Zig Syrup Bci

zig-syrup-bci

Multimodal brain-computer interface pipeline for zig-syrup. Parses, processes, and classifies signals from EEG, fNIRS, eye tracking, and body pose modalities with GF(3) trit conservation.

Modules

GF(3) Conservation

eeg(0) + fnirs(+1) + eye(-1) = 0 mod 3 ✓

Verified across module boundaries in bci_integration_test.zig (16 tests).

Quick Start

# Run all BCI tests
zig build test-bci

# With real PhysioNet data (downloads 1.2MB EDF)
curl -sL -o testdata/S001R01.edf \
  "https://physionet.org/files/eegmmidb/1.0.0/S001/S001R01.edf"
zig build test-bci

Test Fixtures

Key APIs

DSI-24 Parser

const sample = try dsi24.parseDSI24Packet(&packet_84bytes);
// sample.eeg_channels[0..21] — µV values
// sample.aux_channels[0..3]
// sample.sample_counter, .timestamp_us

fNIRS Modified Beer-Lambert

const config = fnirs.WavelengthPair.plux(); // 660/860nm, DPF 6.51/5.60
const hemo = fnirs.beerLambert(delta_od1, delta_od2, config);
// hemo.hbo, .hbr, .hbt — µmol/L concentration changes
const reading = fnirs.FNIRSReading.fromConcentration(hemo, timestamp_ms, threshold);
// reading.trit — .plus (activation), .zero (baseline), .minus (deactivation)

Eye Tracking IVT

const result = eye.classifyIVT(current_gaze, prev_gaze, .{});
// result.event — .fixation, .saccade, .blink
// result.velocity — degrees/second
// result.event.toTrit() — .zero (fixation), .plus (saccade), .minus (blink)

EDF Reader

const edf = try edf_reader.EDFFile.parse(file_bytes);
// edf.n_channels, .n_records, .record_duration
// edf.channels[i].labelStr(), .unitStr(), .samples_per_record
const digital = try edf.getSample(record, channel, sample_idx);
const physical_uv = edf.toPhysical(channel, digital);

ERC (Ensemble Reservoir Computing)

var reservoir = erc.Cyton.init(.entropy_weighted);
const result = reservoir.processFromBandPowers(all_bands);
// result.trit, .confidence, .logits[3], .ensemble_entropy

// Online adaptation (NLMS — learning rate independent of feature scale)
const config = erc.LearningConfig{ .learning_rate = 0.5 };
const mse = reservoir.adaptFromBandPowers(all_bands, .plus, config);
// mse → convergence monitor; weights adapt to real EEG data

// Propagator integration
const cv = reservoir.toCellValue(); // → CellValue(f32)

LSL StreamSynchronizer

var sync = lsl.StreamSynchronizer.init();
const eeg_id = try sync.addStream(.{ .stream_type = .eeg, .nominal_rate = 300.0, ... });
// StreamType.trit(): eeg→0, fnirs→+1, eye_tracking→-1

SDF Verification (Ch1-Ch8)

Additive Design: ✓

New modalities added without modifying existing modules. Each sensor is a SensorConfig struct registered in UniversalReceiver.init().

Abstraction Barriers: ✓

Three clear layers: acquisition (parsers) → processing (mBLL/IVT/FFT) → classification (trit).

gx10 Deployment

Validated on 4x NVIDIA GB10 nodes (aarch64-linux, 128GB unified memory each):

# Install zig on gx10 node
curl -sL -o /tmp/zig.tar.xz 'https://ziglang.org/download/0.15.2/zig-aarch64-linux-0.15.2.tar.xz'
mkdir -p ~/.local && tar xf /tmp/zig.tar.xz -C ~/.local/
ln -sf ~/.local/zig-aarch64-linux-0.15.2/zig ~/.local/bin/zig

# Clone and test
git clone -b feat/bci-multimodal-pipeline https://github.com/plurigrid/zig-syrup.git
cd zig-syrup && zig build test-bci

gx10 BCI Use Cases

• Headless BCI acquisition server: Run LSL bridge + EDF writer on idle nodes
• Cross-compile target: Native aarch64 build, same arch as embedded targets
• Parallel dataset processing: Parse/classify large EDF archives across 4 nodes
• LoLa integration: BCI trit streams as input features for autoencoder training

Related Skills

GF(3) Triads

zig-syrup-bci(0) ⊗ sdf(-1) ⊗ bci-colored-operad(+1) = 0 ✓
zig-syrup-bci(0) ⊗ edf-reader(-1) ⊗ fnirs-processor(+1) = 0 ✓

PR

https://github.com/plurigrid/zig-syrup/pull/2