World Sleep Industry ENCYCLOPEDIA

Millimeter-Wave Radar Modules

6.1.1

Definition, composition, and industry role

A millimeter-wave radar module integrates an RF transmitter and receiver, antennas, clocks and power, analog-to-digital conversion, digital processing, memory, and host interfaces into a sensing component that can be embedded in a product. Some modules use antenna-in-package construction, placing the RF integrated circuit and antenna array in one package. Others combine a radar chip with printed-circuit-board antennas and a microcontroller or digital signal processor. A module may expose raw or intermediate-frequency samples, or perform local processing and output range, velocity, angle, point clouds, target tracks, presence, body movement, or respiratory trends.

In sleep products, mmWave radar can support contactless presence, bed entry and exit, gross movement, sleep-zone occupancy, and sensing of small chest or abdominal displacement. Radio sensing works in darkness, does not require a visible-light image, and does not attach to the skin. With suitable engineering, the sensor can sit behind a nonmetallic cover or furniture surface. However, “camera-free” does not mean privacy-free, and sensitivity to micromotion does not inherently deliver sleep stages, diagnose apnea, or produce clinical-grade vital signs.

The module is a component, not a finished system. Results depend on spectrum rules, antenna orientation, mounting height, bed materials, subject range, indoor multipath, algorithms, firmware, and the product enclosure. A developer should distinguish an evaluation board, reference design, generic module, and certified production module. A supplier demonstration with one still person at an ideal position is not automatically an end-product specification. A medical device, alarm, or safety-critical use adds risk management, clinical or performance validation, and regulatory obligations in the intended market.

Operating principles and observable quantities

Many modules use frequency-modulated continuous-wave radar. The transmitter produces a linear frequency sweep, or chirp. The receiver captures target reflections, and mixing the transmitted and received signals produces a beat frequency related to propagation delay and therefore range. Greater usable sweep bandwidth generally improves theoretical range resolution. Phase changes across chirps carry Doppler velocity information, while phase differences across receive channels support angle-of-arrival estimation. Some products use pulse radar, which has different timing, duty-cycle, and compliance considerations.

Sleep sensing relies heavily on phase sensitivity to minute displacement. Periodic motion from breathing modulates the phase in a selected range bin. Phase unwrapping, detrending, filtering, and spectral or time-domain analysis can estimate a respiratory rhythm. Cardiac displacement is smaller and faster and is readily contaminated by respiratory harmonics, mattress transmission, posture, bedding, and gross motion. The radar directly observes electromagnetic reflections and displacement—not airflow, oxygen saturation, brain activity, or sleep state. Physiological meaning comes from a model and the evidence used to validate it.

Range, velocity, and angle figures are coupled rather than independent marketing numbers. Maximum unambiguous range is shaped by sampling and chirp slope, velocity coverage by chirp repetition, and angular resolution by aperture and array geometry. Improving one dimension may increase data volume, power, or frame duration or reduce performance elsewhere. Static-clutter removal may improve person detection while also attenuating slow respiratory content that the application needs. Waveform design should work backward from the sleep task, installation, and allowed latency instead of accepting an evaluation-board default.

Specifications, selection, and interfaces

Selection begins with the operating band, bandwidth, equivalent radiated power, power spectral density, duty cycle, and indoor, outdoor, fixed, or mobile conditions allowed in every target jurisdiction. Commercial devices may operate near 24 GHz, 60 GHz, or higher automotive bands, but available spectrum and permitted uses are not globally uniform. The same chip can fall under different test conditions because of firmware, antenna gain, or enclosure. A procurement specification should lock the actual frequency limits, transmit and receive channels, antenna gain, and radiation pattern rather than state only “mmWave.”

Sensing requirements include minimum and maximum range, range and velocity resolution, field of view, angle accuracy, update rate, target capacity, static-presence sensitivity, startup time, and end-to-end latency. A sleep use case also needs usable conditions for one and two occupants, side sleeping, heavy bedding, a running fan, pets, and an adjacent bed. A respiratory output should define units, update period, valid observation window, confidence, invalidity criteria, and recovery after gross motion, not only an average error from a controlled demonstration.

Interfaces may include SPI, UART, CAN, USB, or Ethernet. The output layer may range from ADC and IF samples through radar cubes, point clouds, and tracks to events. Raw access supports independent algorithm development but imposes bandwidth, storage, computation, and data-protection costs. Edge events conserve resources but deepen dependence on supplier algorithms and firmware versions. Power rails, peak current, low-power wake behavior, temperature limits, thermal resistance, dimensions, connectors, boot straps, debug ports, SDK licensing, and production-tool support also belong in the selection matrix.

Sleep-processing chain and scenario limitations

A representative processing chain performs RF calibration, DC and static-clutter handling, a range transform, cross-chirp Doppler processing, beamforming or angle estimation, detection, and tracking. A micromotion branch selects signals around the person's range-angle region, unwraps and differences phase, flags motion-corrupted intervals, filters appropriate bands, and estimates spectra or temporal periodicity. The product layer fuses presence, bed-zone, motion, and time-series features into events. It should retain synchronized timestamps, configuration and model versions, and quality indicators instead of emitting only a confident-looking label.

