The reviewed record of science sign in
Pith

arxiv: 2605.28432 · v1 · pith:AABCPADQ · submitted 2026-05-27 · eess.SP

Transformer-Based Heartbeat Monitoring with FMCW Radar Under Random Body Motion

Matteo Pardi , Amir Hosein Oveis , Saba Kharabadze , Ajeet Kumar This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-06-29 10:24 UTCgrok-4.3pith:AABCPADQrecord.jsonopen to challenge →

classification eess.SP
keywords FMCW radarheartbeat monitoringtransformer networkrandom body motionPPG reconstructioncontactless vital signs
0
0 comments X

The pith

A CNN-Transformer network reconstructs PPG-like signals from 77 GHz FMCW radar data even under random body motion.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents a hybrid system that first applies model-based processing to extract chest displacement and high-level motion features from raw FMCW radar returns, then passes those features through a CNN-Transformer network to produce a reconstructed photoplethysmography waveform. Training uses a simultaneously recorded PPG trace as the target signal. The approach is tested on the IEEE AESS Radar Challenge datasets across stationary, deep-breathing, and random-body-motion conditions, where it produces usable average heart rate and heart-rate-variability estimates in cases where prior methods break down and records the highest aggregate score among the compared techniques.

Core claim

The proposed architecture reliably reconstructs the PPG signal in all scenarios, achieving high fidelity in controlled conditions and maintaining robust performance under motion. This enables reliable average heart rate (AHR) and heart rate variability (HRV) estimation even where benchmark methods fail, and leads to the highest total score among the compared approaches.

What carries the argument

Hybrid framework that combines model-based extraction of chest displacement and motion features with a CNN-Transformer network trained to map those features onto a synchronized PPG waveform.

If this is right

  • Reliable AHR and HRV estimates become available from radar in stationary, deep-breathing, and random-body-motion conditions.
  • The method outperforms existing benchmark techniques on the official IEEE AESS Radar Challenge figures of merit.
  • Contactless cardiac monitoring remains feasible when respiration and body motion are present.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same feature-extraction-plus-transformer pipeline could be retrained on other reference waveforms if a different ground-truth sensor were substituted.
  • Deployment on embedded radar hardware would require quantifying the computational cost of the CNN-Transformer stage under real-time constraints.

Load-bearing premise

A synchronized PPG signal supplies accurate ground truth for the cardiac component present in the radar chest-displacement signal.

What would settle it

Independent ECG recordings taken simultaneously with the radar under random body motion; if the reconstructed waveform deviates substantially from the ECG-derived heartbeat intervals, the reconstruction claim would be refuted.

Figures

Figures reproduced from arXiv: 2605.28432 by Ajeet Kumar, Amir Hosein Oveis, Matteo Pardi, Saba Kharabadze.

Figure 1
Figure 1. Figure 1: Overview of the chest-motion information embedded [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Let the complex radar data acquired over a CPI be [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figure 2
Figure 2. Figure 2: End-to-end scheme of the proposed method. Starting from the radar range–time matrix, the chest range bin is tracked and [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Schematic of the signal processing steps used to extract [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Feature representation obtained from a sample CPI. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Estimated PPG signal (CNN–Transformer output) com [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
read the original abstract

Millimeter-wave Frequency Modulated Continuous Wave (FMCW) radar enables contactless cardiac monitoring, but heartbeat estimation becomes challenging when respiration and random body motion (RBM) distort the radar signal. In this paper, we propose a hybrid framework for 77 GHz FMCW radar that combines model-based signal processing with a Convolutional Neural Network (CNN)-Transformer network. The first block extracts chest displacement and constructs meaningful high-level motion features from raw radar data, while the second block reconstructs a photoplethysmography (PPG)-like signal from the extracted features. In this study, a synchronized PPG signal is used as the ground truth for heartbeat monitoring in supervised training. The method is evaluated following the IEEE AESS Radar Challenge Problem I protocol using the official datasets and figures of merit across three motion scenarios: stationary, deep breathing, and RBM. Results show that the proposed architecture reliably reconstructs the PPG signal in all scenarios, achieving high fidelity in controlled conditions and maintaining robust performance under motion. This enables reliable average heart rate (AHR) and heart rate variability (HRV) estimation even where benchmark methods fail, and leads to the highest total score among the compared approaches.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 1 minor

