{"id":"36b6dcf3-d3a0-49ad-b400-2978486d257f","arxiv_id":"2412.00093","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A streaming open-source software lock-in demodulator achieves roughly two orders of magnitude lower 1/f background noise at 1 Hz than an SR830 commercial lock-in, enabling lower-noise resistance fluctuation measurements.","lead":"A software-defined lock-in amplifier, built from an off-the-shelf DAQ and open-source Python, delivers about 100 times lower low-frequency background noise than a commercial SR830 lock-in at 1 Hz. It also provides live preview and streaming downsampling, making it a drop-in replacement for hardware lock-ins in resistance noise experiments.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Two-order noise improvement is only measured with the ADC fed directly by the function generator; the SR560 preamplifier used in real DUT measurements is absent from the background test, so the headline system-level claim is not yet supported.","rationale":"The reader's weakest assumption is exactly the one I would flag. The background test is the only evidence for the headline improvement, and it bypasses the one active component that differs between the test configuration and the application. The paper is careful and transparent about this, and the matching test in Fig. 4a verifies the demodulator's frequency response, so this is not a correctness failure or an internal inconsistency. But the abstract and conclusion state the improvement as a property of the system, and the conclusion even suggests that multi-stage amplifiers will exploit the +/-10 V input range; that claim is not measured. A controlled repeat with the SR560 in the signal path would settle whether the two-order margin is preserved. Secondary issues (single instrument comparison, no error bars) are real but less load-bearing: even a 2x uncertainty would not erase a 100x effect, and the SR830 was chosen as a common high-end reference. If the SR560 test shows the floor is dominated by the preamplifier, the correct statement would be 'software lock-in with a suitable low-noise preamplifier' rather than 'our system', and the quantitative claim would need to be scoped to the direct-ADC case. That is why the conditional verdict is appropriate and no change to the reader's verdict is needed.","tokens_in":6996,"tokens_out":6883,"duration_ms":65134,"concrete_test":"Repeat the Section III background-noise measurement with the SR560 inserted between the DS360 and the NI 9239, using the same RMS voltage at the ADC input (e.g., 1 V) and the same source impedance as the real-DUT test; then compute SV(1 Hz)/<V>^2 at the software lock-in output. If the value remains near 2e-13 Hz^-1, the concern is resolved; if it rises by more than about 10x toward the SR830 floor, the abstract's 'system' claim is not supported in the actual measurement chain.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim—two orders of magnitude lower SV(1 Hz)/<V>^2 than the SR830—rests on Fig. 4c, where the DS360 is connected directly to the NI 9239 ADC. The full measurement chain in Fig. 2b and in the real-DUT comparison (Fig. 4a) inserts an SR560 preamplifier before the ADC. Section III explicitly states the direct connection, so the omission is transparent, but it is also load-bearing: the advertised advantage is a property of the acquisition system, not just of the demodulator algorithm. If the SR560's output noise at the carrier frequency, or any amplitude-dependent artifact it introduces, is comparable to or larger than the SR830's 2e-11 Hz^-1 floor, the as-stated system improvement would not survive in the configuration actually used for resistance-noise measurements. The observed 1/f floor in Fig. 5 is attributed to either the common reference channel or the sine source, not to the ADC; that attribution is speculative. The real-DUT match to within 5% cannot rescue the improvement claim because that DUT's noise is several orders above the instrument floor and only verifies the transfer function.