{"id":"1c8cef59-bbdf-4e0d-b3a7-bc7d374c10ca","arxiv_id":"2501.06755","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A MIMO FMCW radar framework with a custom phantom achieves multi-person localization and vital signs estimation, reporting 94% respiration and 87% heart rate accuracy within 2 bpm in multi-person human trials.","lead":"This paper presents a radar system that finds multiple people in a room and reads their breathing and heart rates without contact, using a custom mechanical dummy to test the system. The algorithm combines sparse signal recovery with a dictionary method that suppresses breathing harmonics, and it is tested in 12 phantom and 12 human trials.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Headline accuracies are not end-to-end: NCVSM is evaluated with true target locations, so the reported ASR/RMSE values do not verify the localization-plus-estimation pipeline.","rationale":"The reader's weakest assumption was stationarity (A-1), which is a genuine limitation. My primary concern is complementary but more directly tied to the headline numbers: the NCVSM evaluation protocol deliberately uses true locations, so the reported accuracies are component-level, not system-level. The reader's rationale does mention this as a weakness, but the reader's 'weakest assumption' field emphasizes the motion issue. I therefore mark partial agreement: the concerns are related, but I weight the oracle-location decoupling as the most load-bearing because it affects the interpretation of the central quantitative claim even for perfectly stationary subjects. In good faith, the paper has real strengths: the phantom is a genuine testbed contribution, RaLU-JSR is a sensible convex joint-sparse formulation, E-VSDR is coherently designed to suppress respiration harmonics, and the comparisons in Table III are internally consistent. The concern is not an internal mathematical contradiction; it is that the paper's strongest claim is stated as a full pipeline while the validation protocol separates the two stages. Keeping the reader's CONDITIONAL verdict is appropriate: the paper should be accepted only if the authors provide end-to-end results with estimated localizations and quantitative localization metrics across all trials, or explicitly scope the claims accordingly.","tokens_in":24642,"tokens_out":6977,"duration_ms":75030,"concrete_test":"Re-run all c4 multi-person human trials end-to-end: estimate S from the first Tloc=5 s with RaLU-JSR exactly as in Section V-B; extract {v_z} via (16)-(17) using that estimated support; apply E-VSDR; then recompute ASR2/3/4 and ARMSE for HR and RR and compare with the c4 rows of Table III. Also tabulate per-trial detection counts and range/angle errors for RaLU-JSR in all c3/c4 trials instead of only two example maps. If end-to-end ASR values remain within ~2 bpm thresholds (or within a few percentage points) of the oracle-location values, the central pipeline claim is supported; if they drop materially, the reported headline overstates the framework.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that the RaLU-JSR localization plus E-VSDR estimation pipeline reliably localizes multiple people and estimates their vital signs. The headline multi-person human accuracies (RR ASR2/3/4: 94.14/98.12/98.69%; HR ASR2/3/4: 87.10/94.12/95.54%; ARMSE 0.98/1.33 bpm) are not end-to-end results. Section V-C states that, for a fair comparison of NCVSM methods, 'we assume that all considered subjects were accurately detected and positioned' and uses the true locations of the examined subjects to extract thoracic vibrations. Consequently, Table III c4 quantifies E-VSDR given perfect localization, not the full framework. Localization is supported only by selected maps (Fig. 8, c3 trial #2; Fig. 11, c4 trial #3) and a textual claim that RaLU-JSR was the only method to detect all subjects; no detection rate, false-alarm rate, or position RMSE is reported across the 12 multi-person trials. If the support estimated by RaLU-JSR is off by one range/angle bin in any trial, the beamformer (16) will mix in clutter or a neighbor's vibration and E-VSDR estimates can degrade; the paper provides no way to quantify this from the reported numbers. This matters because the abstract and title promise a full framework, and a user would run it end-to-end, not with oracle locations. The related stationarity assumption A-1 (fixed support, no body movement) narrows the same issue: even under that scope, the claimed end-to-end accuracy is unvalidated.