... Blinowska

Laboratory of Medical Physics, Institute of Experimental Physics, Warsaw University, ul. Hoza 69, 00-681 Warszawa, Poland, http://brain.fuw.edu.pl.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... (Hz)

For the angular frequency this constant would be equal to $\frac 1 4$

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... expansion

In practice this sum will be of course finite; in this study we used 2 different stopping criteria: for the Dataset I iterations were stopped when 99.9% of signal's energy was explained, which resulted in about 400-500 waveforms per trial. For the Dataset II number of iterations was fixed at 1300. These criterions were chosen to ensure proper repreentation of high frequency structures at a minimum computational cost (number of iterations. Details of the MP procedure can be found in [14]

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... stationarity.

Standard test for stationarity relies on dividing the questioned epoch into subepochs of the length exceeding the period of the lowest frequency present in the signal, and then applying a non-parametric test (e.g. sign test) to statistical descriptors of these subepochs [18]. Usual length of the reference epoch does not exceed seconds, so considreing the presence of low EEG frequencies (order of few Hz) we would have too few subepochs for a reasonable application of a low-power non-parametric test

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... $\char93 \{\textrm{ref}\}$

$\char93 \{\textrm{ref}\}$ is the time length of the reference period measured in chosen time widths of resels.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

....

For the brevity we omit the distinction between the exact value $\alpha$ which would be estimated from all the possible repetitions, and the actually calculated

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... faster.

In the pseudo-$t$ approach we estimate the distribution of the statistics for the null hypothesis only once for each frequency ( $\mathcal{O}$ $(N_\textrm{rep})$), and then the test for each resel is only $\mathcal{O}$$(1)$. The full resampling estimating distributions for each resel is $\mathcal{O}(N_\textrm{rep})$ $\times$ the number of investigated resels.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... Internet.

Playing with different settings on the two presented datasets involves only changing designated parameters in configuration files; adapting the system to a different dataset requires writing a new configuration file. One has to be also aware that the applied statistical procedures are computer-intensive, therefore in some cases computations can take up to hours.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.