Design And Implementation Of A Power Amplifier Model Based On Transfer Learning

Power amplifiers(PAs)are essential devices in the wireless communication system,which has a crucial impact on the signal quality of the communication system.In order to better analyze the characteristics of power amplifiers,various behavior modeling methods have been proposed.Among them,due to the good fitting ability of the artificial neural network to nonlinear function,the neural network behavior model has been widely used in power amplifier modeling.With the popularization and application of the fifth-generation mobile communication technology,the transmission bandwidth of wireless communication system increases,which makes the nonlinearity and memory effect of power amplifiers stronger.To improve the modeling ability for the broadband power amplifier,using the deep neural network(DNN)can achieve good modeling performance.Still,the problem of long training time can not be ignored.In order to reduce the modeling time of the neural network model for the broadband power amplifier,a design method for the power amplifier model based on transfer learning is proposed in this thesis.This method can effectively reduce the model training complexity and model modeling time,realize the fast adaptive modeling of power amplifiers.Firstly,a new power amplifier model is established based on the deep neural network.Then,the general characteristics of the power amplifier are extracted by transfer learning,and the pre-design filter is established by fixing part of the network in the pre-trained DNN model.The power amplifier model based on transfer learning mainly comprises two parts.The first part of the model is the pre-design filter to extract the characteristics of the PAs,and the second part is the adaptation layers to fit the output of the authentic PAs.The experiment results show that the proposed model can significantly reduce the complexity of the network structure to be trained,reduce the training time,and guarantee the performance compared with the power amplifier DNN model.