The light on the back of the watch
Turn a smartwatch over and you will usually see small glowing lights and a few dark windows. This is an optical heart rate sensor, and the technique is called photoplethysmography, or PPG. The idea is simple. The lights, often green, shine into the skin on the wrist. Blood absorbs more of that light than the surrounding tissue does. Each heartbeat sends a pulse of blood through the small vessels under the sensor, so the amount of light reaching the photodiodes changes slightly with every beat. Software counts those rhythmic changes and reports beats per minute.
Green light is common partly because blood absorbs it strongly, which gives a clearer signal at the wrist. Some devices also use red or infrared light, which can be used in estimates of blood oxygen. Those estimates are more sensitive to error, and consumer versions are not substitutes for clinical measurements.
Turning light into a number
The raw signal is a wavy line, and it is messy. It includes the pulse, but also changes from breathing, tiny movements and shifts in contact between the watch and the skin. The watch must separate the pulse from everything else. The usual steps are:
- filtering out frequencies that cannot be a heartbeat,
- using the motion sensor to identify and subtract movement-related interference,
- picking out peaks that represent beats, and
- averaging over a short window to produce a stable reading.
Because that last step smooths the data, the number on screen can lag behind a quick change, such as the start of a sprint. Many watches also sample less often at rest and more often during a workout to conserve battery.
Developers often fine-tune these algorithms with machine learning, using examples where a reference measurement was recorded alongside the wrist signal.
How the watch senses movement
Movement is measured mainly by an accelerometer, a tiny component that reacts to changes in speed and to gravity. Many watches pair it with a gyroscope that senses rotation. Together they produce a stream of numbers describing how the wrist is moving in three dimensions.
Steps. Walking produces a regular rhythm in the accelerometer signal as the arm swings. Software looks for repeating peaks of the right size and timing and counts them as steps. This works reasonably well for ordinary walking but can be fooled by activities that produce similar rhythms, and it can miss steps when the arm is still, as when you push a cart or hold a handrail.
Activity type. By comparing patterns in the signal against known examples, software can often tell walking from running, cycling or swimming. This is a pattern-recognition task, which is typically handled with machine learning methods such as neural networks. When the pattern is unusual, the watch may guess wrong or need you to choose the activity manually.
Distance and pace. Without a satellite fix, distance is estimated from step counts and an assumed stride length. With a receiver for satellite navigation, the watch can compute speed and distance outdoors much more directly, though tall buildings, tree cover and canyons can degrade the signal.
Falls and other events. Some watches look for a sharp impact followed by stillness. These features can be useful but may produce false alarms or miss a real event.
Elevation and floors. Many watches include a barometric pressure sensor, since air pressure falls slightly with altitude. Weather changes can also move the reading.
Why readings go wrong
Several factors are widely recognized as affecting wrist-based optical sensing.
- Motion. Vigorous, irregular movement, such as weight training or gripping a handle, can mimic or hide a pulse. Steady activities such as walking usually work better than interval training.
- Fit. A watch that is too loose lets light leak in and slides across the skin. A very tight strap can restrict blood flow. A snug but comfortable fit is generally recommended.
- Skin and tissue. Skin tone, tattoos, scars, hair and wrist size can influence how much light is absorbed or scattered. Researchers have discussed how differences in skin tone may affect accuracy in some optical devices, and it is a reason to treat any single reading with caution.
- Temperature and circulation. Cold weather narrows blood vessels in the extremities, weakening the signal.
- Sensor placement. Where the watch sits on the wrist matters, and so does which wrist.
A chest strap, which detects the heart’s electrical signal, is commonly considered more reliable during hard exercise. Even so, no consumer device is perfect.
What the readings are good for
At rest and during steady activity, wrist readings are often close enough to be useful for noticing patterns: how your resting pulse changes over weeks, how hard a familiar walk feels compared with last month, or how consistently you move during the day. They are less trustworthy for exact numbers during intense training.
Some watches also try to flag an irregular heartbeat by looking for unevenness between beats, and some can record an electrical signal when you touch a contact. These features are intended as prompts to seek advice, not diagnoses. A normal result does not rule out a problem, and an alert does not confirm one. If you have symptoms such as chest pain, fainting or a racing heart, seek medical attention rather than relying on a gadget.
A note on the data
A heart rate and movement record can say a lot about a person’s habits, and where it is stored matters. Some processing happens on the watch itself and some in the cloud, and the privacy implications are covered in the discussion of wearable health data. For an overview of the full range of body-worn devices, the survey of wearable devices is the place to start.