Blind source separation of multiplicative mixtures of non- stationary surface EMG signals

Electromyographic (EMG) signals detected over the skin are mixtures of signals generated by many active muscles due to the phenomena related to volume conduction. Separation of the sources is necessary when single muscle activity has to be detected. Signals generated by different muscles may be considered uncorrelated but have a largely overlapping bandwidth. When many muscles are active, no a priori information is available about the mixing matrix. Under certain assumptions, mixtures of surface EMG signals can be considered multiplicative. In this study we apply blind source separation (BSS) methods to separate the signals generated by two active muscles. An algorithm based on cross time-frequency representations will be used on simulated and experimental non-stationary EMG signals. The experimental signals were collected from muscles which could be activated selectively. The contractions performed by the subjects allowed objective validation of the methods. From the simulated signals, optimal performance was obtained. Correlation coefficients between the reference and reconstructed sources were higher than 0.9 even for sources whose spectral and temporal support largely overlapped. In the experimental case, in the reconstructed source the contribution of the other source was significantly decreased after the application of the BSS methods. The ratio between root mean square (RMS) values of the signals from the two sources increased from (mean ± standard deviation) 2.33 ± 1.04 to 4.51 ± 1.37 and from 1.55 ± 0.46 to 2.72 ± 0.65 for wrist flexion and rotation, respectively. This increment was statistically significant. It was concluded that BSS approaches are promising for the separation of surface EMG signals, with applications which go from the muscle assessment, detection of muscle activation intervals, and prosthetic control.