NEURAL NETWORK BASED FILTER MODELING AND OPTIMIZATION FOR 5G AND BEYOND APPLICATIONS

Abstract

Designing high-performance microwave and millimeter-wave filters is difficult because small changes in geometric dimensions and electrical sizes can significantly affect the filter’s characteristic. Typically, in filter design, the initial values of design variables are optimized to achieve the desired performance. In the field of high-frequency RF device modeling, the use of machine learning (ML) through artificial neural networks (ANN) has gained popularity in recent years. Unlike other RF modeling techniques, ANN-based models require training with sufficient datasets to achieve the desired accuracy level. The input data could be the device’s dimensions, while the output could be the S-parameters. Once trained, the ANN-based model can provide EM-level accuracy and equivalent-circuit-level speed. Additionally, it is highly scalable, allowing for the introduction of more input parameters to make the model more versatile and complex. Therefore, the ANN-based model is an excellent option for high-frequency RF modeling compared to other techniques. The main objective of this research project is to develop an AAN that can be used in design of RF Filters.