Figure 3.

This figure summarizes various aspects of the proposed methodology. The montage defined by the m vector is illustrated by the leftward band displaying binary weights including frontal, parietal, and central channels and excluding the other channels from the mixture. Below the band, a scalp map shows the topological coverage of this montage (red dots for included electrodes, pink for excluded). At the right of the montage band, all available channels are stack in a matrix S. The mixture signal s^(m) is obtained by matrix multiplication of m and S as described in (1). Applying the CWT to s^(m) and taking the modulus, we obtain a time-frequency map |W^(m) as shown. Using (4) on |W^(m), we compute the d(t_j) detection function used to determine the times t^start, t^max, and t^end. $\tilde{x}$^ampn is obtained as d(t^max_n) and f^max_n as f_i with $i=\underset{\begin{array}{c}1\le i\le N_f\\ t_j=t_n^{max}\end{array}}{\text{argmax}}\left|w_{i,j}^{(m)}\right|$.