%201-1.avif)
Every insight starts as a signal.
Circular Ring 2 samples your physiology around the clock and turns raw photons, voltage and motion into metrics a clinician recognises. This is how that pipeline is built, how it holds up against reference-standard equipment, and who checks our work.
Since you opened this page
Participants across our internal research programmes, spanning ages 19 to 74, all skin tones and both sexes.
Hours of labelled physiological signal used to train and stress-test our models.
Independent research sites — university hospitals, sleep laboratories and sports institutes.
Peer-reviewed papers and preprints co-authored or supported by the Circular research team.
Four sensors, one continuous picture.
The finger is the most vascular place a wearable can sit. Arterial pulsation there is roughly a hundred times stronger than at the wrist, which is why a ring can resolve beat-to-beat detail a watch has to infer.
Photoplethysmography
Green, red and infrared LEDs sample blood volume at 25 Hz. Pulse rate, beat-to-beat interval, HRV, blood oxygen and respiratory rate are all derived from this one optical signal.
3 wavelengths · 25 Hz
Single-lead ECG
Two dry electrodes record the heart’s electrical activity at 510 Hz. A 40-second recording is enough to read rhythm morphology and screen for atrial fibrillation.
510 Hz · 40 s recording
Skin temperature
A thermistor in contact with the finger resolves 0.01 °C every second. Deviation from your own nightly baseline is what matters — not an absolute reading.
0.01 °C · 1 Hz
3-axis motion
A 3-axis accelerometer at 25 Hz classifies activity type, cadence and intensity, and separates genuine stillness from quiet wakefulness at night.
Accelerometer · 25 Hz
What 510 hertz looks like.
Sampling rate sets the ceiling on what can be seen. A rhythm irregularity lasting a fifth of a second is invisible to a sensor that only looks a few times a second, and no model recovers what was never recorded.
Each track runs at its true relative speed. Everything is timestamped against one on-ring clock, which is what allows an optical beat to be lined up with an electrical one.
The finger is not a smaller wrist.
Optical heart measurement depends on how much arterial blood sits between the light source and the sensor. On the palmar side of the finger there is a great deal of it, close to the surface, in tissue that barely moves when you are still.
Denser arterial bed
Digital arteries sit close to the surface with no intervening tendon sheath, so the pulsatile component of the optical signal is one to two orders of magnitude larger.
Less motion contamination
A ring is clamped to bone at a fixed distance from the skin. A watch slides and rotates, and every millimetre of movement is noise a model has to guess through.
Shorter optical path
Less tissue between emitter and detector means more photons return, which is what makes a 510 Hz electrical channel and a clean beat-to-beat interval possible at all.
Amplitude figures are indicative of the physiology rather than a benchmark of any particular device.
All of it fits in 4.5 grams of titanium.
There is no room for a compromise you can hide. Every component earns its volume, and the optical bench sits where the arterial signal is strongest.
Outer shell
Grade-5 titanium with a diamond-like carbon coating, 2.6 mm profile. Hypoallergenic and dimensionally stable across the temperature range the finger actually sees.
Optical bench
Three LEDs and two photodiodes behind a sapphire window, positioned on the palmar side where arterial pulsation is strongest.
Electrode ring
Two dry contacts — one on the inner surface, one on the outer shell. Touching the outside with the opposite hand closes the circuit for an ECG.
Core
Microcontroller, inertial unit, thermistor and a 24 mAh cell. Up to seven days between charges with continuous sampling on.
Five stages between the sensor and the sentence you read.
Every number in the app can be traced back through this chain. Nothing is smoothed into meaninglessness, and nothing is shown when the underlying signal quality does not support it.
Signal-quality gating means a night with poor contact produces a shorter report rather than a confident guess.
We throw data away.
When you roll over, adjust the ring or wash your hands, the optical signal is destroyed for a few seconds. The tempting move is to smooth it into something presentable. We cut it out and record a gap instead.
A night with heavy exclusion produces a shorter report and a note explaining why, rather than a confident number built on nothing.
The protocol matters more than the number.
Three rules govern every validation study we run. They are the reason our figures are lower than some of the numbers you will see quoted elsewhere.
Where a cohort is too small to support a claim, we publish the cohort size next to the claim and let you judge it.
What a validation plot actually looks like.
A correlation coefficient can hide a systematic offset, so it should never travel alone. Here is the raw scatter against the reference instrument, the line of identity, and the Bland–Altman figures that belong beside it.
HRV, RMSSD against lead-II ECG
Each point is one participant-night: ring on the vertical axis, simultaneous clinical ECG on the horizontal. Points sitting above the dashed line mean the ring read higher than the reference, below it means lower. What matters is that they scatter evenly around the line rather than drifting to one side, because a drift is a bias you would carry into every reading.
Bland–Altman plots and per-cohort residuals for every metric in the table above are included in the validation pack.
What each signal actually tells you.
Pick a metric to see the shape of the raw data behind it and how we read it.
Nightly RMSSD, not a single reading
Heart rate variability is only interpretable against your own history. We compute RMSSD from beat-to-beat intervals during the most stable portion of sleep, then compare it with your trailing 14-night band. A single low night is noise; three consecutive nights below the band is a signal worth acting on.
