【作者】 金哲；

【导师】 宋执环； 何加铭；

【作者基本信息】 浙江大学 ， 控制科学与工程， 2007， 博士

【摘要】 随着用户数量的迅速增长和宽带通信业务的开展，现有的通信频段已经变得越来越拥挤。为了提高频谱利用率，现代无线通信系统中广泛采用了非恒包络的线性调制方式和多载波技术，这对射频功率放大器的线性度提出了很高的要求。而为了提高功率效率，一般让射频功率放大器工作在接近饱和点，此时非线性变得十分严重。要解决功率效率和频谱效率间的矛盾，线性化技术十分关键。预失真是一种有广阔发展前景的线性化技术，其效果的好坏，与所建立的功放模型有着密切的联系。在近年来迅速发展的WCDMA和OFDM等宽带通信系统中，射频功率放大器的记忆效应显著，传统的无记忆建模和预失真技术，很难获得理想的效果。本论文运用非线性系统的相关理论和方法，对宽带通信系统中，有记忆射频功放的建模和预失真方法进行了研究，主要内容如下。1．分析了射频功率放大器的主要非线性特性，阐述了放大器建模的基本理论。2．提出了射频功率放大器记忆效应的分析方法。在宽带通信系统中，放大器的记忆效应十分显著，传统的无记忆模型无法分析由记忆效应引起的现象。本文在深入分析记忆效应的表现形式和产生机理的基础上，提出了基于Hammerstein模型和Volterra级数模型的记忆效应分析方法，分别对有记忆功放的特性曲线和IMD不对称现象进行分析，使理论模型和实际观测到的现象能够很好地吻合。3．提出了基于小波网络的射频功率放大器建模方法。在该方法中，采用延时复值小波网络构造有记忆功放模型，并将训练样本、小波函数和放大器本身的特性相结合，对模型进行初始化，然后运用梯度算法训练模型中的复参数。由于复值网络能直接处理放大器输入、输出信号的复包络，所以模型的结构比较简单。仿真结果表明，相比于文献中的BP神经网络模型，所建立的模型有较快的收敛速度和较好的时、频域性能。4．阐述了预失真技术的原理，着重研究了基于查询表和基于工作函数的预失真器实现方法，并对二者的特点进行了比较。5．提出了基于简化Volterra级数的放大器预失真线性化方法。一般形式的Volterra模型由于参数数量庞大，很难直接用于高阶有记忆预失真器设计。本文建立了一种基于简化Volterra级数的预失真方法。该方法首先结合放大器自身的特性，对一般形式的Volterra模型进行简化，并在此基础上构建预失真器。预失真系统的实现采用间接学习结构，并运用限定记忆递推最小二乘算法更新预失真器参数，以提高线性化系统实时跟踪功放特性变化的能力，减小在线辨识所需的数据存储空间和计算量。最后，通过计算机仿真，验证了所建立的预失真方法的有效性。更多还原

【Abstract】 Since the number of users and broadband communication traffic rise rapidly,radio frequency resources are becoming more and more scarce. To enhance spectrum efficiency, linear modulation and multicarrier techniques are widely adopted in modem wireless communication systems. Their non-constant envelope signals require to be amplified linearly. However, in order to increase power efficiency, PAs are often driven into the saturation region, where nonlinear distortion is severe.Linearization techniques play a key role to resolve the conflict between spectrum efficiency and power efficiency.Predistortion is a promising technique to linearize power amplifiers. Its effect heavily depends on the corresponding PA model. In broadband communication systems, such as WCDMA and OFDM, the memory effects can no longer be ignored and the performance of traditional memoryless modeling and predistortion method is seriously degraded.In this paper, methodologies of nonlinear systems are applied to model and predistort PAs with memory effects in broadband communication systems. The contents of the paper are as follows.1. The main nonlinear characteristics of power amplifiers and the basic modeling theory are described.2. Methods of analyzing memory effects of PAs are proposed. In broadband communication systems, memory effects of PAs are significant and memoryless model cannot analyze the phenomena resulted by memory effects. After investigating the phenomena and the causes of PA memory effects, methods based on Hammerstein model and Volterra series are proposed to analyze the dynamic PA characteristic curves and IMD asymmetry respectively. Therefore, the theoretical models can explain the observation results well.3. A wavelet network based method is developed to model RF PAs. Firstly, a complex wavelet network with tapped delay lines is applied to construct the model of PAs with memory. Secondly, the network is initialized according to the training data set, the wavelet function and the characteristics of the PA. Finally, the established network is trained by gradient techniques. Since the proposed mode can process complex signals directly, much simpler network architecture is achieved. The simulation results show that compared with the previously published BP model, the proposed model provides faster convergence rate and better performance in the time domain and frequency domain.4. The principles of predistortion techniques are described. The look-up-table based and workfunction based predistorter are discussed and compared in detail.5. A predistortion method based on simplifed Volterra series is proposed to linearize power amplifiers. The general Volterra model can hardly be utilized to design the higher order predistors with memory because of its high computational complexity. In this paper, a predistortion method using simplifed Volterra series is developed. The general Volterra model is simplified according to the PA characteristics and then the predistorter is designed using the simplified model. The indirect learning architechture is applied to implement the linearization system and the recursive least squares algorithm with a sliding rectangular window is utilized to identify the predistorter parameters. Therefore, the predistortion system can trace the variation of the PA characteristics effectively and the data storage space and computational complexity are also decreased. Finally, the performance of the proposed predistortion method is validated through computer simulations.更多还原