描述了一种基于实数延时模糊神经网络的有记忆效应的功率放大器模型。该模糊神经系统即自适应模糊神经推理系统,采用模糊c类均值聚类方法来减少模型的规则数目和简化模型结构。在训练过程中,采用最小二乘和反向传播相结合的高效算法提取模型参数。在测试平台上用三载波WCDMA宽带信号对射频功率放大器进行测试,并借助矢量信号分析仪采样功率放大器输入和输出数据,成功地对模型进行了训练和验证。通过和实数延时神经网络模型(RVTDNN)比较,该模型的收敛速度远快于这些前馈结构的神经网络模型。比较和分析时域和频域结果表明模型有很好的性能,其归一化均方误差达-38dB。
This paper describes the implementation of a real-valued time-delay neuro-fuzzy system for behavioral modeling of power amplifiers with memory effects. The neuro-fuzzy system is called adaptive neuro-fuzzy inference system ( ANFIS), and the fuzzy c-means clustering method is adopted tO decrease the number of rules and simplify the structure of the system. In the training process, the efficient hybrid algorithm is used to identify the parameters, which applies a combination of the least-squares method and the back-propagation gradient descent method. The proposed model has been successfully trained and validated with three-carrier WCDMA signal in the test bench, in which the input and output signals of PA are sampled by a vector signal analyzer (VSA). The convergence speed is much higher than that of feed forward neural networks in comparison with the real-valued time-delay neural networks (RVTDNN). The good performance has been achieved in validation with the normalized mean square error (NMSE) of -38dB by comparison between measurements and modeled results in time and frequency domain.