论文部分内容阅读
针对内部结构不详、器件参数未知的复杂电子电路电磁脉冲响应建模这一难点问题,笔者采用NARX神经网络建立动力学模型,并提出了采用正弦波扫频信号及其电路响应作为训练数据的方法,同时给出了NARX神经网络建模的理论基础及设计步骤,证明了集总参数电路响应模型可用NARX神经网络所建立的动力学模型替代,从而得到了基于数据的电子电路电磁脉冲响应建模方法。运用ADS软件完成滤波器电路及射频放大电路的设计与仿真,建立NARX神经网络模型并得到了较好的预测效果,验证了该方法适用于集总参数电路的电磁脉冲响应预测。对NARX神经网络的缺陷进行简要分析,并提出使用遗传算法优化网络参数和使用支持向量机或极限学习机替代NARX神经网络中前馈神经网络部分的改进方法,为后续研究工作指引方向。
Aiming at the difficult problem of modeling electromagnetic impulse response of complex electronic circuits with unknown internal structure and unknown device parameters, the author uses NARX neural network to establish a dynamic model and proposes a method of using sinusoidal swept signal and its circuit response as training data At the same time, the theoretical foundation and design steps of NARX neural network modeling are given. It is proved that the lumped parameter circuit response model can be replaced by the dynamic model established by NARX neural network, and the electromagnetic impulse response of electronic circuit based on data is obtained method. The ADS software is used to design and simulate the filter circuit and the RF amplifier circuit. The NARX neural network model is established and a good prediction result is obtained. The proposed method is validated to predict the electromagnetic impulse response of the lumped parameter circuit. This paper briefly analyzes the defects of NARX neural network and proposes an improved method that uses genetic algorithm to optimize the network parameters and replace the feedforward neural network part of NARX neural network by using SVM or extreme learning machine, which will guide the research work in the future.