---
title: Multi-Path Vital Sign Sensing
url: https://www.emergentmind.com/topics/multi-path-vital-sign-sensing
type: topic
---

# Multi-Path Vital Sign Sensing

Multi-path vital sign sensing denotes the estimation of respiration and heartbeat from wireless or photonic measurements in which multiple propagation paths, multiple body scattering points, or multiple sensing viewpoints are explicitly exploited rather than treated only as interference. In this setting, chest micro-motion modulates phase, amplitude, channel impulse response, or micro-Doppler energy along direct and reflected paths, and system design centers on preserving, separating, and fusing those paths across delay, angle, frequency, subcarrier, or spatial dimensions so that cardiopulmonary motion remains observable under non-line-of-sight (NLOS), unfavorable orientation, subject motion, and multiperson coexistence [2310.05507][2303.13816][2509.10088].

## 1. Physical basis and signal models

At the propagation level, multi-path vital sign sensing is governed by path-dependent modulation of the received field by chest displacement. In MEDUSA’s IR-UWB formulation, the complex baseband received signal for transmitter \(i\), receiver \(j\), and path \(p\) is modeled as
\[
s_{ij}(t)=\sum_p \alpha_{ij,p} e^{j(2\pi f_c\tau_{ij,p}+\phi_{ij,p})} e^{j\,\phi_p(t)} + n_{ij}(t),
\]
with
\[
\phi_p(t)=\frac{4\pi}{\lambda} d_p(t),
\qquad
d(t)=A_r \sin(2\pi f_r t)+A_h \sin(2\pi f_h t),
\]
where \(f_r \in [0.1,0.5]\) Hz and \(f_h \in [1,2]\) Hz. The path projection is explicitly bistatic, \(d_p(t)=(u_{tx}+u_{rx})\!\cdot x(t)\), so different paths observe different effective amplitudes and phases even for the same underlying chest motion [2310.05507].

Bistatic OFDM sensing makes the same geometry dependence explicit in another form. In the 26.5 GHz OFDM JCAS system, the phase sensitivity of path \(p\) is
\[
\Delta \phi_p(t)\approx \frac{2\pi}{\lambda} x(t)\,[\cos\theta_{tx,p}+\cos\theta_{rx,p}],
\]
with the monostatic special case \(\Delta \phi(t)\approx \frac{4\pi}{\lambda}x(t)\). This relation explains why heart and breathing detection depend strongly on aspect angle, why NLOS reflections can still encode vital signs, and why a path that is geometrically weak for one orientation can become useful when indirect reflections change the effective incidence and scattering directions [2509.11767].

WiFi CSI formulations express the same phenomenon in frequency-selective multipath form. For OFDM subcarrier \(k\),
\[
H_k(t)=\sum_p \alpha_{k,p}(t)e^{-j2\pi f_k \tau_p(t)} = H_{s,k}+H_{d,k}(t),
\]
and first-order differentiation removes the static component \(H_{s,k}\), isolating the motion-modulated part. This is important because indoor wireless vital sign sensing rarely observes a single specular return; instead it observes a superposition of static clutter, chest-coupled reflections, and body-motion-induced path changes whose observability varies across subcarriers and antenna streams [2003.09386].

## 2. Architectures that exploit multipath rather than suppress it

Distributed coherent radar is one major architectural response to the fragility of single-view sensing. MEDUSA begins from a \(16\times16\) antenna array that can be partitioned into dispersed coherent sub-arrays, such as four \(4\times4\) sub-arrays or eight \(2\times8/8\times2\) sub-arrays, and uses low-power IR-UWB front-ends with per-element range up to about 6 m. Its central claim is that fully synchronized distributed MIMO arrays create multi-view coverage so that some Tx–Rx paths remain chest-coupled even when others are corrupted by blockage, motion, or orientation changes; this is the architectural basis for coherent multi-path vital sign sensing in realistic rooms [2310.05507].

A second architectural line treats the chest itself as a multi-scattering object rather than a point target. Pi-ViMo formalizes a multi-scattering point model in which the torso spans multiple range bins and angles, especially at close range, and therefore single-bin selection is structurally lossy. Its modified CFAR-based candidate selection, in-band/out-of-band bin qualification, and phase-aligned coherent combining convert multi-bin scattering diversity into an SNR gain and make estimation possible for subjects at any location in the radar field of view under micro-level random body movements (RBM) [2303.13816].

Propagation control can also be made explicit through reconfigurable surfaces. The space-time-coding RIS system uses a 1-bit, \(32\times32\)-element RIS at approximately 3.5 GHz to generate frequency-orthogonal harmonic beams, assigning a distinct harmonic to each detected person and thereby separating intertwined echoes at the physical layer before baseband decomposition. A related RIS-assisted radar configuration at 7.15 GHz introduces a second, controllable sensing path in addition to the direct radar path; in the reported geometry, the direct path impinges at \(\theta_d \approx 78.75^\circ\) while the RIS-reflected path is designed for near-normal incidence \(\theta_r \approx 0^\circ\), making respiration observable when the direct path is nearly insensitive to chest motion [2401.07422][2509.10088].

