基于神经网络的功率放大器预失真建模

提出一种神经网络结合分离信号对功率放大器预失真建模的方法。将输入/输出信号的线性与非线性部分分开处理,利用神经网络良好的逼近能力,采用LM算法,拟合出功率放大器特性曲线,进而建立预失真模型,使非线性功率放大器的输入/输出曲线整体呈线性化。在保证输出幅度限制和输出功率最大化的前提下,与未作信号分离的神经网络建模方法、多项式建模方法以及Saleh函数模型方法相比较,发现信号分离神经网络建模方法能得到较小的归一化均方误差和误差矢量幅度。仿真结果表明,采用信号分离神经网络对功率放大器及其预失真建模,整体线性化误差较小、精度高、效果更佳。

Power Amplifier Predistortion Modeling Based on Neural Network

A new method was presented to establish the power amplifier predistortion model by combining neural network with the method of separating the signal. The proposed method handled the linear part and the nonlinear part of input-output signals separately. With the good approximation ability of the neural network and the LM algorithm, the power amplifier characteristic curves were fitted, and then the predistortion model was established, which could make the overall input-output curve of the nonlinear power amplifier be linear. Under the premise of ensuring output amplitude limit and output power maximization, the experimental results of the neural network model for signal separation was compared with the results of three models, including the neural network model for signal non-separation, the polynomial model and the Saleh model. It showed that the neural network model for signal separation had smaller normalized mean square error and smaller error vector magnitude. The simulation results showed that the overall linearization of the power amplifier and its predistortion model had smaller error, higher accuracy and better effect when the signal separate neural network was adopted.