Abstract
A number of simplified models, based on perturbation theory, have been proposed for the fiber-optical channel and have been extensively used in the literature. Although these models are mainly developed for the low-power regime, they are used at moderate or high powers as well. It remains unclear to what extent the capacity of these models is affected by the simplifying assumptions under which they are derived. In this paper, we consider single channel data transmission based on three continuous-time optical models: i) a regular perturbative channel, ii) a logarithmic perturbative channel, and iii) the stochastic nonlinear Schrödinger (NLS) channel. We apply two simplifying assumptions on these channels to obtain analytically tractable discrete-time models: we neglect the channel memory (fiber dispersion) and we use a sampling receiver. We investigate the per-sample capacity of these simplified discrete-time models, establishing tight bounds for the regular perturbative channel, obtaining the capacity of the logarithmic perturbative channel, and presenting a novel upper bound on the capacity of the zero-dispersion NLS channel. Our results illustrate that the capacity of these models departs from each other at high powers because they yield different capacity pre-logs. Since all three models are based on the same physical channel, our results highlight that care must be exercised in using simplified channel models in the high-power regime.
Introduction
Fiber-optic cables serve as the backbone of global data traffic, yet determining their fundamental data capacity remains one of the most significant open problems in communication theory. The physics of light propagation in glass is governed by the Stochastic Nonlinear Schrödinger (NLS) equation, a complex model involving a non-trivial interplay between signal power, noise, and chromatic dispersion. Because the NLS equation is mathematically cumbersome for direct capacity analysis, researchers frequently rely on simplified models to estimate performance limits.
However, these simplifications—often derived under low-power assumptions—are frequently applied to moderate or high-power regimes where their validity is questionable. This paper investigates three such models to determine how these mathematical shortcuts distort capacity estimates as signal power increases.
Simplified Channel Models
The study focuses on three discrete-time models derived from the NLS equation under the assumptions of zero dispersion (memoryless) and a sampling receiver:
- Regular Perturbative Channel (RPC): This model treats nonlinearity as a small additive perturbation to the signal. It is widely used due to its mathematical simplicity but assumes that nonlinear effects do not fundamentally change the signal structure.
- Logarithmic Perturbative Channel (LPC): This model treats nonlinearity primarily as a phase shift. It is often employed in scenarios where phase noise is the dominant impairment.
- Memoryless NLS Channel (MNC): This is a more physically grounded simplification. It retains the core structure of the NLS equation but sets the dispersion parameter to zero, focusing purely on the interaction between nonlinearity and noise.
Methodology
To evaluate these models, the authors employ information-theoretic tools to establish tight upper and lower bounds on "per-sample capacity."
- RPC Analysis: Using the Entropy Power Inequality, the authors derive a closed-form lower bound. They further utilize a "dual capacity" expression to establish a tight upper bound, providing a clear window into the model's behavior.
- LPC Analysis: The authors demonstrate that the LPC is mathematically equivalent to a standard Additive White Gaussian Noise (AWGN) channel, allowing for a direct calculation of its capacity.
- MNC Analysis: The study introduces a novel upper bound for the MNC that significantly improves upon previous benchmarks in the literature (e.g., the work of Kramer et al.), offering a more accurate characterization of the "true" simplified fiber capacity.
Key Insights: The Pre-log Divergence
The most striking finding of this research is the divergence of the "capacity pre-log"—the factor determining the rate at which capacity grows as signal power ($P$) increases. While all three models are derived from the same physical foundation, their predicted growth rates at high power are fundamentally different:
- The RPC yields a pre-log of 3, suggesting that capacity explodes as power increases. This is identified as an artifact of the model; the additive treatment of nonlinearity causes an artificial increase in perceived signal power.
- The LPC yields a pre-log of 1, behaving identically to a linear AWGN cable.
- The MNC yields a pre-log of only 1/2, representing a much more conservative and physically realistic growth rate.
Numerical analysis shows that these models begin to depart from one another significantly at relatively low power levels. The RPC starts to diverge from the more accurate MNC at approximately -15 dBm, while the LPC holds its consistency until roughly -5 dBm.
Conclusion and Implications
The results prove that perturbative models (RPC and LPC) are only reliable in low-power regimes. In high-power scenarios, the RPC in particular becomes "grossly inaccurate," overestimating capacity due to its mathematical structure. This highlights a significant danger for engineers: using these simplified models to design high-capacity, high-power coherent systems may lead to substantial design errors.
While the "zero-dispersion" assumption used here is a mathematical convenience that does not capture the full complexity of real fiber (which possesses memory), this study serves as a critical first step. Future work must bridge the gap by incorporating non-zero dispersion and more practical, band-limited receivers to refine our understanding of the fundamental limits of optical communication.
Summary updated