Synthesis of Refiectarray Based on Deep Learning Technique

In this work, we investigate feasibility of applying deep learning techniques to synthesis of reflectarrays. A deep convolutional neural network is proposed based on AlexNet to predict phase-shift in an reflectarray antenna given a reflected direction. The proposed network takes radiation pattern and beam direction as input, with training and testing data obtained by array theory. After cafefully training, the proposed network demonstrates strong approximation ability and makes correct prediction of phase-shift. Preliminary numerical experiments show that the prediction error of phase-shift can reach below 0.4%. This paper shows that deep convolutional neural networks can mimic the phase synthesis process of reflectarrays and it has a great potential for real-time phase prediction in more complex problems of array synthesis.