Signal quality assessment of surface electromyography envelope extraction across multiple noise conditions

Abstract

Envelope extraction is a fundamental preprocessing step in surface electromyography (sEMG) for myoelectric control, yet its signal quality under varying noise conditions remains insufficiently characterized using classification-independent metrics. This study assessed the signal quality of seven envelope extraction configurations, spanning instantaneous, static low-pass, and adaptive Kalman-based smoothing, using a dual-metric framework combining intrinsic signal-to-noise ratio (SNR) and power-reference correlation (p). Twenty subjects from the Ninapro DB2 dataset were examined under clean conditions and four noise types (power-line interference (PLI), motion artifact (MA), white Gaussian noise (WGN), and their combination), all scaled to 10 dB SNR. Static low-pass filtering achieved the highest intrinsic SNR (up to 4.4 dB), whereas the instantaneous tier attained the highest fidelity (p up to 0.985) but strongly negative SNR (-6 to -13 dB); the adaptive Kalman tier was intermediate (1.5-1.9 dB, p=0.87). Within the adaptive tier, a Hilbert-based front-end outperformed rectification under PLI (+0.43 dB), MA (+0.29 dB), and combined noise (+0.22 dB; all p<0.0083), but not under WGN (-0.05 dB, p=0.070). These findings indicate that the benefit of phase-coherent estimation depends on the deterministic structure of interference rather than its magnitude.