Research On Predistortion Of RF Power Amplifiers In Wireless Communications

Orthogonal frequency division multiplexing (OFDM) with both high spectrum efficiency and robustness to multi-path fading is widely used in high-data-rate wireless communication systems. However, its high Peak-to-Average Power Ratio (PAPR) will result in out-of-band spectral regrowth and in-band distortion when OFDM signal passes through a power amplifier with nonlinear range. Thus, it is significant to further conduct research on the nonlinear character of amplifiers and implement the linearization technology in more ways. In addition, the technology of digital predistortion is very promising for its superiority in accuracy, complexity and adaptability. This dissertation focuses on the digital predistortion in OFDM systems. The research includes predistortion model, identification and its implementation.The research work and achievements are presented in the following:1.When Output-Backoff (OBO) is decreased, the performance of memoryless polynomial predistorter deteriorates. In order to improve the predistortion performance, this thesis proposes a dynamic coefficient polynomial predistorter based on subsection compensation. Different polynomial coefficient combination of amplitude changes is selected as the input signal to minimize the compensation error. A dynamic coefficient polynomial model is given with simplified coefficient estimation algorithm based on direct learning architecture. Theoretical and simulation analysis show that it can achieve good performance with limited order. Therefore, its hardware implementation is flexible and the design of the postpositional low-pass-filter is easier.2.A predistorter design is proposed based on Generalized Normalized Gradient Descent (GNGD) algorithm. The merit of GNGD is that its learning rate provides compensation for the assumptions in the derivation of normalized least mean square (NLMS), and therefore it improves the robustness of coefficient estimation. A detailed derivation of GNGD and its hardware architecture are given. The theoretical and simulation analysis show that it solves the initial parameter-sensitive problem existed in the predistorter using NLMS algorithm. The proposed predistorter also gets better linearization performance than the predistorter using NLMS algorithm. Its complexity is also superior to LS algorithm.3.The memory polynomial and generalised polynomial predistorter have disadvantages of power-amplifier-model selectivity and limited performance. A fractional memory polynomial predistorter design is proposed, the merit of which is that it can be more accurate and smooth with certain order. Theoretical and simulation analysis show that the proposed predistorter gets a better linearization performance without model selectivity.4.In digital predistorter, the polynomial model is very popular due to its simplicity and easiness to implement. The polynomial effective order determines the design of low pass filter and the effectiveness of the linearization results. As far as we know, there is no paper at resent concerning the polynomial effective order in predistortion. For the indirect learning predistorter, this thesis proposes an effective order estimation method. The effective order can be acquired according to normalized singular value after singular value decomposition (SVD). A detaild derivation is presented. Theoretical and simulation analysis show its efficiency.