G.654.E-Compliant Fibre for Ultra-Wideband Links

• G.654.E-compliant fibre is a modern single-mode fibre with ultra-low attenuation, a large effective area, and an elevated cutoff wavelength affecting S-band use.
• Experimental studies show negligible multipath interference (MPI) in S-band transmission and demonstrate feasibility over spans up to 1552 km.
• Comparative metrics indicate up to a 3.17 dB SNR improvement over G.652.D, highlighting its advantages in reducing loss and nonlinear impairments.

Searching arXiv for the cited papers to ground the article in the current literature. G.654.E-compliant fibre is an ITU-T single-mode fibre class positioned for long-haul terrestrial and submarine transmission, and in recent S-band and S+C+L-band studies it is characterized by the combination of ultra-low attenuation and large effective area, alongside a potentially problematic long cable cutoff wavelength that can extend into the S-band. In the 2025 experimental literature, the central technical question is whether this elevated cutoff introduces higher-order-mode-related multipath interference (MPI) severe enough to compromise S-band operation. Two complementary demonstrations address that question: single-channel S-band transmission over selected high-cutoff G.654.E fibres with no meaningful SNR degradation attributable to MPI (Aparecido et al., 27 Jan 2025), and long-haul SCL-band transmission over Corning Vascade EX2500 with a cutoff wavelength around 1520 nm, where signals below cutoff were reported to incur negligible MPI penalty (Yang et al., 29 Jul 2025). Taken together, these studies present G.654.E-compliant fibre as not only compatible with S-band use under the tested conditions, but also advantageous relative to G.652.D in regimes where lower attenuation and larger effective area dominate system performance.

1. Definition and transmission-relevant properties

In the cited work, G.654.E-compliant fibre is presented as a modern low-loss, large-effective-area single-mode fibre class. Relative to conventional G.652.D fibre, the key reported properties are lower attenuation and larger effective area. The papers quote 0.148 dB/km at 1550 nm and 125 \,\mu\text{m}^2 for G.654.E, versus approximately 0.2 dB/km attenuation at 1550 nm and approximately 80 \,\mu\text{m}^2 effective area for G.652.D (Aparecido et al., 27 Jan 2025). In the S-band single-channel study, the two actual G.654.E fibre types were Corning TXF and Corning Vascade EX2500, with attenuation at 1490 nm of 0.185 dB/km and 0.168 dB/km, respectively (Aparecido et al., 27 Jan 2025).

These parameters matter because lower attenuation reduces span loss and accumulated ASE burden, while larger effective area reduces nonlinear penalties, including inter-channel stimulated Raman scattering and more general Kerr-limited nonlinear interference. The same structural choices, however, tend to shift the cable cutoff or cutoff wavelength upward. The experimental papers therefore treat G.654.E as a fibre class defined by a tradeoff: the transmission medium is intrinsically favorable for ultra-wideband coherent transport, but its long cutoff wavelength raises a specific concern for S-band operation (Yang et al., 29 Jul 2025).

A further transmission-relevant property discussed in the long-haul SCL study is the higher water absorption peak in the E-band compared with low-water-peak G.652.D. That feature degrades practical Raman pumping efficiency at E-band pump wavelengths and makes G.654.E less naturally compatible with strong distributed Raman amplification, even while its lower loss and larger effective area improve intrinsic transmission performance (Yang et al., 29 Jul 2025).

2. Cutoff wavelength, higher-order modes, and MPI risk

The defining caveat in these studies is the upward-shifted cutoff. In the single-channel S-band experiment, the introduction states that cable cutoff can extend potentially up to 1530 nm, and the tested fibres were intentionally chosen toward the upper end of the manufacturing distribution, with average cable cutoff values around 1480–1500 nm (Aparecido et al., 27 Jan 2025). In the long-haul SCL experiment, the fibre is described as having a cutoff wavelength around 1520 nm, with parts of the transmitted S-band therefore lying below cutoff (Yang et al., 29 Jul 2025).

The concern is standard in single-mode fibre terminology: below cutoff, higher-order mode guidance can occur. In the papers’ framing, that may lead to coupling between guided content and higher-order modes, producing multipath interference and degrading received signal quality. The single-channel study explicitly links the issue to practical cable bend conditions, since below-cutoff behaviour is not merely a nominal waveguide property but can interact with bending-induced coupling under installed-fibre-like conditions (Aparecido et al., 27 Jan 2025).

The prior modelling result cited in the S-band study gives the analytical risk threshold that motivates the experiments: even in adverse loose-tube-cable-like bend conditions, MPI levels in the S-band are below $-70~\text{dB/km}$, corresponding to an SNR penalty of less than 0.1 dB (Aparecido et al., 27 Jan 2025). The later experiments do not derive this criterion, but they are structured as direct validations of whether such modal penalties remain negligible in practice.

