【作者】 刘策含； 金鹏飞； 周俊； 张雅鑫； 杨梓强；

【Author】 Cehan Liu;Pengfei Jin;Jun Zhou;Yaxin Zhang;Ziqiang Yang;School of Electronic Science and Engineering, University of Electronic Science and Technology of China;

【机构】 电子科技大学电子科学与工程学院；

【摘要】 本研究提出了一种基于神经网络的微带滤波器性能预测方法,针对一个太赫兹频段微带滤波器进行预测。该神经网络结合多任务训练和共享权重机制,通过利用ADS仿真生成S参数(S11和S21)数据集,以物理结构参数和频率作为输入,S参数作为输出,实现了微带滤波器性能指标的预测。相比传统全连接神经网络,本研究提出的神经网络架构提高了泛化能力,有效感知出全频段特性并能预测出S参数趋势,实现了较为精确的频率响应预测。仿真结果表明,该方法所预测出的S参数与ADS仿真结果拟合度高,验证了其在微带滤波器建模方面的准确性和有效性,为微带滤波器优化设计和后续的逆向设计提供了一种高效的智能建模方法。更多还原

【Abstract】 This study proposes a neural network-based performance prediction method for microstrip filters, targeting a THz-band microstrip filter for prediction. The neural network combines multi-task training and shared weight mechanisms, utilizing an ADS simulation-generated S-parameter(S11 and S21) dataset. Physical structure parameters and frequency are used as inputs, with S-parameters as outputs, to predict the performance metrics of the microstrip filter. Compared to traditional fully connected neural networks, the proposed neural network architecture enhances generalization capability, effectively captures full-band characteristics, and predicts the trend of S-parameters, achieving accurate frequency response predictions. Simulation results show that the predicted S-parameters match well with the ADS simulation results, validating the accuracy and effectiveness of this method in microstrip filter modeling. This approach provides an efficient intelligent modeling method for microstrip filter optimization and subsequent inverse design.更多还原