Modeling And Linearization of Radio Frequency Power Amplifiers Using Machine Learning Approaches

This paper presents the application of contemporary machine learning algorithms to model and linearize the characteristics of Radio Frequency Power Amplifiers (RFPAs). RFPAs are required by wireless systems to increase the signal power for long-distance transmission. However, they cause signals with high peak-to-average power ratios (PAPR) to be distorted due to their inherent nonlinearity qualities and memory effects, raising bit-error rates (BER) in wireless systems. Hence, a key step in reducing signal distortion is linearising its behavioral features utilizing the digital pre-distortion (DPD) technique. Support vector regression (SVR) and Kernel regression (KR), two well-known kernel-based machine learning approaches, are used in this study to model and linearize RF PAs. The modeling and linearising capabilities of both techniques are demonstrated on a class AB GaN-based RFPA. It was found that linearising the RFPA characteristics using SVR with radial basis kernel function yields improved linearization performance compared to the kernel regression-based and standard memory polynomialbased method.