Deep Learning Based Efficient Beam Steering Algorithm for Deformed Curved Array Antenna

In this paper, a physically driven deep learning method is proposed for the beam steering problem of deformed conformal array antenna. The loss design is carried out by three constraints: gain, main lobe direction and side lobe level (SLL), and the softmax method is chosen for the loss design in order to solve the gradient truncation problem in the neural network. The specific direction map is obtained by active element pattern (AEP) algorithm after the output amplitude and phase of the neural network. The results show that the trained neural network output performs well under three constraints, and the gain and SLL obtained by this method are better than the superior single constraint.