Generalization of a wavelet-based algorithm to adaptively detect activation intervals in weak and noisy myoelectric signals

This study introduces an adaptive implementation of a Continuous Wavelet Transform (CWT) decomposition technique used to estimate the timing of muscular activation in weak and noisy myoelectric signals. The algorithm updates automatically the threshold based on the statistical properties of the EMG data, through an iterative estimation of the Signal-to-Noise Ratio (SNR). Moreover, it includes a stopping criterion for the number of CWT decomposition levels, and this allows a relevant decrease of the computational burden. This algorithm was applied to both synthetic and semi-synthetic signals, and compared against the original formulation of the CWT-based technique and a common threshold-based technique for the detection of muscle activations. Performance of these techniques was assessed by using Bias, Relative Timing Error and Accuracy of the detection. Bias values resulted lower than 18 ms, Relative Timing Error lower than 5% and Accuracy greater than 97% for all the tested SNR (ranging from −2 dB to 10 dB), and with a substantial independence from SNR levels. The performance was shown to hold also if the hypothesis of absence of muscular activation in the reference window cannot be guaranteed. The results show that the proposed approach, which is adaptive, operator-independent and iterative, performs properly when applied to weak and noisy myoelectric signals, and is thus a valid general solution when dealing with clinical conditions where muscular activity is low, and when recording conditions cannot be entirely controlled.