⚡ electrical engineering

The Coherent Polynomials Closed-Form Model for Evaluating Nonlinear Interference in Any Island

This paper presents an improved Coherent Polynomials Closed-Form Model that enhances the accuracy of nonlinear interference evaluation in any optical link by incorporating spectral NLI PSD and coherent accumulation, enabling precise calculations across diverse systems without relying on machine-learning corrections.

Original authors: Yifeng Gao, Yanchao Jiang, Pierluigi Poggiolini

Published 2026-05-05

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Yifeng Gao, Yanchao Jiang, Pierluigi Poggiolini

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to predict how much static noise (interference) will ruin a message sent through a very long, complex fiber-optic cable. In the world of high-speed internet, this noise is called Nonlinear Interference (NLI). If you can't predict it accurately, your internet connection might slow down or drop.

For a long time, engineers had two ways to predict this noise:

The "Supercomputer" Method: Extremely accurate, but it takes so long to calculate that it's useless for real-time planning.
The "Quick Sketch" Method (PCFM1): Very fast, but it makes a lot of rough guesses. It assumes the noise looks the same everywhere and ignores how different parts of the signal interact. This worked fine for old, simple systems, but it started failing badly in modern, high-speed, and low-dispersion systems.

This paper introduces PCFM2, a new "Quick Sketch" method that is just as fast as the old one but as accurate as the supercomputer method.

Here is how the authors achieved this, explained through simple analogies:

1. The Problem: The "Flat Map" vs. The "Mountain Range"

The old method (PCFM1) treated the noise like a flat, white sheet of paper. It assumed that if you measured the noise at the center of a channel, the noise was exactly the same everywhere else in that channel.

The Reality: The noise is actually a mountain range with peaks and valleys. In modern systems, the shape of this "noise mountain" changes depending on where you look and how the signal travels.
The Fix: PCFM2 stops assuming the noise is flat. Instead, it calculates the exact shape of the noise at every specific frequency, like mapping the actual mountains rather than guessing they are all flat.

2. The Problem: The "Rough Box" vs. The "Jigsaw Puzzle"

To calculate the total noise, engineers have to sum up contributions from different parts of the signal.

The Old Way: They tried to fit these complex, diamond-shaped noise areas into simple square boxes. If the noise shape was a weird diamond or a long, thin strip, the square box either missed parts of the noise or included empty space where there was no noise. To fix this, they used a "magic correction factor" (Machine Learning) to guess the error, but this only worked for specific, high-dispersion scenarios.
The New Way (PCFM2): Imagine you have a complex shape (like a jigsaw piece). Instead of forcing it into a square, PCFM2 covers the shape with many small, adjacent rectangles that fit together perfectly.
- Analogy: Think of tiling a bathroom floor. The old method used giant square tiles that left gaps or overlapped. The new method uses small, custom-cut tiles that fit the exact shape of the floor, no matter how weird it is. This allows the math to be solved exactly without needing any "magic guesses."

3. The Problem: The "Solo Walk" vs. The "Choir"

Signals travel through many segments of fiber (spans).

The Old Way: It assumed that the noise generated in one segment was completely independent of the noise in the next segment. It was like listening to a choir where everyone sings a different song, and you just add up the volume of each singer individually.
The Reality: In modern systems, the noise from different segments can "sync up" (cohere), creating a much louder, more destructive roar. It's like a choir where everyone suddenly harmonizes perfectly; the volume spikes much higher than the sum of individual voices.
The Fix: PCFM2 explicitly tracks how the "phase" (the timing) of the noise evolves along the entire link. It calculates how the noise from Segment 1 interacts with Segment 2, 3, and so on, capturing this "harmonizing" effect accurately, even when the fiber has very low dispersion (which usually makes this effect tricky to predict).

The Result

By combining these three improvements—mapping the exact noise shape, using perfect-fitting rectangles, and tracking the "harmonizing" of noise across the whole link—the authors created a model that:

Is fast: It uses a closed-form formula (a direct math equation), so it doesn't need a supercomputer.
Is accurate: It works perfectly even in difficult scenarios like low-dispersion fibers and multi-subcarrier systems (where many data streams are packed together).
Needs no "Magic": It doesn't rely on machine learning or correction factors to fix its mistakes because the math itself is now precise.

In short, PCFM2 is a new, smarter ruler that measures the "noise" in fiber optics with perfect precision, regardless of how complex the system is, allowing engineers to design faster and more reliable internet connections without needing to run slow, heavy simulations.

