Can wearable ultra-wideband radar monitor breathing during motion?

A radar on the sternum reads lung motion

Respiro is a chest-strap device that places a consumer-grade ultra-wideband radar module over the sternum. Unlike non-contact radar setups that sit in a room and watch chest-wall movement from a distance, this system senses the body itself: the paper explains that inflated and deflated lungs present different dielectric properties, so more signal reflects at the muscle-lung interface when the lung is inflated. The radar therefore has an in-body signal to work with, not only a surface displacement.

The device records complex channel impulse responses, the radar's estimate of how reflections arrive over time. Each windowed response is treated as one row of a measurement matrix, and the respiratory rate is extracted from that matrix rather than from a single hand-picked reflection. The authors describe the pipeline as computing the magnitude of each complex response, calibrating the responses so they share the same dominant-path energy, aligning them, then computing an optimal weighting that maximizes energy in the relevant frequency bands before taking the dominant frequency as the respiratory rate.

Twelve participants, six activity conditions

The evaluation enrolled 12 healthy adults: 3 female and 9 male, aged 21 to 29 with a mean age of 25. Each participant completed a fixed sequence twice: two minutes sitting, two minutes standing still, two minutes walking on a treadmill, two minutes running on a treadmill, a break, one minute cycling, a break, one minute of squats, and one minute standing still. Walking started at 5 km/h and running at 10 km/h, with speeds adjusted per participant after a treadmill test.

Ground truth came from a spirometer, the reference the paper treats as a gold standard for respiration monitoring. The radar and accelerometer data were sampled at 32 Hz, the spirometer analog signal at 50 Hz, and a respiration belt at its default 2 kHz. The spirometer was removed for the second iteration of the sequence, so not every activity block carried the same reference recording. The study also fitted an accelerometer to the same device so the authors could compare against two published accelerometer-based respiratory-rate systems.

The 1.11 bpm average hides condition differences

Across all sessions and participants, the UWB method's respiratory-rate error averaged 1.11 breaths per minute, and 71% of all measurements fell below a 1 bpm deviation. Those two numbers answer different questions and should not be swapped: the average describes typical error magnitude across the pooled data, while the 71% figure describes how often an individual measurement crossed the 1 bpm threshold used to classify a successful trace. A method can have a modest average and still miss that threshold often.

The paper's error distributions compare the UWB approach against the accelerometer-based methods of Rahman et al. and Bates et al. across sitting, standing, walking, running, cycling, and squats. The authors report that their UWB-based approach has a higher success rate than those accelerometer systems because it is less prone to motion artifacts. That comparison is against IMU-based baselines only; the paper notes that most contact-based alternatives require extra hardware such as custom antennas, which would introduce confounding variables, so no such contact-based system was run as a baseline here.

Walking cadence collides with breathing rates

The most important failure mode is mechanical, not algorithmic. The authors state that movement of the sensor encasement can still produce motion artifacts, and this is particularly evident during walking, where the cadence range overlaps with possible respiratory rates. When a wearer's step rate falls in the same frequency band as breathing, a method that extracts the dominant frequency in that band has a genuine ambiguity to resolve.

This is also why the paper's claim is narrower than 'motion-immune.' The device is described as less prone to motion artifacts than accelerometer-based methods, not free of them, and the walking condition is where that distinction matters most for a reader judging whether the approach would hold up in free-living use. The paper's own framing is that the results underscore the potential of consumer-grade UWB radar in body-worn devices, which is a statement about promise under tested conditions rather than demonstrated performance outside them.

Sitting accuracy rivals a two-contact system

The most direct accuracy comparison in the paper is in the seated condition. The closest prior work, by Culjak et al., reported a mean absolute percentage error of 8.75% using a transmissive UWB signal through the chest with participants sitting in a chair. In the same seated setting, Respiro observed a mean absolute percentage error of 6.45%.

The authors read this as evidence that a single-point-of-contact wearable can reach accuracy comparable to a two-point-of-contact wearable system. The comparison is specific: same posture, percentage-error metric, and a prior system that used two contact points. It does not establish that the single-contact design matches two-contact designs during walking or running, because the cited prior figure covers seated participants only.

Young healthy adults are the evidence base

All participants were between 21 and 29 years old, and the sample was 12 people. The authors list both the small sample size and the limited age diversity as limitations, and they note that future work should include more participants to investigate participant-specific effects. The method is described as adapting to different body types by targeting respiration-related frequency bands, but that adaptation was not tested across a wide range of body types or ages in this study.

The activity protocol was also controlled: treadmill speeds were set and adjusted by an experimenter, activities ran for fixed durations, and the sensor was positioned by an experimenter on the sternum with a 1-2 cm gap maintained from the respiration belt to avoid contact artifacts. Those conditions describe a lab study of motion, not unrestricted daily living, and the reported error figures should be read within them.

Match the activity to the error threshold

For a reader deciding whether this sensing approach fits a use case, the paper supports a scenario-based reading rather than a single verdict. If the requirement is respiratory-rate tracking within roughly 1 bpm during seated or stationary monitoring, the seated mean absolute percentage error of 6.45% and the pooled 71% success share are the relevant evidence. If the requirement is reliable tracking during walking, the cadence-overlap problem is the specific obstacle the paper identifies, and the pooled average does not resolve it for that condition.

The paper's distinctive contribution is the combination of a single contact point, consumer-grade off-the-shelf radar modules, and evaluation on moving participants, which the authors describe as the first such study of a wearable UWB-based respiratory-rate system. The practical priority that follows is to treat walking as the discriminating test case: it is where accelerometer baselines reportedly degrade, where this method still shows motion artifacts, and where the overlap between step cadence and breathing rate determines whether the measurement can be trusted.

Sources used in this answer