Summary. The manuscript presents a hybrid framework for 77 GHz FMCW radar heartbeat monitoring that combines model-based preprocessing to extract chest displacement and high-level motion features with a CNN-Transformer network to reconstruct a PPG-like waveform. Supervised training uses a synchronized PPG signal as ground truth. Evaluation follows the official IEEE AESS Radar Challenge Problem I protocol and datasets across stationary, deep breathing, and random body motion (RBM) scenarios, claiming reliable PPG reconstruction, robust AHR/HRV estimation where benchmarks fail, and the highest total score among compared methods.

Significance. If the performance claims hold under rigorous validation, the work would advance contactless vital-signs monitoring by demonstrating a practical hybrid approach that maintains fidelity under RBM, where purely model-based or end-to-end methods often degrade; the use of an external challenge protocol and independent PPG reference is a methodological strength.

major comments (3)
  1. Abstract and Evaluation section: the claim of superior performance and 'highest total score' is reported without quantitative error bars, confidence intervals, or statistical significance tests on the figures of merit, leaving the central performance claim only partially supported.
  2. Abstract: the supervised training objective relies on a synchronized PPG signal as ground truth for the cardiac component in the radar displacement signal, yet no analysis or validation is provided of the degree to which mechanical chest motion and peripheral blood-volume pulsation remain equivalent under RBM.
  3. Evaluation section: no ablation studies are presented to quantify the contribution of the Transformer block versus the model-based feature extraction or to test sensitivity to the chosen loss function and data-exclusion rules.
minor comments (1)
  1. Abstract: specify the exact composition of the 'total score' (weights on AHR, HRV, and waveform fidelity) and the official challenge figures of merit.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments. We provide point-by-point responses below and have revised the manuscript to strengthen the statistical reporting and include ablation studies.

read point-by-point responses

  1. Referee: Abstract and Evaluation section: the claim of superior performance and 'highest total score' is reported without quantitative error bars, confidence intervals, or statistical significance tests on the figures of merit, leaving the central performance claim only partially supported.

    Authors: We agree that the performance claims would benefit from statistical support. In the revised manuscript, error bars (standard deviation across subjects), 95% confidence intervals, and results of paired statistical tests (t-tests) comparing our method to benchmarks have been added to the relevant figures and tables in the Evaluation section. The Abstract has been updated to reference these additions. revision: yes

  2. Referee: Abstract: the supervised training objective relies on a synchronized PPG signal as ground truth for the cardiac component in the radar displacement signal, yet no analysis or validation is provided of the degree to which mechanical chest motion and peripheral blood-volume pulsation remain equivalent under RBM.

    Authors: The evaluation strictly follows the IEEE AESS Radar Challenge protocol, which designates synchronized PPG as the reference for all submitted methods. Our model learns the radar-to-PPG mapping under this protocol. A dedicated biomechanical validation of equivalence between chest displacement and peripheral PPG specifically under RBM is not provided, as it lies outside the paper's scope and would require separate multi-modal experiments. A limitations paragraph has been added to the Discussion. revision: partial

  3. Referee: Evaluation section: no ablation studies are presented to quantify the contribution of the Transformer block versus the model-based feature extraction or to test sensitivity to the chosen loss function and data-exclusion rules.

