Benchmarking sEMG Decomposition: A Comprehensive Assessment of KmCKC and SL Across Synchrony, Duration, MVC, and Firing Rate

Purpose: We evaluated two high-performing surface electromyography (sEMG) decomposition methods—k-means convolution kernel compensation (KmCKC) and SVD plus LMMSE (SL)—under challenging conditions, including high motor unit (MU) firing synchrony, short signal durations, elevated firing rates, and varying maximum voluntary contraction (MVC) levels. These findings provide valuable insights for researchers and clinicians. Methods: Four categories of sEMG signals were analyzed using both simulated and experimental data: signals with varying MU firing synchrony (20%, 40%, 60%), short-duration signals (0.3–10 s), signals with different average MU firing rates (up to 50 Hz), and signals acquired at different MVC levels (10%, 30%, 50%). Decomposition performance was evaluated based on the number of identified MUs, precision, and pulse-to-noise ratio (PNR). Results: SL consistently outperformed KmCKC across all conditions. At 20% MU firing synchrony, SL identified 11.3 MUs versus KmCKC’s 5.7. For short-duration signals (< 1.3 s), KmCKC failed to identify any MUs, while SL detected 3.3–6.3 MUs. At a 40 Hz firing rate, SL decomposed 12.7 MUs compared to KmCKC’s 3. Across different MVC levels, SL maintained superior performance with precision exceeding 98.7%. Conclusion: The SL method demonstrates superior decomposition performance compared to KmCKC, particularly for complex sEMG signals. SL is the preferred method for short-duration signals and those with high MU firing synchrony and firing rates.