Sensing characterization of graphene broadband terahertz absorber with E-connected structure

We propose two thin, flexible, broadband, tunable terahertz (THz) metamaterial absorbers, consisting of a patterned graphene layer, a dielectric layer, and a metal substrate laminated. The time-domain finite integration method was used to analyze the absorber characteristics. The simulation results showed that the one-dimensional mirror E-type bridge link structure absorber achieved more than 90% near-perfect absorption at 2.8 to 4.15 THz. On this basis, a polarization-insensitive two-dimensional mirror E-type orthogonal structure absorber was designed. This achieved more than 90% near-perfect absorption at 2.15 to 3.26 THz, and the bandwidth with an absorption rate of A≥95% reached 1.01 THz. The polarization angle of the incident wave varied from 0 to 80 deg, the incident angle of the electromagnetic wave varied from 0 to 20 deg in the transverse electric mode and from 0 to 60 deg in the transverse magnetic mode, and the tangent of the loss angle of the dielectric layer varied from 0.003 to 0.011. The absorption performance and operating bandwidth of the absorber were stable. The adjustment of the graphene chemical potential achieved a stable tuning within 95.4% to 99.8% absorbance. The elevation of the graphene relaxation time led to a decrease in the bandwidth for absorbance of A≥90%, but two distinct absorption peaks appeared at 1.26 and 4.13 THz. Two metrics, sensitivity and dielectric constant prediction accuracy, were used to evaluate the performance of the absorber application, and a sensitivity value of 3.26% was calculated. When the analyte with the dielectric constant of 2 to 11 was applied, the absorption curve was red-shifted. Moreover, three resonance frequencies and a relationship model between the resonance frequency absorption value and the dielectric constant were established using the back propagation (BP) neural network. The accurate prediction of the dielectric constant of the unknown analyte was successfully realized, with an average prediction accuracy of up to 98.49%.