Research On Concurrent Dual-band Digital Predistortion Model And Model Reduction Method

As the most important part of the communication system,the efficiency and linearity of power amplifiers are mutually limiting indicators.The linearity of the power amplifier determines the performance of the entire system.Digital predistortion technology,as a linearization method for power amplifiers,is widely used due to its flexible structure and superior performance.In order to meet the requirements of concurrent dual-band communication systems,power amplifiers that can support two frequency bands have attracted much attention.The corresponding dual-band digital predistortion has become a research hotspot.This paper focuses on the concurrent dual-band digital predistortion model and model reduction method.The main research contents and innovations are as follows:1.The Modified Dynamic Deviation Reducation(MDDR)model was proposed.And the MDDR model is extended to dual band,a concurrent Two Dimensional Dynamic Deviation Reduction(2D-MDDR)model is obtained.The experimental verification shows that the 2D--MDDR model can accurately model the behavior of the power amplifier and well suppress the nonlinear distortion generated by the power amplifier.Especially in the case of strong non-linear distortion and complex transmission signals,the performance advantage is obvious compared with the 2D-DPD model,the normalized mean square error(NMSE)is improved more than 1 dB,and the adjacent channel power ratio(ACPR)is improved about 2-7 dBc.Compared with the DB-DDR model,the NMSE has a 1 dB improvement and the ACPR has a 1-4 dBc improvement.2.Aiming at the problem of large coefficients in the concurrent dual-band digital predistortion model,a model reduction method based on Group Lasso(least absolute shrinkage and selection operation)was proposed.Through this method,the sparse solution of the coefficients of the behavior model of power amplifier and the predistortion model can be estimated,which reduces the model’s coefficient amount.While reducing the model coefficient amount,it also ensures the performance of the model.The performance of the pruned model is similar to that of the complete model.Use this method can achieve the effect of reducing complexity while maintaining performance.Using the model pruned method based on Group Lasso to extract the model coefficients,compared to using the least squares method(LS)to extract the coefficients,it can avoid the occurrence of overfitting.The performance of this model reduction method was verified by simulation.The verification signals used for simulation are the input/output WCDMA signal collected based on the class AB power amplifier and the input/output LTE signal collected based on the Doherty power amplifier.The linearization performance of the 2D-DPD model/2D-MDDR model and the pruned 2D-DPD model/pruned 2D-MDDR model are simulated.The simulation results show that the algorithm can reduce the amount of coefficients by about half,and there will be a little loss of performance.