Research On Nonlinear Characteristics Of Radio Frequency Transmitter For Wireless Communication

With the continuous development of mobile communication system towards the direction of high rate,large capacity and ultra-wideband,the signals processed by these new technologies have many characteristics,such as multi-carrier,multi-level,ultra-high bandwidth and higher peak-to-average power ratio(PAPR),etc.,which puts forward a great challenge for the radio frequency(RF)transmitter used in the communication system.On the one hand,the transmitter excited by wide bandwidth signals will present stronger and deeper nonlinear memory characteristics.On the other hand,the RF transmitter system is inevitably affected by the modulator in-phase/quadrature(I/Q)imbalance,the local oscillator(LO)leakage and the nonlinear distortion of the amplifier etc.Meanwhile,these different nonlinear characteristics interact with each other,which will seriously degrade the performance of the communication system.Therefore,it is the core technology to solve the reliable transmission of wireless communication in the future by researching a new type of broadband efficient linear transmitter,which make the RF front-end of wireless broadband transmission work efficiently under the premise of meeting the strict linearity of the system.In view of the above requiremes,this dissertation firstly applies compressive sensing theories and adaptive signal processing algorithms in the field of signal processing and machine learning theories in the field of artificial intelligence to studying and discussing some problems about nonlinear characteristics identification and joint compensation of broadband transmitter in detail.Secondly,to improve the linearity and efficiency of transmitter under wide bandwidth signal excitation,a joint compensation scheme of the transmitter based on the three-input joint model is proposed,which can fully improve the distortion suppression ability in view of the transmitter joint compensation.The main contributions and innovations of this dissertation are summarized as follows:1.An adaptive sparse predistorter design method based on compressed sensing theory and adaptive signal processing algorithm is proposed.Aiming at the inherent batch operation mode in the traditional compressive sensing PA model simplification algorithm,the integration of compressed sensing greedy algorithm and adaptive signal processing theory is studied,and a broadband adaptive sparse predistortion system is constructed.The two sparse adaptive parameters updating algorithm is designed by using conjugate gradient and stochastic gradient descent algorithm combined with subspace tracking greedy algorithm respectively.Simulation and experimental results show that the proposed algorithm can effectively construct a sparse adaptive predistorter with only a few parameters,and the nonlinear distortions and memory effects of the PA in the predistortion system can be compensated adaptively.Compared with the non-sparse full model predistortion technique and the batch pruning method,the proposed algorithm has a faster convergence speed,and can improve the tracking ability and reduce the number of parameters of the original model by more than 60%,which fully verifies the advantages of the proposed adaptive sparse method.2.A power amplifier(PA)behavioral model pruning method based on sparse principal component analysis(SPCA)in machine learning theory is proposed.In this method,the linear transformation of the PA behavioral model is carried out for achieving the purpose of data matrix dimensionality reduction without loss of information.In this scheme,the model coefficients are reduced by transforming the model from high dimension to low dimension,and the sparse algorithm is introduced to reduce the calculation amount in the process of data matrix dimension reduction.On the one hand,this method can retain the important information of the original model structure to the greatest extent.On the other hand,SPCA sparse method greatly reduces the computational complexity of the traditional PCA method in the process of linear transformation of dimensional reduction by sparse processing of load vectors,and the number of model parameters can be decreased to one third of the original full model.Simulation and experimental results show that the simplified model not only has the same precision as the original full model,but also has higher numerical stability.3.A method for RF transmitter behavioral modeling based on least squares support vector machine(LSSVM)is proposed to improve the accuracy of the transmitter behavioral model.In this method,by using support vector machine(SVM)model in machine learning theory instead of traditional PA behavioral model based on Voltterra series,a sparse LSSVM algorithm based behavioral model for solving large-scale data training is proposed,and the basic principle of sparse LSSVM model and detailed parameter extraction algorithm are given.Using only a limited number of training samples,the proposed model can represent the characteristics of transmitters with I/Q imbalance and dc-offset with high accuracy.Simulation and experimental results show that the model can achieve the normalized mean aquare error(NMSE)of-36.76 dB,and better model performance than traditional generalized memory polynomial(GMP)model(-18.3 dB),conjugate generalized memory polynomial(CGMP)model(-27.91 dB),ordinary support vector regression(SVR)model(-31.2 dB)and ordinary LSSVM model(-35.12 dB)when I/Q imbalance,dc-offset and PA nonlinear distortion are all taken into consideration.In addition,the training time and running time of the proposed model are more than 90% lower than the similar kind model.4.To overcome the nonlinear interactive effects of RF transmitter impairments,a three-input joint compensation model,which is composed of the nonlinear frequency-dependent cross terms between the I and Q branches as well as the magnitude of the input signal,is proposed.In this new model structure,the enhanced model base sets generated by the increased input signal amplitude of the transmitter can be used to mimick the dynamic AM/AM and AM/PM characteristics of the PA sub-module,and the overall performance of the model is superior to the traditional I/Q imbalance model obviously.Based on that,an adaptive greedy algorithm based on robust quasi-Newton(RQN)theory is used to prune model parameters online,and the model coefficients can be reduced to less than 38% of the full model without reducing the performance of the system.Simulation and experimental results show that the proposed sparse model predistorter has higher linearization performance than other popular joint compensation models,which provides a highly efficient and accurate linearization solution for direct frequency conversion transmitters.