{"id":"95023eb1-3918-400b-9801-9dfa60eadafd","arxiv_id":"2507.05655","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"A Raspberry Pi 4B works as controller, recorder, and processor for the PRATUSH lab-model digital spectrometer, achieving millikelvin-level thermal-noise-limited residuals.","lead":"The PRATUSH space-radiometer lab model replaces its computer with a Raspberry Pi 4B to control the digital spectrometer, read out FPGA spectra, and store data. Integrated with the analog receiver, the system reaches thermal-noise-limited residuals near 12.5 mK over 44 hours, supporting the viability of a low-power SBC design for space.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central performance claim is unverifiable because the maximally smooth function and dynamic-flagging thresholds are not specified, so the 12.5 mK residual could be an artifact of an arbitrarily flexible fit.","rationale":"The reader's weakest-assumption analysis identifies the unspecified maximally smooth function and tuned flagging thresholds as the key risk, and I concur. The paper's performance claim is not independently supported by machine-checked proofs or released code, and the residual rms is a post-fit quantity whose meaning depends entirely on the fit's flexibility. The concern is not about author intent; rather, the paper as written does not permit a reader to verify that the 12.5 mK residual is not an artifact of the smooth-function fit and flagging choices. A conditional decision requiring the model specification and a robustness test is appropriate, and I see no basis to move the verdict in either direction.","tokens_in":28433,"tokens_out":4441,"duration_ms":51875,"concrete_test":"Ask the authors to publish the exact definition of the maximally smooth function used for Figures 11 and 13 (basis, number of parameters, smoothness regularization, fitting algorithm) and to recompute the residual rms after varying the model flexibility, e.g., halving and doubling the effective degrees of freedom while keeping all else fixed. If the 12.5 mK figure changes by more than roughly 30%, or if the residuals cease to be Gaussian/white under this variation, the thermal-noise-limited claim is not robust. An additional injection test—adding a synthetic smooth spectral feature of amplitude ~50 mK to the raw spectra before the same preprocessing and checking whether the pipeline recovers it—would determine whether the smooth fit can mask real spectral structure.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The conclusion that the PRATUSH digital system is suitable for science rests on residual rms values of 12.5 mK (30.51 kHz) and 3.4 mK (610 kHz) after subtracting a 'maximally smooth function' from the 44-hour averaged spectrum (Section 6, Figs. 11-13). The function is never defined: no basis, order, number of parameters, smoothness penalty, or fitting procedure is given, and the dynamic-flagging thresholds (MAD multiplier and contiguous-flagged-channel count in Section 5.1) are described as 'tuned' without quoting their values. The residual statistics therefore depend on two unstated degrees of freedom. If the smooth model is sufficiently flexible, it can absorb smooth instrumental systematics—and, in the actual science case, a smooth cosmological signal—so a small Gaussian residual does not demonstrate the absence of systematics; it only shows that whatever remains after an arbitrary fit is noise-like. The Gaussianity and simulated-noise comparisons in Figs. 11-12 test statistical properties of the post-fit residual, not whether meaningful spectral structure was removed. This is the most load-bearing assumption: the 'devoid of systematics' claim is not verifiable without specifying and testing the fit model and flagging thresholds.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper describes the design, implementation, and laboratory validation of the digital correlation spectrometer for the PRATUSH global 21-cm experiment, using a Raspberry Pi 4 Model B as the master controller, data recorder, and real-time processor. The authors detail the pSPEC-based hardware (10-bit ADCs, Virtex-6 FPGA, 16384-channel FFT correlator), the SBC selection, the firmware and software architecture, and a custom 'dynamic flagging' algorithm that removes data-drop artifacts caused by the SBC's limited performance. Validation results with standard terminations and an antenna simulator show residual rms values of about 72 mK after an 8-hour run and 12.5 mK (3.4 mK after smoothing) after 44 hours of effective data, with residuals reported as Gaussian