Reinforcement learning for optimizing the design of ceramic antennas: Sintering behavior and dielectric properties of Li5Mg21TiAl3O30 ceramics

In this research, we successfully synthesized a novel Li 5 Mg 21 TiAl 3 O 30 ceramic via the traditional solid-state reaction. This ceramic, when sintered at 1360 °C, demonstrated favorable microwave dielectric properties (MDPs), characterized by an εᵣ of 10.73, a Q×f of 81,598 GHz (at 12.61 GHz), and a τ f of −27.65 ppm/°C. In addition, the Deep Deterministic Policy Gradient (DDPG) algorithm was employed to optimize a ceramic-based dielectric resonator antenna (DRA) intelligently. Through this approach, the antenna parameters were effectively optimized, achieving a design with a center frequency of 7.90 GHz, a low return loss (S 11 ) of –53.6 dB, and a wide bandwidth of 1050 MHz, and a radiation efficiency of 92.55%, aligning the requirements of the 6 G band. The innovative application of deep learning technology improved the DRA design optimization process, offering a new paradigm for the intelligent design of ceramic-based microwave communication devices.