Russian Scientists Propose Method to Speed Up Microwave Filter Design

© iStock

Researchers at HSE MIEM, in collaboration with colleagues from the Moscow Technical University of Communications and Informatics (MTUCI), have implemented a novel approach to designing microwave filters—generative synthesis using machine learning tools. The proposed method reduces the filter development cycle from several days to just a few minutes and in the future could be applied to the design of other microwave electronic devices. The results were presented at the IEEE International Conference '2026 Systems of Signals Generating and Processing in the Field of on Board Communications.'

Microstrip components are topological elements of a printed circuit board that enable the transmission, reception, and processing of radiofrequency or microwave signals. They are widely used in telecommunications, satellite navigation, wireless communication, radar, and other systems. These components are valued for their compact size, relatively low production cost, and the ability to be fabricated directly on the circuit board without the need for additional parts or complex assembly.

However, designing such devices is not straightforward. Their performance strongly depends on topology—the geometric dimensions of the traces, the spacing between them, and the shape and type of resonators. Even minor deviations in these parameters can significantly affect device characteristics, such as tuning to the desired frequency range or the level of reflected signal. Therefore, precise parameter selection is essential.

Analytical methods and computer-aided design tools are typically used for this purpose. However, the process cannot be fully automated, and engineers still need to perform numerous electromagnetic simulations and manually adjust parameters. The task becomes particularly challenging when multiple factors must be considered simultaneously. In addition, the relationship between device topology and performance is often nonlinear: even small changes in geometric dimensions can have a disproportionately large impact on operation, making the outcome difficult to predict in advance. As a result, the design process turns into a lengthy search through many possible configurations. 

Generative synthesis of microwave devices using machine learning helps simplify this process. This approach was developed and applied by a research team at HSE MIEM led by Prof. Andrey Yelizarov and student project supervisor Artiom Katsnelson, Senior Lecturer at MIEM, in collaboration with colleagues from MTUCI—Prof. Grigory Aristarkhov and Senior Lecturer Oleg Arinin. By solving the synthesis problem with ML algorithms, the method enables automatic generation of device topology and geometric dimensions based on specified electrical characteristics and parameters. 

Andrey Yelizarov

'Most tools for designing microwave devices address the analysis problem. However, this typically involves a long series of iterations: for a given device topology, numerous computer simulations are performed, and the resulting characteristics are then used to optimise and refine the geometric dimensions of the device. We propose replacing this process by solving the inverse problem—generative synthesis of microwave devices using machine learning algorithms. In this approach, the topology is modelled and its geometric dimensions are calculated from the specified electrical characteristics and filter parameters,' comments Prof. Yelizarov. 

The researchers compiled a structured dataset of 16,250 parametric configurations of microstrip filters, generated using an automated pipeline they developed in Python along with CST Studio Suite software. Using this dataset, they trained and compared four machine learning algorithms, of which XGBoost proved to be the most accurate, achieving an average error of just 0.51% across the ten target filter parameters. 

Further verification confirmed that the approach proposed by the authors captures real physical patterns rather than relying on coincidences in the training data. This enables the synthesis of both the topology and geometric dimensions of a filter based on specified characteristics, as well as the evaluation of the resulting structure’s properties. According to the authors, generative synthesis using machine learning can reduce design time from several days of engineers’ work to just a few minutes. In the future, the developed software could be integrated into computer-aided design systems and applied to the development of other microwave electronic devices.