Research On Adaptive Digital Predistortion Technology Of Power Amplifier Based On Neural Network

With the development of mobile communication and the requirement of energy saving and environmental protection,how to solve the contradiction between the efficiency and linearity of radio frequency power amplifiers(RF PA)in mobile communication system has become an increasingly important research direction.Therefore,the linearization technology of PA has been widely concerned,and Neural Network(NN)is considered as a kind of predistortion model of PA with great application prospect due to its flexible and high-precision nonlinear fitting ability.This paper mainly studies the adaptive digital predistortion technology of PA based on NN.Firstly,the nonlinear characteristics,evaluation index,behavior model and digital predistortion principle of PA are introduced.The back propagation(BP)real-valued time-delay Neural Network model for PA is introduced,and the influence of the number of hidden layers and the number of neurons on the model accuracy is compared and analyzed.Secondly,the NN PA model is improved for the problem of computation complexity.SNN(Simplified Neural Network)model and ASNN(Augmented SNN)model are proposed by adding a classifier to the model.In addition,a new fusion model of NN-ASNN is proposed,which is compared with the traditional NN PA model and greatly improves the accuracy of PA modeling.Thirdly,the Instantaneous Dynamic Gain(IDG)concept of amplifier is introduced for the optimization of traditional direct learning architecture(DLA).It avoids the complex process of PA modeling and updates the predistorter coefficient by replacing the output error of the predistorter with the output error of PA.A new IDG-DLA predistortion scheme suitable for NN model is proposed,and compared with the predistortion scheme based on the traditional DLA,the results show that the new scheme can reduce the computational complexity and achieve better PA linearization effect.Finally,the experimental results show that the new NN predistortion architecture based on IDG not only guarantees the performance but also reduces the computational complexity,and the predistortion algorithm based on data fusion has better predistortion effect than the original algorithm.