Reflectarray Antenna Design Using the Deep Learning ControlNet Diffusion Model

In this paper, we apply the ControlNet diffusion model, a generative deep learning framework that integrates both text and image inputs for enhanced conditional control, to design reflectarray antennas. This generative deep learning-enabled method facilitates the design of pixelated unit cells with a high degree of freedom, enabling the generation of diverse patterns that satisfy the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$S$</tex> -parameters requirements for reflectarray antennas. The generated unit cells cover a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\sim 2\pi$</tex> phase range across the 9.0 GHz to 11.0 GHz. A single-layer reflectarray antenna with dimensions of 300 mm <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\times$</tex> 300 mm and a focal length of 180 mm was designed, achieving a maximum gain of 28.0 dBi at 10 GHz, an aperture efficiency of 50.3% and 1-dB bandwidth of 7.5%. The co-polarized antenna pattern has sidelobe levels of 22.2 dB and 19.4 dB in the E-plane and H-plane respectively.