Cadence Lock: When Your Heart-Rate Monitor Starts Counting Your Steps

Mostafa Dhouib·7 min read

Cadence Lock: When Your Heart-Rate Monitor Starts Counting Your Steps

The short answer. An optical heart-rate sensor on a moving body sees the cardiac pulse and the motion artefact mixed together, and the usual way to separate them is to find the dominant periodic component and reject the one that matches the accelerometer. That works while the two frequencies are far apart. In field sports, step cadence climbs into the same range as heart rate, and when they converge the algorithm can lock onto cadence and report it as heart rate. The reading stays smooth, plausible and stable, which is exactly what makes it dangerous, because an unstable wrong reading is at least visibly wrong. The condition where the frequencies meet is the load-bearing row of the validation protocol, and it is the row most protocols omit.

An optical heart-rate band works beautifully on a jog. Take it onto a pitch and the number occasionally becomes fiction, in a way that looks completely normal.

Why the signal has to be separated at all

Optical sensing shines light into tissue and measures what comes back. Blood volume changes with each heartbeat, modulating the returned light, and that modulation is the cardiac signal.

It is small. Motion modulates the same measurement far more strongly, because movement changes the optical path: the sensor shifts against the skin, pressure varies, tissue deforms. On a moving body the motion artefact is routinely larger than the signal being measured.

So the algorithm has to separate them. The standard approach uses the accelerometer: find the periodic components in the optical signal, identify which one matches the motion, subtract or reject it, and take what is left as cardiac.

The assumption underneath is that the motion frequency and the cardiac frequency are distinguishable. Most of the time they are.

The convergence

Walking is around 2 Hz, roughly 120 steps per minute. Resting heart rate is around 1 Hz. Comfortably separated.

Running raises cadence toward 2.5 to 3 Hz, or 150 to 180 steps per minute. Working hard raises heart rate toward 2.5 to 3 Hz as well, or 150 to 180 beats per minute.

In field sports, both are in that band at the same time, repeatedly, for extended periods. The two signals the algorithm is supposed to distinguish are now at approximately the same frequency, and frequency is the only thing it was using to tell them apart.

When they converge, cancelling the motion component can cancel the cardiac component with it, because they are the same component as far as the separation is concerned. What remains, and what the algorithm reports, can be the cadence.

Why the failure is well disguised

If the sensor reported noise, somebody would notice. Cadence lock does the opposite.

It is stable. Step cadence is a cleaner, more regular signal than heart rate, so the output looks better than a correct reading. Smooth, low variance, no artefacts.

It is plausible. It reports 165 while the athlete is working hard. That is a reasonable heart rate for that effort. Nothing about the value is suspicious.

It is physiologically coherent. Cadence and heart rate rise and fall together with effort, so the trace tracks exertion. It looks like a heart-rate curve because it is driven by the same underlying variable.

It recovers by itself. When the athlete stops, cadence goes to zero, the frequencies separate, and the algorithm reacquires the real heart rate. By the time anyone looks at the data, it is correct again.

The combination means the only way to detect it is to be measuring something else at the same time.

What this does to validation

If your protocol is steady-state treadmill work at easy and moderate paces, you will not see this. The frequencies stay separated, the algorithm performs well, and your reported error is genuinely achieved under the conditions you tested.

The protocol therefore has to contain the condition where cadence and heart rate converge, and it has to be the row you look at first. The relevant consensus standard for wearable heart-rate validation sets out the structure for this: an electrocardiogram reference, a defined error metric, a validity threshold, and mandatory stratification by skin type, sex and body mass index. The stratification matters independently because optical absorption varies with skin pigmentation and tissue coupling varies with body composition, so an aggregate figure over an unrepresentative cohort means less than it appears to.

The practical addition is a column alongside the error metric: how often, and for how long, did the reported value track the accelerometer's dominant frequency rather than the reference? That is the cadence-lock rate, and it is the number that distinguishes a sports product from a treadmill product.

What can be done about it

Report a confidence or quality signal, and act on it. Where the frequencies are converged, the algorithm's own separation is unreliable and it generally knows. Exposing that is more useful than smoothing over it.

Abstain rather than interpolate. A gap in the trace is honest. A plausible number during convergence is not, and a coach cannot distinguish them.

Improve the mechanical coupling. A substantial share of the artefact comes from relative motion between the sensor and the skin. Placement, strap design and pressure change the ratio between signal and artefact, which widens the range over which separation works at all. This is why sensor placement is an algorithm decision rather than an industrial-design one.

Use the accelerometer as a flag, not only as a subtrahend. When the dominant motion frequency approaches the reported heart rate, that is a condition worth marking in the output rather than only an input to the cancellation.

The honest limit

None of this eliminates the problem. When two signals occupy the same frequency and one is far stronger, no amount of processing separates them reliably from a single optical channel. That is an information limit, not an implementation shortfall.

What the above buys is knowing when you are in that condition and saying so, which is the difference between a product that is wrong sometimes and a product that is confidently wrong without telling anyone. Multi-wavelength sensing and better mechanical design widen the operating envelope. They do not remove the boundary.

FAQ

Why does my heart-rate monitor show my step rate during exercise? Because optical heart-rate algorithms separate the cardiac signal from motion by frequency, and in hard running cadence rises into the same range as heart rate, around 150 to 180 per minute. When the two converge, the separation cannot distinguish them, and the algorithm can report the cadence as the heart rate.

Why is cadence lock so hard to notice? Because the wrong reading is smoother and more plausible than a correct one. Step cadence is a cleaner signal than heart rate, the value is reasonable for the effort, it rises and falls with exertion, and it corrects itself as soon as the athlete stops, so the recorded data looks fine afterwards.

Will a treadmill validation protocol catch this? No. At easy and moderate steady-state paces the frequencies stay separated and the algorithm performs well, so the reported error is real for the conditions tested and silent about the one that matters. The protocol has to include the condition where cadence and heart rate converge.

What should a wearable heart-rate validation protocol include? An electrocardiogram reference, a defined error metric with a stated validity threshold, cohort stratification by skin type, sex and body mass index, and a column reporting how often and for how long the output tracked the accelerometer's dominant frequency rather than the reference.

Can cadence lock be eliminated? Not from a single optical channel. When two signals occupy the same frequency and one is far stronger, separating them reliably is an information limit rather than an implementation shortfall. Better mechanical coupling and multi-wavelength sensing widen the operating envelope without removing the boundary.

Why is sensor placement an algorithm decision? Because a large part of the motion artefact comes from relative movement between the sensor and the skin, so placement, strap design and contact pressure set the ratio between the cardiac signal and the artefact. That ratio determines the range over which frequency separation works at all, which is an algorithmic property decided by mechanical design.