Photonic implementations extend the same principle to distributed sensing by engineering path delays. A dual-domain microwave photonic radar uses optical distribution and photonic de-chirping so that distinct fiber lengths or geometric delays yield distinct IF center frequencies for different subjects or access points; the system simultaneously monitored respiratory and pulse rate of two male volunteers. A separate WDM-based architecture integrates contact FBG sensing and contactless radar generation/de-chirping on the same optical channel and demonstrated simultaneous respiration and heartbeat monitoring of three people, with the distributed optical network providing the scaling mechanism [2510.26283][2405.15271].

## 3. Path selection, separation, and fusion algorithms

The canonical processing problem is to identify which paths carry clean cardiopulmonary modulation and which carry clutter, bulk motion, or unstable multipath. MEDUSA processes per-pair channel impulse responses \(h_{ij}(\tau,t)\), extracts phase traces \(\theta_{ij,p}(t)=\arg\{h_{ij}(\tau_{ij,p},t)\}\), and uses delay, angle, and coherence to prefer chest-coupled paths. The coherence metric
\[
C_{ij,p}=\frac{\left|\sum_t r_{ij,p}(t)e^{-j2\pi f_r t}\right|}{\sum_t |r_{ij,p}(t)|}
\]
formalizes this intuition, while synchronized virtual-array beamforming and an attention-based self-supervised contrastive model weight Tx–Rx pairs and paths according to periodicity and cross-view agreement [2310.05507].

Pi-ViMo performs path selection at the range-bin level. Candidate bins are found by modified CFAR, then qualified by whether the peak frequency lies in the respiration band \([0.1,0.8]\) Hz or heartbeat band \([0.8,2]\) Hz with \(E_{in}/E_{out}\ge 5\). Cross-correlations among accepted bins estimate delays and phase lags, after which phase-rotated coherent summation produces a single combined displacement trace. The final respiration and heartbeat separation is not spectral-only: it uses physiology-inspired template matching, with respiration modeled by an RC process and heartbeat by a Van der Pol relaxation oscillator, thereby constraining solutions when RBM distorts naive Fourier peaks [2303.13816].

Other frameworks replace phase demodulation with energy-domain observables. The 77-GHz FMCW micro-Doppler energy approach forms per-target energy streams
\[
E(i)=\sum_{rx=1}^{N_{RX}} \sum_{r=r_0-\Delta r}^{r_0+\Delta r} \sum_{d=d_0-\Delta d}^{d_0+\Delta d} |S_{doppler}(rx,d,r,i)|^2,
\]
then uses STAP to suppress clutter and coherent multipath subspaces, MUSIC for high-resolution AoA estimation, and adaptive spectral filtering to estimate respiration and heart rates. Its explicit rationale is that energy aggregates over frequency and space, so phase fluctuations induced by multipath tend to average out while slow cardiopulmonary energy variations persist [2510.19639].

At the dataset and benchmarking level, the Sensing Dataset Protocol preserves multipath structure instead of collapsing it during preprocessing. It maps heterogeneous measurements into canonical tensors \(X\in\mathbb{C}^{A\times K\times T}\), applies phase sanitization and alignment, and then uses CP-ALS pooling with
\[
\mathcal{X}\approx \sum_{r=1}^{R} a_r \circ b_r \circ c_r
\]
to retain separable spatial, spectral, and temporal components. In this view, the temporal factors \(c_r\) can be treated as candidate vital-sign rhythms and the spatial/spectral factors as multipath signatures, making the representation explicitly path-preserving across modalities [2512.12180].

## 4. NLOS operation, multiperson sensing, and in-the-wild performance

A defining aim of multi-path vital sign sensing is robustness outside single-subject, line-of-sight laboratory conditions. MEDUSA was proposed precisely because existing radar systems struggled with NLOS, changes in location and orientation, subject movement, and simultaneous tracking of multiple targets. Its evaluation involved 21 participants and over 200 hours of collected data, totaling 3.75 TB, and reported an average gain of 20% compared to existing systems employing COTS radar sensors, which the paper attributes to the spatial diversity of synchronized distributed MIMO in familiar and unfamiliar indoor environments [2310.05507].

For explicit multi-target radar, the 77-GHz micro-Doppler energy framework reports accurate detection and separation of up to four targets within 5 m, with mean absolute errors of 1.2 beats per minute for respiration and 2.3 beats per minute for heart rate, average tracking accuracy of approximately 96.7%, and consistent operation at 10 fps on an i7 CPU. The same work reports that targets down to approximately 0.8 m separation can be resolved, which is significant because vital sign mixtures become difficult to disentangle when range, angle, and Doppler support overlap [2510.19639].

RIS-based systems address the same multiperson problem by altering the propagation channel itself. The STC RIS prototype monitored vital signs of up to four persons simultaneously; the abstract reports RR and HR estimation errors below 1 RPM and 5 BPM, while the detailed summary reports average RR error of approximately 1.1 RPM and average HR error of approximately 4.7 BPM across trials, with robustness in a corridor containing chairs, desks, and occasional passersby. The central distinction from algorithm-only separation is that each person is illuminated by a frequency-orthogonal harmonic beam, so the echoes are physically tagged before decomposition [2401.07422].

