PPG for wearable products: wavelengths, quality, and raw access
How photoplethysmography works in wearable products: optical stack tradeoffs, motion and perfusion limits, multi-wavelength design, and when raw PPG access matters.
Quick answer: photoplethysmography (PPG) is an optical measurement of pulsatile blood-volume change in tissue. It can support pulse rate, interval timing, and, in purpose-built multi-wavelength systems, oxygen-saturation estimates. It is not ECG, and a green pulse waveform is not a pulse-oximeter.
What the waveform actually is
PPG is a noninvasive optical technique. An LED illuminates tissue. A photodetector measures transmitted or reflected light. Arterial volume rises in systole and falls in diastole, so the effective optical path changes. The detector sees that as a varying intensity signal (1, 2).
The resulting trace can support heart-rate estimation, pulse-interval analysis, morphology features, perfusion-related trends, and, in multi-wavelength systems designed for it, oxygen-saturation estimation. It is useful because it is simple to acquire on a wrist or finger. It is limited because it is indirect.
PPG is not a generic "heart signal." It does not measure cardiac electricity. If the product question is electrical timing, use ECG. If the question is oxygen saturation, use a true multi-wavelength pulse-oximetry design. The side-by-side lives in PPG vs ECG vs pulse oximetry.
Fingertip transmission traces are the historical teaching case (5). Wrist reflectance traces are what most wearable products actually ship. Do not treat a fingertip figure as a wrist performance claim.
How the signal forms
A PPG channel combines three things: an optical emitter, a detector, and a biological optical path. That path includes skin, connective tissue, venous blood, arterial blood, and local anatomy. As the arterial volume changes, so does the mix of absorption and scattering the detector sees (4).
The waveform is therefore shaped by more than the heartbeat:
- arterial pulsation
- tissue optical properties
- wavelength-dependent penetration depth
- local perfusion
- sensor geometry (emitter-detector spacing, photodiode area)
- contact mechanics (strap tension, pressure, intermittent coupling)
- ambient-light contamination and instrumentation noise
Treat a PPG trace as a composite optical phenomenon, not a pure pressure wave and not a direct measure of cardiac electricity. That framing is what keeps derived features honest later.
Contact versus non-contact geometries change the path again. Imaging and remote PPG exist in the literature (4). Wearable products almost always stay in contact reflectance or fingertip transmission. This page stays with those two.
Transmission versus reflectance
Two geometries dominate practical systems.
Transmission
The light source and detector sit on opposite sides of the tissue. Light passes through the volume before it reaches the detector. This is the classic fingertip pulse-oximeter architecture (5).
Strengths: strong signal at thin, well-perfused sites. Mature path for pulse oximetry. Often cleaner than wrist reflectance.
Constraints: limited to sites where transmission geometry is practical. Less suitable for low-friction continuous everyday wear.
Reflectance
The emitter and detector sit on the same side of tissue. The detector captures light that has scattered back. This is the dominant architecture for wrist wearables and many adhesive or patch systems (1).
Strengths: compact hardware. Flexible placement. Practical for continuous wear.
Constraints: more sensitive to motion, pressure changes, and placement variability. The optical path is less stable than a fingertip transmission setup.
If you copy a fingertip validation pack onto a wrist product, you have not validated the wrist product. You have borrowed a geometry you do not ship.
AC and DC components
PPG is often decomposed into a pulsatile AC component and a slower DC component.
| Component | What it mainly reflects | Why it matters |
|---|---|---|
| DC | Baseline tissue optics, nonpulsatile blood, venous volume, temperature effects, contact conditions, slow drift | Sets the background. Shifts with placement, pressure, and physiology |
| AC | Beat-synchronous pulsatile arterial variation | Carries most of the information used for pulse timing, morphology, and pulse-oximetry style calculations |
In many systems the AC component is small relative to the total optical signal. Practical PPG engineering is often about recovering a small physiological signal from a much larger background plus several noise sources (3).
That ratio is why "we filtered it" is not a quality strategy. Filtering can make a weak AC look smooth. It cannot invent pulsatile contrast that contact, perfusion, or motion already destroyed.
Wavelength selection is application-dependent
Different wavelengths interact differently with tissue, blood, and depth.
- Green light is commonly used in wrist wearables because it often provides strong superficial pulsatile contrast for heart-rate tracking.
- Red and infrared are central to many multi-wavelength systems and pulse-oximetry architectures.
- No single wavelength is universally best. The right choice depends on body site, target metric, motion environment, power budget, and optical stack design.
Copying another device's wavelength choice without matching its use case is usually a mistake. Green-only pulse timing and a red/IR saturation estimate are different products that happen to share a photodetector family.
On Sensor Bio V2, published wavelength centers are green 530 nm, red 655 nm, and infrared 940 nm. Green is the default primary channel. Infrared and red appear in the raw package when those channels are enabled on the organization profile. Wavelength center is a hardware fact. It is not a validation study.
