Data-Driven Optimization Strategies for Tunable RF Systems

The analysis and application of three different optimization algorithms are presented for tunable RF systems along with a comparison of the different tuning control schemes. These approaches are explored both theoretically and in measurement and are applied to a tunable matching network (TMN) to demonstrate how they address dynamic loading conditions in RF systems. Specifically, we present a data-driven (zeroth-order) projected gradient descent (ZO-PGD) algorithm, a feedforward neural network (FNN), and a novel hybrid approach that combines ZO-PGD and an FNN. The techniques are compared in the measurement of the representative TMN system for a variety of different load impedance trajectories and at different rates of change, and the relative merits of each technique are discussed.