What Your Wrist Sensor Actually Sees: How a PPG Sensor in a Smartwatch Works

Turn your smartwatch over and press it against a dark surface. You will see flickering green and red lights through the back cover. Those lights are not decorative. They are the front end of a photoplethysmography sensor, a tiny optical laboratory that fires photons into your skin and reads what bounces back. Understanding how a PPG sensor smartwatch how does it work means tracing the path from that first photon all the way to the health metrics on your screen. The numbers, heart rate, blood oxygen, stress level, even estimated blood pressure, all start with that pulse of light.

Yet few people know what happens between the photon and the number. Most explanations available online stop at "light measures blood flow," which is true but thin. A green LED plus a photodiode gets you a heartbeat, but getting from there to SpO2 and heart rate variability requires physics, signal processing, and a chain of inferences that is far more interesting than the final display suggests.

Why Green Light? The Hemoglobin Question

If you designed a heart rate sensor from scratch, you might not choose green. Red LEDs are cheaper and infrared penetrates deeper. But evolution has made a specific choice for us: hemoglobin, the oxygen-carrying protein in red blood cells, absorbs green light with particular strength. At roughly 530 nanometers, the absorption coefficient of hemoglobin hits a local maximum. Green photons do not travel far into tissue, about 0.5 to 1 millimeter, but they do not need to. The capillary bed just beneath the skin surface is where the action is.

Each time the heart contracts, a pressure wave pushes a fresh bolus of blood through the capillaries. Blood volume in the illuminated tissue increases momentarily, and more green photons get absorbed. The photodiode on the sensor records a dip in reflected light intensity. Between beats, blood volume decreases, less light gets absorbed, and the reflected signal rises. Plot this over time and you get a photoplethysmogram: a waveform whose peaks map to heartbeats, whose troughs map to the diastolic pause, and whose shape encodes a surprising amount of information about your cardiovascular state.

Green was not an arbitrary choice. Research comparing wavelengths across the visible spectrum found that green light produces the strongest AC-to-DC ratio in the PPG signal. The AC component is the pulsatile part, the heartbeat-driven change, and the DC component is the steady baseline from tissue, bone, and venous blood. A higher AC/DC ratio means the heartbeat signal stands out more clearly against the noise floor. Green also resists motion artifact better than longer wavelengths because it penetrates less deeply. It samples a more localized tissue volume, which is less susceptible to displacement when you move your wrist.

Take a device like the Geelouxian MT500 as a concrete example. It uses a 530nm green LED for heart rate, sampling the reflected light 100 times per second. That 100Hz sampling rate is not overkill. Heart rate alone requires only a few samples per second to count beats, but measuring the precise timing between beats, the foundation of heart rate variability analysis, demands millisecond-level resolution.

Reflective vs. Transmissive: Why the Wrist Is Harder Than the Fingertip

If you have ever used a fingertip pulse oximeter in a clinic, you have experienced transmissive PPG. The device clips onto your finger with a light source on one side and a detector on the other. Photons travel straight through the tissue, following a relatively short and well-defined path. The signal is strong and clean.

Your smartwatch cannot clip onto anything. It sits on top of your wrist, firing light in and measuring what scatters back out. This is reflective PPG, and it is fundamentally harder.

In the reflective geometry, photons enter the skin, scatter through multiple tissue layers, epidermis, dermis, subcutaneous fat, muscle, and only a fraction find their way back to the photodiode. The path length is longer, more variable, and heavily influenced by local anatomy. Blood vessels in the wrist are deeper and more dispersed than in the fingertip. The signal is weaker by roughly an order of magnitude.

Ambient light compounds the problem. Sunlight, fluorescent office lighting, even the display of the watch itself produce broadband noise that can swamp the tiny AC signal from pulsatile blood flow. The sensor enclosure uses a light-blocking gasket, and the LEDs pulse at specific frequencies so the system can mathematically subtract ambient light. The margin is thin. This is why your watch asks you to keep still during a SpO2 reading. It needs every photon it can get, and motion introduces variables the algorithm cannot fully correct.

