{"id":"64951a20-ed1d-4c27-9998-47cf0617e087","arxiv_id":"2608.11659","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Measured satellite emissions already exceed single-dish radio astronomy protection criteria at multiple frequencies, and the second harmonic of a 2620 MHz direct-to-device downlink causes at least 59% data loss for geodetic VLBI today.","lead":"This paper uses new telescope measurements of radio emissions from satellite constellations to estimate how much they damage radio astronomy and geodetic VLBI, and to propose how quiet each satellite must be. It finds some telescope observations are already fully blocked and gives limits in the units regulators use.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The per-satellite limits in Table 10 and the 59%/100% VLBI data loss both rest on treating the 5240 MHz DTD harmonic as a constant, near-isotropic radiator; the strongest check is to re-derive the aggregate from per-pass flux statistics rather than the single compressed peak.","rationale":"The reader's weakest_assumption focuses on the DTD share of future fleets (Section 4.2's explicit anti-conservative direction). That is a real but secondary issue: the paper itself flags it, and the scaling is linear, so a larger DTD fraction merely strengthens the conclusion. The more load-bearing spot is the per-satellite amplitude and isotropy of the 5240 MHz harmonic, which is the single input that drives the 59%/100% VLBI data loss and the Table 10 ceiling. The text states the ATCA flux is a compressed lower bound, and the exact per-satellite level used is the peak of 19/8/5 tracked passes, converted to an isotropic EIRP at unity duty cycle. A compressed peak from 21-of-36 compressing passes is a plausible but unverified representation of the fleet average; it is the one place where the central quantitative claims could move by more than the quoted error bars. The paper's internal conservatism arguments about beam-forming do not cover this, since the harmonics are treated as isotropic precisely to justify using one number. This is not an objection to the paper's existence or direction; the conditional verdict stands, but for a different reason than the reader's. A direct per-pass distribution check would settle whether the 59% figure is robust, and is cheap to run from the existing data.","tokens_in":59491,"tokens_out":1785,"duration_ms":17280,"concrete_test":"Recompute the 5240 MHz EPFD aggregate using the full per-pass flux time series from SNIFFLES-I for the 36 DTD passes, not just the compressed peak Y-factor value. Specifically: (a) report the distribution of range-corrected per-pass peak and mean spectral flux densities; (b) re-run the Hobart Monte-Carlo with per-satellite levels drawn from that distribution (mean and median rather than peak) and with a duty-cycle derived from the 21-of-36 compression fraction; (c) if the resulting present-fleet VLBI data loss drops below 50% or the per-satellite ceiling rises by more than 5 dB, the headline claim rests on the peak-level convention and needs to be re-bracketed.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central VLBI result is the 5240 MHz DTD second harmonic: 59% data loss today, 100% at ~100k–300k satellites, and a per-satellite ceiling of ~65–70 dB(uV/m) at 10 m. The per-satellite level for that harmonic is a single measurement given as a lower bound (Table 5: ATCA >= -76.1 dB(W/Hz), E_meas = 98.7 dB(uV/m)), described in Section 7.1 as driving the ATCA back-end into compression in 21 of 36 DTD passes. Using a compressed, peak-selected level as the representative isotropic EIRP for every DTD satellite, at unity duty cycle, is the load-bearing step for both the present-fleet exceedance and the future scaling. The paper argues the isotropic convention is conservative for beamformed emission, but for the harmonics it argues near-isotropy from 'flat to within ~2 dB' tracking of 19, 8 and 5 passes respectively. With only 5–19 passes and no per-pass flux distribution or off-axis dependence published, the near-isotropy claim is not fully demonstrated; if the harmonic is instead patchy in elevation or azimuth, the aggregate EPFD and the 59% figure could shift. This is the weakest load-bearing link because all downstream numbers (Tables 10–11 limits, Figure 7 curves) scale linearly with this single per-satellite input, and the paper itself flags the input as a compressed lower bound.