Machine Learning-Based Design of an Ultrawideband Optically Transparent Microwave Metamaterial Absorber

An ultrawideband, optically transparent metamaterial absorber (MMA) was designed and fabricated, employing a convolutional neural network (CNN) in conjunction with a particle swarm optimization algorithm (PSO). The CNN enables a rapid prediction of reflection coefficients, thereby significantly simplifying the design process. PSO performs global optimization to fine-tune structural parameters based on CNN predictions, achieving configurations that meet specified absorption targets. The combination of CNN and PSO has formed an innovative reverse inverse process, which can achieve on-demand design of the MMA according to the desired absorption spectra. The designed MMA employs an optically transparent indium tin oxide (ITO) conductive film. With a thickness of 8.5 mm, the absorber achieves absorptivity exceeding 95% across the ultrawideband frequency range from 14.2 to 37.8 GHz, corresponding to a relative bandwidth of 90.8%, and exhibits excellent angular stability. Based on the simulation results, physical prototypes with dimensions of 250 × 250 mm2 were fabricated. Experimental measurements demonstrate strong agreement with the simulated reflection coefficients and confirm the broadband absorption performance. The proposed MMA exhibits both high optical transparency and efficient broadband electromagnetic absorption, making it highly suitable for applications in multispectral stealth technology and electromagnetic compatibility engineering.