{"id":"7119b9cb-b4d8-4003-ad48-fdef91c43dc5","arxiv_id":"2505.05825","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"Multi-rate sub-Nyquist sampling with compressive sensing extends a free-induction-decay magnetometer's resolvable frequency range from 251 Hz to 3000 Hz while suppressing aliasing.","lead":"This paper shows that multi-rate under-sampled measurements combined with compressive sensing can extend the frequency range of a free-induction-decay magnetometer beyond the usual Nyquist limit. A generalist might read it because it offers a practical way to keep high sensitivity while gaining wide bandwidth in quantum sensors.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The reconstruction model Y=ΦX is not justified for the actual data: the 41 one-second DFT spectra have unknown relative phase, so they are not linear projections of a common spectrum.","rationale":"The reader's sparsity/stationarity concern is real, but the decisive issue is more specific: the linear measurement model itself lacks a defined phase reference. This is a correctness risk, not a disagreement with consensus; if the model is mis-specified, the single 800 Hz demonstration cannot validate the method. The Eq. (3) inconsistency (with T=1 s and f_SCL=251 Hz, τ≈2 ms, Eq. (3) predicts f_CS≈125 kHz, not 3 kHz) and the lack of error bars are secondary concerns. The multi-rate alias-matching idea is plausible, and the reported 0.6% amplitude consistency is encouraging, so I would not reject outright; the paper should be conditional on a phase-coherent or power-spectrum formulation and on release of raw data for independent reconstruction.","tokens_in":10048,"tokens_out":11174,"duration_ms":132080,"concrete_test":"Obtain the 41 raw one-second records along with the recorded trigger timestamps or the relative phase of the 800 Hz source. Compute complex DFTs, correct each record to a common time origin using the measured phase offsets, and reconstruct X by solving the resulting complex linear system (split into real and imaginary parts). Then repeat the reconstruction using magnitude DFTs as implied by the paper. If the phase-corrected complex reconstruction does not place a calibrated peak at 800 Hz, the published amplitude-only result is an artifact of phase-insensitive processing. If raw phase data are unavailable, regenerate synthetic one-second records at the stated rates with independent random initial phases and test whether the paper's procedure still recovers the tone.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing assumption is that the concatenated DFT spectra Y satisfy Y=ΦX for a common sparse spectrum X (Eq. 14 and the sentence 'These spectra are vertically concatenated to form Y' in Experimental Results). The experiment acquires 41 one-second records at different repetition rates with no stated phase lock between the sampling clock and the 800 Hz tone. If the record start times are not synchronized to the tone, each complex DFT acquires a record-dependent phase factor exp(iθ_i), so the true relation is Y_i = D_i Φ_i X, not Y_i = Φ_i X; concatenating then invalidates the solver. If instead Y_i denotes the magnitude spectrum (as Fig. 4C/E/G suggest), the mapping is nonlinear: |Σ a_j| ≠ Σ|a_j|, so two aliased components in one bin do not add linearly. Neither case is compatible with the stated linear system, and Lawson-Hanson NNLS does not solve a complex linear system without specifying how real and imaginary parts are handled. Because the 800 Hz peak is the entire experimental evidence for the central claim, the linear model must be established before the reconstruction can be accepted.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript reports a demonstration of compressive sensing (CS) with multi-rate sub-Nyquist sampling applied to a free-induction-decay (FID) 87Rb magnetometer. The authors model the undersampled DFT spectra from 41 prime repetition rates (263–503 Hz) as linear projections of a common high-resolution spectrum, solve the concatenated system with Lawson–Hanson non-negative least squares, and reconstruct a spectrum extending to 3000 Hz. They identify an 800 Hz applied modulation tone outside the nominal 251 Hz Nyquist range, suppress alias peaks seen in individual undersampled spectra, and report unchanged photon-shot-noise-limited sensitivity of 4 pT/√Hz at 100 Hz. A scaled frequency-response