SCL-Band Long-Haul Transmission

• SCL-band long-haul transmission is the simultaneous use of S, C, and L bands in DWDM systems to expand capacity through integrated spectral management.
• Advanced designs leverage hybrid amplification (Raman plus lumped gain) and closed-form GN/EGN models to accurately predict and optimize performance amid ISRS and nonlinear effects.
• Experimental and modeled studies validate that dynamic launch power optimization and gain equalization yield significant throughput gains and energy per bit improvements.

Searching arXiv for recent and foundational S+C+L-band long-haul transmission papers to ground the article. Search query: "SCL-band long-haul transmission Raman GN model S+C+L" SCL-band long-haul transmission denotes ultra-wideband coherent DWDM in which the S, C, and L wavelength bands are operated simultaneously on a single fibre over many spans, typically with band-specific lumped amplifiers and, in many realizations, distributed Raman amplification or other hybrid gain mechanisms. In recent work, SCL is treated not as three isolated bands but as a single strongly coupled spectral system whose performance is set by wavelength-dependent loss, amplifier noise, chromatic dispersion, Kerr nonlinearity, and inter-channel stimulated Raman scattering (ISRS); accordingly, launch-power shaping, gain equalization, and closed-form GN/EGN modeling have become central design tools (Sohanpal et al., 30 Jul 2025, Yang et al., 2024).

1. Spectral scope and system architectures

SCL-band systems are defined by simultaneous use of the S, C, and L bands in a fully loaded multi-band DWDM configuration rather than by time- or space-separated operation. In one explicit definition, a CL-band system spans 1530–1620 nm with 73 channels, whereas an SCL-band system spans 1460–1620 nm with 127 channels, requiring a thulium-doped fibre amplifier in S-band and EDFAs in C- and L-band (Sohanpal et al., 30 Jul 2025). Other recent realizations occupy 12.4 THz across S+C+L in a 1065 km recirculating-loop experiment, 15.08 THz from 1480 to 1620 nm over 1552 km on G.654.E fibre, 18 THz over 1000 km in closed-form C+L+S optimization, and 20 THz over 7000 km in a transoceanic proxy analysis (Yang et al., 2024, Yang et al., 29 Jul 2025, Jiang et al., 2024, Semrau et al., 2020).

The link architecture is correspondingly heterogeneous. A representative experimental hybrid system uses low-OH peak SSMF, 71 km spans, three gain blocks per span, and four backward Raman pumps at 1365, 1385, 1405, and 1425 nm with total pump power 1.5 W (Yang et al., 2024). A distinct long-haul realization on G.654.E uses 18 spans of 86.2 km with three gain blocks per span composed of an S-band TDFA and low-gain C- and L-band EDFAs, with optional backward Raman pumping (Yang et al., 29 Jul 2025). At the analytical level, a 1000 km terrestrial-style C+L+S design uses 10 spans of 100 km SMF with 50 channels per band at 100 GBaud, while a transoceanic 7000 km study considers 364 channels at 50 GBd across a 20 THz S+C+L window (Jiang et al., 2024, Semrau et al., 2020).

These architectures already indicate a defining feature of SCL long-haul transmission: the usable spectrum is enlarged, but the amplification problem becomes intrinsically multi-band. Separate amplifier types, band transitions, guard regions, WSS shaping, and Raman-induced spectral coupling are no longer peripheral implementation details; they are part of the transmission design itself.

2. Dominant physical mechanisms across S, C, and L

The central wideband impairment is ISRS, which transfers power from shorter wavelengths to longer wavelengths along the fibre. In the 20 THz transoceanic S+C+L analysis, shorter-wavelength S-band channels act as Raman pump and lose power, while longer-wavelength L-band channels gain power; with non-ideal gain equalization this leads to cumulative spectral tilt and an end-of-link SNR penalty, especially in S-band (Semrau et al., 2020). In a field-deployed 15.6 THz S+C+L metro experiment, a 5 dB launch pre-tilt was experimentally optimized so that S-band launched with higher power density and the resulting ISRS produced a more balanced received spectrum (Yang et al., 2024).

