World Sleep Industry ENCYCLOPEDIA

BCG sensing modules

6.1.1

Definition, industry role, and adjacent boundaries

A ballistocardiographic sensing module is a component assembly that acquires small mechanical reactions of the body caused by cardiac ejection and blood acceleration, then converts, conditions, and digitizes them as a ballistocardiogram. It can be installed in a mattress, topper, bed deck, bed leg, chair, weighing scale, or wearable carrier. It commonly senses respiration and body movement as well, providing input for low-burden cardiopulmonary trend monitoring during sleep.

A module typically combines one or more sensing elements—such as piezoelectric or electromechanical film, pressure or force sensors, load cells, or accelerometers—with an analog front end, filtering, analog-to-digital conversion, clock, interface, and quality detection. Some modules also run heartbeat, respiration, movement, or occupancy algorithms. Within WSIC it is a specialized component in the research, manufacturing, and supply chain pathway. Outputting a heart-rate number does not automatically make it a medical diagnostic device.

This entry differs from Smart-Mattress Piezoelectric Sensing Films. A film is one possible raw sensing element; a BCG module additionally manages conditioning and interfaces. It also differs from ECG: ECG records electrical cardiac activity, whereas BCG records mechanical reaction of the body and support structure. Peaks, timing, and pathologic meaning are not interchangeable. A complete sleep-monitoring system still needs applications, data governance, product-level algorithms, and clinical validation.

Physiological source, waveform, and interpretation limits

Cardiac contraction accelerates blood and changes its direction within major vessels, producing reaction forces transmitted through the body. BCG waveforms are often described with H, I, J, K, and other labels, but actual morphology depends on measurement axis, posture, body type, mattress, frame, and sensor placement. A bed module observes the combined transfer of body, bed, and transducer, not a direct copy of intracardiac pressure or blood flow.

Respiration produces slower motion of the torso and bed, while turning, limb movement, coughing, and external impacts usually exceed cardiac recoil by a large margin. A channel contains a mixture of BCG, respiration, motion, structural resonance, and electrical noise. A J-like peak visible in a quiet interval may deform, split, or disappear after a postural change or in disease, so every local extremum cannot be assumed to be a heartbeat.

BCG can support estimates of heart rate, interbeat intervals, breathing rate, and movement and can be used to study selected relative changes. J-wave amplitude is not equivalent to stroke volume, however, and the R–J interval depends on sensor method and transfer path. Blood pressure, cardiac output, arrhythmia, sleep stages, and deterioration alerts require validation of the particular hardware, algorithm, and population against appropriate references. Suppliers should separate directly measured quantities from derived interpretation.

Sensors, analog front end, and digital interface

Piezoelectric and electromechanical films capture dynamic deformation of a bed, load cells measure forces at supports, accelerometers sense local vibration, and hydraulic or pneumatic structures transmit body mechanics to a pressure transducer. These routes differ in sensitive direction, bandwidth, preload, noise, and static response. Combining sensors may broaden coverage or aid posture recognition while introducing channel matching, synchronization, and fusion problems.

The analog front end must span weak cardiac recoil and very large turning signals. High-impedance piezoelectric elements commonly need charge or high-input-impedance amplification. Bridge force sensors need stable excitation and low-drift instrumentation amplification, while accelerometers need suitable range and noise density. Bias, antialias filtering, mains, electromagnetic and electrostatic protection, saturation recovery, and electrical isolation near the user also belong in the front-end design.

Sampling rate, clock stability, channel latency, and filter phase affect interbeat timing and synchronized reference comparison. Module output may include raw and conditioned waveforms, signal quality, motion flags, and derived metrics. The protocol must define units, range, timestamps, invalid values, and version. An average heart rate without valid duration and quality state prevents the host system from distinguishing true stability from algorithmic filling of missing intervals.

Bed installation, adaptation, and multichannel management

Sensors above or below a mattress, on a deck, and under bed legs observe different force axes and waveforms. Mattress material, thickness, firmness, protectors, springs, and frame change transmission. The distance between sensor and thorax, occupant height and weight, and lateral, supine, or prone posture alter quality. Modules need a validated mounting range, orientation, and bed compatibility rather than a generic instruction to place them under a mattress.

