Fig 1.
Effects of stationary versus non-stationary noise on signal detection.
A) Signal detection with stationary noise, B) signal detection with non-stationary noise.
Fig 2.
Flow diagram illustrating the calculation of the SNR and SNR-wall.
A) Time domain signal for the SNR calculation takes the power Tt(c) of the P300 peak as the consciously generated EEG change. B) Frequency domain signal power for the SNR calculation takes the power Tt(c) and assumes a conscious 40% reduction of the signal power in a narrow band around a peak. C) For the overall SNR the nominal variance is calculated over the whole EEG recording. To calculate the SNR-wall the maximum
variance and minimum
variance is detected with a sliding window sample by sample. D) If the SNR is greater than the SNR-wall then a conscious change in the EEG can be detected. If the SNR is less than the SNR-wall then it is impossible to detect a conscious change.
Fig 3.
Complete dataset of a representative recording (subject 20).
A) Time domain of the different recordings (d[n] in μV against s). B) Power spectra of d[n] of the corresponding recordings (V2/Hz). C) Raw P300 recording (μV against s). D) Averaged P300 response (μV against ms).
Fig 4.
SNR and SNR-wall calculations for the “jaw clench” task.
Columns A-M: electrode signals in the time domain with lines for the maximum variance and minimum variance
while using a sliding window of 2 sec. Columns B-N: frequency power spectra of the corresponding time domain signals of columns A-M. Columns C-O: SNR-wall and SNR calculations of the corresponding time domain signals of columns A-M.
Fig 5.
SNR and SNR-wall calculations for the “reading” task.
Columns A-M: electrode signals in the time domain with lines for the maximum variance and minimum variance
while using a sliding window of 2 sec. Columns B-N: frequency power spectra of the corresponding time domain signals of columns A-M. Columns C-O: SNR-wall and SNR calculations of the corresponding time domain signals of columns A-M.
Fig 6.
SNR versus SNR-wall for A) full range electrode signal with only highpass filtering at 0.1 Hz and 50 Hz removal, B) 0.1–3 Hz bandpass filtered, C) 8–18 Hz bandpass filtered, D) narrow bandpass 10±2 Hz and E) derivative. The bars show the average for the SNR-wall and the SNR with error bars for standard deviation. A t-test was used to determine if conscious EEG changes can be detected at p < 0.05 and the “*” identifies the experimental and filtering conditions where this is significantly possible.
Table 1.
Detection of conscious EEG changes significantly possible ordered by task.