In this paper, a machine learning-based algorithm, K-Nearest Neighbor (KNN), is employed to predict the actual patterns of the linear arrays with arbitrary positions and varied element numbers. Based on the assumption that the active element pattern (AEP) is only susceptible to adjacent elements, the AEP of each element is predicted according to its subarray geometry features. And the array pattern is calculated correspondingly. To verify the validity of the method, the proposed methodology was employed to predict the patterns in several array synthesis cases using a dataset constructed from a series of full-wave simulated 17-element microstrip antenna arrays. Numerical results demonstrate that the proposed method is capable of predicting the pattern of arrays with arbitrary geometries and varied element numbers ranging from 7 to 127 with promising precision, which indicates that the proposed method is a feasible and effective array pattern calculation approach and applicable for all linear array synthesis problems with position optimization ability and mutual coupling effect considered.
Read full abstract