A Universal and Lightweight Automated Design Framework for Multi-Band Power Amplifiers

In power amplifier (PA) design, the repeated trial-and-error process and time-consuming electromagnetic (EM) simulation are major obstacles to achieving efficient design. In this study, based on a small microstrip line dataset, a lightweight deep neural network (DNN) is trained to accurately predict the S-parameters of layout-level matching networks (MNs). Leveraging the DNN’s high-speed prediction capabilities and an ordinary particle swarm optimization (PSO) algorithm, the entire optimization process is completed without the involvement of EM simulation software, enabling highly efficient automated design of PAs. This method is applied to the automated synthesis of narrowband, tri-band, and broadband PAs, demonstrating high flexibility. Among these, the broadband PA (a 0.5-5 GHz GaN PA) is fabricated for validation. Measurement results show that within the operating frequency range, the output power reaches 39-42.2 dBm, and the drain efficiency (DE) ranges from 47.2% to 74.6%, with an average of 56.5%.