Asynchronous non-invasive high-speed BCI speller with robust non-control state detection

<b>1. General description</b>Data was recorded using BCI2000 with g.USBamp (g.tec, Austria) EEG amplifier. 32 electrodes were used. Sampling rate was set to 600 Hz and data was bandpass filtered by the amplifier between 0.1 Hz and 60 Hz using a Chebyshev filter of order 8 and notch-filtered at 50 Hz. Data was stored as MATLAB mat-File, whereas each file is the data of one participant.<br><br><b>2. Experimental description</b>The experiment was split in several parts using two different keyboard layouts. Each target was modulated with random stimulation patterns, which were presented with 60 bits per second.<br>32 target matrix-keyboard layout (all runs consits of 32 trials in lexicographic order):- 1 spatial filter run- 3 model training runs- 1 threshold optimization run- 6 asynchronous spelling runs- 4 asynchronous spelling runs with 30 s of non-control state at the beginning and end<br>55 target German QWERTZ-keyboard layout:<br>- 3 copy-spelling runs with the task to write "Asynchron BCI" (case-sensitive)<b><br></b><b>3. Variable description</b>Each file VP*.mat contains the following variables:- spfilter_data_xcontains the raw EEG data of the spatial filter runs split by each m-sequence cyle. Data was recorded for 32 trials with 3 m-sequence cycles each, which results in 32*3=96 m-sequence cycles in total. Since the m-sequence was shifted by 2 bits for each succesive target, the trials were lag-fixed previously. As the m-sequence consists of 63 bits, the data has a length of 63/60*600=630 samples. Therefore the matrix has the following dimension: #m-sequences X #channels X #samples- spfilter_data_ycontains the stimulation pattern for each m-sequenze cycle upsampled to be synchronized with the EEG data. The matrix has the following dimension: #m-sequences X #samples- train_data_xcontains the raw EEG data of the training runs split by trials. The matrix has the following dimension: #trials X #channels X #samples- train_data_ycontains the stimulation pattern for each target, upsampled to be synchronized with the EEG data. The matrix has the following dimension: #trials X #targets X #samples- test_async_data_xcontains the raw EEG data of the asynchronous spelling runs split by trials. The trial duration varies due to the asynchronous control. The variable contains one entry for each trial of the following dimension: #channels X #samples- test_async_data_ycontains the stimulation pattern for each target, upsampled to be synchronized with the EEG data. The variable contains one entry for each trial of the following dimension: #targets X #samples- test_nc_data_xcontains the raw EEG data of the asynchronous spelling runs including non-control test, split by trials. The trial duration varies due to the asynchronous control. Furthermore, classifications occured during the non-control state for VP05 and VP07, which is why the data contains 1 and 2 additional trials, respectively. The variable contains one entry for each trial of the following dimension: #channels X #samples- test_nc_data_ycontains the stimulation pattern for each target, upsampled to be synchronized with the EEG data. The variable contains one entry for each trial of the following dimension: #targets X #samples- test_qwertz_data_xcontains the raw EEG data of the asynchronous copy-spelling runs using the QWERTZ-layout, split by trials. The trial duration as well as the number of trials vary, as the subjects were tasked to spell case-sensitve and to correct erros. The variable contains one entry for each trial of the following dimension: #channels X #samples- test_qwertz_data_ycontains the stimulation pattern for each target, upsampled to be synchronized with the EEG data. The variable contains one entry for each trial of the following dimension: #targets X #samples- pValThreshold<br>contains the threshold determined during the online experiment- maxTrialTime<br>contains the upper sub-trial duration in samples, which was determined during the online experiment.<br>There are two additional files "layout_matrix.mat" and "layout_qwertz.mat". Each contains the number of samples for each target used to correct the raster latency as wll as the labels of each target.