| 摘要: |
| 为解决广义频分复用(Generalized Frequency Division Multiplexing，GFDM)系统中由于高功率放大器(High Power Amplifier，HPA)引起的非线性失真，在考虑放大器测量噪声的情况下，提出了一种基于实部反馈和列文伯格-马奎尔特算法(Real Valued Feedback Levenberg Marquard Predistortion,R LM PD)的自适应预失真方案。该方案采用记忆多项式模型(Memory Polynomial，MP)模拟HPA的逆函数，只利用输出反馈信号和期望信号的实部分量计算预失真器系数。同时，该方案选择收敛速度快、精确度高的LM算法进行参数辨识。仿真结果表明，该方案相比传统直接学习结构可以减少一个反馈支路，在信噪比为16 dB时，误比特率可达到5.1 ×10 -6 ，归一化均方误差相较无预失真时降低了约 17 dB。 与现有的一些补偿方案相比，该方案具有更好的线性化和抗噪声性能。 |
| 关键词: 广义频分复用 数字预失真 实部反馈 列文伯格-马奎尔特算法 测量噪声 |
| DOI： |
|
| 基金项目: |
|
| An adaptive predistortion scheme based on real valued feedback and Levenberg-Marquard algorithm in GFDM system |
| MIAO Shuo,NIU Andong,LIU Jianing,LI Yingshan |
| (1.College of Electronic Information and Optical Engineering,Nankai University
2.Tianjin Key Laboratory of Optoelectronic Sensor and Sensing Network Technology;1.College of Electronic Information and Optical Engineering,Nankai University
3.Tianjin Key Laboratory of Optoelectronic Sensor and Sensing Network Technology;1.College of Electronic Information and Optical Engineering,Nankai University
4.Tianjin Key Laboratory of Optoelectronic Sensor and Sensing Network Technology;1.College of Electronic Information and Optical Engineering,Nankai University
5.Tianjin Key Laboratory of Optoelectronic Sensor and Sensing Network Technology) |
| Abstract: |
| In order to solve the nonlinear distortion caused by the high power amplifier(HPA) in the generalized frequency division multiplexing(GFDM) system,this paper proposes an adaptive predistortion scheme based on real valued feedback and Levenberg Marquard predistortion(R LM PD) algorithm,considering the measurement noise of the amplifier.This scheme uses a memory polynomial(MP) model to simulate the inverse function of HPA,and only uses the real part of the output feedback signal and the expected signal to calculate the predistorter coefficients.At the same time,it selects the Levenberg Marquard(LM) algorithm with fast convergence speed and high accuracy for parameter identification.The simulation results show that the scheme can reduce one feedback branch compared with the traditional direct learning structure,and the bit error rate can reach 5.1×10 -6 when the signal to noise ratio is 16 dB,and the normalized mean square error is about 17 dB lower than that without predistortion.Compared with some existing compensation schemes,this scheme has better linearization and anti noise performance. |
| Key words: GFDM digital predistorter real valued feedback Levenberg Marquard algorithm measurement noise |