Overnight oxygen saturation, event by event
Rather than a single nightly average, the ring records the full saturation curve and counts discrete desaturation events of 3% or more. Event frequency, not the mean, is what correlates with disrupted breathing during sleep.
A hypnogram, scored in 30-second epochs
Sleep staging combines pulse-wave morphology, HRV, respiratory rate, temperature and motion in a single model, scored on the same 30-second epochs a sleep technician uses. Agreement with blind-scored polysomnography reaches 87.4% across four classes.
Deviation from your own baseline
Absolute skin temperature is a poor signal; deviation is a good one. The ring builds a 28-day personal baseline, then reports departures from it — the same pattern that tracks cycle phase, altitude adaptation and the days before you notice you are unwell.
“What made the collaboration workable was access to the raw signal. Most consumer devices hand you a score and ask you to trust it. We could check the waveform ourselves.”
Principal investigator, sleep medicine
University hospital sleep laboratory
The literature we build on, and add to.
Work co-authored or supported by our team, alongside the peer-reviewed evidence our methodology rests on.
10 of 10 shown
Preprint
Circular Research, with two university sleep laboratories. 312 nights, blind-scored. Reports 87.4% epoch agreement and per-stage confusion analysis.
Submitted
Circular Research with a hospital arrhythmia unit. 604 recordings compared against cardiologist-read 12-lead ECG.
Sensors
Method paper describing the feature set behind nightly respiratory rate, benchmarked against respiratory-impedance plethysmography.
Transl Androl Urol
Case report using continuous ring-derived sleep and temperature data as part of the clinical picture.
Lancet Digital Health
Ferguson T, Olds T, Curtis R, Blake H, Crozier AJ, Dankiw K, Dumuid D, Kasai D, O’Connor E, Virgara R, Maher C.
JMIR Mhealth Uhealth
Huhn S, Axt M, Gunga HC, Maggioni MA, Munga S, Obor D, Sié A, Boudo V, Bunker A, Sauerborn R, Bärnighausen T, Barteit S.
Int J Med Inform
Choudhury A, Asan O. Volume 156, 104612.
J Clin Sleep Med
Berryhill S, Morton CJ, Dean A, Berryhill A, Provencio-Dean N, Patel SI, Estep L, Combs D, Mashaqi S, Gerald LB, Krishnan JA, Parthasarathy S.
Healthc Inform Res
Pardamean B, Soeparno H, Budiarto A, Mahesworo B, Baurley J.
JMIR Mhealth Uhealth
Kim JW, Ryu B, Cho S, Heo E, Kim Y, Lee J, Jung SY, Yoo S.
What is running right now.
Six active protocols, each with a named reference standard and a defined read-out date. Partners can request the protocol summary for any of them.
Sleep apnoea risk stratification
Ring-derived desaturation events and respiratory variability against attended polysomnography with AHI scoring. Target n = 400, university hospital sleep laboratory.
Nocturnal AFib burden after ablation
Continuous rhythm monitoring in post-ablation patients versus implantable loop recorder. n = 180, hospital arrhythmia unit.
Cycle-phase detection from temperature and HRV
Multi-signal phase estimation validated against urinary LH testing and serum progesterone. n = 260, women’s health clinic.
Early illness detection from multi-signal deviation
Prospective cohort testing whether combined temperature, HRV and respiratory-rate deviation precedes self-reported symptom onset. Target n = 1,000.
Shift-work recovery in an occupational cohort
Sleep debt and autonomic recovery across rotating shift patterns, run with an enterprise partner’s occupational health team. n = 340.
Training load and readiness in endurance athletes
Ring-derived readiness against laboratory measures of autonomic recovery and performance testing. n = 92, national sports institute.
What we will not claim.
The fastest way to lose a clinical partner is to overstate what a sensor on a finger can do. So here is the other side of the ledger, in plain language.
It does not diagnose anything
An irregular rhythm notification is a prompt to see a clinician, not a finding. Diagnosis needs a clinical assessment and, for arrhythmia, a physician-read recording.
It cannot replace a sleep study
Desaturation events and respiratory variability can flag disturbed breathing. Grading sleep apnoea requires attended polysomnography with airflow and effort channels we do not have.
Absolute SpO2 is not a clinical reading
Our oxygen saturation is calibrated for trend and event detection during sleep. It is not a substitute for a medical pulse oximeter when an accurate absolute value matters.
HRV does not compare between people
RMSSD varies several-fold across healthy adults for reasons that have nothing to do with fitness or recovery. Your number is only meaningful against your own history.
Temperature is a deviation, not a fever
We report distance from your own baseline in tenths of a degree. Skin temperature at the finger is not core body temperature and should not be read as one.
No metabolic claims
We do not infer blood glucose, blood pressure or hydration from optical data. The published evidence does not support those claims from a device of this kind, so we do not make them.
Bring us a protocol.
We supply devices, raw signal access under agreement and engineering time for studies that hold us to a reference standard. Clinical, academic and enterprise partners are all welcome.
Typical response within five working days. For press and investor enquiries, use the contact page.