Machine learning-driven design of wide-angle impedance matching structures for wide-angle scanning arrays.

This paper introduces a versatile and efficient design methodology for optimizing wide-angle impedance matching (WAIM) configurations, enhancing the scanning range of arbitrary antenna arrays. The three-layered structure is modeled using the generalized scattering matrices (GSMs) of the layers, incorporating sufficient excited modes for efficient input impedance calculation. To broaden the method's applicability and meet manufacturing requirements, it also considers dielectric materials other than air between the array and WAIM. Machine learning (ML) algorithms are integrated to evaluate WAIM characteristics, reducing calculation time and resources while enhancing adaptability to new structures with minimal designer intervention. Decision Tree-based models are chosen to provide accurate prediction while minimizing the dataset preparation time. The methodology involves training a network using three ML algorithms, including decision tree, bagging, and random forest. Optimal WAIM parameters are efficiently determined using a genetic algorithm (GA). Three matching layers are designed and validated for several arrays operating at the frequency range between 9 and 11 GHz. The random forest model shows the best performance in predicting the WAIM behavior, with RMSE, [Formula: see text] scores, MAPE of 0.033, 0.916, and 2.161, respectively. Results demonstrate that the designed WAIMs effectively enhance the scanning range of both microstrip and waveguide arrays within the desired frequency range. The method achieves a calculation time of 0.3 s per angle, significantly faster than previous approaches, with a total runtime under an hour and minimal RAM usage (9.7 MB). This method offers an efficient framework for developing tools to design wide-angle scanning arrays and expand their applications.