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper combines the SNIFFLES-I survey measurements of intended emissions, spurious emissions, and unintended electromagnetic radiation (UEMR) from NGSO constellations with a Monte-Carlo equivalent power flux density (EPFD) model to quantify interference to Australian radio astronomy facilities and to the AuScope geodetic VLBI (VGOS) array across 1–14 GHz. It then scales the present catalogued fleet to 100,000 and 300,000 satellites and inverts the ITU-R RA.769 protection criteria to derive per-satellite emission and radiation ceilings in dB(µV/m) at 10 m in a 1 MHz reference bandwidth. The central claims are that the single-dish continuum criterion is already exceeded at all measured bright detections, that the 5240 MHz second harmonic of the 2620 MHz Direct-to-device (DTD) downlink causes at least 59% VLBI data loss today and 100% at 300,000 satellites, that the corresponding per-satellite VLBI ceiling is 65–70 dB(µV/m) while current spacecraft exceed it by 20–74 dB, and that narrowband UEMR is diluted against the 32 MHz VGOS channel and does not breach the VLBI criterion. A separate, more speculative analysis treats proposed orbital data centres as a UEMR-dominated megaconstellation.","tokens_in":59830,"tokens_out":6709,"duration_ms":74248,"significance":"If the results hold, this paper provides the first directly measurement-anchored, quantitative bridge between operational NGSO constellation emissions and the per-satellite limits that standard-setting bodies could apply, which is timely and policy-relevant for the proposed WRC-31 agenda item on geodetic VLBI. The manuscript has clear strengths: the EPFD engine is validated against a dedicated 2098 s observing run (§4.5); the linear-in-N scaling is checked against direct Monte-Carlo simulations of genuinely up-scaled fleets (§6.1); and the per-satellite inversion in Eq. (7) is genuinely parameter-free with respect to the measured level, since the measured EIRP cancels. The full data grid and plot set are available, supporting reproducibility. The main risks are concentrated in the small-sample near-isotropy assumption for the DTD harmonics and in the frozen future-fleet composition, both of which the authors partially acknowledge but which affect the headline numerical limits.","major_comments":[{"comment":"The 5240 MHz DTD second harmonic is the principal VLBI threat, but its per-satellite input is a single compressed peak-selected lower bound (ATCA ≥ −76.1 dB(W/Hz); E_meas = 98.7 dB(µV/m)) that drove the calibrating back-end into compression in 21 of 36 passes. The paper uses this value as a constant, near-isotropic, unity-duty-cycle radiator for every DTD satellite, and the near-isotropy claim rests on only 19, 8, and 5 tracked passes for the second, third, and fourth harmonics, with no per-pass flux distribution or off-axis dependence shown. Because the aggregate, the 59% VLBI loss, N_breach ≈ 188, and the +32.0 dB exceedance in Table 10 all scale linearly with this single input, I ask for per-pass flux statistics, an explicit compression correction or bound, and a demonstration that the conclusions are unchanged if the harmonic is patchy in elevation or azimuth. The direction of the error from compression is conservative (true loss is higher), but the numerical limits and breach populations are not protected by that direction.","section":"§4.3, Table 5, §7.1"},{"comment":"The 300,000-satellite scenario fixes the DTD-capable share at 5.4% of the fleet and the measured active fraction at 0.869. The authors correctly state in §4.2 that this is the one direction in which the assumptions are anti-conservative, and they quote dB shifts for larger DTD shares in the text, but the headline per-satellite ceilings (65–70 dB(µV/m) for VLBI; 18–46 dB(µV/m) for single-dish) are tabulated without this dependence. Since these ceilings are the paper's central deliverable for standard-setting bodies, the tables should state the assumed DTD share and include at least a one-line sensitivity (for example, 50% and 100% DTD share) so that a regulator does not apply a ceiling that is too loose under a plausible future composition. I do not regard this as an internal inconsistency; it is an applicability gap in the presentation of the main numerical product.","section":"§4.2, Tables 10 and 11"},{"comment":"The authors show that the rigid 10 log10(N/N0) shift underestimates the true VLBI data loss for the not-yet-saturated detections: the 7860 MHz third harmonic reaches a true 92% loss at 300,000 satellites against 71% from the linear estimate, and the corrected compound-resampling method reproduces direct simulations to within a few points. Table 8 and the main text nevertheless quote the linear-estimate value of 68% at 300,000. Given that the corrected method is already validated and described in the same section, the tabulated and narrative loss figures should either be the corrected values or be explicitly labelled as analytic lower bounds. This matters because 68% is one of the headline quantitative claims for the 300,000-satellite scenario and the paper already possesses the machinery to state the more accurate number.","section":"§6.1, Figure 10, Table 8"}],"minor_comments":[{"comment":"There is a typo in the sentence describing the beam alignment: 'closely aligned towawrds telescope boresight' should read 'closely aligned towards telescope