EEG Plot

Each frame, EEG Plot runs a two-stage GPU pipeline and then draws text overlays.

1. Compute (min-max decimation, "chalkboard" mode). The kernel eeg_plot_minmax reduces the raw ring buffer to one min/max pair per output pixel column per channel. Output width is the viewport width, capped at 4096 columns; samples per column are visibleFrames / outputWidth, where visibleFrames = sampleRate × Time Window (clamped to the buffer's maxFrames). Crucially, a column is computed once, when the sweep passes over it, and never recomputed — old data stays exactly where it was drawn, which is what produces the chart-recorder look. Only columns between the previous and current sweep position are updated each frame.

2. DC offset (incremental rolling mean). When DC Filter is on, eeg_plot_dc_incremental maintains a per-channel running sum over the visible window, adding new samples and subtracting expired ones — O(delta) work per frame rather than O(window). The resulting mean is subtracted from every sample before decimation, so baseline drift cannot push traces out of their lane. NaN and Inf samples are skipped.

3. Render. One instanced 6-vertex quad per displayed channel; the fragment shader reads the min/max buffer, applies optional smoothing by blending neighbouring columns, interpolates the centerline and the min-max envelope with a Catmull-Rom spline, and evaluates a signed-distance line. Values are normalized as value / scale clamped to ±0.45 of the channel slot, so a signal at exactly the scale limit fills the lane. Output is premultiplied alpha; anti-aliasing uses fwidth() scaled by Grafit's global AA multiplier.

Two automatic visual cues are baked into the shader:

An eraser bar — a soft gap roughly 1.5% of the plot width — travels with the write head, separating the newest data from the oldest.

Coordinates and outputs

X/Y place the plot's center in Grafit coordinates (origin at viewport center, shorter viewport edge maps to [-1, 1]); Width/Height are the plot's extent. When labels are enabled the shader reserves a 12% left margin and an 8% right margin, so traces are drawn only in the middle band. Like every generator, EEG Plot renders to an offscreen MTLTexture that becomes visible only through a Canvas node, and setting Enable to false clears that texture.

Besides Image, the node emits a Values JSON output: the latest per-electrode value after the same notch-average and DC subtraction the display uses, converted to µV when in µV mode. Keys are the electrode labels supplied by the data source and are wrapped in ordered JSON so anatomical front-to-back order (Fp, F, C, P, T, O) is preserved.

Display modes (Settings panel)

Style, smoothing, filters (Settings panel)

The µV conversion is µV per count = (VREF × 1,000,000) / (2²³ × Gain), i.e. 4.5 V at 24x yields about 0.02235 µV per ADC count. The inspector shows the computed value live.

Dynamic scale port

The amplitude port renames itself with the unit mode: it is µV Scale (default 120.0) in µV mode and RAW Scale (default 500.0) in Raw ADC mode. Each mode's value is remembered separately, so switching back and forth never carries a raw-count scale into microvolts. Connections survive the rename.

Interactions worth knowing: Time Window and viewport width jointly set samplesPerPixel — a long window on a narrow viewport means more samples collapsed per column, so the min-max envelope grows thicker while the centerline stays smooth. Height divided by channel count sets the lane height in Stacked mode, and the minimum line thickness is expressed relative to that lane, so very short plots with many channels render proportionally chunkier traces. Label Size only matters when Show Labels is on; turning labels off also removes the margins, giving traces the full width.

• BrainBrick Device Input → EEG Plot → Canvas — the standard live 8-channel monitor with lanes, values, and time markers
• BrainBrick Device Input → EEG Plot with Overlay mode — superimpose all electrodes on one ruler to spot channels that diverge
• EEG Plot (Values) → JSON Array Get → Remap → Canvas position — drive visuals from the same filtered per-electrode value that the trace displays
• BrainBrick Device Input → FFT EEG → EEG Spectrum → Canvas, alongside an EEG Plot on a second Canvas layer — time domain and frequency domain side by side
• EEG Plot in Single mode with a large µV Scale on a projection layer — a single dramatic hero trace for a performance
• EEG Plot (Values) → Smooth → Array/Math chain — condition the live per-electrode readout before triggering events

• Nothing renders without a Canvas node on the Image output; Canvas also owns layer order and blend mode.
• Connect MTL Data straight from BrainBrick Device Input — the buffer is shared unified memory, so no per-frame copy happens on the CPU.
• If traces look flat, the scale is too large; if they clip red, it is too small. Red tinting starts at 90% of full scale by design.
• Keep DC Filter enabled for electrode recordings: without it, slow baseline wander drags traces out of their lanes within seconds.
• The notch filter and DC filter are display-only. Anything you record via the recording nodes still contains the untouched raw signal.
• A hard reset (clearing all buffers and redrawing) is triggered automatically when the data buffer pointer changes, when a cable is reconnected, or when the write head jumps more than a full sweep — so unplugging and replugging a device does not corrupt the sweep.
• Anti-aliasing follows Grafit's global AA multiplier; there is no per-node AA quality control.