Abstract
Modern radio-frequency (RF) systems increasingly require impedance matching networks that are not only efficient, but also adaptive, reconfigurable, and robust to changing operating conditions. Applications such as high-power wireless communication, spectrum sharing, and integrated sensing and communication (ISAC) place stringent demands on bandwidth, power handling, and dynamic impedance control. Traditional RF design methodologies, which rely on analytical models and static circuit configurations, face growing challenges in addressing nonlinear device behavior, large design spaces, and environment-dependent system responses.
This dissertation presents a unified approach to the design and control of reconfigurable RF impedance matching circuits and components through the integration of machine learning, hardware innovation, and stability-aware analysis. First, a framework for machine learning model selection in RF and antenna engineering is developed, identifying suitable model classes and evaluation criteria for data-driven RF applications. This framework is then applied to the modeling of varactor diodes, where a machine learning-based approach is used to accurately capture nonlinear capacitance-voltage behavior and predict circuit performance with reduced computational cost.
Second, a soft-robotics-enabled reconfigurable capacitor is introduced as a hardware platform for continuous impedance tuning. The proposed design leverages pneumatic actuation of a compliant structure to achieve a wide capacitance tuning range, demonstrating a deformation-based approach to reconfigurable RF components that complements conventional semiconductor and MEMS technologies.
Third, the stability and robustness of advanced matching networks are investigated through a simulation framework for non-Foster circuits under component tolerances. By combining circuit simulation, full-wave electromagnetic modeling, and Monte Carlo analysis, the proposed method provides direct insight into how parameter uncertainty affects system stability and enables the construction of stability margins for robust design.
Together, these contributions establish a design paradigm that combines data-driven modeling, physically reconfigurable hardware, and stability-aware analysis to address the challenges of modern high-power and adaptive RF systems. The results demonstrate that integrating machine learning with reconfigurable circuit design enables more efficient exploration of complex design spaces and supports the development of intelligent RF systems capable of maintaining optimal performance under dynamic operating conditions.