Why Your Watch Can't Measure Your Breathing Rate
Most modern running watches will show you a respiration rate. It sits in the same list as heart rate, cadence and pace, formatted the same way, updated on the same screen. Everything about the presentation suggests it belongs there.
It does not. Your watch is not measuring your breathing. It is guessing at it, from a signal that gets worse exactly when your training gets interesting.
Here is what that looks like on a real session.
What you are looking at
This is one run. A 15.4 km session lasting 93 minutes, built around 20 × 400 m at roughly 3:15 per kilometre with 40 seconds of recovery between reps. Average heart rate for the session was 164 bpm and it peaked at 196.
Both lines were recorded at the same time, on the same athlete, into the same file. The watch was a Forerunner 255. The Tymewear VitalPro was paired to it over Bluetooth as the heart rate source, so the beat-to-beat data underneath both numbers traces back to the same strap on the same ribcage.
The only thing that differs between the red line and the blue line is how each one arrives at a breathing rate.
The red line is Garmin's respiration rate, inferred from the timing between heartbeats. The blue line is the Tymewear breathing rate, measured directly from the expansion and contraction of the ribcage.
They do not agree. They barely relate to each other at all.
Across the 4,723 seconds where both signals are present, the two disagree by an average of 15 breaths per minute. The correlation between them is r = 0.17, which is close enough to nothing that you could not use one to predict the other. Tymewear spends 54% of the run above 50 breaths per minute. Garmin spends 0.2% of the run there, and never once reports a value above 51.
That last number is worth sitting with. During twenty hard 400s at 3:15 pace, with heart rate touching 196, this athlete's watch never registered a breathing rate above 51 breaths per minute. The actual peak was 75.
How watches arrive at a respiration rate
The method is called ECG-derived respiration, and it is genuinely clever. It rests on a real physiological phenomenon called respiratory sinus arrhythmia, or RSA.
When you inhale, your heart rate briefly speeds up. When you exhale, it briefly slows down. The interval between consecutive beats therefore rises and falls in time with your breathing. If you record those beat-to-beat intervals accurately enough, you can extract the rhythm hiding inside them and report it as a breathing rate.
This is what Garmin does. The Firstbeat algorithms that sit inside Garmin devices describe the input plainly in their own technical documentation: "beat-by-beat R-R derived respiration rate." No sensor is watching your chest. The algorithm is reading the spacing of your heartbeats and inferring how often you must have breathed to produce that pattern.
At rest and during sleep, this works reasonably well. You are still, the beat detection is clean, and the respiratory modulation of heart rate is large and easy to see. Checked against overnight sleep studies, wrist-derived respiration rates land within roughly a breath or two per minute, which is the use case these algorithms were built and tuned for.
Then you start running, and three things go wrong at once.
Problem one: the signal shrinks when you need it
RSA exists because of the vagus nerve. Parasympathetic activity modulates the sinus node beat by beat, and breathing modulates that modulation.
The moment you begin exercising, vagal activity withdraws. Casadei and colleagues documented this directly: the absolute power of the high-frequency spectral component of heart rate variability, which is the RSA signal, falls at the onset of exercise and keeps falling as the work rate climbs. Hatfield and colleagues found the same thing across trained and untrained men, with RSA decreasing systematically through progressive exercise stages.
The relationship is not perfectly monotonic. Blain and colleagues showed that RSA amplitude declines up to roughly 62% of VO₂peak and then partially recovers at the highest intensities, driven not by nerve traffic but by the mechanical effect of large lung inflations tugging on the heart. That is a real effect, and it is part of why the estimate does not collapse entirely.
But it changes what the algorithm is reading. At low intensity it is tracking a neural rhythm. At high intensity it is tracking a mechanical artefact of a rhythm that has largely gone quiet. The thing being measured is not stable across the intensity range you actually train in.
Problem two: there are not enough heartbeats
This one is arithmetic, and it is unforgiving.
An RR-interval series is sampled once per heartbeat. That is the entire data rate available. To reconstruct a rhythm, you need several samples per cycle of that rhythm.
At the end of a hard 400, this athlete's heart rate was around 190 bpm and their breathing rate was around 62 breaths per minute. That is roughly three heartbeats per breath. Three samples per cycle is not enough to faithfully describe a waveform, and it gets worse as breathing rate climbs toward heart rate.
So the very moment breathing becomes most informative, when an athlete is deep into a threshold session and their ventilation is telling you something their pace cannot, is the moment the method runs out of resolution.
Problem three: the algorithm hides its own uncertainty
Faced with a noisy, low-resolution estimate, the sensible engineering response is to smooth heavily and constrain the output to a plausible range. That is clearly what is happening here, and you can see it in the shape of the red line.
In this file, Garmin's respiration rate is unchanged from one second to the next 80% of the time. The Tymewear signal is unchanged 46% of the time. There are stretches in the chart, several minutes long, where the red line is perfectly flat, holding a single value while the athlete's actual breathing moves around underneath it.
