Research On Predistortion And Equalization In Nonlinear Satellite Channels With Memory

In order to make full use of the power resources of ground station and satellite transponder for wide band satellite communication system,the operating point of power amplifier is usually driven near to the saturation region which leads to nonlinear characteristic.And,the memory effects are caused by wide-band signal,whilst nonlinear memory characteristic is revealed in the ground station and the satellite transponder simultaneously.For the sake of countering the imperfect distortion of channel,the baseband signal predistortion technique is studied at the transmitter,while the equalizer technique under nonlinear interference is studied at the receiver.With regard to improve the practicability,the research priorities are how to design stable predistorter with simple structure.In order to utilize the power of channel coding,the research priorities are how to design turbo equalizer by acquiring useful information from nonlinear interference terms.The main work and innovated research achievements in this thesis are as follows,1.Hybrid predistorters consisting of low order polynomial and small size look-up-table are proposed,the complexity and instability of predistorter is reduced;QR decomposition-recursive least square algorithms are adopted to solve the coefficients of memory polynomial,the instability and complexity of least square predistorter is reduced.Owing to the low accuracy of polynomial predistorter and slow convergence rate of look-up-table predistorter,polynomial with low order and look-up-table with small size are adopted to construct hybrid predistorters,while both the complexity and instability is reduced.According to different cascade structure and training order,four hybrid predistorters are proposed,and the best one has the structure of polynomial followed by look-up-table,while the polynomial part is trained firstly.Simulation results show that the hybrid predistorter outperforms the traditional one with single structure while having a lower complexity.It is instability to calculate the inversion matrix of autocorrelation matrix about the output data from power amplifier in least square predistorter,whereas,the coefficents are solved after QR-decomposing the data matrix.Matrices with low dimension are operated in predistorters using QR decomposition-recursive least square based on Givens rotation and QR decomposition – block recursive least square based on Householder transformation.Compared with traditional algorithms belonging to least square family,better performance is achieved.2.Extracting extrinsic information strictly,turbo equalizers based on linear minimum mean square error(LMMSE)for nonlinear channel with memory are proposed;Exploiting useful information from the nonlinear interference terms,turbo equalizers based on soft interference cancellation-minimum mean square error(SIC-MMSE)and joint Gaussian(JG)are proposed.Based on the affine transform,LMMSE turbo equalizers with zero and nonzero bias are derived.The extrinsic information is calculated obeying the rule ‘extrinsic information should be independent with the prior information' strictly,and the prior information is excluded in both the linear and nonlinear terms.By cancelling the inference in two steps,the SIC-MMSE turbo equalizers are derived by reserving the linear and nonlinear terms related to the detected symbol in the first step.Both the linear and nonlinear terms that contain the detected symbol are considered as desired signals,while the other terms and channel noise are considered as a total noise which is assumed to obey joint Gaussian distribution and the JG turbo equalizer is achieved.Simulation results show that the proposed turbo equalizers for nonlinear memory channel achieve better performance than the existing LMMSE turbo equalizer.Moreover,the nonlinear terms that related to the detected symbol can be exploited to improve the performance of turbo equalization.3.Aim at nonlinear inference channel,under the unify framework of factor graph(FG)and message passing algorithms,turbo equalizer based on belief propagation(BP)algorithm is proposed,passing discrete message in a tree FG,maximum a posteriori turbo equalizer is obtained;turbo equalizer based on variational message passing-belief propagation(VMP-BP)algorithm is proposed,passing continuous message in a cycle FG,turbo equalizer with low complexity is obtained.In order to easy the derivation of ‘extrinscin information',the probabilistic model of satellite communication system is represented by factor graph.BP algorithm is adopted for deriving the message related to the mapping and modulation nodes.By introducing the state nodes,the factor node of equalizer part is opened in a tree structure.The forward and backward message are derived under BP algorithm,the marginal distribution of symbol is calculated exactly,while obtaining the maximum a posteriori turbo equalizer passing discrete message.By illustrating the relationship between symbols and observations,the factor node of equalizer part is opened in a cyclic structure.The iterative messages are derived under VMP algorithm,and the continuous messages are parametrized by the canonical parameters and sufficient statistics of exponential family in a concise form.The discrete and continuous messages related to symbol node and mapping node are updated,while obtaining the turbo equalizer with hybrid message passing.Simulation results show that the proposed turbo equalizer based on a combined VMP-BP algorithm achieved better performance than the traditional LMMSE turbo equalizer with a lower complexity.