
How to choose a wearable for research
A research wearable checklist: raw waveform access, export formats, wear-time realities, and battery under study settings. Built for study ops.
Quick answer: Pick the wearable that can deliver the signal your analysis needs, under the wear-time and power budget your protocol can sustain. Start with raw waveform access, export paths, wear-time reporting, and battery life under study settings. Lock the capture profile before enrollment so every device writes the same data model.
Start with the analysis, not the brand
A study fails quietly when the device only ever stored someone else's score. If you cannot re-derive features, audit artifact windows, or hand a waveform to a collaborator, you bought a dashboard, not a research instrument.
Write the analysis plan first. Name the features (overnight rMSSD, respiratory rate from beat intervals, steps and cadence, a custom optical model). Then ask every vendor the same four questions:
- Can we store the raw waveform (or at least beat intervals) when the protocol needs them?
- Can we export into our warehouse, EDC, or notebook in a format and window we can actually run?
- Will people wear it long enough for the n in the SAP to be real?
- What is battery and on-device storage under the sample rate and duty cycle we will actually use?
Those four items are the checklist. Industrial design, app polish, and a recovery-score screenshot come after. If a vendor cannot answer them in writing, keep shopping.
This page is the shortlist version of the job. The setup version (once you have picked a stack) is how to configure a research wearable. The RFP version is wearables for clinical trials. The endpoint-definition version is digital biomarkers.
Sensor Bio is one stack you can score against the same four items: Capture on V2 hardware, Connect through SDK and API, Build as an org-level device profile, Act as operator views after the signal exists. Score us the same way you score anyone else.
Checklist item 1. Raw waveform access
Ask whether optical PPG leaves the device as a waveform you can store, not only as heart rate or a recovery index.
Derived scores are hard to re-analyze later. If the protocol may need re-derivation, algorithm work, or external review of the optical signal, you need the waveform (or at least beat intervals) stored on purpose. "We can send you a PDF of the recovery trend" is not waveform access.
For Sensor Bio programs, raw PPG export includes unfiltered amplitudes for green, infrared, and red channels when those wavelengths are enabled. Timestamps are nanosecond-resolution in the raw download. Green is the default primary channel. Infrared and red appear when configured. On V2 hardware, published wavelength centers are green 530 nm, red 655 nm, and infrared 940 nm.
| Raw field | Meaning |
|---|---|
ts | Nanosecond timestamp |
green | Unfiltered green amplitude (AU) |
ir | Unfiltered infrared amplitude (AU), when enabled |
red | Unfiltered red amplitude (AU), when enabled |
No filtering is applied in that raw package. Accelerometer raw is a separate enable: x, y, z in m/s² with millisecond timestamps. Ask for it when artifact review or activity-trace re-analysis is in scope. Motion context is how you explain a bad optical window. It is not optional for a model team and often optional for a coaching program that only needs computed steps.
Two operational facts matter before first enrollment:
- Raw PPG is disabled by default in both daytime Passive and overnight Rest Mode. Turn it on for the organization if the protocol needs it. A vendor who says "we have raw" and ships with raw off in sleep will give you computed nights and an empty waveform archive. Write the on-switch into site initiation, not into a late amendment.
- Sample rate is configurable per organization at 25, 50, or 100 Hz (default 50 Hz). Pick from the analysis, not from a marketing default. A rate that is too coarse cannot be fixed later if the raw signal was never stored.
If your team only needs beat-to-beat intervals and computed metrics (heart rate, HRV as rMSSD, respiratory rate, steps, cadence), say so explicitly. On-device and near-device SDK options also cover sleep staging, HRV, beat-to-beat, and active minutes. Do not pay the power and storage cost of raw capture without a use for it.
Vendor questions that belong in the first call:
- Which channels leave the device as a file we can archive: raw PPG, beat intervals, computed metrics, raw accel?
- Who turns raw on: org admin, wearer app, or a support ticket after first visit?
- What is the sample rate, who sets it, and can it change per device after enrollment?
- Are Passive (awake) and Rest Mode (sleep) separate defaults? Do both need an explicit raw enable?
If the answers are "our app shows heart rate" and "sample rate is not exposed," you are looking at a consumer stack. That can still be the right buy for a coarse activity endpoint. It is the wrong buy for a re-derivable optical record. For the partner file path, see raw PPG access. For the optical primer, see photoplethysmography (PPG).
Checklist item 2. Export that your pipeline can use
Research stacks fail on formats and windows as often as they fail on sensors. A waveform you cannot land in the warehouse is a waveform you do not have.
