{"id":"25cc0d50-6cba-409f-a391-37f19da29a57","arxiv_id":"2507.21865","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"First demonstration of SCL-band long-haul (1552 km) transmission over G.654.E-compliant fibre, achieving 100.85 Tb/s GMI with negligible penalty below the 1520 nm cutoff.","lead":"Researchers sent S, C, and L band signals through a low-loss G.654.E optical fibre over 1552 km, reaching an information-theoretic rate of about 101 Tb/s. The work suggests that ultrawideband long-haul links could rely on simpler lumped amplifiers instead of power-hungry Raman pumping.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 100.8 Tb/s headline rests on ASE-emulated WDM neighbours; if real modulated channels induce different XPM/FWM, the throughput and G.654.E-vs-G.652.D comparison could shift.","rationale":"The reader's weakest assumption is the faithfulness of ASE-emulated WDM neighbours, and I agree this is the most load-bearing point. Every quantitative claim in the paper—the 100.85 Tb/s GMI figure, the 97.02/92.8 Tb/s numbers, and the comparison between G.654.E and G.652.D—is obtained by measuring one real channel while all other channels are ASE noise. If nonlinear interference from real modulated neighbours differs from that of shaped ASE, the reported throughput is not necessarily what a fully populated system would deliver. This is not an accusation of error: the experiment is internally consistent, and the qualitative feasibility conclusion could still hold. But the central quantitative claim would be on firmer ground only after a targeted check with real neighbours. I do not find a separate, stronger objection that would change the reader's CONDITIONAL verdict; the same condition (verify the loading fidelity) is already captured, so no verdict adjustment is needed. The test I propose is deliberately small: a few real neighbour channels and/or a split-step simulation would settle whether the concern actually lands.","tokens_in":927,"tokens_out":1069,"duration_ms":110087,"concrete_test":"Repeat the transmission at the paper's optimised launch powers with the CUT surrounded by a small set of real modulated neighbours, e.g., three 112 GBd GS-QAM channels on each side at the paper's channel spacing, and compare the CUT's SNR/GMI with the ASE-loaded case for both G.654.E and G.652.D fibres. If per-channel GMI shifts by more than roughly 0.1 bit/4D-sym (or SNR by more than ~0.5 dB), the headline throughput and fibre comparison need re-evaluation. A complementary split-step simulation with real modulated neighbours versus shaped ASE over 1552 km could isolate the physics before a lab rerun.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim (abstract: 100.8 Tb/s GMI over 1552 km; Fig. 6) is measured with one real 112 GBd channel per run, while all co-propagating WDM channels are replaced by spectrally-shaped ASE (Fig. 2). Total throughput is then the sum of single-channel GMI values. ASE loading fixes only the average power spectral density of the interfering channels. It does not reproduce the temporal intensity fluctuations or phase relationships of modulated neighbours: cross-phase modulation depends on instantaneous power, and four-wave-mixing products involving two or more neighbours depend on their modulation statistics. A stationary ASE source is close to a circular Gaussian process and thus to the GN-model assumption, whereas real GS-16/64-QAM neighbours have different excess kurtosis and symbol-rate-dependent coherence. The paper also uses a simplified ISRS-GN model [19] to optimise powers, so both experiment and model share the Gaussian-noise assumption; format-dependent nonlinear interference would therefore affect both the absolute rate and the G.654.E-vs-G.652.D comparison in Fig. 6. The S-band is below the fibre cutoff, so MPI/mode-coupling could also interact differently with real modulated co-propagating channels than with ASE. The manuscript does not provide evidence that ASE loading is faithful for these inter-channel nonlinear effects.