Optimization of RF PAs Using CNN Based Surrogate Model and Genetic Optimization Algorithm

ABSTRACT

This paper proposes a novel optimized design method combining a convolutional neural network (CNN) surrogate model with a genetic algorithm (GA), which is applied to the optimized design of power amplifiers (PAs) to address the inherent limitations of traditional PA design, including high computational costs and low optimization efficiency caused by reliance on high‐fidelity electromagnetic simulations. The core application workflow of the proposed CNN‐GA collaborative design framework of PAs consists of two key modules. First, a high‐precision CNN surrogate model is established to learn the nonlinear mapping between the geometric parameters of PAs and their critical electrical performance indicators, such as power‐added efficiency (PAE) and output power ( P _out ). The constructed model replaces time‐consuming full‐wave electromagnetic simulations and realizes rapid prediction of PA performance parameters. On this basis, a GA method is adopted to optimize the physical dimensions of PA matching networks. A dedicated fitness function oriented to wideband performance specifications is designed to conduct systematic iterative optimization under full operating band constraints, screen structural parameters that meet all frequency‐domain performance requirements, avoid local optimal solutions, and achieve global performance optimization of PAs. Experimental results demonstrate that the proposed CNN‐GA intelligent design framework improves the PA optimization efficiency by an order of magnitude while maintaining design accuracy and satisfying wideband performance thresholds. This study fully verifies the feasibility and superiority of the CNN‐GA hybrid algorithm in the intelligent optimization design of microwave PAs, providing an effective technical scheme and reference for the rapid and intelligent design of high‐performance microwave devices.