A Simultaneous ECG-PCG Acquisition System with Real-Time Burst-Adaptive Noise Cancellation

Cardiac auscultation is an essential clinical skill, requiring excellent hearing to distinguish subtle differences in timing and pitch of heart sounds. However, diagnosing solely from these sounds is often challenging due to interference from surrounding noise, and the information may be limited. Most of the existing solutions that adaptively cancel external noise are either non-real-time or computationally intensive, making them unsuitable for implementation in a portable system. This work proposes an end-to-end system with a real-time adaptive noise cancellation pipeline integrated into a device that simultaneously acquires electrocardiogram (ECG) and phonocardiogram (PCG) signals. We employ a burst adaptive normalized least mean square algorithm that adjusts its adaptation in response to high-energy, non-stationary hospital noise. The algorithm’s performance was initially assessed using datasets with artificially induced noise. Subsequently, the complete end-to-end system was validated using real-world hospital recordings captured with the dual-modality device. For PCG and ECG signals recorded from the device in noisy hospital settings, the proposed system achieved signal-to-noise ratio improvements of 37.01 dB and 30.32 dB, respectively. Furthermore, complexity analysis confirms the pipeline’s suitability for embedded implementation. These results demonstrate the system’s effectiveness in enabling reliable and accessible cardiac screening in noisy hospital environments typical of resource-constrained settings.

💡 Research Summary

The paper presents a complete, low‑cost, portable system that simultaneously records electrocardiogram (ECG) and phonocardiogram (PCG) signals while performing real‑time adaptive noise cancellation (ANC) suitable for noisy hospital environments, especially in resource‑limited settings. The hardware integrates a medical‑grade 24‑bit ADS1294 ADC for single‑lead ECG acquisition via three dry Ag/AgCl electrodes, and a dual‑microphone array (digital MEMS microphones CMM‑4030DT) mounted on a Littmann III stethoscope head for heart‑sound capture. Ambient noise is recorded by a separate microphone array. Signals are sampled at 500 Hz (ECG) and 8 kHz (audio), down‑sampled to 2 kHz for processing on an STM32F746 Cortex‑M7 microcontroller. The MCU uses DMA and double‑buffering to stream data via a BLE module to a PC for visualization.

The signal‑processing pipeline consists of two parallel paths. The ECG path applies three elliptic IIR filters (0.5 Hz high‑pass, 150 Hz low‑pass, 49.5‑50.5 Hz notch) to remove baseline drift, high‑frequency noise, and power‑line interference. The PCG path employs a novel burst‑adaptive normalized least‑mean‑square (BA‑NLMS) filter. Unlike conventional NLMS, BA‑NLMS estimates the average reference‑signal energy with a moving average, detects sudden energy bursts using a threshold η, and temporarily scales the step size by a factor β (>1) during bursts. This dynamic step‑size adjustment enables rapid convergence when non‑stationary, high‑energy hospital noises occur, while preserving stability during quieter periods. The weight update follows:

( \bar E(n)=\alpha\bar E(n-1)+(1-\alpha)|r(n)|^2 )

( \mu_{\text{eff}}(n)=\begin{cases}\mu_0\beta & \text{if } |r(n)|^2 > \eta\bar E(n)\ \mu_0 & \text{otherwise}\end{cases} )

( w(n+1)=w(n)+\mu_{\text{eff}}(n)\frac{e(n)r(n)}{\epsilon+|r(n)|^2} )

where (r(n)) is the reference noise, (e(n)) the error (desired‑minus‑output), and (\epsilon) a small constant.

Performance evaluation proceeded in three stages. First, synthetic experiments mixed clean PCG recordings from the PhysioNet/CinC 2016 dataset with a large collection of real hospital ambient noises (562 recordings) at SNRs ranging from –10 dB to +5 dB. BA‑NLMS achieved a normalized mean‑square error (NMSE) of 0.0229 ± 0.0201, ΔSNR of 20.05 dB, correlation coefficient (CC) of 0.9891, and normalized mean absolute error (NMAE) of 0.1285, outperforming standard LMS, conventional NLMS, and a deep‑learning LU‑Net baseline. Notably, during simulated burst events the algorithm maintained rapid adaptation, reducing NMSE by ~30 % compared with NLMS.

Second, the complete device was tested on real ECG‑PCG recordings obtained from pediatric patients at Lady Ridgeway Hospital for Children, Colombo, Sri Lanka. In these real‑world recordings, the system improved PCG SNR by an average of 37.01 dB and ECG SNR by 30.32 dB, confirming robustness against actual clinical acoustic coupling and non‑stationary noise sources.

Third, computational complexity was analyzed. BA‑NLMS retains O(N) complexity (N = filter length) identical to NLMS, with only a few extra arithmetic operations for energy averaging and threshold comparison. Benchmarks on the STM32F746 showed that the full pipeline (both ECG and PCG paths) consumes well under the MCU’s processing budget, allowing continuous operation at the target sampling rates with a modest power draw of ~90 mA from a 1200 mAh Li‑Po battery, yielding >13 hours of runtime.

In summary, the authors deliver an end‑to‑end solution that couples a thoughtfully engineered dual‑modality acquisition front‑end with a lightweight, burst‑aware adaptive filter. The system achieves clinically relevant SNR gains for both ECG and PCG in noisy, resource‑constrained hospital settings, and its low computational load makes it suitable for embedded deployment. This work advances the feasibility of portable, real‑time cardiac screening and paves the way for broader tele‑cardiology applications in low‑ and middle‑income regions.