    Authors: We acknowledge this omission. The revised manuscript adds a dedicated ablation subsection with three experiments: (i) full model vs. CNN-only and model-based-only variants, (ii) alternative loss functions, and (iii) varying data-exclusion thresholds. Quantitative results appear in a new table, confirming the Transformer's contribution and sensitivity to design choices. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation or evaluation chain

full rationale

The paper trains a CNN-Transformer to regress a PPG-like waveform from model-based radar features, using an external synchronized PPG reference as supervised ground truth and evaluating AHR/HRV via the independent IEEE AESS Radar Challenge protocol on official datasets. No self-definitional mappings, fitted inputs renamed as predictions, load-bearing self-citations, or ansatz smuggling appear in the provided text; the central claim rests on empirical performance against benchmarks rather than reducing to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that PPG provides a suitable supervised target for radar-derived cardiac signals; no free parameters or invented entities are mentioned.

axioms (1)
  • domain assumption Synchronized PPG signal is a valid ground truth for radar-based heartbeat estimation.
    Explicitly used for supervised training as stated in the abstract.

pith-pipeline@v0.9.1-grok · 5748 in / 1118 out tokens · 30285 ms · 2026-06-29T10:24:49.605032+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

19 extracted references · 2 canonical work pages · 1 internal anchor

  1. [1]

    Short Range De- vices (SRD); Automotive Short Range Radar operating in the 77 GHz to 81 GHz band; Harmonised Standard,

    European Telecommunications Standards Institute, “Short Range De- vices (SRD); Automotive Short Range Radar operating in the 77 GHz to 81 GHz band; Harmonised Standard,” ETSI EN 302 264 v1.1.1, 2008

    2008

  2. [2]

    Amendment of the Commis- sion’s Rules Regarding the 76–81 GHz Band,

    Federal Communications Commission, “Amendment of the Commis- sion’s Rules Regarding the 76–81 GHz Band,” FCC Report and Order, ET Docket No. 15-26, 2017

    2017

  3. [3]

    An Overview of Algorithms for Contactless Cardiac Feature Extraction from Radar Signals: Advances and Chal- lenges,

    Yuanyuan Zhang et al., “An Overview of Algorithms for Contactless Cardiac Feature Extraction from Radar Signals: Advances and Chal- lenges,”IEEE Transactions on Instrumentation and Measurement, vol. 72, pp. 1–20, 2023, arXiv:2508.02122 [eess]

    work page arXiv 2023

  4. [4]

    Noncontact Cardiac Parameters Estimation Using Radar Acoustics for Healthcare IoT,

    Yu Rong, Isabella Lenz, and Daniel W. Bliss, “Noncontact Cardiac Parameters Estimation Using Radar Acoustics for Healthcare IoT,”IEEE Internet of Things Journal, vol. 11, no. 5, pp. 7630–7639, Mar. 2024

    2024

  5. [5]

    Motion-robust non-contact heartbeat sensing using radar acoustics for healthcare iot,

    Yu Rong, Kawon Han, and Daniel W. Bliss, “Motion-robust non-contact heartbeat sensing using radar acoustics for healthcare iot,”IEEE Internet of Things Journal, pp. 1–1, 2026

    2026

  6. [6]

    Fast Acquisition of Heart Rate in Non- contact Vital Sign Radar Measurement Using Time-Window-Variation Technique,

    Jianxuan Tu and Jenshan Lin, “Fast Acquisition of Heart Rate in Non- contact Vital Sign Radar Measurement Using Time-Window-Variation Technique,”IEEE Transactions on Instrumentation and Measurement, vol. 65, no. 1, pp. 112–122, Jan. 2016

    2016

  7. [7]

    Vital signs monitoring of multiple people using a FMCW millimeter-wave sensor,

    Adeel Ahmad et al., “Vital signs monitoring of multiple people using a FMCW millimeter-wave sensor,” in2018 IEEE Radar Conference (RadarConf18), Oklahoma City, OK, Apr. 2018, pp. 1450–1455, IEEE

    2018

  8. [8]

    Enhancement of Remote Vital Sign Monitoring Detection Accuracy Using Multiple-Input Multiple-Output 77 GHz FMCW Radar,

    Toan K. V o Dai et al., “Enhancement of Remote Vital Sign Monitoring Detection Accuracy Using Multiple-Input Multiple-Output 77 GHz FMCW Radar,”IEEE Journal of Electromagnetics, RF and Microwaves in Medicine and Biology, vol. 6, no. 1, pp. 111–122, Mar. 2022

