Digital Predistortion for RF Power Amplifiers Based on Enhanced Orthonormal Hermite Polynomial Basis Neural Network

摘要

In this paper, a novel digital baseband predistorter for RF power amplifiers (PAs) based on enhanced orthonormal Hermite polynomial basis neural network (EOHPBNN) is proposed. Digital predistortion technique based on neural network has been a hot topic in recent years, but the commonly used neural network predistorters employs feedforward neural networks (FNNs) with sigmoid function as the hidden neurons' activation function, which have limited linearization performance. The new proposed predistorter utilizes an orthonormal Hermite polynomial basis neural network where the orthonormal Hermite polynomial terms are chosen as the hidden neurons' activation functions. Taking advantage of the universal approximation capability of Hermite polynomial, the EOHPBNN predistorter shows superior linearization performance to the traditional NN-based predistorter. Also, the design of the EOHPBNN predistorter is combined with the AM/AM and AM/PM distortion characteristics, showing an improved linearization performance. The experimental results on a class-AB power amplifiers using wideband CMMB test signal demonstrate the excellent linearization performance.