基于深度神经空间映射的小型化微波无源器件自动化设计方法研究

With the development of microwave technology, microwave passive components in communication and radar systems are moving toward high-frequency, high-speed, miniaturization and multi-function. The performance degradation caused by parasitic effects and dispersion effects in the passive devices is becoming more and more serious. The design of these kinds of complex devices are more difficult and time consuming, severely constraining R&D efficiency and increasing R&D costs. This project is intended to conduct in-depth research in the modeling, synthesis, diagnosis and automation design of miniaturized passive components. The details include: 1) Studying the physical and mathematical hybrid modeling methods of miniaturized passive components, realizing the rapid and accurate prediction of the performance of the quasi-lumped filters, coupled resonator filters, wide-band power dividers and couplers. 2) Investigating on the synthesis methods of passive components with specific topology and the stable and fast diagnosis methods. 3) Research on the deep neural space mapping methods and parallel acceleration technology to realize rapid optimization of complex passive components with high dimensional design parameters. With the deep neural space mapping algorithm as the core, a rapid design platform for microwave passive components is constructed. The research of this project will provide a complete set of synthesis, modeling, simulation and rapid optimization technologies for the design of high-performance miniaturized passive components, and realize automation with no-human-in-the-loop as far as possible.

随着微波技术的发展，通信和雷达系统中微波无源器件不断往高频高速、小型化、多功能方向发展，器件内的寄生效应和色散效应引起的性能恶化问题越来越严峻，设计这样的复杂器件更加困难和耗时，严重制约了研发效率，增加了研发成本。本项目拟在小型化无源器件的建模、综合、诊断以及自动化设计等方面进行深入研究。具体包括：1）研究小型化无源器件的物理、数学混合建模方法，实现准集总滤波器，耦合谐振器滤波器、宽带功分器和耦合器性能的快速准确预测；2）研究无源器件特定拓扑结构的综合技术以及稳定、快速的诊断分析方法；3）研究深度神经空间映射方法、并行加速技术，实现具有高维度设计参数的复杂无源器件的快速优化。并以深度神经空间映射算法为核心，构建微波无源器件自动化快速设计平台。本项目的研究将为高性能小型化无源器件的设计提供一整套综合、建模、仿真和快速优化技术，并实现自动化基本做到无人干预。

随着信息技术不断演进，通信系统中微波无源器件不断往小型化、多功能方向发展，器件内的寄生和色散效应引起的性能恶化问题越来越严峻，设计这样的复杂器件更加困难和耗时，严重制约了研发效率，增加了成本。本项目针对无源器件自动化设计的核心科学问题从器件结构建模、电路综合与诊断与调试算法等方面进行深入研究。包括：.1）研究了深度神经空间映射算法的基本原理、架构以及训练优化算法，实现了基于神经网络和物理、数学模型的无源器件建模方法；.2）研究并实现了准集总滤波器，耦合谐振器滤波器、阶梯阻抗功分器和耦合器等无源器件的综合方法与优化调试算法；.3）构建了针对IPD工艺的滤波器自动化设计软件。.该软件为IPD滤波器的设计提供一整套综合、诊断与快速调试技术。相比于商业软件的优化算法，IPD自动设计软件能减少迭代次数约80%，有效的提高了设计效率，基本做到无人干预。本项目研究成果共发表IEEE论文5篇，申请中国发明专利2项，申请软件著作权3项。