and consistent with thermal noise; the paper concludes that the digital system is suitable for PRATUSH's science requirements.","tokens_in":28747,"tokens_out":6245,"duration_ms":67005,"significance":"If the reported residual performance holds, this is a valuable engineering demonstration that a low-power, commercially available single-board computer can serve as the digital back-end for a global 21-cm radiometer. The paper provides substantial concrete detail on the hardware, firmware resources, data rates, and power/EMI considerations, and the dynamic flagging algorithm is a useful contribution to handling data-integrity issues in such systems. The comparison against 100 Gaussian noise realizations and the Gaussianity tests are good falsifiable checks of the statistical properties of the residuals. However, the central 'devoid of systematics' claim rests on an unspecified fitting procedure and unquoted flagging thresholds, and the 'thermal noise limited' claim is not tied to a radiometer-equation prediction, so the scientific significance of the demonstration is not yet fully established as written.","major_comments":[{"comment":"The 'maximally smooth function' subtracted from the averaged spectra is never defined: the authors give no basis, order, number of parameters, smoothness penalty, or fitting procedure. Since the central quantitative claims (residual rms of 72 mK, 12.5 mK, and 3.4 mK, and the statement that residuals are 'devoid of systematics') all depend on this subtraction, the claim is not verifiable as written. A sufficiently flexible smooth model can absorb smooth instrumental systematics and, in the science case, a smooth cosmological signal, so a small Gaussian residual after an arbitrary fit does not by itself demonstrate the absence of systematics. Please specify the function completely and demonstrate robustness of the residual rms to the fit's degrees of freedom, and test on simulated spectra with injected smooth components that the fitting procedure does not remove signal-like structure.","section":"Section 6 (Figs. 11-13)"},{"comment":"The claim that residuals are 'consistent with thermal noise expectations' is not supported by a calculation of the expected residual rms from the radiometer equation (system temperature, bandwidth, integration time, number of channels). The comparison against 100 Gaussian realizations with variance matched to the data tests Gaussianity and channel independence, but it does not test whether the absolute residual level equals the thermal noise prediction. Additionally, the 8-hour (72 mK) and 44-hour (12.5 mK) runs use different input terminations (50-ohm load vs antenna simulator), so the reduction in rms cannot be attributed to integration time alone; please provide the predicted rms for each configuration and compare directly with the measured values.","section":"Section 6 (Figs. 11-12)"},{"comment":"The dynamic flagging thresholds—the MAD multiplier used to set the deviation threshold and the number of consecutive flagged channels required to drop a full spectrum—are described as 'tuned to minimize false or erroneous flagging' but the actual values are never quoted. The moving-window size and step are given only as illustrative examples ('20 spectra' and 'one spectrum'). Because the final residual statistics depend on which spectra are retained, please report the exact parameter values used in the 8-hour and 44-hour analyses, and include a robustness check showing that the residual rms is insensitive to reasonable variations of these parameters.","section":"Section 5.1, Eqs. (1)-(2)"},{"comment":"The conclusion that the system is 'suitable for the science requirements of PRATUSH' is not tied to a quantitative statement of those requirements in this paper. Please state the target sensitivity or allowable systematic residual for the PRATUSH global 21-cm measurement and show explicitly how the measured 12.5 mK (30.51 kHz) and 3.4 mK (610 kHz) residuals satisfy that requirement.","section":"Section 6 and Section 8"}],"minor_comments":[{"comment":"The window function is referred to as 'Nuttal'; the standard spelling is 'Nuttall' (see also reference [66]).","section":"Throughout"},{"comment":"The footnote 'Sathyanarayana Rao et al., 2023' should be an explicit citation to reference [1] rather than a bare author-year mention.","section":"Table 6 caption"},{"comment":"Please define 'acquisition' in the post-processing timing statement. If it denotes a single switching state of about 8 seconds, 150 acquisitions correspond to about 20 minutes, consistent with the