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proposes an end-to-end MIMO FMCW radar pipeline for multi-person localization and non-contact vital sign monitoring. The signal model in Eq. (10) is used to formulate localization as joint-sparse recovery solved by RaLU-JSR (Algorithm 1); the estimated support feeds the beamformer in Eq. (16) to extract thoracic Doppler phases, from which E-VSDR (Algorithm 2) estimates respiration and heart rates using harmonic-resilient dictionary recovery and adaptive temporal refinement. The authors also contribute a custom three-unit hardware phantom driven by recorded impedance signals and validate the pipeline in 12 phantom and 12 human trials. The headline multi-person human results are RR ASR2/3/4 of 94.14/98.12/98.69% and HR ASR2/3/4 of 87.10/94.12/95.54%, with ARMSE of 0.98 and 1.33 bpm.","tokens_in":25000,"tokens_out":5419,"duration_ms":57881,"significance":"The phantom is a genuine experimental contribution: it provides a repeatable, controllable testbed with realistic cardiopulmonary waveforms and was used to tune the algorithm before human experiments. The algorithmic ideas, especially the joint-sparse localization and dictionary-based harmonics suppression, are plausible, and the head-to-head comparison with six existing methods is a useful benchmark. If the full pipeline were validated end-to-end, the work would advance practical radar-based monitoring of multiple stationary people in cluttered indoor settings. At present, however, the strongest numerical claims are conditional on oracle localization, and the localization evidence is qualitative; the significance of the results therefore depends on completing the end-to-end evaluation.","major_comments":[{"comment":"The reported NCVSM metrics are not end-to-end. The text explicitly states that, for a fair comparison, all subjects were assumed to be accurately detected and positioned, and that the extracted thoracic vibrations used the true locations. Consequently, the headline ASR and ARMSE values in Table III validate E-VSDR conditional on perfect localization, not the full RaLU-JSR plus E-VSDR framework promised in the abstract and title. Since the beamformer in Eq. (16) depends directly on the support estimate, an off-by-one bin in the RaLU-JSR support could degrade the vital-sign estimates. Please add an end-to-end evaluation in which the support produced by Algorithm 1 is used in Eq. (16), and report the resulting ASR/RMSE; alternatively, if conditional results are intended, restrict the abstract and conclusion claims accordingly.","section":"Section V-C, Table III"},{"comment":"Localization success is asserted on the basis of illustrative maps from one multi-person phantom trial and one multi-person human trial, with statements such as 'only the proposed RaLU-JSR detects and positions all 3 subjects.' No detection rate, false-alarm rate, or position RMSE is reported across the 12 trials. This is load-bearing because localization errors propagate directly into the beamformer in Eq. (16) and therefore into the vital-sign estimates. Please report quantitative localization metrics over all trials, including per-trial detection/position errors, and discuss sensitivity to the peak-detection thresholds and to the regularization parameter gamma.","section":"Section V-B, Figs. 8 and 11"},{"comment":"The framework assumes that monitored individuals remain stationary, with only slight thoracic movements, and the human protocol asked subjects to breathe calmly and avoid large movements. Because the support is recovered once from the first five seconds and then used in a fixed beamformer, any body sway or repositioning breaks the joint-support assumption. The abstract's 'real-world, cluttered environments' and the term 'robust' therefore overstate the validated scope. Either add experiments with natural body movement or explicitly state in the abstract and conclusion that the results apply to stationary subjects.","section":"Section II-B, A-1; Section V-A"},{"comment":"The multi-person human results are based on nine subjects (three trials of three subjects), and the reported ASR and ARMSE values are point estimates with no confidence intervals or per-trial variability. Differences such as the HR ASR2 gap between E-VSDR (87.10%) and PhaseReg+ (71.58%) could, with this sample size, be subject to considerable sampling variability. Please provide per-trial results and confidence intervals (e.g., bootstrap or per-subject standard deviations) for the headline metrics, or present the uncertainty explicitly.","section":"Table III, Figs. 10 and 13"}],"minor_comments":[{"comment":"The expression for the vital-based spectral filter is dimensionally ambiguous: Pi is described as a length-L window, but it is multiplied elementwise with the L-by-N matrix F_L Y^(k)^T. Please specify the intended broadcasting or define Pi as a matrix.","section":"Eq. (13)"},{"comment":"The indexing of matrix B is inconsistent: Eq. (5) and Eq. (10) define B(p,k), while the text after Eq. (12) writes B(k,p) = exp(...). Please align the notation.","section":"Section III-A"},{"comment":"The supplementary material containing the single-person trial results is referenced but is not included with this submission; please make it available or summarize the single-person results in the main text.","section":"Section