Research On Digital Predistortion Of Power Amplifier Based On Attention Recurrent Neural Network

Power amplifier is one of the important components to ensure the complete function of wireless communication system.However,due to some modulation techniques and thermal effects,the power amplifier has nonlinearity.Due to the nonlinearity of the power amplifier,the bit error rate of the whole wireless communication system will be greatly increased,and it will also interfere with the neighboring channels,so the nonlinearity of the power amplifier needs to be alleviated urgently.There are many linearization schemes to alleviate the nonlinearity of power amplifier,among which digital predistortion technology is a more effective linearization scheme.In this thesis,the Long short-term Memory(LSTM)network,which are good at processing time series in the field of deep learning,and attention mechanism are used to improve the modeling accuracy of power amplifier behavioral model,so as to achieve the purpose of improving the existing digital predistortion technology.The nonlinearity of power amplifier has corresponding evaluation indexes.This thesis first gives a detailed description of the commonly used evaluation indexes of power amplifier nonlinearity,and summarizes several power amplifier behavioral models with and without memory effect.Then,the traditional BP neural network and LSTM are used to model the behavior of power amplifier respectively,and the modeling effects of the two models are compared under the condition that the model structure and the memory depth of the model are equal.The experiment proves that the LSTM model is better than the traditional BP neural network in the behavior modeling of power amplifier.There are many delayed input items in the power amplifier behavioral model.The purpose of attention mechanism is to assign different weights to these delayed input items and then input them into the model,so as to improve the effect of the model.Since experiments have shown that the effect of LSTM model is better than that of the traditional BP neural network,the subsequent experiments are based on the LSTM power amplifier behavioral model.The power amplifier was then modelled by adding an attention mechanism to the LSTM.Comparing the effects of the three models,it is found that the effect of the LSTM model with attention mechanism is significantly better than that of the LSTM model without attention mechanism.In addition,an important conclusion is that the attentional mechanism can prune the model and significantly reduce the number of model parameters.Finally,LSTM model with attentional mechanism is used to build the inverse model of power amplifier behavioral model,and this model is used as a predistorter.The experimental results show that after the original signal is corrected by the pre-distorter in this way,the Adjacent Channel Power Ratio(ACPR)of the output signal of the power amplifier is obviously reduced,which illustrate the feasibility of LSTM model with attentional mechanism in predistortion system.