由於正交分頻多工(Orthogonal Frequency Division Multiplexing, OFDM)訊號具有較高的峰值對均值功率比(Peak-to-Average Power Ratio, PAPR)，容易受到功率放大器非線性的影響，而產生訊號的失真，導致系統效能降低。在過去的文獻中，使用預失真技術用於補償非線性功率放大器造成的失真為主要的研究方向。在本篇論文以有記憶性多項式模型當作功率放大器，並利用有記憶性多項式作為預失真器的模型，並利用有記憶性多項式作為預失真器的模型，提出NM單純型搜索法 (Nelder and Mead Simplex Search Method)直接搜索預失真參數，以及使用高斯-牛頓線性化法(Gaussian-Newton method)推導出有記憶性多項式模型下的演算法，最後提出最大期望値法(Expectation Maximization)結合NM單純型搜索法，並且分別利用直接學習架構(Direct Learning Architecture,DLA)以及間接學習架構(Indirect Learning Architecture,ILA)找到預失真的參數，最後本篇論文比較補償非線性功率放大器演算法的效率。 ;The characteristic of high peak-to-average power ratio (PAPR) of orthogonal frequency division multiplexing (OFDM) signals is well known to seriously degrade system performance. The predistortion (PD) technique for compensating the nonlinear power amplifiers (PAs) has become a main approach in the literature. In this paper, we consider a memory polynomial model for the PAs. Taking into account the direct learning and indirect learning structures for the PD, we study some algorithms for the PD coefficients, including the expectation maximum (EM) algorithm, Gaussian-Newton linearization method, and the Nelder-Mead simplex search method. In this thesis, the algorithm efficiency and computational complexity are compared by applying those methods for the PA compensation problem.