The Beer-Lambert Law: How Two Wavelengths Become a Blood Oxygen Number

Heart rate is a counting problem. Count the pulses per minute and you are done. Blood oxygen saturation is a spectroscopy problem, and it requires physics that goes back to the 18th century.

The Beer-Lambert Law states that the amount of light absorbed by a substance depends on three factors: the concentration of the absorbing species, the distance the light travels through it, and a wavelength-specific constant called the extinction coefficient. Written out:

A equals epsilon multiplied by c multiplied by l

Where A is absorbance, epsilon is the molar extinction coefficient, c is concentration, and l is path length. The intensity of light exiting the tissue follows:

I equals I-subscript-zero multiplied by 10 raised to the power of negative epsilon times c times l

Now here is the insight that makes pulse oximetry possible. Oxygenated hemoglobin and deoxygenated hemoglobin have different extinction coefficients at different wavelengths. At 660 nanometers, red light, deoxygenated hemoglobin absorbs more strongly. At 940 nanometers, near-infrared, oxygenated hemoglobin absorbs more strongly.

By flashing red and infrared LEDs alternately and measuring the ratio of light absorbed at each wavelength, the sensor can estimate what fraction of hemoglobin is carrying oxygen. The math works because you are measuring the same tissue, at the same moment, with the same path length. The only variable that changes between the two readings is the absorption difference between Hb and HbO2.

But a wrist sensor adds a complication the Beer-Lambert Law was not designed for. The law assumes a known, constant path length, which is reasonable in a transmissive finger clip where the light goes straight through. In a reflective wrist sensor, photons scatter unpredictably through multiple tissue types. The path length becomes a statistical estimate rather than a known constant. This is the central reason that wrist SpO2 readings carry a wider error margin, typically plus or minus 3 to 5 percent, compared to medical finger clips at plus or minus 2 percent. Studies published in Lancet Digital Health and PLOS Digital Health have confirmed that consumer smartwatch SpO2 can deviate from clinical measurements by several percentage points, particularly at lower saturation levels and in users with darker skin tones.

The MT500 uses a 660nm red LED and a 940nm infrared LED alongside its green heart-rate LED, implementing this dual-wavelength ratio method on a reflective wrist platform. The physics is identical to what a hospital pulse oximeter uses. The geometry is not, and that difference defines the accuracy ceiling. When someone asks how a PPG sensor smartwatch how does it work and stops at the LED colors, they miss the entire chain of reasoning that turns two absorption ratios into a single percentage on the display.

Heart Rate Variability: What the Gaps Between Beats Reveal

Counting beats per minute is the shallowest layer of cardiovascular monitoring. The gaps between beats tell a richer story.

Your heart does not beat like a metronome. If your resting heart rate is 60 beats per minute, the intervals between consecutive beats might be 0.98 seconds, then 1.04 seconds, then 0.96 seconds, then 1.01 seconds. This beat-to-beat variation is heart rate variability, or HRV, and it reflects the ongoing tug-of-war between two branches of your autonomic nervous system.

The sympathetic branch accelerates your heart and prepares your body for action. The parasympathetic branch, mediated by the vagus nerve, slows it down and promotes recovery. A healthy autonomic system keeps both branches active, producing high variability. When stress, illness, or poor recovery tilt the balance toward sympathetic dominance, variability drops.

PPG sensors measure HRV indirectly. Instead of detecting the electrical R-wave peak, as an ECG chest strap would, the sensor detects the mechanical pulse wave arriving at the wrist. The time from one pulse peak to the next is the pulse-to-pulse interval, an approximation of the R-R interval that ECG-based HRV analysis uses. At rest, this approximation correlates well with ECG-derived HRV. Studies report correlation coefficients around 0.90 to 0.97 in controlled conditions. During movement, the correlation degrades because the pulse arrival time at the wrist is affected by changes in blood pressure and vascular tone.

