Artifacts and extra beats: why HRV readings go wrong

HRV is worked out from the gaps between normal heartbeats. Two kinds of bad gap spoil it: artifacts artifacts Artifact An error in a heartbeat recording, such as a missed, extra or mistimed beat, often caused by movement or poor sensor contact. Artifacts can badly distort HRV numbers. More about Artifact , which are errors in the measurement, and extra beats, which are real heartbeats that start in the wrong place. Both are common: people move and talk while they measure, and healthy hearts produce the odd early beat. This page shows what each one looks like, how much a single bad gap moves the number, how software finds and fixes it, and why a device sometimes refuses to give a reading at all.

Two different things

An artifact is a measurement error. The heart beat normally, but the device got the timing wrong. Typical causes are movement, poor skin contact, a dry or loose chest strap, a detector that missed a beat or counted one twice, and Bluetooth packets lost on the way to the phone. Research guidelines call these technical artifacts: they come from poorly attached electrodes or from movement (Laborde and colleagues, 2017).

An ectopic beat ectopic 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 is a real heartbeat that starts in the wrong place. A normal beat starts in the sinus node sinus node Sinus node The heart's natural pacemaker, a small patch of cells that starts every normal heartbeat. Left alone it fires about 100 times a minute; the vagus nerve holds it lower at rest. More about Sinus node , the heart’s own pacemaker. An ectopic beat, also called an extrasystole or a premature contraction, starts somewhere else and usually comes early. If it starts in the upper chambers it is a premature atrial contraction (PAC); in the lower chambers, a premature ventricular contraction (PVC). These beats are common in healthy people. In a study of 2,048 healthy adults aged 25 to 41 who wore a 24-hour ECG 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 , 69% had at least one PVC, and the typical person had two (von Rotz and colleagues, 2017). The NHS describes the skipped or extra beats people sometimes feel as common and usually harmless.

Neither one belongs in an HRV number. HRV is meant to describe how the nervous system speeds up and slows down the sinus node, so the formulas use only normal-to-normal (NN) intervals. Beats the sinus node did not start are called physiological artifacts, and they have to be dealt with as well (Quigley and colleagues, 2024). Laborde and colleagues list ectopic beats, atrial fibrillation, sighs and coughs among them.

What each problem looks like in the tachogram

A tachogram tachogram Tachogram A line chart of heartbeat intervals in the order they happened, one point per beat, in milliseconds. It shows HRV as a wavy line instead of a single number. More about Tachogram is the list of gaps between beats drawn as dots, one dot per gap, with longer gaps higher up. At rest the dots rise and fall gently with your breathing. Each problem breaks that pattern in its own way, and the shape is what software looks for (Lipponen and Tarvainen, 2019).

normal gap wrong gap what the heart did

• Missed beat Measurement error One gap about twice as long: the detector skipped a beat.
• False beat Measurement error One gap split into two short ones: something that was not a beat was counted.
• Ectopic beat Real heartbeat A short gap, then a long pause: an early beat that started outside the normal pacemaker.
• Motion noise Measurement error A run of scattered values, too long and too short: movement swamped the signal.
• Atrial fibrillation No normal rhythm Irregular from start to end. There is no normal rhythm left, so HRV does not mean what it usually does.

5 small charts of the time between heartbeats, one per problem, each drawn against a faint line of a normal resting rhythm. Missed beat: one value about twice as high as its neighbours, and one dot fewer. False beat: one normal value replaced by two values about half as high. Ectopic beat: one value well below the line followed at once by one well above it, a short gap and then a long pause. Motion noise: a run of values scattered far above and below the line. Atrial fibrillation: every value is different and there is no steady line at all.

• A missed beat joins two gaps into one about twice as long. One dot jumps up, and there is one dot fewer.
• An extra or false detection splits one gap into two short ones. Two dots drop down.
• An ectopic beat comes early, so its gap is short, and the heart then pauses before the next normal beat, so the gap after it is long. The short-long pair is the signature. On a watch or ring the early beat often pushes out too little blood to be seen, so the pair turns into one long gap that looks like a missed beat. PPG explained shows why.
• Motion noise scatters a run of dots far above and below the line. Movement makes the sensor’s signal swing more than the pulse does, so the detector finds beats in the wrong places, or misses them.
• Atrial fibrillation is different. The gaps are irregular from start to end, because the sinus node is no longer setting the pace. There is no normal stretch to compare against, so cleaning cannot help: HRV numbers from such a recording do not mean what they usually mean. Peltola’s review lists atrial fibrillation as a physiological artifact that can last minutes to hours, and reports the advice to remove whole disturbed segments, or the whole recording, when bad beats are frequent (Peltola, 2012).

Why they happen in everyday life

Most bad readings have an ordinary cause.