correction recovers the applied 800 Hz amplitude to within 0.6%, and a separate linearity check at 137 Hz gives slope 1.006. The paper claims a factor-of-12 extension of the resolvable frequency range and a general route beyond the 'spin coherence limit.'","tokens_in":10275,"tokens_out":8391,"duration_ms":88671,"significance":"The core idea — using sparse recovery over multiple sub-Nyquist rates to convert long coherence time from a bandwidth limitation into a reconstruction advantage — is potentially useful for atomic magnetometry and other FID-based sensors. The paper includes real experimental data, a concrete 800 Hz recovery above the nominal Nyquist limit, an explicit comparison of aliased spectra before and after reconstruction, and a sensitivity measurement. These strengths make the empirical demonstration worth taking seriously. However, the theoretical scaling law in Eq. (3) is inconsistent with the experimental parameters by about a factor of 40, and the linear measurement model in Eq. (10) requires a common absolute phase reference that is not documented. Both points are load-bearing for the central claim, so the manuscript needs substantive revision before the result can be accepted.","major_comments":[{"comment":"Eq. (3) claims f_CS ≃ T/(2τ²), but this scaling is contradicted by the experimental parameters. The conventional limit is quoted as 251 Hz at a pump-probe repetition rate of 503 Hz, which gives τ ≈ 1/(2×251) ≈ 2.0 ms. Inserting T = 1 s and τ ≈ 2.0 ms into Eq. (3) yields f_CS ≈ 125 kHz, not the demonstrated 3000 Hz. The text does not identify the actual coherence time used in the experiment or explain why the realized extension is about 40 times smaller than the predicted limit. The stated design condition M_i ∼ sqrt(N_s) is also not satisfied: for f_CS = 3000 Hz and T = 1 s one has N_s = 6000 and sqrt(N_s) ≈ 77, while the 41 experimental sampling rates are primes from 263 to 503 Hz. Please correct the scaling or provide the actual parameters and an explanation of the discrepancy.","section":"Measurement Process and Compressive Sensing, Eq. (3)"},{"comment":"The construction of Y by vertical concatenation of the 41 DFT spectra is the key step that turns the experiment into a CS problem, but the linear relation Y_i = Φ_i X is not justified in the manuscript. Fig. 4(C,E,G) shows 'amplitude spectra', whereas Eq. (10) is a coherent linear relation between complex spectra. If the concatenated Y are magnitudes, the relation is nonlinear because the magnitude of a sum of aliased components is not the sum of their magnitudes. If Y are complex DFT values, each 1-s record contains an arbitrary initial phase φ0 from Eq. (4) and the start times of records at different sampling rates are not stated to be synchronized to the 800 Hz source; with record-dependent phases the true relation is Y_i = D_i Φ_i X, and the concatenated system does not have the assumed form. The Lawson–Hanson solver is applied without specifying how complex data are handled. Please specify the exact contents of Y and the phase reference, and either establish phase coherence experimentally or reformulate the reconstruction to be invariant to record-dependent phases.","section":"Experimental Results / Methods Eq. (10)"},{"comment":"The amplitude-linearity test in Fig. 7 is performed at 137 Hz, which lies inside the conventional 251 Hz Nyquist band. It therefore does not test whether the CS reconstruction preserves amplitudes in the extended range beyond the SCL. The only above-Nyquist amplitude validation is the single 800 Hz point, and that point uses an independently calibrated low-pass response correction. To support the claim that the method recovers the correct amplitude at the correct frequency in the extended band, the authors should provide amplitude validation at several frequencies above 251 Hz, or at least quantify the reconstruction error over the full 3000 Hz range.","section":"Reconstruction consistency, Fig. 7"}],"minor_comments":[{"comment":"The caption lists panels (B), (C), and (F) where the text refers to (B,D,F) and (C,E,G); the caption should be corrected.","section":"Fig. 4 caption"},{"comment":"The DFT formula in Eq. (5) uses the summation index k but the exponent contains t and N_i; the sum should be over the sampled time index consistently.","section":"Methods, Eq. (5)"},{"comment":"Reference 15 is malformed ('K. M. Budker, D.'), reference 2 is duplicated as reference 6, and reference 10 concatenates two distinct works that should be separated.","section":"References"},{"comment":"The reconstruction dimension N_s, the number of frequency bins, and the sparsity level k are not defined in the experimental section; these parameters should be stated explicitly, including how k = 8 is determined and whether it is known before or after reconstruction.","section":"Experimental Results"},{"comment":"There are several typographical errors, including 'detunning' and 'denots', that should be corrected in a final pass.","section":"Text"}],"recommendation":"major_revision","confidential_remarks":"The manuscript reports an interesting experiment, but the two technical issues — the Eq. (3) scaling mismatch and the unspecified phase reference in Y = ΦX — are serious. I would like the authors to provide the raw complex DFT data or a clear description of phase synchronization before recommending publication. If the phase reference cannot be established, the 800 Hz recovery may not be a valid solution of the stated linear system. This is the main risk to the paper's central claim."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The idea is sound: use multi-rate sub-Nyquist sampling plus compressive sensing to extend the bandwidth of an FID magnetometer without giving up the long coherence time that gives sensitivity. That is a legitimate extension of established CS/MASS work to a quantum sensing platform, and the paper shows real data claiming an 800 Hz tone recovered above the nominal 251 Hz limit, with amplitude error of 0.6% and sensitivity apparently unchanged. Those are the reasons to take it seriously.\n\nThe problem is that the reconstruction model as written is not justified by the experimental procedure. The authors acquire 41 one-second records at different pump-probe rates and concatenate their DFT spectra to form Y in the linear system Y = ΦX. For that to hold, each record's complex spectrum must be a linear projection of the same underlying complex spectrum X. But the 800 Hz tone is not phase-locked to the pump-probe cycle (no synchronization is stated), so each record starts at a different phase of that tone. The complex DFT of each record then carries a record-dependent phase factor; the concatenated Y is not a linear function of a single X. If the authors instead used magnitude spectra, the linearity also fails because the magnitude of a sum of aliased complex components is not the sum of magnitudes. Either way, the central evidence for the claim is unsupported as reported.\n\nThe scaling formula f_CS ≈ T/(2τ²) is also inconsistent with the experiment: with T = 1 s and τ around 2 ms it predicts ~125 kHz, while the paper only demonstrates 3000 Hz. That may be a deliberate choice, but it is unexplained. The single-tone validation is nice but does not show that the method handles multiple components in a full band, and there are no error bars anywhere. The \"general technique\" framing is stronger than what one tone at one amplitude supports.\n\nIf the authors can clarify how the phase issue is handled—perhaps they did synchronize the sampling clock to the 800 Hz source, or they used a nonlinear sparse solver that accounts for phases—then the result could be a useful demonstration. As written, the paper needs that clarification before the central claim is credible. The raw idea is worth refereeing, and a good referee would catch this. I would not cite it in its current form, but I would bring it to a reading group to discuss the interaction between CS assumptions and real quantum sensors.","headline":"The paper's central claim rests on a multi-rate reconstruction whose linear model is not justified by the experimental timing; worth refereeing, but the authors must fix the phase problem and validation gaps.","tokens_in":10822,"tokens_out":4810,"would_cite":false,"duration_ms":54155,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["07.55.Ge"],"model":"deepseek-v4-flash","headline":"Compressive sensing extends free-induction-decay magnetometry's resolvable frequency range from 251 Hz to 3000 Hz while keeping sensitivity at 4 pT/√Hz.","keywords":["compressive sensing","free induction decay magnetometry","spin coherence limit","sub-Nyquist sampling","frequency aliasing","atomic magnetometer","sparse spectrum