Movesense ECG/HRV Live Dashboard

Real-time ECG, HRV, and motion streaming from a chest-worn Movesense sensor to a browser dashboard over Bluetooth

✨ Motivation

Movesense ships a medical-grade chest sensor with ECG, IMU9, heart rate, and temperature channels, but the vendor SDK only targets Android and iOS. There is no first-party way to get live biosignals onto a laptop for quick experimentation, so this project builds that missing path: a Python backend that talks to the sensor directly over Bluetooth Low Energy, and a browser dashboard for watching and recording the streams in real time.

How do you read a wearable’s biosignal streams on a laptop when the vendor never built that path — and how do you trust the decoded values once you have them? The measurement payloads ride on Movesense’s proprietary GSP protocol, whose byte layout isn’t published, so decoding it correctly was as much the project as building the dashboard itself.

🔧 Hardware

A Movesense MD sensor clipped into a certified chest belt; the two electrode pads on the inside pick up the ECG.

🖥️ Live Dashboard

Sampling-rate and stream controls, heart rate / HRV / temperature tiles, and live ECG, accelerometer, gyroscope, and magnetometer traces — 15 seconds into a session labeled "jumping." The accelerometer dips mark the jumps.

🧊 Architecture Overview

Movesense MD sensor  --BLE-->  Python backend (bleak)  --WebSocket-->  Browser dashboard
                                     |
                                     +-- FastAPI serves /ws and /status

Two data paths run side by side. Heart rate and RR-intervals arrive over the standard Bluetooth Heart Rate Service, pre-decoded; the backend keeps a rolling window of the last 60 RR-intervals and reports a live SDNN as an HRV indicator. ECG, IMU9, and temperature arrive over GSP, Movesense’s own protocol for clients that can’t run their mobile SDK - all three share one GATT notify characteristic, distinguished by a reference code chosen at subscribe time, with binary payload layouts that had to be decoded from live captures rather than a published spec. The backend forwards everything over a WebSocket to a single-file HTML/Chart.js frontend, which also drives labeled-session recording: stopping a capture returns a ZIP with one CSV per stream plus a meta.json and a per-recording README.

🔍 Verifying an Undocumented Protocol

Movesense’s GSP spec documents the command/response envelope (HELLO, SUBSCRIBE, response codes) but not the byte layout of the measurement payload itself. Guessing at a binary format for a biosignal and trusting it blindly seemed worse than not decoding it at all, so each channel was checked against something independently known to be true rather than assumed correct:

  • Accelerometer: sqrt(x²+y²+z²) should read ~9.8 m/s² on a roughly stationary sensor (Earth’s gravity) — confirmed across multiple samples.
  • ECG: raw int16 samples scaled to mV using Movesense’s published conversion factor, checked against the shape of a normal resting ECG trace.
  • Magnetometer: flagged as unverified rather than silently assumed correct — an uncalibrated MEMS magnetometer carries a hard-iron offset that the obvious sanity check (magnitude in Earth’s ~25-65 µT field range) can’t confirm on a stationary sensor.
  • Sampling-rate safety: GSP has no GET verb to ask the sensor which rates it supports, and subscribing at an unsupported rate is silently accepted and delivers nothing - so starting a recording resubscribes, waits for packets to actually arrive, and refuses to start (restoring the previous working rate) if none do.

⚙️ Technical Stack

  • Hardware: Movesense MD sensor, certified chest belt with ECG electrode pads
  • Protocols: Bluetooth Low Energy, standard BLE Heart Rate Service, Movesense GSP (reverse-engineered payload decoding)
  • Backend: Python 3.10+, FastAPI, bleak, WebSocket
  • Frontend: Single-file HTML dashboard, Chart.js (vendored, no CDN dependency)
  • Data export: Multi-stream CSV + JSON metadata per recorded session
  • Tests: unittest, run via GitHub Actions on every push

🔗 Links