Moving. Walking, typing, gesturing and shifting in a chair all move the sensor against the skin. Optical sensors in watches and rings suffer most. In a study of consumer and research wearables, heart rate error during activity was on average 30% higher than at rest (Bent and colleagues, 2020), and a review of pulse-based HRV found that physical activity made its agreement with the ECG much worse, often unacceptably so (Schäfer and Vagedes, 2013).

A loose fit. A watch or ring that slides around loses steady contact with the skin, and a chest strap that slips loses its electrical contact. Both lead to missed and false beats. How chest straps measure HRV covers dry electrodes, static and lost Bluetooth packets.

Cold hands. Cold narrows the small vessels in the skin. In 16 people, cooling and warming the skin changed the PPG 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 signal (Jeong and colleagues, 2014). A weaker pulse is easier to miss.

Talking, coughing and sighing. Each one changes your breathing and moves your chest for a few seconds. Laborde and colleagues list sighs and coughs as physiological artifacts, and ask people not to speak or move during a resting recording.

Sleep position. Lying on the arm that wears the watch can press or twist the sensor, and pressing a fingertip hard onto a phone camera squeezes the blood out of it. Both change the pulse the sensor sees. Fingertip HRV explains the pressure problem.

Caffeine and alcohol, for extra beats. In a randomised trial, 100 adults wore a continuous ECG for two weeks while text messages told them, at random, which days to drink coffee. On coffee days they had on average 154 PVCs a day, against 102 on days without caffeine; the number of PACs did not differ significantly (Marcus and colleagues, 2023). In 202 young adults followed with an ECG through a planned bout of heavy drinking, PVCs were more frequent during the drinking hours, while early atrial beats were more frequent in the comparison periods (Brunner and colleagues, 2024). The NHS lists alcohol, caffeine, nicotine, stress, anxiety and lack of sleep as common causes of palpitations.

How much one bad beat changes the numbers

Here is a short series of 12 resting gaps, in milliseconds:

950, 980, 1010, 1000, 970, 940, 930, 960, 990, 1010, 990, 960

Now make the seventh beat come early, at 60% of its normal gap. The gap of 930 ms becomes 558 ms, and the next gap, 960 ms, carries the rest: 1332 ms. The total time is the same. The table shows RMSSD 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 and SDNN 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 , worked out with this site’s own formulas, for the clean series, with the ectopic beat left in, and after each way of cleaning it. The last two rows do the same for a missed beat instead.

One early beat multiplied RMSSD about 11 times and SDNN about 6 times. RMSSD reacts most because it is built from the difference between each gap and the next. In the clean series those differences are 10 to 30 ms. The ectopic pair adds three jumps of hundreds of milliseconds, and RMSSD squares each jump before averaging, so the big ones swamp everything else. SDNN measures how far each gap sits from the average gap. The pair moves two gaps far from the average, but the average itself barely moves, so SDNN rises less. Every way of cleaning brings both numbers back close to the clean values, and none brings them back exactly.

Twelve gaps is a very short reading, which is why the effect is so large. In the five-minute teaching recording used in the signal chain lesson, 341 gaps long, the same early beat touches 0.6% of the gaps and still raises RMSSD from 33.4 to 57.9 ms, up 73%. SDNN rises 36%, HF power HF power HF power The variance of the heartbeat intervals that falls in the 0.15 to 0.40 Hz band, in ms²: the range of normal breathing. It mainly reflects vagal activity when breathing is 9 to 24 breaths a minute. More about HF power 38% and LF power LF power LF power The variance of the heartbeat intervals that falls in the 0.04 to 0.15 Hz band, in ms²: rhythms repeating every 7 to 25 seconds. Once read as sympathetic activity, which is now disputed. More about LF power 30%. Interpolating the pair brings RMSSD back to 33.4 ms. RMSSD, built from beat-to-beat differences, reacts most. The spectrum reacts too, because one sharp jump spreads false power across all the bands, including HF.

The research agrees. Modelling missed and false R-wave detections, Berntson and Stowell (1998) found that a single artifact in a 128-second series can add more false variance than the person’s real resting variability, in all the commonly analysed frequency bands and in time-domain measures too. The Kubios HRV user’s guide shows that just two artifacts in a five-minute segment significantly change SDNN and RMSSD. And in short recordings, editing even under 5% of the intervals can shift HF, LF, pNN50 pNN50 pNN50 The percentage of neighbouring NN intervals that differ by more than 50 milliseconds. It tracks fast, vagally driven changes, like RMSSD, but as a share rather than a size. More about pNN50 and RMSSD (Peltola, 2012, citing Salo and colleagues, 2001).

How correction works

Correction has two steps: find the bad gaps, then decide what to do with each one.

Finding the bad gaps

Software uses one or more of these rules, roughly from the simplest to the most careful.