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper describes a software-defined lock-in demodulator implemented in Python, using an NI 9239 data acquisition system and streaming signal processing with real-time preview. The authors validate the frequency response against an SR830 lock-in on a real DUT showing 1/f resistance noise, and they measure the background noise of the software chain versus the SR830. The central claim is that the software lock-in achieves a relative voltage noise power spectral density SV(1 Hz)/<V>^2 of about 2e-13 Hz^-1, roughly two orders of magnitude below the SR830's 2e-11 Hz^-1, and that this improvement enables low-frequency resistance noise measurements without bridge circuits.","tokens_in":7176,"tokens_out":5080,"duration_ms":48956,"significance":"If the claimed improvement holds, this work is practically significant for low-frequency resistance noise spectroscopy: it offers an open-source, off-the-shelf alternative to commercial lock-in amplifiers with a much lower 1/f background. The paper's strengths include a reproducible open-source implementation, a streaming architecture with downsampling and real-time preview (a useful advance over earlier packages), and a direct, same-source comparison with a widely used commercial instrument. The frequency-response match to within 5% on a real DUT is a convincing transfer-function check, and the use of cross-correlation in Fig. 5 is a reasonable diagnostic. The main limitation is that the headline two-order improvement is demonstrated only for the ADC/demodulator chain without the SR560 preamplifier that is present in the actual noise-measurement setup.","major_comments":[{"comment":"The two-order background-noise improvement is measured with the DS360 function generator connected directly to the NI 9239 ADC, whereas the full measurement chain in Fig. 2b and the real-DUT comparison in Fig. 4a insert an SR560 preamplifier before the ADC. The text explicitly states the direct connection, but the abstract and conclusion claim a system-level improvement ('our system offers superior low-frequency noise performance'). Because the SR560's noise, particularly any amplitude-dependent or 1/f component at the carrier frequency, could dominate the demodulated spectrum, the as-stated improvement is not yet supported for the configuration actually used in resistance-noise measurements. Please either repeat the background test with the SR560 in the signal path or clearly qualify the claim as applying only to the demodulator/ADC chain.","section":"Section III, Figs. 2b and 4c"},{"comment":"The residual 1/f-like upturn in Fig. 5 is attributed to either the common reference channel or the sine source, based on the observation that it is fully correlated across channels. However, because all three demodulators share the same reference channel, the cross-correlation cannot distinguish between a source artifact and a reference-path artifact, nor can it exclude a contribution from the ADC's common-mode or clock path. This attribution is therefore speculative. The relative comparison with the SR830 is unaffected, but the absolute value SV(1 Hz)/<V>^2 ≈ 2e-13 Hz^-1 should be presented as an upper bound for the demodulator's own noise floor, not as a demonstrated property of the demodulator alone.","section":"Section III, Fig. 5"},{"comment":"The headline quantitative claim rests on single smoothed spectra with no repeated measurements, error bars, or confidence intervals. While the two-order margin is large enough that modest run-to-run variations would not change the conclusion, the absence of any uncertainty estimate makes the claimed value (and the implied universality of the 2e-13 Hz^-1 floor) less robust. Please provide at least one repeat measurement or an uncertainty estimate for the background spectra, or explicitly state that the values are representative single traces.","section":"Section III, Figs. 4 and 5"}],"minor_comments":[{"comment":"The legend 'Software (0.1 V, 0.3 V, 1 V)' is ambiguous because it is not clear which line corresponds to which amplitude, especially if the curves overlap. Please use distinct markers or a table to identify each amplitude.","section":"Section III, Fig. 4c"},{"comment":"The paper says the software background 'does not exhibit any such dependence on the input amplitude', but the plotted curves appear to have a small spread at low frequencies. Please clarify whether this spread is within the expected smoothing/statistical scatter, or quantify any residual amplitude dependence.","section":"Section III, Fig. 4c and Fig. 5"},{"comment":"The phrase 'the 1/f noise magnitude is only approximately SV(f = 1 Hz)/<V>^2 ≈ 2e-13 Hz^-1' conflates the measured total background with the demodulator's own noise. Consider rewording to 'the measured background' or 'the system background'.","section":"Section III, text near Fig. 5"},{"comment":"The relative difference of about 5% is quoted without specifying how the comparison