Data-driven modeling of band-pass filter for sub-5G applications

Abstract

Radiofrequency noise is one of the challenging problems in the design of high-performance wireless communication systems, for which microstrip band-pass filters are one of the most commonly used design solutions for this challenge. However, for having high-performance designs. usage of a 3D Electromagnetic simulation tool is a must in which the computation efficiency of the whole design optimization process might not be acceptable or even feasible. An efficient solution is utilization of AI-based algorithms to create data-driven surrogate models of the handled problem. In this paper, for achieving design optimization of an edge-coupled band-pass filter AI-based algorithms have been used to create a data-driven surrogate model. To achieve this, by using a 3D full-wave simulator a data set for the aimed bandpass filter is generated. Then a series of state-of-the-art regression algorithms, Support Vector Regression Machine, MultiLayer Perceptron, Ensemble Learning, Gaussian Process Regression, and Convolutional Neural Network have been used to create a data-driven surrogate model for the aimed filter design. In the third step, the obtained data-driven surrogate model is used to assist an optimization process directed by the Bayesian optimization technique to optimally determine geometrical design parameters of the desired band-pass filter for sub-5G applications at frequency of 3.4 GHz. The obtained results of the surrogate model are compared with experimental results and found to be in high agreement level. Furthermore, the performance of the optimally designed filter is compared with the counterpart designs in literature. Thus, based on the obtained results, it can be said that the proposed surrogate-assisted optimization process is not only an efficient method in terms of computational costs but also is an efficient method to obtain high-performance microwave filter designs.