↑↓ move↵ open[] previous / next lesson? all shortcuts
Appearance
Theme
Type size
Reading

Focus mode hides the sidebars. Press f to toggle it anywhere.

Keyboard shortcuts
/
Search
[ ]
Previous / next lesson
t
Jump to the table of contents
r
Review queue
f
Focus mode
?
This panel
ProjectOne afternoon to a demo, a week to a speller$0 beyond the amplifierTier 2

Project A: The SSVEP speller

Boxes flicker at different rates; the back of your head flickers with whichever one you look at. A decoder reads the frequency out of the EEG and types the letter. Works in an afternoon, and it is the BCI that actually works.

AssumesProject A: The ADS1299 boardLab Streaming Layer and PsychoPySpineDecoding / signal processing / ML
On this pageOne afternoon to a demo, a week to a speller
  1. The signal
  2. The decoder
  3. The stimulus
  4. The speller

Stare at a square flashing ten times a second and your visual cortex oscillates at ten hertz, strongly enough that a single at the back of the head shows a clean peak in the within two seconds. Put four squares on the screen at four different rates, decide which rate the contains, and you have a four-way selector. Arrange letters behind the squares and you have a speller. The is the most reliable non-invasive paradigm, it needs no training, and it works for nearly everyone with working eyes.

Simulated SSVEP decoderSimulated signal
This interactive needs JavaScript. If you are reading a printout, the caption describes what it shows.
Figure 1. The simulated version. Click a box to 'look at' it; the decoder fits sines and cosines at each candidate frequency to the last two seconds and picks the best. Optionally turn on real flicker to see what the stimuli look like.
Try this
  1. Click different boxes and watch how long the takes to follow. That lag is the window length.
  2. Lower the strength until the starts making mistakes. Real people vary this much.
  3. Note that the ’s score for the true frequency includes its second harmonic. Real has harmonics too.
Predict before you look

Your monitor refreshes at 60 Hz. Which set of flicker frequencies can it display exactly?

Only frequencies that divide 60 evenly, because the screen can only change on a refresh. 10 Hz is on for three frames and off for three. 8.57 Hz (60/7) is on for three or four frames alternately, which still works. 9 Hz cannot be shown steadily on a 60 Hz screen and the actual stimulus will smear across nearby frequencies. This is the first practical constraint every designer meets.

The signal

SSVEP sites
This interactive needs JavaScript. If you are reading a printout, the caption describes what it shows.
Figure 2. Over visual cortex: Oz strongest, O1 and O2 beside it. One electrode at Oz with an earlobe reference is enough for a two-target speller.

The response sits at the flicker frequency and its harmonics, over occipital (Oz strongest, O1 and O2 next). Amplitude is a few microvolts, which is small, but it is a pure tone, and pure tones are what spectra find best. Two to four seconds of data is enough to decide among four to eight frequencies for most people.

The decoder

is the standard. For each candidate frequency, build reference signals: sine and cosine at f, at 2f, and at 3f. Find the combination of channels that correlates most with the combination of those references. The frequency with the highest correlation wins. With one channel it reduces to fitting the references by least squares and comparing the fit, which is what the simulator does. With several occipital channels the multichannel version is meaningfully better.

import numpy as np
from sklearn.cross_decomposition import CCA

def refs(f, fs, n, harmonics=3):
    t = np.arange(n) / fs
    return np.column_stack([g(2*np.pi*h*f*t) for h in range(1, harmonics+1) for g in (np.sin, np.cos)])

def decode(X, freqs, fs):           # X: samples × channels, a 2–4 s window
    scores = []
    for f in freqs:
        cca = CCA(n_components=1).fit(X, refs(f, fs, len(X)))
        u, v = cca.transform(X, refs(f, fs, len(X)))
        scores.append(np.corrcoef(u[:, 0], v[:, 0])[0, 1])
    return freqs[int(np.argmax(scores))], scores

The stimulus

, full screen, black background, four to eight white squares each toggling on a fixed frame schedule, with a letter or word under each. Push an marker whenever the layout changes. Frame-locked toggling is the whole trick; do it inside the flip loop, never with timers.

The speller

A four-target selector spells slowly. The classic layout is a tree: four squares each holding a group of letters; look at a group to select it, then the letters in that group spread across the four squares, then select the letter. Two selections per character, about eight seconds each with a comfortable window: a character every fifteen to twenty seconds. Slow, and it works on the first try for most people, which no other paradigm can say.

Measure and report two numbers: accuracy (fraction of selections correct) and in bits per minute, which combines accuracy, number of targets, and speed into one figure that lets you compare against the literature. The classification explainer has the formula.

  1. Set up the pipeline from Phase 2: stream, marker stream, LabRecorder.
  2. Write the stimulus with four frequencies from the divisors of your refresh rate (6, 7.5, 10, 12 Hz is a good set). Verify the flicker rates with a photodiode on a spare channel; the of the photodiode channel should show a sharp line at each frequency.
  3. Record yourself looking at each square for ten seconds in turn. Plot the Oz for each. The peak should move.
  4. Implement the and run it offline on the recording with two-second windows. Report accuracy per window.
  5. Close the loop: decode live, highlight the winning square. Tune the window length for the best speed-accuracy trade for you.
  6. Build the two-level letter tree. Spell your name. Time it. Compute accuracy and ITR.
  7. Test on two friends (with the photosensitivity screening and under your lab’s protocol). Note how much the amplitude varies between people.
Recall
Why is SSVEP the most reliable non-invasive BCI paradigm?
The response is a pure tone at a known frequency in occipital EEG, which spectral methods detect very well within seconds, it needs no user training, and it works for nearly everyone with working vision.
Recall
Why must flicker frequencies divide the monitor's refresh rate?
The screen can only change state on a refresh, so only frequencies that fit whole frames are displayed steadily; others smear across nearby frequencies.
Recall
What safety screening does an SSVEP system require and why?
Ask about personal or family history of seizures or reactions to flashing light, since flicker at 3 to 60 Hz can trigger photosensitive epilepsy in about one in four thousand people; exclude anyone who answers yes and stop on any discomfort.