was made after the SR830 spectrum was corrected for its filter response. Please state the correction method or reference.","section":"Section III, Fig. 4a"},{"comment":"There are minor typographical errors, e.g., 'behavoiur' in Section III and 'futhermore' in the Fig. 1 caption. Please proofread.","section":"Throughout"},{"comment":"The abstract says 'compared to high-end commercial lock-in instruments' (plural), but the comparison is with a single SR830 unit. Please temper the wording to 'compared to a high-end commercial lock-in amplifier' unless more instruments are tested.","section":"Abstract and Introduction"}],"recommendation":"major_revision","confidential_remarks":"The paper is a good fit for physics.ins-det and the software artifact is a genuine contribution. The main issue is the missing preamplifier test; if the authors add a background measurement with the SR560 in the chain, or explicitly scope the claim to the demodulator/ADC path, the paper would likely be acceptable. There is no evidence of circular reasoning or fitted parameters; the comparison is direct and the code is open source. The 'two orders of magnitude' claim is large, so the authors should be encouraged to provide any data that would confirm it under the conditions of actual use."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The useful new thing here is the streaming, state-preserving demodulator: it keeps filter state across block boundaries, decimates on the fly, and gives a live preview. That is a real step beyond off-line ULIA and microcontroller versions, and the implementation is careful (buffering, Observer pattern, HDF5 output). The code is released, which matters for a methods paper. The 5% match with the SR830 on a real DUT is a solid, direct validation of the transfer function.\n\nThe soft spot is exactly where the stress-test note lands. The two-order noise improvement (Fig. 4c) was measured with the function generator connected straight to the ADC, bypassing the SR560. The real measurement chain in Fig. 2b includes that preamp, and the real-DUT spectrum is several orders above the instrument floor, so it only verifies the frequency response, not the claimed floor. If the SR560 adds amplitude-dependent or 1/f noise comparable to the SR830 floor, the as-stated system-level improvement would not hold. The paper is transparent about the direct connection, so this is not concealment, but it is load-bearing. Also, the comparison is against a single SR830 unit, there are no repeat runs or error bars, and the cross-correlation attribution of the residual 1/f noise is speculative rather than tested.\n\nThese are all addressable. The demodulator itself appears to work as advertised, and the engineering is honest. The paper deserves a serious referee; an only mildly demanding revision would ask for (1) the background test repeated with the SR560 inserted, (2) at least a few repeated spectra to show run-to-run spread, and (3) a more cautious statement about the residual 1/f source. I would not desk-reject it, and I would not block publication on the missing preamp test as long as the claim is scoped to the acquisition chain and not the full system.\n\nThis is a solid, useful methods paper for the low-frequency noise community. It is not a breakthrough, but it is a genuine drop-in alternative to an expensive instrument. Send it to review.","headline":"A genuinely useful streaming software lock-in with clean code and a solid SR830 cross-check, but the headline two-order noise improvement is measured on the bare ADC and may not survive the SR560 preamp in the real measurement chain.","tokens_in":7709,"tokens_out":1604,"would_cite":true,"duration_ms":17783,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A software-defined lock-in amplifier replaces commercial lock-in hardware and lowers the 1 Hz background voltage-noise power spectral density by about two orders of magnitude.","keywords":["software-defined lock-in amplifier","1/f noise","resistance noise","low-frequency noise measurements","digital demodulation","data acquisition","real-time signal processing","noise spectroscopy"],"falsifier":"Measure the background voltage-noise power spectral density at 1 Hz for the full chain — signal generator, SR560 preamplifier, and the software-defined lock-in — at $\\langle V\\rangle = 1\\,\\mathrm{V}$; if $S_V(1\\,\\mathrm{Hz})/\\langle V\\rangle^2$ rises from about $2\\times10^{-13}$ toward $2\\times10^{-11}\\,\\mathrm{Hz}^{-1}$, the claimed two-order advantage