Research On Blind Identification Of Nonlinear Distortion Of Digital Reciever Based On Second And Higher-Order Statistic

This dissertation focuses on the issue of blind identification of digital receiverr. According to the theory of system identification, digital receiver is considered as "black box", namely, we consider the whole digital receiver as a nonlinear system, rather than researching the nonlinearity from each sub-module. By using an algorithm based on Second-Order-Statistic, the identify performance of Hammerstein model and Volterra model are compared. The memory effect and nonlinearity of Hammerstein is blindly identified based on Higher-Order-Statistic, we derives several blind identification and compensation method. The main contributions of this dissertation can be concluded as follows:1,Analyzing and summrizationg several typical nonlinear models, including nonlinear model without or with memory effect, such as power series model, orthogonal polynomial model, Volterra model and block-oriental model. This thesis analyzes the characteristics of those different models, such as calculation complexity, relationship of those nonlinear models. Summarizes the inverse models of those nonlinear models; analyzes the sampling requirement of nonlinear system identification and compensation, the theorital derivation and simulation results show that though Nonlinear systems usually cause spectral spreading resulting in an output signal bandwidth that is greater than the input signal bandwidth, sampling at the Nyquist rate of the output signal is usually not necessary, and that a nonlinear system can be identified and compensated at the Nyquist rate of the input signal.2,A blind identification and compensation algorithm based on second order statistic is developed. The coefficients of nonlinear system are identified blindly by applying a least-squares criterion to the out-of-band spectral content of the nonlinear compensator output. The performance of algorithm is analyzed. By using different nonlinear model in nonlinear compensator, those models'performances are compared. The simulation and test results show that as to the digital transceive, the performance of Hammerstein model is better than Volterra model.3,Two typical HOS identify algorithms are derived, such as analitical algorithm and itterative algorithm. The simulation results are given. But those two kinds of algorithms could not be used to determinate the memory depth and nonlinear order of unknow system. A new HOS algorithm is proposed to identify the memory effect of Hammerstein model, including memory depth determination and coefficients extraction. The memory depth determination is converted into finding the rank of extended matrix constructed by cumulants of system output. In order to yield the rank of cumulant matrix, we propose NPODE (Nonlinear Product of Diagonal Entry) method which is the extension of PODE, and simulation verifies its robustness by comparing with GM method and inflexion method. Finally linear block coefficients extraction method is given, the theoretical derivation and simulation indicated that the extraction process is not affected by the strength of nonlinearity of Hammerstein model. In order to extract coefficients more robustly, by using higher order cumulant of output signal, two sets linear equations are proposed to extract coefficients of linear block with memory. Theoretical derivation shows that those two sets of linear equations have unique solutions. They could be used alternately to identification the memory effect of Hammerstein model, and the identification process does not affected by memory-less nonlinear block. Finally, simulations verify that the new developments have higher performance than direct extraction method.4,The normalized kurtosis and its application in blind identification of weak nonlinear system is derived. The definition of normalized kurtosis and its some useful properties in system identification are presented;according to the definition and properties, the influences of memory depth and nonlinear order on normalized kurtosis are derived theoretically, and simulation result demonstrates the rule of normalized kurtosis varying with the change of system characteristics. This shows that normalized kurtosis has the ability to identify weak nonlinear system accurately. Accordingly, this paper proposes a step by step method to blindly identify extremely weak nonlinear system whose SFDR(Spurs Free Dynamic Range) is up to 85dBFS(dB Full Scale) by using normalized kurtosis. Finally, combined with the proposed methods, this paper analyzes the potential value and accuracy advantage in blindly identifying and compensating the weak nonlinear Hammerstein system.