Analyze Steady-State Visual Evoked Potentials (SSVEPs)

Steady-state stimulation is frequently used for sensory stimulation in the visual (SSVEP), auditory (SSAEP), and somatosensory (SSSEP) domains. On this page, we will first present an example analysis strategy for a 64-channel SSVEP dataset. Subsequently, we will make some simulated data and use another analysis strategy.

Analyze experimental data

The design of a trial is shown in the figure below. On each trial, a ring is flashing at 8.33 Hz for 4.56 s (at 100 Hz, on for 6 frames and off for 6 frames). During this time, a 100-ms rectangle is shown about every 1 s for four times. In a block of 40 trials, the task is either low load (i.e., count color, e.g., black) or high load (i.e., count a certain combination, e.g., black horizontal rectangle and white vertical rectangle). After each trial, subjects report if they counted 2 or 3 targets. Within each block, the ring is shown at four eccentricities: 2, 3, 4, and 6 visual degrees (10 trials at each excentricity), in pseudorandom order for each subject. Load levels alternate between blocks (i.e., LHLHLHLH or HLHLHLHL , counterbalanced across subjects).

There are 2 loads x 4 blocks x 4 ring eccentricities x 10 trials = 320 SSVEP trials. Because 4 rectangles are shown within each trial, this gives 4 x 320 = 1280 rectangle trials. This gives a total of 320 + 1280 = 1600 trials.

In the analysis, consider whether the frequency tagged (or steady state ) stimulus is

1. phase consistent over trials. If so, average and then do wavelet/mtmconvol/mtmfft, or time-domain regression. (In the example, each 4.56-s SSVEP trial has the same phase.)
2. not phase consistent across trials, but the phase of the stimulus is known. If so, Fourier decompose to get the complex representation, deal with single-trial phase differences, then average. (In the example, the onset of the rectangles is jittered relative to the onset of a SSVEP trial. Thus, the phase of the flashing rings varies between rectangle trials).
3. not phase consistent across trials, and the phase of the stimulus is not known. If so, Fourier decompose to get the power and average over trials.

Furthermore, consider whether the cortical response

1. is assumed to be constant within the trial
2. changes over time within the trial

Finally, consider whether the stimulation contains

1. a single frequency, as in a traditional SSVEP
2. a mixture of multiple frequencies, as in frequency tagging

Using time-domain analysis

Using frequency analysis

Create and analyze a simulated steady-state dataset

fsample = 1000;
nsample = 100*fsample;

% start with a continuous representation of 100 seconds of data
data.label = {'trigger', 'eeg'};
data.time = {(1:nsample)/fsample};
data.trial = {zeros(length(data.label), nsample)};

% add a trigger every 100ms
i = 100;
while i<nsample
  data.trial{1}(1,i) = 1;
  i = i + 100;
end

% create some sort of SSVEP signal
data.trial{1}(2,:) = ft_preproc_bandpassfilter(data.trial{1}(1,:), fsample, [3 18], [], [], 'onepass');
data.trial{1}(2,:) = data.trial{1}(2,:) + 0.02*randn(size(data.trial{1}(2,:)));

% cut into one-second snippets
cfg = [];
cfg.length = 1;
data = ft_redefinetrial(cfg, data);

plot(data.time{1}, data.trial{1})
legend(data.label)

cfg = [];
cfg.method = 'mtmfft';
cfg.taper = 'hanning';
cfg.output = 'powandcsd';
cfg.foilim = [1 100];
freq = ft_freqanalysis(cfg, data);

plot(freq.freq, freq.powspctrm);
legend(freq.label)

% normalize the CSD for the power in the trigger sequence
freq.crsspctrm = freq.crsspctrm ./ freq.powspctrm(1,:);

plot(freq.freq, abs(freq.crsspctrm));
xlabel('frequency (Hz)')
ylabel('phase-locked amplitude (a.u.)')