Machine Learning-Aided Synthesis Approach for 3D Printed Ridge Gap Waveguides

In this paper, we propose a scalable machine learning (ML)-based inverse design approach for the automatic design of ridge gap waveguide (RGW) unit cells, targeting frequency bands and applications ranging from 3 to 300 GHz. This approach aims to improve reliability and reduce the high computational costs typically associated with the trial-and-error method or conventional optimization techniques in electromagnetic (EM) simulators, especially when the target frequency band changes. The proposed ML-based approach significantly reduces computation time, making predictions in less than 1 minute, compared to the 120, 72, and 96 hours required by the trial-anderror method, PSO, and GA optimizers, respectively. To validate this method, we designed a 2-port RGW structure operating within the 24-26 GHz range, tailored for the 5G n258 band (24-24.25 GHz) and midband applications (24.2-25.5 GHz). We conducted an experimental validation by fabricating the 2-port RGW structure with overall dimensions of 15 mm × 15 mm × 5.5 mm using laser powder bed fusion (LPBF) technology and measuring it with standard methods. The results demonstrate a measured return loss below-25 dB across the operating band and an average insertion loss of ≈ 1.5 dB, underscoring the effectiveness of the proposed ML-based design approach and the potential of 3D-printed technology for RGW-based component prototyping.