Demand clarity on:
- File formats (Sensor Bio exports CSV and JSON for standard downloads)
- Windows per request (computed metrics: 1-30 days; raw metrics: 1-7 days)
- Concurrency and link lifetime (Sensor Bio: 3 concurrent jobs per organization; links expire and can be refreshed)
- API access for automation (Sensor Bio REST base URL:
https://api.sensorbio.com/v1/, API key header) - Mobile paths when the phone is in the loop (iOS via SPM, Android Kotlin Flow BLE SDKs)
- Who is allowed to pull data (Sensor Bio raw/computed downloads require a Super Admin role)
- Whether third-party apps use a consent flow (Sensor Bio documents OAuth 2.0 consent for third-party apps)
If your warehouse, EDC, or notebook cannot ingest the vendor's default package without a custom reverse-engineering project, keep shopping.
| Path | When it fits a research program |
|---|---|
| Batch CSV / JSON | Study ops and partner validation before a full connector ships |
| REST | Server-side pipelines, warehouses, scheduled pulls |
| Mobile SDK | Phone mediates BLE sync; on-device or near-device processing inside your app |
Cadence is part of export, even when people file it under "device settings." Live, 1 minute, 5 minute, and 15 minute cadences change how soon an operator sees a gap. They do not replace the archive. A live stream with no batch export is a coaching product. A 15-minute cadence with a raw file you can pull later is a research product. Ask for both the live path and the archive path if you need both. Real-time physiological data streaming covers the live side. Wearable API for health products covers REST versus SDK versus aggregator architecture.
A practical first test: ask the vendor to walk one short capture from device to a file you open on your laptop. If that walk takes a professional services SOW, you do not have an export story yet.
For research and product teams
Put the raw signal in your study or product
If your shortlist already needs waveform export and a battery number under study settings, we can walk those four checks against a capture profile.
Checklist item 3. Wear-time and adherence
Signal you never capture is missing data, not a rounding error.
Ask how the device and program tooling report:
- Hours worn per day
- Charge and sync gaps
- Dropout after week one
Comfort, strap fit, and charge friction decide whether your n is real. A device that can write 100 Hz raw and sits in a drawer after night four is a failed instrument. Plan follow-up before the cohort starts. Do not wait for the closeout meeting to discover half the nights are empty.
Wear-time is a protocol object, not a hope. Write target hours and target nights with a valid window the same way you write visit windows. Then ask the vendor what the operator actually sees when a device misses a night: last sync time, battery, wear-time proxy, capture profile still matching the protocol. Without that view, support tickets become forensic science.
Three wear-time failure modes show up early:
Charge copy that does not match the profile. Participants were told the default-settings story. The profile is closer to max recording. They charge weekly. Nights go empty. This is a checklist item 4 leak that presents as a checklist item 3 problem.
Sync that requires a phone nobody carries at night. If the mobile SDK path is in the loop, say so in the participant instruction. If cloud-to-cloud REST is the archive path, say who triggers the pull and how stale a device can get before on-device storage wraps. Published V2 on-device storage is about 2.5 days under default settings and about 1.2 hours under max recording. A weekend without sync on a max-recording profile is not a small gap.
No owner for dropout. Someone has to watch week one. Adherence watchlists exist for this. If the vendor only offers a consumer app and a monthly CSV, you are the watchlist. Budget that time.
PPG signal quality is the optical-quality companion to this item. Wear-time is whether the device was on. Signal quality is whether the optical window is usable while it was on. You need both.
Checklist item 4. Battery under study settings
Vendors love "up to X days." Protocols love high sample rates and raw export. Those two sentences fight each other.
Battery under default wellness settings is not battery under max recording or high sample-rate capture. Ask every vendor for both. If they only quote the default, treat the number as a consumer marketing figure until they show the study-settings figure.
For Sensor Bio V2, published device comparison figures are:
- 5 to 7 days of battery under default settings
- Up to 2 days under max recording settings
- On-device storage about 2.5 days under default settings, about 1.2 hours under max recording settings
Treat those as planning bounds, not promises for every duty cycle. Higher sample rates, raw channels, and continuous capture shorten the week. A 25 Hz computed-only profile and a 100 Hz raw-on profile do not share a charge instruction. Put the tradeoff in the protocol and in the participant instructions.
| Planning question | Default-settings envelope | Max-recording envelope |
|---|---|---|
| Battery (published V2 comparison) | 5 to 7 days | About 2 days |
| On-device storage (published V2 comparison) | About 2.5 days | About 1.2 hours |
| Typical research posture that lands here | Computed metrics, raw off, 50 Hz default | Raw on, high sample rate, continuous capture |
| Charge / sync copy you can defend | Weekly charge may be plausible | Nightly charge and frequent sync |
You do not get a published per-rate hour table beyond those two envelopes. Do not invent one. Use the bounds to pick a posture, then validate on a canary device under your duty cycle before you print participant materials.
