Privacy-preserving gait biometrics via synchronous mechanical and bioelectrical co-sensing and ciphertext-domain inference

Gait information is rooted in muscle contraction dynamics and neuromuscular regulation, and represents a reliable biometric. However, mechanical and bioelectrical signals acquired using conventional wearables are prone to replication and forgery during storage and transmission. Moreover, the degradation of recognition performance under standard encryption can occur due to limited feature dimensionality and fidelity. In this study, we present a spatiotemporally synchronous strain-surface electromyography (sEMG) co-sensing system (S-SESS), which co-locates and phase-aligns kinematic and neuromuscular signals from the gastrocnemius to deliver secure gait recognition. The strain sensing unit achieves a low hysteresis (1.6% at 100% strain) and high cyclic fidelity (> 10,000 cycles) using hierarchical supramolecular aggregates with multiple hydrogen bonds and dynamic π-π stacking that ensure precise phase tracking. The adaptive sEMG unit exhibits high interfacial conformity, suppresses motion artifacts, and achieves a signal-to-noise ratio (SNR) greater than 40 dB. The bimodal features are locally encrypted using the Advanced Encryption Standard-128 (AES-128) algorithm and processed directly in the ciphertext domain via a dedicated ByteEmbedded Temporal Convolutional Network (ByteEmbedded-TCN). This directly maps ciphertext streams into byte-embedded sequences and leverages causal and dilated convolutions to construct multi-scale receptive fields for temporal alignment and classification. The end-to-end framework achieves 89.7% ± 1.2% gait identification accuracy under encryption. The findings of this study provide a privacy-preserving wearable paradigm for trustworthy identity authentication.