A PPG PPG An optical method that shines light into the skin and detects the pulse of blood with each heartbeat. Most watches, rings and phone-camera apps use it to estimate HRV. More about PPG sensor is the light on the back of a watch or inside a ring. It shines into your skin and measures how much light comes back, and each heartbeat pushes a little extra blood into the skin that dims the light for a moment, so the sensor sees one wave per beat. Everything a watch tells you about your heart, from heart rate and HRV to blood oxygen and irregular-rhythm alerts, starts with the timing and shape of those waves.
What a PPG sensor does
PPG stands for photoplethysmography: “photo” for light, “plethysmo” for volume. The hardware is small. A light-emitting diode (LED) shines into the tissue and a photodiode measures the light that comes out. Blood absorbs light, so when more blood is in the tissue, less light reaches the photodiode (Allen, 2007).
There are two ways to arrange the parts:
- Transmission. The LED and the photodiode sit on opposite sides of a fingertip or earlobe. Hospital pulse oximeters and finger clips work this way.
- Reflection. The LED and the photodiode sit side by side on the same surface, and the photodiode catches light that scatters back out. Watches, rings and the camera-plus-flash trick on a phone work this way (Allen, 2007).
The colour of the light matters. Red and infrared light go deep into the tissue, which is what a pulse oximeter needs. Green light is absorbed strongly by blood and does not go as deep, so at the wrist it gives a cleaner pulse and holds up better during movement. That is why the back of most watches glows green for heart rate and switches to red and infrared only for blood oxygen (Castaneda and colleagues, 2018; Charlton and colleagues, 2023).
The signal has two parts. Most of the light that returns is a steady level, called the DC component, set by the skin, bone and the blood that is always there. On top of it rides a small ripple, the AC component, that follows each heartbeat. The ripple is a small fraction of the total, which is why the sensor’s electronics must amplify it and why small disturbances can swamp it (Allen, 2007).
The shape of one pulse
Turn the ripple upside down so that more blood points up, and each heartbeat looks like this.
- Systolic peak
- Dicrotic notch
- Diastolic peak
- Pulse-to-pulse interval, from one systolic peak to the next
A smooth wave with two pulses. Each pulse rises steeply to a systolic peak, falls to a small dip called the dicrotic notch, rises again to a lower diastolic peak, and then settles back to the baseline. The gap between the two systolic peaks is marked as 1.00 seconds.
Each beat draws the same three landmarks (Elgendi, 2012):
- Systolic peak. The heart contracts and sends a pressure wave down the arteries. Blood volume in the skin rises fast to its peak.
- Dicrotic notch. The aortic valve closes and the wave from the heart fades. The volume starts to fall.
- Diastolic peak. The pressure wave reflects back from branch points and the lower body and arrives as a second, smaller hump before the volume settles.
Algorithms rarely work on the raw curve. They pick out points on it: the foot where the wave starts to rise, the peak, the point of steepest rise, and the timing and height of the notch and the diastolic peak. One popular method takes the second derivative of the wave, which turns gentle bends into clear spikes labelled a, b, c, d and e, and compares their sizes (Elgendi, 2012; Takazawa and colleagues, 1998).
How the wave changes with age, stiffness and fitness
The reflected wave is the interesting part. In an elastic young artery the pressure wave travels slowly, so the reflection comes back late and arrives as a separate diastolic hump with a clear notch in front of it. In a stiffer artery the wave travels faster, the reflection comes back sooner and merges into the main wave, and the notch fills in.
Doctors noticed this long before wearables. In 1973 Dawber and colleagues sorted arterial pulse waves into four classes, from a distinct notch to no notch at all, in a study of coronary heart disease (Dawber and colleagues, 1973). A clear notch is the usual finding in people with healthy, compliant arteries (Elgendi, 2012).
Later work put numbers on it. Millasseau and colleagues measured the time from the systolic peak to the diastolic peak and divided the person’s height by it to get a stiffness index. In 87 adults without symptoms, aged 21 to 68, the index rose with age and tracked the speed of the pressure wave along the aorta, the standard measure of large-artery stiffness. They also described a reflection index, the height of the diastolic hump as a fraction of the systolic peak, which responds to drugs that change the tone of the arteries (Millasseau and colleagues, 2002; Millasseau and colleagues, 2006). Takazawa’s team worked from the second derivative instead and combined the a to e spikes into an “aging index”, (b − c − d − e)/a, which rose with age in 600 people (Takazawa and colleagues, 1998).
