| With the advent of mobile internet and the information explosion, the limited spectrum resources are required to carry more and more data traffic. In order to meet this requirement, a variety of broadband non-constant envelope modulation techniques have been implemented to the next generation wireless communication. Compared with the traditional communication system, the new generation wireless communication systems usually transmit signals with wider bandwidth and higher peak to average power ratio. Since the analog devices in the transmit pathway and such as power amplifier are offer with memory and nonlinear, thess transmitted signals will suffer from serious in-band distortion and out-band spectrum extension. Thus, power amplifier linearization technique is a key technology for the transmitter design. Baseband digital predistortion, as one of the most cost-efficient linearization techniques, has attracted much attention recently, due to it flexibility and functional integrity.This dissertation applies the nonlinear inverse system theory in the field of automatic control and adaptive filtering techniques in the field of signal processing to the nonlinear behavioral model designation and parameters extraction. The main contributions and innovations of this dissertation are summarized as follows:1. The dissertation systematically summarizes the main theory and techniques of the digital predistortion and nonlinear behavioral model, as well as the latest research in the field.2. Taking nonlinear auto-regressive moving average (NARMA) model as an example, three parameter extraction algorithms are derived based on three learning architectures. The effectiveness of the algorithms and the superior performance of the model are illustrated through simulation experiments. The three algorithms basically meet the needs of the parametric predistorter modeling. Specifically, the nonlinear filtered-X recursive least squares (NFXRLS) algorithm, used to predistorter parameter extraction in direct learning architecture, has fast convergence and strong stability, which is suitable for many parametric predistorter model structures and not sensitive to amplifier modeling accuracy. 3. A novel nonlinear behavioral models, which is the model of simplicial canonical piecewise linear (SCPWL) function, is introduced. Various analyses, experiments and discussions for the model are presented in the dissertation. Two special complex valued SCPWL models are deduced for the first time without increasing the scale of its dimension and parameters through basis fitting method. Finally, the memory nonlinear behavioral model is constructed through the special complex valued SCPWL model, and then expanded to be a model classs.4. Aiming at some problem about the multi-lookup table NARMA model and taking into account the characters of NARMA structure and special complex valued SCPWL function, a kind of new predistorter implementation architecture is proposed. The architecture is constructed by a number of BPCs (BPC, basic predistortion cell) and can be implemented on a FPGA or DSP chip. The feasibility and effectiveness of the design is verified by a FPGA evaluation module.5. For the synchronization problem between the digital input signal and the output feedback digital signal, a loop-delay estimation scheme, including the interger loop-delay estimation and the fraction loop-delay estimation, is proposed. The scheme can be flexibly configured according to the practical situation, and achieve higher final estimation accuracy. The software and hardware simulation environment eventually verified the validity and reliability of the loop-delay estimation and adjustment.6. In order to verify the theories and algorithms, software simulation and hardware experimental platform is built. Especially the hardware experimental platform provides a necessary means and a reference for the actual performance test and product design. Parts of the test validation work have been completed by the hardware test bench. |
PDF Full Text Request