Kerr nonlinearity remains the second defining mechanism, but its distribution is highly wavelength dependent in SCL systems. In the 18 THz C+L+S closed-form study, S-band channels are penalized by higher loss, higher effective $\gamma$, lower dispersion, and stronger ISRS depletion into C/L, which explains why S-band initially underperforms in DFA-only configurations (Jiang et al., 2024). The experimentally validated 1065 km hybrid Raman system further shows that maximum nonlinear interference occurs around 1505 nm, matching the peak Raman on/off gain and local power maximum there; C- and L-band see lower NLI because their launch powers are set about 2 dB below S-band (Yang et al., 2024).

ASE noise also becomes explicitly band dependent because the gain distribution is hybrid. In the 1065 km experiment, Raman gain supplies most of the gain in S-band and part of C-band, while L-band relies much more heavily on lumped gain; as a result, the main contributor to ASE in L-band is the high lumped gain in gain block 2, whereas S-band combines higher TDFA noise figure with reduced need for lumped gain (Yang et al., 2024). In the G.654.E experiment, Raman contributes less strongly than on G.652.D because the Raman gain coefficient is lower and pump wavelengths in the E-band see higher attenuation, yet the fibre’s ultra-low loss and large effective area partly compensate by reducing both ASE accumulation and nonlinear interference (Yang et al., 29 Jul 2025).

A practical implication is that SCL transmission cannot be characterized by a single “best band.” C-band often has the best conventional amplifier performance, L-band benefits from ISRS transfer, and S-band contributes large additional bandwidth but is simultaneously the most sensitive to amplifier limitations, Raman design, and spectral tilt. This suggests that the relevant optimization variable is not merely total launch power but the entire wavelength-dependent power evolution.

3. Analytical models and optimization frameworks

Recent SCL design work relies heavily on closed-form or semi-analytical GN/EGN models that explicitly include ISRS, Raman gain, and wavelength-dependent fibre parameters. The experimentally validated hybrid Raman model adopts the standard GN relation

$P_\text{NLI} \approx \eta P^3,$

and extends it to arbitrary $(z,\lambda)$-dependent power profiles, hybrid amplification, and wavelength-dependent $\beta_2$, $\beta_3$, and loss (Yang et al., 2024). In that formulation, the total received SNR of channel $f_0$ is modeled as

$$\text{SNR}(f_0)=\frac{P_\text{signal}(f_0)}{P_\text{TRx}(f_0)+P_\text{ASE}(f_0)+P_\text{NLI}(f_0)},$$

with Raman-generated ASE and lumped-amplifier ASE handled separately (Yang et al., 2024).

For broader optimization, the closed-form CFM/EGN framework used for 18 THz C+L+S links treats launch powers and Raman pump parameters as coupled design variables. Two objective functions are used: throughput-only optimization,

$$f_{\text{obj}}=\mathrm{mean}\left(\mathrm{IR}_{\text{Rx}}^n\right),$$

and throughput-plus-flatness optimization,

$$f_{\text{obj}}= \mathrm{mean}\left(\mathrm{IR}_{\text{Rx}}^n\right) - \left|\mathrm{IR}_{\text{Rx}}^{\max}-\mathrm{IR}_{\text{Rx}}^{\min}\right|,$$

which explicitly penalizes GSNR/IR spread across channels (Jiang et al., 2024). In the 20 THz transoceanic study, a modulation-format dependent closed-form ISRS GN model is used together with particle swarm optimization followed by gradient descent to optimize per-channel launch powers and code-rate assignment under S-band 16-QAM and C+L-band 64-QAM assumptions (Semrau et al., 2020).