Phase-based OFDM JCAS demonstrates the same tradeoff in a different regime. At 26.5 GHz, NLOS experiments showed that breathing rate could still be reliably estimated, but heart rate was too weak to extract in the reported tests; two subjects at different ranges could be separated when 1 GHz bandwidth was used, and the farther subject’s HR could be misidentified because a breathing harmonic dominated the heart band. In WiFi CSI sleep monitoring, a much lower-frequency sensing substrate nonetheless remained viable across more than 550 hours of data from 5 users over 80 nights, with approximately 45% LOS and approximately 55% NLOS, median whole-night respiration error below 1.19 BPM, and 95th percentile below 1.9 BPM [2509.11767][2003.09386].

## 5. Misconceptions, limitations, and failure modes

A persistent misconception is that multipath is purely detrimental. The literature is more specific: uncontrolled multipath can cause phase jitter, fading, path mixing, and false harmonics, but deliberately selected or engineered multipath can provide diversity, alternative chest-sensitive views, and constructive combining. MEDUSA uses dispersed coherent sub-arrays to ensure that at least some paths retain high-SNR vital-sign modulation; Pi-ViMo uses multiple chest-area bins rather than a single “best” bin; RIS-assisted radar creates an additional path specifically to restore sensitivity when the direct path is geometrically weak. This suggests that the key distinction is not between “with” and “without” multipath, but between unmanaged superposition and controlled path exploitation [2310.05507][2303.13816][2509.10088].

Another misconception is that wider bandwidth automatically improves all aspects of vital-sign sensing. The OFDM JCAS experiments report no significant influence of bandwidth on phase-based vital sign extraction quality for a selected path, because the vital-sign information is encoded in phase; bandwidth chiefly determines range resolution and therefore the ability to separate multiple subjects or reflections. By contrast, the 77-GHz micro-Doppler energy framework depends on high-resolution range–Doppler processing and adaptive spatial filtering, so bandwidth and array processing directly affect target separation performance [2509.11767][2510.19639].

Across platforms, heart-rate estimation remains structurally harder than respiration. The reasons stated across the papers are consistent: heartbeat amplitudes are much smaller than respiratory motion; respiration harmonics contaminate the heart band; bulk motion masks micro-motion; and synchronization or phase instability can translate directly into displacement error. MEDUSA gives the residual-error relation
\[
\sigma_d \approx \frac{\lambda}{4\pi}\sigma_\phi,
\]
illustrating why coherent distributed sensing requires tight time/frequency/phase alignment. Pi-ViMo is limited to single-subject scenarios and targets micro-RBMs rather than macro-motions. The RIS-assisted 7.15 GHz system notes that heart-rate extraction is constrained by a 4 Hz slow-time sampling rate, which is borderline for the heart band. More generally, severe bulk motion, deep shadowing, insufficient STAP training snapshots, calibration errors, and overlapping multi-user returns remain standard failure modes across modalities [2310.05507][2303.13816][2510.19639][2509.10088].

## 6. Distributed implementations and benchmarked research directions

Recent work extends multi-path vital sign sensing beyond single-radar prototypes toward distributed infrastructures. The integrated FBG-plus-radar system places the FBG at the center of both pathways: chest wall motion modulates optical intensity for contact sensing and also participates in radar generation and photonic de-chirping for contactless sensing. The proof-of-concept simultaneously realized contact and contactless respiration and heartbeat monitoring of three people, with maximum absolute measurement errors of 1.6 respirations per minute and 2.3 beats per minute during 60 s monitoring, and the measurement error did not have an obvious change even when the monitoring time was decreased to 5 s [2405.15271].

Microwave-photonic radar makes delay engineering itself a multiplexing primitive. The dual-domain system reported average accuracies of 98.1% for pulse detection and 99.85% for respiratory detection in two-subject experiments, achieved rapid indirect blood-pressure estimation with a coefficient of determination of 0.905, and argued that low-loss optical-fiber distribution enables scaling into a distributed, non-contact, multiple-vital-sign monitoring platform. It also provided an explicit scalability expression and stated that, with the reported parameters, the system theoretically supports at least 517 simultaneously monitored individuals [2510.26283].

Standardization has become a parallel research requirement because multipath-aware pipelines are otherwise difficult to compare. The Sensing Dataset Protocol defines a canonical data-block schema, lightweight synchronization, frequency–time alignment, and CP-ALS pooling that preserve multipath, spectral, and temporal structure across modalities. Its benchmark uses a cross-user split and reports that variance drops by approximately 88% across seeds while maintaining competitive accuracy and latency. A plausible implication is that future progress in multi-path vital sign sensing will depend not only on better propagation control and fusion algorithms, but also on reproducible, modality-agnostic evaluation regimes that preserve rather than erase multipath diversity [2512.12180].

Source: https://www.emergentmind.com/topics/multi-path-vital-sign-sensing