A Dual-Band Reconfigurable Antenna Optimization Using Machine Learning Techniques

This study presents an innovative methodology for the optimization of a reconfigurable antenna capable of dynamically adapting to four distinct radiation states: activation at the lower frequency band, activation at the higher frequency band, simultaneous activation at both bands, and a deactivated state. To achieve this adaptability, the antenna design incorporates two PIN diodes, facilitating seamless reconfiguration across multiple operational modes. A comprehensive dataset comprising of 2400 samples was generated to develop a surrogate model that accurately predicts the antenna's performance metrics. Utilizing this surrogate model, the Deep Deterministic Policy Gradient (DDPG) algorithm was applied to refine the antenna's structural parameters, ensuring optimal performance across all operational states. The proposed framework capitalizes on the surrogate model to expedite performance evaluations, substantially minimizing the computational burden typically associated with full-wave electromagnetic simulations. Results confirm the efficacy of the DDPG-driven optimization, demonstrating significant performance improvements across the designated frequency bands of 1.9 GHz and 2.4 GHz. This research highlights the transformative potential of reinforcement learning in the design and refinement of reconfigurable antennas, showcasing its applicability to complex, multi-objective engineering challenges.