Walkeeg v3.0

WalkEEG

An open-source wearable biosignal platform for EEG, ECG and EMG.

Schematics, firmware, mobile app, cloud backend — and a 500 Hz live browser dashboard. Built for BCI research, education, and anyone curious about the electrical activity inside their body.

What is this?

WalkEEG captures electrophysiological signals from the body, streams them wirelessly, and makes them visible in real time — on a phone, in a browser, or in a desktop oscilloscope.

These span four orders of magnitude in amplitude. A front end that handles all three rates its noise against the weakest (EEG) while keeping enough headroom for the strongest (EMG) — which is what an 80–120 dB CMRR analog front end delivers.

What's in the box

walkeeg/
├── 01-hardware/     KiCad schematics — analog front end, power, connectors
├── 02-firmware/     nRF52840 firmware (Zephyr / Nordic Connect SDK)
├── 03-PC program/   Desktop oscilloscope (LabVIEW, EN + Chinese)
├── 04-MATLAB program/ Analysis scripts (legacy, optional)
├── 05-Flutter program/ iOS + Android app
├── 06-website/      www.walkeeg.com (static, S3 + CloudFront)
├── 07-backend/      AWS SAM — one deployment example, not the platform
├── 08-realtime/     500 Hz WebSocket relay + live browser dashboard
├── algorithms/      Platform-neutral signal statistics (stdlib Python)
└── docs/            Design notes and interface specs
     └── algorithms/ Algorithm contracts + conformance tests

Quick start

Firmware

The firmware targets the nRF52840 and builds with Nordic Connect SDK. A complete, verified toolchain lives at /opt/nordic/ncs/v2.5.99-dev1 on macOS — but any recent NCS works.

# 1. Source the NCS environment (adjust paths to your install)
export ZEPHYR_BASE=/path/to/ncs/zephyr
export ZEPHYR_TOOLCHAIN_VARIANT=zephyr
export ZEPHYR_SDK_INSTALL_DIR=/path/to/ncs/toolchains/*/opt/zephyr-sdk

# 2. Build from inside the NCS west workspace — not from the app dir
west build -b nrf52840dk/nrf52840 -d /tmp/build-walkeeg 02-firmware

# 3. Flash (Nordic boards)
nrfutil device recover   --serial-number <SN>
nrfutil device program   --serial-number <SN> --firmware /tmp/build-walkeeg/zephyr/zephyr.hex
nrfutil device reset     --serial-number <SN>

Building with J-Link instead? Older J-Link firmware (e.g. V9) can fail to program the nRF52840 — erase succeeds, writes silently don't stick. Use nrfutil instead.

Startup diagnostics. The firmware blinks the run LED N times to report how far it got:

Mobile app

cd "05-Flutter program"
cp lib/config/app_config.example.dart lib/config/app_config.dart  # then fill in your AWS values
flutter pub get
flutter run

Requires Bluetooth permissions. On iOS, add NSBluetoothAlwaysUsageDescription to Info.plist.

Live dashboard

cd 08-realtime
npm install
node server.js          # http://0.0.0.0:8787
node simulate.js        # optional: synthetic 500 Hz source

Open http://<your-lan-ip>:8787. The phone relays frames over WebSocket to /ingest; browsers subscribe at /live.

How it works

    AD8232                  nRF52840                 Phone / PC
 ┌──────────┐   analog   ┌────────────┐    BLE    ┌────────────┐
 │ Electrodes│ ────────▶ │  SAADC     │ ────────▶ │  Flutter   │
 │  + AFE    │  10-1000µV│  + NUS GATT│  2 Mbps   │  app       │
 └──────────┘            └────────────┘           └─────┬──────┘
                                                        │ WebSocket
                                                        ▼
                                              ┌──────────────────┐
                                              │ 08-realtime      │
                                              │ browser dashboard│
                                              └──────────────────┘

1. Acquire — electrodes pick up microvolt-level potentials; the AFE amplifies and filters them.
2. Transmit — the nRF52840 digitises at 500 Hz – 4 kHz and streams over BLE 5.0 using Nordic UART Service, with 2M PHY, data-length extension (251-byte payloads) and a 498-byte MTU.
3. Visualise — the Flutter app renders live waveforms with QRS detection and HRV metrics.
4. Store — sessions upload to S3 via Cognito-authenticated direct PUT; metadata lands in DynamoDB.
5. Share — the app can relay raw frames to a local WebSocket server, so any browser on the network sees the same signal live.

BLE packet format

magic 'W' (0x57) | version (1) | seq (u16 LE) | t_ms (u64 LE) | n (u8) | ch[8][n] (u16 LE)
└──────────────────────────── 13-byte header ─────────────────────────────┘

8 channels, 500 Hz, n = 50 samples per batch → ~10 frames per second.

Repository map

Contributing

Issues and pull requests are welcome. Before opening a PR:

1. Keep the firmware buildable against current NCS — note the NCS version you tested with.
2. For hardware changes, attach the updated schematic PDF alongside the KiCad files.
3. For protocol changes, update docs/ in the same commit.

Where does algorithm code belong?

Signal processing is specified once in docs/algorithms/ and implemented once in algorithms/. That pair is the only place the math should exist.

• Real-time algorithms run next to the data (app DSP, relay) — latency forbids a network round trip. They reimplement the contract for performance, and must match it.
• Offline / batch algorithms belong in algorithms/. 07-backend/ is just one way to run them, on one cloud. Swap it for cron, a Kubernetes job, or a laptop script without touching the math.
• Every implementation must satisfy docs/algorithms/conformance.md against the same fixtures.

Run the contract tests at any time:

python3 algorithms/python/conformance_test.py

See CONTRIBUTING.md for details (coming soon).

License

MIT — see LICENSE.

Copyright (c) 2026 the WalkEEG team · www.walkeeg.com