Researches On Compensation Techniques For Nonlinear Distortion In Receiver Front-end

Wide-band digital receivers usually face the problem that the desired weak signals are probably submerged by the nonlinear distortions generated by large signals or strong interferes within the same band, trying to increase the spurs-free dynamic range of the receiver front-end could greatly improve that situation. This thesis firstly introduces nonlinear system identification techniques to the compensation of those nonlinear imperfections caused by the receiver front-end as a weakly nonlinear system. The technique is implemented by way of adaptively adjusting the parameters of a digital nonlinear compensation model subsequent to the digital receiver front-end to characterize the inverse of nonlinear transfer function of the receiver front-end for efficiently reducing or eliminating the nonlinearity of the receiver front-end. The main researches of this thesis can be concluded as follows:By means of qualitatively analysis and simulation, various error sources of noise and nonlinear distortion products in the output of digital receiver front-end and their proportion are defined and investigated. The influence of these two dominant products on the overall weak signals detection performance of the digital receiver are analyzed, which indicates that it is the nonlinear distortions that seriously limit the spurs-free dynamic range of the digital receiver front-end.Several commonly used nonlinear models with memory effect are detailed described and summarized, which are memory polynomial model,box model and Volterra model. These parametric models are also compared with each other in the fitting precision and computational complexity. The theoretical analysis shows that Volterra model has better performance in both universality and precision compared with the other models with the same order and memory depth. However, it also pays the price in its higher computational complexity. Based on that, this thesis provides a method for simplifying the Volterra model:if the frequency domain Volterra kernels are separable, the Volterra model can be simplified to parallel Wiener model; and if the frequency domain Volterra kernels are linear functions of the combined frequencies, the model can then be simplified to parallel Hammerstein model. The inverse model of the Volterra model and its simplified models are determined, which are Volterra model and parallel Wiener-Hammerstein model respectively. Experimental measurements on both an analog-to-digital device AD6645 and a direct sampling receiver front-end indicate that the changing law of their frequency domain Volterra kernels with one frequency of the input has the symmetry properties of parallel Wiener model. Therefore, the inverse model of their nonlinear transfer function for digital compensation is parallel Wiener-Hammerstein model.This thesis designs three adaptive Volterra model identification and compensation techniques for the nonlinear distortion in the digital receiver front-end. They use identification criterions of minimum mean-squared error,minimum kurtosis difference and the one based on second-order and third-order statistics respectively. The performance and advantages of these model identification methods are evaluated by means of theoretical derivation, simulation and comparison experiment. The simulation results indicate that linear performance of the systems to be compensated could all get different degrees of improvement after the model identification and distortion compensation. Among that, performance of the minimum mean-squared error based algorithm is better than the other algorithms that use statistics information, because it directly uses instantaneous amplitude information of the input.This thesis proposes two blind Volterra model identification and compensation techniques for the nonlinear distortion in the digital receiver front-end. They use identification criterions of Minimum the Power of Out-of-Band signals and Minimum the Power of signals Other than the large ones respectively. The algorithm based on the former critertion limits the frequency range of input signals, and then takes the total power of the out-of-band signals as the goal function. The algorithm based on the latter critertion distinguishes the frequency locations of large signals, and then takes the total power of the signals outside the locations as the goal function. Performances of these two criterions are evaluated by means of theoretical derivation and simulation and comparison experiment. The algorithms are tested on a direct sampling receiver front-end hardware circuits. Experimental results with multi-tone sinusoidal input and band-pass input show that with the proposed algorithms, the Spurs-Free Dynamic Range of the system would generally achieve nearly 14dB to 24dB improvement after the model identification and distortion compensation.A hybrid parallel Wiener-Hammerstein model identification and compensation techniques for the nonlinear distortion in the digital receiver front-end is proposed, the algorithm could efficiently avoid the difficulty in the direct identification techniques caused by that the output is not linear related with parameters of the modular model. The hybrid model identification employs a two-step process:the first step uses a single-tone sinusoidal signal as the test input, and then accomplishes the identification of the Wiener module in the parallel Wiener-Hammerstein model according to harmonic distortion counteract based criterion. The second step implements the adaptive identification of the linear memory module in the model by applying the blind MPOT criterion. Performance of this algorithm is tested on a direct sampling receiver front-end hardware circuits. Experimental results with multi-tone sinusoidal input indicate that with the proposed algorithm, the SFDR of the system would achieve beyond 30dB improvement after the model identification and distortion compensation.