Machine Learning-Assisted Synthesis of Large Nonuniform Antenna Arrays Including Mutual Coupling With an Equivalent Small Array Model

Mutual coupling in nonuniform antenna arrays is much more complicated and difficult to describe than their uniform counterparts. Therefore, traditional nonuniform antenna array synthesis methods are usually based on isolated element patterns and do not consider mutual coupling effects. It will inevitably cause various errors in radiation performance, especially for nonuniform antenna arrays with close element spacings. In this article, a machine learning (ML)-assisted method is proposed for the synthesis of large nonuniform antenna arrays in the presence of mutual coupling. An equivalent small array (ESA) model is developed to predict the element patterns considering mutual coupling under arbitrary array layouts. Compared with traditional ML-assisted methods, the proposed method establishes an efficient and accurate offline model to significantly reduce the full-wave simulation time for training dataset development. The remaining optimization procedure does not require full-wave simulation. Furthermore, the ESA model could be directly applied to antenna arrays with various antenna element numbers and spacings. The effectiveness of the proposed method is verified through linear and planar antenna arrays. Numerical and experimental results demonstrated that the mutual coupling is fully considered in the synthesis method and the synthesis results highly agree with the full-wave simulation and measurement results.