V-A"},{"comment":"The claim of 'angular error of less than 3 degrees' for the illustrated phantom trial should be accompanied by a description of how the angular error is measured and by the corresponding values for the other trials.","section":"Section V-D"},{"comment":"There are minor typographical and caption errors, including 'corrspondingly' in Section III-B and 'produces' in the caption of Fig. 8; a careful proofread would improve presentation.","section":"General"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is a competent, incremental extension of the authors' own VSDR work, and the custom impedance-driven phantom is the most distinctive contribution. If you build radar NCVSM validation rigs, the phantom is worth knowing about. The signal model and RaLU-JSR/E-VSDR pipeline are clearly specified, and the experimental design is honest in structure: external ECG/PPG/respiration-belt ground truth, consistent grids and bands, and baseline methods tested both with and without the refinement procedure. Twelve phantom and twelve human trials is a reasonable validation effort; using phantom trials to tune parameters before human sessions is good practice.\n\nThe stress-test concern holds up. The headline ASR/RMSE numbers in the abstract and Table III are not end-to-end. Section V-C says explicitly that, for a fair comparison, the NCVSM comparison assumes all subjects were accurately detected and positioned and uses true locations. So Table III quantifies E-VSDR given perfect localization, not the full framework. Localization support comes only from selected maps (Fig. 8 and Fig. 11) and a textual claim that RaLU-JSR was the only method to detect all subjects; no detection rate, false-alarm rate, or position RMSE is reported. If the estimated support is off by one range/angle bin, the beamformer in (16) can mix in a neighbor's vibration or clutter, and the downstream estimates degrade; the data as presented cannot quantify that. This is the load-bearing soft spot.\n\nOther soft spots, in proportion. The multi-person human evidence is three trials with nine subjects; the headline percentages are fragile and have no confidence intervals. Assumption A-1 is explicit (stationary subjects, fixed support), and the protocol asked subjects to avoid movement, so the robustness claim is really robustness to static clutter, not body motion; that scope is stated in the model but undersold by the title and abstract. No code or data are released, which limits independent replication. The adaptive refinement uses its own estimates as future search centers, a mild feedback loop, but the comparison against refined baselines mitigates it.\n\nThe math is coherent; the bilinear JSR extension is standard FISTA with a 3D soft-threshold, and the harmonic dictionary construction is well specified. I see no circular derivation or invented entities. Self-citation is not a problem here because the paper is an explicit extension of [18].\n\nWho benefits: radar/NCVSM researchers working on multi-person vital signs or on repeatable testbeds. It deserves a serious referee. The right outcome is a revise-and-resubmit: report end-to-end accuracies using estimated support, tabulate localization detection and position-error metrics, add confidence intervals, and ideally release code/data.","headline":"Solid incremental radar NCVSM paper with a genuinely useful phantom and a sensible harmonics-resilient estimator, but the headline accuracy numbers are not end-to-end because localization is verified separately and vital signs are extracted from true locations.","tokens_in":25518,"tokens_out":3046,"would_cite":true,"duration_ms":30555,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A multi-person radar pipeline that localizes stationary people in cluttered rooms and estimates respiration and heart rates via joint sparse recovery and harmonic-cancelling dictionary estimation, validated with a thoracic-motion phantom…","keywords":["frequency-modulated continuous wave radar","MIMO radar","multi-person localization","vital signs monitoring","joint sparse recovery","respiratory harmonics","phantom validation","non-contact sensing"],"falsifier":"Run the identical pipeline in a room where one of the three seated subjects shifts posture, sways, or briefly stands during the 30-second monitoring window, while the support is fixed from the first five seconds; if heart-rate accuracy within 2 bpm remains above 85%, the robustness claim survives, whereas a sharp drop would show that the claim depends on immobility.","tokens_in":24485,"feed_emoji":"📡","tokens_out":8684,"duration_ms":79986,"temperature":0.7,"pith_summary":"Non-contact radar monitoring of several people at once usually fails in cluttered rooms because static furniture reflections masquerade as people and because respiration harmonics swamp the weak heartbeat signal. This paper argues that both failures can be solved with a single