Multi-objective Efficiency and Phase Distortion Optimizations for Automated Design of Power Amplifiers through Deep Neural Networks

This study provides a comprehensive multi-objective optimization process for enhancing efficiency and minimizing phase distortion concurrently for high power amplifiers (HPAs) using deep neural networks (DNNs). The proposed optimization-oriented strategy consists of two sequential optimization phases: predicting the suitable HPA structure using a classification DNN and optimizing components values of the selected model using regression DNNs. In sizing design parameters two simultaneous regression DNNs are applied that are based on multi-objective particle swarm optimization (PSO) and multi-objective pareto front using modified quicksort (PFUMQ) methods for optimizing efficiency and phase distortion, respectively. The proposed design method reduces the phase distortion without significantly worsening efficiency and gain performances in HPAs. The successive optimization method is fully automated giving rise to an electromagnetic (EM)-verified post-layout generation, and prepares an easy-of-use procedure with reduced dependency to designers' experiences. For validating the proposed method, a wideband HPA is optimized for 1-2.3 GHz range frequency. It reveals larger than 55% drain efficiency at 40 dBm output power, and around 40% phase variation reduction in the operation band.