boresight'.","section":"§4.3"},{"comment":"The caption text 'all expressed at 10 min a 1 MHz reference bandwidth' should read 'all expressed at 10 m in a 1 MHz reference bandwidth'; the current phrasing is confusing.","section":"Figure 14 caption"},{"comment":"The table uses red text to mark placeholder and floor-only entries, but this information will be lost in monochrome print; please add a symbol or footnote marker in addition to colour.","section":"Table 5"}],"recommendation":"major_revision","confidential_remarks":"The manuscript fits the journal's scope and I see no novelty-disclosure concern; the reliance on the authors' own SNIFFLES-I survey is an observational input, not a circular derivation. The main risk is the small-sample near-isotropy claim for the harmonics and the use of a compressed lower bound as the representative per-satellite level; if the authors cannot supply per-pass flux statistics, the 59%/100% VLBI numbers should be cast consistently as lower bounds throughout, including in the abstract. The ODC section is more speculative than the rest of the paper, but it is clearly labelled as such and does not carry the main conclusions."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis is the first paper I've seen that turns the satellite-interference problem into actual numbers a regulator can use: per-satellite field-strength limits in dB(µV/m) at 10 m, scaled to constellation size. The central finding—that single-dish RA.769 protection is already exceeded across the sky at the present fleet, and that the 5240 MHz DTD second harmonic already costs geodetic VLBI at least 59% of observing time—is well supported. The paper does real work: it uses the SNIFFLES-I measured per-satellite levels, validates its Monte-Carlo EPFD engine against a dedicated 2000 s stare, checks the linear scaling against direct up-scaled simulations, and is explicit about where its assumptions are conservative and where they are anti-conservative.\n\nThe genuinely new pieces are the extension of aggregate-interference results from 150 MHz and 1400 MHz up to 1-14 GHz, the identification of the DTD harmonic threat to VGOS, and the inversion of EPFD into per-satellite emission ceilings. The ODC analysis is a bonus, and the discussion of how per-satellite masks fit the existing MIL-STD-461/CISPR framework is sensible.\n\nThe soft spots are real but not load-bearing. The near-isotropy of the 5240 MHz harmonic rests on tracking of only 5-19 passes and is anchored to a single compressed peak; the paper itself flags it as a lower bound. I would like to see per-pass flux statistics or at least a sensitivity analysis on that input before the 59% number is used in a regulation. Some other inputs are detection-floor placeholders without formal error bars, and the 300,000-satellite projection assumes today's radiating composition is frozen—an assumption the authors admit is anti-conservative. None of these change the qualitative picture: the margins for single-dish astronomy are tens of decibels, so even a factor-of-a-few correction in the harmonic level leaves that conclusion intact. The VLBI loss figure has more uncertainty, but the direction of the bias (compression-limited lower bound) means the problem is likely worse, not better.\n\nThis paper deserves a serious referee. It is aimed at spectrum managers, ITU-R/CISPR participants, and the VLBI and radio-astronomy communities. I'd bring it to reading group and I'd cite it.\n\nRecommendation: send to peer review. Ask the authors to make the per-pass tracking data and code available, or at least include a formal sensitivity analysis on the harmonic isotropy assumption.","headline":"Serious, measurement-grounded interference study that gives regulators usable per-satellite limits; the 59% VLBI data-loss figure is a lower bound that will shift with better harmonic characterization, but the overall result holds.","tokens_in":60432,"tokens_out":2634,"would_cite":true,"duration_ms":27163,"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":"Measured satellite emissions already breach radio astronomy protection thresholds across 1–14 GHz, and the second harmonic of one downlink already costs geodetic VLBI at least 59% of its data.","keywords":["geodetic VLBI","VGOS","radio frequency interference","satellite constellations","Earth orientation parameters","radio astronomy","unintended electromagnetic radiation","direct-to-device downlink"],"falsifier":"Track a large sample of Direct-to-device satellites with a receiver that does not compress, such as a 16-bit digitizer or an attenuated calibration path, to recover the true level and bandwidth of the 5240 MHz second harmonic over many passes; if the uncompressed level is more than about 12 dB below the apparent measured value, the paper's headline 59% VLBI data loss is an overestimate, and a direct measurement of the fraction of 2000 s VGOS scans whose aggregate exceeds the VLBI