World Sleep Industry ENCYCLOPEDIA

Wi-Fi CSI sleep sensing modules

6.1.1

Definition and industrial scope

A Wi-Fi channel state information sleep-sensing module is a component or subsystem that uses small changes in wireless propagation around a room and a human body to infer sleep-related states such as bed occupancy, gross movement, turning, and resting respiratory rhythm. Channel state information, or CSI, normally represents the complex channel response for multiple subcarriers and transmit-receive antenna paths at a particular time. It may expose amplitude, phase, or features derived from them. CSI retains finer frequency and spatial structure than received signal strength indication, which compresses a link into a coarse value. It nevertheless measures radio propagation, not airflow, respiratory effort, electrocardiography, electroencephalography, or oxygen saturation directly.

The word module in this entry means more than a Wi-Fi chip that can exchange network traffic. It denotes a traceable sensing chain containing controlled transmitting and receiving functions, antennas and radio-frequency front ends, CSI acquisition firmware, timing and buffering, edge computation or a host-processing interface, and quality and configuration management. An ordinary router, a network module that reports only RSSI, a dedicated millimeter-wave radar, and a complete polysomnography system should not all be described as CSI sleep-sensing modules. When a product reports sleep stages, respiratory events, or other health conclusions, the native measurement capability of the module must be distinguished from the upstream algorithm, clinical interpretation, and regulated product claim.

Sensing mechanism and system architecture

A wireless signal reaches the receiver through a direct path and through reflections from walls, furniture, the bed, and the body. Small periodic chest and abdominal motion, larger limb or trunk movement, posture changes, and people entering or leaving the area alter the length, attenuation, and phase relationships of some paths. Those changes modulate the CSI time series. Sensitivity depends on the person's location in the propagation geometry. The same respiratory motion may be prominent at one position and fall into a low-sensitivity region or blind spot at another. CSI therefore cannot be interpreted independently of node placement, antenna orientation, frequency band, channel bandwidth, room layout, and the number of people on the bed.

A typical architecture contains one or more transmitting nodes, one or more receiving nodes, and a sensing processor. The transmitter produces measurement-capable frames at a controlled cadence. The receiver records the response of subcarriers and antenna paths together with packet sequence, timestamp, and link state. The processor performs synchronization, cleaning, path selection, feature extraction, and state estimation before sending results to a mattress controller, bedside device, gateway, or cloud service. Transmit and receive functions may also reside in access points, client devices, or dedicated nodes. The design must still identify which component controls a measurement session, where raw measurements are held, and how communication traffic and sensing acquisition coexist.

Hardware, interfaces, and deployment conditions

Hardware selection should consider whether the chipset exposes stable CSI, supported bands and bandwidths, antenna count and spacing, automatic-gain behavior, sampling continuity, timestamp precision, buffer capacity, thermal drift, and continuous-power requirements. Different chipsets and firmware may expose different subcarrier sets, quantization formats, and phase biases. Data from two implementations cannot be assumed interchangeable merely because both are called CSI. The host interface should carry device identity, path and antenna indexes, center frequency, bandwidth, packet sequence, acquisition time, loss indicators, and firmware version, with an explicit version for every data-format revision.

Installation instructions should define node height, distance, direction, and obstructions relative to the bed, while fixing antenna orientation and power-supply placement. The installer should record the bed frame, mattress, walls, and major furniture. Commissioning should cover an empty-room baseline, an unoccupied bed, a person lying still, and representative turns. If multiple links are used to reduce blind spots, their complementarity should be demonstrated. Moving a router, automatic channel changes, new metal objects, changed doors or windows, a relocated fan, or a new bed partner can all change the multipath field. The product should detect a material configuration change and request review instead of applying an old calibration indefinitely.

Signal processing, quality control, and outputs

The processing chain normally starts by checking packet order and timing continuity. It then corrects phase- or frequency-offset artifacts, suppresses outliers and slow drift, and selects or fuses components from subcarriers and antenna paths that are responsive to the intended movement. Respiratory estimation may use periodicity within a plausible band, amplitude and phase changes, component analysis, or a learned model. Gross-motion detection emphasizes wider-band, higher-amplitude transients. Filter limits, analysis-window duration, resampling, and model version should match the target population, environment, and latency requirement, and every reproducible setting should be retained.