A minor textual inconsistency appears in the first paper regarding the assignment of exact average cable cutoff values to TXF and EX2500: the setup section and conclusion swap which fibre is associated with the value near 1500 nm and which with 1480 nm. The tested cutoff range itself is nevertheless clear: the fibres were selected precisely because their cable cutoffs lay in the S-band region and near its upper end (Aparecido et al., 27 Jan 2025).

3. Experimental evidence from single-channel S-band transmission

The first direct experimental assessment of G.654.E fibre for S-band transmission used a coherent single-channel transmission test across 22 channels from 1474 nm to 1525 nm, covering most of the S-band of interest (Aparecido et al., 27 Jan 2025). The optical carrier was generated by an external cavity laser (ECL) with linewidth < 100 kHz, boosted by a thulium-doped fibre amplifier (TDFA), and applied through a polarization controller to a dual-polarization IQ Mach–Zehnder modulator with 35 GHz 3-dB bandwidth. Electrical drive came from 92 GSa/s, 8-bit DACs, and the transmitted format was dual-polarization 16QAM at 16 GBaud or 64 GBaud. Reception used a 70 GHz coherent receiver frontend, waveform capture with a 10-bit, 256 GSa/s Keysight UXR real-time oscilloscope, and pilot-based DSP that fully compensated chromatic dispersion (Aparecido et al., 27 Jan 2025).

The central measurement strategy was a fibre-versus-VOA substitution. An optical switch selected either the actual fibre span or a second VOA path whose attenuation was tuned to match the loss of the fibre path, using a power monitor in the switch. Because the VOA path introduces attenuation but not fibre-induced MPI, comparing the received SNR in the two cases isolates whether the fibre span contributes any excess impairment beyond attenuation and amplifier noise (Aparecido et al., 27 Jan 2025).

The tested distances were approximately 80 km and 160 km. The exact combinations were:

Fibre	Distance(s)
Vascade EX2500	86.2 km, 173.5 km
TXF	81.7 km, 164.6 km
G.652.D reference fibre	161.3 km

The principal metrics were received SNR and OSNR, with OSNR measured in a 0.1 nm reference bandwidth. The paper states that the primary metric was average SNR, measured from five received traces (Aparecido et al., 27 Jan 2025). The rationale was that OSNR captures signal power and ASE over the signal bandwidth, whereas SNR reflects all relevant impairments. Under this criterion, if the SNR after fibre transmission overlaps with the attenuation-matched VOA result and any residual mismatch tracks OSNR differences, then there is no evidence of MPI penalty.

The reported result was that across all 22 wavelengths, both baud rates, and both tested distances, the measured SNRs after transmission through the G.654.E fibres predominantly overlapped with those obtained using the VOA-only path. Quantitatively, the measured SNR mismatch between fibre and VOA was below 0.4 dB for every channel in all configurations. The largest discrepancy occurred for the 64 GBaud, 160 km TXF case at 1474 nm, where the fibre case was 0.39 dB lower than the VOA case; however, the corresponding OSNR difference was 0.33 dB, and the authors therefore attribute the difference to ordinary launch-power or modulator-bias variation rather than MPI (Aparecido et al., 27 Jan 2025).

Under the paper’s experimental criterion, this constitutes MPI-penalty-free S-band transmission over the tested G.654.E fibres. More precisely, the absence of excess SNR degradation relative to the attenuation-matched VOA path implies that ASE noise and transceiver noise are the only significant impairments, with no residual penalty attributable to modal interference detectable under the tested conditions (Aparecido et al., 27 Jan 2025).

4. Long-haul SCL-band feasibility and below-cutoff operation

The second study extends the question from single-channel S-band validation to long-haul ultra-wideband transmission. It demonstrates 1552 km SCL-band transmission over Corning Vascade EX2500 G.654.E fibre, despite the fibre’s cutoff wavelength around 1520 nm, and reports 100.85 Tb/s from GMI and 92.8 Tb/s decoded net rate with Raman assistance (Yang et al., 29 Jul 2025). The total spectral occupancy is stated as 15.08 THz, and the transmission employed 112 GBd signalling, with GS-16QAM in S-band and GS-64QAM in C- and L-bands (Yang et al., 29 Jul 2025).