These models are not only predictive but operationally important because brute-force SSFM over 12–20 THz and many spans is computationally prohibitive for iterative design. The 1065 km Raman validation is especially significant in this respect: the semi-analytical Raman power profile matches the measured 71 km span power evolution with average error 0.23 dB, and the resulting channel-wise SNR prediction over 12.4 THz remains within 0.38 dB after 355 km and 0.60 dB after 1065 km on average (Yang et al., 2024). This fixes one of the traditional objections to GN-style planning in multi-band Raman-amplified links, namely that arbitrary wavelength-dependent gain/loss profiles would make closed-form prediction too inaccurate to be useful.

4. Experimental and modeled performance regimes

Recent literature spans metropolitan field fibre, laboratory long-haul loops, ultra-low-loss terrestrial fibre, and transoceanic proxy analyses. The resulting performance envelope is summarized below.

Regime	Configuration	Reported outcome
Experimental long-haul	12.4 THz S+C+L, 15 spans of 71 km, hybrid Raman + TDFA/EDFA	94.28 Tb/s at 1065 km; average SNR estimation error 0.60 dB (Yang et al., 2024)
Experimental long-haul	15.08 THz SCL on G.654.E, 18 × 86.2 km	100.85 Tb/s GMI and 92.8 Tb/s net at 1552 km (Yang et al., 29 Jul 2025)
Modeled long-haul	18 THz C+L+S, 10 × 100 km, optimized S-only Raman	119.0 Tb/s at 1000 km; GSNR within $\pm 0.5$ dB under flatness-oriented optimization (Jiang et al., 2024)
Modeled transoceanic	20 THz S+C+L, 100 × 70 km	119.5 Tb/s with ideal gain equalization; 112.3 Tb/s with non-ideal gain equalization (Semrau et al., 2020)
Experimental field fibre	15.6 THz S+C+L over 39 km deployed metro fibre	202.3 Tb/s GMI and 189.5 Tb/s net with lumped amplification only (Yang et al., 2024)

Two experimental results are particularly important for long-haul interpretation. First, the 1065 km hybrid-amplified system demonstrates that the analytically predicted Raman tilt, ASE, and NLI can be matched to measurement with sub-dB SNR accuracy across 12.4 THz, which makes the model a practical engineering tool rather than merely a qualitative guide (Yang et al., 2024). Second, the 1552 km G.654.E experiment shows that SCL transmission remains feasible even though the fibre cutoff wavelength is around 1520 nm; the reported outcome is 100.85 Tb/s in GMI and 92.8 Tb/s net over 15.08 THz, and the achievable-information-rate with lumped amplification is comparable to that of G.652.D links with distributed Raman amplification (Yang et al., 29 Jul 2025).

The 39 km field demonstration is not a long-haul experiment, but it clarifies deployment-relevant mechanisms that remain present at longer reach: 32 GBaud channelization on a 32.5 GHz grid, band-specific lumped amplification, strong ISRS over 125 nm, and the need for experimentally optimized pre-tilt under realistic power-safety restrictions (Yang et al., 2024). A plausible implication is that metro and long-haul SCL systems share the same spectral-management problem, but differ in whether transceiver limits or ASE/NLI accumulation dominate.

5. Amplification strategies, gain equalization, and energy per bit

Three amplification paradigms recur in SCL long-haul work: lumped-only per-band amplification, hybrid lumped plus distributed Raman amplification, and low-loss-fibre operation that reduces the need for Raman. In the 1065 km hybrid system, Raman supplies the main gain in S-band and part of C-band, while a TDFA and low-gain EDFAs provide residual lumped compensation; the measured SNR dip at 1505 nm shows the canonical trade-off, because the Raman gain peak improves signal power while also creating a local NLI hotspot (Yang et al., 2024). In the 18 THz optimization study, adding only three backward pumps with total power below 1 W raises throughput from 93.0 Tb/s in DFA-only C+L+S to 119.0 Tb/s and reduces GSNR spread to within $P_\text{NLI} \approx \eta P^3,$0 dB under the flatness-aware objective (Jiang et al., 2024).