disappears in the configuration users would actually run.","tokens_in":6762,"feed_emoji":"📉","tokens_out":15823,"duration_ms":116319,"temperature":0.7,"pith_summary":"This paper claims that a software-defined lock-in amplifier can replace commercial lock-in hardware in low-frequency resistance noise measurements and outperform it. The key result is a background relative voltage-noise power spectral density at 1 Hz of about $2\\times10^{-13}\\,\\mathrm{Hz}^{-1}$, roughly two orders of magnitude below the $\\sim2\\times10^{-11}\\,\\mathrm{Hz}^{-1}$ of the SR830 instrument, because the digital demodulator does not add amplitude-dependent $1/f$ noise. If correct, this would let researchers measure very low-noise devices without bridge circuits, using only an off-the-shelf data acquisition system and open-source code. The claim is supported by a comparison on a real device showing the two spectra agree within about 5%.","feed_headline":"Software lock-in cuts 1 Hz noise power 100-fold","feed_subtitle":"Digital demodulation on off-the-shelf hardware reaches a 2e-13 Hz^-1 noise floor, enabling low-noise resistance measurements.","key_machinery":"The central mechanism is the streaming digital demodulation chain: a 24-bit $\\Delta$-Sigma ADC samples the amplified DUT voltage and a simultaneous reference channel; the DUT stream is mixed element-wise with the reference, amplitude-corrected using an RMS detector that tracks the reference amplitude, low-pass filtered, and decimated. Filter state is carried across time-series blocks so the output is continuous, and a block-wise Observer pattern permits quasi-real-time display and low data rates. This replaces the analog mixer and filter of a hardware lock-in, and because the demodulation is numeric, the instrument's own semiconductor and algorithm noise does not scale with input amplitude.","core_discovery":"The paper establishes that performing the lock-in demodulation entirely in software yields a cleaner low-frequency background than the commercial instrument it replaces. With a pure sine input and identical source and load conditions, the software-defined system shows no amplitude-dependent $1/f$ noise up to $\\langle V\\rangle = 1\\,\\mathrm{V}$, whereas the SR830 background scales as $S_V(1\\,\\mathrm{Hz})/\\langle V\\rangle^2 \\approx 2\\times10^{-11}\\,\\mathrm{Hz}^{-1}$. At 1 Hz the software background is $S_V(1\\,\\mathrm{Hz})/\\langle V\\rangle^2 \\approx 2\\times10^{-13}\\,\\mathrm{Hz}^{-1}$, about two orders of magnitude lower. On a real DUT with strong $1/f$ resistance noise, the software and hardware spectra match to within about 5% over the measured range, confirming the digital processing preserves the signal.","pith_inferences":["If the residual correlated $1/f$ floor originates in the signal source, as the paper speculates, swapping in a lower-phase-noise generator should push the floor below $2\\times10^{-13}\\,\\mathrm{Hz}^{-1}$; this direct test is not performed in the paper.","The two-order advantage was demonstrated without the SR560 in the signal path; repeating the background test with the preamplifier would show whether the full measurement chain keeps the advantage.","Because demodulation is software-only, the same acquired data stream can be demodulated at multiple carrier frequencies or with a software phase-locked loop, enabling multi-frequency noise spectroscopy or low-frequency impedance measurements without hardware changes; the paper notes the PLL extension but not multi-frequency demodulation."],"forward_implications":["Devices whose relative resistance noise is below roughly $10^{-11}\\,\\mathrm{Hz}^{-1}$ at 1 Hz become measurable without bridge circuits or cross-correlation setups.","The demodulator runs quasi-real-time on a general-purpose computer with off-the-shelf DAQ hardware, so it can replace lock-in hardware without custom electronics.","Decimation during acquisition keeps the output data rate low, making multi-hour or multi-day noise runs practical to store and process.","The author suggests that adding cross-correlation with multiple channels and preamplifiers, or using multiple reference channels, could suppress the residual correlated $1/f$ contribution further."],"supporting_citations":[{"why":"Supplies the AC modulation method that shifts the DUT noise spectrum away from the amplifier 1/f noise.","marker":"[6]"},{"why":"Reviews limits of conventional lock-in noise measurements and the need for bridge or cross-correlation configurations when the