Experimental Protocol

This dataset --created and contributed by Luca Citi, Riccardo Poli, and Caterina Cinel-- was generated as part of a study aimed at identifying  the factors limiting the performance of brain-computer interfaces (BCIs) based on event-related potentials (ERPs), in order to improve the transfer rate and the usability of these interfaces.
 
In this study, each participant was asked to spell a total of 20 characters using a traditional matrix speller (see speller.png). Each record in the dataset contains the signals, triggers and annotations corresponding to a single run, i.e. the spelling of a single character. Target characters were randomly chosen before the beginning of the run. Each row and column of the standard 6x6 matrix of characters was randomly intensified without replacement for 100 ms with a gap of 50 ms, leading to a stimulus onset asynchrony (SOA) of 150 ms. During a run, each row and column flashed approximately 20 times. During that period, subjects were asked to focus on the target character and to mentally count the number of times it was highlighted. From subject 03 onwards, subjects were asked to report their final count at the end of the run.

During the recordings, subjects were seated comfortably with their neck supported by a C-shaped inflatable travel pillow to reduce muscular artefacts. The eyes were at approximately 80 cm from a 22-inch LCD screen with 60 Hz refresh rate.
Data were collected using a BioSemi ActiveTwo EEG system sampled at 2048 Hz. The signals recorded comprise 64 EEG electrodes (locations can be downloaded from http://www.biosemi.com/download/Cap_coords_all.xls ), two earlobes (EARL and EARR) for referencing, and vertical and horizontal electrooculograms (VEOGL, VEOGR, HEOGL, and HEOGR) for artefacts removal.
The signals are referenced to the CMS electrode (see http://www.biosemi.com/faq/cms&drl.htm ). As this mode does not provide the full common mode rejection ratio (CMRR), the signals should be re-referenced via software (e.g., wrt the average of EARL and EARR) to achieve the full CMRR and improve the signal-to-noise ratio. 

Data format

The data are provided here in EDF+ format (containing the signals and an annotation channel). For use with PhysioToolkit software, rdedfann was used to generate a separate PhysioBank-compatible annotation file (with the suffix .event) for each record. The .event files and the annotation channels in the corresponding .edf files contain identical data. 
In each record, annotations have the following structure:

    0:01.588     3252 #TgtP_RC01_SOA63
    0:02.091     4282 #start
    0:05.535    11336 MNOPQR
    0:05.685    11643 STUVWX
...
    0:25.178    51565 FLRX4_
    0:30.122    61689 #counted22of22
    0:30.375    62207 #end

The first annotation reports the target character ("P" in the example), the name of the run (RC01) and information about the duration of the stimuli timing (stimuli are on for 6 refresh cycles, i.e. 100 ms, and off for 3, i.e. 50 ms).
The second annotation marks the beginning of the run while the last one its end. In the example, the penultimate annotation reports the number of target flashes counted by the subject and the actual number displayed.
All other annotations report the timing of each intensification and the set of characters (in a row or column) being intensified.


Acknowledgments

This dataset is described in:

    Citi L, Poli R, and Cinel C. Documenting, modelling and exploiting P300 amplitude changes due to variable target delays in Donchin's speller. Journal of Neural Engineering, vol. 7, p. 056006 [http://dx.doi.org/10.1088/1741-2560/7/5/056006]; 2010.

Please cite this publication and the standard citation for PhysioNet:

    Goldberger AL, Amaral LAN, Glass L, Hausdorff JM, Ivanov PCh, Mark RG, Mietus JE, Moody GB, Peng C-K, Stanley HE. PhysioBank, PhysioToolkit, and PhysioNet: Components of a New Research Resource for Complex Physiologic Signals. Circulation 101(23):e215-e220 [Circulation Electronic Pages; http://circ.ahajournals.org/cgi/content/full/101/23/e215]; 2000.