reconstruction","quantum sensing"],"falsifier":"Apply a field whose two-sided spectrum contains more than 21 nonzero components (e.g., 11 distinct positive tones above 251 Hz plus their negative mirrors), or a single tone that drifts by more than 1 Hz over the 41 s acquisition; if the reconstruction shows spurious peaks or omits true tones at the expected amplitudes, the central claim is refuted.","tokens_in":9844,"feed_emoji":"🧲","tokens_out":10894,"duration_ms":109802,"temperature":0.7,"pith_summary":"Free-induction-decay magnetometry normally faces a trade-off: a longer coherence time improves sensitivity but, by the sampling theorem, forces a narrower resolvable frequency range and causes frequency aliasing. This paper reports that compressive sensing breaks that trade-off. By recording the same decay at 41 distinct sub-Nyquist repetition rates and jointly reconstructing a sparse spectrum, the resolvable range of an alkali-vapor FID magnetometer is extended from the $251\\,\\mathrm{Hz}$ limit to $3000\\,\\mathrm{Hz}$, with alias-induced spurious peaks suppressed. The recovered $800\\,\\mathrm{Hz}$ test tone appears at the right frequency and, after correcting the magnetometer's low-pass response, at the right amplitude within 0.6%. Sensitivity stays at $4\\,\\mathrm{pT}/\\sqrt{\\mathrm{Hz}}$ at $100\\,\\mathrm{Hz}$, so the frequency extension comes at no measured sensitivity cost.","feed_headline":"Magnetometer frequency range boosted 12-fold at same sensitivity","feed_subtitle":"Compressive sensing lifts the resolvable band from 251 Hz to 3000 Hz and suppresses alias peaks.","key_machinery":"The load-bearing mechanism is multi-rate asynchronous sub-Nyquist sampling (MASS): 41 measurement matrices $\\Phi_i$ are built from prime sample lengths $M_i$ of order $\\sqrt{N_s}$, chosen so different sampling rates fold high-frequency components into distinguishable alias patterns. Concatenating the records gives $Y = \\Phi X$, a linear system whose solution $X$ is recovered by a least-squares solver. The scaling law $f_{\\mathrm{CS}} \\simeq T f_{\\mathrm{SCL}} / \\tau$ carries the argument: total acquisition time $T$ buys enough independent low-rate projections to replace the coherence-time-limited range with a sparsity-limited range, provided the spectrum has at most $k$ nonzero lines and the number of rates $v$ exceeds $2k-1$.","core_discovery":"The central claim is that the spin coherence limit is not a hard ceiling for frequency range: a linear-system reconstruction can recover spectral components far above the single-shot sampling rate as long as the underlying spectrum is sparse. The authors demonstrate this with a free-induction-decay $^{87}\\mathrm{Rb}$ magnetometer whose coherence time sets the per-shot sampling rate near $500\\,\\mathrm{Hz}$. They take 41 one-second records, each sampled at a different prime repetition rate between $263\\,\\mathrm{Hz}$ and $503\\,\\mathrm{Hz}$, form the vertical stack $Y$ from them, and solve $Y = \\Phi X$ for the two-sided spectrum $X$ using a least-squares reconstruction. The reconstructed spectrum reaches $3000\\,\\mathrm{Hz}$, a factor of 12 beyond the $251\\,\\mathrm{Hz}$ limit set by the sampling rate, and shows the applied $800\\,\\mathrm{Hz}$ modulation plus 50 Hz power-line harmonics; the same signal appears as different aliases (12 Hz, 43 Hz, 207 Hz) in individual undersampled records, demonstrating that the joint reconstruction rather than any single record resolves it. Sensitivity is unchanged at $4\\,\\mathrm{pT}/\\sqrt{\\mathrm{Hz}}$ at $100\\,\\mathrm{Hz}$, limited by photon shot noise, and the reconstructed amplitude matches the applied field after a low-pass frequency-response correction.","pith_inferences":["The same multi-rate reconstruction should transfer to other long-coherence FID platforms such as NV-center or noble-gas-spin sensors, where the per-shot sampling rate is even smaller relative to the frequencies of interest; the paper motivates but does not demonstrate this transfer.","The formula $f_{\\mathrm{CS}} \\simeq (T/\\tau)f_{\\mathrm{SCL}}$ implies the 12-fold gain is not a ceiling: longer total acquisition time $T$ should push the resolvable range higher, provided the spectrum stays sparse and stationary over that longer window.","A stress test with a slowly drifting tone or a sparse-but-time-varying spectrum would quantify the failure threshold of