Terahertz group-delay readout based on a graphene-programmable spoof surface plasmon polariton frequency-domain kernel for refractive-index classification

Conventional terahertz biosensing schemes mainly rely on resonance frequency or phase-shift readouts, which often suffer from limited stability toward phase noise, weak dispersion exploitation and reduced interpretability in classification-oriented sensing. To address this limitation, a terahertz group-delay readout framework is proposed based on a graphene-programmable spoof surface plasmon polariton (SSPP) platform for binary classification of biological refractive indices. By integrating a graphene tuning layer into an SSPP waveguide, the frequency-domain response becomes electrically reconfigurable, enabling dispersion-engineered sensing. The differential group delay Δτ is adopted as the primary observable to capture band-localized, sign-definite and dispersion-slope variations induced by refractive-index perturbations. Through sub-band aggregation, the continuous spectral response is compressed into a low-dimensional, physically interpretable feature vector for linear discriminant classification. Compared with phase-difference readouts, the proposed Δτ-based scheme exhibits strong dispersion-sensitive discriminative capability and maintains usable performance under controlled measurement perturbations, while practical phase-noise suppression and reference calibration remain important for stable implementation. By structurally embedding dispersion control and group-delay feature extraction into a programmable SSPP architecture, the proposed design directly addresses the stability and feature-utilization limitations of existing terahertz readout methods. These results show that dispersion-structured frequency-domain features provide a stable and interpretable route toward programmable terahertz label-free biosensing.