Research On Digital Predistortion Technology Based On Artificial Intelligence Theory

Power amplifier is an important part of wireless communication system.Due to the inherent non-linear characteristics of the power amplifier,the signal will produce nonlinear distortion,causing in-band distortion of the signal and out-of-band spectral spread,resulting in an increase in the bit error rate of the signal.Therefore,linearization techniques need to be used to improve the non-linear effects of the power amplifier.In the existing power amplifier linearization technology,the predistortion technology has been widely used for its advantages of high accuracy,stable performance and easy implementation.In this paper,the digital predistortion technology is studied based on the Neural Network(NN)theory and Support Vector Machine(SVM)theory in the field of artificial intelligence.Firstly,theoretically study the nonlinearity and digital predistortion of the power amplifier from a mathematical perspective,and then on this basis,the main work and innovation of this article are carried out,as follows:On the basis of summarizing the existing BP neural network structure in deep learning,an Elman neural network model building method with internal delay feedback structure is proposed.In order to prove the feasibility of the Elman neural network model,the BP neural network was simultaneously introduced into the behavior modeling research of the power amplifier,and GA was used to optimize it to construct the GA model,and then compared with the Elman neural model for relevant experimental verification The comprehensive comparison of the performance and convergence time of the model proves that the Elman neural network model has faster convergence,stronger dynamic modeling ability and higher modeling accuracy than other models.Then this paper proposes a model establishment method based on the Twin Support Tensor Regression(TSTR)algorithm.Because the tensor mode can better maintain the spatial structure of the data,it can achieve better results.This paper uses the sample data set based on the Twin Support Vector Regression(TSVR)algorithm in machine learning.The tensor mode instead of the vector mode is used for experimental analysis.The TSTR algorithm is used to study the model establishment of the power amplifier.Finally,the performance of the simulation experiment is improved compared with the GMP model and the TSVR model,which proves the feasibility of the TSTR model.The pre-distortion verification platform was used to verify the proposed method.Two types of power amplifiers,Class F power amplifier and Doherty power amplifier,were used to verify the performance of the proposed algorithm,and the 10 M bandwidth WCDMA and 10 M bandwidth LTE signals were used for verification.The experimental results show that after the predistorter correction of the proposed method,the ACPR output signal of Class F power amplifier is improved by about 15.81 d B,and the ACPR output signal of Doherty power amplifier is improved by about 12 d B,proving that the Elman neural network model has a good performance compared with the GMP model.Non-linear compensation performance.