Machine learning-assisted design and optimization of a graphene-based tunable bandpass filter for millimeter-wave applications

Abstract With the advancement of communication technologies, high-performance and tunable microstrip bandpass filters in the W-band (75-110 GHz) are increasingly vital for millimeter-wave systems. However, their development faces key challenges, including complex multi-parameter dependencies that hinder performance understanding, the lack of systematic analysis on graphene integration strategies, and uncertainties in achieving optimal tunability. This study employs machine learning tools to investigate how structural and coupling parameters influence core filter characteristics such as bandwidth, insertion loss, and out-ofband rejection. Based on the data-driven insights, several graphene-based integration schemes are evaluated to realize voltage-controlled amplitude modulation. Among them, a resonator-coupled configuration achieves a modulation depth of 15.10 dB at 2.5 eV, while an extended variant reaches 10.75 dB at just 0.6 eV. These results demonstrate a quantifiable trade-off between modulation depth and driving power, providing a practical reference for designing energy-efficient, reconfigurable mmWave front-end components.