A novel structure parameter estimation method for microstrip bandpass filter (BPF) based on multilayer fully connected network (FCN) is proposed in this letter. We design a network called the main-sub (MS)-Net, which includes the main-network to estimate the structure parameters and the subnetwork to predict the frequency responses of the BPFs. Compared with other neural network-based optimization methods, the MS-Net can generate its own data during the learning process without the need of collecting data sets and pretraining the network. The structure parameters estimated by Main-Net will gradually satisfy the design specifications in the directly iterative learning process. To demonstrate the validity of the proposed method, it was used for designing a third-order and a fifth-order microstrip BPFs. The experiment results show that the proposed method is valid and effective.

中文翻译：

---

一种基于多层FCN的微带BPF结构参数估计方法

本文提出了一种基于多层全连接网络（FCN）的微带带通滤波器（BPF）结构参数估计方法。我们设计了一个称为主子 (MS)-Net 的网络，其中包括用于估计结构参数的主网络和用于预测 BPF 频率响应的子网络。与其他基于神经网络的优化方法相比，MS-Net 可以在学习过程中生成自己的数据，而无需收集数据集和预训练网络。Main-Net 估计的结构参数会在直接迭代学习过程中逐渐满足设计规范。为了证明所提出方法的有效性，它被用于设计三阶和五阶微带 BPF。