Machine Learning–Assisted Optimization of RF Power Amplifiers for High-Efficiency Wireless Transmitters

Abstract

The power consumption of power amplifiers (PAs) dominates the modern wireless transmitters and thus there is a strong desire to improve efficiency as a design goal. But, high power-added efficiency (PAE) and gain and linearity are difficult to meet in RF design spaces, where the spaces are strongly nonlinear and non-dimensional. Traditional methods, e.g. load-pull analysis and tedious parametric sweeps, are computationally intensive and may be inefficient in multi-parameter optimization. The proposal in this paper is a machine learning-aided optimization of a 2.4 GHz Class-AB RF power amplifier that was designed in 65 nm CMOS technology. An artificial neural network (ANN) surrogate model which is trained with harmonic-balance simulation data will provide a high accuracy prediction of important performance quantities of a circuit, such as PAE, gain, and output power. Optimal bias voltages and matching network component values are then found with a performance constraint using Bayesian optimization. The suggested framework reaches up to 10.8% enhancement in PAE in contrast to a traditional grid-based optimization as well as decreasing the optimization time in general by about 35 percent. The surrogate model portrays an excellent prediction accuracy, and low mean absolute error all through the design space. The findings show that machine learning-assisted optimization can provide a quicker convergence, enhanced efficiency, and scalable RF design methodologies, which will provide an efficient way out to high-efficiency wireless transmitter architectures.