From the sequence of intervals, software computes standardized metrics. RMSSD, the root mean square of successive differences, reflects short-term, vagally-mediated variability and is the primary metric for recovery tracking. SDNN, the standard deviation of NN intervals, captures total variability over a measurement window. The LF/HF ratio, derived from frequency-domain analysis, estimates the balance between sympathetic and parasympathetic activity, though its interpretation remains debated in the research community.

A 2021 systematic review of 35 studies by Hickey and colleagues found that HRV-based stress detection achieved 85 to 93 percent sensitivity during overnight sleep, the most favorable measurement window, but dropped to 65 to 75 percent during ambulatory conditions when no accelerometer context was available. Adding motion data improved accuracy to 82 to 92 percent. This tells you something about the sensor: it works best when you are still, and the accelerometer is not a backup. It is a co-equal data source.

From Photons to Digits: The Signal Processing Chain

The gap between a flickering LED and a HRV score on your screen is filled by a signal processing pipeline that most explanations of how a PPG sensor smartwatch how does it work leave completely unexamined.

Step one is the analog front end. The photodiode converts reflected light into a tiny current, which an amplifier boosts to a usable voltage. An analog-to-digital converter samples this voltage at a fixed rate. On a 100Hz system, that means 100 measurements per second, each one a snapshot of how much light made it back through the tissue.

The raw signal is noisy. Ambient light at 50 or 60 hertz from indoor lighting introduces a periodic hum. Respiration modulates the signal at roughly 0.2 to 0.4 hertz. Body movement produces lower-frequency baseline drift. The useful heartbeat signal lives in the 0.5 to 8 hertz band for resting adults, and a band-pass filter isolates this range while rejecting everything outside it.

After filtering, a peak detection algorithm scans the waveform for local maxima. Each detected peak marks one heartbeat. But not every peak is real. Motion artifacts can produce false peaks, and a missed peak shifts the entire interval calculation. The algorithm cross-references the PPG signal with accelerometer data. If the accelerometer registers a sharp wrist movement at the same instant the optical signal spikes, the system flags that data point as unreliable.

Once peaks are validated, the intervals between them form a time series. From this series, the standard HRV metrics including RMSSD, SDNN, and LF/HF are computed over windows of typically one to five minutes. Each metric requires statistical aggregation, and the choice of window length affects the result. Shorter windows respond faster to changes but are noisier. Longer windows are more stable but lag behind real physiological shifts.

The 100Hz sampling rate provides 10-millisecond resolution between consecutive samples. For heart rate measurement, counting beats over a minute, this is far more than necessary. For HRV calculation, where R-R interval differences of 20 to 50 milliseconds are physiologically meaningful, 10-millisecond precision is the minimum viable resolution. Sampling at 25Hz, by contrast, would quantize intervals into 40-millisecond bins and mask the subtle variations that define healthy autonomic function.

When the Sensor Struggles: The Physical Limits

A PPG sensor is not broken when it gives inconsistent readings during a run. It is operating at the edge of a design constraint rooted in physics.

Consider running at 180 steps per minute. Your wrist accelerates and decelerates 3 times per second. If your heart rate is 150 beats per minute, 2.5 hertz, the motion frequency and the heart rate frequency sit uncomfortably close. The photodiode cannot cleanly separate a change in reflected light caused by a heartbeat from one caused by your wrist shifting position relative to the sensor. The signal-to-noise ratio collapses. The displayed heart rate may spike, flatline, or lock onto your cadence instead of your pulse.

Skin tone introduces a second constraint. Melanin in the epidermis absorbs light across the visible and near-infrared spectrum. In darker skin, less light reaches the capillary bed and less returns to the photodiode. The effective signal is weaker, and the algorithm must extract the pulsatile component from a smaller absolute signal. The Lancet Digital Health study from 2022 documented that SpO2 accuracy in consumer wearables degrades measurably in users with darker skin, a finding that has prompted calls for more inclusive sensor calibration datasets. This is not a failure of any single device. It is a systematic bias in the optical approach itself.