Fitness pulls the other way. In 146 healthy volunteers aged 21 to 96, arterial stiffness rose with age, but at any age it was lower in people with higher aerobic capacity, and endurance-trained men aged 54 to 75 had less stiff arteries than sedentary men of the same age (Vaitkevicius and colleagues, 1993). In another study, endurance-trained middle-aged and older men had central arteries 20% to 35% more compliant than less active men of the same age, and when 20 sedentary men took up regular aerobic exercise, mostly walking, for three months, their arterial compliance rose to a similar level (Tanaka and colleagues, 2000).
Two cautions. Those fitness studies measured stiffness directly, with pressure sensors and ultrasound at the neck and leg, not from a watch. And PPG contour indexes change for reasons that have nothing to do with the arteries: where the sensor sits, how hard it presses, skin temperature and ambient light all move them, and so do drugs that act on the vessels (Charlton and colleagues, 2023; Millasseau and colleagues, 2006). They describe groups well. They are not a test of your own vascular age, and no watch can tell you how trained your arteries are from the shape of a pulse.
Try it: read a pulse like a watch does
The explorer below is a teaching model, not a recording. Each pulse is the sum of two bell curves, one for the systolic wave and one for the reflection, which is a standard way to build synthetic PPG for testing algorithms (Tang and colleagues, 2020). A simple detector then finds the peaks, times the gaps between them and works out RMSSD RMSSD Root mean square of successive differences: an HRV metric built from how much each heartbeat interval differs from the next, in milliseconds. It mainly reflects vagal activity. More about RMSSD , the same way an app would.
Read a pulse like a watch does
A teaching model, not a recording
Detected gaps 1010 · 1040 · 980 · 960 · 1010 · 1040 · 980 ms
- Pulse rate
- 60 bpm
- RMSSD, measured
- 45 ms
- RMSSD, true
- 42 ms
At 100 Hz the samples are 10 ms apart and the detector finds every beat, so RMSSD reads 45 ms, within 3 ms of the model's true 42 ms.
A clear dicrotic notch and a separate diastolic wave (reflection index 0.50): the picture of elastic arteries.
The sliders work when JavaScript is on.
A line chart of 8 seconds of a synthetic PPG signal with one bump per heartbeat. Short ticks mark the detected pulses. A second chart plots the gap between detected pulses as a line, with the model's true gaps as a dashed line. Readouts give the pulse rate, the RMSSD of the detected gaps and the true RMSSD, followed by a sentence explaining any difference and a sentence describing the shape of each pulse. Five sliders set arterial stiffness, heart rate, signal strength, movement and sampling rate, and a checkbox turns on interpolation between samples.
Try these settings in turn. Push stiffness up and watch the notch vanish. Drop the signal strength, as cold hands would, until the detector loses beats. Add movement and see the gaps go wild. Then set the sampling rate Sampling rate How many times a second a sensor measures the signal, in hertz (Hz). It sets how precisely each beat can be timed: about ±1 ms at 1000 Hz, ±4 ms at 250 Hz and ±20 ms at a 25 Hz camera. More about Sampling rate to 25 Hz: every beat is still found, yet RMSSD changes, because each peak time is only known to the nearest 40 ms.
Pulse rate variability is not quite HRV
HRV, strictly, is measured from the heart’s electrical signal on an ECG ECG A recording of the heart's electrical activity, taken with electrodes on the skin. ECG-based RR intervals are the reference standard for measuring HRV. More about ECG , as the gaps between R peaks. A PPG sensor sees the pressure wave that the beat sends out, and that wave takes time to reach the wrist or finger. The delay, called the pulse transit time, is not constant. It changes with blood pressure, with posture, with the tone of the arteries, and a little with every breath. So the gap between two pulse waves is the gap between two heartbeats plus the change in transit time, and the variation you measure from a watch is pulse rate variability Pulse rate variability Variation in the time between pulse beats, measured optically by watches, rings and phone cameras. It closely tracks HRV at rest but drifts from it during movement. More about Pulse rate variability , PRV, not HRV (Mejía-Mejía and colleagues, 2020; Yuda and colleagues, 2020).
In practice the two agree well when you keep still. A review of studies that compared PPG with ECG found PRV an acceptable stand-in for HRV at rest, with a tendency to read short-term measures such as RMSSD a little higher, which the authors traced to breathing acting on the circulation. During physical activity and some mental tasks the agreement got much worse, sometimes to an unacceptable degree (Schäfer and Vagedes, 2013). Yuda and colleagues go further and argue that PRV carries extra information about the blood vessels, so it should be treated as its own measure rather than a copy of HRV (Yuda and colleagues, 2020).
For your trend this means two things. A resting reading from a good watch, ring or camera app is a fair estimate of your HRV. And a watch and a chest strap will never agree to the millisecond, even on the same beats.
What goes wrong
Every PPG number rests on finding the right peak at the right time. These are the usual ways that fails.