Technical Summary: The Coherent Polynomials Closed-Form Model for Evaluating Nonlinear Interference in Any Island

Problem Statement
Accurate modeling of nonlinear interference (NLI) is critical for the design of modern optical fiber systems, particularly in emerging scenarios involving ultra-wideband, low-dispersion, and multi-subcarrier transmissions. While the Gaussian Noise (GN) and Enhanced GN (EGN) models provide reliable benchmarks, their reliance on numerical integration makes them computationally prohibitive for real-time applications. Consequently, Closed-Form Models (CFMs) have been developed to enable fast estimation. However, existing CFMs, including the Polynomial Closed-Form Model (PCFM1), rely on simplifying assumptions that degrade accuracy in practical, complex scenarios. Specifically, these models typically: (i) assume the NLI power spectral density (PSD) is spectrally flat and evaluate it only at the channel center frequency; (ii) approximate GN-model integration "islands" with simple rectangles or stripes; (iii) neglect multi-channel interference (MCI); and (iv) assume incoherent accumulation of NLI across spans or model coherence crudely. These approximations fail significantly in low-dispersion and digital subcarrier systems where MCI and coherent accumulation effects are dominant.

Methodology
The authors propose an upgraded framework, termed PCFM2, which removes these limitations within a unified analytical structure. The methodology proceeds as follows:

Accumulated-Dispersion Representation: Starting from the full-link GN-model formulation, the authors introduce an accumulated-dispersion representation that explicitly captures phase evolution along the link. This allows for the decomposition of NLI contributions into intra-span and inter-span terms, both described by a common analytical structure.
Arbitrary Rectangular Integration: Unlike PCFM1, which uses square approximations for Self-Channel Interference (SCI) islands, PCFM2 represents any integration island (SCI, XCI, or MCI) as a union of adjacent rectangular sub-domains. This allows the model to handle arbitrary integration regions and frequency offsets without relying on Machine Learning Correction Factors (MLCF).
Polynomial Expansion of Spatial Power Profile (SPP): The effective SPP is represented as a polynomial expansion. This transforms the complex double integrals over frequency and span coordinates into closed-form expressions.
Coherent Accumulation Formulation: The model explicitly retains the coherent accumulation of NLI across spans. It derives a generalized kernel, $R^{(v,w)}_x(f)$ , which describes the cross-correlation between any pair of spans ( $v$ and $w$ ). This kernel is evaluated by reducing the frequency integrals to sine-integral functions and the spatial integrals to piecewise polynomials, resulting in a fully closed-form solution.
Mathematical Derivation: The derivation involves defining local span coordinates, introducing scaled variables for dispersion, and mapping rectangular domains to intervals to solve the integrals analytically. The solution handles both positive and negative dispersion values and includes a limiting expression for zero-dispersion scenarios.

Key Contributions

Removal of Spectral Flatness Assumption: PCFM2 evaluates NLI PSD at arbitrary frequencies within the channel, accurately capturing spectral variations that PCFM1 misses.
Generalized Integration Domains: By covering arbitrary integration islands with adjacent rectangles, the model accurately handles complex geometries arising in multi-subcarrier systems and non-uniform channel spacing.
Explicit Coherent Modeling: The framework provides a unified analytical treatment for both intra-span and inter-span coherence, eliminating the need for incoherent accumulation assumptions.
Elimination of MLCF Dependence: The model achieves high accuracy through intrinsic analytical improvements rather than relying on machine-learning correction factors, which are often ineffective in low-dispersion regimes.

Results
The paper validates PCFM2 against the numerical GN-model reference through three specific tests:

SCI PSD Accuracy: In a 140 GBaud single-channel system over 100 km, PCFM2 accurately computes the full SCI PSD at any frequency offset. While PCFM1 (even with MLCF) shows significant errors (approx. 1 dB) in low-dispersion regimes ( $|\beta_{2,eff}| \in [0, 25]$ ps²/km), PCFM2 maintains high accuracy across the entire range, including near-zero dispersion.
Multi-Subcarrier Systems: In an 8-subcarrier system, PCFM2 accurately predicts the NLI PSD, capturing substantial deviations from flatness and asymmetric shapes caused by XCI and MCI. PCFM1 fails to capture these features, and MLCF does not effectively correct PCFM1 in these low-dispersion, multi-subcarrier scenarios.
Multi-Span Coherence: In a 10-span link, PCFM2 closely tracks the coherent GN-model reference. The results demonstrate that PCFM2 correctly captures the constructive accumulation of NLI due to phase correlation, showing that the gap between coherent and incoherent accumulation widens as dispersion decreases.

Significance
The paper claims that PCFM2 provides a computationally efficient and analytically rigorous tool for analyzing NLI accumulation in practical systems. By removing the restrictive assumptions of previous CFMs, the model is particularly relevant for low-dispersion and digital subcarrier scenarios where traditional models fail. The ability to evaluate NLI over any rectangular integration domain without machine-learning corrections offers a robust solution for the design and optimization of next-generation optical fiber systems.

1. The Problem: The "Flat Map" vs. The "Mountain Range"

2. The Problem: The "Rough Box" vs. The "Jigsaw Puzzle"

3. The Problem: The "Solo Walk" vs. The "Choir"

The Result

More like this