Zeroth-Order Optimization for Varactor-Tuned Matching Network

A data-driven projected gradient descent method is proposed to minimize the input reflection coefficient of a tunable matching network. The approach selects the network’s two varactor tuning states to minimize the input reflection coefficient in response to an unknown, time-varying load impedance. Fixed and adaptive step size strategies for the data-driven gradient method are compared in a hardware-in-loop experiment operating at 850 MHz, and measurements with both static and dynamic load impedances are presented.