Dynamic gain equalization is similarly decisive in transoceanic-style SCL systems. In the 7000 km analysis, ideal equalization restores the transmitted launch power profile after every span and yields about 119.5 Tbit/s, whereas compensating only 50% of the ISRS power transfer per span and fully resetting every fifth span reduces throughput to about 112.3 Tbit/s, a roughly 6% penalty (Semrau et al., 2020). This is not a secondary impairment: it quantifies how much throughput is lost when ISRS is acknowledged but not fully neutralized.

The energetic dimension introduces a further qualification. A detailed power-consumption study using measured benchtop amplifiers and Raman pumps shows that hybrid Raman+lumped SCL links can reduce energy per bit by up to about 26% relative to lumped-only links, but only in the long-haul regime (Sohanpal et al., 30 Jul 2025). At 100 spans, the total SCL-system energy-per-bit reduction for a 4-pump hybrid design is 12%, 20%, and 26% for $P_\text{NLI} \approx \eta P^3,$1, 2, and 8 W, respectively; at 1 span, however, additional Raman pumps increase energy per bit because the system remains largely transceiver-noise limited (Sohanpal et al., 30 Jul 2025). One common misconception is therefore that Raman is always energetically preferable in SCL systems. The available evidence is narrower: Raman is advantageous when accumulated ASE, ISRS, and nonlinear penalties materially constrain throughput, not when the link is already dominated by transceiver SNR.

6. Constraints, misconceptions, and future directions

A first misconception is that adding S-band is automatically equivalent to a proportional increase in useful long-haul capacity. The modeled record is more nuanced. Extending a 12 THz C+L design to 18 THz C+L+S without Raman raises throughput from 67.5 Tb/s to 93.0 Tb/s, but the incremental S-band contribution is markedly less efficient because S-band channels suffer higher loss, higher effective $P_\text{NLI} \approx \eta P^3,$2, lower dispersion, and stronger ISRS depletion; only after S-focused backward Raman is introduced does the C+L+S system reach 119.0 Tb/s with nearly flat GSNR (Jiang et al., 2024). The transoceanic 20 THz study reaches the same conclusion in a different form: S+C+L gains are large, but only when launch powers, modulation formats, and dynamic gain equalization are jointly optimized (Semrau et al., 2020).

A second misconception is that a cutoff wavelength in the S-band precludes SCL operation on ultra-low-loss G.654.E fibre. The 1552 km experiment explicitly contradicts that view: despite a cutoff around 1520 nm, the system reports no significant MPI-induced penalty below cutoff and achieves 100.8 Tb/s GMI over 15.08 THz (Yang et al., 29 Jul 2025). The compensating mechanism is not Raman strength—Raman is actually weaker in G.654.E than in G.652.D—but rather lower attenuation and lower nonlinearity from the larger effective area (Yang et al., 29 Jul 2025).

A third constraint lies in DSP and hardware, not only in fibre propagation. Equalization-enhanced phase noise scales as

$P_\text{NLI} \approx \eta P^3,$3

so modern long-haul S/C/L operation at high baud rates and higher-order modulation tightens the coupling between linewidth budget, chromatic-dispersion compensation strategy, and carrier-phase recovery (Xu et al., 2016). At the transceiver side, field and laboratory wideband experiments also show that transceiver SNR ceilings can become comparable to span noise in parts of the spectrum, especially in C and L (Yang et al., 2024).

The near-term trajectory points in two directions. One is refinement inside SCL itself: better wavelength-dependent noise-figure and insertion-loss models, flatter S-band amplification, and faster joint optimization of launch powers and Raman pumps, all of which are directly motivated by the residual band-edge discrepancies seen in current validation work (Yang et al., 2024). The other is bandwidth extension beyond SCL. Four-band S-to-U long-haul transmission over 22.05 THz and 1040 km has already been demonstrated using hybrid PPLN-based OPAs and EDFAs (Shimizu et al., 10 Dec 2025). This suggests that SCL is best understood not as a terminal architecture but as the present operating point in a broader multiband expansion of standard single-mode fibre systems.