As an essential power device in communication systems,RF amplifiers must typically balance high efficiency and high linearity in signal transmission.Operating near saturation,RF amplifiers achieve higher efficiency but suffer from reduced linearity.One approach to improve linearity is to pre-distort the input signal,introducing an opposite pre-distortion to counteract amplifier nonlinearity.Traditional analog pre-distortion cannot fully address the memory effects inherent in amplifiers.To mitigate the inevitable high-order nonlinear distortion in semiconductor amplifiers,this paper proposes a delay neural network model that leverages current and prior data points to capture higher-order nonlinearities for amplifier modeling.This model accounts for the dynamic high-order nonlinear effects arising from amplifier memory effects.A four-layer neural network is constructed and trained in a supervised manner by fitting the input-output sample dataset of the amplifier.After pre-distortion processing,the normalized mean square error(NMSE)improved from an initial-21.83dB to-37.77dB,demonstrating the model's strong linearization capability.
关键词
功率放大器/高阶非线性/神经网络/数字预失真/记忆效应
Key words
power amplifier/higher-order nonlinear/neural network/digital pre distortion/memory effects
维普
万方数据