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports an experimental S+C+L-band (SCL-band) transmission over 1552 km of G.654.E-compliant Vascade EX2500 fibre, using 112 GBd channels with one real modulated channel under test per run and all co-propagating WDM channels emulated by spectrally shaped ASE. The authors report 100.85 Tb/s total throughput estimated from GMI with Raman amplification, 97.02 Tb/s without Raman, and decoded net rates of 92.8 Tb/s and 88.8 Tb/s, respectively. These results are compared with a G.652.D low-water-peak fibre link with Raman amplification, which gave 89.9 Tb/s decoded rate. The central claim is that G.654.E-compliant fibre, despite its ~1520 nm cutoff, supports long-haul SCL-band transmission with performance comparable to G.652.D with distributed Raman amplification, but without requiring Raman pumps.","tokens_in":6758,"tokens_out":4712,"duration_ms":59796,"significance":"If the result holds, it is a significant experimental demonstration: it would show that ultra-low-loss, large-effective-area G.654.E fibre can extend three-band WDM transmission to long-haul distances with only lumped amplification, potentially simplifying UWB system architectures. The paper has clear strengths: it directly compares G.654.E and G.652.D fibres in the same recirculating-loop setup, reports Raman on-off gain measurements, distinguishes GMI estimates from decoded net rates, and includes both Raman-amplified and DFA-only scenarios. The main caveat is that the headline throughput is a GMI sum over single-channel measurements with ASE-emulated neighbours, so the fully populated WDM performance is not directly measured; this is the load-bearing assumption that needs explicit validation or qualification.","major_comments":[{"comment":"The reported totals of 100.85 Tb/s and 97.02 Tb/s are computed by summing per-channel GMI values measured with only one real 112 GBd channel per run, while all co-propagating WDM channels are emulated by spectrally shaped ASE. The paper does not demonstrate that ASE loading reproduces the inter-channel nonlinear effects of real modulated neighbours, specifically cross-phase modulation, four-wave mixing, ISRS, and, in the below-cutoff S-band, MPI. Since the central claim concerns the feasibility and throughput of a fully populated SCL-band WDM system, this assumption is load-bearing. A validation experiment with two or more real modulated neighbours, or a measurement of nonlinear-interference scaling with channel count for ASE versus modulated loading, is needed before the headline throughput can be taken at face value.","section":"Experimental setup (Fig. 2) and Results (Fig. 6)"},{"comment":"No repeated measurements, error bars, or uncertainty estimates are reported. The claims of negligible MPI penalty, the 0.86 bit/4D-symbol S-band improvement with Raman, and the 0.67 bit/4D-symbol C-band advantage of G.654.E over G.652.D rest on single measurements per scenario. Given the small differences involved, the absence of any repeatability or statistical confidence statement makes it difficult to assess whether these comparisons are significant.","section":"Results (Fig. 6)"},{"comment":"The definition of 'decoded net data rate' is incomplete. The text states only that pilot overhead was deducted; it does not specify the FEC code rate, FEC overhead, or any other overheads included in the adaptive rate decoding. Since the abstract and conclusion emphasize throughput, the relation between the GMI-based estimate (100.85 Tb/s) and the decoded rate (92.8 Tb/s) should be stated precisely.","section":"Results, decoded net rate"}],"minor_comments":[{"comment":"The inset shows '116 GHz' while the text states a baud rate of 112 GBaud; please clarify whether 116 GHz is the WDM grid spacing and, if so, state the number of WDM channels and the total bandwidth accordingly.","section":"Fig. 2 inset"},{"comment":"The band labels 'SCL', 'CLU', 'SCLU', and 'SCLUX' in the figure legend are not defined in the caption; please define them for readers unfamiliar with the notation.","section":"Fig. 1"},{"comment":"The abstract correctly qualifies the headline as '100.8 Tb/s (GMI)', but the conclusion repeats 'more than 100 Tb/s throughput' without the GMI qualifier; recommend consistently noting that this figure is a GMI-based estimate, not a decoded rate.","section":"Abstract and Conclusion"},{"comment":"The geometric shaping parameters for GS-16-QAM and GS-64-QAM (e.g., number