instrument floor limits sensitivity.","marker":"[7]"},{"why":"Previous software lock-in implementation that lacks streaming output and filter-state retention; serves as the baseline for the real-time and data-rate improvements.","marker":"[8]"},{"why":"Microcontroller-based lock-in achieving downsampling but requiring custom electronics; motivates the off-the-shelf DAQ approach.","marker":"[9]"},{"why":"Describes DC bias-independence checks and bridge configurations needed to validate AC resistance noise measurements.","marker":"[11]"},{"why":"Explains the Delta-Sigma ADC principle used by the data acquisition hardware.","marker":"[12]"},{"why":"Reports the strong 1/f resistance noise of the organic charge-transfer salt used as the real DUT in the method comparison.","marker":"[20]"},{"why":"Provides the cross power spectral density technique used to show the residual low-frequency noise is correlated across input channels.","marker":"[22]"}],"fun_headline_variants":["Software lock-in beats hardware 1/f noise by 100x","Open-source demodulator quiets 1 Hz noise 100-fold","Digital lock-in trims low-freq noise two orders","Free software lock-in cuts low-freq noise 100x","Software lock-in: 100x quieter at 1 Hz than commercial"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The two-order improvement was measured with the signal generator connected directly to the ADC, bypassing the SR560 preamplifier used in the real measurement chain; if that preamplifier contributes amplitude-dependent 1/f noise comparable to the SR830's, the claimed improvement would not hold for actual DUT measurements.","fun_headline_variants_meta":{"raw":{"variants":["Software lock-in beats hardware 1/f noise by 100x","Open-source demodulator quiets 1 Hz noise 100-fold","Digital lock-in trims low-freq noise two orders","Free software lock-in cuts low-freq noise 100x","Software lock-in: 100x quieter at 1 Hz than commercial"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000911,"raw_usage":{"total_tokens":3875,"prompt_tokens":866,"completion_tokens":3009,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":482,"completion_tokens_details":{"reasoning_tokens":2934}},"tokens_in":482,"tokens_out":3009,"duration_ms":19963,"temperature":1.0,"reasoning_tokens":2934,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T11:17:36.485144+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the background voltage-noise power spectral density at 1 Hz for the full chain — signal generator, SR560 preamplifier, and the software-defined lock-in — at $\\langle V\\rangle = 1\\,\\mathrm{V}$; if $S_V(1\\,\\mathrm{Hz})/\\langle V\\rangle^2$ rises from about $2\\times10^{-13}$ toward $2\\times10^{-11}\\,\\mathrm{Hz}^{-1}$, the claimed two-order advantage disappears in the configuration users would actually run.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the AC modulation method that shifts the DUT noise spectrum away from the amplifier 1/f noise."},{"cited_title":"Scandurra , author C","cited_arxiv_id":null,"evidence_quote":"Reviews limits of conventional lock-in noise measurements and the need for bridge or cross-correlation configurations when the instrument floor limits sensitivity."},{"cited_title":"Uhl , author L","cited_arxiv_id":null,"evidence_quote":"Previous software lock-in implementation that lacks streaming output and filter-state retention; serves as the baseline for the real-time and data-rate improvements."},{"cited_title":"O’Donoghue \\ and\\ author P","cited_arxiv_id":null,"evidence_quote":"Microcontroller-based lock-in achieving downsampling but requiring custom electronics; motivates the off-the-shelf DAQ approach."},{"cited_title":"Thyzel , author M","cited_arxiv_id":null,"evidence_quote":"Describes DC bias-independence checks and bridge configurations needed to validate AC resistance noise measurements."},{"cited_title":"Kester ,\\ @noop journal journal Analog Devices Analog Dialogue \\ ( year 2005 ) NoStop","cited_arxiv_id":null,"evidence_quote":"Explains the Delta-Sigma ADC principle used by the data acquisition hardware."},{"cited_title":"Thomas , author Y","cited_arxiv_id":null,"evidence_quote":"Reports the strong 1/f resistance noise of the organic charge-transfer salt used as the real DUT in the method comparison."},{"cited_title":"Sampietro , author L","cited_arxiv_id":null,"evidence_quote":"Provides the cross power spectral density technique used to show the residual low-frequency noise is correlated across input channels."}],"review_version":1}