This is the most misleading part. Heavy smoothing makes the output look confident. A clean, stable line reads as a good measurement. In this case the stability is not the signal being steady, it is the algorithm having very little to say and saying it slowly.
What it costs you during a workout
Averages over a whole run understate the problem. Zoom in on four consecutive reps.
The blue line does what breathing does. It climbs through each rep, peaks as the athlete finishes, drops through the recovery jog, then climbs again. Across all twenty reps it swings an average of 24 breaths per minute between peak and trough, and it moves in lockstep with the heart rate trace underneath.
The red line swings 7 breaths per minute, and it lags. Over the twenty reps, the average peak Tymewear records during a rep is 62 breaths per minute. Garmin's average rep peak is 42.
If you were using respiration rate to judge whether an athlete was holding the session together or starting to come apart, one of these signals would tell you and the other would not.
This matters because breathing is not a vanity metric. Ventilation is the field-accessible window into the ventilatory thresholds, the individual boundaries between intensity domains that determine whether a session builds aerobic capacity or just accumulates fatigue. Thresholds are identified from inflections in breathing, and an inflection cannot be found in a signal that has been flattened into a straight line.
What accurate actually looks like
The honest comparison for a breathing measurement is not another wearable. It is a metabolic cart, which is what exercise physiology labs have used for decades.
We ran that comparison and published it. Twenty-six male and female athletes completed ramped VO₂max tests wearing the Tymewear VitalPro while simultaneously being measured by a Cosmed K5, the gold-standard portable metabolic cart.
For breathing rate, the VitalPro agreed with the Cosmed to within a mean absolute error of 1.2 breaths per minute. For minute ventilation, the pooled correlation across all athletes was r = 0.973 (r² = 0.947).
The full methods, per-athlete results and Bland-Altman analysis are in the validation study.
That is the gap in a sentence. Against a laboratory metabolic cart, the difference is around one breath per minute. Against a watch estimating from heartbeat timing, on the run above, the difference is fifteen.
The reason is not that Garmin's engineers are careless. It is that they are solving a much harder problem. Recovering breathing from the spacing of heartbeats is an inference, and inferences degrade when the underlying signal weakens. Measuring the circumference of a ribcage as it expands is a direct observation, and it does not care how hard you are working or where your vagal tone has gone.
What to do with this
Do not throw away the respiration number on your watch. Use it where it works.
For sleep and resting trends, wrist-derived respiration is reliable enough to act on. A rising overnight breathing rate across several days is a genuine signal, and that is the use case these algorithms were designed and validated for.
During training, treat it as a rough trend indicator and nothing more. Do not use it to set intensity, do not use it to find a threshold, and do not read anything into a value that has not changed in four minutes.
If you want breathing data that holds up during the sessions that actually drive adaptation, you need something that measures breathing rather than inferring it. That is what the VitalPro does.
Final thoughts
The chart at the top of this article is not a story about one watch having a bad day. Both lines are working exactly as designed. One is reporting a direct measurement of chest expansion. The other is reporting the output of an algorithm doing its best with heartbeat timing, in conditions where heartbeat timing has stopped carrying much respiratory information.
The problem is that they are displayed identically, in the same units, next to each other in the same app. Nothing on the screen tells you that one of these numbers is a measurement and the other is an estimate that quietly stopped tracking reality somewhere around your second interval.
Breathing is the most responsive signal an endurance athlete has. It moves faster than heart rate, it is not blunted by cardiac drift, and it marks the intensity boundaries that matter. It is worth measuring properly.
References
Blain, G., Meste, O., & Bermon, S. (2005). Influences of breathing patterns on respiratory sinus arrhythmia in humans during exercise. American Journal of Physiology-Heart and Circulatory Physiology, 288(2), H887-H895. https://doi.org/10.1152/ajpheart.00767.2004
Casadei, B., Cochrane, S., Johnston, J., Conway, J., & Sleight, P. (1995). Pitfalls in the interpretation of spectral analysis of the heart rate variability during exercise in humans. Acta Physiologica Scandinavica, 153(2), 125-131. https://doi.org/10.1111/j.1748-1716.1995.tb09843.x
Firstbeat Technologies. VO2 estimation method based on heart rate measurement. White paper. https://firstbeat.com/wp-content/uploads/2015/10/white_paper_vo2_estimation.pdf
Hatfield, B. D., Spalding, T. W., Santa Maria, D. L., Porges, S. W., Potts, J. T., Byrne, E. A., Brody, E. B., & Mahon, A. D. (1998). Respiratory sinus arrhythmia during exercise in aerobically trained and untrained men. Medicine & Science in Sports & Exercise, 30(2), 206-214. https://doi.org/10.1097/00005768-199802000-00006
Tymewear. Validation of the Tymewear VitalPro chest strap against the Cosmed K5 metabolic cart for ventilation monitoring. https://www.tymewear.com/blogs/validation-studies/tymewear-internal-validation-study-of-breathing-metrics