of constellation points, shaping block length) are not given; adding these details would improve reproducibility.","section":"References [16] and modulation formats"}],"recommendation":"major_revision","confidential_remarks":"The paper is a useful experimental contribution and is within scope. The main substantive issue is the ASE-emulated WDM loading: the central throughput claim depends on the unverified assumption that ASE loading faithfully reproduces the nonlinear interference of real modulated neighbours. The lack of uncertainty quantification compounds this concern. I do not see grounds for rejection, but the authors should either validate the emulation with real neighbouring channels or explicitly re-frame the headline as an ASE-loaded estimate, and they should add repeatability information."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my read of arXiv:2507.21865. It's a solid record-type demonstration: first SCL-band long-haul (1552 km) transmission over G.654.E fiber, with 100.85 Tb/s from GMI and 92.8 Tb/s decoded. The main practical claim is that G.654.E with only lumped amplification roughly matches G.652.D with 1.5 W of Raman pumping. That's an interesting result for future line-system design.\n\nWhat's genuinely new is the extension of the group's S-band-only work to S+C+L over G.654.E, plus a direct head-to-head measurement against a low-water-peak G.652.D fiber. The Raman on/off comparison and the per-channel SNR/GMI are internally consistent. The claim of negligible MPI penalty below cutoff is supported indirectly by the similar S-band SNR to G.652.D; it's not a direct MPI measurement, but it's reasonable.\n\nThe soft spots are the ones the stress-test flags. The headline throughput is a GMI estimate, and the WDM neighbors are spectrally-shaped ASE, not real modulated channels. That's a genuine limitation: ASE loading fixes the average power spectral density but doesn't reproduce the temporal statistics that drive XPM and FWM. Since the power optimization also uses the group's own ISRS-GN model—which rests on the same Gaussian-noise assumption—the absolute rates and the G.654.E vs G.652.D comparison could shift in a fully populated system with real neighbors. I don't think this invalidates the qualitative feasibility claim, but it does mean the numbers should be treated as conditional.\n\nAlso minor: no error bars or repeated runs are reported, which is common in this kind of demonstration but worth noting.\n\nOverall, the paper is honest and clearly written. It deserves a serious referee: the comparison is useful, and the record is real. For peer review, I'd want the authors to address the ASE-emulation fidelity head-on, add a repeat measurement or at least a discussion of variability, and be careful to label the 100.85 Tb/s as GMI-based in the abstract to avoid overclaiming.","headline":"A credible first SCL-band long-haul result on G.654.E fiber; the headline rate is GMI-based with ASE-emulated WDM neighbors, but the qualitative feasibility claim holds.","tokens_in":7356,"tokens_out":2013,"would_cite":true,"duration_ms":22816,"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":"This paper reports the first SCL-band long-haul transmission over G.654.E-compliant fibre, reaching 100.85 Tb/s of GMI over 1552 km despite the fibre's 1520 nm cutoff, and shows the link can match a Raman-amplified G.652.D system with…","keywords":["SCL-band transmission","G.654.E fibre","ultra-wideband WDM","stimulated Raman scattering","lumped amplification","geometric constellation shaping","multipath interference","1552 km transmission"],"falsifier":"Transmit a fully populated or densely sampled grid of real 112 GBd modulated channels across S, C, and L bands over the same 18-span G.654.E loop and compare per-channel SNR and total GMI; if the real full-grid S-band SNR or total throughput falls below the shaped-ASE result by more than the system margin, the reported 100.85 Tb/s is not a faithful capacity for this fibre.","tokens_in":6307,"feed_emoji":"📡","tokens_out":10494,"duration_ms":105570,"temperature":0.7,"pith_summary":"The paper sets out to show that G.654.E-compliant fibre, despite a cutoff wavelength near 1520 nm that intrudes into the S-band, can support ultra-wideband