    2022

  9. [9]

    Radar-Based High-Accuracy Cardiac Activity Sensing,

    Weijie Xia, Yi Li, and Shiqi Dong, “Radar-Based High-Accuracy Cardiac Activity Sensing,”IEEE Transactions on Instrumentation and Measurement, vol. 70, pp. 1–13, 2021

    2021

  10. [10]

    Fmcw-based contactless heart rate monitoring,

    Zhanjun Hao et al., “Fmcw-based contactless heart rate monitoring,” Scientific Reports, vol. 15, no. 1, pp. 2616, 2025

    2025

  11. [11]

    A radar vital signs detection method in complex environments,

    Chaoyan Zhang et al., “A radar vital signs detection method in complex environments,”Scientific Reports, vol. 16, no. 1, pp. 2333, 2026

    2026

  12. [12]

    A Supervised Machine Learning Algorithm for Heart-Rate Detection Using Doppler Motion-Sensing Radar,

    Justin Saluja, Joaquin Casanova, and Jenshan Lin, “A Supervised Machine Learning Algorithm for Heart-Rate Detection Using Doppler Motion-Sensing Radar,”IEEE Journal of Electromagnetics, RF and Microwaves in Medicine and Biology, vol. 4, no. 1, pp. 45–51, Mar. 2020

    2020

  13. [13]

    Heart Rate Variability Monitoring Based on Doppler Radar Using Deep Learning,

    Sha Yuan et al., “Heart Rate Variability Monitoring Based on Doppler Radar Using Deep Learning,”Sensors, vol. 24, no. 7, pp. 2026, Jan. 2024, Publisher: Multidisciplinary Digital Publishing Institute

    2026

  14. [14]

    Segmentation of Radar-Recorded Heart Sound Signals Using Bidirectional LSTM Networks,

    Kilin Shi et al., “Segmentation of Radar-Recorded Heart Sound Signals Using Bidirectional LSTM Networks,” in2019 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), Berlin, Germany, July 2019, pp. 6677–6680, IEEE

    2019

  15. [15]

    Multi-Feature Transformer-Based Learning for Remote and Accurate Vital Sign Es- timation Using mmWave FMCW Radar,

    Yuh-Shyan Chen, Jhuo-Han Lyu, and Chih-Shun Hsu, “Multi-Feature Transformer-Based Learning for Remote and Accurate Vital Sign Es- timation Using mmWave FMCW Radar,” in2025 IEEE Wireless Communications and Networking Conference (WCNC), Milan, Italy, Mar. 2025, pp. 1–6, IEEE

    2025

  16. [16]

    Radar-based heartbeat monitoring in dynamic scenar- ios,

    Yu Rong et al., “Radar-based heartbeat monitoring in dynamic scenar- ios,”IEEE Aerospace and Electronic Systems Society (AESS) Challenge Problem I, 2025, Supported by IEEE AESS and 2025 IEEE International Radar Conference

    2025

  17. [17]

    Tune: A Research Platform for Distributed Model Selection and Training

    Richard Liaw et al., “Tune: A research platform for distributed model selection and training,”arXiv preprint arXiv:1807.05118, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  18. [18]

    A system for massively parallel hyperparameter tuning,

    Liam Li et al., “A system for massively parallel hyperparameter tuning,” inProceedings of the Third Conference on Machine Learning and Systems, MLSys 2020, Austin, TX, USA, March 2-4, 2020, Inderjit S. Dhillon, Dimitris S. Papailiopoulos, and Vivienne Sze, Eds. 2020, mlsys.org

    2020

  19. [19]

    Making a science of model search: Hyperparameter optimization in hundreds of dimensions for vision architectures,

    James Bergstra, Daniel Yamins, and David Cox, “Making a science of model search: Hyperparameter optimization in hundreds of dimensions for vision architectures,” inInternational conference on machine learning. PMLR, 2013, pp. 115–123

    2013