A point module may lose cardiac activity when the user moves outside its region. Multisite arrays expand coverage and permit adaptive channel selection but require calibration of gain, noise, latency, and position. The selected channel and reason for switching should remain traceable. Fusion should not be based on choosing the clearest signal retrospectively; unknown posture, failed channels, and inconsistent signals should produce an insufficient-quality state rather than a forced metric.

Bed partners, pets, adjustable beds, massage, heating, and care-bed motion are separate challenges. Two people's cardiac and respiratory signals can couple through the structure, and one wide sensor may not separate them. A partner's turn can contaminate the opposite side. Bed exit, sitting, edge loading, and environmental vibration need explicit states. A product validated only for one person on a static bed should say so rather than silently claiming every household configuration.

Signal processing, quality gating, and output metrics

A typical chain first identifies occupancy and gross movement, excludes saturation and recovery, and then extracts respiratory and cardioballistic components from quieter windows. Time-domain peak detection is intuitive, but morphology changes with posture. Spectral, autocorrelation, or cepstral approaches estimate average periodicity but may not provide accurate beat-to-beat intervals. Wavelets, empirical decomposition, and machine learning can address nonstationarity while adding training-data, versioning, and explainability requirements.

Quality gating should consider amplitude range, noise, periodic stability, channel agreement, movement, contact, and front-end state and preserve the valid proportion in output. Invalid periods should not be interpolated and then used for HRV, respiratory events, or alarms. Filters, window length, refractory period, outlier rules, and fusion strategy must be versioned. Algorithm updates require renewed end-to-end validation from raw input to final metrics.

Heart and respiratory rates are usually the most basic outputs. Interbeat intervals, HRV, respiratory variability, occupancy time, and movement statistics need stricter point-level quality. Models that infer awakening, sleep stages, apnea, rhythm problems, or hemodynamics are higher-level product functions, not intrinsic module capabilities. Data latency, refresh window, unmeasurable conditions, and suitability for real-time alarms should be stated.

Validation, quality system, and procurement acceptance

Bench testing should cover sensitivity, bandwidth, noise, dynamic range, linearity or repeatability, channel matching, synchronization, saturation recovery, temperature and humidity, power, and interface faults. Mechanical testing should include preload, impact, bending, compression, bed-frame resonance, and installation repeatability. Sensor and front-end specifications should be established separately, followed by end-to-end standard excitation so software correction does not conceal hardware drift.

Human validation should cover ages, sexes, body shapes, weights, mattresses, postures, overnight movement, and repeated nights. Heart rate and interbeat interval require synchronized ECG reference, breathing rate requires a validated respiratory channel, and sleep parameters require fit-for-purpose PSG. In addition to mean error and correlation, reports should include agreement, epoch-level error, valid-data proportion, postmovement recovery, failed participants, and worst cases. Training, tuning, and independent test data must remain distinct.

Procurement specifications should define sensing elements, mounting, channel count, raw output, sampling and filters, interface, clock, gain, quality flags, power, environment, life, and whether algorithms reside in firmware. Acceptance should confirm hardware and firmware versions, serial identity, calibration, waveform replay, and fault codes. A supplier's best case or average heart rate cannot replace system validation on the target bed and application.

Safety, compliance, manufacturing, and data governance

Manufacturing controls should cover sensor lots, bonding and preload, amplifier parts, conversion, clock, shielding, connectors, and encapsulation. Under-bed modules experience compression, impact, damp heat, cleaning fluids, shipping, and aging support surfaces. Functional self-test should cover open and short channels, offset, noise, and stored configuration. Supplier or algorithm changes require risk assessment, bridging, and traceable release.

Safety scope includes isolation, leakage, electrostatic discharge, electromagnetic compatibility, temperature rise, liquid ingress, flammability, sharp edges, wireless and charging states, and interference with powered beds or other medical equipment. Medical, research, wellness, and consumer uses carry different market-access, labeling, alarm, and quality duties. “Contactless” does not mean risk-free or maintenance-free, and the module cannot substitute for emergency monitoring without evidence.

BCG can reveal routines, co-sleeping, movement, and health trends over long periods. Raw signals, quality flags, and derived metrics need data minimization, access control, retention limits, deletion, and audit. Remote updates must not change outputs through unvalidated firmware or models. The field is moving toward multichannel sensing, self-calibration, edge quality assessment, and multimodal fusion, but maturity should be judged by visible failures, explicit limits, and repeated cross-bed validation—not the uninterrupted production of numbers.