Optical stack design
Most writeups start at the waveform and the algorithm. The stack before that (how light enters tissue, interacts with it, and returns to a detector) determines what is even available to process.
| Design variable | Why it matters | Typical tradeoff |
|---|---|---|
| LED current and duty cycle | Affects signal amplitude, battery life, and thermal load | More light can improve SNR but increases power cost and saturation risk |
| Photodiode area | Influences sensitivity and noise behavior | Larger area can collect more light but may add capacitance and front-end constraints |
| Emitter-detector spacing | Shapes sampling volume and depth sensitivity | Too small or too large can reduce usable pulsatile contrast |
| Sampling timing and duty cycle | Controls ambient-light rejection and dynamic range | Weak timing design creates avoidable artifacts |
More optical power is not a universal fix. Balance signal amplitude, power budget, thermal behavior, and analog front-end headroom together. Optimizing any one variable in isolation is how you get a bright LED and a saturated diode.
Ambient-light rejection is a hidden differentiator. Effective rejection typically combines optical barriers in the enclosure, synchronized LED sampling (read the photodetector when the LED is on), dark-frame subtraction (a lights-off measurement to remove background), and housing that closes leakage paths. This engineering rarely appears on a product slide. It often decides whether field data is usable.
Body site tradeoffs
Site selection is often the difference between a working device and a failing deployment.
| Site | Benefits | Challenges |
|---|---|---|
| Finger (transmission) | Often strong pulsatile signal. Mature optical use case | Less practical for all-day continuous wear |
| Wrist | Comfortable and familiar for consumer wearables | Motion, tendon movement, variable perfusion, pressure instability |
| Ear | Can provide stable pulse information in some applications | Comfort and acceptance constraints |
| Forehead | Often useful in monitored settings | Less natural for continuous consumer wear |
| Upper arm or patch | May improve stability in some designs | Form-factor and long-term adherence tradeoffs |
The best site in the lab is not always the best site for long-term wear. The most comfortable site is not always the most optically reliable. Write both in the protocol: the site you validated, and the site people will actually keep on.
What PPG can support, and what it should not be sold as
Under appropriate conditions, PPG can support:
- pulse rate and heart-rate style outputs
- beat-to-beat interval estimates when signal quality is adequate
- pulse waveform morphology features
- perfusion-oriented trends
- respiratory modulation features under some conditions
- oxygen saturation when the system is explicitly designed and validated for multi-wavelength pulse oximetry
These are meaningful capabilities. Frame them as estimates from an optical path, not as direct physiologic truth.
PPG is not:
- a direct ECG replacement
- a standalone disease diagnosis engine
- a direct blood-pressure measurement without model assumptions and calibration
- equally reliable across all motion states, temperatures, skin properties, and perfusion levels
- automatically suitable for SpO2 estimation just because an optical pulse waveform exists
A green-channel heart-rate pipeline and an SpO2 pipeline share a family of optics. They do not share a validation pack. If you need saturation, design and validate for it. Do not infer it from a pulse waveform you already had.
Derived features are easy to compute and easy to over-trust
Once a waveform is available, teams compute pulse-interval statistics, amplitude and contour features, upstroke and decay timing, respiratory modulation of amplitude or baseline, pulse-arrival timing when combined with ECG, and signal-quality indices for confidence-aware estimation.
The technical issue is not whether these features can be computed. It is whether they remain valid across real-world confounders. Interval variability computed on a motion-corrupted wrist trace is a number. It is not HRV research.
Beat detection is its own subject. If you will swap a detector, store the waveform. See PPG beat detection. Quality scoring that gates the detector is PPG signal quality.
Failure modes you should design for
PPG quality degrades fast outside controlled settings. Fine et al. catalog the usual inaccuracy sources for continuous optical monitoring: motion, perfusion, contact, ambient light, and population variability among them (3). Design as if those are the default environment.
Motion artifact
Motion changes the optical path and often overwhelms the true pulsatile component, especially at the wrist. Exercise is the obvious case. Talking, typing, and strap bounce are enough. If the protocol includes a motion condition, validate that condition. Do not borrow a seated fingertip figure.
Contact pressure
Strap tension, compression, or intermittent coupling alters local perfusion and waveform shape. A "too tight" night and a "too loose" commute produce different artifacts that can look like physiology if you only store a smoothed rate.
Low perfusion
Cold exposure, vasoconstriction, or site-specific circulatory changes shrink the pulsatile signal. The DC stays. The AC disappears. Algorithms that always emit a number will invent a pulse.
Ambient light leakage
Poor housing, shielding, or timing injects large nonphysiological fluctuations. Outdoor wrist wear is a different optical problem than a covered fingertip probe.