Cold weather adds another variable. In low temperatures, peripheral blood vessels constrict to conserve core heat. Blood flow to the wrist capillaries drops, and the AC component of the PPG signal, the part driven by pulsatile blood volume changes, shrinks. A sensor that tracks heart rate reliably indoors may struggle to lock on during a winter walk. The photodiodes and LEDs are working. The tissue they are measuring has simply changed its optical properties.

Sensor fit is the variable most under user control. Too loose, and ambient light leaks into the optical chamber, raising the DC baseline and compressing the dynamic range available for the AC signal. Too tight, and the band compresses capillaries, reducing blood flow and weakening the signal. The usable range is narrow. Finding it requires a few seconds of adjustment followed by a check that the displayed heart rate looks physiologically plausible.

These are not defects. They are the physical boundary conditions of optical sensing on a moving, variable, living surface. Understanding them shifts the experience from "my watch is inaccurate" to "my watch is measuring a difficult signal in challenging conditions."

Blood Pressure: The Estimation That Is Not a Measurement

Many smartwatches now display a blood pressure number. The feature is easy to misunderstand.

Traditional blood pressure measurement uses an inflatable cuff to physically occlude the brachial artery. As the cuff deflates, a sensor detects the return of blood flow, marking systolic pressure, and the point where flow becomes continuous, marking diastolic pressure. This is direct, mechanical, and clinically validated.

Optical smartwatch blood pressure monitoring cannot occlude an artery. Instead, it estimates pressure indirectly. One method, Pulse Transit Time or PTT, measures the delay between the heart's electrical activation and the arrival of the pulse wave at the wrist. Stiffer arteries transmit the pulse faster, and arterial stiffness correlates with higher blood pressure. A shorter PTT suggests elevated pressure. Another approach, Pulse Wave Analysis or PWA, examines the shape of the PPG waveform. A sharper systolic upstroke correlates with arterial stiffness.

Both methods face the same fundamental problem: the relationship between PTT or waveform shape and absolute blood pressure is different for every person. It depends on age, arterial health, vessel geometry, and moment-to-moment vascular tone. Without individual calibration against a real cuff reading, an optical blood pressure estimate is a relative trend at best. Even with calibration, the relationship drifts over hours to days as vascular tone changes.

No smartwatch on the market holds FDA clearance for diagnostic blood pressure measurement. The feature provides a directional signal. A reading that is consistently above your personal baseline in the morning might reflect real physiological change. A single number taken in isolation tells you almost nothing. This is not a flaw in the sensor design. It is a limit imposed by estimating pressure from pulse timing in a medium as complex as living arterial tissue.

The green light on the back of your watch travels less than a millimeter into your skin before it starts its journey back. In that sub-millimeter round trip, it encodes information about your heart's rhythm, your blood's oxygen content, your nervous system's balance, and the stiffness of your arteries. It does this not through magic but through a chain of physics that extends from the quantum absorption properties of hemoglobin, through the Beer-Lambert Law, through analog signal conditioning and digital filtering, and finally to statistical metrics validated against clinical studies with known error margins.

Knowing this chain does not make the numbers more accurate, but it makes them more useful. When you understand that a wrist SpO2 reading carries a wider confidence interval than a fingertip measurement, you stop treating the two as interchangeable. When you understand that HRV reflects autonomic balance rather than a simple stress score, you learn to read low morning readings as a signal to prioritize recovery rather than a reason to worry. When you understand that motion artifacts are not sensor failure but a signal processing challenge rooted in frequency overlap, you stop blaming the device and start positioning your wrist more carefully.

A good sensor is honest about what it cannot measure. A good user knows what questions to ask of the numbers. The green light does the measuring. Physics does the rest.