threshold would settle the matter.","tokens_in":59258,"feed_emoji":"📡","tokens_out":9931,"duration_ms":92016,"temperature":0.7,"pith_summary":"Protecting radio astronomy from satellite constellations has been argued in the abstract; this paper makes it quantitative from direct measurements. It claims that at the present fleet of roughly 12,000 satellites, the summed emissions from non-geostationary constellations already exceed the single-dish protection criteria of the radio-astronomy service across the whole sky at every frequency tested, including two primary protected bands. For geodetic VLBI, the new-generation VGOS system that anchors Earth orientation and reference frames, the binding threat is not the broad platform noise but a specific spurious harmonic: the second harmonic of the 2620 MHz Direct-to-device downlink, at 5240 MHz, which already produces at least 59% data loss and would reach 100% at about 300,000 satellites. The paper inverts the aggregate protection analysis into per-satellite emission ceilings, 65–70 dB(µV/m) at 10 m in 1 MHz for VLBI and 18–46 dB(µV/m) for single-dish astronomy at 300,000 satellites, and shows current spacecraft exceed them by tens of decibels. The consequence is that genuine coexistence requires per-satellite emission control at the spacecraft design stage, a discipline already used for launch-vehicle compatibility, rather than scheduling or filtering.","feed_headline":"One satellite harmonic already causes 59% data loss for geodetic VLBI","feed_subtitle":"At 300,000 satellites the loss reaches 100%; the paper derives per-satellite emission limits to prevent it.","key_machinery":"The carrying mechanism is a Monte-Carlo equivalent power flux density (EPFD) simulation that propagates the actual catalogued orbits of the present fleet, weights each satellite's measured emission or radiation by the receiving antenna gain toward it using a standard reference pattern, and compares the summed aggregate against the protection thresholds interpolated across 1–14 GHz. The per-satellite inputs come from a companion survey that measured flux-calibrated emission levels from about 4,600 tracked observations, separated into intended emissions, spurious emissions (harmonics and frequency-conversion products), and unintended electromagnetic radiation. An analytic scaling, verified by direct simulation of cloned fleets, extends the aggregate linearly with constellation size, and the analysis is inverted to express the protection requirement as a per-satellite field-strength limit in dB(µV/m) at 10 m in a 1 MHz bandwidth. A bandwidth correction is central: broadband harmonics fill the 32 MHz VGOS channel, while narrowband radiation is diluted and largely spares VLBI, so the bright downlink harmonics, not the platform noise, set the VLBI threat.","core_discovery":"The paper's central claim is that the aggregate radio emission and radiation from low-Earth-orbit satellite constellations, measured directly at the telescopes rather than taken from filings, already exceeds the radio-astronomy protection criteria (RA.769) for single-dish observations across 1–14 GHz, and that the dominant threat to geodetic VLBI is the spurious second harmonic of the 2620 MHz Direct-to-device downlink at 5240 MHz. Using a Monte-Carlo equivalent power flux density model over the catalogued orbits of about 12,000 satellites and scaling to 100,000 and 300,000 satellites, the authors find 59% data loss for VGOS VLBI today at 5240 MHz, a lower bound because the measurement compresses the calibrating receiver, rising to 100% at 300,000; the 7860 MHz third harmonic rises from 7.3% to 68%. Inverting the same analysis gives a per-satellite emission ceiling of about 65–70 dB(µV/m) at 10 m in 1 MHz for geodetic VLBI at 300,000 satellites, and 18–46 dB(µV/m) for single-dish radio astronomy, with current spacecraft exceeding the single-dish ceilings by 20–74 dB at every calibrated detection. The authors argue this is a spacecraft-engineering problem with a known solution, radiated-emission masks of the type spacecraft already meet for launch-vehicle compatibility, and propose a per-satellite mask that tightens as constellations grow.","pith_inferences":["If the Direct-to-device share of future fleets grows beyond today's 5.4%, every numerical limit in the paper tightens further; the 300,000-satellite column would then understate the threat, and the VLBI-friendliest DTD band near 852–960 MHz conflicts directly with the 21 cm hydrogen-line observations that occupy the same low spectrum.","The paper's bandwidth criterion suggests a cheap pre-launch certification test: measure not only the carrier power of each downlink but the bandwidth of every harmonic and clock product, since narrowband radiation is harmless to VLBI while broadband harmonics are not.","The same inversion method, turning an aggregate protection threshold into a per-unit emission ceiling, could