记忆功放的BP神经网络分离预失真方法

Separate BPNN-predistortion method for nonlinear HPA with memory

针对OFDM无线通信系统中的记忆非线性功率放大器预失真困难以及预失真精度不高的问题,充分考虑了记忆非线性功率放大器的失真特性及其结构特点,提出了神经网络分离预失真方法,并给出了该方法的原理、实现结构和自适应算法.该方法把记忆非线性功率放大器的预失真分为3个模块来实施,即抵消记忆效应模块、AM/AM失真矫正模块和AM/PM失真矫正模块,并基于神经网络预失真器非直接学习结构,利用Levenberg-Marquardt BP算法确定各个神经网络预失真器.仿真结果表明该方法可降低邻信道互调功率约30 dB.

To circumvent the predistortion limitation for nonlinear high power amplifiers with memory （NHPAM） in OFDM communication systems, a separate BPNN （back propagation neural network） predistortion method with its structure and algorithm and their principles is proposed by considering the distortion characteristics and structure features of NHPAM. This method separates the whole predistortion for NHPAM into memory, AM/AM and AM/PM predistortion subsystems. Then the three NN predistorters are identified by the Levenberg-Marquardt BP algorithm based on the indirect training method. Simulation results show this proposed method can reduce ACEPR （Adjacent Channel Error Power Ratio） 30dB.