Multiple occupants are a central bedroom challenge. Reflections can merge when two people have similar range or angle, and one person's large movement can obscure the other's micromotion. Pets, oscillating fans, curtains, HVAC vanes, cleaning robots, and a bedside caregiver can produce periodic or moving returns. Springs, metal bed frames, electric blankets, thick mattresses, mosquito nets, and metal-coated covers modify propagation and multipath. The algorithm needs explicit states for unsupported scene, lost target, and insufficient quality rather than fabricating a continuous value.

Radar can contribute movement, presence, and respiration-related features to a sleep-stage model, but these are not electroencephalography, electrooculography, or electromyography. Labeling wake, light, deep, and rapid-eye-movement sleep as ground truth from radar alone requires independent validation against an appropriate reference in the intended population. A claim to detect apnea, provide a life-safety alarm, or guide treatment needs a defined intended use, analysis of missed and false alerts, a response chain, and the applicable medical-device route. A consumer “sleep trend” must not imply that disease has been excluded.

Antenna, mechanical, and end-product integration

The antenna region determines whether a module can achieve its stated performance. Board implementation must follow supplier requirements for stack-up, substrate, copper, impedance, grounding, vias, and antenna keep-outs. Metal, cables, batteries, and lossy materials should not obstruct the RF aperture. Antenna-in-package parts reduce board-level RF work but still require controlled orientation, nearby structure, and assembly tolerance. Headboard, ceiling, side-wall, and furniture-integrated positions have different coverage and occlusion and require measurement in real rooms rather than an open-space calculation.

A radome or product cover must be assessed at the operating band for permittivity, loss, thickness, curvature, coatings, seams, and moisture. Metallic pigments, conductive fabric, coated glass, and structural ribs can attenuate the main beam or create sidelobes; resin filler and moisture variations can cause lot-to-lot shifts. After enclosure assembly, the team should remeasure pattern, coverage, blind zones, and clutter. Testing should include objects placed by users, wall angles, moved furniture or beds, and residue from normal cleaning.

Power ripple, digital clocks, communication radios, and fast processors can raise noise or create spurs, while radar transmissions must coexist with Wi‑Fi, other radars, and the product's radios. Integration includes filtering, clock planning, grounding and shielding, thermal management, and electromagnetic-compatibility pre-scans. Hospitals, hotels, or showrooms with many units require mutual-interference tests and timing, waveform, channel, or interference-detection strategies. Through-wall or cross-room sensing is not an automatic benefit; range gates and spatial masks should keep collection inside the intended area.

Data, privacy, cybersecurity, and compliance

Radar normally does not produce directly recognizable facial images, but raw samples, point clouds, room occupancy, bed-entry timing, and respiratory trends can still reveal behavior and health information. Privacy begins at the sensing boundary, using physical direction, range gates, and zone masks to minimize collection. Processing that can occur on the module or a local gateway should not default to cloud upload of raw data. External transmission can be limited to events, necessary aggregates, or reduced data with clear retention, access, and deletion rules.

Firmware security affects measurement integrity and privacy. Production devices benefit from secure boot, signed updates, controlled debug access, device identity, encrypted communication, and least privilege. Radar configurations and models should be protected from unauthorized modification. An update process needs version, rollback, and compatibility controls so a remote change cannot silently alter the sensing boundary, transmit parameters, or alarm thresholds. Cloud, mobile applications, and third-party analytics belong in the threat model; account deletion, resale, and vendor exit require credential revocation and associated-data deletion.

Radio compliance is a shared module and end-product responsibility. A certified module may reduce some work, but antenna, cover, supply, firmware, emission settings, or installation changes can require reassessment. Market entry includes the applicable radio, exposure, electromagnetic compatibility, electrical safety, environmental, and labeling requirements, plus additional obligations for medical, elder-care, workplace, or child-related settings. The compliance file should identify hardware, antenna, firmware configuration, test mode, and installation restrictions so the tested sample matches production.

Validation, manufacturing, and supply-chain control

Validation begins with traceable requirements that decompose “presence,” “bed exit,” or “respiratory trend” by distance, posture, bedding, occlusion, number of people, motion, environment, and latency. Bench work checks RF output, noise, clocks, and interfaces; a chamber or controlled space measures patterns and target response; scenario testing uses real bedrooms, varied beds, and multi-person interference. Algorithm datasets need varied body sizes, ages, clothing, bedding, and reflectivity conditions. Splitting training and test data by person prevents fragments from the same subject inflating performance.

The reference must match the claim. Presence and bed exit can use synchronized human annotation and suitable sensors; respiratory trends need a verified respiration reference; sleep stages and medical events demand more rigorous clinical references, protocols, and statistics. Reports should include failure rate, coverage, confidence intervals, subgroup performance, multi-person and post-motion recovery, false and missed events, and handling of invalid data—not just mean error. Demonstrations, engineering verification, and regulatory evidence should remain distinct.

Production control includes material traceability, optical and electrical inspection, RF calibration, transmit-receive self-test, sampled antenna or end-to-end tests, and firmware programming verification. Golden units, limits, and measurement-system analysis help detect lot drift and tester differences. A change to wafer, package, antenna material, memory, algorithm library, or development tool can affect performance and needs supplier notification, impact review, and revalidation. The deliverable is not only hardware; it includes pinned configuration and algorithms, calibration, compliance evidence, defect response, and a lifecycle support commitment.