中文摘要：

针对高功率放大器（High Power Amplifier,HPA）的神经网络（Neural Network,NN）预失真器非直接学习方法中存在的预失真性能缺陷和直接学习方法中存在的计算复杂的弊端,本文基于非直接方法得到了HPA后逆滤波器的精确辨识,利用非线性算子的运算性质及一种近似方法分别推导出了新的NN预失真器学习结构及其相应的自适应算法。该算法由HPA的后逆滤波器辅助,直接产生HPA的前逆滤波器的输出。与直接学习方法相比,它大大简化了计算复杂度。仿真结果表明,本文提出的NN预失真器学习方法可以有效地改善非直接学习方法的预失真效果,进一步降低邻信道功率比约5dB。

英文摘要：

To circumvent the predistortion limitation of high power amplifiers （HPAs） in the indirect neural network （NN） predistorter learning methods and the computational complexity of direct methods, a novel learning structure and its corresponding algorithm are derived by the nonlinear inverse operator property and an approximation formula, respectively. This method is based on the precise identification of the HPA post-filter, and directly generates HPA pre-filter output, which greatly reduces the computational complexity of current direct learning methods. Simulations show the proposed method outperforms the indirect learning method in the term of about 5dB adjacent channel power ratio improvement.