High-density surface EMG decomposition based on a convolutive blind source separation approach

A novel automatic approach is developed in the present study to decompose high density surface electromyography (EMG) signals into motor unit (MU) firing patterns. The observed surface EMG signals are first modeled as a convolutive mixture of active MU sources. Contrast function maximization is employed to extract the first source, and separation of other sources is then carried out by an iterative deflation approach. Each extracted source is further processed and verified with the characteristics of motor unit action potential and firing patterns. The performance of the proposed automatic approach is evaluated in well-designed computer simulation. Results show that 4.7±0.5 and 7.1±0.6 MUs were correctly identified in the case of 5 and 10 active MUs respectively.