Algorithm 2: Calculation of a coupling matrix in the source space.

Input: EEG source signals $S=[s_1(t),s_2(t),\cdots ,s_c(t)]\in {ℝ}^{n\times c}$ and measured fNIRS signals $Y=[y_1(t),y_2(t),\cdots ,y_d(t)]\in {ℝ}^{n\times d}$ Output: $C_S$ 1: for $i=1,\cdots ,c$, do 2:       Calculate the time-frequency power spectrum $P_i(t,f)$ for $s_i(t)$ by Equation (8). 3:       Calculate the normalized time-varying power $P_i(t)$ for $s_i(t)$ by Equation (9). 4:       Calculate the predicted fNIRS signal ${\tilde{y}}_i(t)$ by Equation (10). 5: end for 6: Create matrix $\tilde{Y}$ by utilizing all ${\tilde{y}}_i(t)$. 7: Segment $\tilde{Y}$ and $Y$ into $l$ samples; each sample contains ${\tilde{Y}}^{\prime }$ and $Y^{\prime }$. 8: for $i=1,\cdots ,l$, do 9:          Fit $Y^{\prime }$ within general linear model by Equation (13). 10:       Calculate the coupling matrix in the source space $C_S$ by Equation (14). 11: end for