What to ask other vendors, in the same language:
- Battery at the sample rate you will actually set
- Battery with raw waveform on versus derived-only
- On-device storage when the phone is out of range
- Whether Passive and Rest Mode change the power story
- What happens when storage fills: wrap, stop, or drop to computed-only
If the answer is a single "up to 7 days" with no study-settings counterpart, you do not have a battery number. You have a brochure.
A short decision table
| Question | Consumer black-box stack | Research / infrastructure stack |
|---|---|---|
| Can you store raw PPG? | Often no | Yes, when enabled |
| Who sets sample rate? | Vendor app defaults | Your org profile |
| Export into your warehouse? | Limited or delayed | API, SDK, and batch export |
| Battery quote matches study mode? | Usually not | Ask for default vs max-recording |
| Wear-time visible to operators? | Wearer app, maybe | Hours worn, sync gaps, dropout |
| Can you version the capture model? | Rarely | Profile applied once across the cohort |
Score a vendor (including Sensor Bio) by walking one protocol through the table. An overnight recovery program that only needs computed rMSSD can accept raw off, 50 Hz, 5 or 15 minute cadence, and the default-settings battery envelope. An algorithm program that will reopen green / ir / red amplitudes needs raw on, a written sample rate, Super Admin export, and the max-recording envelope in the charge SOP. A motion-forward activity program needs honest accel access and a cadence matched to coaches, not a copied 100 Hz raw profile "just in case."
Those three postures are program illustrations, not care protocols. Details live in how to configure a research wearable.
How the four items interact
Teams treat the checklist as four independent boxes. The failures come from the joints.
Raw on (item 1) without Super Admin export (item 2) means the waveform exists in a place you cannot pull. Export that only works as a live stream (item 2) without wear-time views (item 3) means you will not know which subjects are silently off-wrist. Wear-time targets (item 3) written against a 5 to 7 day charge story while the profile is max recording (item 4) produce empty nights that look like non-adherence. Battery bounds (item 4) quoted without naming sample rate and raw on/off (item 1) are not comparable across vendors.
Write one capture profile that names all four. Then run a one-device rehearsal:
- Apply the org profile (rate, cadence, sensors, raw on/off).
- Wear it through a Passive window and a Rest Mode window.
- Pull one computed file (1-30 day window rules) and one raw file if raw is on (1-7 day window rules).
- Open the file. Confirm columns. Confirm the interval matches the rate you set.
- Leave the phone away long enough to learn the storage behavior under this profile.
- Write the charge and sync sentence from what you observed, bounded by the published 5 to 7 day / 2-day and 2.5-day / 1.2-hour figures.
If that rehearsal fails, the vendor failed the checklist. If it passes, you have a protocol object you can put in the SAP.
Failure modes in vendor selection
Score shopping. A recovery index with a pretty confidence interval is not a channel list. Ask for the upstream signal.
Default-settings battery in the protocol. The brochure number is the default envelope. Your profile may be the other envelope.
Raw promised, raw off at shipment. Defaults keep Sensor Bio raw off in Passive and Rest Mode. Put the on-switch in the initiation checklist.
Export as a portal screenshot. If the only path is a human clicking through a vendor UI, you do not have automation. You have a bottleneck.
One sample rate for every study. 25, 50, and 100 Hz are different products in power and storage. Match the rate to the analysis. See configuration for how to lock it.
Buying the algorithm profile for a coaching study. Extra raw channels and 100 Hz feel serious. They also move you toward the 2-day / 1.2-hour envelope. Serious is the profile that matches the SAP.
No canary device. First enrollment is a poor time to learn that Rest Mode raw was still off.
Next steps
Use the companion buyer guide on wearables for clinical trials when you are writing an RFP. When partners need the access path for waveforms, see raw PPG access. For profile setup before enrollment, read how to configure a research wearable. For architecture (REST, SDK, aggregator), see wearable API for health products.
Get started is the product conversation when the four checks already point at a configurable stack. Technical reference stays on developers.sensorbio.com.











