Research On Low Sampling Rate Method For Predistortion Of Broadband Power Amplifier

Digital predistortion(DPD)is the most popular power amplifier linearization method in recent years,which can effectively improve the linearity of the power amplifier and improve the efficiency of the power amplifier.In broadband scenarios,the band-limited DPD limits the bandwidth of the output signal of the power amplifier through an external filter,which can effectively reduce the sampling rate requirement of the feedback loop ADC.However,due to the introduction of filters in the behavioral model,the numerical stability of the bandlimited DPD model is poor.This paper is devoted to improving the numerical stability of the band-limited DPD model,so as to improve the linearization performance in the case of low sampling bandwidth and low sampling rate.The main work contents and innovations are as follows:1.The reason of numerical instability in solving band-limited DPD using traditional least squares(LS)method is analyzed,and a band-limited DPD identification method based on Truncated Singular Value Decomposition(TSVD)is proposed.Firstly,perform singular value decomposition on the ill-conditioned band-limited matrix in the model identification process,and then use the L-curve method to determine the optimal truncation parameters.By discarding a certain number of small singular values,the modeling accuracy of forward PA and the numerical stability of band-limited DPD are improved,and the linearization performance under the condition of limited feedback bandwidth is improved.The proposed method is tested and verified on two types of power amplifier linearization platforms based on class AB and Doherty power amplifiers.Compared with the LS,adaptive iterative and Gauss Seidel iterative solution methods,the proposed method has the best linearization performance,and the proposed band-limited model identification method requires less computation during training.2.A band-limited DPD method based on amplitude segmentation is proposed,which divides the amplitude of the input signal of the power amplifier into multiple intervals,and uses low model orders to model the power amplifier characteristics in each interval.In each amplitude interval,the coefficients of the low-order band-limited model are identified by the LS method.The linearization test and verification of the proposed method are carried out based on the class AB power amplifier and the Doherty power amplifier.The test results show that when using the "full-band linearization" structure,compared with the performance of the traditional non-segmented band-limited DDR model,the condition number of the inverse matrix is reduced by more than 6100 times,and the ACPR is improved by 3.53/2.53dB and 5.8/1.9 dB,the NMSE is improved by 0.42 dB and 2.0 dB,respectively,where the feedback bandwidth is only 1.2 and 1.33 times the bandwidth of the input signal.When using the "limited bandwidth linearization"structure,the linearized ACPR performance of the proposed band-limited DPD method is close to that of the full sampling case,where the bandwidth of the feedback is 1.33 times the bandwidth of the input signal.The method proposed in this paper to improve the numerical stability of the band-limited DPD model can improve the linearization performance of lowsampling rate application scenarios,and help relieve the pressure on the feedback loop ADC of high-sampling rate systems in 5G,6G and other large-bandwidth application scenarios.