Movement
Moving the arm shifts the sensor against the skin and sloshes blood around, which adds swings to the signal far larger than the pulse. In a study of 53 people wearing six devices, heart rate error was on average 30% higher during activity than at rest, and the type of activity and the device model made the biggest difference (Bent and colleagues, 2020). Rhythmic movement is the worst case because a step rate of 160 a minute looks like a heart rate of 160 (Zhang and colleagues, 2015).
A weak pulse: cold hands and poor perfusion
Cold narrows the small vessels in the skin, so less blood reaches the sensor and the ripple shrinks while the sensor’s own noise stays the same. Allen’s review lists skin temperature and perfusion among the main things that change the size of the pulse wave (Allen, 2007). A camera reading taken in a cold room can fail outright.
Contact pressure and ambient light
The sensor must sit flat and firm. Too loose, and the LED’s light leaks out, daylight leaks in and the signal drifts as the strap shifts. Too tight, and the pressure squeezes the small vessels and flattens the pulse (Allen, 2007; Charlton and colleagues, 2023).
Skin tone and tattoos
Melanin absorbs light, so darker skin returns less of it. Whether this changes the readings has been studied most for heart rate, and the evidence is mixed. The 53-person study above measured skin tone and found no statistically significant difference in heart rate accuracy across tones (Bent and colleagues, 2020). A 2023 systematic review found ten studies with 469 people in all: four reported lower accuracy in darker skin, four found no effect and two were mixed, and few people with the darkest skin had been tested (Koerber and colleagues, 2023). Blood oxygen is a clearer story. In records from one US hospital, when the pulse oximeter read 92% to 96%, a blood test showed oxygen below 88% in 11.7% of measurements from Black patients and 3.6% from White patients; a multi-hospital dataset gave 17.0% and 6.2% (Sjoding and colleagues, 2020). Treat a watch’s SpO2 as a rough guide and heart rate as the more trustworthy number.
Dark tattoo ink absorbs light too. Apple’s own guidance says the ink, pattern and saturation of some tattoos can block light from the sensor and make readings unreliable (Apple Support).
Sampling rate and interpolation
A sensor reads the light a fixed number of times a second. At 25 samples a second a peak can only be placed to the nearest 40 ms, and HRV is built from differences of a few tens of milliseconds. In a pilot study that down-sampled 10 kHz PPG recordings, PRV measures did not differ significantly from ECG HRV at 25 Hz and above, while RMSSD started to differ at 15 Hz and below (Choi and Shin, 2017). A 2021 study of 57 five-minute recordings put the limit higher: SDNN SDNN The standard deviation of all NN intervals in a recording, in milliseconds. It measures overall variability, and its value depends strongly on how long the recording lasts. More about SDNN and RMSSD stayed within 5% of the 1 kHz value down to 50 samples a second without interpolation, and down to 10 and 20 samples a second when the peak position was interpolated between samples (Béres and Hejjel, 2021). Few makers say how often their sensor samples.
Extra beats and atrial fibrillation
A premature beat Ectopic beat A heartbeat that starts outside the heart's normal pacemaker and usually comes early. Ectopic beats are common, but they distort HRV unless removed or corrected. More about Ectopic beat pushes out less blood than a normal one, so its pulse wave is small and the detector often misses it. The recorder then sees one long gap instead of two short ones. HRV standards say extra beats and the gaps around them must be removed before any HRV number is calculated (Task Force, 1996), because a single misplaced beat can change the result (Laborde and colleagues, 2017). In atrial fibrillation the gaps are irregular beat after beat, and PRV stops meaning what it usually means; that irregularity is what the watch alerts described below look for.
What watches work out from PPG
One signal feeds many numbers. Each algorithm below has its own method and its own limits.
Beat detection and heart rate. The simplest approach finds each peak or foot and counts. During exercise many watches switch to a frequency approach instead: they look for the strongest regular rhythm in the signal over a few seconds, and use the accelerometer to cancel the rhythm that matches your steps. TROIKA, a well-known method of this kind, reached an average error of 2.34 beats a minute in 12 people running at up to 15 km/h (Zhang and colleagues, 2015). Heart rate is the hardest PPG number to get badly wrong, because an error in one peak barely moves a 10-second average.
HRV, really PRV. The app lists the gaps between peaks, throws out gaps that look wrong, and applies a formula such as RMSSD. Because one mistimed peak changes RMSSD a lot, most makers measure during sleep and reject stretches with movement. See how HRV is measured for the formulas.
Blood oxygen (SpO2). Oxygen-rich and oxygen-poor blood absorb red and infrared light differently. The sensor compares the size of the pulse ripple at the two colours, a “ratio of ratios”, and converts it to a percentage using a curve fitted on volunteers (Allen, 2007). The skin-tone bias above lives in this step.
