in different positions. To address these performance requirements, this article proposes a fast beam pointing control method based on the deep learning model. Specifically, the ResNet-18 framework is improved as the core to construct the mapping relationship between the positions of moving target within the effective power transmission region and the feeding amplitudes and phases of the phased array antenna. To improve the accuracy of the antenna pattern synthesis, the eigen-driven analysis method (EDM) is proposed, in which the coupling effects among the array elements are considered. To generate numerous and accurate training samples for the deep learning model, an autoencoder combining the convolutional neural network (CNN) with EDM is established, in which a setting mask function is set to quickly and accurately determine the main-lobe and sidelobe intervals at different pointing angles.