AEGS-HI: An adaptively Enhanced Filtering Algorithm for Removing FES Artifacts and M-waves in Surface EMG under Non-Integer Sampling Ratios

Closed-loop electromyogram-controlled functional electrical stimulation (FES) is hindered by unreliable voluntary EMG (vEMG) extraction under dynamically varying stimulation without hardware synchronization. Conventional methods excel under strict periodicity but suffer severe degradation, including transient spikes, phase drift, or structural failure, when assumptions are violated. We propose the Adaptive Enhanced Gram-Schmidt with Hybrid Interpolation (AEGS-HI), integrating numerically stable artifact suppression, dual-modal time-frequency residual detection, and context-aware adaptive inpainting. Across benchmarks spanning steady-state to intermittent variable-frequency stimulation (20-50 Hz, 8-14 mA), AEGS-HI demonstrates robust performance where competitors fail: under the most challenging paradigm, it achieves RMS_cc = 0.66, nearly doubling RLS (0.34). This dynamic robustness entails a principled trade-off: lower steady-state SNR than RLS reflects deliberate prioritization of stability over static optimality, while eliminating preonset distortions inherent to GS-series methods. In situ validation without clean vEMG reference confirms effective harmonic suppression, preserved physiological spectral content (20-400 Hz), and envelope dynamics consistent with voluntary activation (r_env = 0.718 ± 0.093). By providing a single-channel, software-only solution decoupling stimulation adaptability from signal-processing constraints, AEGS-HI removes a critical barrier to clinically viable closed-loop FES neuroprostheses.