pipeline: localize the thoraces by recovering a jointly sparse range-angle map using a 3D $\\ell^2$,1-regularized least-squares problem, then estimate vitals with a dictionary that explicitly subtracts respiration and its harmonics. The claim is supported by a custom hardware phantom that replays recorded thoracic impedance signals, and by human trials. In three-person trials the method reports respiration-rate accuracy of 94.14%, 98.12%, and 98.69% within 2, 3, and 4 breaths per minute, and heart-rate accuracy of 87.10%, 94.12%, and 95.54% within the same thresholds in beats per minute, with average RMSE of 0.98 and 1.33 bpm. If this holds, radar-based vital-sign monitoring becomes credible for waiting rooms and smart-home health sensing.","feed_headline":"Three people's heart and breath rates read remotely by radar","feed_subtitle":"Joint-sparse localization plus harmonic-resistant estimation gives 94-98% respiration accuracy.","key_machinery":"The load-bearing object is the joint-sparse bilinear signal model $Y_l = A X_l B + W_l$, in which each frame shares the same unknown support across slow time; RaLU-JSR recovers the sparse tensor $X$ by minimizing a 3D $\\ell_{2,1}$-regularized least-squares cost with a fast proximal-gradient acceleration. The second mechanism is the harmonic-resilient dictionary estimator E-VSDR, which builds respiration and heartbeat dictionaries on a 1-bpm grid, estimates respiration by a sparse peak, subtracts the respiration fundamental and all its harmonics that fall in the heartbeat band, and then selects the remaining sparse heartbeat tone. The hardware phantom, three vibration units driven by recorded thoracic impedance signals, functions as a repeatable ground-truth stand-in for human thoraces and was used to tune the algorithm before human trials.","core_discovery":"The paper's central claim is that multi-person vital-sign monitoring by MIMO FMCW radar reduces to two coupled estimation problems that can be solved robustly in clutter: recovering the joint sparse range-angle support of stationary people, and estimating each person's respiration and heartbeat from the phase of the support beamformer while actively cancelling respiration harmonics. Localization is performed once on the first five seconds using RaLU-JSR, which solves for the 3D tensor of complex amplitudes from Y_l = AX_lB + W_l with a joint $\\ell^2$,1 penalty across slow-time frames and a vital-frequency clutter filter. Vital signs are then estimated continuously by E-VSDR, an extension of the Vital Signs Dictionary Recovery method that splits the heartbeat band into interfered and non-interfered frequencies using the estimated respiration rate, cancels the harmonics by least squares, and selects the remaining sparse heartbeat tone. The authors report that, in the multi-person human trials, only the proposed localization detected and positioned all three subjects, and the E-VSDR estimator outperformed FFT, phase regression, and orthogonal-projection baselines, with or without the added refinement, in both average success rates and RMSE.","pith_inferences":["The stationary-subject assumption (A-1) is the real boundary of the claimed robustness; a natural extension is online support re-estimation that tracks small posture shifts, and this paper does not yet demonstrate that.","The phantom could be reused as a standardized benchmark for other radar sensing tasks, since it generates ground-truth thoracic motion with known cardiopulmonary content, but the paper only demonstrates its use for this pipeline.","On populations with irregular breathing patterns or arrhythmias, the harmonic-cancellation step may subtract energy near the true heartbeat; testing E-VSDR against the pathological signals the phantom can replay would settle whether the reported margins persist.","The dictionary-based estimator's advantage over FFT baselines should grow as heartbeat-to-noise ratio falls, so the largest performance gap is expected precisely in the noisiest real deployments."],"forward_implications":["A single in-phase channel can be used for both localization and Doppler extraction, sidestepping I/Q imbalance without sacrificing accuracy.","A harmonics-aware dictionary estimator provides heart-rate estimates in multi-person, cluttered settings at roughly 87% success within 2 bpm, a level that makes radar plausible for unsupervised waiting-room monitoring.","The phantom provides a repeatable validation path for radar vital-sign systems, so algorithmic parameters can be tuned and claims compared without recruiting human subjects each time.","If the reported accuracy holds, continuous monitoring windows of 30 s at 0.05 s intervals can produce stable vital-sign curves over a two-minute session.","The localization step needs only five seconds of data, after which continuous monitoring can reuse the fixed support."],"supporting_citations":[{"why":"Supplies the sparsity-based VSDR vital-sign