Breathing rate Breathing rate How many breaths you take a minute. It moves HRV on its own: slower, deeper breathing makes the beat-to-beat swing larger, and below about 9 breaths a minute it shifts from the HF band into LF. More about Breathing rate . Breathing leaves three marks on the PPG: the baseline Baseline Your own usual HRV level and range, built from weeks of readings taken the same way. Comparing a reading with your baseline says more than comparing it with other people. More about Baseline drifts up and down, the pulses grow and shrink, and the gaps between them lengthen and shorten. Algorithms pull out one or more of these and count the cycles. Movement and poor sensor contact corrupt the signal they rely on, and few have been tested on data from people going about their day (Charlton and colleagues, 2018).
Irregular rhythm alerts. The watch checks the gaps between pulses, in the background, for the chaotic pattern of atrial fibrillation and, if several checks agree, sends a notification. In the Apple Heart Study of 419,297 people, 0.52% got a notification over a median of 117 days. Of the 450 who then returned a readable ECG patch, 34% had atrial fibrillation on it, and when a new notification arrived while the patch was on, the ECG showed atrial fibrillation at that moment 84% of the time (Perez and colleagues, 2019). In the Fitbit Heart Study of 455,699 people, 1% got a detection. Of the 1,057 who returned a readable patch, 32.2% had atrial fibrillation on it, and a fresh detection during patch wear matched atrial fibrillation on the ECG 98.2% of the time (Lubitz and colleagues, 2022). A notification is a prompt to get an ECG, not a diagnosis, and no notification is not an all-clear.
Cuffless blood pressure. There are two routes. One times the pressure wave between two points, such as an ECG spike at the chest and a PPG peak at the wrist. A stiffer, more pressurised artery carries the wave faster, so a shorter transit time suggests higher pressure, but each person needs a cuff calibration and the device must be recalibrated from time to time (Mukkamala and colleagues, 2015). The other route reads the shape of the single PPG wave. An IEEE standard for cuffless devices has existed since 2014 (IEEE 1708), and it does ask for tests across a rise and a fall in pressure, but experts have pointed out that it leaves the way those changes are produced undefined, so a device can pass without showing it tracks the kinds of change that matter (Mukkamala and colleagues, 2021). The European Society of Hypertension says these devices should not be used for clinical decisions until they pass proper validation (Stergiou and colleagues, 2022).
“Vascular age” and stress scores. Some apps turn the stiffness index, the reflection index or the second-derivative aging index into a vascular age, or mix PRV with other signals into a stress score. The indexes do track age in groups, as the studies above show. Whether a given person’s score means anything has not been validated, and the same person can get a different vascular age by pressing the sensor harder or warming their hand. Read them as a curiosity.
PPG, ECG and chest straps for HRV
| Wrist or ring PPG | Phone camera PPG | Chest strap (ECG) | |
|---|---|---|---|
| What it times | Pulse wave in the skin | Pulse wave in the fingertip | The heart’s electrical R peak |
| Measure | PRV | PRV | HRV |
| When | Sleep, or spot readings | A reading you start | A reading you start |
| Samples per second | Rarely published | The camera’s frame rate, usually 30 or 60 | 250 to 1,000, the range the ECG standards advise |
| At rest | Close to ECG | Close to ECG in a 29-person study | The home reference |
| While moving | Unreliable | Not usable | Good, if the strap is wet and snug |
| Extra beats | Often missed | Often missed | Visible and removable |
The ECG standards ask for 250 to 500 samples a second or more, and accept a lower rate only with interpolation (Task Force, 1996). The camera figure is the usual phone frame rate, and in a study of 29 people at rest a camera app’s RMSSD agreed closely with an ECG (Plews and colleagues, 2017). A clinical ECG, with several leads, remains the reference for all three. Device by device evidence is on HRV on Oura, Whoop, Garmin and Apple Watch.
What this means for your readings
- Compare a PPG device only with itself. Its HRV is PRV, measured on its own schedule with its own filters, so it will not match a strap or another watch. The lesson on why your devices disagree works through an example.
- Keep still and warm for any reading you start yourself, and wear the strap snug but not tight, as the makers advise (Apple Support). How to take a good HRV reading has the routine.
- Trust sleep readings over daytime spot checks. Still hours give the detector its best chance.
- Treat a one-off strange value as a sensor problem first. Movement, a cold hand or a missed beat explains most of them.
- Use heart rate with confidence, HRV as a trend, SpO2 as a rough guide, and vascular age and stress scores as a curiosity.
- Take an irregular-rhythm alert to a doctor for an ECG. Do not use the absence of alerts as reassurance if you have symptoms.
- If you want ECG-grade HRV at home, a chest strap and a free app give it. See is wrist HRV as accurate as a chest strap? and can a phone camera measure HRV?