Enhancing K-Band Dual-Input Doherty PA Performance by Bayesian Optimization

We present a methodology based on Bayesian optimization (BO) for the performance enhancement of a dual-input Doherty power amplifier (DIDPA) operating at 24 GHz under wideband modulation. First, we implement a dual-input control (DIC) algorithm, which allows for the user-defined arbitrary nonlinear shaping between the two inputs of the DIDPA. Then, an iterative procedure for the performance optimization is configured, where the tuning variables are the bias voltages and the parameters defining the dual-input shaping. Such a procedure jointly targets the improvement of power-added efficiency (PAE) while maintaining linearity by digital predistortion (DPD). To study the feasibility of the method in terms of convergence time and related tradeoffs in performance, the BO is applied in two fashions, namely, one leveraging a quasistatic model (QSM) of the DIDPA and the other directly on modulated signal acquisitions. The experimental results for high peak-to-average power ratio (PAPR) multitone broadband excitations show that the optimized settings can lead to an absolute improvement of several percentage points in PAE compared to the nonoptimized case, while still maintaining a suitable performance in terms of linearity and output power.