• Range filter. Drop any gap outside a plausible range, such as 300 to 2,000 ms (200 to 30 beats a minute). Piskorski and Guzik (2005) call this the square filter. It catches gross errors but misses an ectopic beat whose gaps are both in range.
• Percentage of the previous gap. Flag a gap that differs from the one before by more than a set share, such as 20%. Piskorski and Guzik’s quotient filter removes a beat when the ratio of neighbouring gaps falls outside 0.8 to 1.2. They describe it as the most aggressive of their filters, yet the one most likely to leave bad beats in: when several similar bad gaps come in a row, some of them look normal to a rule that only compares neighbours.
• Distance from a local median. Compare each gap with the median of the gaps around it, and flag it if it is too far away. Kubios HRV compares each gap with a local average and offers thresholds from 0.45 s (“very low”) to 0.05 s (“very strong”), set for 60 beats a minute and scaled to the heart rate. This site’s HRV calculator uses the range rule above, then drops any gap more than 25% from the median of the five gaps on each side.
• The person’s own spread. Berntson and colleagues (1990) set the cut-off from each person’s own distribution of beat-to-beat differences, using percentile-based measures that bad values distort less than an ordinary average. Their rule flagged all 1,494 simulated and real artifacts in their test data, with a false alarm rate under 0.3%.
• Detect and classify. Lipponen and Tarvainen (2019) combined thresholds that adapt over time with a classifier that recognises the shape each problem makes. It found all simulated missed and extra beats and 96.96% of real atrial and ventricular ectopic beats, with a specificity of 99.94%. After correction, the mean error in HRV measures was under 2% for missed and extra beats. The automatic correction in Kubios HRV Premium also works on the differences between successive gaps, and its user’s guide cites this paper as its validation.

The threshold matters. Alcantara and colleagues (2020) processed the same recordings from 312 people with each Kubios setting and found that the results differed significantly; between the two extreme settings, mean SDNN differed by 39 ms in the children, 27 ms in the young adults and 8 ms in the middle-aged adults. A strict rule removes real beats, and a loose one lets bad ones through. That is why research guidelines recommend checking automatic corrections by eye (Quigley and colleagues, 2024).

Fixing a flagged gap

There are three basic options.

• Delete it. The bad gaps are dropped and the series closes up. It is simple and adds nothing invented. But the record gets shorter, and two gaps that were never neighbours now sit side by side, which gives the spectrum a false step. Salo and colleagues (2001) did not recommend deletion for RMSSD and pNN50, and in 24-hour spectra only 5% of the data could be deleted without a significant error, while interpolation performed well.
• Replace it by interpolation. The bad gaps are replaced with values made from their neighbours: their mean, a straight line between them (linear) or a smooth curve through the surrounding points (cubic spline). The timing and the number of beats are kept, which spectra need. The cost is smoothing: the filled stretch is calmer than real beats, and linear and spline fills have been reported to raise low-frequency power (Peltola, 2012).
• Insert a beat. For a missed beat, the long gap is split back into two. This is right only when the long gap really hides one undetected beat. An ectopic pause must never be split, because no beat is missing.

1. Delete Ectopic beat RMSSD 29.6 ms The short and long intervals are dropped and the gap is closed: 10 intervals remain. Cost: The record is shorter, and two beats that were never neighbours now sit side by side. Spectra see a false step.
2. Interpolate Ectopic beat RMSSD 23.8 ms The two intervals are replaced by points on a straight line between the good neighbours: 957 and 973 ms. Cost: The timing and the beat count are kept, but the filled stretch is smoother than real beats.
3. Split Missed beat RMSSD 27.1 ms The doubled interval, 1970 ms, is cut into two: 985 and 985 ms. A beat is put back. Cost: Right only when the long gap really hides one beat. An ectopic pause must never be split.

wrong interval value put in where the gap was closed

Three panels, each a small chart of twelve beat-to-beat intervals near 1,000 ms. Delete: an ectopic beat made one interval of 558 ms and one of 1332 ms; both are removed and the line closes up, two dots shorter, RMSSD 29.6 ms. Interpolate: the same two intervals are replaced by 957 and 973 ms on a straight line between their neighbours, RMSSD 23.8 ms. Split: a missed beat left one interval of 1970 ms; it is cut into two of 985 ms, RMSSD 27.1 ms. The clean series has an RMSSD of 25.8 ms.

For short readings the choice matters less than the amount: with two gaps fixed in twelve, every method in the table lands within a few milliseconds of the clean value, while leaving the beat in makes RMSSD more than ten times too high.

