Reconstruction of Respiratory Flow from an Impedance-Based Contactless Sensor System

Abstract

Conventional respiratory monitoring is often invasive, while most non-contact technologies like radar or cameras are limited to estimating respiratory rate, failing to reconstruct the detailed waveform of the respiratory flow itself. This gap limits their clinical utility for advanced diagnostics. We introduce a novel system that bridges this gap by combining a contactless, impedance-based sensor (the Thoraxmonitor) with a dedicated machine learning framework to directly reconstruct the full respiratory flow signal. Operating at 433 MHz, the system's antenna array detects subtle changes in thoracic impedance, which are then translated into a quantitative flow signal by a Multilayer Perceptron Regressor. Based on data from 17 subjects benchmarked against a gold-standard flowmeter, our system accurately detected 98% of respiratory cycles. It achieved remarkable precision in timing respiratory events, with mean deviations of + 60 ms (±79 ms) for inspiration and + 50 ms (±63 ms) for expiration, making it suitable for time-critical applications. While a systematic bias in absolute tidal volume prediction currently limits inter-subject comparisons, the system excels at tracking relative intra-subject changes. Crucially, our model quantifies its own reliability, providing an intrinsic self-assessment mechanism. This work demonstrates a significant step beyond simple rate detection towards comprehensive, comfortable, and reliable respiratory analysis in clinical and everyday settings.

Keywords: ISM radio band; contactless sensor; machine learning; monitoring; non-invasive; respiratory signal detection; soft sensor.

Figure 1

Illustration of the experimental setup and signal processing pipeline. The top left image shows the subject seated with the contactless sensor system, capturing respiratory flow using a 433 $\mathrm{M}$$\mathrm{Hz}$ antenna array whereas simultaneously measuring the flow signal as a reference. The top right plot displays the flow signal, highlighting inspiration and expiration phases, as well as their onsets. The next section illustrates the process of modulated carrier wave demodulation to extract phase information. Subsequent plots present the raw flow and Q/I signals. The following section outlines the preprocessing pipeline applied to each subject or dataset, including resampling, bandpass filtering, transformation, and feature extraction. The final plot depicts the Monte Carlo process using a Multilayer Perceptron, showing the reconstructed flow signal along with reliability intervals.

Figure 2

Data acquisition workflow combining the Thoraxmonitor and a flowmeter. (a) The experimental setup with the Thoraxmonitor antennas and the concurrent reference flowmeter recording, highlighting the relative sensor placement and subject posture. The Thoraxmonitor electronics are mounted on the rear of the chair and connected via U.FL cables to a transmit (Tx) and receive (Rx) UHF antenna pair separated by 35 $\mathrm{c}$$\mathrm{m}$. The antennas are fixed 30 $\mathrm{c}$$\mathrm{m}$ above the seat level to couple through the thoracic region while the participant remains in everyday clothing. A nose mask with an inline flowmeter provides the reference flow signal, which, together with the Thoraxmonitor data, is recorded on a laptop positioned in front of the subject. (b) The electronics chain from the 433 $\mathrm{M}$$\mathrm{Hz}$ synthesizer to the digitization stage, illustrating the signal processing path that yields the in-phase and quadrature data used throughout this work. A synthesizer (LMX2571) generates the 433 $\mathrm{M}$$\mathrm{Hz}$ carrier signal, which is transmitted via the TX antenna. The signal received at the RX antenna, modulated by thoracic impedance changes, is processed by an I/Q demodulator (LTC5585). The resulting baseband signals are amplified by differential amplifiers (OPA2316ID) and digitized by the Analog-to-Digital Converter (ADC) of an Arduino Due for further processing.

Figure 3

Series of graphical illustrations categorizing respiratory conditions and depicting the model’s performance. For each category, the figure shows the annotated reference sensor flow (green line), the predicted flow (orange line), the reliability score (grey line), and the reliability threshold (red dashed line). Intervals where the reliability score exceeds the threshold are highlighted with a red background in the predicted flow plot. Both, the reference and predicted flow are annotated with the onset of inspiration (red triangle) and expiration (blue triangle), calculated as described in Section 3.1.3.

Figure 4

Violin plots comparing the duration of respiratory cycles measured and predicted across all subjects and the corresponding errors as a boxplot below. These plots reveal the variability in breathing patterns and the model’s capability to accurately predict respiratory cycle durations.

Figure 5

Comparative analysis of measured and predicted tidal volumes in different subjects, represented by violin plots. The plots illustrate the performance of the model in estimating tidal volumes and show a systematic bias in the predictions.

Figure 6

Bland–Altman plot showcasing the agreement between measured and predicted tidal volumes, identifying a systematic underestimation bias.