stated '~25 minutes of observation'; if it denotes a full six-state cycle, the timing would be different.","section":"Section 5"},{"comment":"The table is difficult to read because several cells are concatenated and the column alignment is inconsistent; please reformat it into separate rows and columns.","section":"Table 5"},{"comment":"'L VDS' should be 'LVDS' in the description of the GPIO connection to the FPGA.","section":"Section 4.3"},{"comment":"The caption says that white lines denote flagged spectra, but the lower panel's vertical lines appear in a darker color; please ensure the figure colors match the caption or adjust the wording.","section":"Figure 10 caption"}],"recommendation":"major_revision","confidential_remarks":"This is a solid engineering validation paper with a useful contribution in the dynamic flagging algorithm and a clear demonstration that an RPi4B can handle the data-acquisition role. The main obstacle is that the headline residual numbers depend on an unspecified smooth fit and unquoted flagging thresholds; these are fixable within the manuscript's scope by adding the missing specification and a radiometer-equation check. The extrapolation that a 'space-qualified SBC with specifications similar to the RPi4B will perform comparably' is an assumption that should be clearly labeled as such. The paper fits the journal's scope and, with the required additions, would be acceptable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the hardware work is real, the dynamic flagging idea is genuinely new, and the 44-hour dataset is a useful validation. But the paper's central claim — residuals at 12.5 mK 'devoid of systematics' — cannot be checked as written. The 'maximally smooth function' is never defined, and the dynamic flagging thresholds are described as 'tuned' without numbers. The stress-test note is right: a sufficiently flexible smooth model can absorb smooth instrumental systematics (and, in the sky case, a smooth cosmological signal), so a small Gaussian residual after the fit does not by itself establish the absence of systematics.\n\nWhat the paper does well: the RPi4B replacement for a laptop is a sensible SWaP move, the SBC survey in Tables 4–5 is thorough, and the dynamic flagging algorithm targets real artifacts (data drops) that standard RFI flagging misses. The 44-hour residual measurement against 100 Gaussian realizations is a reasonable statistical check — it tests Gaussianity of the post-fit residual, which is necessary but not sufficient for the claim. The authors are also honest about limitations: 3–10% data corruption, slow RPi4B post-processing, and the need for ground-based high-performance processing. The self-citation to SARAS/pSPEC is appropriate given the shared lineage.\n\nSoft spots, in proportion. The undefined smooth function is the load-bearing one; it must be specified with basis, order, smoothness penalty, and number of parameters. The flagging thresholds (MAD multiplier, contiguous-channel count, window size/step) also need values and ideally a sensitivity study. Without those, the residual statistics are not reproducible. There is also an internal consistency issue: 72 mK over 8 h implies ~31 mK over 44 h for thermal noise, yet the paper reports 12.5 mK. That is a factor of ~2.4 improvement, which either means the two runs had different system temperatures (plausible, but not stated) or the smooth fit absorbed more in the longer dataset. The paper should reconcile this. Finally, calling the RPi4B a 'real-time processor' is an overclaim given the text says 25 min of data takes ~100 min to process on the Pi; a one-line clarification that real-time refers to acquisition, not processing, fixes it.\n\nVerdict: this deserves a serious referee, not a desk reject. The engineering is significant for PRATUSH and the dynamic flagging algorithm has broader applicability. But the revised version needs to supply the missing fit/threshold specifications, address the scaling inconsistency, and soften the 'devoid of systematics' claim to something the evidence supports.","headline":"Useful engineering with an honest write-up, but the central 12.5 mK 'devoid of systematics' claim is unverifiable until the smooth fit and flagging thresholds are specified.","tokens_in":29291,"tokens_out":3760,"would_cite":false,"duration_ms":42708,"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 Raspberry Pi 4B can act as the master controller, real-time processor, and data recorder for the PRATUSH digital correlation spectrometer, with 44-hour residual spectra of 12.5 mK that are