The long-haul system used a recirculating fibre loop with 18 recirculations, each containing one 86.2 km span, giving 18 \times 86.2~\text{km} = 1551.6~\text{km}, reported as 1552 km. The transmitter employed an external-cavity laser (ECL) and a dual-polarisation IQ thin-film lithium niobate modulator with 3 dB bandwidth ~80 GHz, driven by an AWG. Fourth-order Volterra digital pre-distortion compensated nonlinear response of the DAC, driver amplifier, and modulator, achieving a maximum back-to-back SNR of ~19 dB across all three bands. Co-propagating WDM channels were emulated by spectrally shaped ASE noise from a wideband ASE source shaped by WaveShapers used as WSSs (Yang et al., 29 Jul 2025).

Within the recirculating loop were a pair of AOMs, a polarisation scrambler, three gain blocks—each including an S-band TDFA, a low-gain C-band EDFA, and a low-gain L-band EDFA—together with three backward Raman pumps, WSSs, and VOAs for power balancing (Yang et al., 29 Jul 2025). For G.654.E, the Raman pump wavelengths were 1365 nm, 1385 nm, 1405 nm, with maximum output powers ~500 mW each, all set to maximum because of reduced Raman efficiency. For G.652.D, pumps at 1365, 1385, 1405, 1425 nm were used, with total pump power ~1.5 W (Yang et al., 29 Jul 2025).

The paper’s core below-cutoff finding is explicit: “Negligible penalty due to multipath interference (MPI) was observed for signals below the cutoff wavelength.” The empirical basis is twofold. First, the system transmits over 1552 km with part of the S-band below the nominal cutoff. Second, S-band performance on G.654.E with Raman is reported as similar to that on G.652.D with Raman, which the authors interpret as evidence of negligible MPI-induced penalties for below-cutoff operation (Yang et al., 29 Jul 2025).

This suggests that, at least for the tested EX2500 fibre and the experimental architecture employed, a nominal cutoff near 1520 nm does not preclude practical long-haul SCL operation. The result is not formulated as a universal theorem about all G.654.E links, but as an experimental demonstration that below-cutoff transmission can remain feasible in practice (Yang et al., 29 Jul 2025).

5. Comparison with G.652.D and system-level consequences

Both papers use G.652.D as the practical benchmark, but they do so in different regimes. In the single-channel S-band experiment, the comparison isolates loss and linear-regime SNR behaviour. At 1490 nm over approximately similar distances—173.5 km for Vascade EX2500, 164.6 km for TXF, and 161.3 km for G.652.D—the total span losses were 29.15 dB, 30.45 dB, and 33.87 dB, respectively (Aparecido et al., 27 Jan 2025). This gave EX2500 roughly 2 dB OSNR advantage over TXF and TXF a similar OSNR advantage over G.652.D, with corresponding linear-regime SNR improvements.

For 16 GBaud, the measured maximum SNR values were 20.91 dB for Vascade EX2500, 20.05 dB for TXF, and 17.74 dB for G.652.D, with optimum launch powers of 8.5 dBm, 9.5 dBm, and 9 dBm, respectively. Thus EX2500 improved SNR over G.652.D by about 3.17 dB, while TXF improved it by about 2.31 dB (Aparecido et al., 27 Jan 2025). For 64 GBaud, the available TDFA gain limited launch power to 11 dBm, so optimum launch power could not be determined, but at that capped level the measured SNR values were 17.2 dB for Vascade EX2500, 16.3 dB for TXF, and 14.4 dB for G.652.D, implying gains of 2.8 dB and 1.9 dB over G.652.D (Aparecido et al., 27 Jan 2025).

In the long-haul SCL study, the comparison becomes architectural. The measured total span loss at 1550 nm including splices was 13.0 dB for the 86.2 km G.654.E span and 16.5 dB for the G.652.D span. The paper also states that G.654.E has approximately 0.04 dB/km lower attenuation across the signal wavelengths than the comparison G.652.D fibre (Yang et al., 29 Jul 2025). At the same time, G.652.D is more Raman-friendly, whereas G.654.E has a higher water peak in the E-band and a lower Raman gain coefficient, “approximately half that of the G.652.D fibre” (Yang et al., 29 Jul 2025).

The resulting tradeoff is central. G.652.D offers efficient E-band Raman pumping and, in this experiment, achieves more than 20 dB maximum Raman on-off gain. G.654.E, by contrast, offers lower loss and lower nonlinearity because of its larger effective area. The paper’s principal system-level comparison is therefore that G.654.E with only lumped DFA achieves a throughput comparable to that of the G.652.D fibre with Raman amplification, specifically 88.8 Tb/s decoded for G.654.E without Raman versus 89.9 Tb/s decoded for G.652.D with Raman (Yang et al., 29 Jul 2025).

This is not presented as a blanket superiority claim. Rather, the papers indicate a structured tradeoff: G.652.D is more compatible with strong distributed Raman gain, while G.654.E is intrinsically a better transmission medium in terms of attenuation and effective area. A plausible implication is that fibre choice can substitute, to some degree, for amplifier complexity in future ultra-wideband systems.

