Pattern Correction for Impaired Conformal Array Using Physics-Encoded Neural Network

Conformal arrays highly match the carrier’s surface, improving space utilization and enlarging the beam coverage. However, the impairment of antenna elements frequently occurs, resulting in deterioration of the radiation characteristics. In this work, a physics-encoded machine learning method is proposed to correct the degraded pattern with the working elements. The proposed method encodes the superposition principle into a neural network, increasing the interpretability and considering of the mutual coupling effect. By minimizing the value of the loss function, a set of excitations is produced to recover the radiation pattern. Simulation results verify the effectiveness of the proposed method for both cylindrical and spherical conformal arrays.