consistent with thermal noise.","keywords":["Raspberry Pi 4 Model B","single-board computer","FPGA","digital correlation spectrometer","PRATUSH","dynamic flagging","21-cm global signal","size weight and power"],"falsifier":"Split the 44-hour dataset into two independent halves and process each through the same calibration, dynamic flagging, and maximally smooth fit; if the residuals are pure thermal noise, each half should have rms near $\\sqrt{2}\\times 12.5$ mK, and the difference of the two half-spectra should be Gaussian with rms near 25 mK, with no correlation to the full-spectrum residuals, whereas any correlated structure would reveal a fixed fitting or instrumental artifact.","tokens_in":28265,"feed_emoji":"📡","tokens_out":11114,"duration_ms":116531,"temperature":0.7,"pith_summary":"PRATUSH is a proposed space radiometer that will try to detect the sky-averaged 21-cm signal from Cosmic Dawn, an extremely faint signal buried under bright foregrounds. This paper argues that the laboratory model of its digital receiver, a pSPEC FPGA spectrometer paired with a Raspberry Pi 4B single-board computer in place of a laptop, is ready for that task. The paper validates the integrated system against precision terminations and an antenna simulator, and reports that after calibration, dynamic flagging, and a maximally smooth fit, 44 hours of effective data leave Gaussian residuals with an rms of 12.5 mK at native resolution, and 3.4 mK after smoothing to 610 kHz. The claim is that this residual is dominated by thermal noise rather than receiver systematics, so the SBC-based digital system meets PRATUSH's science requirements.","feed_headline":"Raspberry Pi controller hits 12.5 mK radio noise floor","feed_subtitle":"A low-power SBC passes the PRATUSH 21-cm receiver test, with residuals at millikelvin level.","key_machinery":"The argument is carried by the full digital receiver chain: two 10-bit ADCs sampling at 250 Msps feed a Virtex-6 FPGA that applies a four-term window, computes a 16,384-point FFT, and accumulates self- and cross-power spectra in an X-engine with 134 ms on-chip integration; the integrated spectra are packetized as UDP over 1 GbE and received by the Raspberry Pi 4B, which generates calibration control signals, runs the acquisition script, and executes a time-domain dynamic flagging algorithm that flags channels deviating by a threshold number of median absolute deviations from the median spectrum and drops any spectrum with 16 consecutive flagged channels. The validation metric is the behavior of residuals after subtracting a maximally smooth function: their Gaussianity and the decrease of their rms with frequency smoothing are taken to show the residuals are thermal noise.","core_discovery":"The paper's central claim is that the PRATUSH laboratory digital receiver, built around the pSPEC platform and controlled by a Raspberry Pi 4B, achieves thermal-noise-limited performance suitable for the experiment. On 44 hours of data taken with an antenna simulator load, subtracting a maximally smooth function from the calibrated, averaged, dynamically flagged spectrum leaves residuals with a Gaussian distribution and an rms of 12.5 mK at the native 30.51 kHz resolution, falling to 3.4 mK after 610 kHz boxcar averaging. The authors present this as evidence that the FPGA spectrometer and SBC-based acquisition and processing pipeline introduce no non-smooth systematic features above the millikelvin level, and that a space-qualified SBC with comparable specifications should be viable for the flight model.","pith_inferences":["An implication the authors leave implicit is that the same FPGA-plus-SBC architecture could serve as a low-power template for other lunar or deep-space radio instruments, since the data rate here, about 8 MBps and 0.2 TB per 8 hours, is modest by modern standards.","If the residual Gaussianity survives a fully specified maximally smooth fit, the pipeline could set meaningful upper limits on spectrally structured signals at the few-millikelvin scale after longer integrations, since thermal noise continues to fall as the inverse square root of integration time.","A natural testable extension is to move the dynamic flagging algorithm from post-processing onto the SBC or into the FPGA itself, so data-quality screening happens in real time and downlink volume can be reduced; the paper only demonstrates the algorithm offline.","The current validation used standard terminations and an antenna simulator; the decisive next test the paper has not yet performed is an antenna-connected sky observation, where