Wito — bioimpedance research wristband for clinical-grade signal capture
Constraint
The box they were trapped in
Wito needed a wrist-worn wearable that captures non-invasive bioimpedance and physiological signals clean enough to feed downstream analytics and ML models — across study participants, across days, with the kind of timing discipline a regulated medical-device pathway eventually demands. The hard parts were synchronising five independent sensors on one I²C bus, validating signal quality on the fly, and respecting IEC 60601 / ISO 13485 design hygiene from day one rather than retrofitting it later.
Approach
How we attacked it
ESP32-D2WD as the host MCU with multi-rail power domains (VDD33 + VDD30) on a custom wearable PCB. Bioimpedance and ECG are read off a MAX30009; body and ambient temperature off an NCT75; motion off a BMA400; light off an EM7028; and a PCF8563 RTC anchors the timeline. The firmware brings up I²C with fixed sensor addressing, captures samples interrupt-driven instead of polled, and time-aligns every channel against the RTC so a downstream analyst can trust that what was happening at 12:01:03.045 across all six streams really did happen together. Signal validation runs in-firmware so a wrist-off or noisy bioimpedance read doesn't pollute the dataset. BLE ships the structured packets to the companion processing layer.
Decisions
What we picked, and what we rejected
Interrupt-driven capture with RTC-anchored timestamps
Polling five sensors at the rates bioimpedance research wants would have burned battery and still smeared the time alignment. Interrupts let each sensor signal when it has data, and stamping every sample against the PCF8563 means the downstream team can trust what 12:01:03.045 means across all channels — a non-negotiable for ML training and any future clinical comparison.
On-device signal validation before the data leaves the wrist
A wrist-off event or a single noisy bioimpedance window can poison a study if it makes it into the dataset unflagged. Validating in firmware lets us tag or drop those samples at source rather than trying to clean them up after they've crossed BLE — and saves the analyst from chasing data-quality issues that look like algorithm problems.
I²C over SPI for the multi-sensor bus
I²C with fixed addressing keeps a wrist-sized PCB routable and lets us add or swap a sensor without re-spinning the board. SPI would have given more bandwidth, but none of the streams here need it — and the routing complexity on a wearable form factor would have been the bigger cost.
Designed against IEC 60601 / ISO 10993 / ISO 13485 from day one, without certifying yet
Wito is building a research foundation, not a cleared product yet — but the second the device starts collecting data on humans, the design choices that aren't medical-device-friendly become expensive to undo. Following the standards from the start keeps the regulatory pathway open without paying for certification before there's a product to certify.
Trade-off
What we didn't build
We respected medical-device design practices (IEC 60601, ISO 10993, ISO 13485) without certifying the device — Wito was building a research foundation, not a regulated product yet. We left the higher-level analytics, model training, and clinical pipelines on Wito's side; our job was the device and the structured data feed it produces. We picked I²C over SPI for the sensor mix to keep the routing tractable on a wrist-sized PCB, accepting the lower bandwidth because none of the channels need it.
Outcome
What changed after we shipped
A research-grade wearable foundation that lets Wito collect consistent, time-aligned physiological data across study participants and moves the bioimpedance research toward clinical validation. Signal quality is good enough that the team can train models on the raw streams without re-doing the capture stack.
Talk to us
Have a similar project in mind?
Tell us what you're working on. We'll let you know whether it's a fit.