To show an example of this effect, 2-min of synthetic MEG recordings, sampled at 512 Hz, were simulated by using the head model and MEG sensor layout of one subject taken from the Human Connectome Project dataset (Larson-Prior et al., 2013). The time series for two interacting dipolar sources in the cortex, with orientation perpendicular to the local cortical surface, were generated as follows: the time series of source 1 was sampled from a Gaussian distribution; the time series of source 2 was set to a Finite-Impulse-Response (FIR) filtered version of the time series of the first source; the FIR filter coefficients were randomly drawn from a standard normal distribution and a filter order P = 5 was used. The location of source 1 was kept fixed (black dot in panel A of Figure 1), while the location of source 2 was varied across all the remaining cortical locations. The source time series were then projected to the sensor level by solving the MEG forward problem (Nolte, 2003). Uncorrelated sensor noise was added to sensor signals. From the synthetic recordings,
