[Paper Review] A Dynamic Parametric Simulator for Fetal Heart Sounds

Yang Zhou, Yiang Zhou+3 more|arXiv (Cornell University)|Jan 23, 2026

Phonocardiography and Auscultation Techniques0 citations

TL;DR

The paper introduces a reproducible dynamic parametric simulator for abdominal fetal phonocardiograms (fPCG) that combines cycle-level fetal S1/S2 event synthesis with a transmission model and configurable noise, calibrated from real recordings, to enable controlled benchmarking.

ABSTRACT

Research on fetal phonocardiogram (fPCG) is challenged by the limited number of abdominal recordings, substantial maternal interference, and marked transmissioninduced signal attenuation that complicate reproducible benchmarking. We present a reproducible dynamic parametric simulator that generates long abdominal fPCG sequences by combining cycle-level fetal S1/S2 event synthesis with a convolutional transmission module and configurable interference and background noise. Model parameters are calibrated cyclewise from real abdominal recordings to capture beat-to-beat variability and to define data-driven admissible ranges for controllable synthesis. The generated signals are validated against real recordings in terms of envelope-based temporal structure and frequency-domain characteristics. The simulator is released as open software to support rapid, reproducible evaluation of fPCG processing methods under controlled acquisition conditions.

Motivation & Objective

• Provide a reproducible, interpretable simulator for abdominal fPCG signals under controlled interference and SNR conditions.
• Calibrate cycle-level fetal heart sound parameters from real abdominal recordings to capture beat-to-beat variability.
• Incorporate an explicit abdomen-to-sensor transmission model and a configurable noise mechanism.
• Enable long-sequence synthesis with data-driven admissible parameter ranges for benchmarking fPCG processing methods.

Proposed method

• Model observed abdominal fPCG as x(t)=h(t)*(xf(t)+xm(t))+n(t) with separate fetal, maternal, and noise components.
• Represent fetal S1/S2 events with an asymmetric damped-sinusoid kernel driven by cycle-specific parameters (A, f0, Ta, tau).
• Use per-cycle parameter vectors theta^(k) to control S1/S2 amplitude and envelope, with options to share envelopes for S1/S2 across cycles.
• Model maternal cardiac interference as a second two-event source with fixed nuisance hyperparameters and a global scaling factor.
• Apply a cascaded exponential impulse response h(t) to simulate abdominal transmission, normalized to fix overall gain.
• Add colored AR(1) noise with optional slow gain modulation to mimic recording disturbances, with SNR control via RMS scaling.
• Introduce cycle-to-cycle variability by sampling theta^(k) from admissible ranges, enabling long-sequence synthesis.
• Specify heart-rate through RR series with weak HRV and optional time-stretching to align cycles.
• Provide two usage modes: preset-driven interactive synthesis and configuration-driven batch generation.

Experimental results

Research questions

• RQ1 Can a data-driven, cycle-wise parametric model reproduce both temporal and spectral characteristics of real abdominal fPCG?
• RQ2 How well does the simulator match real fPCG in envelope-based temporal structure and frequency-domain characteristics under identical preprocessing?
• RQ3 What are the data-driven parameter ranges and correlations that support realistic, reproducible long-sequence fPCG generation?
• RQ4 To what extent can a simple LTI transmission model and noise scheme capture the variability seen in real abdominal recordings?

Key findings

• The simulator reproduces dominant envelope morphology and beat-to-beat modulation in multi-cycle envelopes of real fPCG.
• Cycle-averaged envelope ACFs and PSDs of simulated signals align with those from real recordings within the tested preprocessing pipeline.
• Fitted per-cycle S1/S2 parameters show interpretable correlations, informing calibrated sampling ranges for generation.
• Long-sequence synthesis preserves RR structure and S1–S2 timing constraints, with residual discrepancies attributable to cycle boundary detection and simplified propagation/noise models.
• Envelope-based temporal statistics and frequency-domain characteristics (low-to-mid frequency content) are maintained in simulation, supporting its use for algorithm development and benchmarking.
• The software is released as both a Python package and an interactive web front end, enabling reproducible fPCG benchmarking under controlled conditions.