A Dual DNNs Method for Real-Time Beam Scanning and Stabilization of Deformed Conformal Array Antenna

This letter presents a method utilizing dual deep neural networks (DNNs) for real-time beam stabilization and scanning capability restoration in deformed conformal array antenna. Initially, the physical-driven deep neural networks based on active element pattern (AEP) are trained to swiftly generate the phase for directional control. Based on the pre-trained neural networks, the physical-driven deep neural networks semidefinite relaxation (PDNNs-SDR) algorithm is proposed to generate the amplitude that meets the requirements of the desirable radiation pattern. The data-driven deep neural networks are subsequently employed to achieve real-time amplitude output based on the existing amplitude as training data. Simulation and measured results demonstrate that the proposed methods can rapidly output the amplitude and phase corresponding to the desired radiation pattern. It combines the advantages of deep neural networks and convex optimization algorithm and achieves a good trade-off between both methods.