Appropriate outputs include occupancy, movement or turn events, respiratory rate during qualified resting intervals, analyzable duration, signal quality, and invalidity reasons. Each number or event should carry its interval and a confidence or quality grade. During gross motion, nearby walking, acquisition gaps, network congestion, link changes, or room reconfiguration, the module should declare an invalid or degraded state. Smoothing and interpolation must not manufacture a continuous, convincing physiological trace where no valid measurement exists. Heart rate, apnea candidates, or sleep stages, if offered, must be identified as model inferences with stated reference signals and limits of use.

Single-person, multi-person, and interference management

A single still occupant is the easiest condition to interpret, yet side or prone posture, thick bedding, reduced respiratory excursion, and a low-sensitivity node geometry can still weaken the signal. A bed partner, pet, caregiver, person in a neighboring bed, or oscillating fan can add periodic or nonperiodic changes. Motion-related CSI variation is often much larger than respiration-related variation. A system should therefore segment the scene and exclude gross-motion intervals before estimating resting respiration, rather than passing every fluctuation into a single rate estimator.

Multi-person monitoring requires both separation of periodic components and correct attribution of each component to a bed position or person. Finding two spectral peaks does not demonstrate identity assignment. Spatially separated links, antenna arrays, known bed geometry, or an auxiliary sensor may help, but performance must be tested when respiratory rates are close, both people turn, one leaves, and people exchange positions. When reliable attribution is unavailable, the module should fall back to room-level presence or activity, or explicitly withhold individual vital-sign values. A plausible but incorrectly assigned result can be more harmful than a declared lack of data.

Verification, acceptance, and maintenance

Verification should use synchronized references selected for each claim. Respiratory rate may be compared with a qualified respiratory belt, airflow channel, or clinical reference. Movement and posture may use synchronized video or adjudicated observation. Sleep staging and respiratory-event claims require the relevant polysomnographic signals and expert scoring. The evaluation population and sites should span body sizes, ages, postures, bed frames and mattresses, room dimensions, node distances, line-of-sight and non-line-of-sight geometries, network loads, and ordinary interferers. Training, tuning, and test partitions should be separated by person and home so that room fingerprints do not leak into the reported result.

Respiratory-rate reports should give the error distribution, valid coverage, and definition of a resting interval. Event reports should include sensitivity, specificity, positive predictive value, false-alarm burden, and timing tolerance. Occupancy and motion reports should quantify missed and false detections over complete nights. A mean accuracy alone cannot replace stratified outcomes and failure cases. Production acceptance should also test antenna assembly variation, firmware updates, power recovery, extended packet loss, channel switching, and clock drift. Deployed fleets should be watched through quality statistics, configuration fingerprints, and drift alerts, with renewed validation after a material algorithm or radio-configuration change.

Safety, privacy, compliance, and supply chain

CSI sensing does not make images, but it may reveal whether a room is occupied, when people move, and potentially their respiratory rhythm. It therefore handles sensitive information. A system should obtain clear consent before activation, show a visible operating state, and provide a practical way to disable sensing. Data minimization and local feature extraction should be defaults. Raw CSI, device identifiers, and derived records need encryption, least-privilege access, key rotation, audit logs, defined retention, and secure deletion. Remote updates require signature verification, rollback protection, and a vulnerability-response process. The absence of a camera does not remove the duty to inform partners, visitors, and other bystanders.

The finished product must be assessed for applicable radio, electromagnetic compatibility, electrical safety, cybersecurity, privacy, and medical-device requirements in every market where it is sold. WLAN sensing standardization can improve measurement sessions and information exchange between compatible devices, but it does not establish the clinical validity of a sleep algorithm. Supply-chain records should freeze chipsets, antennas, radio parameters, firmware, models, calibration, and interface versions, and replacements for discontinued parts need equivalence review. Marketing should distinguish wellness trends, screening prompts, and diagnostic claims. Abnormal results should lead to an appropriately validated care pathway, not to a diagnosis or treatment decision made by the module alone.