Population and anatomy
Skin properties, tissue thickness, hair, site choice, and intersubject variability all affect signal quality and algorithm behavior. Stratify validation. Do not hide the split.
Algorithmic false confidence
A heavily smoothed output can look reassuring even when the underlying signal is unreliable. This is one of the most important product risks. Filtering is necessary. It is not a quality strategy on its own.
Abstention logic
A system that sometimes reports "signal unavailable" is more credible than one that always produces a number. Credible pipelines include beat-level or window-level confidence scoring, quality indices that gate downstream outputs, and abstention logic that declines to report a metric when the waveform is not trustworthy. Silent bad data is worse than missing data.
Motion context helps you explain a bad optical window. On Sensor Bio, accelerometer raw (x, y, z in m/s², millisecond timestamps) is a separate enable from raw PPG. If you will reject high-motion windows, turn accel on and join on time in your pipeline.
When raw access matters
Derived scores are hard to re-analyze later. If you cannot reopen the optical record, you cannot change a filter, hand a segment to a collaborator, or defend a derived feature when someone asks how it was built.
You need the amplitudes when at least one of these is true:
- You will write or swap a beat detector, filter, or quality model after collection starts.
- A reviewer will ask to see the optical record, not only rMSSD or heart rate.
- Morphology features or multi-wavelength ratios you compute yourself are in scope.
You do not need the amplitudes when the protocol is satisfied by stored beat-to-beat intervals plus computed heart rate, rMSSD, respiratory rate, steps, or cadence. Raw is a cost in power and storage. Pay it when you will open the file.
For Sensor Bio programs, raw PPG export includes unfiltered amplitudes for green, infrared, and red when those wavelengths are enabled. Timestamps are nanosecond-resolution. Field names are ts, green, ir, and red. No filtering is applied in that package. Values are arbitrary units (AU). Do not treat AU as a calibrated blood-volume number.
Two operational facts belong in the capture profile, not in a week-two ticket:
- Raw PPG is disabled by default in Passive (awake) and Rest Mode (sleep). Turn it on for the organization if the protocol needs waveforms.
- Sample rate is configurable per organization at 25, 50, or 100 Hz (default 50 Hz). A rate that is too coarse cannot be fixed later if the raw signal was never stored.
Computed downloads cover 1-30 days per request. Raw downloads cover 1-7 days. Concurrent jobs are limited to 3 per organization. Super Admin is required for those batch pulls. REST lives at https://api.getsensr.io/v1/ with an API key header. iOS SDK is via SPM. Android is Kotlin Flow BLE.
Confirm field names and routes on Developers before you freeze a connector. Partner wiring is raw PPG access. The SDK and API references are the live contracts. This page states published product facts. If the live reference and this article disagree, trust the live reference.
Battery and storage move with the profile. Published V2 comparison figures are about 7 days battery and about 2.5 days on-device storage under default settings, versus about 2 days and about 1.2 hours under max recording. Treat those as planning bounds. A 100 Hz raw-on profile and a 50 Hz computed-only profile do not share a charge instruction. Trial-side RFP language is in wearables for clinical trials. Profile setup is in how to configure a research wearable.
What to document in a study or product protocol
Write the measurement context before first enrollment:
- geometry (transmission vs reflectance) and body site
- wavelengths enabled, and the published centers if you rely on them
- sample rate and whether raw amplitudes are stored
- cadence for computed outputs (live, 1 min, 5 min, 15 min)
- artifact handling, quality flags, and abstention rules
- missingness and wear-time definitions
- reference method for each endpoint, separately
- participant placement and wear-time instructions
- export path, role, and pull windows
- privacy boundary (what leaves the device, what stays on-device)
Claims should stay inside the validation you actually ran. Pulse-rate agreement is not interval-timing agreement. Interval timing is not morphology. Saturation is not "we had a red LED."
Sensor Bio V2 is a general wellness device, not FDA cleared for diagnosis. Keep product language inside research and program design.
Related reading
- PPG signal quality
- PPG vs ECG vs pulse oximetry
- PPG beat detection
- Raw PPG access for partners
- SpO2 monitoring
- Wearables for clinical trials
- How to configure a research wearable
- Digital biomarkers
References
1. Tamura T, Maeda Y, Sekine M, Yoshida M. Wearable Photoplethysmographic Sensors: Past and Present. Electronics. 2014. Source
2. Allen J. Photoplethysmography and its application in clinical physiological measurement. Physiological Measurement. 2007. Source
3. Fine J, Branan KL, Rodriguez AJ, et al. Sources of inaccuracy in photoplethysmography for continuous cardiovascular monitoring. Biosensors. 2021. Source
4. Sun Y, Thakor N. Photoplethysmography Revisited: From Contact to Noncontact, From Point to Imaging. IEEE TBME. 2016. Source
5. Elgendi M. On the analysis of fingertip photoplethysmogram signals. Current Cardiology Reviews. 2012. Source