A Smart Cushion for Real-Time Heart Rate Monitoring

This paper presents a smart cushion for real time heart rate monitoring. The cushion comprises of an integrated micro-bending fiber sensor, which records the BCG (Ballistocardiogram) signal without direct skin-electrode contact, and an optical transceiver that does signal amplification, digitization, and pre-filtering. To remove the artifacts and extract heart rate from BCG signal, a computationally efficient heart rate detection algorithm is developed. The system doesn’t require any pre-training and is highly responsive with the outputs updated every 3 sec and initial response within first 10 sec. Tests conducted on human subjects show the detected heart rate closely matches the one from a commercial SpO2 device.

💡 Research Summary

The paper introduces a novel “smart cushion” that enables real‑time, contact‑free heart‑rate monitoring by exploiting a micro‑bending fiber‑optic sensor to capture ballistocardiogram (BCG) signals. The motivation stems from the high prevalence of sleep disorders and the need for unobtrusive physiological monitoring; heart‑rate variability (HRV) is a proven indicator of sleep quality, yet conventional electrocardiogram (ECG) approaches require skin‑electrode contact, limiting their suitability for home or bedside use. BCG, which records minute body vibrations caused by cardiac ejection, offers a non‑contact alternative, but existing electronic BCG sensors suffer from electromagnetic interference, especially in environments such as MRI suites.

The proposed system integrates a micro‑bending multimode fiber sandwiched between two micro‑benders. When a subject lies on or leans against the cushion, cardiac‑induced micro‑movements modulate the light intensity traveling through the fiber. An embedded optical transceiver (LED source, photodiode detector, low‑noise amplifiers, FIR band‑pass filter, and a 12‑bit ADC) converts this optical modulation into a digital signal sampled at 50 Hz and streams it via USB to a host computer. The transceiver’s built‑in 1.6 Hz high‑pass component removes the dominant respiratory component, while a 10 Hz low‑pass limits high‑frequency noise, leaving the 2–10 Hz band where BCG peaks reside.

Signal processing proceeds through a six‑stage pipeline: (1) 2–10 Hz FIR band‑pass filtering (40 dB attenuation), (2) cubic amplitude scaling (x³) to accentuate peak heights without altering zero‑crossings, (3) a short 0.06 s moving‑average to smooth transient spikes, (4) absolute‑value conversion followed by a 0.3 s moving‑average to produce a smooth envelope suitable for thresholding, (5) adaptive threshold determination, and (6) peak detection. The adaptive threshold is initialized as 25 % of the maximum value among the first 300 samples. After each detection, the threshold is updated to 25 % of the average amplitude of the last eight confirmed peaks. To prevent runaway thresholds, any newly measured peak exceeding twice the previous peak is capped, and if no peaks are found over a J‑J interval the threshold is reduced by 10 % to avoid lock‑out.

Peak detection itself looks for a monotonically rising edge (three consecutive increasing samples) followed within 0.25 s (≈12 samples) by a monotonically falling edge, satisfying specific sample‑difference conditions. If two detected peaks lie within 0.3 s, the lower‑amplitude one is discarded, mitigating false positives caused by noise or motion artifacts.

The algorithm was evaluated on five healthy volunteers (ages 25–49, heights 154–172 cm, weights 53–68 kg) who rested on the cushion for five minutes while a commercial fingertip pulse‑oximeter recorded reference heart rates. Performance metrics—sensitivity (Se), positive predictive value (+P), and detection error rate (DER)—were computed from true positives (TP), false positives (FP), and false negatives (FN). Results (Table I) show sensitivities ranging from 96.0 % to 99.0 %, +P from 96.4 % to 100 %, and DER between 0.013 % and 0.080 %. Table II demonstrates that the average heart rate derived from the cushion differed from the SpO₂ reference by less than 2 bpm in all cases.

Compared with prior BCG processing approaches—wavelet transforms, machine‑learning classifiers, and template‑matching—the presented method achieves comparable or superior accuracy while requiring far fewer computational resources, making it amenable to low‑power ASIC implementation. The fiber‑optic sensor also offers intrinsic immunity to electromagnetic interference, enabling potential use in MRI environments where electronic sensors fail.

Limitations include the controlled, static testing conditions; the algorithm’s robustness to large body movements, multi‑user scenarios, or long‑duration monitoring was not assessed. Future work should explore multi‑channel fiber arrays for spatial filtering, adaptive motion‑artifact cancellation, and integration of additional physiological signals (e.g., respiration) to broaden clinical applicability.

In conclusion, the authors deliver a compact, low‑cost smart cushion that captures BCG via a micro‑bending fiber sensor and extracts heart rate with a lightweight, adaptive detection algorithm. The system delivers >95 % accuracy with updates every three seconds and an initial response within ten seconds, matching commercial pulse‑oximeter measurements and demonstrating strong potential for home‑based sleep and health monitoring.