S+C+L transmission over long haul. It reports the first such demonstration: 112 GBd channels over 1552 km of G.654.E fibre, with 100.85 Tb/s of total GMI (generalised mutual information, a rate estimate) and 92.8 Tb/s decoded when E-band Raman pumps are used, and 88.8 Tb/s decoded with only lumped amplifiers. The practical point is that the fibre's ultra-low loss and larger effective area compensate for its higher water-absorption peak and weaker Raman response, so a G.654.E link with plain lumped amplification can match a Raman-amplified G.652.D link. If correct, this would remove one of the main complexities of long-haul ultra-wideband systems.","feed_headline":"First S+C+L long-haul link on G.654.E fibre hits 100.85 Tb/s","feed_subtitle":"Lumped-only amplification matches Raman-amplified standard fibre, so ultra-wideband long-haul systems may skip Raman pumps.","key_machinery":"The central object is the G.654.E-compliant fibre span itself: a 86.2 km Vascade EX2500 fibre with about 0.148 dB/km attenuation at 1550 nm and about 125 square micrometres of effective area. Its low loss and low nonlinearity are what allow a DFA-only SCL-band link to match a Raman-amplified G.652.D system, while its 1520 nm cutoff does not produce a measurable MPI penalty. The experimental machinery around it is a recirculating loop with three lumped gain blocks (an S-band TDFA plus C- and L-band EDFAs), optional backward E-band Raman pumps, and spectrally-shaped ASE noise standing in for co-propagating WDM channels; a simplified ISRS-aware Gaussian-noise model sets the pump powers and per-band launch powers. These components together determine how much of the reported throughput comes from the fibre's linear advantages rather than from Raman gain.","core_discovery":"In this paper, the authors claim that G.654.E-compliant fibre can support SCL-band long-haul transmission despite a cutoff wavelength around 1520 nm. Using 18 spans of Vascade EX2500 fibre totalling 1552 km, they transmit 112 GBd channels across 15.08 THz of S+C+L spectrum and obtain 100.85 Tb/s of total GMI with E-band Raman assistance (92.8 Tb/s after decoding), and 88.8 Tb/s decoded with only lumped S-, C-, and L-band amplifiers. The decoded throughput with lumped amplification is effectively equal to the 89.9 Tb/s decoded from a Raman-amplified low-water-peak G.652.D link on the same span length, while the GMI comparison is 97.02 Tb/s versus 97.90 Tb/s. The authors conclude that the fibre's ultra-low loss and larger effective area compensate for its higher water-absorption peak and roughly halved Raman gain coefficient, making DFA-only UWB long-haul transmission viable on G.654.E.","pith_inferences":["Editorial extension: since co-propagating neighbours were shaped ASE rather than live modulated channels, the 100.85 Tb/s GMI is a projected capacity; a fully populated WDM comb could shift SNR through cross-phase modulation and four-wave mixing that broadband noise does not reproduce.","Editorial extension: the same fibre's weaker Raman response means designs that retain distributed Raman should re-optimise pump wavelengths around the E-band water peak rather than simply scaling G.652.D pump plans.","Editorial extension: if production G.654.E fibres replicate this span loss and effective area, ultra-wideband long-haul repeater sites could drop Raman hardware altogether, replacing a roughly 1.5 W pump stage with lumped DFA-only gain blocks.","Editorial extension: the simplified ISRS GN-model optimisation could be validated against per-channel SNR in a fully populated system; agreement would make the model a practical design tool for G.654.E UWB links."],"forward_implications":["With Raman pumps, the G.654.E link achieves 100.85 Tb/s GMI (92.8 Tb/s decoded) over 1552 km across 15.08 THz of S+C+L spectrum at 112 GBd.","Without Raman, the same G.654.E link delivers 88.8 Tb/s decoded, comparable to the 89.9 Tb/s of a Raman-amplified G.652.D link, so distributed Raman is not required to match standard fibre performance.","Signals below the 1520 nm cutoff suffer negligible multipath-interference penalty, so the S-band remains usable on cutoff-shifted G.654.E fibre.","E-band Raman pumping mainly helps the short-wavelength S-band, improving average