Abstract
Interest in reflectarray antennas (RAs) has surged recently; unlike phased arrays, they need no feeding network, though design remains the main bottleneck. RA design is challenging because it requires adjusting hundreds of unit-cell geometry parameters to achieve desired reflection phases. Standard decomposition methods compute phases for individual cells analytically, but neglect mutual coupling, leading to suboptimal performance in gain, efficiency, and sidelobes. Optimization of the complete array at the level of full-wave electromagnetic (EM) analysis ensures reliability but is computationally heavy, and commercial solvers struggle with handling multiple objectives (gain, sidelobe levels). This work addresses direct EM-driven RA optimization for maximum gain and minimum sidelobes using a trust-region gradient search with numerical derivatives and customized objective function formulated to handle the design goal of choice properly. Extensive verification experiments involving several RAs implemented in microstrip technology and a range of conventional optimization methods demonstrate the competitive performance of the presented method, both in terms of the design quality rendered during the optimization process and the associated computational expenses.