SSVEP Frequency-Tagged BCI Speller

Attended Target

Signal Parameters

Attention gain (SSVEP amplitude) 0.70

EEG noise level 0.40

Analysis window 2.0 s

Decoder Output

Decoded target: —
Confidence: 0%

Correct decodes

0 / 0

Decode rate

0.5 Hz

How it works

An SSVEP speller flickers each on-screen target at its own fixed frequency f_k. When the user's gaze/attention locks onto target k, the visual cortex's occipital EEG (electrode Oz) starts oscillating at f_k and its harmonics — the "steady-state visual evoked potential":

EEG(t) = G·[sin(2πf_k t) + 0.35·sin(2·2πf_k t)] + Σ 0.06·sin(2πf_j t) + N(t)·σ_noise
  G      = attention gain (signal strength)
  σ_noise = EEG noise level
  N(t)   = white noise sample

The decoder never sees which target was attended — it only has the single simulated Oz channel. Over a sliding window of W seconds it runs the Goertzel algorithm (a single-frequency DFT bin, far cheaper than a full FFT) at each of the 4 stimulus frequencies:

for each candidate frequency f:
  s[n] = x[n] + 2cos(2πf/fs)·s[n-1] − s[n-2]
  power(f) = s[N-1]² + s[N-2]² − 2cos(2πf/fs)·s[N-1]·s[N-2]

The frequency with the highest Goertzel power is the decoded command — the same principle real SSVEP spellers use (in practice via canonical correlation analysis, CCA, which is more robust but mathematically related). A longer window averages out noise and raises accuracy, but a BCI speller needs a decode roughly every 1-2 seconds to be usable — the Analysis window slider makes that speed/accuracy trade-off explicit.

Flicker target (attended = bright ring) Simulated Oz EEG trace

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.