antenna reflection and environmental systematics enter the same residual analysis."],"forward_implications":["An SBC-class computer can replace a laptop as the controller and data recorder of a high-dynamic-range correlation spectrometer, cutting the digital receiver's size, weight, and power by about 40%.","The dynamic flagging algorithm can remove the broadband step-like artifacts caused by SBC- and SD-card-related data drops, which ordinary per-channel RFI flagging misses, recovering 3 to 10% of a long observation run.","The reported residual floor implies that the integrated analog-plus-digital receiver chain is sensitive at the millikelvin level, meeting the dynamic range and spectral smoothness PRATUSH requires to search for a 21-cm signal.","Because the residual rms scales with integration time and smoothing in the way thermal noise does, longer observations should continue to improve sensitivity, and the firmware is portable to space-grade FPGAs with reported resource margins.","On the RPi4B, post-processing a 25-minute observation takes about 100 minutes, so the authors conclude that raw data should be downlinked and processed on the ground whenever bandwidth allows."],"supporting_citations":[{"why":"This reference defines PRATUSH science requirements, the 55-110 MHz band, the calibration states, and the signal amplitude targets the digital receiver must meet.","marker":"[1]"},{"why":"This reference provides the pSPEC board design and the FPGA-based digital correlation spectrometer architecture that the RPi4B controls.","marker":"[52]"},{"why":"This reference supplies the six-state bandpass calibration methodology and receiver design adapted for the PRATUSH laboratory model.","marker":"[51]"},{"why":"This reference establishes the earlier global 21-cm radiometer context and the acquisition and flagging approach the SBC pipeline extends.","marker":"[10]"},{"why":"This reference defines the four-term window applied before the FFT to suppress spectral leakage, a key step in controlling spectral systematics.","marker":"[66]"},{"why":"This reference provides the post-processing software used for flagging, calibration, averaging, and file handling in the pipeline.","marker":"[89]"},{"why":"This reference introduces the pSPEC platform as a generic precision spectrometer, the hardware basis of the PRATUSH digital receiver.","marker":"[54]"}],"fun_headline_variants":["Raspberry Pi drives 12.5 mK noise floor for PRATUSH","SBC-based controller hits millikelvin precision for 21-cm study","PRATUSH receiver with Pi 4B achieves 12.5 mK residuals","Low-power SBC passes PRATUSH's millikelvin test","Pi-based spectrometer reaches 3.4 mK after averaging"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The 12.5 mK residual claim rests on the unstated assumption that the maximally smooth curve subtracted before the residual analysis absorbs all smooth instrumental response without hiding real spectral structure, since the paper does not specify that curve's form or flexibility.","fun_headline_variants_meta":{"raw":{"variants":["Raspberry Pi drives 12.5 mK noise floor for PRATUSH","SBC-based controller hits millikelvin precision for 21-cm study","PRATUSH receiver with Pi 4B achieves 12.5 mK residuals","Low-power SBC passes PRATUSH's millikelvin test","Pi-based spectrometer reaches 3.4 mK after averaging"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000173,"raw_usage":{"total_tokens":1275,"prompt_tokens":936,"completion_tokens":339,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":552,"completion_tokens_details":{"reasoning_tokens":239}},"tokens_in":552,"tokens_out":339,"duration_ms":4006,"temperature":1.0,"reasoning_tokens":239,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T19:21:54.424800+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Split the 44-hour dataset into two independent halves and process each through the same calibration, dynamic flagging, and maximally smooth fit; if the residuals are pure thermal noise, each half should have rms near $\\sqrt{2}\\times 12.5$ mK, and the difference of the two half-spectra should be Gaussian with rms near 25 mK, with no correlation to the full-spectrum residuals, whereas any correlated structure would reveal a fixed fitting or instrumental artifact.","supporting_citations":[{"cited_title":"XXIth URSI General Assembly and Scientific Symposium (URSI GASS) (2014)","cited_arxiv_id":null,"evidence_quote":"This reference introduces the pSPEC platform as a generic precision spectrometer, the hardware basis of the PRATUSH digital receiver."}],"review_version":1}