Real-Time Pattern Synthesis for Large-Scale Conformal Arrays Based on Interpolation and Artificial Neural Network Method

Real-time pattern synthesis for large-scale conformal arrays has long been a challenging task. Two efficient methods for large-scale conformal arrays pattern synthesis in a real-time way are, respectively, proposed in this article, including the cubic spline interpolation (CSI)-based method and generalized regression neural network (GRNN)-based method. The purpose of the synthesis is to find a set of amplitude excitations that yields a pattern to maximize directivity at a given scanning angle with desired low sidelobe level (SLL). The synthesis problem is first transformed into a convex problem (CVX), and then, the solution to the problem is obtained by convex optimization algorithm to generate sample dataset. The CSI-based method sets the data in the sample set as interpolation points to construct interpolation functions, while the GRNN-based method uses the input-output pairs in the sample dataset to train the model offline, learning the laws of the problem. Then, the desired amplitude excitations can be potentially obtained in a real-time way, either by the well-constructed interpolation function or the well-trained GRNN model. Finally, the representative numerical validations of a 514-element conformal array are presented to assess the effectiveness of the proposed two methods. Comparison with other advanced methods in open literatures is also provided.