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

In this paper, a direct machine-learning (ML)-assisted inverse-synthesis approach is proposed for ridge gap waveguide (RGW) unit-cell design. The proposed approach maps the prescribed lower and upper stopband frequencies directly to the main geometrical parameters of the RGW unit cell, namely (a), (p), and (w), thereby reducing the need for repeated parametric sweeps or geometry-recovery optimization when the target frequency band changes. After the initial EM-database generation and model training, the proposed approach predicts the unit-cell dimensions in less than 1 min, compared with approximately 120, 72, and 96 h required by conventional trial-and-test, particle swarm optimization (PSO), and genetic algorithm (GA)-based procedures, respectively. The proposed approach can be constructed for target bands within 3–300 GHz, provided that a suitable EM-generated database is established for the selected design domain. Its broader applicability is supported through computational prediction and full-wave electromagnetic (EM) verification at representative target bands, while experimental validation is carried out using a two-port RGW structure operating from 24 to 26 GHz. The prototype, with overall dimensions of (15mm × 15mm × 5.5mm), is fabricated using laser powder-bed fusion (LPBF) technology and characterized through vector network analyzer measurements. The measured results show good agreement with the predicted and simulated responses, with return loss above 25 dB across the operating band and an average insertion loss of approximately 1.5 dB. These results demonstrate the effectiveness of the proposed inverse-synthesis approach for accelerating RGW unit-cell design and highlight the potential of metallic additive manufacturing for compact RGW-based component prototyping.