Historically, vital sign monitoring as a routine practice involves wearable devices with contact-based sensors. In various situations, however, with patients affected by severe burns, for example, wearable devices aren’t ideal. That’s how contactless radar technology became a thing in healthcare for continuously monitoring vital signs such as heart and respiration rates.

The concept of relying on radar to detect human cardiopulmonary motion isn’t brand new — it dates back to the 1970s. Since then, the use of this technology has been explored and applied to search-and-rescue operations, adult and neonatal sleep monitoring, and lung cancer radiotherapy.

The initial idea to use radar took advantage of the Doppler effect, or the changes in frequency produced in relation to an observer moving relative to the wave’s source. Doppler radars, by extension, allowed people to detect micro-shifts when a reflected wave bounced off a moving chest wall.

What if we could make it more accurate? What if we could move beyond just measuring respiratory rates? That was the goal of Thanh Trúc Trần, a PhD candidate in the Department of Artificial Intelligence in Biomedical Engineering at the Friedrich-Alexander University of Erlangen-Nürnberg in Erlangen, Germany. That team published results in Scientific Reports.

“Our study builds on the foundational concept that electromagnetic waves can capture submillimeter displacements of the chest wall caused by respiration,” said Trần. “Our work differentiates since we also consider other respiratory metrics such as inspiratory times, expiratory times, and inspiratory-to-expiratory time ratios, which give a more holistic picture of a patient’s health state and, in a next step, can also be important to analyze irregular breathing patterns.”

How They Did It

Trần’s team used a portable continuous-wave radar operating at 24 GHz. “The technology relies on six-port interferometry, which measures phase shifts in reflected electromagnetic waves to capture micrometric displacements on the chest wall caused by respiration and cardiac activity,” said Trần. The most important factors driving the accuracy of this system are high phase sensitivity and submillimeter displacement resolution.

For the radar system to work, it required line-of-sight. So, as participants lay still on a horizontal tilt table and breathed calmly for at least 10 minutes, the radar system was mounted above them, capturing their bodily motions by measuring phase changes between transmitted and received signals.

The phase angle was derived from the in-phase (I) and quadrature (Q) components of the radar raw signal through arctangent demodulation, a means of extracting and quantifying information.

“In ideal conditions, a target (here: chest wall) moving back and forth traces a perfect unit circle in the I/Q complex plane, where the phase angle directly maps to the physical target displacement,” said Trần. That’s not how real-world receiver circuits work, unfortunately. Amplitude imbalances, channel crosstalk, and other issues cause the I/Q trajectory to form a distorted ellipse instead of a perfect circle.

“If you apply arctangent demodulation directly to an uncorrected ellipse, it introduces severe nonlinear phase distortions,” Trần explained. Accurate displacement measurement becomes impossible.

To avoid this, the team developed a means of “ellipse reconstruction” to act as a calibration preprocessing step. “It mathematically estimates the distortion parameters (ellipse center, axis tilt, and amplitude ratio) and projects the data back into a normalized unit circle,” said Trần. “Once corrected, arctangent demodulation can accurately recover the true, linear physical chest motion down to submillimeter scales.”

When the raw radar distance signal is demodulated, it reveals a mixture of respiration, micro-cardiac movements, slow posture drifts, and high-frequency noise. Then, by applying a fourth-order Butterworth bandpass filter — so only certain signals get through — the team was able to isolate physiologic components of interest. The team focused on four discrete, clinically relevant breathing metrics: inspiratory time (TI), expiratory time (TE), respiratory rate, and the inspiratory-to-expiratory time (I:E) ratio.

The Butterworth filter ensured “the actual shape and phase of the respiratory waveform (which is vital for timing metrics like inspiratory and expiratory durations) are preserved without artificial ripple artifacts or distortion,” Trần said.

Defining a ‘Cycle’

Afterward, to detect individual respiratory cycles, the team used peak-valley identification on the radar-derived respiratory waveform and reference signals.

Inhalation was marked by an increasing displacement curve, with the onset identified at the signal valley or local minimum, and the end occurring at the signal peak or local maximum. Exhalation spanned from the peak down to the subsequent valley. A single complete respiratory cycle was defined from one valley to the next.

The radar and reference respiratory signals were not perfectly smooth, however, and sometimes exhibited local extrema that were artifact-related, disrupting the respiratory cycles. To suppress this, physiologically motivated constraints on amplitude prominence, temporal distance, and peak width were applied during extremum detection.

“These constraints were chosen based on expected resting breathing characteristics in healthy adults and refined empirically to ensure robust detection across subjects,” said Trần.

From these segmented cycles, the team extracted exact temporal features for every breath. “While our algorithm detects individual breath cycles, occasional missed cycles or artifacts mean we cannot achieve a reliable 1-to-1 matching between individual radar breaths and reference breaths,” said Trần. “To overcome this, we used 2-minute sliding windows to compute stable average metrics for comparison while still retaining temporal resolution.” 