GMI by 0.86 bit/4D-symbol for 1480 to 1500 nm channels, while C- and L-band performance is largely unchanged.","These results position G.654.E fibre as suitable for DFA-only long-haul ultra-wideband transmission, reducing amplifier complexity relative to Raman-based UWB links."],"supporting_citations":[{"why":"Establishes the theoretical feasibility of S-band transmission over G.654.E despite the cutoff wavelength, which this work extends to the full SCL band.","marker":"[10]"},{"why":"Prior experimental demonstration of MPI-penalty-free S-band transmission over G.654.E fibres, the direct precursor to the full SCL result.","marker":"[11]"},{"why":"Documents the low loss and large effective area of G.654.E fibres for long-haul, high-data-rate terrestrial systems.","marker":"[9]"},{"why":"Quantifies the lower Raman gain coefficient of G.654.E compared with G.652.D, explaining the maximum-pump-power operating point and the Raman inefficiency.","marker":"[13]"},{"why":"Supplies the closed-form ISRS Gaussian-noise model used to optimise pump powers and per-band launch powers for both fibre types.","marker":"[19]"},{"why":"Provides the DFA-only S+C+L-band transmission baseline and adaptive-rate decoding approach used for throughput estimation.","marker":"[5]"},{"why":"Recent Raman-amplified triple-band transmission over standard single-mode fibre that serves as the performance benchmark for the G.654.E result.","marker":"[4]"}],"fun_headline_variants":["SCL-band on G.654.E fibre: 100.8 Tb/s over 1552 km","Lumped amps rival Raman on G.654.E SCL-band link at 100.8 Tb/s","100.8 Tb/s SCL-band over G.654.E despite 1520 nm cutoff","G.654.E SCL-band: DFA-only 88.8 Tb/s equals Raman G.652.D","First SCL-band long haul on G.654.E: 100.8 Tb/s GMI"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The 100.85 Tb/s headline assumes that spectrally-shaped ASE noise faithfully emulates the co-propagating WDM channels, so that the measured SNR and GMI reflect true multi-channel nonlinear propagation rather than a stand-in that could miss cross-phase modulation and four-wave mixing.","fun_headline_variants_meta":{"raw":{"variants":["SCL-band on G.654.E fibre: 100.8 Tb/s over 1552 km","Lumped amps rival Raman on G.654.E SCL-band link at 100.8 Tb/s","100.8 Tb/s SCL-band over G.654.E despite 1520 nm cutoff","G.654.E SCL-band: DFA-only 88.8 Tb/s equals Raman G.652.D","First SCL-band long haul on G.654.E: 100.8 Tb/s GMI"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001032,"raw_usage":{"total_tokens":4303,"prompt_tokens":860,"completion_tokens":3443,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":476,"completion_tokens_details":{"reasoning_tokens":3308}},"tokens_in":476,"tokens_out":3443,"duration_ms":28631,"temperature":1.0,"reasoning_tokens":3308,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T12:16:06.938596+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Transmit a fully populated or densely sampled grid of real 112 GBd modulated channels across S, C, and L bands over the same 18-span G.654.E loop and compare per-channel SNR and total GMI; if the real full-grid S-band SNR or total throughput falls below the shaped-ASE result by more than the system margin, the reported 100.85 Tb/s is not a faithful capacity for this fibre.","supporting_citations":[{"cited_title":"G. 654. E optical fibers for high-data-rate terrestrial transmission systems with long reach","cited_arxiv_id":null,"evidence_quote":"Documents the low loss and large effective area of G.654.E fibres for long-haul, high-data-rate terrestrial systems."},{"cited_title":"Performance analysis of lower Raman gain coefficient G.654.E fiber with distributed Raman amplifiers","cited_arxiv_id":null,"evidence_quote":"Quantifies the lower Raman gain coefficient of G.654.E compared with G.652.D, explaining the maximum-pump-power operating point and the Raman inefficiency."},{"cited_title":"Net 107.7- Tb/s triple-band WDM transmission over 1200-km single- mode fiber with forward- and backward-pumped dis- tributed Raman amplifiers","cited_arxiv_id":null,"evidence_quote":"Recent Raman-amplified triple-band transmission over standard single-mode fibre that serves as the performance benchmark for the G.654.E result."}],"review_version":1}