6. Amplification, modelling assumptions, and practical limits

The long-haul paper provides the clearest view of how G.654.E fibre interacts with amplification architecture. Raman assistance remains beneficial on G.654.E, despite its lower Raman efficiency: throughput rises from 97.02 Tb/s from GMI and 88.8 Tb/s decoded net rate without Raman to 100.85 Tb/s from GMI and 92.8 Tb/s decoded net rate with Raman (Yang et al., 29 Jul 2025). The largest Raman benefit occurs for channels from 1480 to 1500 nm, which gain an average of 0.86 bit/4D-symbol in GMI, while the rest of the S-band and the C/L bands show similar performance (Yang et al., 29 Jul 2025).

Launch powers in that work were optimized using a simplified ISRS GN model, with the stated simplification that “Wavelength-dependent gains or losses were replaced with average values per band in the model to simplify optimisation and system design.” The optimized total launch powers by band for G.654.E were 19.7 dBm in S-band, 19.4 dBm in C-band, and 18.5 dBm in L-band; for G.652.D they were 19.7 dBm, 16.6 dBm, and 18.5 dBm, respectively (Yang et al., 29 Jul 2025). The paper does not provide explicit equations in the text at hand, so the modelling role remains qualitative in the present record: it is used to optimize pump powers and launch powers, not to furnish a closed-form theory of cutoff or MPI (Yang et al., 29 Jul 2025).

The practical relevance of the single-channel S-band study is strengthened by the fact that the fibres were measured on shipping reels, whose bend diameter is said to be similar to that in a loose-tube cable, making the experiment representative of practical installed-fibre conditions (Aparecido et al., 27 Jan 2025). At the same time, both studies delimit their own scope. The first is based on single-channel measurements across 22 tested wavelengths, rather than a fully loaded WDM system, and infers MPI from absence of excess SNR degradation relative to a VOA reference rather than from direct modal decomposition (Aparecido et al., 27 Jan 2025). The second demonstrates below-cutoff SCL transmission on a specific fibre type—Corning Vascade EX2500—in a recirculating-loop architecture, and its headline 100.85 Tb/s is explicitly a GMI-derived throughput estimate, not a decoded net payload; the corresponding decoded value is 92.8 Tb/s (Yang et al., 29 Jul 2025).

These caveats do not negate the findings, but they define the evidentiary status of current claims. The literature supports the practical feasibility of S-band and SCL-band transmission over G.654.E-compliant fibre under the tested conditions, while leaving open broader questions about generalization across fibre variants, loading conditions, and network architectures.

7. Interpretation and significance

The combined evidence from the 2025 studies establishes a coherent technical picture. G.654.E-compliant fibre is attractive for ultra-wideband coherent systems because it combines ultra-low attenuation and large effective area, yielding lower span loss, lower ASE accumulation, and reduced nonlinear penalties relative to G.652.D (Aparecido et al., 27 Jan 2025). The principal concern—that elevated cable cutoff or cutoff wavelength could render S-band operation impractical through higher-order-mode propagation and MPI—was the main obstacle to broader acceptance of the fibre class for S-band use.

The experiments reported in (Aparecido et al., 27 Jan 2025) and (Yang et al., 29 Jul 2025) directly address that obstacle. In the first, two intentionally high-cutoff G.654.E fibres with cable cutoffs around 1480–1500 nm showed no measurable MPI penalty over approximately 80 km and 160 km, at 16 GBaud and 64 GBaud, across 1474–1525 nm. In the second, 1552 km SCL-band transmission over EX2500 remained feasible despite a cutoff near 1520 nm, and the authors explicitly report negligible penalty due to multipath interference (MPI) for below-cutoff signals (Yang et al., 29 Jul 2025).

The broader significance is that G.654.E does not merely tolerate S-band use; under the investigated conditions it can improve performance. In the linear regime, lower attenuation translated directly into SNR gains over G.652.D in the S-band single-channel study (Aparecido et al., 27 Jan 2025). In the long-haul wideband regime, the intrinsic fibre advantages were strong enough that lumped-only amplification on G.654.E approached the throughput of Raman-assisted G.652.D, despite G.654.E being less Raman-efficient (Yang et al., 29 Jul 2025).

This suggests that future ultra-wideband terrestrial systems may treat G.654.E not as an S-band-compromised speciality fibre, but as a viable and in some cases preferable transmission medium whose principal liability—the long cutoff—requires empirical validation rather than categorical exclusion. The present experimental record supports that view, while also indicating that Raman architecture choices remain nuanced, especially toward the short-wavelength edge of the S-band (Yang et al., 29 Jul 2025).