Research On Linearization Technology Of Dual-input Doherty Transmitter

As the core of wireless communication system,RF transmitter has a crucial influence on communication quality.Now entering the 5G communication era,various efficient spectrum modulation technologies are used to fulfill the growing communication expectations,which makes the design of RF transmitter system facing great challenges.Dual-input Doherty transmitter is a new star among various technologies that enhance efficiency recently.Due to the advantages of large bandwidth,high performance and low circuit complexity,it is expected to become a high-quality RF transmitter solution.However,the signal linearity of the transmitter system is very poor,which limits its development and application in communication system.Digital predistortion,as a simple and efficient linearization technique,is expected to be a linearization solution for dualinput Doherty transmitters.This paper focuses on the nonlinear characteristics of the dualinput Doherty transmitter to explore its digital pre-distortion implementation method.The main contribution and innovation of this thesis are summed up as follows:1)This thesis researches the nonideal characteristics of dual-input Doherty power amplifier,then discusses the strength and weakness of several power amplifier behavior models with memory,points out the limitation of traditional memory polynomial model in the strong nonlinear modeling,and puts forward a solution which uses the neural network as the behavior model of such Doherty power amplifier.2)The digital pre-distortion of BP neural network is implemented,in which the structure of two hidden layers is selected to reduce the number of parameters as much as possible while ensuring the model accuracy.The LM algorithm was applied to accelerate the convergence of the network and save time cost.The experimental results indicate that the NMSE and ACPR of the output of this power amplifier are decreased by 13 d B and10 d B respectively after the BP model predistortion,and the signal performance is improved about 5-7d B compared with the ordinary memory polynomial model.3)Based on the deep learning theory,the LSTM network is used as the predistortion model when power amplifier is with strong nonlinearity of memory,and the advanced input data is inserted into samples to simulate the delay phenomenon in the transmission process of the circuit.The Adam algorithm is introduced to identify the parameters,which can extremely strengthen the performance of the model.In the meantime,a dropout layer structure is applied in the network to increase the sparsity and the adaptability of new samples of this network.4)Building a predistortion verification platform for dual-input Doherty power amplifiers,and using BP and LSTM network as the predistortion models of the power amplifier.The experimental results indicate that,for modulation signal with a total bandwidth of 20 MHz,after BP model and LSTM model compensation,the NMSE of power amplifier output signal is-42 d B,-46 d B,ACPR is-51 d Bc,-55 d Bc,respectively.For test signal with a total bandwidth of 40 MHz,the NMSE of output predistorted by BP model and LSTM model are-34 d B and-38 d B respectively,while ACPR are-40 d Bc and-45 d Bc respectively.