be applied to other passive users of the spectrum or to analogous aggregate problems such as reflected sunlight from very large constellations affecting optical astronomy."],"forward_implications":["Adopting the proposed per-satellite ceilings would require satellite operators to suppress spurious emissions and unintended radiation by tens of decibels, a level the paper argues is within existing aerospace electromagnetic-compatibility practice.","If the fleet reaches 300,000 satellites with today's share of Direct-to-device payloads, VGOS geodetic VLBI loses 100% of its data at the 5240 MHz harmonic and 68% at 7860 MHz, degrading the reference frames and UT1–UTC products that underpin GNSS and timing services.","High-sensitivity single-dish radio astronomy in several 1–14 GHz bands is already foreclosed today rather than merely degraded, because the measured emissions push the whole-sky aggregate above the protection criterion at the present fleet.","A per-satellite limit that falls 10 dB per decade of fleet growth gives regulators a concrete, non-gameable target that shifts the compliance burden onto the largest constellations while letting small operators fly within a uniform mask."],"supporting_citations":[{"why":"Supplies the directly measured per-satellite emission and radiation levels, detection rates, and active fractions that drive every EPFD input.","marker":"(Indermuehle and Lourenc ¸o 2026)"},{"why":"Defines the RA.769-2 single-dish and VLBI protection thresholds against which the aggregate EPFD is compared.","marker":"(ITU-R 2003)"},{"why":"Derives VGOS-specific geodetic VLBI thresholds from the RA.769 methods, corroborating the paper's interpolated criteria to better than 1 dB.","marker":"(ITU-R 2022)"},{"why":"Established low-frequency unintended electromagnetic radiation from Starlink and the earlier aggregate-interference picture that this paper extends to 1–14 GHz.","marker":"(Di Vruno et al. 2023)"},{"why":"Documents generation dependence of unintended radiation levels, supporting the paper's use of per-version measured levels and active fractions.","marker":"(Bassa et al. 2024)"},{"why":"The filed 100,000-satellite Gen3 system and its Ku-band downlink provide the concrete large-fleet scenario and a direct intended-emission path into VGOS band D.","marker":"(Space Exploration Holdings, LLC 2026)"},{"why":"The published Gen2 Direct-to-cell peak e.i.r.p. bounds the single-satellite in-beam worst case that sets the front-end saturation analysis.","marker":"(SpaceX 2023)"}],"fun_headline_variants":["59% VLBI data loss from one satellite harmonic today","Satellite harmonic already swamps geodetic VLBI by 59%","Per-satellite emission limits derived to save radio astronomy","LEO constellations already exceed radio astronomy limits","The satellite downlink harmonic that breaks VLBI: 5240 MHz"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The projections assume the future fleet looks like today's in composition, with satellites carrying the Direct-to-device downlink remaining about 5.4% of the total and each measured emission appearing in the same fraction of the fleet as observed; if the DTD share grows, the quoted limits tighten and 100% VLBI data loss arrives before 300,000 satellites.","fun_headline_variants_meta":{"raw":{"variants":["59% VLBI data loss from one satellite harmonic today","Satellite harmonic already swamps geodetic VLBI by 59%","Per-satellite emission limits derived to save radio astronomy","LEO constellations already exceed radio astronomy limits","The satellite downlink harmonic that breaks VLBI: 5240 MHz"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000219,"raw_usage":{"total_tokens":1607,"prompt_tokens":1273,"completion_tokens":334,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":889,"completion_tokens_details":{"reasoning_tokens":249}},"tokens_in":889,"tokens_out":334,"duration_ms":3253,"temperature":1.0,"reasoning_tokens":249,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T00:31:50.052123+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Track a large sample of Direct-to-device satellites with a receiver that does not compress, such as a 16-bit digitizer or an attenuated calibration path, to recover the true level and bandwidth of the 5240 MHz second harmonic over many passes; if the uncompressed level is more than about 12 dB below the apparent measured value, the paper's headline 59% VLBI data loss is an overestimate, and a direct measurement of the fraction of 2000 s VGOS scans whose aggregate exceeds the VLBI threshold would settle the matter.","supporting_citations":[{"cited_title":"SpaceX Gen2 direct-to-cellular system, attachment a: Technical information to supplement schedule s","cited_arxiv_id":null,"evidence_quote":"The published Gen2 Direct-to-cell peak e.i.r.p. bounds the single-satellite in-beam worst case that sets the front-end saturation analysis."}],"review_version":1}