Go deeper Why an ectopic beat cannot always be repaired

Interpolation assumes that the rhythm around an ectopic beat is otherwise normal. Quigley and colleagues (2024) point out that an ectopic beat can reset or alter the ongoing rhythm, so summing or smoothing the intervals around it cannot always restore what a normal beat would have done. ECG and Holter HRV describes how the effect can spill into the beats that follow, one reason clinical analysis looks at the ECG trace and not only the list of intervals. When ectopic beats are frequent, the safer choice is to leave that stretch out, or the whole recording (Peltola, 2012), and to report how much was removed (Quigley and colleagues, 2024).

Plant a missed, extra or ectopic beat in the five-minute teaching recording below, then pick a correction method and compare RMSSD, SDNN and the spectrum with the clean recording.

Clean one bad beat

The teaching recording, 341 intervals

Gap before each beat, first 60 seconds Flagged

Clean recording Missed beat Extra beat Ectopic beat

Leave it in Delete Mean of neighbours Linear interpolation Cubic spline

Flagged: 1 of 341 intervals

Measure	Clean	Bad beat	Fixed
RMSSD (ms)	33.4	70.2	70.2
SDNN (ms)	28.7	51.7	51.7
pNN50 (%)	8.2	8.8	8.8
LF power (ms²)	120	1241	1241
HF power (ms²)	635	2314	2314

A missed beat makes one interval about twice as long as its neighbours. Left in, it lifts RMSSD from 33.4 to 70.2 ms, 2.1 times the true value, while SDNN moves less.

The buttons work when JavaScript is on.

A line of the gap between beats over the first 60 seconds of the teaching recording, with the planted artifact as a spike, a circle on each flagged interval and the corrected line drawn over it. A table gives RMSSD, SDNN, pNN50, LF power and HF power for the clean recording, with the artifact left in, and after correction, plus how many intervals were flagged. One group of buttons plants a missed beat, an extra beat or an ectopic beat, or restores the clean recording; another chooses the correction method: leave it in, delete, mean of neighbours, linear interpolation or cubic spline. A missed beat left in roughly doubles RMSSD; every method brings it back within a few percent.

To see artifacts across a whole day, open the HRV workbench and pick the “Recording problems” day: early beats from the heart, a loose strap, and movement noise on walks and a run. To check one of your own recordings, paste its intervals into the HRV calculator: it lists every beat it removed, with the time and the reason.

Quality limits, and why devices skip readings

How much correction is too much? There is no single agreed figure, but the guidance points the same way:

• Kubios HRV recommends keeping the number of corrected beats under 5% and, by default, warns about low-quality data above that share.
• Most researchers require at least 80% normal intervals, and published studies have edited anywhere from 1% to 30% of beats (Peltola, 2012).
• The 2024 psychophysiology guidelines encourage researchers to report the share of beats deleted or corrected, and to test whether editing changed their results whenever it passed 1% (Quigley and colleagues, 2024).

Devices apply the same logic, which explains several things people notice.

Why watches and rings measure HRV during sleep or stillness. An optical sensor agrees with the ECG at rest and much less during movement (Schäfer and Vagedes, 2013). So most wearables take their HRV from quiet stretches of the night, or from a reading you start while sitting still.

Why a night sometimes shows no HRV, or a gap. If you slept restlessly, wore the band loose, or your hands were cold, the device may not find enough stretches of clean beats. Makers rarely publish their exact rules, but a good device leaves a gap instead of reporting a number built on guesses.

Why a strap or app rejects a reading. This site’s chest-strap test refuses to show a result when fewer than 40 clean beats arrived, or when more than 10% of the gaps had to be removed. It then says why, for example “Too many beats looked wrong (missed or extra beats)”, and how to try again. The phone camera test allows up to 20%, because camera pulses are noisier, but it also demands a clean pulse wave.

To get readings that pass:

1. Sit or lie still, and stay silent for the whole reading.
2. Breathe normally.
3. Wet a chest strap’s electrodes and wear it snug, just below the chest muscles.
4. Wear a watch or ring snugly, and warm cold hands first.
5. If a reading is rejected, wait a minute and take it again. A rejection usually reflects how the reading was taken.
6. Judge your HRV by its trend over days and weeks. One odd value is more often a measurement problem than a change in you.

When extra beats matter

An HRV app or wearable cannot tell what kind of extra beat you had or whether it matters. A strap sends only the beat times, not the ECG trace, and a watch often misses extra beats entirely. Frequent extra beats can also make HRV look suddenly high, as why is my HRV suddenly high? explains.

Extra beats that you feel as a skipped, fluttering or thumping heartbeat are common and usually harmless (NHS). See a doctor if they keep coming back or happen more often, if they last longer than a few minutes, or if you have a heart condition or a family history of heart problems. Get urgent help if palpitations do not go away or come with chest pain, shortness of breath, feeling faint or fainting. The page can HRV detect heart problems? gives the full list. These steps apply whatever your HRV shows.