the stationarity assumption; the paper does not provide such a robustness curve.","Because the method is a post-processing reconstruction rather than a change to the spin system, it could in principle be combined with existing coherence-extension techniques such as spin locking or dynamic decoupling, pushing sensitivity and bandwidth simultaneously."],"forward_implications":["For any FID-based sensor, the resolvable frequency range can be lifted from about $1/(2\\tau)$ to roughly $(T/\\tau)\\cdot f_{\\mathrm{SCL}}$, so longer total averaging time directly buys more bandwidth at fixed sensitivity.","Frequency-alias artifacts that would otherwise appear inside the original band—here, spurious peaks at 12 Hz, 43 Hz, and 207 Hz from an 800 Hz signal—are removed by the joint reconstruction.","A tone above the old limit is recovered with correct frequency and, after low-pass response correction, correct amplitude to within 0.6%.","The sensitivity of the CS-enhanced magnetometer remains photon-shot-noise limited at $4\\,\\mathrm{pT}/\\sqrt{\\mathrm{Hz}}$ at $100\\,\\mathrm{Hz}$, meaning the range extension does not degrade sensitivity.","The sparsity requirement quantifies the trade-off: with $v$ sampling rates, the bilateral spectrum must contain at most $k=(v+1)/2$ nonzero components for the MASS reconstruction to be valid."],"supporting_citations":[{"why":"establishes the sampling-rate bound $f_{\\mathrm{SCL}}=1/(2\\tau)$ that defines the spin coherence limit being overcome.","marker":"(20)"},{"why":"supplies the compressive-sensing guarantee that sparse spectra can be reconstructed from highly incomplete frequency information.","marker":"(21)"},{"why":"introduces compressed sensing and the sparsity and incoherence conditions used to justify sub-Nyquist recovery.","marker":"(22)"},{"why":"formalizes the sparsity and incoherence requirements that the measurement matrix must satisfy.","marker":"(26)"},{"why":"provides the multi-rate asynchronous sub-Nyquist (MASS) sampling scheme with prime-length projections that the experiment adopts.","marker":"(40)"},{"why":"provides the least-squares algorithm used to solve the linear system $Y=\\Phi X$.","marker":"(41)"},{"why":"supplies the damped-sinusoid FID signal model and the choice of probe time equal to the coherence time that sets the single-shot sampling rate.","marker":"(6)"},{"why":"gives the low-pass frequency response used to correct the reconstructed amplitude and to characterize sensitivity across frequency.","marker":"(42)"}],"fun_headline_variants":["Compressive sensing lifts magnetometer range 12x past spin limit","Beyond spin coherence limit: magnetometer range up 12-fold","Sparse recovery breaks magnetometer frequency ceiling 12x","Alias-free magnetometry: range 251 Hz to 3000 Hz via CS","12x wider bandwidth for FID magnetometer via compressive sensing"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The reconstruction treats the 41 one-second records as different projections of one fixed, sparse magnetic-field spectrum; the central claim collapses if the field drifts in frequency or contains many closely spaced tones during the 41-second acquisition.","fun_headline_variants_meta":{"raw":{"variants":["Compressive sensing lifts magnetometer range 12x past spin limit","Beyond spin coherence limit: magnetometer range up 12-fold","Sparse recovery breaks magnetometer frequency ceiling 12x","Alias-free magnetometry: range 251 Hz to 3000 Hz via CS","12x wider bandwidth for FID magnetometer via compressive sensing"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000173,"raw_usage":{"total_tokens":1299,"prompt_tokens":987,"completion_tokens":312,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":603,"completion_tokens_details":{"reasoning_tokens":221}},"tokens_in":603,"tokens_out":312,"duration_ms":3644,"temperature":1.0,"reasoning_tokens":221,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T22:55:05.737492+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Apply a field whose two-sided spectrum contains more than 21 nonzero components (e.g., 11 distinct positive tones above 251 Hz plus their negative mirrors), or a single tone that drifts by more than 1 Hz over the 41 s acquisition; if the reconstruction shows spurious peaks or omits true tones at the expected amplitudes, the central claim is refuted.","supporting_citations":[],"review_version":1}