“Outlier cycles (containing the minimum or maximum TI or TE in the window, respectively) were excluded before averaging to minimize the distortion caused by occasional extreme values without discarding information from most cycles,” she continued. In doing so, the research team was able to prove that noncontact radar could reliably capture complex temporal breathing dynamics, specifically the I:E ratio and individual phase times (TI, TE), and not just basic respiratory rates.

Findings Suggest Even Greater Potential

Statistical equivalence testing expectedly demonstrated a 97.2% equivalence for respiratory rates when comparing radar to reference impedance pneumography, confirming the reliability of radar systems for routine respiratory monitoring. What’s arguably more exciting, however, was obtaining an 85.7% equivalence for the I:E ratio, demonstrating the potential for new possibilities for radar-based monitoring.

“Our study advances the field by demonstrating that radar signal processing can reliably extract detailed phase timings of individual breath cycles with clinically validated accuracy,” said Trần. This methodological benchmark is particularly promising for those with complex pulmonary conditions such as chronic obstructive pulmonary disease, asthma, or respiratory distress, where monitoring breathing phase dynamics is crucial.

Liron Sinvani, MD, aging researcher and professor of medicine at the Feinstein Institutes for Medical Research at Northwell Health in Manhasset, New York, agreed (she was not involved in the study).

“These finer-grained measurements tell us much more about how a patient is breathing and can provide early clues to conditions like airway obstruction, respiratory depression, or worsening lung disease,” she said. “The study builds on previous work by demonstrating that radar can extract these clinically meaningful details with accuracy comparable to a standard contact-based reference, which is an important step toward making radar a truly comprehensive respiratory monitoring tool rather than just a breathing rate counter.”

Moeness Amin, PhD, professor of electrical and computer engineering at Villanova University in Villanova, Pennsylvania, has been involved in radar-based health monitoring for almost 20 years (he was also not involved in the study). While he recognized the promising aspects of this proof-of-concept validation study, he acknowledged the researchers have a long way to go before radar fully works as they suggested.

“The type of radar requires the patients to be in a resting scenario, laying relaxed on the horizontal tilt table, breathing calmly for at least 10 minutes while avoiding large movements,” he said. “This is a very controlled environment. Extension of the results in an uncontrolled environment while the person is nonstationary on the table or broadly moving is more practical.” He also notes the study did not consider estimation of cardiac activities such as heart rates.

In an upcoming follow-up study, Trần and her team plan to address these constraints and validate their advanced data analytics models using an under-the-mattress radar system to test its continuous monitoring capabilities in clinical environments.

Beyond Respiration

Rukiye Kökkiz, a PhD candidate from Fenerbahçe University in Istanbul, Turkey, is currently conducting a study to develop and clinically validate a system that enables continuous and noninvasive cardiac function monitoring in surgical patients during the preoperative and postoperative periods using contactless radar sensor technology.

“Different algorithms will be used in the process of monitoring and extracting cardiac function measurements (respiratory rate, heart rate, blood pressure, chest wall micro-vibrations, stroke volume, and cardiac output estimations) through a single system,” said Kökkiz.

Beyond monitoring cardiac and respiratory functions, radar may even be capable of detecting nocturnal hypoglycemia in diabetes. Researchers from the University of Bern in Bern, Switzerland conducted a scoping review of studies published between 2007 and 2025 that examined how radar monitored physiologic parameters during sleep or sleep-related disorders with overlapping autonomic and motor physiologic signatures.

By synthesizing several radar-based sleep physiology studies, they concluded that radar was well positioned to enable indirect nocturnal hypoglycemia detection at both the physiologic and disease levels.

Alex Dimitriu, MD, sleep medicine physician, was independent from the hypoglycemia study and wasn’t yet convinced. “Given the significance of hypoglycemia, we’d need to see reliable, repeatable detection that accounts for all the variables that could shift this physiologic data — like sleep stages, sleep apnea, or other medical conditions that could alter the signals,” he said.

Nevertheless, all the radar-based monitoring studies discussed here reinforce radar’s potential as a multi-parametric monitoring tool. “The breath-by-breath phase timing and waveform reconstruction methodologies we validated in our paper provide the exact foundational data quality needed to feed future machine learning models aimed at detecting complex events like nocturnal hypoglycemia,” said Trần.

This is especially important because there are unfortunately still not many large-scale, systematic validation studies on radar vital sign monitoring due to the lack of public datasets and standardized protocols. Collecting radar data is logistically complex and often labor-intensive, and the wide variation of hardware parameters — frequency, antenna configurations, and processing algorithms — across studies makes benchmarking difficult.

Still, the potential remains. “Continuous monitoring is critically important for detecting problems early,” said Sinvani. “For older adults being monitored at home, in nursing homes, in the hospital, patients in palliative care, a system that works through clothing and bedding without any patient burden could be a game-changer.”

Sinvani, Amin, and Dimitriu reported having no conflicts. Disclosure information for the study authors is available in the original study publications.