Research And Implementation Of Predistortion Technology For Power Amplifier Based On BP Neural Network

In modern wireless communication systems,RF power amplifiers are important components of base station transmitters.Affected by the non-linear characteristics of the power amplifier,the input signal would have serious in-band distortion and out-of-band distortion after being amplified,which increases the bit error rate of the communication system and interferes with adjacent channels.Therefore,it is very important to linearize the power amplifier.At the same time,the linearization technology of the power amplifier has become a key technology in broadband wireless communication.Among them,baseband digital predistortion has become a hotspot in linearization research in recent years due to its advantages of low cost and easy implementation.After analyzing the nonlinear characteristics of the power amplifier,this paper proposes a digital pre-distortion solution for the power amplifier based on the BP neural network model.The innovation and main work of this paper include the following aspects:1.This thesis analyzes the nonlinear characteristics of the power amplifier,discusses the modeling accuracy and complexity of several existing power amplifier models,and points out the limitations of each model.At the same time,a BP neural network model is proposed.This model has great advantages in modeling accuracy.The model is used to model a set of power amplifier input and output signals.Simulation in MATLAB found that its NMSE is reduced by more than 10 dB compared to the memory polynomial model.2.In this thesis,the power amplifier memory effect is reflected in the BP neural network,and the Levinberg-Marquardt algorithm is introduced to identify the model coefficients,which can greatly reduce the number of iterations.A digital predistortion system for power amplifier is established in MATLB,and a neural network model with the number of neurons in the first hidden layer as 7 and the number of neurons in the second hidden layer as 5 is used to predistort the power amplifier.The simulation results show that the BP neural network model can reduce the output signal ACPR by 20 dB.3.This thesis implements the simulation and verification of the effective BP neural network predistorter in FPGA,and proposes a symbol separation method to multiply the signed fixed-point number to reduce resource consumption.For the hidden function of the neuron hidden layer tansig,After analyzing the limitations of traditionallookup tables,this paper proposes a segmented uniform quantization lookup table method,which can not only reduce resource consumption,but also ensure that data accuracy is not affected.The parameter recognition part of the neural network predistorter is completed in MATLAB,and then the coefficients are imported into the BP neural network model in the FPGA to predistort the power amplifier.4.Build the predistortion test platform of the power amplifier with the existing devices in the laboratory,cascade the BP neural network predistorter and the class F power amplifier,input the LTE signal with a center frequency of 2.4GHz and a bandwidth of 20 MHz,and perform the output signal of the power amplifier,Measurements show that the ACPR of the output signal of the power amplifier after predistortion can be reduced by more than 13 dB compared to the original power amplifier.