I/Q Imbalance Compensation And Novel Gated Neural Network Modeling For Digital Predistortion

A power amplifiers(PA)is the most critical device in wireless transmitter front-end that is nonlinear.Digital predistortion(DPD)has become the popular technique of power amplifier linearization because of its good linearization performance,flexi-ble programming and low cost.In the future,the development trend of communication system is characterized by wide bandwidth,miniaturization and low cost.These char-acteristics bring new challenges to the traditional DPD technique.The trend of wide bandwidth and low cost makes the problem of I/Q imbalance more serious,and the wideband signal excitation with high peak to average ratio makes the power amplifier facing stronger and more complex nonlinear and memory effect.Aiming at these two problems,this paper focuses on the compensation technology of I/Q imbalance and the behavior modeling of neural network model with low complexity and high performance.In the traditional I/Q imbalance compensation model,most of them only consider the quadrature modulation imperfections(QMI)in the transmission link,not the quadra-ture demodulation imperfections(QDMI)in the feedback loop.This paper focuses on the compensator model of feedback loop QDMI,and an accurate inverse modeling ap-proach is proposed to eliminate the effect of I/Q imbalance.On this basis,it is extended to the scene of large-scale MIMO architecture.Considering the development of multi-band communication system,a new model for jointly compensating the non-ideal char-acteristics of the modulator and the concurrent multi-band PA distortion is proposed.And the performance of the proposed model is verified by experiments.At present,the traditional digital predistortion model is difficult to meet the needs of ultra wideband scene applications.Considering the strong nonlinear fitting ability of neural network,this paper proposes an instant gated neural network model based on RNN(recurrent neural network)model.Considering that the performance of traditional neural network model based on MLP(multilayered perceptron)structure is limited by the weakness of MLP structure itself in fitting complex memory effect,this paper mainly studies the structure based on RNN.In this paper,the disadvantages of RNN structure are analyzed,and the nonlinear state of power amplifier is divided into two parts:static state and memory state.Then two state control units are introduced.Based on RNN network,a new instant gated recurrent neural network structure is proposed.Then,ac-cording to the characteristics of the new structure,the corresponding behavior model is proposed.Then a series of optimization designs are carried out,and the augmented model with stronger modeling ability is proposed.Because the instant gated recurrent neural network model proposed in this paper is based on the nonlinear behavior mecha-nism of power amplifiers,the proposed models can achieve better linearization perfor-mance compared with RNN model and traditional GMP model,and have comparable performance with lower computational complexity compared with the state-of-the-art RNN-based variant models,such as gated recurrent unit(GRU)model,and these are also verified in the experiment.This dissertation reviews the existing neural network models based on vector de-composition,and analyzes some possible problems when the model input is decom-posed into I and Q inputs.In order to avoid these problems,a new instant gated neural network model based on vector decomposition is proposed in this paper.This neural network model based on vector decomposition will be more in line with the physical characteristics of the power amplifier from the physical mechanism,so the total number of parameters of the model will be less than that of the neural network model in the form of I and Q inputs,and the performance is quite or even better.Then through the anal-ysis and comparison of the traditional single-band model and the dual-band model,the dual-band neural network behavior model is constructed by constructing the dual-band input.And through the experimental comparison,it is verified that the proposed models have good performance...