estimator and single-channel signal model that E-VSDR extends.","marker":"[18]"},{"why":"Supplies the sparse non-contact localization approach and SIMO model generalized here to MIMO.","marker":"[19]"},{"why":"Supplies the recorded thoracic impedance and ECG signals that the phantom replays and that serve as ground truth.","marker":"[44]"},{"why":"Supplies the joint sparse reconstruction algorithm that RaLU-JSR adapts to the bilinear radar model.","marker":"[47]"},{"why":"Supplies the fast proximal-gradient acceleration and Lipschitz step choice used to solve the l2,1-regularized least-squares problem.","marker":"[48]"},{"why":"Provides the FFT and arctangent-demodulation baseline and phase-unwrapping procedure used in Doppler extraction.","marker":"[13]"},{"why":"Provides the orthogonal-projection harmonic suppression method compared against and extended by E-VSDR.","marker":"[36]"},{"why":"Provides the phase-regression vital-sign estimation baseline compared in the trials.","marker":"[40]"},{"why":"Provides the angle-FFT localization method used as a baseline in the localization comparisons.","marker":"[31]"}],"fun_headline_variants":["Radar with phantom-trained algorithm tracks multiple vital signs in clutter","MIMO FMCW radar reads multiple heart and breath rates via joint sparsity","Harmonic-resistant radar estimates multiple vital signs with 94-98% accuracy","Phantom-assisted radar framework localizes and monitors multiple people"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Subjects remain stationary throughout the session, with only their chests moving from breathing and heartbeat, and the locations found in the first five seconds stay valid for all later estimates; the trials also instructed subjects to breathe calmly and avoid large movements.","fun_headline_variants_meta":{"raw":{"variants":["Radar with phantom-trained algorithm tracks multiple vital signs in clutter","MIMO FMCW radar reads multiple heart and breath rates via joint sparsity","Harmonic-resistant radar estimates multiple vital signs with 94-98% accuracy","Phantom-assisted radar framework localizes and monitors multiple people"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000448,"raw_usage":{"total_tokens":2339,"prompt_tokens":1100,"completion_tokens":1239,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":716,"completion_tokens_details":{"reasoning_tokens":1162}},"tokens_in":716,"tokens_out":1239,"duration_ms":10783,"temperature":1.0,"reasoning_tokens":1162,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T20:49:59.922714+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the identical pipeline in a room where one of the three seated subjects shifts posture, sways, or briefly stands during the 30-second monitoring window, while the support is fixed from the first five seconds; if heart-rate accuracy within 2 bpm remains above 85%, the robustness claim survives, whereas a sharp drop would show that the claim depends on immobility.","supporting_citations":[{"cited_title":"Sparsity-based multi-person non-contact vital signs monitoring via fmcw radar,","cited_arxiv_id":null,"evidence_quote":"Supplies the sparsity-based VSDR vital-sign estimator and single-channel signal model that E-VSDR extends."},{"cited_title":"Sparse non-contact multiple people localization and vital signs monitoring via fmcw radar,","cited_arxiv_id":null,"evidence_quote":"Supplies the sparse non-contact localization approach and SIMO model generalized here to MIMO."},{"cited_title":"A dataset of clinically recorded radar vital signs with synchronised reference sensor signals,","cited_arxiv_id":null,"evidence_quote":"Supplies the recorded thoracic impedance and ECG signals that the phantom replays and that serve as ground truth."},{"cited_title":"Rapid quantum image scanning microscopy by joint sparse reconstruction,","cited_arxiv_id":null,"evidence_quote":"Supplies the joint sparse reconstruction algorithm that RaLU-JSR adapts to the bilinear radar model."},{"cited_title":"Remote monitoring of human vital signs using mm-wave fmcw radar,","cited_arxiv_id":null,"evidence_quote":"Provides the FFT and arctangent-demodulation baseline and phase-unwrapping procedure used in Doppler extraction."},{"cited_title":"Vital signs detection with difference beamforming and orthogonal projection filter based on simo-fmcw radar,","cited_arxiv_id":null,"evidence_quote":"Provides the orthogonal-projection harmonic suppression method compared against and extended by E-VSDR."},{"cited_title":"Smart homes that monitor breathing and heart rate,","cited_arxiv_id":null,"evidence_quote":"Provides the phase-regression vital-sign estimation baseline compared in the trials."},{"cited_title":"Vital signs monitoring of multiple people using a fmcw millimeter-wave sensor,","cited_arxiv_id":null,"evidence_quote":"Provides the angle-FFT localization method used as a baseline in the localization comparisons."}],"review_version":1}